Print this page
dccp: properties
| Split |
Close |
| Expand all |
| Collapse all |
--- old/usr/src/uts/common/inet/ip/ip_if.c
+++ new/usr/src/uts/common/inet/ip/ip_if.c
1 1 /*
2 2 * CDDL HEADER START
3 3 *
4 4 * The contents of this file are subject to the terms of the
5 5 * Common Development and Distribution License (the "License").
6 6 * You may not use this file except in compliance with the License.
7 7 *
8 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9 9 * or http://www.opensolaris.org/os/licensing.
10 10 * See the License for the specific language governing permissions
11 11 * and limitations under the License.
12 12 *
13 13 * When distributing Covered Code, include this CDDL HEADER in each
14 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15 15 * If applicable, add the following below this CDDL HEADER, with the
16 16 * fields enclosed by brackets "[]" replaced with your own identifying
17 17 * information: Portions Copyright [yyyy] [name of copyright owner]
18 18 *
19 19 * CDDL HEADER END
20 20 */
21 21 /*
22 22 * Copyright (c) 1991, 2010, Oracle and/or its affiliates. All rights reserved.
23 23 * Copyright (c) 1990 Mentat Inc.
24 24 */
25 25
26 26 /*
27 27 * This file contains the interface control functions for IP.
28 28 */
29 29
30 30 #include <sys/types.h>
31 31 #include <sys/stream.h>
32 32 #include <sys/dlpi.h>
33 33 #include <sys/stropts.h>
34 34 #include <sys/strsun.h>
35 35 #include <sys/sysmacros.h>
36 36 #include <sys/strsubr.h>
37 37 #include <sys/strlog.h>
38 38 #include <sys/ddi.h>
39 39 #include <sys/sunddi.h>
40 40 #include <sys/cmn_err.h>
41 41 #include <sys/kstat.h>
42 42 #include <sys/debug.h>
43 43 #include <sys/zone.h>
44 44 #include <sys/sunldi.h>
45 45 #include <sys/file.h>
46 46 #include <sys/bitmap.h>
47 47 #include <sys/cpuvar.h>
48 48 #include <sys/time.h>
49 49 #include <sys/ctype.h>
50 50 #include <sys/kmem.h>
51 51 #include <sys/systm.h>
52 52 #include <sys/param.h>
53 53 #include <sys/socket.h>
54 54 #include <sys/isa_defs.h>
55 55 #include <net/if.h>
56 56 #include <net/if_arp.h>
57 57 #include <net/if_types.h>
58 58 #include <net/if_dl.h>
59 59 #include <net/route.h>
60 60 #include <sys/sockio.h>
61 61 #include <netinet/in.h>
62 62 #include <netinet/ip6.h>
63 63 #include <netinet/icmp6.h>
64 64 #include <netinet/igmp_var.h>
65 65 #include <sys/policy.h>
66 66 #include <sys/ethernet.h>
67 67 #include <sys/callb.h>
68 68 #include <sys/md5.h>
69 69
70 70 #include <inet/common.h> /* for various inet/mi.h and inet/nd.h needs */
71 71 #include <inet/mi.h>
72 72 #include <inet/nd.h>
73 73 #include <inet/tunables.h>
74 74 #include <inet/arp.h>
75 75 #include <inet/ip_arp.h>
76 76 #include <inet/mib2.h>
77 77 #include <inet/ip.h>
78 78 #include <inet/ip6.h>
79 79 #include <inet/ip6_asp.h>
80 80 #include <inet/tcp.h>
81 81 #include <inet/ip_multi.h>
82 82 #include <inet/ip_ire.h>
83 83 #include <inet/ip_ftable.h>
84 84 #include <inet/ip_rts.h>
85 85 #include <inet/ip_ndp.h>
86 86 #include <inet/ip_if.h>
87 87 #include <inet/ip_impl.h>
88 88 #include <inet/sctp_ip.h>
89 89 #include <inet/ip_netinfo.h>
90 90 #include <inet/ilb_ip.h>
91 91
92 92 #include <netinet/igmp.h>
93 93 #include <inet/ip_listutils.h>
94 94 #include <inet/ipclassifier.h>
95 95 #include <sys/mac_client.h>
96 96 #include <sys/dld.h>
|
↓ open down ↓ |
96 lines elided |
↑ open up ↑ |
97 97 #include <sys/mac_flow.h>
98 98
99 99 #include <sys/systeminfo.h>
100 100 #include <sys/bootconf.h>
101 101
102 102 #include <sys/tsol/tndb.h>
103 103 #include <sys/tsol/tnet.h>
104 104
105 105 #include <inet/rawip_impl.h> /* needed for icmp_stack_t */
106 106 #include <inet/udp_impl.h> /* needed for udp_stack_t */
107 +#include <inet/dccp_stack.h> /* needed for dccp_stack_t */
107 108
108 109 /* The character which tells where the ill_name ends */
109 110 #define IPIF_SEPARATOR_CHAR ':'
110 111
111 112 /* IP ioctl function table entry */
112 113 typedef struct ipft_s {
113 114 int ipft_cmd;
114 115 pfi_t ipft_pfi;
115 116 int ipft_min_size;
116 117 int ipft_flags;
117 118 } ipft_t;
118 119 #define IPFT_F_NO_REPLY 0x1 /* IP ioctl does not expect any reply */
119 120 #define IPFT_F_SELF_REPLY 0x2 /* ioctl callee does the ioctl reply */
120 121
121 122 static int nd_ill_forward_get(queue_t *, mblk_t *, caddr_t, cred_t *);
122 123 static int nd_ill_forward_set(queue_t *q, mblk_t *mp,
123 124 char *value, caddr_t cp, cred_t *ioc_cr);
124 125
125 126 static boolean_t ill_is_quiescent(ill_t *);
126 127 static boolean_t ip_addr_ok_v4(ipaddr_t addr, ipaddr_t subnet_mask);
127 128 static ip_m_t *ip_m_lookup(t_uscalar_t mac_type);
128 129 static int ip_sioctl_addr_tail(ipif_t *ipif, sin_t *sin, queue_t *q,
129 130 mblk_t *mp, boolean_t need_up);
130 131 static int ip_sioctl_dstaddr_tail(ipif_t *ipif, sin_t *sin, queue_t *q,
131 132 mblk_t *mp, boolean_t need_up);
132 133 static int ip_sioctl_slifzone_tail(ipif_t *ipif, zoneid_t zoneid,
133 134 queue_t *q, mblk_t *mp, boolean_t need_up);
134 135 static int ip_sioctl_flags_tail(ipif_t *ipif, uint64_t flags, queue_t *q,
135 136 mblk_t *mp);
136 137 static int ip_sioctl_netmask_tail(ipif_t *ipif, sin_t *sin, queue_t *q,
137 138 mblk_t *mp);
138 139 static int ip_sioctl_subnet_tail(ipif_t *ipif, in6_addr_t, in6_addr_t,
139 140 queue_t *q, mblk_t *mp, boolean_t need_up);
140 141 static int ip_sioctl_plink_ipmod(ipsq_t *ipsq, queue_t *q, mblk_t *mp,
141 142 int ioccmd, struct linkblk *li);
142 143 static ipaddr_t ip_subnet_mask(ipaddr_t addr, ipif_t **, ip_stack_t *);
143 144 static void ip_wput_ioctl(queue_t *q, mblk_t *mp);
144 145 static void ipsq_flush(ill_t *ill);
145 146
146 147 static int ip_sioctl_token_tail(ipif_t *ipif, sin6_t *sin6, int addrlen,
147 148 queue_t *q, mblk_t *mp, boolean_t need_up);
148 149 static void ipsq_delete(ipsq_t *);
149 150
150 151 static ipif_t *ipif_allocate(ill_t *ill, int id, uint_t ire_type,
151 152 boolean_t initialize, boolean_t insert, int *errorp);
152 153 static ire_t **ipif_create_bcast_ires(ipif_t *ipif, ire_t **irep);
153 154 static void ipif_delete_bcast_ires(ipif_t *ipif);
154 155 static int ipif_add_ires_v4(ipif_t *, boolean_t);
155 156 static boolean_t ipif_comp_multi(ipif_t *old_ipif, ipif_t *new_ipif,
156 157 boolean_t isv6);
157 158 static int ipif_logical_down(ipif_t *ipif, queue_t *q, mblk_t *mp);
158 159 static void ipif_free(ipif_t *ipif);
159 160 static void ipif_free_tail(ipif_t *ipif);
160 161 static void ipif_set_default(ipif_t *ipif);
161 162 static int ipif_set_values(queue_t *q, mblk_t *mp,
162 163 char *interf_name, uint_t *ppa);
163 164 static int ipif_set_values_tail(ill_t *ill, ipif_t *ipif, mblk_t *mp,
164 165 queue_t *q);
165 166 static ipif_t *ipif_lookup_on_name(char *name, size_t namelen,
166 167 boolean_t do_alloc, boolean_t *exists, boolean_t isv6, zoneid_t zoneid,
167 168 ip_stack_t *);
168 169 static ipif_t *ipif_lookup_on_name_async(char *name, size_t namelen,
169 170 boolean_t isv6, zoneid_t zoneid, queue_t *q, mblk_t *mp, ipsq_func_t func,
170 171 int *error, ip_stack_t *);
171 172
172 173 static int ill_alloc_ppa(ill_if_t *, ill_t *);
173 174 static void ill_delete_interface_type(ill_if_t *);
174 175 static int ill_dl_up(ill_t *ill, ipif_t *ipif, mblk_t *mp, queue_t *q);
175 176 static void ill_dl_down(ill_t *ill);
176 177 static void ill_down(ill_t *ill);
177 178 static void ill_down_ipifs(ill_t *, boolean_t);
178 179 static void ill_free_mib(ill_t *ill);
179 180 static void ill_glist_delete(ill_t *);
180 181 static void ill_phyint_reinit(ill_t *ill);
181 182 static void ill_set_nce_router_flags(ill_t *, boolean_t);
182 183 static void ill_set_phys_addr_tail(ipsq_t *, queue_t *, mblk_t *, void *);
183 184 static void ill_replumb_tail(ipsq_t *, queue_t *, mblk_t *, void *);
184 185
185 186 static ip_v6intfid_func_t ip_ether_v6intfid, ip_ib_v6intfid;
186 187 static ip_v6intfid_func_t ip_ipv4_v6intfid, ip_ipv6_v6intfid;
187 188 static ip_v6intfid_func_t ip_ipmp_v6intfid, ip_nodef_v6intfid;
188 189 static ip_v6intfid_func_t ip_ipv4_v6destintfid, ip_ipv6_v6destintfid;
189 190 static ip_v4mapinfo_func_t ip_ether_v4_mapping;
190 191 static ip_v6mapinfo_func_t ip_ether_v6_mapping;
191 192 static ip_v4mapinfo_func_t ip_ib_v4_mapping;
192 193 static ip_v6mapinfo_func_t ip_ib_v6_mapping;
193 194 static ip_v4mapinfo_func_t ip_mbcast_mapping;
194 195 static void ip_cgtp_bcast_add(ire_t *, ip_stack_t *);
195 196 static void ip_cgtp_bcast_delete(ire_t *, ip_stack_t *);
196 197 static void phyint_free(phyint_t *);
197 198
198 199 static void ill_capability_dispatch(ill_t *, mblk_t *, dl_capability_sub_t *);
199 200 static void ill_capability_id_ack(ill_t *, mblk_t *, dl_capability_sub_t *);
200 201 static void ill_capability_vrrp_ack(ill_t *, mblk_t *, dl_capability_sub_t *);
201 202 static void ill_capability_hcksum_ack(ill_t *, mblk_t *, dl_capability_sub_t *);
202 203 static void ill_capability_hcksum_reset_fill(ill_t *, mblk_t *);
203 204 static void ill_capability_zerocopy_ack(ill_t *, mblk_t *,
204 205 dl_capability_sub_t *);
205 206 static void ill_capability_zerocopy_reset_fill(ill_t *, mblk_t *);
206 207 static void ill_capability_dld_reset_fill(ill_t *, mblk_t *);
207 208 static void ill_capability_dld_ack(ill_t *, mblk_t *,
208 209 dl_capability_sub_t *);
209 210 static void ill_capability_dld_enable(ill_t *);
210 211 static void ill_capability_ack_thr(void *);
211 212 static void ill_capability_lso_enable(ill_t *);
212 213
213 214 static ill_t *ill_prev_usesrc(ill_t *);
214 215 static int ill_relink_usesrc_ills(ill_t *, ill_t *, uint_t);
215 216 static void ill_disband_usesrc_group(ill_t *);
216 217 static void ip_sioctl_garp_reply(mblk_t *, ill_t *, void *, int);
217 218
218 219 #ifdef DEBUG
219 220 static void ill_trace_cleanup(const ill_t *);
220 221 static void ipif_trace_cleanup(const ipif_t *);
221 222 #endif
222 223
223 224 static void ill_dlpi_clear_deferred(ill_t *ill);
224 225
225 226 /*
226 227 * if we go over the memory footprint limit more than once in this msec
227 228 * interval, we'll start pruning aggressively.
228 229 */
229 230 int ip_min_frag_prune_time = 0;
230 231
231 232 static ipft_t ip_ioctl_ftbl[] = {
232 233 { IP_IOC_IRE_DELETE, ip_ire_delete, sizeof (ipid_t), 0 },
233 234 { IP_IOC_IRE_DELETE_NO_REPLY, ip_ire_delete, sizeof (ipid_t),
234 235 IPFT_F_NO_REPLY },
235 236 { IP_IOC_RTS_REQUEST, ip_rts_request, 0, IPFT_F_SELF_REPLY },
236 237 { 0 }
237 238 };
238 239
239 240 /* Simple ICMP IP Header Template */
240 241 static ipha_t icmp_ipha = {
241 242 IP_SIMPLE_HDR_VERSION, 0, 0, 0, 0, 0, IPPROTO_ICMP
242 243 };
243 244
244 245 static uchar_t ip_six_byte_all_ones[] = { 0xFF, 0xFF, 0xFF, 0xFF, 0xFF, 0xFF };
245 246
246 247 static ip_m_t ip_m_tbl[] = {
247 248 { DL_ETHER, IFT_ETHER, ETHERTYPE_IP, ETHERTYPE_IPV6,
248 249 ip_ether_v4_mapping, ip_ether_v6_mapping, ip_ether_v6intfid,
249 250 ip_nodef_v6intfid },
250 251 { DL_CSMACD, IFT_ISO88023, ETHERTYPE_IP, ETHERTYPE_IPV6,
251 252 ip_ether_v4_mapping, ip_ether_v6_mapping, ip_nodef_v6intfid,
252 253 ip_nodef_v6intfid },
253 254 { DL_TPB, IFT_ISO88024, ETHERTYPE_IP, ETHERTYPE_IPV6,
254 255 ip_ether_v4_mapping, ip_ether_v6_mapping, ip_nodef_v6intfid,
255 256 ip_nodef_v6intfid },
256 257 { DL_TPR, IFT_ISO88025, ETHERTYPE_IP, ETHERTYPE_IPV6,
257 258 ip_ether_v4_mapping, ip_ether_v6_mapping, ip_nodef_v6intfid,
258 259 ip_nodef_v6intfid },
259 260 { DL_FDDI, IFT_FDDI, ETHERTYPE_IP, ETHERTYPE_IPV6,
260 261 ip_ether_v4_mapping, ip_ether_v6_mapping, ip_ether_v6intfid,
261 262 ip_nodef_v6intfid },
262 263 { DL_IB, IFT_IB, ETHERTYPE_IP, ETHERTYPE_IPV6,
263 264 ip_ib_v4_mapping, ip_ib_v6_mapping, ip_ib_v6intfid,
264 265 ip_nodef_v6intfid },
265 266 { DL_IPV4, IFT_IPV4, IPPROTO_ENCAP, IPPROTO_IPV6,
266 267 ip_mbcast_mapping, ip_mbcast_mapping, ip_ipv4_v6intfid,
267 268 ip_ipv4_v6destintfid },
268 269 { DL_IPV6, IFT_IPV6, IPPROTO_ENCAP, IPPROTO_IPV6,
269 270 ip_mbcast_mapping, ip_mbcast_mapping, ip_ipv6_v6intfid,
270 271 ip_ipv6_v6destintfid },
271 272 { DL_6TO4, IFT_6TO4, IPPROTO_ENCAP, IPPROTO_IPV6,
272 273 ip_mbcast_mapping, ip_mbcast_mapping, ip_ipv4_v6intfid,
273 274 ip_nodef_v6intfid },
274 275 { SUNW_DL_VNI, IFT_OTHER, ETHERTYPE_IP, ETHERTYPE_IPV6,
275 276 NULL, NULL, ip_nodef_v6intfid, ip_nodef_v6intfid },
276 277 { SUNW_DL_IPMP, IFT_OTHER, ETHERTYPE_IP, ETHERTYPE_IPV6,
277 278 NULL, NULL, ip_ipmp_v6intfid, ip_nodef_v6intfid },
278 279 { DL_OTHER, IFT_OTHER, ETHERTYPE_IP, ETHERTYPE_IPV6,
279 280 ip_ether_v4_mapping, ip_ether_v6_mapping, ip_nodef_v6intfid,
280 281 ip_nodef_v6intfid }
281 282 };
282 283
283 284 static ill_t ill_null; /* Empty ILL for init. */
284 285 char ipif_loopback_name[] = "lo0";
285 286
286 287 /* These are used by all IP network modules. */
287 288 sin6_t sin6_null; /* Zero address for quick clears */
288 289 sin_t sin_null; /* Zero address for quick clears */
289 290
290 291 /* When set search for unused ipif_seqid */
291 292 static ipif_t ipif_zero;
292 293
293 294 /*
294 295 * ppa arena is created after these many
295 296 * interfaces have been plumbed.
296 297 */
297 298 uint_t ill_no_arena = 12; /* Setable in /etc/system */
298 299
299 300 /*
300 301 * Allocate per-interface mibs.
301 302 * Returns true if ok. False otherwise.
302 303 * ipsq may not yet be allocated (loopback case ).
303 304 */
304 305 static boolean_t
305 306 ill_allocate_mibs(ill_t *ill)
306 307 {
307 308 /* Already allocated? */
308 309 if (ill->ill_ip_mib != NULL) {
309 310 if (ill->ill_isv6)
310 311 ASSERT(ill->ill_icmp6_mib != NULL);
311 312 return (B_TRUE);
312 313 }
313 314
314 315 ill->ill_ip_mib = kmem_zalloc(sizeof (*ill->ill_ip_mib),
315 316 KM_NOSLEEP);
316 317 if (ill->ill_ip_mib == NULL) {
317 318 return (B_FALSE);
318 319 }
319 320
320 321 /* Setup static information */
321 322 SET_MIB(ill->ill_ip_mib->ipIfStatsEntrySize,
322 323 sizeof (mib2_ipIfStatsEntry_t));
323 324 if (ill->ill_isv6) {
324 325 ill->ill_ip_mib->ipIfStatsIPVersion = MIB2_INETADDRESSTYPE_ipv6;
325 326 SET_MIB(ill->ill_ip_mib->ipIfStatsAddrEntrySize,
326 327 sizeof (mib2_ipv6AddrEntry_t));
327 328 SET_MIB(ill->ill_ip_mib->ipIfStatsRouteEntrySize,
328 329 sizeof (mib2_ipv6RouteEntry_t));
329 330 SET_MIB(ill->ill_ip_mib->ipIfStatsNetToMediaEntrySize,
330 331 sizeof (mib2_ipv6NetToMediaEntry_t));
331 332 SET_MIB(ill->ill_ip_mib->ipIfStatsMemberEntrySize,
332 333 sizeof (ipv6_member_t));
333 334 SET_MIB(ill->ill_ip_mib->ipIfStatsGroupSourceEntrySize,
334 335 sizeof (ipv6_grpsrc_t));
335 336 } else {
336 337 ill->ill_ip_mib->ipIfStatsIPVersion = MIB2_INETADDRESSTYPE_ipv4;
337 338 SET_MIB(ill->ill_ip_mib->ipIfStatsAddrEntrySize,
338 339 sizeof (mib2_ipAddrEntry_t));
339 340 SET_MIB(ill->ill_ip_mib->ipIfStatsRouteEntrySize,
340 341 sizeof (mib2_ipRouteEntry_t));
341 342 SET_MIB(ill->ill_ip_mib->ipIfStatsNetToMediaEntrySize,
342 343 sizeof (mib2_ipNetToMediaEntry_t));
343 344 SET_MIB(ill->ill_ip_mib->ipIfStatsMemberEntrySize,
344 345 sizeof (ip_member_t));
345 346 SET_MIB(ill->ill_ip_mib->ipIfStatsGroupSourceEntrySize,
346 347 sizeof (ip_grpsrc_t));
347 348
348 349 /*
349 350 * For a v4 ill, we are done at this point, because per ill
350 351 * icmp mibs are only used for v6.
351 352 */
352 353 return (B_TRUE);
353 354 }
354 355
355 356 ill->ill_icmp6_mib = kmem_zalloc(sizeof (*ill->ill_icmp6_mib),
356 357 KM_NOSLEEP);
357 358 if (ill->ill_icmp6_mib == NULL) {
358 359 kmem_free(ill->ill_ip_mib, sizeof (*ill->ill_ip_mib));
359 360 ill->ill_ip_mib = NULL;
360 361 return (B_FALSE);
361 362 }
362 363 /* static icmp info */
363 364 ill->ill_icmp6_mib->ipv6IfIcmpEntrySize =
364 365 sizeof (mib2_ipv6IfIcmpEntry_t);
365 366 /*
366 367 * The ipIfStatsIfindex and ipv6IfIcmpIndex will be assigned later
367 368 * after the phyint merge occurs in ipif_set_values -> ill_glist_insert
368 369 * -> ill_phyint_reinit
369 370 */
370 371 return (B_TRUE);
371 372 }
372 373
373 374 /*
374 375 * Completely vaporize a lower level tap and all associated interfaces.
375 376 * ill_delete is called only out of ip_close when the device control
376 377 * stream is being closed.
377 378 */
378 379 void
379 380 ill_delete(ill_t *ill)
380 381 {
381 382 ipif_t *ipif;
382 383 ill_t *prev_ill;
383 384 ip_stack_t *ipst = ill->ill_ipst;
384 385
385 386 /*
386 387 * ill_delete may be forcibly entering the ipsq. The previous
387 388 * ioctl may not have completed and may need to be aborted.
388 389 * ipsq_flush takes care of it. If we don't need to enter the
389 390 * the ipsq forcibly, the 2nd invocation of ipsq_flush in
390 391 * ill_delete_tail is sufficient.
391 392 */
392 393 ipsq_flush(ill);
393 394
394 395 /*
395 396 * Nuke all interfaces. ipif_free will take down the interface,
396 397 * remove it from the list, and free the data structure.
397 398 * Walk down the ipif list and remove the logical interfaces
398 399 * first before removing the main ipif. We can't unplumb
399 400 * zeroth interface first in the case of IPv6 as update_conn_ill
400 401 * -> ip_ll_multireq de-references ill_ipif for checking
401 402 * POINTOPOINT.
402 403 *
403 404 * If ill_ipif was not properly initialized (i.e low on memory),
404 405 * then no interfaces to clean up. In this case just clean up the
405 406 * ill.
406 407 */
407 408 for (ipif = ill->ill_ipif; ipif != NULL; ipif = ipif->ipif_next)
408 409 ipif_free(ipif);
409 410
410 411 /*
411 412 * clean out all the nce_t entries that depend on this
412 413 * ill for the ill_phys_addr.
413 414 */
414 415 nce_flush(ill, B_TRUE);
415 416
416 417 /* Clean up msgs on pending upcalls for mrouted */
417 418 reset_mrt_ill(ill);
418 419
419 420 update_conn_ill(ill, ipst);
420 421
421 422 /*
422 423 * Remove multicast references added as a result of calls to
423 424 * ip_join_allmulti().
424 425 */
425 426 ip_purge_allmulti(ill);
426 427
427 428 /*
428 429 * If the ill being deleted is under IPMP, boot it out of the illgrp.
429 430 */
430 431 if (IS_UNDER_IPMP(ill))
431 432 ipmp_ill_leave_illgrp(ill);
432 433
433 434 /*
434 435 * ill_down will arrange to blow off any IRE's dependent on this
435 436 * ILL, and shut down fragmentation reassembly.
436 437 */
437 438 ill_down(ill);
438 439
439 440 /* Let SCTP know, so that it can remove this from its list. */
440 441 sctp_update_ill(ill, SCTP_ILL_REMOVE);
441 442
442 443 /*
443 444 * Walk all CONNs that can have a reference on an ire or nce for this
444 445 * ill (we actually walk all that now have stale references).
445 446 */
446 447 ipcl_walk(conn_ixa_cleanup, (void *)B_TRUE, ipst);
447 448
448 449 /* With IPv6 we have dce_ifindex. Cleanup for neatness */
449 450 if (ill->ill_isv6)
450 451 dce_cleanup(ill->ill_phyint->phyint_ifindex, ipst);
451 452
452 453 /*
453 454 * If an address on this ILL is being used as a source address then
454 455 * clear out the pointers in other ILLs that point to this ILL.
455 456 */
456 457 rw_enter(&ipst->ips_ill_g_usesrc_lock, RW_WRITER);
457 458 if (ill->ill_usesrc_grp_next != NULL) {
458 459 if (ill->ill_usesrc_ifindex == 0) { /* usesrc ILL ? */
459 460 ill_disband_usesrc_group(ill);
460 461 } else { /* consumer of the usesrc ILL */
461 462 prev_ill = ill_prev_usesrc(ill);
462 463 prev_ill->ill_usesrc_grp_next =
463 464 ill->ill_usesrc_grp_next;
464 465 }
465 466 }
466 467 rw_exit(&ipst->ips_ill_g_usesrc_lock);
467 468 }
468 469
469 470 static void
470 471 ipif_non_duplicate(ipif_t *ipif)
471 472 {
472 473 ill_t *ill = ipif->ipif_ill;
473 474 mutex_enter(&ill->ill_lock);
474 475 if (ipif->ipif_flags & IPIF_DUPLICATE) {
475 476 ipif->ipif_flags &= ~IPIF_DUPLICATE;
476 477 ASSERT(ill->ill_ipif_dup_count > 0);
477 478 ill->ill_ipif_dup_count--;
478 479 }
479 480 mutex_exit(&ill->ill_lock);
480 481 }
481 482
482 483 /*
483 484 * ill_delete_tail is called from ip_modclose after all references
484 485 * to the closing ill are gone. The wait is done in ip_modclose
485 486 */
486 487 void
487 488 ill_delete_tail(ill_t *ill)
488 489 {
489 490 mblk_t **mpp;
490 491 ipif_t *ipif;
491 492 ip_stack_t *ipst = ill->ill_ipst;
492 493
493 494 for (ipif = ill->ill_ipif; ipif != NULL; ipif = ipif->ipif_next) {
494 495 ipif_non_duplicate(ipif);
495 496 (void) ipif_down_tail(ipif);
496 497 }
497 498
498 499 ASSERT(ill->ill_ipif_dup_count == 0);
499 500
500 501 /*
501 502 * If polling capability is enabled (which signifies direct
502 503 * upcall into IP and driver has ill saved as a handle),
503 504 * we need to make sure that unbind has completed before we
504 505 * let the ill disappear and driver no longer has any reference
505 506 * to this ill.
506 507 */
507 508 mutex_enter(&ill->ill_lock);
508 509 while (ill->ill_state_flags & ILL_DL_UNBIND_IN_PROGRESS)
509 510 cv_wait(&ill->ill_cv, &ill->ill_lock);
510 511 mutex_exit(&ill->ill_lock);
511 512 ASSERT(!(ill->ill_capabilities &
512 513 (ILL_CAPAB_DLD | ILL_CAPAB_DLD_POLL | ILL_CAPAB_DLD_DIRECT)));
513 514
514 515 if (ill->ill_net_type != IRE_LOOPBACK)
515 516 qprocsoff(ill->ill_rq);
516 517
517 518 /*
518 519 * We do an ipsq_flush once again now. New messages could have
519 520 * landed up from below (M_ERROR or M_HANGUP). Similarly ioctls
520 521 * could also have landed up if an ioctl thread had looked up
521 522 * the ill before we set the ILL_CONDEMNED flag, but not yet
522 523 * enqueued the ioctl when we did the ipsq_flush last time.
523 524 */
524 525 ipsq_flush(ill);
525 526
526 527 /*
527 528 * Free capabilities.
528 529 */
529 530 if (ill->ill_hcksum_capab != NULL) {
530 531 kmem_free(ill->ill_hcksum_capab, sizeof (ill_hcksum_capab_t));
531 532 ill->ill_hcksum_capab = NULL;
532 533 }
533 534
534 535 if (ill->ill_zerocopy_capab != NULL) {
535 536 kmem_free(ill->ill_zerocopy_capab,
536 537 sizeof (ill_zerocopy_capab_t));
537 538 ill->ill_zerocopy_capab = NULL;
538 539 }
539 540
540 541 if (ill->ill_lso_capab != NULL) {
541 542 kmem_free(ill->ill_lso_capab, sizeof (ill_lso_capab_t));
542 543 ill->ill_lso_capab = NULL;
543 544 }
544 545
545 546 if (ill->ill_dld_capab != NULL) {
546 547 kmem_free(ill->ill_dld_capab, sizeof (ill_dld_capab_t));
547 548 ill->ill_dld_capab = NULL;
548 549 }
549 550
550 551 /* Clean up ill_allowed_ips* related state */
551 552 if (ill->ill_allowed_ips != NULL) {
552 553 ASSERT(ill->ill_allowed_ips_cnt > 0);
553 554 kmem_free(ill->ill_allowed_ips,
554 555 ill->ill_allowed_ips_cnt * sizeof (in6_addr_t));
555 556 ill->ill_allowed_ips = NULL;
556 557 ill->ill_allowed_ips_cnt = 0;
557 558 }
558 559
559 560 while (ill->ill_ipif != NULL)
560 561 ipif_free_tail(ill->ill_ipif);
561 562
562 563 /*
563 564 * We have removed all references to ilm from conn and the ones joined
564 565 * within the kernel.
565 566 *
566 567 * We don't walk conns, mrts and ires because
567 568 *
568 569 * 1) update_conn_ill and reset_mrt_ill cleans up conns and mrts.
569 570 * 2) ill_down ->ill_downi walks all the ires and cleans up
570 571 * ill references.
571 572 */
572 573
573 574 /*
574 575 * If this ill is an IPMP meta-interface, blow away the illgrp. This
575 576 * is safe to do because the illgrp has already been unlinked from the
576 577 * group by I_PUNLINK, and thus SIOCSLIFGROUPNAME cannot find it.
577 578 */
578 579 if (IS_IPMP(ill)) {
579 580 ipmp_illgrp_destroy(ill->ill_grp);
580 581 ill->ill_grp = NULL;
581 582 }
582 583
583 584 if (ill->ill_mphysaddr_list != NULL) {
584 585 multiphysaddr_t *mpa, *tmpa;
585 586
586 587 mpa = ill->ill_mphysaddr_list;
587 588 ill->ill_mphysaddr_list = NULL;
588 589 while (mpa) {
589 590 tmpa = mpa->mpa_next;
590 591 kmem_free(mpa, sizeof (*mpa));
591 592 mpa = tmpa;
592 593 }
593 594 }
594 595 /*
595 596 * Take us out of the list of ILLs. ill_glist_delete -> phyint_free
596 597 * could free the phyint. No more reference to the phyint after this
597 598 * point.
598 599 */
599 600 (void) ill_glist_delete(ill);
600 601
601 602 if (ill->ill_frag_ptr != NULL) {
602 603 uint_t count;
603 604
604 605 for (count = 0; count < ILL_FRAG_HASH_TBL_COUNT; count++) {
605 606 mutex_destroy(&ill->ill_frag_hash_tbl[count].ipfb_lock);
606 607 }
607 608 mi_free(ill->ill_frag_ptr);
608 609 ill->ill_frag_ptr = NULL;
609 610 ill->ill_frag_hash_tbl = NULL;
610 611 }
611 612
612 613 freemsg(ill->ill_nd_lla_mp);
613 614 /* Free all retained control messages. */
614 615 mpp = &ill->ill_first_mp_to_free;
615 616 do {
616 617 while (mpp[0]) {
617 618 mblk_t *mp;
618 619 mblk_t *mp1;
619 620
620 621 mp = mpp[0];
621 622 mpp[0] = mp->b_next;
622 623 for (mp1 = mp; mp1 != NULL; mp1 = mp1->b_cont) {
623 624 mp1->b_next = NULL;
624 625 mp1->b_prev = NULL;
625 626 }
626 627 freemsg(mp);
627 628 }
628 629 } while (mpp++ != &ill->ill_last_mp_to_free);
629 630
630 631 ill_free_mib(ill);
631 632
632 633 #ifdef DEBUG
633 634 ill_trace_cleanup(ill);
634 635 #endif
635 636
636 637 /* The default multicast interface might have changed */
637 638 ire_increment_multicast_generation(ipst, ill->ill_isv6);
638 639
639 640 /* Drop refcnt here */
640 641 netstack_rele(ill->ill_ipst->ips_netstack);
641 642 ill->ill_ipst = NULL;
642 643 }
643 644
644 645 static void
645 646 ill_free_mib(ill_t *ill)
646 647 {
647 648 ip_stack_t *ipst = ill->ill_ipst;
648 649
649 650 /*
650 651 * MIB statistics must not be lost, so when an interface
651 652 * goes away the counter values will be added to the global
652 653 * MIBs.
653 654 */
654 655 if (ill->ill_ip_mib != NULL) {
655 656 if (ill->ill_isv6) {
656 657 ip_mib2_add_ip_stats(&ipst->ips_ip6_mib,
657 658 ill->ill_ip_mib);
658 659 } else {
659 660 ip_mib2_add_ip_stats(&ipst->ips_ip_mib,
660 661 ill->ill_ip_mib);
661 662 }
662 663
663 664 kmem_free(ill->ill_ip_mib, sizeof (*ill->ill_ip_mib));
664 665 ill->ill_ip_mib = NULL;
665 666 }
666 667 if (ill->ill_icmp6_mib != NULL) {
667 668 ip_mib2_add_icmp6_stats(&ipst->ips_icmp6_mib,
668 669 ill->ill_icmp6_mib);
669 670 kmem_free(ill->ill_icmp6_mib, sizeof (*ill->ill_icmp6_mib));
670 671 ill->ill_icmp6_mib = NULL;
671 672 }
672 673 }
673 674
674 675 /*
675 676 * Concatenate together a physical address and a sap.
676 677 *
677 678 * Sap_lengths are interpreted as follows:
678 679 * sap_length == 0 ==> no sap
679 680 * sap_length > 0 ==> sap is at the head of the dlpi address
680 681 * sap_length < 0 ==> sap is at the tail of the dlpi address
681 682 */
682 683 static void
683 684 ill_dlur_copy_address(uchar_t *phys_src, uint_t phys_length,
684 685 t_scalar_t sap_src, t_scalar_t sap_length, uchar_t *dst)
685 686 {
686 687 uint16_t sap_addr = (uint16_t)sap_src;
687 688
688 689 if (sap_length == 0) {
689 690 if (phys_src == NULL)
690 691 bzero(dst, phys_length);
691 692 else
692 693 bcopy(phys_src, dst, phys_length);
693 694 } else if (sap_length < 0) {
694 695 if (phys_src == NULL)
695 696 bzero(dst, phys_length);
696 697 else
697 698 bcopy(phys_src, dst, phys_length);
698 699 bcopy(&sap_addr, (char *)dst + phys_length, sizeof (sap_addr));
699 700 } else {
700 701 bcopy(&sap_addr, dst, sizeof (sap_addr));
701 702 if (phys_src == NULL)
702 703 bzero((char *)dst + sap_length, phys_length);
703 704 else
704 705 bcopy(phys_src, (char *)dst + sap_length, phys_length);
705 706 }
706 707 }
707 708
708 709 /*
709 710 * Generate a dl_unitdata_req mblk for the device and address given.
710 711 * addr_length is the length of the physical portion of the address.
711 712 * If addr is NULL include an all zero address of the specified length.
712 713 * TRUE? In any case, addr_length is taken to be the entire length of the
713 714 * dlpi address, including the absolute value of sap_length.
714 715 */
715 716 mblk_t *
716 717 ill_dlur_gen(uchar_t *addr, uint_t addr_length, t_uscalar_t sap,
717 718 t_scalar_t sap_length)
718 719 {
719 720 dl_unitdata_req_t *dlur;
720 721 mblk_t *mp;
721 722 t_scalar_t abs_sap_length; /* absolute value */
722 723
723 724 abs_sap_length = ABS(sap_length);
724 725 mp = ip_dlpi_alloc(sizeof (*dlur) + addr_length + abs_sap_length,
725 726 DL_UNITDATA_REQ);
726 727 if (mp == NULL)
727 728 return (NULL);
728 729 dlur = (dl_unitdata_req_t *)mp->b_rptr;
729 730 /* HACK: accomodate incompatible DLPI drivers */
730 731 if (addr_length == 8)
731 732 addr_length = 6;
732 733 dlur->dl_dest_addr_length = addr_length + abs_sap_length;
733 734 dlur->dl_dest_addr_offset = sizeof (*dlur);
734 735 dlur->dl_priority.dl_min = 0;
735 736 dlur->dl_priority.dl_max = 0;
736 737 ill_dlur_copy_address(addr, addr_length, sap, sap_length,
737 738 (uchar_t *)&dlur[1]);
738 739 return (mp);
739 740 }
740 741
741 742 /*
742 743 * Add the pending mp to the list. There can be only 1 pending mp
743 744 * in the list. Any exclusive ioctl that needs to wait for a response
744 745 * from another module or driver needs to use this function to set
745 746 * the ipx_pending_mp to the ioctl mblk and wait for the response from
746 747 * the other module/driver. This is also used while waiting for the
747 748 * ipif/ill/ire refcnts to drop to zero in bringing down an ipif.
748 749 */
749 750 boolean_t
750 751 ipsq_pending_mp_add(conn_t *connp, ipif_t *ipif, queue_t *q, mblk_t *add_mp,
751 752 int waitfor)
752 753 {
753 754 ipxop_t *ipx = ipif->ipif_ill->ill_phyint->phyint_ipsq->ipsq_xop;
754 755
755 756 ASSERT(IAM_WRITER_IPIF(ipif));
756 757 ASSERT(MUTEX_HELD(&ipif->ipif_ill->ill_lock));
757 758 ASSERT((add_mp->b_next == NULL) && (add_mp->b_prev == NULL));
758 759 ASSERT(ipx->ipx_pending_mp == NULL);
759 760 /*
760 761 * The caller may be using a different ipif than the one passed into
761 762 * ipsq_current_start() (e.g., suppose an ioctl that came in on the V4
762 763 * ill needs to wait for the V6 ill to quiesce). So we can't ASSERT
763 764 * that `ipx_current_ipif == ipif'.
764 765 */
765 766 ASSERT(ipx->ipx_current_ipif != NULL);
766 767
767 768 /*
768 769 * M_IOCDATA from ioctls, M_ERROR/M_HANGUP/M_PROTO/M_PCPROTO from the
769 770 * driver.
770 771 */
771 772 ASSERT((DB_TYPE(add_mp) == M_IOCDATA) || (DB_TYPE(add_mp) == M_ERROR) ||
772 773 (DB_TYPE(add_mp) == M_HANGUP) || (DB_TYPE(add_mp) == M_PROTO) ||
773 774 (DB_TYPE(add_mp) == M_PCPROTO));
774 775
775 776 if (connp != NULL) {
776 777 ASSERT(MUTEX_HELD(&connp->conn_lock));
777 778 /*
778 779 * Return error if the conn has started closing. The conn
779 780 * could have finished cleaning up the pending mp list,
780 781 * If so we should not add another mp to the list negating
781 782 * the cleanup.
782 783 */
783 784 if (connp->conn_state_flags & CONN_CLOSING)
784 785 return (B_FALSE);
785 786 }
786 787 mutex_enter(&ipx->ipx_lock);
787 788 ipx->ipx_pending_ipif = ipif;
788 789 /*
789 790 * Note down the queue in b_queue. This will be returned by
790 791 * ipsq_pending_mp_get. Caller will then use these values to restart
791 792 * the processing
792 793 */
793 794 add_mp->b_next = NULL;
794 795 add_mp->b_queue = q;
795 796 ipx->ipx_pending_mp = add_mp;
796 797 ipx->ipx_waitfor = waitfor;
797 798 mutex_exit(&ipx->ipx_lock);
798 799
799 800 if (connp != NULL)
800 801 connp->conn_oper_pending_ill = ipif->ipif_ill;
801 802
802 803 return (B_TRUE);
803 804 }
804 805
805 806 /*
806 807 * Retrieve the ipx_pending_mp and return it. There can be only 1 mp
807 808 * queued in the list.
808 809 */
809 810 mblk_t *
810 811 ipsq_pending_mp_get(ipsq_t *ipsq, conn_t **connpp)
811 812 {
812 813 mblk_t *curr = NULL;
813 814 ipxop_t *ipx = ipsq->ipsq_xop;
814 815
815 816 *connpp = NULL;
816 817 mutex_enter(&ipx->ipx_lock);
817 818 if (ipx->ipx_pending_mp == NULL) {
818 819 mutex_exit(&ipx->ipx_lock);
819 820 return (NULL);
820 821 }
821 822
822 823 /* There can be only 1 such excl message */
823 824 curr = ipx->ipx_pending_mp;
824 825 ASSERT(curr->b_next == NULL);
825 826 ipx->ipx_pending_ipif = NULL;
826 827 ipx->ipx_pending_mp = NULL;
827 828 ipx->ipx_waitfor = 0;
828 829 mutex_exit(&ipx->ipx_lock);
829 830
830 831 if (CONN_Q(curr->b_queue)) {
831 832 /*
832 833 * This mp did a refhold on the conn, at the start of the ioctl.
833 834 * So we can safely return a pointer to the conn to the caller.
834 835 */
835 836 *connpp = Q_TO_CONN(curr->b_queue);
836 837 } else {
837 838 *connpp = NULL;
838 839 }
839 840 curr->b_next = NULL;
840 841 curr->b_prev = NULL;
841 842 return (curr);
842 843 }
843 844
844 845 /*
845 846 * Cleanup the ioctl mp queued in ipx_pending_mp
846 847 * - Called in the ill_delete path
847 848 * - Called in the M_ERROR or M_HANGUP path on the ill.
848 849 * - Called in the conn close path.
849 850 *
850 851 * Returns success on finding the pending mblk associated with the ioctl or
851 852 * exclusive operation in progress, failure otherwise.
852 853 */
853 854 boolean_t
854 855 ipsq_pending_mp_cleanup(ill_t *ill, conn_t *connp)
855 856 {
856 857 mblk_t *mp;
857 858 ipxop_t *ipx;
858 859 queue_t *q;
859 860 ipif_t *ipif;
860 861 int cmd;
861 862
862 863 ASSERT(IAM_WRITER_ILL(ill));
863 864 ipx = ill->ill_phyint->phyint_ipsq->ipsq_xop;
864 865
865 866 mutex_enter(&ipx->ipx_lock);
866 867 mp = ipx->ipx_pending_mp;
867 868 if (connp != NULL) {
868 869 if (mp == NULL || mp->b_queue != CONNP_TO_WQ(connp)) {
869 870 /*
870 871 * Nothing to clean since the conn that is closing
871 872 * does not have a matching pending mblk in
872 873 * ipx_pending_mp.
873 874 */
874 875 mutex_exit(&ipx->ipx_lock);
875 876 return (B_FALSE);
876 877 }
877 878 } else {
878 879 /*
879 880 * A non-zero ill_error signifies we are called in the
880 881 * M_ERROR or M_HANGUP path and we need to unconditionally
881 882 * abort any current ioctl and do the corresponding cleanup.
882 883 * A zero ill_error means we are in the ill_delete path and
883 884 * we do the cleanup only if there is a pending mp.
884 885 */
885 886 if (mp == NULL && ill->ill_error == 0) {
886 887 mutex_exit(&ipx->ipx_lock);
887 888 return (B_FALSE);
888 889 }
889 890 }
890 891
891 892 /* Now remove from the ipx_pending_mp */
892 893 ipx->ipx_pending_mp = NULL;
893 894 ipif = ipx->ipx_pending_ipif;
894 895 ipx->ipx_pending_ipif = NULL;
895 896 ipx->ipx_waitfor = 0;
896 897 ipx->ipx_current_ipif = NULL;
897 898 cmd = ipx->ipx_current_ioctl;
898 899 ipx->ipx_current_ioctl = 0;
899 900 ipx->ipx_current_done = B_TRUE;
900 901 mutex_exit(&ipx->ipx_lock);
901 902
902 903 if (mp == NULL)
903 904 return (B_FALSE);
904 905
905 906 q = mp->b_queue;
906 907 mp->b_next = NULL;
907 908 mp->b_prev = NULL;
908 909 mp->b_queue = NULL;
909 910
910 911 if (DB_TYPE(mp) == M_IOCTL || DB_TYPE(mp) == M_IOCDATA) {
911 912 DTRACE_PROBE4(ipif__ioctl,
912 913 char *, "ipsq_pending_mp_cleanup",
913 914 int, cmd, ill_t *, ipif == NULL ? NULL : ipif->ipif_ill,
914 915 ipif_t *, ipif);
915 916 if (connp == NULL) {
916 917 ip_ioctl_finish(q, mp, ENXIO, NO_COPYOUT, NULL);
917 918 } else {
918 919 ip_ioctl_finish(q, mp, ENXIO, CONN_CLOSE, NULL);
919 920 mutex_enter(&ipif->ipif_ill->ill_lock);
920 921 ipif->ipif_state_flags &= ~IPIF_CHANGING;
921 922 mutex_exit(&ipif->ipif_ill->ill_lock);
922 923 }
923 924 } else {
924 925 inet_freemsg(mp);
925 926 }
926 927 return (B_TRUE);
927 928 }
928 929
929 930 /*
930 931 * Called in the conn close path and ill delete path
931 932 */
932 933 static void
933 934 ipsq_xopq_mp_cleanup(ill_t *ill, conn_t *connp)
934 935 {
935 936 ipsq_t *ipsq;
936 937 mblk_t *prev;
937 938 mblk_t *curr;
938 939 mblk_t *next;
939 940 queue_t *wq, *rq = NULL;
940 941 mblk_t *tmp_list = NULL;
941 942
942 943 ASSERT(IAM_WRITER_ILL(ill));
943 944 if (connp != NULL)
944 945 wq = CONNP_TO_WQ(connp);
945 946 else
946 947 wq = ill->ill_wq;
947 948
948 949 /*
949 950 * In the case of lo0 being unplumbed, ill_wq will be NULL. Guard
950 951 * against this here.
951 952 */
952 953 if (wq != NULL)
953 954 rq = RD(wq);
954 955
955 956 ipsq = ill->ill_phyint->phyint_ipsq;
956 957 /*
957 958 * Cleanup the ioctl mp's queued in ipsq_xopq_pending_mp if any.
958 959 * In the case of ioctl from a conn, there can be only 1 mp
959 960 * queued on the ipsq. If an ill is being unplumbed flush all
960 961 * the messages.
961 962 */
962 963 mutex_enter(&ipsq->ipsq_lock);
963 964 for (prev = NULL, curr = ipsq->ipsq_xopq_mphead; curr != NULL;
964 965 curr = next) {
965 966 next = curr->b_next;
966 967 if (connp == NULL ||
967 968 (curr->b_queue == wq || curr->b_queue == rq)) {
968 969 /* Unlink the mblk from the pending mp list */
969 970 if (prev != NULL) {
970 971 prev->b_next = curr->b_next;
971 972 } else {
972 973 ASSERT(ipsq->ipsq_xopq_mphead == curr);
973 974 ipsq->ipsq_xopq_mphead = curr->b_next;
974 975 }
975 976 if (ipsq->ipsq_xopq_mptail == curr)
976 977 ipsq->ipsq_xopq_mptail = prev;
977 978 /*
978 979 * Create a temporary list and release the ipsq lock
979 980 * New elements are added to the head of the tmp_list
980 981 */
981 982 curr->b_next = tmp_list;
982 983 tmp_list = curr;
983 984 } else {
984 985 prev = curr;
985 986 }
986 987 }
987 988 mutex_exit(&ipsq->ipsq_lock);
988 989
989 990 while (tmp_list != NULL) {
990 991 curr = tmp_list;
991 992 tmp_list = curr->b_next;
992 993 curr->b_next = NULL;
993 994 curr->b_prev = NULL;
994 995 wq = curr->b_queue;
995 996 curr->b_queue = NULL;
996 997 if (DB_TYPE(curr) == M_IOCTL || DB_TYPE(curr) == M_IOCDATA) {
997 998 DTRACE_PROBE4(ipif__ioctl,
998 999 char *, "ipsq_xopq_mp_cleanup",
999 1000 int, 0, ill_t *, NULL, ipif_t *, NULL);
1000 1001 ip_ioctl_finish(wq, curr, ENXIO, connp != NULL ?
1001 1002 CONN_CLOSE : NO_COPYOUT, NULL);
1002 1003 } else {
1003 1004 /*
1004 1005 * IP-MT XXX In the case of TLI/XTI bind / optmgmt
1005 1006 * this can't be just inet_freemsg. we have to
1006 1007 * restart it otherwise the thread will be stuck.
1007 1008 */
1008 1009 inet_freemsg(curr);
1009 1010 }
1010 1011 }
1011 1012 }
1012 1013
1013 1014 /*
1014 1015 * This conn has started closing. Cleanup any pending ioctl from this conn.
1015 1016 * STREAMS ensures that there can be at most 1 active ioctl on a stream.
1016 1017 */
1017 1018 void
1018 1019 conn_ioctl_cleanup(conn_t *connp)
1019 1020 {
1020 1021 ipsq_t *ipsq;
1021 1022 ill_t *ill;
1022 1023 boolean_t refheld;
1023 1024
1024 1025 /*
1025 1026 * Check for a queued ioctl. If the ioctl has not yet started, the mp
1026 1027 * is pending in the list headed by ipsq_xopq_head. If the ioctl has
1027 1028 * started the mp could be present in ipx_pending_mp. Note that if
1028 1029 * conn_oper_pending_ill is NULL, the ioctl may still be in flight and
1029 1030 * not yet queued anywhere. In this case, the conn close code will wait
1030 1031 * until the conn_ref is dropped. If the stream was a tcp stream, then
1031 1032 * tcp_close will wait first until all ioctls have completed for this
1032 1033 * conn.
1033 1034 */
1034 1035 mutex_enter(&connp->conn_lock);
1035 1036 ill = connp->conn_oper_pending_ill;
1036 1037 if (ill == NULL) {
1037 1038 mutex_exit(&connp->conn_lock);
1038 1039 return;
1039 1040 }
1040 1041
1041 1042 /*
1042 1043 * We may not be able to refhold the ill if the ill/ipif
1043 1044 * is changing. But we need to make sure that the ill will
1044 1045 * not vanish. So we just bump up the ill_waiter count.
1045 1046 */
1046 1047 refheld = ill_waiter_inc(ill);
1047 1048 mutex_exit(&connp->conn_lock);
1048 1049 if (refheld) {
1049 1050 if (ipsq_enter(ill, B_TRUE, NEW_OP)) {
1050 1051 ill_waiter_dcr(ill);
1051 1052 /*
1052 1053 * Check whether this ioctl has started and is
1053 1054 * pending. If it is not found there then check
1054 1055 * whether this ioctl has not even started and is in
1055 1056 * the ipsq_xopq list.
1056 1057 */
1057 1058 if (!ipsq_pending_mp_cleanup(ill, connp))
1058 1059 ipsq_xopq_mp_cleanup(ill, connp);
1059 1060 ipsq = ill->ill_phyint->phyint_ipsq;
1060 1061 ipsq_exit(ipsq);
1061 1062 return;
1062 1063 }
1063 1064 }
1064 1065
1065 1066 /*
1066 1067 * The ill is also closing and we could not bump up the
1067 1068 * ill_waiter_count or we could not enter the ipsq. Leave
1068 1069 * the cleanup to ill_delete
1069 1070 */
1070 1071 mutex_enter(&connp->conn_lock);
1071 1072 while (connp->conn_oper_pending_ill != NULL)
1072 1073 cv_wait(&connp->conn_refcv, &connp->conn_lock);
1073 1074 mutex_exit(&connp->conn_lock);
1074 1075 if (refheld)
1075 1076 ill_waiter_dcr(ill);
1076 1077 }
1077 1078
1078 1079 /*
1079 1080 * ipcl_walk function for cleaning up conn_*_ill fields.
1080 1081 * Note that we leave ixa_multicast_ifindex, conn_incoming_ifindex, and
1081 1082 * conn_bound_if in place. We prefer dropping
1082 1083 * packets instead of sending them out the wrong interface, or accepting
1083 1084 * packets from the wrong ifindex.
1084 1085 */
1085 1086 static void
1086 1087 conn_cleanup_ill(conn_t *connp, caddr_t arg)
1087 1088 {
1088 1089 ill_t *ill = (ill_t *)arg;
1089 1090
1090 1091 mutex_enter(&connp->conn_lock);
1091 1092 if (connp->conn_dhcpinit_ill == ill) {
1092 1093 connp->conn_dhcpinit_ill = NULL;
1093 1094 ASSERT(ill->ill_dhcpinit != 0);
1094 1095 atomic_dec_32(&ill->ill_dhcpinit);
1095 1096 ill_set_inputfn(ill);
1096 1097 }
1097 1098 mutex_exit(&connp->conn_lock);
1098 1099 }
1099 1100
1100 1101 static int
1101 1102 ill_down_ipifs_tail(ill_t *ill)
1102 1103 {
1103 1104 ipif_t *ipif;
1104 1105 int err;
1105 1106
1106 1107 ASSERT(IAM_WRITER_ILL(ill));
1107 1108 for (ipif = ill->ill_ipif; ipif != NULL; ipif = ipif->ipif_next) {
1108 1109 ipif_non_duplicate(ipif);
1109 1110 /*
1110 1111 * ipif_down_tail will call arp_ll_down on the last ipif
1111 1112 * and typically return EINPROGRESS when the DL_UNBIND is sent.
1112 1113 */
1113 1114 if ((err = ipif_down_tail(ipif)) != 0)
1114 1115 return (err);
1115 1116 }
1116 1117 return (0);
1117 1118 }
1118 1119
1119 1120 /* ARGSUSED */
1120 1121 void
1121 1122 ipif_all_down_tail(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *dummy_arg)
1122 1123 {
1123 1124 ASSERT(IAM_WRITER_IPSQ(ipsq));
1124 1125 (void) ill_down_ipifs_tail(q->q_ptr);
1125 1126 freemsg(mp);
1126 1127 ipsq_current_finish(ipsq);
1127 1128 }
1128 1129
1129 1130 /*
1130 1131 * ill_down_start is called when we want to down this ill and bring it up again
1131 1132 * It is called when we receive an M_ERROR / M_HANGUP. In this case we shut down
1132 1133 * all interfaces, but don't tear down any plumbing.
1133 1134 */
1134 1135 boolean_t
1135 1136 ill_down_start(queue_t *q, mblk_t *mp)
1136 1137 {
1137 1138 ill_t *ill = q->q_ptr;
1138 1139 ipif_t *ipif;
1139 1140
1140 1141 ASSERT(IAM_WRITER_ILL(ill));
1141 1142 /*
1142 1143 * It is possible that some ioctl is already in progress while we
1143 1144 * received the M_ERROR / M_HANGUP in which case, we need to abort
1144 1145 * the ioctl. ill_down_start() is being processed as CUR_OP rather
1145 1146 * than as NEW_OP since the cause of the M_ERROR / M_HANGUP may prevent
1146 1147 * the in progress ioctl from ever completing.
1147 1148 *
1148 1149 * The thread that started the ioctl (if any) must have returned,
1149 1150 * since we are now executing as writer. After the 2 calls below,
1150 1151 * the state of the ipsq and the ill would reflect no trace of any
1151 1152 * pending operation. Subsequently if there is any response to the
1152 1153 * original ioctl from the driver, it would be discarded as an
1153 1154 * unsolicited message from the driver.
1154 1155 */
1155 1156 (void) ipsq_pending_mp_cleanup(ill, NULL);
1156 1157 ill_dlpi_clear_deferred(ill);
1157 1158
1158 1159 for (ipif = ill->ill_ipif; ipif != NULL; ipif = ipif->ipif_next)
1159 1160 (void) ipif_down(ipif, NULL, NULL);
1160 1161
1161 1162 ill_down(ill);
1162 1163
1163 1164 /*
1164 1165 * Walk all CONNs that can have a reference on an ire or nce for this
1165 1166 * ill (we actually walk all that now have stale references).
1166 1167 */
1167 1168 ipcl_walk(conn_ixa_cleanup, (void *)B_TRUE, ill->ill_ipst);
1168 1169
1169 1170 /* With IPv6 we have dce_ifindex. Cleanup for neatness */
1170 1171 if (ill->ill_isv6)
1171 1172 dce_cleanup(ill->ill_phyint->phyint_ifindex, ill->ill_ipst);
1172 1173
1173 1174 ipsq_current_start(ill->ill_phyint->phyint_ipsq, ill->ill_ipif, 0);
1174 1175
1175 1176 /*
1176 1177 * Atomically test and add the pending mp if references are active.
1177 1178 */
1178 1179 mutex_enter(&ill->ill_lock);
1179 1180 if (!ill_is_quiescent(ill)) {
1180 1181 /* call cannot fail since `conn_t *' argument is NULL */
1181 1182 (void) ipsq_pending_mp_add(NULL, ill->ill_ipif, ill->ill_rq,
1182 1183 mp, ILL_DOWN);
1183 1184 mutex_exit(&ill->ill_lock);
1184 1185 return (B_FALSE);
1185 1186 }
1186 1187 mutex_exit(&ill->ill_lock);
1187 1188 return (B_TRUE);
1188 1189 }
1189 1190
1190 1191 static void
1191 1192 ill_down(ill_t *ill)
1192 1193 {
1193 1194 mblk_t *mp;
1194 1195 ip_stack_t *ipst = ill->ill_ipst;
1195 1196
1196 1197 /*
1197 1198 * Blow off any IREs dependent on this ILL.
1198 1199 * The caller needs to handle conn_ixa_cleanup
1199 1200 */
1200 1201 ill_delete_ires(ill);
1201 1202
1202 1203 ire_walk_ill(0, 0, ill_downi, ill, ill);
1203 1204
1204 1205 /* Remove any conn_*_ill depending on this ill */
1205 1206 ipcl_walk(conn_cleanup_ill, (caddr_t)ill, ipst);
1206 1207
1207 1208 /*
1208 1209 * Free state for additional IREs.
1209 1210 */
1210 1211 mutex_enter(&ill->ill_saved_ire_lock);
1211 1212 mp = ill->ill_saved_ire_mp;
1212 1213 ill->ill_saved_ire_mp = NULL;
1213 1214 ill->ill_saved_ire_cnt = 0;
1214 1215 mutex_exit(&ill->ill_saved_ire_lock);
1215 1216 freemsg(mp);
1216 1217 }
1217 1218
1218 1219 /*
1219 1220 * ire_walk routine used to delete every IRE that depends on
1220 1221 * 'ill'. (Always called as writer, and may only be called from ire_walk.)
1221 1222 *
1222 1223 * Note: since the routes added by the kernel are deleted separately,
1223 1224 * this will only be 1) IRE_IF_CLONE and 2) manually added IRE_INTERFACE.
1224 1225 *
1225 1226 * We also remove references on ire_nce_cache entries that refer to the ill.
1226 1227 */
1227 1228 void
1228 1229 ill_downi(ire_t *ire, char *ill_arg)
1229 1230 {
1230 1231 ill_t *ill = (ill_t *)ill_arg;
1231 1232 nce_t *nce;
1232 1233
1233 1234 mutex_enter(&ire->ire_lock);
1234 1235 nce = ire->ire_nce_cache;
1235 1236 if (nce != NULL && nce->nce_ill == ill)
1236 1237 ire->ire_nce_cache = NULL;
1237 1238 else
1238 1239 nce = NULL;
1239 1240 mutex_exit(&ire->ire_lock);
1240 1241 if (nce != NULL)
1241 1242 nce_refrele(nce);
1242 1243 if (ire->ire_ill == ill) {
1243 1244 /*
1244 1245 * The existing interface binding for ire must be
1245 1246 * deleted before trying to bind the route to another
1246 1247 * interface. However, since we are using the contents of the
1247 1248 * ire after ire_delete, the caller has to ensure that
1248 1249 * CONDEMNED (deleted) ire's are not removed from the list
1249 1250 * when ire_delete() returns. Currently ill_downi() is
1250 1251 * only called as part of ire_walk*() routines, so that
1251 1252 * the irb_refhold() done by ire_walk*() will ensure that
1252 1253 * ire_delete() does not lead to ire_inactive().
1253 1254 */
1254 1255 ASSERT(ire->ire_bucket->irb_refcnt > 0);
1255 1256 ire_delete(ire);
1256 1257 if (ire->ire_unbound)
1257 1258 ire_rebind(ire);
1258 1259 }
1259 1260 }
1260 1261
1261 1262 /* Remove IRE_IF_CLONE on this ill */
1262 1263 void
1263 1264 ill_downi_if_clone(ire_t *ire, char *ill_arg)
1264 1265 {
1265 1266 ill_t *ill = (ill_t *)ill_arg;
1266 1267
1267 1268 ASSERT(ire->ire_type & IRE_IF_CLONE);
1268 1269 if (ire->ire_ill == ill)
1269 1270 ire_delete(ire);
1270 1271 }
1271 1272
1272 1273 /* Consume an M_IOCACK of the fastpath probe. */
1273 1274 void
1274 1275 ill_fastpath_ack(ill_t *ill, mblk_t *mp)
1275 1276 {
1276 1277 mblk_t *mp1 = mp;
1277 1278
1278 1279 /*
1279 1280 * If this was the first attempt turn on the fastpath probing.
1280 1281 */
1281 1282 mutex_enter(&ill->ill_lock);
1282 1283 if (ill->ill_dlpi_fastpath_state == IDS_INPROGRESS)
1283 1284 ill->ill_dlpi_fastpath_state = IDS_OK;
1284 1285 mutex_exit(&ill->ill_lock);
1285 1286
1286 1287 /* Free the M_IOCACK mblk, hold on to the data */
1287 1288 mp = mp->b_cont;
1288 1289 freeb(mp1);
1289 1290 if (mp == NULL)
1290 1291 return;
1291 1292 if (mp->b_cont != NULL)
1292 1293 nce_fastpath_update(ill, mp);
1293 1294 else
1294 1295 ip0dbg(("ill_fastpath_ack: no b_cont\n"));
1295 1296 freemsg(mp);
1296 1297 }
1297 1298
1298 1299 /*
1299 1300 * Throw an M_IOCTL message downstream asking "do you know fastpath?"
1300 1301 * The data portion of the request is a dl_unitdata_req_t template for
1301 1302 * what we would send downstream in the absence of a fastpath confirmation.
1302 1303 */
1303 1304 int
1304 1305 ill_fastpath_probe(ill_t *ill, mblk_t *dlur_mp)
1305 1306 {
1306 1307 struct iocblk *ioc;
1307 1308 mblk_t *mp;
1308 1309
1309 1310 if (dlur_mp == NULL)
1310 1311 return (EINVAL);
1311 1312
1312 1313 mutex_enter(&ill->ill_lock);
1313 1314 switch (ill->ill_dlpi_fastpath_state) {
1314 1315 case IDS_FAILED:
1315 1316 /*
1316 1317 * Driver NAKed the first fastpath ioctl - assume it doesn't
1317 1318 * support it.
1318 1319 */
1319 1320 mutex_exit(&ill->ill_lock);
1320 1321 return (ENOTSUP);
1321 1322 case IDS_UNKNOWN:
1322 1323 /* This is the first probe */
1323 1324 ill->ill_dlpi_fastpath_state = IDS_INPROGRESS;
1324 1325 break;
1325 1326 default:
1326 1327 break;
1327 1328 }
1328 1329 mutex_exit(&ill->ill_lock);
1329 1330
1330 1331 if ((mp = mkiocb(DL_IOC_HDR_INFO)) == NULL)
1331 1332 return (EAGAIN);
1332 1333
1333 1334 mp->b_cont = copyb(dlur_mp);
1334 1335 if (mp->b_cont == NULL) {
1335 1336 freeb(mp);
1336 1337 return (EAGAIN);
1337 1338 }
1338 1339
1339 1340 ioc = (struct iocblk *)mp->b_rptr;
1340 1341 ioc->ioc_count = msgdsize(mp->b_cont);
1341 1342
1342 1343 DTRACE_PROBE3(ill__dlpi, char *, "ill_fastpath_probe",
1343 1344 char *, "DL_IOC_HDR_INFO", ill_t *, ill);
1344 1345 putnext(ill->ill_wq, mp);
1345 1346 return (0);
1346 1347 }
1347 1348
1348 1349 void
1349 1350 ill_capability_probe(ill_t *ill)
1350 1351 {
1351 1352 mblk_t *mp;
1352 1353
1353 1354 ASSERT(IAM_WRITER_ILL(ill));
1354 1355
1355 1356 if (ill->ill_dlpi_capab_state != IDCS_UNKNOWN &&
1356 1357 ill->ill_dlpi_capab_state != IDCS_FAILED)
1357 1358 return;
1358 1359
1359 1360 /*
1360 1361 * We are starting a new cycle of capability negotiation.
1361 1362 * Free up the capab reset messages of any previous incarnation.
1362 1363 * We will do a fresh allocation when we get the response to our probe
1363 1364 */
1364 1365 if (ill->ill_capab_reset_mp != NULL) {
1365 1366 freemsg(ill->ill_capab_reset_mp);
1366 1367 ill->ill_capab_reset_mp = NULL;
1367 1368 }
1368 1369
1369 1370 ip1dbg(("ill_capability_probe: starting capability negotiation\n"));
1370 1371
1371 1372 mp = ip_dlpi_alloc(sizeof (dl_capability_req_t), DL_CAPABILITY_REQ);
1372 1373 if (mp == NULL)
1373 1374 return;
1374 1375
1375 1376 ill_capability_send(ill, mp);
1376 1377 ill->ill_dlpi_capab_state = IDCS_PROBE_SENT;
1377 1378 }
1378 1379
1379 1380 void
1380 1381 ill_capability_reset(ill_t *ill, boolean_t reneg)
1381 1382 {
1382 1383 ASSERT(IAM_WRITER_ILL(ill));
1383 1384
1384 1385 if (ill->ill_dlpi_capab_state != IDCS_OK)
1385 1386 return;
1386 1387
1387 1388 ill->ill_dlpi_capab_state = reneg ? IDCS_RENEG : IDCS_RESET_SENT;
1388 1389
1389 1390 ill_capability_send(ill, ill->ill_capab_reset_mp);
1390 1391 ill->ill_capab_reset_mp = NULL;
1391 1392 /*
1392 1393 * We turn off all capabilities except those pertaining to
1393 1394 * direct function call capabilities viz. ILL_CAPAB_DLD*
1394 1395 * which will be turned off by the corresponding reset functions.
1395 1396 */
1396 1397 ill->ill_capabilities &= ~(ILL_CAPAB_HCKSUM | ILL_CAPAB_ZEROCOPY);
1397 1398 }
1398 1399
1399 1400 static void
1400 1401 ill_capability_reset_alloc(ill_t *ill)
1401 1402 {
1402 1403 mblk_t *mp;
1403 1404 size_t size = 0;
1404 1405 int err;
1405 1406 dl_capability_req_t *capb;
1406 1407
1407 1408 ASSERT(IAM_WRITER_ILL(ill));
1408 1409 ASSERT(ill->ill_capab_reset_mp == NULL);
1409 1410
1410 1411 if (ILL_HCKSUM_CAPABLE(ill)) {
1411 1412 size += sizeof (dl_capability_sub_t) +
1412 1413 sizeof (dl_capab_hcksum_t);
1413 1414 }
1414 1415
1415 1416 if (ill->ill_capabilities & ILL_CAPAB_ZEROCOPY) {
1416 1417 size += sizeof (dl_capability_sub_t) +
1417 1418 sizeof (dl_capab_zerocopy_t);
1418 1419 }
1419 1420
1420 1421 if (ill->ill_capabilities & ILL_CAPAB_DLD) {
1421 1422 size += sizeof (dl_capability_sub_t) +
1422 1423 sizeof (dl_capab_dld_t);
1423 1424 }
1424 1425
1425 1426 mp = allocb_wait(size + sizeof (dl_capability_req_t), BPRI_MED,
1426 1427 STR_NOSIG, &err);
1427 1428
1428 1429 mp->b_datap->db_type = M_PROTO;
1429 1430 bzero(mp->b_rptr, size + sizeof (dl_capability_req_t));
1430 1431
1431 1432 capb = (dl_capability_req_t *)mp->b_rptr;
1432 1433 capb->dl_primitive = DL_CAPABILITY_REQ;
1433 1434 capb->dl_sub_offset = sizeof (dl_capability_req_t);
1434 1435 capb->dl_sub_length = size;
1435 1436
1436 1437 mp->b_wptr += sizeof (dl_capability_req_t);
1437 1438
1438 1439 /*
1439 1440 * Each handler fills in the corresponding dl_capability_sub_t
1440 1441 * inside the mblk,
1441 1442 */
1442 1443 ill_capability_hcksum_reset_fill(ill, mp);
1443 1444 ill_capability_zerocopy_reset_fill(ill, mp);
1444 1445 ill_capability_dld_reset_fill(ill, mp);
1445 1446
1446 1447 ill->ill_capab_reset_mp = mp;
1447 1448 }
1448 1449
1449 1450 static void
1450 1451 ill_capability_id_ack(ill_t *ill, mblk_t *mp, dl_capability_sub_t *outers)
1451 1452 {
1452 1453 dl_capab_id_t *id_ic;
1453 1454 uint_t sub_dl_cap = outers->dl_cap;
1454 1455 dl_capability_sub_t *inners;
1455 1456 uint8_t *capend;
1456 1457
1457 1458 ASSERT(sub_dl_cap == DL_CAPAB_ID_WRAPPER);
1458 1459
1459 1460 /*
1460 1461 * Note: range checks here are not absolutely sufficient to
1461 1462 * make us robust against malformed messages sent by drivers;
1462 1463 * this is in keeping with the rest of IP's dlpi handling.
1463 1464 * (Remember, it's coming from something else in the kernel
1464 1465 * address space)
1465 1466 */
1466 1467
1467 1468 capend = (uint8_t *)(outers + 1) + outers->dl_length;
1468 1469 if (capend > mp->b_wptr) {
1469 1470 cmn_err(CE_WARN, "ill_capability_id_ack: "
1470 1471 "malformed sub-capability too long for mblk");
1471 1472 return;
1472 1473 }
1473 1474
1474 1475 id_ic = (dl_capab_id_t *)(outers + 1);
1475 1476
1476 1477 if (outers->dl_length < sizeof (*id_ic) ||
1477 1478 (inners = &id_ic->id_subcap,
1478 1479 inners->dl_length > (outers->dl_length - sizeof (*inners)))) {
1479 1480 cmn_err(CE_WARN, "ill_capability_id_ack: malformed "
1480 1481 "encapsulated capab type %d too long for mblk",
1481 1482 inners->dl_cap);
1482 1483 return;
1483 1484 }
1484 1485
1485 1486 if (!dlcapabcheckqid(&id_ic->id_mid, ill->ill_lmod_rq)) {
1486 1487 ip1dbg(("ill_capability_id_ack: mid token for capab type %d "
1487 1488 "isn't as expected; pass-thru module(s) detected, "
1488 1489 "discarding capability\n", inners->dl_cap));
1489 1490 return;
1490 1491 }
1491 1492
1492 1493 /* Process the encapsulated sub-capability */
1493 1494 ill_capability_dispatch(ill, mp, inners);
1494 1495 }
1495 1496
1496 1497 static void
1497 1498 ill_capability_dld_reset_fill(ill_t *ill, mblk_t *mp)
1498 1499 {
1499 1500 dl_capability_sub_t *dl_subcap;
1500 1501
1501 1502 if (!(ill->ill_capabilities & ILL_CAPAB_DLD))
1502 1503 return;
1503 1504
1504 1505 /*
1505 1506 * The dl_capab_dld_t that follows the dl_capability_sub_t is not
1506 1507 * initialized below since it is not used by DLD.
1507 1508 */
1508 1509 dl_subcap = (dl_capability_sub_t *)mp->b_wptr;
1509 1510 dl_subcap->dl_cap = DL_CAPAB_DLD;
1510 1511 dl_subcap->dl_length = sizeof (dl_capab_dld_t);
1511 1512
1512 1513 mp->b_wptr += sizeof (dl_capability_sub_t) + sizeof (dl_capab_dld_t);
1513 1514 }
1514 1515
1515 1516 static void
1516 1517 ill_capability_dispatch(ill_t *ill, mblk_t *mp, dl_capability_sub_t *subp)
1517 1518 {
1518 1519 /*
1519 1520 * If no ipif was brought up over this ill, this DL_CAPABILITY_REQ/ACK
1520 1521 * is only to get the VRRP capability.
1521 1522 *
1522 1523 * Note that we cannot check ill_ipif_up_count here since
1523 1524 * ill_ipif_up_count is only incremented when the resolver is setup.
1524 1525 * That is done asynchronously, and can race with this function.
1525 1526 */
1526 1527 if (!ill->ill_dl_up) {
1527 1528 if (subp->dl_cap == DL_CAPAB_VRRP)
1528 1529 ill_capability_vrrp_ack(ill, mp, subp);
1529 1530 return;
1530 1531 }
1531 1532
1532 1533 switch (subp->dl_cap) {
1533 1534 case DL_CAPAB_HCKSUM:
1534 1535 ill_capability_hcksum_ack(ill, mp, subp);
1535 1536 break;
1536 1537 case DL_CAPAB_ZEROCOPY:
1537 1538 ill_capability_zerocopy_ack(ill, mp, subp);
1538 1539 break;
1539 1540 case DL_CAPAB_DLD:
1540 1541 ill_capability_dld_ack(ill, mp, subp);
1541 1542 break;
1542 1543 case DL_CAPAB_VRRP:
1543 1544 break;
1544 1545 default:
1545 1546 ip1dbg(("ill_capability_dispatch: unknown capab type %d\n",
1546 1547 subp->dl_cap));
1547 1548 }
1548 1549 }
1549 1550
1550 1551 /*
1551 1552 * Process the vrrp capability received from a DLS Provider. isub must point
1552 1553 * to the sub-capability (DL_CAPAB_VRRP) of a DL_CAPABILITY_ACK message.
1553 1554 */
1554 1555 static void
1555 1556 ill_capability_vrrp_ack(ill_t *ill, mblk_t *mp, dl_capability_sub_t *isub)
1556 1557 {
1557 1558 dl_capab_vrrp_t *vrrp;
1558 1559 uint_t sub_dl_cap = isub->dl_cap;
1559 1560 uint8_t *capend;
1560 1561
1561 1562 ASSERT(IAM_WRITER_ILL(ill));
1562 1563 ASSERT(sub_dl_cap == DL_CAPAB_VRRP);
1563 1564
1564 1565 /*
1565 1566 * Note: range checks here are not absolutely sufficient to
1566 1567 * make us robust against malformed messages sent by drivers;
1567 1568 * this is in keeping with the rest of IP's dlpi handling.
1568 1569 * (Remember, it's coming from something else in the kernel
1569 1570 * address space)
1570 1571 */
1571 1572 capend = (uint8_t *)(isub + 1) + isub->dl_length;
1572 1573 if (capend > mp->b_wptr) {
1573 1574 cmn_err(CE_WARN, "ill_capability_vrrp_ack: "
1574 1575 "malformed sub-capability too long for mblk");
1575 1576 return;
1576 1577 }
1577 1578 vrrp = (dl_capab_vrrp_t *)(isub + 1);
1578 1579
1579 1580 /*
1580 1581 * Compare the IP address family and set ILLF_VRRP for the right ill.
1581 1582 */
1582 1583 if ((vrrp->vrrp_af == AF_INET6 && ill->ill_isv6) ||
1583 1584 (vrrp->vrrp_af == AF_INET && !ill->ill_isv6)) {
1584 1585 ill->ill_flags |= ILLF_VRRP;
1585 1586 }
1586 1587 }
1587 1588
1588 1589 /*
1589 1590 * Process a hardware checksum offload capability negotiation ack received
1590 1591 * from a DLS Provider.isub must point to the sub-capability (DL_CAPAB_HCKSUM)
1591 1592 * of a DL_CAPABILITY_ACK message.
1592 1593 */
1593 1594 static void
1594 1595 ill_capability_hcksum_ack(ill_t *ill, mblk_t *mp, dl_capability_sub_t *isub)
1595 1596 {
1596 1597 dl_capability_req_t *ocap;
1597 1598 dl_capab_hcksum_t *ihck, *ohck;
1598 1599 ill_hcksum_capab_t **ill_hcksum;
1599 1600 mblk_t *nmp = NULL;
1600 1601 uint_t sub_dl_cap = isub->dl_cap;
1601 1602 uint8_t *capend;
1602 1603
1603 1604 ASSERT(sub_dl_cap == DL_CAPAB_HCKSUM);
1604 1605
1605 1606 ill_hcksum = (ill_hcksum_capab_t **)&ill->ill_hcksum_capab;
1606 1607
1607 1608 /*
1608 1609 * Note: range checks here are not absolutely sufficient to
1609 1610 * make us robust against malformed messages sent by drivers;
1610 1611 * this is in keeping with the rest of IP's dlpi handling.
1611 1612 * (Remember, it's coming from something else in the kernel
1612 1613 * address space)
1613 1614 */
1614 1615 capend = (uint8_t *)(isub + 1) + isub->dl_length;
1615 1616 if (capend > mp->b_wptr) {
1616 1617 cmn_err(CE_WARN, "ill_capability_hcksum_ack: "
1617 1618 "malformed sub-capability too long for mblk");
1618 1619 return;
1619 1620 }
1620 1621
1621 1622 /*
1622 1623 * There are two types of acks we process here:
1623 1624 * 1. acks in reply to a (first form) generic capability req
1624 1625 * (no ENABLE flag set)
1625 1626 * 2. acks in reply to a ENABLE capability req.
1626 1627 * (ENABLE flag set)
1627 1628 */
1628 1629 ihck = (dl_capab_hcksum_t *)(isub + 1);
1629 1630
1630 1631 if (ihck->hcksum_version != HCKSUM_VERSION_1) {
1631 1632 cmn_err(CE_CONT, "ill_capability_hcksum_ack: "
1632 1633 "unsupported hardware checksum "
1633 1634 "sub-capability (version %d, expected %d)",
1634 1635 ihck->hcksum_version, HCKSUM_VERSION_1);
1635 1636 return;
1636 1637 }
1637 1638
1638 1639 if (!dlcapabcheckqid(&ihck->hcksum_mid, ill->ill_lmod_rq)) {
1639 1640 ip1dbg(("ill_capability_hcksum_ack: mid token for hardware "
1640 1641 "checksum capability isn't as expected; pass-thru "
1641 1642 "module(s) detected, discarding capability\n"));
1642 1643 return;
1643 1644 }
1644 1645
1645 1646 #define CURR_HCKSUM_CAPAB \
1646 1647 (HCKSUM_INET_PARTIAL | HCKSUM_INET_FULL_V4 | \
1647 1648 HCKSUM_INET_FULL_V6 | HCKSUM_IPHDRCKSUM)
1648 1649
1649 1650 if ((ihck->hcksum_txflags & HCKSUM_ENABLE) &&
1650 1651 (ihck->hcksum_txflags & CURR_HCKSUM_CAPAB)) {
1651 1652 /* do ENABLE processing */
1652 1653 if (*ill_hcksum == NULL) {
1653 1654 *ill_hcksum = kmem_zalloc(sizeof (ill_hcksum_capab_t),
1654 1655 KM_NOSLEEP);
1655 1656
1656 1657 if (*ill_hcksum == NULL) {
1657 1658 cmn_err(CE_WARN, "ill_capability_hcksum_ack: "
1658 1659 "could not enable hcksum version %d "
1659 1660 "for %s (ENOMEM)\n", HCKSUM_CURRENT_VERSION,
1660 1661 ill->ill_name);
1661 1662 return;
1662 1663 }
1663 1664 }
1664 1665
1665 1666 (*ill_hcksum)->ill_hcksum_version = ihck->hcksum_version;
1666 1667 (*ill_hcksum)->ill_hcksum_txflags = ihck->hcksum_txflags;
1667 1668 ill->ill_capabilities |= ILL_CAPAB_HCKSUM;
1668 1669 ip1dbg(("ill_capability_hcksum_ack: interface %s "
1669 1670 "has enabled hardware checksumming\n ",
1670 1671 ill->ill_name));
1671 1672 } else if (ihck->hcksum_txflags & CURR_HCKSUM_CAPAB) {
1672 1673 /*
1673 1674 * Enabling hardware checksum offload
1674 1675 * Currently IP supports {TCP,UDP}/IPv4
1675 1676 * partial and full cksum offload and
1676 1677 * IPv4 header checksum offload.
1677 1678 * Allocate new mblk which will
1678 1679 * contain a new capability request
1679 1680 * to enable hardware checksum offload.
1680 1681 */
1681 1682 uint_t size;
1682 1683 uchar_t *rptr;
1683 1684
1684 1685 size = sizeof (dl_capability_req_t) +
1685 1686 sizeof (dl_capability_sub_t) + isub->dl_length;
1686 1687
1687 1688 if ((nmp = ip_dlpi_alloc(size, DL_CAPABILITY_REQ)) == NULL) {
1688 1689 cmn_err(CE_WARN, "ill_capability_hcksum_ack: "
1689 1690 "could not enable hardware cksum for %s (ENOMEM)\n",
1690 1691 ill->ill_name);
1691 1692 return;
1692 1693 }
1693 1694
1694 1695 rptr = nmp->b_rptr;
1695 1696 /* initialize dl_capability_req_t */
1696 1697 ocap = (dl_capability_req_t *)nmp->b_rptr;
1697 1698 ocap->dl_sub_offset =
1698 1699 sizeof (dl_capability_req_t);
1699 1700 ocap->dl_sub_length =
1700 1701 sizeof (dl_capability_sub_t) +
1701 1702 isub->dl_length;
1702 1703 nmp->b_rptr += sizeof (dl_capability_req_t);
1703 1704
1704 1705 /* initialize dl_capability_sub_t */
1705 1706 bcopy(isub, nmp->b_rptr, sizeof (*isub));
1706 1707 nmp->b_rptr += sizeof (*isub);
1707 1708
1708 1709 /* initialize dl_capab_hcksum_t */
1709 1710 ohck = (dl_capab_hcksum_t *)nmp->b_rptr;
1710 1711 bcopy(ihck, ohck, sizeof (*ihck));
1711 1712
1712 1713 nmp->b_rptr = rptr;
1713 1714 ASSERT(nmp->b_wptr == (nmp->b_rptr + size));
1714 1715
1715 1716 /* Set ENABLE flag */
1716 1717 ohck->hcksum_txflags &= CURR_HCKSUM_CAPAB;
1717 1718 ohck->hcksum_txflags |= HCKSUM_ENABLE;
1718 1719
1719 1720 /*
1720 1721 * nmp points to a DL_CAPABILITY_REQ message to enable
1721 1722 * hardware checksum acceleration.
1722 1723 */
1723 1724 ill_capability_send(ill, nmp);
1724 1725 } else {
1725 1726 ip1dbg(("ill_capability_hcksum_ack: interface %s has "
1726 1727 "advertised %x hardware checksum capability flags\n",
1727 1728 ill->ill_name, ihck->hcksum_txflags));
1728 1729 }
1729 1730 }
1730 1731
1731 1732 static void
1732 1733 ill_capability_hcksum_reset_fill(ill_t *ill, mblk_t *mp)
1733 1734 {
1734 1735 dl_capab_hcksum_t *hck_subcap;
1735 1736 dl_capability_sub_t *dl_subcap;
1736 1737
1737 1738 if (!ILL_HCKSUM_CAPABLE(ill))
1738 1739 return;
1739 1740
1740 1741 ASSERT(ill->ill_hcksum_capab != NULL);
1741 1742
1742 1743 dl_subcap = (dl_capability_sub_t *)mp->b_wptr;
1743 1744 dl_subcap->dl_cap = DL_CAPAB_HCKSUM;
1744 1745 dl_subcap->dl_length = sizeof (*hck_subcap);
1745 1746
1746 1747 hck_subcap = (dl_capab_hcksum_t *)(dl_subcap + 1);
1747 1748 hck_subcap->hcksum_version = ill->ill_hcksum_capab->ill_hcksum_version;
1748 1749 hck_subcap->hcksum_txflags = 0;
1749 1750
1750 1751 mp->b_wptr += sizeof (*dl_subcap) + sizeof (*hck_subcap);
1751 1752 }
1752 1753
1753 1754 static void
1754 1755 ill_capability_zerocopy_ack(ill_t *ill, mblk_t *mp, dl_capability_sub_t *isub)
1755 1756 {
1756 1757 mblk_t *nmp = NULL;
1757 1758 dl_capability_req_t *oc;
1758 1759 dl_capab_zerocopy_t *zc_ic, *zc_oc;
1759 1760 ill_zerocopy_capab_t **ill_zerocopy_capab;
1760 1761 uint_t sub_dl_cap = isub->dl_cap;
1761 1762 uint8_t *capend;
1762 1763
1763 1764 ASSERT(sub_dl_cap == DL_CAPAB_ZEROCOPY);
1764 1765
1765 1766 ill_zerocopy_capab = (ill_zerocopy_capab_t **)&ill->ill_zerocopy_capab;
1766 1767
1767 1768 /*
1768 1769 * Note: range checks here are not absolutely sufficient to
1769 1770 * make us robust against malformed messages sent by drivers;
1770 1771 * this is in keeping with the rest of IP's dlpi handling.
1771 1772 * (Remember, it's coming from something else in the kernel
1772 1773 * address space)
1773 1774 */
1774 1775 capend = (uint8_t *)(isub + 1) + isub->dl_length;
1775 1776 if (capend > mp->b_wptr) {
1776 1777 cmn_err(CE_WARN, "ill_capability_zerocopy_ack: "
1777 1778 "malformed sub-capability too long for mblk");
1778 1779 return;
1779 1780 }
1780 1781
1781 1782 zc_ic = (dl_capab_zerocopy_t *)(isub + 1);
1782 1783 if (zc_ic->zerocopy_version != ZEROCOPY_VERSION_1) {
1783 1784 cmn_err(CE_CONT, "ill_capability_zerocopy_ack: "
1784 1785 "unsupported ZEROCOPY sub-capability (version %d, "
1785 1786 "expected %d)", zc_ic->zerocopy_version,
1786 1787 ZEROCOPY_VERSION_1);
1787 1788 return;
1788 1789 }
1789 1790
1790 1791 if (!dlcapabcheckqid(&zc_ic->zerocopy_mid, ill->ill_lmod_rq)) {
1791 1792 ip1dbg(("ill_capability_zerocopy_ack: mid token for zerocopy "
1792 1793 "capability isn't as expected; pass-thru module(s) "
1793 1794 "detected, discarding capability\n"));
1794 1795 return;
1795 1796 }
1796 1797
1797 1798 if ((zc_ic->zerocopy_flags & DL_CAPAB_VMSAFE_MEM) != 0) {
1798 1799 if (*ill_zerocopy_capab == NULL) {
1799 1800 *ill_zerocopy_capab =
1800 1801 kmem_zalloc(sizeof (ill_zerocopy_capab_t),
1801 1802 KM_NOSLEEP);
1802 1803
1803 1804 if (*ill_zerocopy_capab == NULL) {
1804 1805 cmn_err(CE_WARN, "ill_capability_zerocopy_ack: "
1805 1806 "could not enable Zero-copy version %d "
1806 1807 "for %s (ENOMEM)\n", ZEROCOPY_VERSION_1,
1807 1808 ill->ill_name);
1808 1809 return;
1809 1810 }
1810 1811 }
1811 1812
1812 1813 ip1dbg(("ill_capability_zerocopy_ack: interface %s "
1813 1814 "supports Zero-copy version %d\n", ill->ill_name,
1814 1815 ZEROCOPY_VERSION_1));
1815 1816
1816 1817 (*ill_zerocopy_capab)->ill_zerocopy_version =
1817 1818 zc_ic->zerocopy_version;
1818 1819 (*ill_zerocopy_capab)->ill_zerocopy_flags =
1819 1820 zc_ic->zerocopy_flags;
1820 1821
1821 1822 ill->ill_capabilities |= ILL_CAPAB_ZEROCOPY;
1822 1823 } else {
1823 1824 uint_t size;
1824 1825 uchar_t *rptr;
1825 1826
1826 1827 size = sizeof (dl_capability_req_t) +
1827 1828 sizeof (dl_capability_sub_t) +
1828 1829 sizeof (dl_capab_zerocopy_t);
1829 1830
1830 1831 if ((nmp = ip_dlpi_alloc(size, DL_CAPABILITY_REQ)) == NULL) {
1831 1832 cmn_err(CE_WARN, "ill_capability_zerocopy_ack: "
1832 1833 "could not enable zerocopy for %s (ENOMEM)\n",
1833 1834 ill->ill_name);
1834 1835 return;
1835 1836 }
1836 1837
1837 1838 rptr = nmp->b_rptr;
1838 1839 /* initialize dl_capability_req_t */
1839 1840 oc = (dl_capability_req_t *)rptr;
1840 1841 oc->dl_sub_offset = sizeof (dl_capability_req_t);
1841 1842 oc->dl_sub_length = sizeof (dl_capability_sub_t) +
1842 1843 sizeof (dl_capab_zerocopy_t);
1843 1844 rptr += sizeof (dl_capability_req_t);
1844 1845
1845 1846 /* initialize dl_capability_sub_t */
1846 1847 bcopy(isub, rptr, sizeof (*isub));
1847 1848 rptr += sizeof (*isub);
1848 1849
1849 1850 /* initialize dl_capab_zerocopy_t */
1850 1851 zc_oc = (dl_capab_zerocopy_t *)rptr;
1851 1852 *zc_oc = *zc_ic;
1852 1853
1853 1854 ip1dbg(("ill_capability_zerocopy_ack: asking interface %s "
1854 1855 "to enable zero-copy version %d\n", ill->ill_name,
1855 1856 ZEROCOPY_VERSION_1));
1856 1857
1857 1858 /* set VMSAFE_MEM flag */
1858 1859 zc_oc->zerocopy_flags |= DL_CAPAB_VMSAFE_MEM;
1859 1860
1860 1861 /* nmp points to a DL_CAPABILITY_REQ message to enable zcopy */
1861 1862 ill_capability_send(ill, nmp);
1862 1863 }
1863 1864 }
1864 1865
1865 1866 static void
1866 1867 ill_capability_zerocopy_reset_fill(ill_t *ill, mblk_t *mp)
1867 1868 {
1868 1869 dl_capab_zerocopy_t *zerocopy_subcap;
1869 1870 dl_capability_sub_t *dl_subcap;
1870 1871
1871 1872 if (!(ill->ill_capabilities & ILL_CAPAB_ZEROCOPY))
1872 1873 return;
1873 1874
1874 1875 ASSERT(ill->ill_zerocopy_capab != NULL);
1875 1876
1876 1877 dl_subcap = (dl_capability_sub_t *)mp->b_wptr;
1877 1878 dl_subcap->dl_cap = DL_CAPAB_ZEROCOPY;
1878 1879 dl_subcap->dl_length = sizeof (*zerocopy_subcap);
1879 1880
1880 1881 zerocopy_subcap = (dl_capab_zerocopy_t *)(dl_subcap + 1);
1881 1882 zerocopy_subcap->zerocopy_version =
1882 1883 ill->ill_zerocopy_capab->ill_zerocopy_version;
1883 1884 zerocopy_subcap->zerocopy_flags = 0;
1884 1885
1885 1886 mp->b_wptr += sizeof (*dl_subcap) + sizeof (*zerocopy_subcap);
1886 1887 }
1887 1888
1888 1889 /*
1889 1890 * DLD capability
1890 1891 * Refer to dld.h for more information regarding the purpose and usage
1891 1892 * of this capability.
1892 1893 */
1893 1894 static void
1894 1895 ill_capability_dld_ack(ill_t *ill, mblk_t *mp, dl_capability_sub_t *isub)
1895 1896 {
1896 1897 dl_capab_dld_t *dld_ic, dld;
1897 1898 uint_t sub_dl_cap = isub->dl_cap;
1898 1899 uint8_t *capend;
1899 1900 ill_dld_capab_t *idc;
1900 1901
1901 1902 ASSERT(IAM_WRITER_ILL(ill));
1902 1903 ASSERT(sub_dl_cap == DL_CAPAB_DLD);
1903 1904
1904 1905 /*
1905 1906 * Note: range checks here are not absolutely sufficient to
1906 1907 * make us robust against malformed messages sent by drivers;
1907 1908 * this is in keeping with the rest of IP's dlpi handling.
1908 1909 * (Remember, it's coming from something else in the kernel
1909 1910 * address space)
1910 1911 */
1911 1912 capend = (uint8_t *)(isub + 1) + isub->dl_length;
1912 1913 if (capend > mp->b_wptr) {
1913 1914 cmn_err(CE_WARN, "ill_capability_dld_ack: "
1914 1915 "malformed sub-capability too long for mblk");
1915 1916 return;
1916 1917 }
1917 1918 dld_ic = (dl_capab_dld_t *)(isub + 1);
1918 1919 if (dld_ic->dld_version != DLD_CURRENT_VERSION) {
1919 1920 cmn_err(CE_CONT, "ill_capability_dld_ack: "
1920 1921 "unsupported DLD sub-capability (version %d, "
1921 1922 "expected %d)", dld_ic->dld_version,
1922 1923 DLD_CURRENT_VERSION);
1923 1924 return;
1924 1925 }
1925 1926 if (!dlcapabcheckqid(&dld_ic->dld_mid, ill->ill_lmod_rq)) {
1926 1927 ip1dbg(("ill_capability_dld_ack: mid token for dld "
1927 1928 "capability isn't as expected; pass-thru module(s) "
1928 1929 "detected, discarding capability\n"));
1929 1930 return;
1930 1931 }
1931 1932
1932 1933 /*
1933 1934 * Copy locally to ensure alignment.
1934 1935 */
1935 1936 bcopy(dld_ic, &dld, sizeof (dl_capab_dld_t));
1936 1937
1937 1938 if ((idc = ill->ill_dld_capab) == NULL) {
1938 1939 idc = kmem_zalloc(sizeof (ill_dld_capab_t), KM_NOSLEEP);
1939 1940 if (idc == NULL) {
1940 1941 cmn_err(CE_WARN, "ill_capability_dld_ack: "
1941 1942 "could not enable DLD version %d "
1942 1943 "for %s (ENOMEM)\n", DLD_CURRENT_VERSION,
1943 1944 ill->ill_name);
1944 1945 return;
1945 1946 }
1946 1947 ill->ill_dld_capab = idc;
1947 1948 }
1948 1949 idc->idc_capab_df = (ip_capab_func_t)dld.dld_capab;
1949 1950 idc->idc_capab_dh = (void *)dld.dld_capab_handle;
1950 1951 ip1dbg(("ill_capability_dld_ack: interface %s "
1951 1952 "supports DLD version %d\n", ill->ill_name, DLD_CURRENT_VERSION));
1952 1953
1953 1954 ill_capability_dld_enable(ill);
1954 1955 }
1955 1956
1956 1957 /*
1957 1958 * Typically capability negotiation between IP and the driver happens via
1958 1959 * DLPI message exchange. However GLD also offers a direct function call
1959 1960 * mechanism to exchange the DLD_DIRECT_CAPAB and DLD_POLL_CAPAB capabilities,
1960 1961 * But arbitrary function calls into IP or GLD are not permitted, since both
1961 1962 * of them are protected by their own perimeter mechanism. The perimeter can
1962 1963 * be viewed as a coarse lock or serialization mechanism. The hierarchy of
1963 1964 * these perimeters is IP -> MAC. Thus for example to enable the squeue
1964 1965 * polling, IP needs to enter its perimeter, then call ill_mac_perim_enter
1965 1966 * to enter the mac perimeter and then do the direct function calls into
1966 1967 * GLD to enable squeue polling. The ring related callbacks from the mac into
1967 1968 * the stack to add, bind, quiesce, restart or cleanup a ring are all
1968 1969 * protected by the mac perimeter.
1969 1970 */
1970 1971 static void
1971 1972 ill_mac_perim_enter(ill_t *ill, mac_perim_handle_t *mphp)
1972 1973 {
1973 1974 ill_dld_capab_t *idc = ill->ill_dld_capab;
1974 1975 int err;
1975 1976
1976 1977 err = idc->idc_capab_df(idc->idc_capab_dh, DLD_CAPAB_PERIM, mphp,
1977 1978 DLD_ENABLE);
1978 1979 ASSERT(err == 0);
1979 1980 }
1980 1981
1981 1982 static void
1982 1983 ill_mac_perim_exit(ill_t *ill, mac_perim_handle_t mph)
1983 1984 {
1984 1985 ill_dld_capab_t *idc = ill->ill_dld_capab;
1985 1986 int err;
1986 1987
1987 1988 err = idc->idc_capab_df(idc->idc_capab_dh, DLD_CAPAB_PERIM, mph,
1988 1989 DLD_DISABLE);
1989 1990 ASSERT(err == 0);
1990 1991 }
1991 1992
1992 1993 boolean_t
1993 1994 ill_mac_perim_held(ill_t *ill)
1994 1995 {
1995 1996 ill_dld_capab_t *idc = ill->ill_dld_capab;
1996 1997
1997 1998 return (idc->idc_capab_df(idc->idc_capab_dh, DLD_CAPAB_PERIM, NULL,
1998 1999 DLD_QUERY));
1999 2000 }
2000 2001
2001 2002 static void
2002 2003 ill_capability_direct_enable(ill_t *ill)
2003 2004 {
2004 2005 ill_dld_capab_t *idc = ill->ill_dld_capab;
2005 2006 ill_dld_direct_t *idd = &idc->idc_direct;
2006 2007 dld_capab_direct_t direct;
2007 2008 int rc;
2008 2009
2009 2010 ASSERT(!ill->ill_isv6 && IAM_WRITER_ILL(ill));
2010 2011
2011 2012 bzero(&direct, sizeof (direct));
2012 2013 direct.di_rx_cf = (uintptr_t)ip_input;
2013 2014 direct.di_rx_ch = ill;
2014 2015
2015 2016 rc = idc->idc_capab_df(idc->idc_capab_dh, DLD_CAPAB_DIRECT, &direct,
2016 2017 DLD_ENABLE);
2017 2018 if (rc == 0) {
2018 2019 idd->idd_tx_df = (ip_dld_tx_t)direct.di_tx_df;
2019 2020 idd->idd_tx_dh = direct.di_tx_dh;
2020 2021 idd->idd_tx_cb_df = (ip_dld_callb_t)direct.di_tx_cb_df;
2021 2022 idd->idd_tx_cb_dh = direct.di_tx_cb_dh;
2022 2023 idd->idd_tx_fctl_df = (ip_dld_fctl_t)direct.di_tx_fctl_df;
2023 2024 idd->idd_tx_fctl_dh = direct.di_tx_fctl_dh;
2024 2025 ASSERT(idd->idd_tx_cb_df != NULL);
2025 2026 ASSERT(idd->idd_tx_fctl_df != NULL);
2026 2027 ASSERT(idd->idd_tx_df != NULL);
2027 2028 /*
2028 2029 * One time registration of flow enable callback function
2029 2030 */
2030 2031 ill->ill_flownotify_mh = idd->idd_tx_cb_df(idd->idd_tx_cb_dh,
2031 2032 ill_flow_enable, ill);
2032 2033 ill->ill_capabilities |= ILL_CAPAB_DLD_DIRECT;
2033 2034 DTRACE_PROBE1(direct_on, (ill_t *), ill);
2034 2035 } else {
2035 2036 cmn_err(CE_WARN, "warning: could not enable DIRECT "
2036 2037 "capability, rc = %d\n", rc);
2037 2038 DTRACE_PROBE2(direct_off, (ill_t *), ill, (int), rc);
2038 2039 }
2039 2040 }
2040 2041
2041 2042 static void
2042 2043 ill_capability_poll_enable(ill_t *ill)
2043 2044 {
2044 2045 ill_dld_capab_t *idc = ill->ill_dld_capab;
2045 2046 dld_capab_poll_t poll;
2046 2047 int rc;
2047 2048
2048 2049 ASSERT(!ill->ill_isv6 && IAM_WRITER_ILL(ill));
2049 2050
2050 2051 bzero(&poll, sizeof (poll));
2051 2052 poll.poll_ring_add_cf = (uintptr_t)ip_squeue_add_ring;
2052 2053 poll.poll_ring_remove_cf = (uintptr_t)ip_squeue_clean_ring;
2053 2054 poll.poll_ring_quiesce_cf = (uintptr_t)ip_squeue_quiesce_ring;
2054 2055 poll.poll_ring_restart_cf = (uintptr_t)ip_squeue_restart_ring;
2055 2056 poll.poll_ring_bind_cf = (uintptr_t)ip_squeue_bind_ring;
2056 2057 poll.poll_ring_ch = ill;
2057 2058 rc = idc->idc_capab_df(idc->idc_capab_dh, DLD_CAPAB_POLL, &poll,
2058 2059 DLD_ENABLE);
2059 2060 if (rc == 0) {
2060 2061 ill->ill_capabilities |= ILL_CAPAB_DLD_POLL;
2061 2062 DTRACE_PROBE1(poll_on, (ill_t *), ill);
2062 2063 } else {
2063 2064 ip1dbg(("warning: could not enable POLL "
2064 2065 "capability, rc = %d\n", rc));
2065 2066 DTRACE_PROBE2(poll_off, (ill_t *), ill, (int), rc);
2066 2067 }
2067 2068 }
2068 2069
2069 2070 /*
2070 2071 * Enable the LSO capability.
2071 2072 */
2072 2073 static void
2073 2074 ill_capability_lso_enable(ill_t *ill)
2074 2075 {
2075 2076 ill_dld_capab_t *idc = ill->ill_dld_capab;
2076 2077 dld_capab_lso_t lso;
2077 2078 int rc;
2078 2079
2079 2080 ASSERT(!ill->ill_isv6 && IAM_WRITER_ILL(ill));
2080 2081
2081 2082 if (ill->ill_lso_capab == NULL) {
2082 2083 ill->ill_lso_capab = kmem_zalloc(sizeof (ill_lso_capab_t),
2083 2084 KM_NOSLEEP);
2084 2085 if (ill->ill_lso_capab == NULL) {
2085 2086 cmn_err(CE_WARN, "ill_capability_lso_enable: "
2086 2087 "could not enable LSO for %s (ENOMEM)\n",
2087 2088 ill->ill_name);
2088 2089 return;
2089 2090 }
2090 2091 }
2091 2092
2092 2093 bzero(&lso, sizeof (lso));
2093 2094 if ((rc = idc->idc_capab_df(idc->idc_capab_dh, DLD_CAPAB_LSO, &lso,
2094 2095 DLD_ENABLE)) == 0) {
2095 2096 ill->ill_lso_capab->ill_lso_flags = lso.lso_flags;
2096 2097 ill->ill_lso_capab->ill_lso_max = lso.lso_max;
2097 2098 ill->ill_capabilities |= ILL_CAPAB_LSO;
2098 2099 ip1dbg(("ill_capability_lso_enable: interface %s "
2099 2100 "has enabled LSO\n ", ill->ill_name));
2100 2101 } else {
2101 2102 kmem_free(ill->ill_lso_capab, sizeof (ill_lso_capab_t));
2102 2103 ill->ill_lso_capab = NULL;
2103 2104 DTRACE_PROBE2(lso_off, (ill_t *), ill, (int), rc);
2104 2105 }
2105 2106 }
2106 2107
2107 2108 static void
2108 2109 ill_capability_dld_enable(ill_t *ill)
2109 2110 {
2110 2111 mac_perim_handle_t mph;
2111 2112
2112 2113 ASSERT(IAM_WRITER_ILL(ill));
2113 2114
2114 2115 if (ill->ill_isv6)
2115 2116 return;
2116 2117
2117 2118 ill_mac_perim_enter(ill, &mph);
2118 2119 if (!ill->ill_isv6) {
2119 2120 ill_capability_direct_enable(ill);
2120 2121 ill_capability_poll_enable(ill);
2121 2122 ill_capability_lso_enable(ill);
2122 2123 }
2123 2124 ill->ill_capabilities |= ILL_CAPAB_DLD;
2124 2125 ill_mac_perim_exit(ill, mph);
2125 2126 }
2126 2127
2127 2128 static void
2128 2129 ill_capability_dld_disable(ill_t *ill)
2129 2130 {
2130 2131 ill_dld_capab_t *idc;
2131 2132 ill_dld_direct_t *idd;
2132 2133 mac_perim_handle_t mph;
2133 2134
2134 2135 ASSERT(IAM_WRITER_ILL(ill));
2135 2136
2136 2137 if (!(ill->ill_capabilities & ILL_CAPAB_DLD))
2137 2138 return;
2138 2139
2139 2140 ill_mac_perim_enter(ill, &mph);
2140 2141
2141 2142 idc = ill->ill_dld_capab;
2142 2143 if ((ill->ill_capabilities & ILL_CAPAB_DLD_DIRECT) != 0) {
2143 2144 /*
2144 2145 * For performance we avoid locks in the transmit data path
2145 2146 * and don't maintain a count of the number of threads using
2146 2147 * direct calls. Thus some threads could be using direct
2147 2148 * transmit calls to GLD, even after the capability mechanism
2148 2149 * turns it off. This is still safe since the handles used in
2149 2150 * the direct calls continue to be valid until the unplumb is
2150 2151 * completed. Remove the callback that was added (1-time) at
2151 2152 * capab enable time.
2152 2153 */
2153 2154 mutex_enter(&ill->ill_lock);
2154 2155 ill->ill_capabilities &= ~ILL_CAPAB_DLD_DIRECT;
2155 2156 mutex_exit(&ill->ill_lock);
2156 2157 if (ill->ill_flownotify_mh != NULL) {
2157 2158 idd = &idc->idc_direct;
2158 2159 idd->idd_tx_cb_df(idd->idd_tx_cb_dh, NULL,
2159 2160 ill->ill_flownotify_mh);
2160 2161 ill->ill_flownotify_mh = NULL;
2161 2162 }
2162 2163 (void) idc->idc_capab_df(idc->idc_capab_dh, DLD_CAPAB_DIRECT,
2163 2164 NULL, DLD_DISABLE);
2164 2165 }
2165 2166
2166 2167 if ((ill->ill_capabilities & ILL_CAPAB_DLD_POLL) != 0) {
2167 2168 ill->ill_capabilities &= ~ILL_CAPAB_DLD_POLL;
2168 2169 ip_squeue_clean_all(ill);
2169 2170 (void) idc->idc_capab_df(idc->idc_capab_dh, DLD_CAPAB_POLL,
2170 2171 NULL, DLD_DISABLE);
2171 2172 }
2172 2173
2173 2174 if ((ill->ill_capabilities & ILL_CAPAB_LSO) != 0) {
2174 2175 ASSERT(ill->ill_lso_capab != NULL);
2175 2176 /*
2176 2177 * Clear the capability flag for LSO but retain the
2177 2178 * ill_lso_capab structure since it's possible that another
2178 2179 * thread is still referring to it. The structure only gets
2179 2180 * deallocated when we destroy the ill.
2180 2181 */
2181 2182
2182 2183 ill->ill_capabilities &= ~ILL_CAPAB_LSO;
2183 2184 (void) idc->idc_capab_df(idc->idc_capab_dh, DLD_CAPAB_LSO,
2184 2185 NULL, DLD_DISABLE);
2185 2186 }
2186 2187
2187 2188 ill->ill_capabilities &= ~ILL_CAPAB_DLD;
2188 2189 ill_mac_perim_exit(ill, mph);
2189 2190 }
2190 2191
2191 2192 /*
2192 2193 * Capability Negotiation protocol
2193 2194 *
2194 2195 * We don't wait for DLPI capability operations to finish during interface
2195 2196 * bringup or teardown. Doing so would introduce more asynchrony and the
2196 2197 * interface up/down operations will need multiple return and restarts.
2197 2198 * Instead the 'ipsq_current_ipif' of the ipsq is not cleared as long as
2198 2199 * the 'ill_dlpi_deferred' chain is non-empty. This ensures that the next
2199 2200 * exclusive operation won't start until the DLPI operations of the previous
2200 2201 * exclusive operation complete.
2201 2202 *
2202 2203 * The capability state machine is shown below.
2203 2204 *
2204 2205 * state next state event, action
2205 2206 *
2206 2207 * IDCS_UNKNOWN IDCS_PROBE_SENT ill_capability_probe
2207 2208 * IDCS_PROBE_SENT IDCS_OK ill_capability_ack
2208 2209 * IDCS_PROBE_SENT IDCS_FAILED ip_rput_dlpi_writer (nack)
2209 2210 * IDCS_OK IDCS_RENEG Receipt of DL_NOTE_CAPAB_RENEG
2210 2211 * IDCS_OK IDCS_RESET_SENT ill_capability_reset
2211 2212 * IDCS_RESET_SENT IDCS_UNKNOWN ill_capability_ack_thr
2212 2213 * IDCS_RENEG IDCS_PROBE_SENT ill_capability_ack_thr ->
2213 2214 * ill_capability_probe.
2214 2215 */
2215 2216
2216 2217 /*
2217 2218 * Dedicated thread started from ip_stack_init that handles capability
2218 2219 * disable. This thread ensures the taskq dispatch does not fail by waiting
2219 2220 * for resources using TQ_SLEEP. The taskq mechanism is used to ensure
2220 2221 * that direct calls to DLD are done in a cv_waitable context.
2221 2222 */
2222 2223 void
2223 2224 ill_taskq_dispatch(ip_stack_t *ipst)
2224 2225 {
2225 2226 callb_cpr_t cprinfo;
2226 2227 char name[64];
2227 2228 mblk_t *mp;
2228 2229
2229 2230 (void) snprintf(name, sizeof (name), "ill_taskq_dispatch_%d",
2230 2231 ipst->ips_netstack->netstack_stackid);
2231 2232 CALLB_CPR_INIT(&cprinfo, &ipst->ips_capab_taskq_lock, callb_generic_cpr,
2232 2233 name);
2233 2234 mutex_enter(&ipst->ips_capab_taskq_lock);
2234 2235
2235 2236 for (;;) {
2236 2237 mp = ipst->ips_capab_taskq_head;
2237 2238 while (mp != NULL) {
2238 2239 ipst->ips_capab_taskq_head = mp->b_next;
2239 2240 if (ipst->ips_capab_taskq_head == NULL)
2240 2241 ipst->ips_capab_taskq_tail = NULL;
2241 2242 mutex_exit(&ipst->ips_capab_taskq_lock);
2242 2243 mp->b_next = NULL;
2243 2244
2244 2245 VERIFY(taskq_dispatch(system_taskq,
2245 2246 ill_capability_ack_thr, mp, TQ_SLEEP) != 0);
2246 2247 mutex_enter(&ipst->ips_capab_taskq_lock);
2247 2248 mp = ipst->ips_capab_taskq_head;
2248 2249 }
2249 2250
2250 2251 if (ipst->ips_capab_taskq_quit)
2251 2252 break;
2252 2253 CALLB_CPR_SAFE_BEGIN(&cprinfo);
2253 2254 cv_wait(&ipst->ips_capab_taskq_cv, &ipst->ips_capab_taskq_lock);
2254 2255 CALLB_CPR_SAFE_END(&cprinfo, &ipst->ips_capab_taskq_lock);
2255 2256 }
2256 2257 VERIFY(ipst->ips_capab_taskq_head == NULL);
2257 2258 VERIFY(ipst->ips_capab_taskq_tail == NULL);
2258 2259 CALLB_CPR_EXIT(&cprinfo);
2259 2260 thread_exit();
2260 2261 }
2261 2262
2262 2263 /*
2263 2264 * Consume a new-style hardware capabilities negotiation ack.
2264 2265 * Called via taskq on receipt of DL_CAPABILITY_ACK.
2265 2266 */
2266 2267 static void
2267 2268 ill_capability_ack_thr(void *arg)
2268 2269 {
2269 2270 mblk_t *mp = arg;
2270 2271 dl_capability_ack_t *capp;
2271 2272 dl_capability_sub_t *subp, *endp;
2272 2273 ill_t *ill;
2273 2274 boolean_t reneg;
2274 2275
2275 2276 ill = (ill_t *)mp->b_prev;
2276 2277 mp->b_prev = NULL;
2277 2278
2278 2279 VERIFY(ipsq_enter(ill, B_FALSE, CUR_OP) == B_TRUE);
2279 2280
2280 2281 if (ill->ill_dlpi_capab_state == IDCS_RESET_SENT ||
2281 2282 ill->ill_dlpi_capab_state == IDCS_RENEG) {
2282 2283 /*
2283 2284 * We have received the ack for our DL_CAPAB reset request.
2284 2285 * There isnt' anything in the message that needs processing.
2285 2286 * All message based capabilities have been disabled, now
2286 2287 * do the function call based capability disable.
2287 2288 */
2288 2289 reneg = ill->ill_dlpi_capab_state == IDCS_RENEG;
2289 2290 ill_capability_dld_disable(ill);
2290 2291 ill->ill_dlpi_capab_state = IDCS_UNKNOWN;
2291 2292 if (reneg)
2292 2293 ill_capability_probe(ill);
2293 2294 goto done;
2294 2295 }
2295 2296
2296 2297 if (ill->ill_dlpi_capab_state == IDCS_PROBE_SENT)
2297 2298 ill->ill_dlpi_capab_state = IDCS_OK;
2298 2299
2299 2300 capp = (dl_capability_ack_t *)mp->b_rptr;
2300 2301
2301 2302 if (capp->dl_sub_length == 0) {
2302 2303 /* no new-style capabilities */
2303 2304 goto done;
2304 2305 }
2305 2306
2306 2307 /* make sure the driver supplied correct dl_sub_length */
2307 2308 if ((sizeof (*capp) + capp->dl_sub_length) > MBLKL(mp)) {
2308 2309 ip0dbg(("ill_capability_ack: bad DL_CAPABILITY_ACK, "
2309 2310 "invalid dl_sub_length (%d)\n", capp->dl_sub_length));
2310 2311 goto done;
2311 2312 }
2312 2313
2313 2314 #define SC(base, offset) (dl_capability_sub_t *)(((uchar_t *)(base))+(offset))
2314 2315 /*
2315 2316 * There are sub-capabilities. Process the ones we know about.
2316 2317 * Loop until we don't have room for another sub-cap header..
2317 2318 */
2318 2319 for (subp = SC(capp, capp->dl_sub_offset),
2319 2320 endp = SC(subp, capp->dl_sub_length - sizeof (*subp));
2320 2321 subp <= endp;
2321 2322 subp = SC(subp, sizeof (dl_capability_sub_t) + subp->dl_length)) {
2322 2323
2323 2324 switch (subp->dl_cap) {
2324 2325 case DL_CAPAB_ID_WRAPPER:
2325 2326 ill_capability_id_ack(ill, mp, subp);
2326 2327 break;
2327 2328 default:
2328 2329 ill_capability_dispatch(ill, mp, subp);
2329 2330 break;
2330 2331 }
2331 2332 }
2332 2333 #undef SC
2333 2334 done:
2334 2335 inet_freemsg(mp);
2335 2336 ill_capability_done(ill);
2336 2337 ipsq_exit(ill->ill_phyint->phyint_ipsq);
2337 2338 }
2338 2339
2339 2340 /*
2340 2341 * This needs to be started in a taskq thread to provide a cv_waitable
2341 2342 * context.
2342 2343 */
2343 2344 void
2344 2345 ill_capability_ack(ill_t *ill, mblk_t *mp)
2345 2346 {
2346 2347 ip_stack_t *ipst = ill->ill_ipst;
2347 2348
2348 2349 mp->b_prev = (mblk_t *)ill;
2349 2350 ASSERT(mp->b_next == NULL);
2350 2351
2351 2352 if (taskq_dispatch(system_taskq, ill_capability_ack_thr, mp,
2352 2353 TQ_NOSLEEP) != 0)
2353 2354 return;
2354 2355
2355 2356 /*
2356 2357 * The taskq dispatch failed. Signal the ill_taskq_dispatch thread
2357 2358 * which will do the dispatch using TQ_SLEEP to guarantee success.
2358 2359 */
2359 2360 mutex_enter(&ipst->ips_capab_taskq_lock);
2360 2361 if (ipst->ips_capab_taskq_head == NULL) {
2361 2362 ASSERT(ipst->ips_capab_taskq_tail == NULL);
2362 2363 ipst->ips_capab_taskq_head = mp;
2363 2364 } else {
2364 2365 ipst->ips_capab_taskq_tail->b_next = mp;
2365 2366 }
2366 2367 ipst->ips_capab_taskq_tail = mp;
2367 2368
2368 2369 cv_signal(&ipst->ips_capab_taskq_cv);
2369 2370 mutex_exit(&ipst->ips_capab_taskq_lock);
2370 2371 }
2371 2372
2372 2373 /*
2373 2374 * This routine is called to scan the fragmentation reassembly table for
2374 2375 * the specified ILL for any packets that are starting to smell.
2375 2376 * dead_interval is the maximum time in seconds that will be tolerated. It
2376 2377 * will either be the value specified in ip_g_frag_timeout, or zero if the
2377 2378 * ILL is shutting down and it is time to blow everything off.
2378 2379 *
2379 2380 * It returns the number of seconds (as a time_t) that the next frag timer
2380 2381 * should be scheduled for, 0 meaning that the timer doesn't need to be
2381 2382 * re-started. Note that the method of calculating next_timeout isn't
2382 2383 * entirely accurate since time will flow between the time we grab
2383 2384 * current_time and the time we schedule the next timeout. This isn't a
2384 2385 * big problem since this is the timer for sending an ICMP reassembly time
2385 2386 * exceeded messages, and it doesn't have to be exactly accurate.
2386 2387 *
2387 2388 * This function is
2388 2389 * sometimes called as writer, although this is not required.
2389 2390 */
2390 2391 time_t
2391 2392 ill_frag_timeout(ill_t *ill, time_t dead_interval)
2392 2393 {
2393 2394 ipfb_t *ipfb;
2394 2395 ipfb_t *endp;
2395 2396 ipf_t *ipf;
2396 2397 ipf_t *ipfnext;
2397 2398 mblk_t *mp;
2398 2399 time_t current_time = gethrestime_sec();
2399 2400 time_t next_timeout = 0;
2400 2401 uint32_t hdr_length;
2401 2402 mblk_t *send_icmp_head;
2402 2403 mblk_t *send_icmp_head_v6;
2403 2404 ip_stack_t *ipst = ill->ill_ipst;
2404 2405 ip_recv_attr_t iras;
2405 2406
2406 2407 bzero(&iras, sizeof (iras));
2407 2408 iras.ira_flags = 0;
2408 2409 iras.ira_ill = iras.ira_rill = ill;
2409 2410 iras.ira_ruifindex = ill->ill_phyint->phyint_ifindex;
2410 2411 iras.ira_rifindex = iras.ira_ruifindex;
2411 2412
2412 2413 ipfb = ill->ill_frag_hash_tbl;
2413 2414 if (ipfb == NULL)
2414 2415 return (B_FALSE);
2415 2416 endp = &ipfb[ILL_FRAG_HASH_TBL_COUNT];
2416 2417 /* Walk the frag hash table. */
2417 2418 for (; ipfb < endp; ipfb++) {
2418 2419 send_icmp_head = NULL;
2419 2420 send_icmp_head_v6 = NULL;
2420 2421 mutex_enter(&ipfb->ipfb_lock);
2421 2422 while ((ipf = ipfb->ipfb_ipf) != 0) {
2422 2423 time_t frag_time = current_time - ipf->ipf_timestamp;
2423 2424 time_t frag_timeout;
2424 2425
2425 2426 if (frag_time < dead_interval) {
2426 2427 /*
2427 2428 * There are some outstanding fragments
2428 2429 * that will timeout later. Make note of
2429 2430 * the time so that we can reschedule the
2430 2431 * next timeout appropriately.
2431 2432 */
2432 2433 frag_timeout = dead_interval - frag_time;
2433 2434 if (next_timeout == 0 ||
2434 2435 frag_timeout < next_timeout) {
2435 2436 next_timeout = frag_timeout;
2436 2437 }
2437 2438 break;
2438 2439 }
2439 2440 /* Time's up. Get it out of here. */
2440 2441 hdr_length = ipf->ipf_nf_hdr_len;
2441 2442 ipfnext = ipf->ipf_hash_next;
2442 2443 if (ipfnext)
2443 2444 ipfnext->ipf_ptphn = ipf->ipf_ptphn;
2444 2445 *ipf->ipf_ptphn = ipfnext;
2445 2446 mp = ipf->ipf_mp->b_cont;
2446 2447 for (; mp; mp = mp->b_cont) {
2447 2448 /* Extra points for neatness. */
2448 2449 IP_REASS_SET_START(mp, 0);
2449 2450 IP_REASS_SET_END(mp, 0);
2450 2451 }
2451 2452 mp = ipf->ipf_mp->b_cont;
2452 2453 atomic_add_32(&ill->ill_frag_count, -ipf->ipf_count);
2453 2454 ASSERT(ipfb->ipfb_count >= ipf->ipf_count);
2454 2455 ipfb->ipfb_count -= ipf->ipf_count;
2455 2456 ASSERT(ipfb->ipfb_frag_pkts > 0);
2456 2457 ipfb->ipfb_frag_pkts--;
2457 2458 /*
2458 2459 * We do not send any icmp message from here because
2459 2460 * we currently are holding the ipfb_lock for this
2460 2461 * hash chain. If we try and send any icmp messages
2461 2462 * from here we may end up via a put back into ip
2462 2463 * trying to get the same lock, causing a recursive
2463 2464 * mutex panic. Instead we build a list and send all
2464 2465 * the icmp messages after we have dropped the lock.
2465 2466 */
2466 2467 if (ill->ill_isv6) {
2467 2468 if (hdr_length != 0) {
2468 2469 mp->b_next = send_icmp_head_v6;
2469 2470 send_icmp_head_v6 = mp;
2470 2471 } else {
2471 2472 freemsg(mp);
2472 2473 }
2473 2474 } else {
2474 2475 if (hdr_length != 0) {
2475 2476 mp->b_next = send_icmp_head;
2476 2477 send_icmp_head = mp;
2477 2478 } else {
2478 2479 freemsg(mp);
2479 2480 }
2480 2481 }
2481 2482 BUMP_MIB(ill->ill_ip_mib, ipIfStatsReasmFails);
2482 2483 ip_drop_input("ipIfStatsReasmFails", ipf->ipf_mp, ill);
2483 2484 freeb(ipf->ipf_mp);
2484 2485 }
2485 2486 mutex_exit(&ipfb->ipfb_lock);
2486 2487 /*
2487 2488 * Now need to send any icmp messages that we delayed from
2488 2489 * above.
2489 2490 */
2490 2491 while (send_icmp_head_v6 != NULL) {
2491 2492 ip6_t *ip6h;
2492 2493
2493 2494 mp = send_icmp_head_v6;
2494 2495 send_icmp_head_v6 = send_icmp_head_v6->b_next;
2495 2496 mp->b_next = NULL;
2496 2497 ip6h = (ip6_t *)mp->b_rptr;
2497 2498 iras.ira_flags = 0;
2498 2499 /*
2499 2500 * This will result in an incorrect ALL_ZONES zoneid
2500 2501 * for multicast packets, but we
2501 2502 * don't send ICMP errors for those in any case.
2502 2503 */
2503 2504 iras.ira_zoneid =
2504 2505 ipif_lookup_addr_zoneid_v6(&ip6h->ip6_dst,
2505 2506 ill, ipst);
2506 2507 ip_drop_input("ICMP_TIME_EXCEEDED reass", mp, ill);
2507 2508 icmp_time_exceeded_v6(mp,
2508 2509 ICMP_REASSEMBLY_TIME_EXCEEDED, B_FALSE,
2509 2510 &iras);
2510 2511 ASSERT(!(iras.ira_flags & IRAF_IPSEC_SECURE));
2511 2512 }
2512 2513 while (send_icmp_head != NULL) {
2513 2514 ipaddr_t dst;
2514 2515
2515 2516 mp = send_icmp_head;
2516 2517 send_icmp_head = send_icmp_head->b_next;
2517 2518 mp->b_next = NULL;
2518 2519
2519 2520 dst = ((ipha_t *)mp->b_rptr)->ipha_dst;
2520 2521
2521 2522 iras.ira_flags = IRAF_IS_IPV4;
2522 2523 /*
2523 2524 * This will result in an incorrect ALL_ZONES zoneid
2524 2525 * for broadcast and multicast packets, but we
2525 2526 * don't send ICMP errors for those in any case.
2526 2527 */
2527 2528 iras.ira_zoneid = ipif_lookup_addr_zoneid(dst,
2528 2529 ill, ipst);
2529 2530 ip_drop_input("ICMP_TIME_EXCEEDED reass", mp, ill);
2530 2531 icmp_time_exceeded(mp,
2531 2532 ICMP_REASSEMBLY_TIME_EXCEEDED, &iras);
2532 2533 ASSERT(!(iras.ira_flags & IRAF_IPSEC_SECURE));
2533 2534 }
2534 2535 }
2535 2536 /*
2536 2537 * A non-dying ILL will use the return value to decide whether to
2537 2538 * restart the frag timer, and for how long.
2538 2539 */
2539 2540 return (next_timeout);
2540 2541 }
2541 2542
2542 2543 /*
2543 2544 * This routine is called when the approximate count of mblk memory used
2544 2545 * for the specified ILL has exceeded max_count.
2545 2546 */
2546 2547 void
2547 2548 ill_frag_prune(ill_t *ill, uint_t max_count)
2548 2549 {
2549 2550 ipfb_t *ipfb;
2550 2551 ipf_t *ipf;
2551 2552 size_t count;
2552 2553 clock_t now;
2553 2554
2554 2555 /*
2555 2556 * If we are here within ip_min_frag_prune_time msecs remove
2556 2557 * ill_frag_free_num_pkts oldest packets from each bucket and increment
2557 2558 * ill_frag_free_num_pkts.
2558 2559 */
2559 2560 mutex_enter(&ill->ill_lock);
2560 2561 now = ddi_get_lbolt();
2561 2562 if (TICK_TO_MSEC(now - ill->ill_last_frag_clean_time) <=
2562 2563 (ip_min_frag_prune_time != 0 ?
2563 2564 ip_min_frag_prune_time : msec_per_tick)) {
2564 2565
2565 2566 ill->ill_frag_free_num_pkts++;
2566 2567
2567 2568 } else {
2568 2569 ill->ill_frag_free_num_pkts = 0;
2569 2570 }
2570 2571 ill->ill_last_frag_clean_time = now;
2571 2572 mutex_exit(&ill->ill_lock);
2572 2573
2573 2574 /*
2574 2575 * free ill_frag_free_num_pkts oldest packets from each bucket.
2575 2576 */
2576 2577 if (ill->ill_frag_free_num_pkts != 0) {
2577 2578 int ix;
2578 2579
2579 2580 for (ix = 0; ix < ILL_FRAG_HASH_TBL_COUNT; ix++) {
2580 2581 ipfb = &ill->ill_frag_hash_tbl[ix];
2581 2582 mutex_enter(&ipfb->ipfb_lock);
2582 2583 if (ipfb->ipfb_ipf != NULL) {
2583 2584 ill_frag_free_pkts(ill, ipfb, ipfb->ipfb_ipf,
2584 2585 ill->ill_frag_free_num_pkts);
2585 2586 }
2586 2587 mutex_exit(&ipfb->ipfb_lock);
2587 2588 }
2588 2589 }
2589 2590 /*
2590 2591 * While the reassembly list for this ILL is too big, prune a fragment
2591 2592 * queue by age, oldest first.
2592 2593 */
2593 2594 while (ill->ill_frag_count > max_count) {
2594 2595 int ix;
2595 2596 ipfb_t *oipfb = NULL;
2596 2597 uint_t oldest = UINT_MAX;
2597 2598
2598 2599 count = 0;
2599 2600 for (ix = 0; ix < ILL_FRAG_HASH_TBL_COUNT; ix++) {
2600 2601 ipfb = &ill->ill_frag_hash_tbl[ix];
2601 2602 mutex_enter(&ipfb->ipfb_lock);
2602 2603 ipf = ipfb->ipfb_ipf;
2603 2604 if (ipf != NULL && ipf->ipf_gen < oldest) {
2604 2605 oldest = ipf->ipf_gen;
2605 2606 oipfb = ipfb;
2606 2607 }
2607 2608 count += ipfb->ipfb_count;
2608 2609 mutex_exit(&ipfb->ipfb_lock);
2609 2610 }
2610 2611 if (oipfb == NULL)
2611 2612 break;
2612 2613
2613 2614 if (count <= max_count)
2614 2615 return; /* Somebody beat us to it, nothing to do */
2615 2616 mutex_enter(&oipfb->ipfb_lock);
2616 2617 ipf = oipfb->ipfb_ipf;
2617 2618 if (ipf != NULL) {
2618 2619 ill_frag_free_pkts(ill, oipfb, ipf, 1);
2619 2620 }
2620 2621 mutex_exit(&oipfb->ipfb_lock);
2621 2622 }
2622 2623 }
2623 2624
2624 2625 /*
2625 2626 * free 'free_cnt' fragmented packets starting at ipf.
2626 2627 */
2627 2628 void
2628 2629 ill_frag_free_pkts(ill_t *ill, ipfb_t *ipfb, ipf_t *ipf, int free_cnt)
2629 2630 {
2630 2631 size_t count;
2631 2632 mblk_t *mp;
2632 2633 mblk_t *tmp;
2633 2634 ipf_t **ipfp = ipf->ipf_ptphn;
2634 2635
2635 2636 ASSERT(MUTEX_HELD(&ipfb->ipfb_lock));
2636 2637 ASSERT(ipfp != NULL);
2637 2638 ASSERT(ipf != NULL);
2638 2639
2639 2640 while (ipf != NULL && free_cnt-- > 0) {
2640 2641 count = ipf->ipf_count;
2641 2642 mp = ipf->ipf_mp;
2642 2643 ipf = ipf->ipf_hash_next;
2643 2644 for (tmp = mp; tmp; tmp = tmp->b_cont) {
2644 2645 IP_REASS_SET_START(tmp, 0);
2645 2646 IP_REASS_SET_END(tmp, 0);
2646 2647 }
2647 2648 atomic_add_32(&ill->ill_frag_count, -count);
2648 2649 ASSERT(ipfb->ipfb_count >= count);
2649 2650 ipfb->ipfb_count -= count;
2650 2651 ASSERT(ipfb->ipfb_frag_pkts > 0);
2651 2652 ipfb->ipfb_frag_pkts--;
2652 2653 BUMP_MIB(ill->ill_ip_mib, ipIfStatsReasmFails);
2653 2654 ip_drop_input("ipIfStatsReasmFails", mp, ill);
2654 2655 freemsg(mp);
2655 2656 }
2656 2657
2657 2658 if (ipf)
2658 2659 ipf->ipf_ptphn = ipfp;
2659 2660 ipfp[0] = ipf;
2660 2661 }
2661 2662
2662 2663 /*
2663 2664 * Helper function for ill_forward_set().
2664 2665 */
2665 2666 static void
2666 2667 ill_forward_set_on_ill(ill_t *ill, boolean_t enable)
2667 2668 {
2668 2669 ip_stack_t *ipst = ill->ill_ipst;
2669 2670
2670 2671 ASSERT(IAM_WRITER_ILL(ill) || RW_READ_HELD(&ipst->ips_ill_g_lock));
2671 2672
2672 2673 ip1dbg(("ill_forward_set: %s %s forwarding on %s",
2673 2674 (enable ? "Enabling" : "Disabling"),
2674 2675 (ill->ill_isv6 ? "IPv6" : "IPv4"), ill->ill_name));
2675 2676 mutex_enter(&ill->ill_lock);
2676 2677 if (enable)
2677 2678 ill->ill_flags |= ILLF_ROUTER;
2678 2679 else
2679 2680 ill->ill_flags &= ~ILLF_ROUTER;
2680 2681 mutex_exit(&ill->ill_lock);
2681 2682 if (ill->ill_isv6)
2682 2683 ill_set_nce_router_flags(ill, enable);
2683 2684 /* Notify routing socket listeners of this change. */
2684 2685 if (ill->ill_ipif != NULL)
2685 2686 ip_rts_ifmsg(ill->ill_ipif, RTSQ_DEFAULT);
2686 2687 }
2687 2688
2688 2689 /*
2689 2690 * Set an ill's ILLF_ROUTER flag appropriately. Send up RTS_IFINFO routing
2690 2691 * socket messages for each interface whose flags we change.
2691 2692 */
2692 2693 int
2693 2694 ill_forward_set(ill_t *ill, boolean_t enable)
2694 2695 {
2695 2696 ipmp_illgrp_t *illg;
2696 2697 ip_stack_t *ipst = ill->ill_ipst;
2697 2698
2698 2699 ASSERT(IAM_WRITER_ILL(ill) || RW_READ_HELD(&ipst->ips_ill_g_lock));
2699 2700
2700 2701 if ((enable && (ill->ill_flags & ILLF_ROUTER)) ||
2701 2702 (!enable && !(ill->ill_flags & ILLF_ROUTER)))
2702 2703 return (0);
2703 2704
2704 2705 if (IS_LOOPBACK(ill))
2705 2706 return (EINVAL);
2706 2707
2707 2708 if (enable && ill->ill_allowed_ips_cnt > 0)
2708 2709 return (EPERM);
2709 2710
2710 2711 if (IS_IPMP(ill) || IS_UNDER_IPMP(ill)) {
2711 2712 /*
2712 2713 * Update all of the interfaces in the group.
2713 2714 */
2714 2715 illg = ill->ill_grp;
2715 2716 ill = list_head(&illg->ig_if);
2716 2717 for (; ill != NULL; ill = list_next(&illg->ig_if, ill))
2717 2718 ill_forward_set_on_ill(ill, enable);
2718 2719
2719 2720 /*
2720 2721 * Update the IPMP meta-interface.
2721 2722 */
2722 2723 ill_forward_set_on_ill(ipmp_illgrp_ipmp_ill(illg), enable);
2723 2724 return (0);
2724 2725 }
2725 2726
2726 2727 ill_forward_set_on_ill(ill, enable);
2727 2728 return (0);
2728 2729 }
2729 2730
2730 2731 /*
2731 2732 * Based on the ILLF_ROUTER flag of an ill, make sure all local nce's for
2732 2733 * addresses assigned to the ill have the NCE_F_ISROUTER flag appropriately
2733 2734 * set or clear.
2734 2735 */
2735 2736 static void
2736 2737 ill_set_nce_router_flags(ill_t *ill, boolean_t enable)
2737 2738 {
2738 2739 ipif_t *ipif;
2739 2740 ncec_t *ncec;
2740 2741 nce_t *nce;
2741 2742
2742 2743 for (ipif = ill->ill_ipif; ipif != NULL; ipif = ipif->ipif_next) {
2743 2744 /*
2744 2745 * NOTE: we match across the illgrp because nce's for
2745 2746 * addresses on IPMP interfaces have an nce_ill that points to
2746 2747 * the bound underlying ill.
2747 2748 */
2748 2749 nce = nce_lookup_v6(ill, &ipif->ipif_v6lcl_addr);
2749 2750 if (nce != NULL) {
2750 2751 ncec = nce->nce_common;
2751 2752 mutex_enter(&ncec->ncec_lock);
2752 2753 if (enable)
2753 2754 ncec->ncec_flags |= NCE_F_ISROUTER;
2754 2755 else
2755 2756 ncec->ncec_flags &= ~NCE_F_ISROUTER;
2756 2757 mutex_exit(&ncec->ncec_lock);
2757 2758 nce_refrele(nce);
2758 2759 }
2759 2760 }
2760 2761 }
2761 2762
2762 2763 /*
2763 2764 * Intializes the context structure and returns the first ill in the list
2764 2765 * cuurently start_list and end_list can have values:
2765 2766 * MAX_G_HEADS Traverse both IPV4 and IPV6 lists.
2766 2767 * IP_V4_G_HEAD Traverse IPV4 list only.
2767 2768 * IP_V6_G_HEAD Traverse IPV6 list only.
2768 2769 */
2769 2770
2770 2771 /*
2771 2772 * We don't check for CONDEMNED ills here. Caller must do that if
2772 2773 * necessary under the ill lock.
2773 2774 */
2774 2775 ill_t *
2775 2776 ill_first(int start_list, int end_list, ill_walk_context_t *ctx,
2776 2777 ip_stack_t *ipst)
2777 2778 {
2778 2779 ill_if_t *ifp;
2779 2780 ill_t *ill;
2780 2781 avl_tree_t *avl_tree;
2781 2782
2782 2783 ASSERT(RW_LOCK_HELD(&ipst->ips_ill_g_lock));
2783 2784 ASSERT(end_list <= MAX_G_HEADS && start_list >= 0);
2784 2785
2785 2786 /*
2786 2787 * setup the lists to search
2787 2788 */
2788 2789 if (end_list != MAX_G_HEADS) {
2789 2790 ctx->ctx_current_list = start_list;
2790 2791 ctx->ctx_last_list = end_list;
2791 2792 } else {
2792 2793 ctx->ctx_last_list = MAX_G_HEADS - 1;
2793 2794 ctx->ctx_current_list = 0;
2794 2795 }
2795 2796
2796 2797 while (ctx->ctx_current_list <= ctx->ctx_last_list) {
2797 2798 ifp = IP_VX_ILL_G_LIST(ctx->ctx_current_list, ipst);
2798 2799 if (ifp != (ill_if_t *)
2799 2800 &IP_VX_ILL_G_LIST(ctx->ctx_current_list, ipst)) {
2800 2801 avl_tree = &ifp->illif_avl_by_ppa;
2801 2802 ill = avl_first(avl_tree);
2802 2803 /*
2803 2804 * ill is guaranteed to be non NULL or ifp should have
2804 2805 * not existed.
2805 2806 */
2806 2807 ASSERT(ill != NULL);
2807 2808 return (ill);
2808 2809 }
2809 2810 ctx->ctx_current_list++;
2810 2811 }
2811 2812
2812 2813 return (NULL);
2813 2814 }
2814 2815
2815 2816 /*
2816 2817 * returns the next ill in the list. ill_first() must have been called
2817 2818 * before calling ill_next() or bad things will happen.
2818 2819 */
2819 2820
2820 2821 /*
2821 2822 * We don't check for CONDEMNED ills here. Caller must do that if
2822 2823 * necessary under the ill lock.
2823 2824 */
2824 2825 ill_t *
2825 2826 ill_next(ill_walk_context_t *ctx, ill_t *lastill)
2826 2827 {
2827 2828 ill_if_t *ifp;
2828 2829 ill_t *ill;
2829 2830 ip_stack_t *ipst = lastill->ill_ipst;
2830 2831
2831 2832 ASSERT(lastill->ill_ifptr != (ill_if_t *)
2832 2833 &IP_VX_ILL_G_LIST(ctx->ctx_current_list, ipst));
2833 2834 if ((ill = avl_walk(&lastill->ill_ifptr->illif_avl_by_ppa, lastill,
2834 2835 AVL_AFTER)) != NULL) {
2835 2836 return (ill);
2836 2837 }
2837 2838
2838 2839 /* goto next ill_ifp in the list. */
2839 2840 ifp = lastill->ill_ifptr->illif_next;
2840 2841
2841 2842 /* make sure not at end of circular list */
2842 2843 while (ifp ==
2843 2844 (ill_if_t *)&IP_VX_ILL_G_LIST(ctx->ctx_current_list, ipst)) {
2844 2845 if (++ctx->ctx_current_list > ctx->ctx_last_list)
2845 2846 return (NULL);
2846 2847 ifp = IP_VX_ILL_G_LIST(ctx->ctx_current_list, ipst);
2847 2848 }
2848 2849
2849 2850 return (avl_first(&ifp->illif_avl_by_ppa));
2850 2851 }
2851 2852
2852 2853 /*
2853 2854 * Check interface name for correct format: [a-zA-Z]+[a-zA-Z0-9._]*[0-9]+
2854 2855 * The final number (PPA) must not have any leading zeros. Upon success, a
2855 2856 * pointer to the start of the PPA is returned; otherwise NULL is returned.
2856 2857 */
2857 2858 static char *
2858 2859 ill_get_ppa_ptr(char *name)
2859 2860 {
2860 2861 int namelen = strlen(name);
2861 2862 int end_ndx = namelen - 1;
2862 2863 int ppa_ndx, i;
2863 2864
2864 2865 /*
2865 2866 * Check that the first character is [a-zA-Z], and that the last
2866 2867 * character is [0-9].
2867 2868 */
2868 2869 if (namelen == 0 || !isalpha(name[0]) || !isdigit(name[end_ndx]))
2869 2870 return (NULL);
2870 2871
2871 2872 /*
2872 2873 * Set `ppa_ndx' to the PPA start, and check for leading zeroes.
2873 2874 */
2874 2875 for (ppa_ndx = end_ndx; ppa_ndx > 0; ppa_ndx--)
2875 2876 if (!isdigit(name[ppa_ndx - 1]))
2876 2877 break;
2877 2878
2878 2879 if (name[ppa_ndx] == '0' && ppa_ndx < end_ndx)
2879 2880 return (NULL);
2880 2881
2881 2882 /*
2882 2883 * Check that the intermediate characters are [a-z0-9.]
2883 2884 */
2884 2885 for (i = 1; i < ppa_ndx; i++) {
2885 2886 if (!isalpha(name[i]) && !isdigit(name[i]) &&
2886 2887 name[i] != '.' && name[i] != '_') {
2887 2888 return (NULL);
2888 2889 }
2889 2890 }
2890 2891
2891 2892 return (name + ppa_ndx);
2892 2893 }
2893 2894
2894 2895 /*
2895 2896 * use avl tree to locate the ill.
2896 2897 */
2897 2898 static ill_t *
2898 2899 ill_find_by_name(char *name, boolean_t isv6, ip_stack_t *ipst)
2899 2900 {
2900 2901 char *ppa_ptr = NULL;
2901 2902 int len;
2902 2903 uint_t ppa;
2903 2904 ill_t *ill = NULL;
2904 2905 ill_if_t *ifp;
2905 2906 int list;
2906 2907
2907 2908 /*
2908 2909 * get ppa ptr
2909 2910 */
2910 2911 if (isv6)
2911 2912 list = IP_V6_G_HEAD;
2912 2913 else
2913 2914 list = IP_V4_G_HEAD;
2914 2915
2915 2916 if ((ppa_ptr = ill_get_ppa_ptr(name)) == NULL) {
2916 2917 return (NULL);
2917 2918 }
2918 2919
2919 2920 len = ppa_ptr - name + 1;
2920 2921
2921 2922 ppa = stoi(&ppa_ptr);
2922 2923
2923 2924 ifp = IP_VX_ILL_G_LIST(list, ipst);
2924 2925
2925 2926 while (ifp != (ill_if_t *)&IP_VX_ILL_G_LIST(list, ipst)) {
2926 2927 /*
2927 2928 * match is done on len - 1 as the name is not null
2928 2929 * terminated it contains ppa in addition to the interface
2929 2930 * name.
2930 2931 */
2931 2932 if ((ifp->illif_name_len == len) &&
2932 2933 bcmp(ifp->illif_name, name, len - 1) == 0) {
2933 2934 break;
2934 2935 } else {
2935 2936 ifp = ifp->illif_next;
2936 2937 }
2937 2938 }
2938 2939
2939 2940 if (ifp == (ill_if_t *)&IP_VX_ILL_G_LIST(list, ipst)) {
2940 2941 /*
2941 2942 * Even the interface type does not exist.
2942 2943 */
2943 2944 return (NULL);
2944 2945 }
2945 2946
2946 2947 ill = avl_find(&ifp->illif_avl_by_ppa, (void *) &ppa, NULL);
2947 2948 if (ill != NULL) {
2948 2949 mutex_enter(&ill->ill_lock);
2949 2950 if (ILL_CAN_LOOKUP(ill)) {
2950 2951 ill_refhold_locked(ill);
2951 2952 mutex_exit(&ill->ill_lock);
2952 2953 return (ill);
2953 2954 }
2954 2955 mutex_exit(&ill->ill_lock);
2955 2956 }
2956 2957 return (NULL);
2957 2958 }
2958 2959
2959 2960 /*
2960 2961 * comparison function for use with avl.
2961 2962 */
2962 2963 static int
2963 2964 ill_compare_ppa(const void *ppa_ptr, const void *ill_ptr)
2964 2965 {
2965 2966 uint_t ppa;
2966 2967 uint_t ill_ppa;
2967 2968
2968 2969 ASSERT(ppa_ptr != NULL && ill_ptr != NULL);
2969 2970
2970 2971 ppa = *((uint_t *)ppa_ptr);
2971 2972 ill_ppa = ((const ill_t *)ill_ptr)->ill_ppa;
2972 2973 /*
2973 2974 * We want the ill with the lowest ppa to be on the
2974 2975 * top.
2975 2976 */
2976 2977 if (ill_ppa < ppa)
2977 2978 return (1);
2978 2979 if (ill_ppa > ppa)
2979 2980 return (-1);
2980 2981 return (0);
2981 2982 }
2982 2983
2983 2984 /*
2984 2985 * remove an interface type from the global list.
2985 2986 */
2986 2987 static void
2987 2988 ill_delete_interface_type(ill_if_t *interface)
2988 2989 {
2989 2990 ASSERT(interface != NULL);
2990 2991 ASSERT(avl_numnodes(&interface->illif_avl_by_ppa) == 0);
2991 2992
2992 2993 avl_destroy(&interface->illif_avl_by_ppa);
2993 2994 if (interface->illif_ppa_arena != NULL)
2994 2995 vmem_destroy(interface->illif_ppa_arena);
2995 2996
2996 2997 remque(interface);
2997 2998
2998 2999 mi_free(interface);
2999 3000 }
3000 3001
3001 3002 /*
3002 3003 * remove ill from the global list.
3003 3004 */
3004 3005 static void
3005 3006 ill_glist_delete(ill_t *ill)
3006 3007 {
3007 3008 ip_stack_t *ipst;
3008 3009 phyint_t *phyi;
3009 3010
3010 3011 if (ill == NULL)
3011 3012 return;
3012 3013 ipst = ill->ill_ipst;
3013 3014 rw_enter(&ipst->ips_ill_g_lock, RW_WRITER);
3014 3015
3015 3016 /*
3016 3017 * If the ill was never inserted into the AVL tree
3017 3018 * we skip the if branch.
3018 3019 */
3019 3020 if (ill->ill_ifptr != NULL) {
3020 3021 /*
3021 3022 * remove from AVL tree and free ppa number
3022 3023 */
3023 3024 avl_remove(&ill->ill_ifptr->illif_avl_by_ppa, ill);
3024 3025
3025 3026 if (ill->ill_ifptr->illif_ppa_arena != NULL) {
3026 3027 vmem_free(ill->ill_ifptr->illif_ppa_arena,
3027 3028 (void *)(uintptr_t)(ill->ill_ppa+1), 1);
3028 3029 }
3029 3030 if (avl_numnodes(&ill->ill_ifptr->illif_avl_by_ppa) == 0) {
3030 3031 ill_delete_interface_type(ill->ill_ifptr);
3031 3032 }
3032 3033
3033 3034 /*
3034 3035 * Indicate ill is no longer in the list.
3035 3036 */
3036 3037 ill->ill_ifptr = NULL;
3037 3038 ill->ill_name_length = 0;
3038 3039 ill->ill_name[0] = '\0';
3039 3040 ill->ill_ppa = UINT_MAX;
3040 3041 }
3041 3042
3042 3043 /* Generate one last event for this ill. */
3043 3044 ill_nic_event_dispatch(ill, 0, NE_UNPLUMB, ill->ill_name,
3044 3045 ill->ill_name_length);
3045 3046
3046 3047 ASSERT(ill->ill_phyint != NULL);
3047 3048 phyi = ill->ill_phyint;
3048 3049 ill->ill_phyint = NULL;
3049 3050
3050 3051 /*
3051 3052 * ill_init allocates a phyint always to store the copy
3052 3053 * of flags relevant to phyint. At that point in time, we could
3053 3054 * not assign the name and hence phyint_illv4/v6 could not be
3054 3055 * initialized. Later in ipif_set_values, we assign the name to
3055 3056 * the ill, at which point in time we assign phyint_illv4/v6.
3056 3057 * Thus we don't rely on phyint_illv6 to be initialized always.
3057 3058 */
3058 3059 if (ill->ill_flags & ILLF_IPV6)
3059 3060 phyi->phyint_illv6 = NULL;
3060 3061 else
3061 3062 phyi->phyint_illv4 = NULL;
3062 3063
3063 3064 if (phyi->phyint_illv4 != NULL || phyi->phyint_illv6 != NULL) {
3064 3065 rw_exit(&ipst->ips_ill_g_lock);
3065 3066 return;
3066 3067 }
3067 3068
3068 3069 /*
3069 3070 * There are no ills left on this phyint; pull it out of the phyint
3070 3071 * avl trees, and free it.
3071 3072 */
3072 3073 if (phyi->phyint_ifindex > 0) {
3073 3074 avl_remove(&ipst->ips_phyint_g_list->phyint_list_avl_by_index,
3074 3075 phyi);
3075 3076 avl_remove(&ipst->ips_phyint_g_list->phyint_list_avl_by_name,
3076 3077 phyi);
3077 3078 }
3078 3079 rw_exit(&ipst->ips_ill_g_lock);
3079 3080
3080 3081 phyint_free(phyi);
3081 3082 }
3082 3083
3083 3084 /*
3084 3085 * allocate a ppa, if the number of plumbed interfaces of this type are
3085 3086 * less than ill_no_arena do a linear search to find a unused ppa.
3086 3087 * When the number goes beyond ill_no_arena switch to using an arena.
3087 3088 * Note: ppa value of zero cannot be allocated from vmem_arena as it
3088 3089 * is the return value for an error condition, so allocation starts at one
3089 3090 * and is decremented by one.
3090 3091 */
3091 3092 static int
3092 3093 ill_alloc_ppa(ill_if_t *ifp, ill_t *ill)
3093 3094 {
3094 3095 ill_t *tmp_ill;
3095 3096 uint_t start, end;
3096 3097 int ppa;
3097 3098
3098 3099 if (ifp->illif_ppa_arena == NULL &&
3099 3100 (avl_numnodes(&ifp->illif_avl_by_ppa) + 1 > ill_no_arena)) {
3100 3101 /*
3101 3102 * Create an arena.
3102 3103 */
3103 3104 ifp->illif_ppa_arena = vmem_create(ifp->illif_name,
3104 3105 (void *)1, UINT_MAX - 1, 1, NULL, NULL,
3105 3106 NULL, 0, VM_SLEEP | VMC_IDENTIFIER);
3106 3107 /* allocate what has already been assigned */
3107 3108 for (tmp_ill = avl_first(&ifp->illif_avl_by_ppa);
3108 3109 tmp_ill != NULL; tmp_ill = avl_walk(&ifp->illif_avl_by_ppa,
3109 3110 tmp_ill, AVL_AFTER)) {
3110 3111 ppa = (int)(uintptr_t)vmem_xalloc(ifp->illif_ppa_arena,
3111 3112 1, /* size */
3112 3113 1, /* align/quantum */
3113 3114 0, /* phase */
3114 3115 0, /* nocross */
3115 3116 /* minaddr */
3116 3117 (void *)((uintptr_t)tmp_ill->ill_ppa + 1),
3117 3118 /* maxaddr */
3118 3119 (void *)((uintptr_t)tmp_ill->ill_ppa + 2),
3119 3120 VM_NOSLEEP|VM_FIRSTFIT);
3120 3121 if (ppa == 0) {
3121 3122 ip1dbg(("ill_alloc_ppa: ppa allocation"
3122 3123 " failed while switching"));
3123 3124 vmem_destroy(ifp->illif_ppa_arena);
3124 3125 ifp->illif_ppa_arena = NULL;
3125 3126 break;
3126 3127 }
3127 3128 }
3128 3129 }
3129 3130
3130 3131 if (ifp->illif_ppa_arena != NULL) {
3131 3132 if (ill->ill_ppa == UINT_MAX) {
3132 3133 ppa = (int)(uintptr_t)vmem_alloc(ifp->illif_ppa_arena,
3133 3134 1, VM_NOSLEEP|VM_FIRSTFIT);
3134 3135 if (ppa == 0)
3135 3136 return (EAGAIN);
3136 3137 ill->ill_ppa = --ppa;
3137 3138 } else {
3138 3139 ppa = (int)(uintptr_t)vmem_xalloc(ifp->illif_ppa_arena,
3139 3140 1, /* size */
3140 3141 1, /* align/quantum */
3141 3142 0, /* phase */
3142 3143 0, /* nocross */
3143 3144 (void *)(uintptr_t)(ill->ill_ppa + 1), /* minaddr */
3144 3145 (void *)(uintptr_t)(ill->ill_ppa + 2), /* maxaddr */
3145 3146 VM_NOSLEEP|VM_FIRSTFIT);
3146 3147 /*
3147 3148 * Most likely the allocation failed because
3148 3149 * the requested ppa was in use.
3149 3150 */
3150 3151 if (ppa == 0)
3151 3152 return (EEXIST);
3152 3153 }
3153 3154 return (0);
3154 3155 }
3155 3156
3156 3157 /*
3157 3158 * No arena is in use and not enough (>ill_no_arena) interfaces have
3158 3159 * been plumbed to create one. Do a linear search to get a unused ppa.
3159 3160 */
3160 3161 if (ill->ill_ppa == UINT_MAX) {
3161 3162 end = UINT_MAX - 1;
3162 3163 start = 0;
3163 3164 } else {
3164 3165 end = start = ill->ill_ppa;
3165 3166 }
3166 3167
3167 3168 tmp_ill = avl_find(&ifp->illif_avl_by_ppa, (void *)&start, NULL);
3168 3169 while (tmp_ill != NULL && tmp_ill->ill_ppa == start) {
3169 3170 if (start++ >= end) {
3170 3171 if (ill->ill_ppa == UINT_MAX)
3171 3172 return (EAGAIN);
3172 3173 else
3173 3174 return (EEXIST);
3174 3175 }
3175 3176 tmp_ill = avl_walk(&ifp->illif_avl_by_ppa, tmp_ill, AVL_AFTER);
3176 3177 }
3177 3178 ill->ill_ppa = start;
3178 3179 return (0);
3179 3180 }
3180 3181
3181 3182 /*
3182 3183 * Insert ill into the list of configured ill's. Once this function completes,
3183 3184 * the ill is globally visible and is available through lookups. More precisely
3184 3185 * this happens after the caller drops the ill_g_lock.
3185 3186 */
3186 3187 static int
3187 3188 ill_glist_insert(ill_t *ill, char *name, boolean_t isv6)
3188 3189 {
3189 3190 ill_if_t *ill_interface;
3190 3191 avl_index_t where = 0;
3191 3192 int error;
3192 3193 int name_length;
3193 3194 int index;
3194 3195 boolean_t check_length = B_FALSE;
3195 3196 ip_stack_t *ipst = ill->ill_ipst;
3196 3197
3197 3198 ASSERT(RW_WRITE_HELD(&ipst->ips_ill_g_lock));
3198 3199
3199 3200 name_length = mi_strlen(name) + 1;
3200 3201
3201 3202 if (isv6)
3202 3203 index = IP_V6_G_HEAD;
3203 3204 else
3204 3205 index = IP_V4_G_HEAD;
3205 3206
3206 3207 ill_interface = IP_VX_ILL_G_LIST(index, ipst);
3207 3208 /*
3208 3209 * Search for interface type based on name
3209 3210 */
3210 3211 while (ill_interface != (ill_if_t *)&IP_VX_ILL_G_LIST(index, ipst)) {
3211 3212 if ((ill_interface->illif_name_len == name_length) &&
3212 3213 (strcmp(ill_interface->illif_name, name) == 0)) {
3213 3214 break;
3214 3215 }
3215 3216 ill_interface = ill_interface->illif_next;
3216 3217 }
3217 3218
3218 3219 /*
3219 3220 * Interface type not found, create one.
3220 3221 */
3221 3222 if (ill_interface == (ill_if_t *)&IP_VX_ILL_G_LIST(index, ipst)) {
3222 3223 ill_g_head_t ghead;
3223 3224
3224 3225 /*
3225 3226 * allocate ill_if_t structure
3226 3227 */
3227 3228 ill_interface = (ill_if_t *)mi_zalloc(sizeof (ill_if_t));
3228 3229 if (ill_interface == NULL) {
3229 3230 return (ENOMEM);
3230 3231 }
3231 3232
3232 3233 (void) strcpy(ill_interface->illif_name, name);
3233 3234 ill_interface->illif_name_len = name_length;
3234 3235
3235 3236 avl_create(&ill_interface->illif_avl_by_ppa,
3236 3237 ill_compare_ppa, sizeof (ill_t),
3237 3238 offsetof(struct ill_s, ill_avl_byppa));
3238 3239
3239 3240 /*
3240 3241 * link the structure in the back to maintain order
3241 3242 * of configuration for ifconfig output.
3242 3243 */
3243 3244 ghead = ipst->ips_ill_g_heads[index];
3244 3245 insque(ill_interface, ghead.ill_g_list_tail);
3245 3246 }
3246 3247
3247 3248 if (ill->ill_ppa == UINT_MAX)
3248 3249 check_length = B_TRUE;
3249 3250
3250 3251 error = ill_alloc_ppa(ill_interface, ill);
3251 3252 if (error != 0) {
3252 3253 if (avl_numnodes(&ill_interface->illif_avl_by_ppa) == 0)
3253 3254 ill_delete_interface_type(ill->ill_ifptr);
3254 3255 return (error);
3255 3256 }
3256 3257
3257 3258 /*
3258 3259 * When the ppa is choosen by the system, check that there is
3259 3260 * enough space to insert ppa. if a specific ppa was passed in this
3260 3261 * check is not required as the interface name passed in will have
3261 3262 * the right ppa in it.
3262 3263 */
3263 3264 if (check_length) {
3264 3265 /*
3265 3266 * UINT_MAX - 1 should fit in 10 chars, alloc 12 chars.
3266 3267 */
3267 3268 char buf[sizeof (uint_t) * 3];
3268 3269
3269 3270 /*
3270 3271 * convert ppa to string to calculate the amount of space
3271 3272 * required for it in the name.
3272 3273 */
3273 3274 numtos(ill->ill_ppa, buf);
3274 3275
3275 3276 /* Do we have enough space to insert ppa ? */
3276 3277
3277 3278 if ((mi_strlen(name) + mi_strlen(buf) + 1) > LIFNAMSIZ) {
3278 3279 /* Free ppa and interface type struct */
3279 3280 if (ill_interface->illif_ppa_arena != NULL) {
3280 3281 vmem_free(ill_interface->illif_ppa_arena,
3281 3282 (void *)(uintptr_t)(ill->ill_ppa+1), 1);
3282 3283 }
3283 3284 if (avl_numnodes(&ill_interface->illif_avl_by_ppa) == 0)
3284 3285 ill_delete_interface_type(ill->ill_ifptr);
3285 3286
3286 3287 return (EINVAL);
3287 3288 }
3288 3289 }
3289 3290
3290 3291 (void) sprintf(ill->ill_name, "%s%u", name, ill->ill_ppa);
3291 3292 ill->ill_name_length = mi_strlen(ill->ill_name) + 1;
3292 3293
3293 3294 (void) avl_find(&ill_interface->illif_avl_by_ppa, &ill->ill_ppa,
3294 3295 &where);
3295 3296 ill->ill_ifptr = ill_interface;
3296 3297 avl_insert(&ill_interface->illif_avl_by_ppa, ill, where);
3297 3298
3298 3299 ill_phyint_reinit(ill);
3299 3300 return (0);
3300 3301 }
3301 3302
3302 3303 /* Initialize the per phyint ipsq used for serialization */
3303 3304 static boolean_t
3304 3305 ipsq_init(ill_t *ill, boolean_t enter)
3305 3306 {
3306 3307 ipsq_t *ipsq;
3307 3308 ipxop_t *ipx;
3308 3309
3309 3310 if ((ipsq = kmem_zalloc(sizeof (ipsq_t), KM_NOSLEEP)) == NULL)
3310 3311 return (B_FALSE);
3311 3312
3312 3313 ill->ill_phyint->phyint_ipsq = ipsq;
3313 3314 ipx = ipsq->ipsq_xop = &ipsq->ipsq_ownxop;
3314 3315 ipx->ipx_ipsq = ipsq;
3315 3316 ipsq->ipsq_next = ipsq;
3316 3317 ipsq->ipsq_phyint = ill->ill_phyint;
3317 3318 mutex_init(&ipsq->ipsq_lock, NULL, MUTEX_DEFAULT, 0);
3318 3319 mutex_init(&ipx->ipx_lock, NULL, MUTEX_DEFAULT, 0);
3319 3320 ipsq->ipsq_ipst = ill->ill_ipst; /* No netstack_hold */
3320 3321 if (enter) {
3321 3322 ipx->ipx_writer = curthread;
3322 3323 ipx->ipx_forced = B_FALSE;
3323 3324 ipx->ipx_reentry_cnt = 1;
3324 3325 #ifdef DEBUG
3325 3326 ipx->ipx_depth = getpcstack(ipx->ipx_stack, IPX_STACK_DEPTH);
3326 3327 #endif
3327 3328 }
3328 3329 return (B_TRUE);
3329 3330 }
3330 3331
3331 3332 /*
3332 3333 * ill_init is called by ip_open when a device control stream is opened.
3333 3334 * It does a few initializations, and shoots a DL_INFO_REQ message down
3334 3335 * to the driver. The response is later picked up in ip_rput_dlpi and
3335 3336 * used to set up default mechanisms for talking to the driver. (Always
3336 3337 * called as writer.)
3337 3338 *
3338 3339 * If this function returns error, ip_open will call ip_close which in
3339 3340 * turn will call ill_delete to clean up any memory allocated here that
3340 3341 * is not yet freed.
3341 3342 */
3342 3343 int
3343 3344 ill_init(queue_t *q, ill_t *ill)
3344 3345 {
3345 3346 int count;
3346 3347 dl_info_req_t *dlir;
3347 3348 mblk_t *info_mp;
3348 3349 uchar_t *frag_ptr;
3349 3350
3350 3351 /*
3351 3352 * The ill is initialized to zero by mi_alloc*(). In addition
3352 3353 * some fields already contain valid values, initialized in
3353 3354 * ip_open(), before we reach here.
3354 3355 */
3355 3356 mutex_init(&ill->ill_lock, NULL, MUTEX_DEFAULT, 0);
3356 3357 mutex_init(&ill->ill_saved_ire_lock, NULL, MUTEX_DEFAULT, NULL);
3357 3358 ill->ill_saved_ire_cnt = 0;
3358 3359
3359 3360 ill->ill_rq = q;
3360 3361 ill->ill_wq = WR(q);
3361 3362
3362 3363 info_mp = allocb(MAX(sizeof (dl_info_req_t), sizeof (dl_info_ack_t)),
3363 3364 BPRI_HI);
3364 3365 if (info_mp == NULL)
3365 3366 return (ENOMEM);
3366 3367
3367 3368 /*
3368 3369 * Allocate sufficient space to contain our fragment hash table and
3369 3370 * the device name.
3370 3371 */
3371 3372 frag_ptr = (uchar_t *)mi_zalloc(ILL_FRAG_HASH_TBL_SIZE + 2 * LIFNAMSIZ);
3372 3373 if (frag_ptr == NULL) {
3373 3374 freemsg(info_mp);
3374 3375 return (ENOMEM);
3375 3376 }
3376 3377 ill->ill_frag_ptr = frag_ptr;
3377 3378 ill->ill_frag_free_num_pkts = 0;
3378 3379 ill->ill_last_frag_clean_time = 0;
3379 3380 ill->ill_frag_hash_tbl = (ipfb_t *)frag_ptr;
3380 3381 ill->ill_name = (char *)(frag_ptr + ILL_FRAG_HASH_TBL_SIZE);
3381 3382 for (count = 0; count < ILL_FRAG_HASH_TBL_COUNT; count++) {
3382 3383 mutex_init(&ill->ill_frag_hash_tbl[count].ipfb_lock,
3383 3384 NULL, MUTEX_DEFAULT, NULL);
3384 3385 }
3385 3386
3386 3387 ill->ill_phyint = (phyint_t *)mi_zalloc(sizeof (phyint_t));
3387 3388 if (ill->ill_phyint == NULL) {
3388 3389 freemsg(info_mp);
3389 3390 mi_free(frag_ptr);
3390 3391 return (ENOMEM);
3391 3392 }
3392 3393
3393 3394 mutex_init(&ill->ill_phyint->phyint_lock, NULL, MUTEX_DEFAULT, 0);
3394 3395 /*
3395 3396 * For now pretend this is a v4 ill. We need to set phyint_ill*
3396 3397 * at this point because of the following reason. If we can't
3397 3398 * enter the ipsq at some point and cv_wait, the writer that
3398 3399 * wakes us up tries to locate us using the list of all phyints
3399 3400 * in an ipsq and the ills from the phyint thru the phyint_ill*.
3400 3401 * If we don't set it now, we risk a missed wakeup.
3401 3402 */
3402 3403 ill->ill_phyint->phyint_illv4 = ill;
3403 3404 ill->ill_ppa = UINT_MAX;
3404 3405 list_create(&ill->ill_nce, sizeof (nce_t), offsetof(nce_t, nce_node));
3405 3406
3406 3407 ill_set_inputfn(ill);
3407 3408
3408 3409 if (!ipsq_init(ill, B_TRUE)) {
3409 3410 freemsg(info_mp);
3410 3411 mi_free(frag_ptr);
3411 3412 mi_free(ill->ill_phyint);
3412 3413 return (ENOMEM);
3413 3414 }
3414 3415
3415 3416 ill->ill_state_flags |= ILL_LL_SUBNET_PENDING;
3416 3417
3417 3418 /* Frag queue limit stuff */
3418 3419 ill->ill_frag_count = 0;
3419 3420 ill->ill_ipf_gen = 0;
3420 3421
3421 3422 rw_init(&ill->ill_mcast_lock, NULL, RW_DEFAULT, NULL);
3422 3423 mutex_init(&ill->ill_mcast_serializer, NULL, MUTEX_DEFAULT, NULL);
3423 3424 ill->ill_global_timer = INFINITY;
3424 3425 ill->ill_mcast_v1_time = ill->ill_mcast_v2_time = 0;
3425 3426 ill->ill_mcast_v1_tset = ill->ill_mcast_v2_tset = 0;
3426 3427 ill->ill_mcast_rv = MCAST_DEF_ROBUSTNESS;
3427 3428 ill->ill_mcast_qi = MCAST_DEF_QUERY_INTERVAL;
3428 3429
3429 3430 /*
3430 3431 * Initialize IPv6 configuration variables. The IP module is always
3431 3432 * opened as an IPv4 module. Instead tracking down the cases where
3432 3433 * it switches to do ipv6, we'll just initialize the IPv6 configuration
3433 3434 * here for convenience, this has no effect until the ill is set to do
3434 3435 * IPv6.
3435 3436 */
3436 3437 ill->ill_reachable_time = ND_REACHABLE_TIME;
3437 3438 ill->ill_xmit_count = ND_MAX_MULTICAST_SOLICIT;
3438 3439 ill->ill_max_buf = ND_MAX_Q;
3439 3440 ill->ill_refcnt = 0;
3440 3441
3441 3442 /* Send down the Info Request to the driver. */
3442 3443 info_mp->b_datap->db_type = M_PCPROTO;
3443 3444 dlir = (dl_info_req_t *)info_mp->b_rptr;
3444 3445 info_mp->b_wptr = (uchar_t *)&dlir[1];
3445 3446 dlir->dl_primitive = DL_INFO_REQ;
3446 3447
3447 3448 ill->ill_dlpi_pending = DL_PRIM_INVAL;
3448 3449
3449 3450 qprocson(q);
3450 3451 ill_dlpi_send(ill, info_mp);
3451 3452
3452 3453 return (0);
3453 3454 }
3454 3455
3455 3456 /*
3456 3457 * ill_dls_info
3457 3458 * creates datalink socket info from the device.
3458 3459 */
3459 3460 int
3460 3461 ill_dls_info(struct sockaddr_dl *sdl, const ill_t *ill)
3461 3462 {
3462 3463 size_t len;
3463 3464
3464 3465 sdl->sdl_family = AF_LINK;
3465 3466 sdl->sdl_index = ill_get_upper_ifindex(ill);
3466 3467 sdl->sdl_type = ill->ill_type;
3467 3468 ill_get_name(ill, sdl->sdl_data, sizeof (sdl->sdl_data));
3468 3469 len = strlen(sdl->sdl_data);
3469 3470 ASSERT(len < 256);
3470 3471 sdl->sdl_nlen = (uchar_t)len;
3471 3472 sdl->sdl_alen = ill->ill_phys_addr_length;
3472 3473 sdl->sdl_slen = 0;
3473 3474 if (ill->ill_phys_addr_length != 0 && ill->ill_phys_addr != NULL)
3474 3475 bcopy(ill->ill_phys_addr, &sdl->sdl_data[len], sdl->sdl_alen);
3475 3476
3476 3477 return (sizeof (struct sockaddr_dl));
3477 3478 }
3478 3479
3479 3480 /*
3480 3481 * ill_xarp_info
3481 3482 * creates xarp info from the device.
3482 3483 */
3483 3484 static int
3484 3485 ill_xarp_info(struct sockaddr_dl *sdl, ill_t *ill)
3485 3486 {
3486 3487 sdl->sdl_family = AF_LINK;
3487 3488 sdl->sdl_index = ill->ill_phyint->phyint_ifindex;
3488 3489 sdl->sdl_type = ill->ill_type;
3489 3490 ill_get_name(ill, sdl->sdl_data, sizeof (sdl->sdl_data));
3490 3491 sdl->sdl_nlen = (uchar_t)mi_strlen(sdl->sdl_data);
3491 3492 sdl->sdl_alen = ill->ill_phys_addr_length;
3492 3493 sdl->sdl_slen = 0;
3493 3494 return (sdl->sdl_nlen);
3494 3495 }
3495 3496
3496 3497 static int
3497 3498 loopback_kstat_update(kstat_t *ksp, int rw)
3498 3499 {
3499 3500 kstat_named_t *kn;
3500 3501 netstackid_t stackid;
3501 3502 netstack_t *ns;
3502 3503 ip_stack_t *ipst;
3503 3504
3504 3505 if (ksp == NULL || ksp->ks_data == NULL)
3505 3506 return (EIO);
3506 3507
3507 3508 if (rw == KSTAT_WRITE)
3508 3509 return (EACCES);
3509 3510
3510 3511 kn = KSTAT_NAMED_PTR(ksp);
3511 3512 stackid = (zoneid_t)(uintptr_t)ksp->ks_private;
3512 3513
3513 3514 ns = netstack_find_by_stackid(stackid);
3514 3515 if (ns == NULL)
3515 3516 return (-1);
3516 3517
3517 3518 ipst = ns->netstack_ip;
3518 3519 if (ipst == NULL) {
3519 3520 netstack_rele(ns);
3520 3521 return (-1);
3521 3522 }
3522 3523 kn[0].value.ui32 = ipst->ips_loopback_packets;
3523 3524 kn[1].value.ui32 = ipst->ips_loopback_packets;
3524 3525 netstack_rele(ns);
3525 3526 return (0);
3526 3527 }
3527 3528
3528 3529 /*
3529 3530 * Has ifindex been plumbed already?
3530 3531 */
3531 3532 static boolean_t
3532 3533 phyint_exists(uint_t index, ip_stack_t *ipst)
3533 3534 {
3534 3535 ASSERT(index != 0);
3535 3536 ASSERT(RW_LOCK_HELD(&ipst->ips_ill_g_lock));
3536 3537
3537 3538 return (avl_find(&ipst->ips_phyint_g_list->phyint_list_avl_by_index,
3538 3539 &index, NULL) != NULL);
3539 3540 }
3540 3541
3541 3542 /*
3542 3543 * Pick a unique ifindex.
3543 3544 * When the index counter passes IF_INDEX_MAX for the first time, the wrap
3544 3545 * flag is set so that next time time ip_assign_ifindex() is called, it
3545 3546 * falls through and resets the index counter back to 1, the minimum value
3546 3547 * for the interface index. The logic below assumes that ips_ill_index
3547 3548 * can hold a value of IF_INDEX_MAX+1 without there being any loss
3548 3549 * (i.e. reset back to 0.)
3549 3550 */
3550 3551 boolean_t
3551 3552 ip_assign_ifindex(uint_t *indexp, ip_stack_t *ipst)
3552 3553 {
3553 3554 uint_t loops;
3554 3555
3555 3556 if (!ipst->ips_ill_index_wrap) {
3556 3557 *indexp = ipst->ips_ill_index++;
3557 3558 if (ipst->ips_ill_index > IF_INDEX_MAX) {
3558 3559 /*
3559 3560 * Reached the maximum ifindex value, set the wrap
3560 3561 * flag to indicate that it is no longer possible
3561 3562 * to assume that a given index is unallocated.
3562 3563 */
3563 3564 ipst->ips_ill_index_wrap = B_TRUE;
3564 3565 }
3565 3566 return (B_TRUE);
3566 3567 }
3567 3568
3568 3569 if (ipst->ips_ill_index > IF_INDEX_MAX)
3569 3570 ipst->ips_ill_index = 1;
3570 3571
3571 3572 /*
3572 3573 * Start reusing unused indexes. Note that we hold the ill_g_lock
3573 3574 * at this point and don't want to call any function that attempts
3574 3575 * to get the lock again.
3575 3576 */
3576 3577 for (loops = IF_INDEX_MAX; loops > 0; loops--) {
3577 3578 if (!phyint_exists(ipst->ips_ill_index, ipst)) {
3578 3579 /* found unused index - use it */
3579 3580 *indexp = ipst->ips_ill_index;
3580 3581 return (B_TRUE);
3581 3582 }
3582 3583
3583 3584 ipst->ips_ill_index++;
3584 3585 if (ipst->ips_ill_index > IF_INDEX_MAX)
3585 3586 ipst->ips_ill_index = 1;
3586 3587 }
3587 3588
3588 3589 /*
3589 3590 * all interface indicies are inuse.
3590 3591 */
3591 3592 return (B_FALSE);
3592 3593 }
3593 3594
3594 3595 /*
3595 3596 * Assign a unique interface index for the phyint.
3596 3597 */
3597 3598 static boolean_t
3598 3599 phyint_assign_ifindex(phyint_t *phyi, ip_stack_t *ipst)
3599 3600 {
3600 3601 ASSERT(phyi->phyint_ifindex == 0);
3601 3602 return (ip_assign_ifindex(&phyi->phyint_ifindex, ipst));
3602 3603 }
3603 3604
3604 3605 /*
3605 3606 * Initialize the flags on `phyi' as per the provided mactype.
3606 3607 */
3607 3608 static void
3608 3609 phyint_flags_init(phyint_t *phyi, t_uscalar_t mactype)
3609 3610 {
3610 3611 uint64_t flags = 0;
3611 3612
3612 3613 /*
3613 3614 * Initialize PHYI_RUNNING and PHYI_FAILED. For non-IPMP interfaces,
3614 3615 * we always presume the underlying hardware is working and set
3615 3616 * PHYI_RUNNING (if it's not, the driver will subsequently send a
3616 3617 * DL_NOTE_LINK_DOWN message). For IPMP interfaces, at initialization
3617 3618 * there are no active interfaces in the group so we set PHYI_FAILED.
3618 3619 */
3619 3620 if (mactype == SUNW_DL_IPMP)
3620 3621 flags |= PHYI_FAILED;
3621 3622 else
3622 3623 flags |= PHYI_RUNNING;
3623 3624
3624 3625 switch (mactype) {
3625 3626 case SUNW_DL_VNI:
3626 3627 flags |= PHYI_VIRTUAL;
3627 3628 break;
3628 3629 case SUNW_DL_IPMP:
3629 3630 flags |= PHYI_IPMP;
3630 3631 break;
3631 3632 case DL_LOOP:
3632 3633 flags |= (PHYI_LOOPBACK | PHYI_VIRTUAL);
3633 3634 break;
3634 3635 }
3635 3636
3636 3637 mutex_enter(&phyi->phyint_lock);
3637 3638 phyi->phyint_flags |= flags;
3638 3639 mutex_exit(&phyi->phyint_lock);
3639 3640 }
3640 3641
3641 3642 /*
3642 3643 * Return a pointer to the ill which matches the supplied name. Note that
3643 3644 * the ill name length includes the null termination character. (May be
3644 3645 * called as writer.)
3645 3646 * If do_alloc and the interface is "lo0" it will be automatically created.
3646 3647 * Cannot bump up reference on condemned ills. So dup detect can't be done
3647 3648 * using this func.
3648 3649 */
3649 3650 ill_t *
3650 3651 ill_lookup_on_name(char *name, boolean_t do_alloc, boolean_t isv6,
3651 3652 boolean_t *did_alloc, ip_stack_t *ipst)
3652 3653 {
3653 3654 ill_t *ill;
3654 3655 ipif_t *ipif;
3655 3656 ipsq_t *ipsq;
3656 3657 kstat_named_t *kn;
3657 3658 boolean_t isloopback;
3658 3659 in6_addr_t ov6addr;
3659 3660
3660 3661 isloopback = mi_strcmp(name, ipif_loopback_name) == 0;
3661 3662
3662 3663 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
3663 3664 ill = ill_find_by_name(name, isv6, ipst);
3664 3665 rw_exit(&ipst->ips_ill_g_lock);
3665 3666 if (ill != NULL)
3666 3667 return (ill);
3667 3668
3668 3669 /*
3669 3670 * Couldn't find it. Does this happen to be a lookup for the
3670 3671 * loopback device and are we allowed to allocate it?
3671 3672 */
3672 3673 if (!isloopback || !do_alloc)
3673 3674 return (NULL);
3674 3675
3675 3676 rw_enter(&ipst->ips_ill_g_lock, RW_WRITER);
3676 3677 ill = ill_find_by_name(name, isv6, ipst);
3677 3678 if (ill != NULL) {
3678 3679 rw_exit(&ipst->ips_ill_g_lock);
3679 3680 return (ill);
3680 3681 }
3681 3682
3682 3683 /* Create the loopback device on demand */
3683 3684 ill = (ill_t *)(mi_alloc(sizeof (ill_t) +
3684 3685 sizeof (ipif_loopback_name), BPRI_MED));
3685 3686 if (ill == NULL)
3686 3687 goto done;
3687 3688
3688 3689 *ill = ill_null;
3689 3690 mutex_init(&ill->ill_lock, NULL, MUTEX_DEFAULT, NULL);
3690 3691 ill->ill_ipst = ipst;
3691 3692 list_create(&ill->ill_nce, sizeof (nce_t), offsetof(nce_t, nce_node));
3692 3693 netstack_hold(ipst->ips_netstack);
3693 3694 /*
3694 3695 * For exclusive stacks we set the zoneid to zero
3695 3696 * to make IP operate as if in the global zone.
3696 3697 */
3697 3698 ill->ill_zoneid = GLOBAL_ZONEID;
3698 3699
3699 3700 ill->ill_phyint = (phyint_t *)mi_zalloc(sizeof (phyint_t));
3700 3701 if (ill->ill_phyint == NULL)
3701 3702 goto done;
3702 3703
3703 3704 if (isv6)
3704 3705 ill->ill_phyint->phyint_illv6 = ill;
3705 3706 else
3706 3707 ill->ill_phyint->phyint_illv4 = ill;
3707 3708 mutex_init(&ill->ill_phyint->phyint_lock, NULL, MUTEX_DEFAULT, 0);
3708 3709 phyint_flags_init(ill->ill_phyint, DL_LOOP);
3709 3710
3710 3711 if (isv6) {
3711 3712 ill->ill_isv6 = B_TRUE;
3712 3713 ill->ill_max_frag = ip_loopback_mtu_v6plus;
3713 3714 } else {
3714 3715 ill->ill_max_frag = ip_loopback_mtuplus;
3715 3716 }
3716 3717 if (!ill_allocate_mibs(ill))
3717 3718 goto done;
3718 3719 ill->ill_current_frag = ill->ill_max_frag;
3719 3720 ill->ill_mtu = ill->ill_max_frag; /* Initial value */
3720 3721 ill->ill_mc_mtu = ill->ill_mtu;
3721 3722 /*
3722 3723 * ipif_loopback_name can't be pointed at directly because its used
3723 3724 * by both the ipv4 and ipv6 interfaces. When the ill is removed
3724 3725 * from the glist, ill_glist_delete() sets the first character of
3725 3726 * ill_name to '\0'.
3726 3727 */
3727 3728 ill->ill_name = (char *)ill + sizeof (*ill);
3728 3729 (void) strcpy(ill->ill_name, ipif_loopback_name);
3729 3730 ill->ill_name_length = sizeof (ipif_loopback_name);
3730 3731 /* Set ill_dlpi_pending for ipsq_current_finish() to work properly */
3731 3732 ill->ill_dlpi_pending = DL_PRIM_INVAL;
3732 3733
3733 3734 rw_init(&ill->ill_mcast_lock, NULL, RW_DEFAULT, NULL);
3734 3735 mutex_init(&ill->ill_mcast_serializer, NULL, MUTEX_DEFAULT, NULL);
3735 3736 ill->ill_global_timer = INFINITY;
3736 3737 ill->ill_mcast_v1_time = ill->ill_mcast_v2_time = 0;
3737 3738 ill->ill_mcast_v1_tset = ill->ill_mcast_v2_tset = 0;
3738 3739 ill->ill_mcast_rv = MCAST_DEF_ROBUSTNESS;
3739 3740 ill->ill_mcast_qi = MCAST_DEF_QUERY_INTERVAL;
3740 3741
3741 3742 /* No resolver here. */
3742 3743 ill->ill_net_type = IRE_LOOPBACK;
3743 3744
3744 3745 /* Initialize the ipsq */
3745 3746 if (!ipsq_init(ill, B_FALSE))
3746 3747 goto done;
3747 3748
3748 3749 ipif = ipif_allocate(ill, 0L, IRE_LOOPBACK, B_TRUE, B_TRUE, NULL);
3749 3750 if (ipif == NULL)
3750 3751 goto done;
3751 3752
3752 3753 ill->ill_flags = ILLF_MULTICAST;
3753 3754
3754 3755 ov6addr = ipif->ipif_v6lcl_addr;
3755 3756 /* Set up default loopback address and mask. */
3756 3757 if (!isv6) {
3757 3758 ipaddr_t inaddr_loopback = htonl(INADDR_LOOPBACK);
3758 3759
3759 3760 IN6_IPADDR_TO_V4MAPPED(inaddr_loopback, &ipif->ipif_v6lcl_addr);
3760 3761 V4MASK_TO_V6(htonl(IN_CLASSA_NET), ipif->ipif_v6net_mask);
3761 3762 V6_MASK_COPY(ipif->ipif_v6lcl_addr, ipif->ipif_v6net_mask,
3762 3763 ipif->ipif_v6subnet);
3763 3764 ill->ill_flags |= ILLF_IPV4;
3764 3765 } else {
3765 3766 ipif->ipif_v6lcl_addr = ipv6_loopback;
3766 3767 ipif->ipif_v6net_mask = ipv6_all_ones;
3767 3768 V6_MASK_COPY(ipif->ipif_v6lcl_addr, ipif->ipif_v6net_mask,
3768 3769 ipif->ipif_v6subnet);
3769 3770 ill->ill_flags |= ILLF_IPV6;
3770 3771 }
3771 3772
3772 3773 /*
3773 3774 * Chain us in at the end of the ill list. hold the ill
3774 3775 * before we make it globally visible. 1 for the lookup.
3775 3776 */
3776 3777 ill->ill_refcnt = 0;
3777 3778 ill_refhold(ill);
3778 3779
3779 3780 ill->ill_frag_count = 0;
3780 3781 ill->ill_frag_free_num_pkts = 0;
3781 3782 ill->ill_last_frag_clean_time = 0;
3782 3783
3783 3784 ipsq = ill->ill_phyint->phyint_ipsq;
3784 3785
3785 3786 ill_set_inputfn(ill);
3786 3787
3787 3788 if (ill_glist_insert(ill, "lo", isv6) != 0)
3788 3789 cmn_err(CE_PANIC, "cannot insert loopback interface");
3789 3790
3790 3791 /* Let SCTP know so that it can add this to its list */
3791 3792 sctp_update_ill(ill, SCTP_ILL_INSERT);
3792 3793
3793 3794 /*
3794 3795 * We have already assigned ipif_v6lcl_addr above, but we need to
3795 3796 * call sctp_update_ipif_addr() after SCTP_ILL_INSERT, which
3796 3797 * requires to be after ill_glist_insert() since we need the
3797 3798 * ill_index set. Pass on ipv6_loopback as the old address.
3798 3799 */
3799 3800 sctp_update_ipif_addr(ipif, ov6addr);
3800 3801
3801 3802 ip_rts_newaddrmsg(RTM_CHGADDR, 0, ipif, RTSQ_DEFAULT);
3802 3803
3803 3804 /*
3804 3805 * ill_glist_insert() -> ill_phyint_reinit() may have merged IPSQs.
3805 3806 * If so, free our original one.
3806 3807 */
3807 3808 if (ipsq != ill->ill_phyint->phyint_ipsq)
3808 3809 ipsq_delete(ipsq);
3809 3810
3810 3811 if (ipst->ips_loopback_ksp == NULL) {
3811 3812 /* Export loopback interface statistics */
3812 3813 ipst->ips_loopback_ksp = kstat_create_netstack("lo", 0,
3813 3814 ipif_loopback_name, "net",
3814 3815 KSTAT_TYPE_NAMED, 2, 0,
3815 3816 ipst->ips_netstack->netstack_stackid);
3816 3817 if (ipst->ips_loopback_ksp != NULL) {
3817 3818 ipst->ips_loopback_ksp->ks_update =
3818 3819 loopback_kstat_update;
3819 3820 kn = KSTAT_NAMED_PTR(ipst->ips_loopback_ksp);
3820 3821 kstat_named_init(&kn[0], "ipackets", KSTAT_DATA_UINT32);
3821 3822 kstat_named_init(&kn[1], "opackets", KSTAT_DATA_UINT32);
3822 3823 ipst->ips_loopback_ksp->ks_private =
3823 3824 (void *)(uintptr_t)ipst->ips_netstack->
3824 3825 netstack_stackid;
3825 3826 kstat_install(ipst->ips_loopback_ksp);
3826 3827 }
3827 3828 }
3828 3829
3829 3830 *did_alloc = B_TRUE;
3830 3831 rw_exit(&ipst->ips_ill_g_lock);
3831 3832 ill_nic_event_dispatch(ill, MAP_IPIF_ID(ill->ill_ipif->ipif_id),
3832 3833 NE_PLUMB, ill->ill_name, ill->ill_name_length);
3833 3834 return (ill);
3834 3835 done:
3835 3836 if (ill != NULL) {
3836 3837 if (ill->ill_phyint != NULL) {
3837 3838 ipsq = ill->ill_phyint->phyint_ipsq;
3838 3839 if (ipsq != NULL) {
3839 3840 ipsq->ipsq_phyint = NULL;
3840 3841 ipsq_delete(ipsq);
3841 3842 }
3842 3843 mi_free(ill->ill_phyint);
3843 3844 }
3844 3845 ill_free_mib(ill);
3845 3846 if (ill->ill_ipst != NULL)
3846 3847 netstack_rele(ill->ill_ipst->ips_netstack);
3847 3848 mi_free(ill);
3848 3849 }
3849 3850 rw_exit(&ipst->ips_ill_g_lock);
3850 3851 return (NULL);
3851 3852 }
3852 3853
3853 3854 /*
3854 3855 * For IPP calls - use the ip_stack_t for global stack.
3855 3856 */
3856 3857 ill_t *
3857 3858 ill_lookup_on_ifindex_global_instance(uint_t index, boolean_t isv6)
3858 3859 {
3859 3860 ip_stack_t *ipst;
3860 3861 ill_t *ill;
3861 3862
3862 3863 ipst = netstack_find_by_stackid(GLOBAL_NETSTACKID)->netstack_ip;
3863 3864 if (ipst == NULL) {
3864 3865 cmn_err(CE_WARN, "No ip_stack_t for zoneid zero!\n");
3865 3866 return (NULL);
3866 3867 }
3867 3868
3868 3869 ill = ill_lookup_on_ifindex(index, isv6, ipst);
3869 3870 netstack_rele(ipst->ips_netstack);
3870 3871 return (ill);
3871 3872 }
3872 3873
3873 3874 /*
3874 3875 * Return a pointer to the ill which matches the index and IP version type.
3875 3876 */
3876 3877 ill_t *
3877 3878 ill_lookup_on_ifindex(uint_t index, boolean_t isv6, ip_stack_t *ipst)
3878 3879 {
3879 3880 ill_t *ill;
3880 3881 phyint_t *phyi;
3881 3882
3882 3883 /*
3883 3884 * Indexes are stored in the phyint - a common structure
3884 3885 * to both IPv4 and IPv6.
3885 3886 */
3886 3887 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
3887 3888 phyi = avl_find(&ipst->ips_phyint_g_list->phyint_list_avl_by_index,
3888 3889 (void *) &index, NULL);
3889 3890 if (phyi != NULL) {
3890 3891 ill = isv6 ? phyi->phyint_illv6: phyi->phyint_illv4;
3891 3892 if (ill != NULL) {
3892 3893 mutex_enter(&ill->ill_lock);
3893 3894 if (!ILL_IS_CONDEMNED(ill)) {
3894 3895 ill_refhold_locked(ill);
3895 3896 mutex_exit(&ill->ill_lock);
3896 3897 rw_exit(&ipst->ips_ill_g_lock);
3897 3898 return (ill);
3898 3899 }
3899 3900 mutex_exit(&ill->ill_lock);
3900 3901 }
3901 3902 }
3902 3903 rw_exit(&ipst->ips_ill_g_lock);
3903 3904 return (NULL);
3904 3905 }
3905 3906
3906 3907 /*
3907 3908 * Verify whether or not an interface index is valid for the specified zoneid
3908 3909 * to transmit packets.
3909 3910 * It can be zero (meaning "reset") or an interface index assigned
3910 3911 * to a non-VNI interface. (We don't use VNI interface to send packets.)
3911 3912 */
3912 3913 boolean_t
3913 3914 ip_xmit_ifindex_valid(uint_t ifindex, zoneid_t zoneid, boolean_t isv6,
3914 3915 ip_stack_t *ipst)
3915 3916 {
3916 3917 ill_t *ill;
3917 3918
3918 3919 if (ifindex == 0)
3919 3920 return (B_TRUE);
3920 3921
3921 3922 ill = ill_lookup_on_ifindex_zoneid(ifindex, zoneid, isv6, ipst);
3922 3923 if (ill == NULL)
3923 3924 return (B_FALSE);
3924 3925 if (IS_VNI(ill)) {
3925 3926 ill_refrele(ill);
3926 3927 return (B_FALSE);
3927 3928 }
3928 3929 ill_refrele(ill);
3929 3930 return (B_TRUE);
3930 3931 }
3931 3932
3932 3933 /*
3933 3934 * Return the ifindex next in sequence after the passed in ifindex.
3934 3935 * If there is no next ifindex for the given protocol, return 0.
3935 3936 */
3936 3937 uint_t
3937 3938 ill_get_next_ifindex(uint_t index, boolean_t isv6, ip_stack_t *ipst)
3938 3939 {
3939 3940 phyint_t *phyi;
3940 3941 phyint_t *phyi_initial;
3941 3942 uint_t ifindex;
3942 3943
3943 3944 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
3944 3945
3945 3946 if (index == 0) {
3946 3947 phyi = avl_first(
3947 3948 &ipst->ips_phyint_g_list->phyint_list_avl_by_index);
3948 3949 } else {
3949 3950 phyi = phyi_initial = avl_find(
3950 3951 &ipst->ips_phyint_g_list->phyint_list_avl_by_index,
3951 3952 (void *) &index, NULL);
3952 3953 }
3953 3954
3954 3955 for (; phyi != NULL;
3955 3956 phyi = avl_walk(&ipst->ips_phyint_g_list->phyint_list_avl_by_index,
3956 3957 phyi, AVL_AFTER)) {
3957 3958 /*
3958 3959 * If we're not returning the first interface in the tree
3959 3960 * and we still haven't moved past the phyint_t that
3960 3961 * corresponds to index, avl_walk needs to be called again
3961 3962 */
3962 3963 if (!((index != 0) && (phyi == phyi_initial))) {
3963 3964 if (isv6) {
3964 3965 if ((phyi->phyint_illv6) &&
3965 3966 ILL_CAN_LOOKUP(phyi->phyint_illv6) &&
3966 3967 (phyi->phyint_illv6->ill_isv6 == 1))
3967 3968 break;
3968 3969 } else {
3969 3970 if ((phyi->phyint_illv4) &&
3970 3971 ILL_CAN_LOOKUP(phyi->phyint_illv4) &&
3971 3972 (phyi->phyint_illv4->ill_isv6 == 0))
3972 3973 break;
3973 3974 }
3974 3975 }
3975 3976 }
3976 3977
3977 3978 rw_exit(&ipst->ips_ill_g_lock);
3978 3979
3979 3980 if (phyi != NULL)
3980 3981 ifindex = phyi->phyint_ifindex;
3981 3982 else
3982 3983 ifindex = 0;
3983 3984
3984 3985 return (ifindex);
3985 3986 }
3986 3987
3987 3988 /*
3988 3989 * Return the ifindex for the named interface.
3989 3990 * If there is no next ifindex for the interface, return 0.
3990 3991 */
3991 3992 uint_t
3992 3993 ill_get_ifindex_by_name(char *name, ip_stack_t *ipst)
3993 3994 {
3994 3995 phyint_t *phyi;
3995 3996 avl_index_t where = 0;
3996 3997 uint_t ifindex;
3997 3998
3998 3999 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
3999 4000
4000 4001 if ((phyi = avl_find(&ipst->ips_phyint_g_list->phyint_list_avl_by_name,
4001 4002 name, &where)) == NULL) {
4002 4003 rw_exit(&ipst->ips_ill_g_lock);
4003 4004 return (0);
4004 4005 }
4005 4006
4006 4007 ifindex = phyi->phyint_ifindex;
4007 4008
4008 4009 rw_exit(&ipst->ips_ill_g_lock);
4009 4010
4010 4011 return (ifindex);
4011 4012 }
4012 4013
4013 4014 /*
4014 4015 * Return the ifindex to be used by upper layer protocols for instance
4015 4016 * for IPV6_RECVPKTINFO. If IPMP this is the one for the upper ill.
4016 4017 */
4017 4018 uint_t
4018 4019 ill_get_upper_ifindex(const ill_t *ill)
4019 4020 {
4020 4021 if (IS_UNDER_IPMP(ill))
4021 4022 return (ipmp_ill_get_ipmp_ifindex(ill));
4022 4023 else
4023 4024 return (ill->ill_phyint->phyint_ifindex);
4024 4025 }
4025 4026
4026 4027
4027 4028 /*
4028 4029 * Obtain a reference to the ill. The ill_refcnt is a dynamic refcnt
4029 4030 * that gives a running thread a reference to the ill. This reference must be
4030 4031 * released by the thread when it is done accessing the ill and related
4031 4032 * objects. ill_refcnt can not be used to account for static references
4032 4033 * such as other structures pointing to an ill. Callers must generally
4033 4034 * check whether an ill can be refheld by using ILL_CAN_LOOKUP macros
4034 4035 * or be sure that the ill is not being deleted or changing state before
4035 4036 * calling the refhold functions. A non-zero ill_refcnt ensures that the
4036 4037 * ill won't change any of its critical state such as address, netmask etc.
4037 4038 */
4038 4039 void
4039 4040 ill_refhold(ill_t *ill)
4040 4041 {
4041 4042 mutex_enter(&ill->ill_lock);
4042 4043 ill->ill_refcnt++;
4043 4044 ILL_TRACE_REF(ill);
4044 4045 mutex_exit(&ill->ill_lock);
4045 4046 }
4046 4047
4047 4048 void
4048 4049 ill_refhold_locked(ill_t *ill)
4049 4050 {
4050 4051 ASSERT(MUTEX_HELD(&ill->ill_lock));
4051 4052 ill->ill_refcnt++;
4052 4053 ILL_TRACE_REF(ill);
4053 4054 }
4054 4055
4055 4056 /* Returns true if we managed to get a refhold */
4056 4057 boolean_t
4057 4058 ill_check_and_refhold(ill_t *ill)
4058 4059 {
4059 4060 mutex_enter(&ill->ill_lock);
4060 4061 if (!ILL_IS_CONDEMNED(ill)) {
4061 4062 ill_refhold_locked(ill);
4062 4063 mutex_exit(&ill->ill_lock);
4063 4064 return (B_TRUE);
4064 4065 }
4065 4066 mutex_exit(&ill->ill_lock);
4066 4067 return (B_FALSE);
4067 4068 }
4068 4069
4069 4070 /*
4070 4071 * Must not be called while holding any locks. Otherwise if this is
4071 4072 * the last reference to be released, there is a chance of recursive mutex
4072 4073 * panic due to ill_refrele -> ipif_ill_refrele_tail -> qwriter_ip trying
4073 4074 * to restart an ioctl.
4074 4075 */
4075 4076 void
4076 4077 ill_refrele(ill_t *ill)
4077 4078 {
4078 4079 mutex_enter(&ill->ill_lock);
4079 4080 ASSERT(ill->ill_refcnt != 0);
4080 4081 ill->ill_refcnt--;
4081 4082 ILL_UNTRACE_REF(ill);
4082 4083 if (ill->ill_refcnt != 0) {
4083 4084 /* Every ire pointing to the ill adds 1 to ill_refcnt */
4084 4085 mutex_exit(&ill->ill_lock);
4085 4086 return;
4086 4087 }
4087 4088
4088 4089 /* Drops the ill_lock */
4089 4090 ipif_ill_refrele_tail(ill);
4090 4091 }
4091 4092
4092 4093 /*
4093 4094 * Obtain a weak reference count on the ill. This reference ensures the
4094 4095 * ill won't be freed, but the ill may change any of its critical state
4095 4096 * such as netmask, address etc. Returns an error if the ill has started
4096 4097 * closing.
4097 4098 */
4098 4099 boolean_t
4099 4100 ill_waiter_inc(ill_t *ill)
4100 4101 {
4101 4102 mutex_enter(&ill->ill_lock);
4102 4103 if (ill->ill_state_flags & ILL_CONDEMNED) {
4103 4104 mutex_exit(&ill->ill_lock);
4104 4105 return (B_FALSE);
4105 4106 }
4106 4107 ill->ill_waiters++;
4107 4108 mutex_exit(&ill->ill_lock);
4108 4109 return (B_TRUE);
4109 4110 }
4110 4111
4111 4112 void
4112 4113 ill_waiter_dcr(ill_t *ill)
4113 4114 {
4114 4115 mutex_enter(&ill->ill_lock);
4115 4116 ill->ill_waiters--;
4116 4117 if (ill->ill_waiters == 0)
4117 4118 cv_broadcast(&ill->ill_cv);
4118 4119 mutex_exit(&ill->ill_lock);
4119 4120 }
4120 4121
4121 4122 /*
4122 4123 * ip_ll_subnet_defaults is called when we get the DL_INFO_ACK back from the
4123 4124 * driver. We construct best guess defaults for lower level information that
4124 4125 * we need. If an interface is brought up without injection of any overriding
4125 4126 * information from outside, we have to be ready to go with these defaults.
4126 4127 * When we get the first DL_INFO_ACK (from ip_open() sending a DL_INFO_REQ)
4127 4128 * we primarely want the dl_provider_style.
4128 4129 * The subsequent DL_INFO_ACK is received after doing a DL_ATTACH and DL_BIND
4129 4130 * at which point we assume the other part of the information is valid.
4130 4131 */
4131 4132 void
4132 4133 ip_ll_subnet_defaults(ill_t *ill, mblk_t *mp)
4133 4134 {
4134 4135 uchar_t *brdcst_addr;
4135 4136 uint_t brdcst_addr_length, phys_addr_length;
4136 4137 t_scalar_t sap_length;
4137 4138 dl_info_ack_t *dlia;
4138 4139 ip_m_t *ipm;
4139 4140 dl_qos_cl_sel1_t *sel1;
4140 4141 int min_mtu;
4141 4142
4142 4143 ASSERT(IAM_WRITER_ILL(ill));
4143 4144
4144 4145 /*
4145 4146 * Till the ill is fully up the ill is not globally visible.
4146 4147 * So no need for a lock.
4147 4148 */
4148 4149 dlia = (dl_info_ack_t *)mp->b_rptr;
4149 4150 ill->ill_mactype = dlia->dl_mac_type;
4150 4151
4151 4152 ipm = ip_m_lookup(dlia->dl_mac_type);
4152 4153 if (ipm == NULL) {
4153 4154 ipm = ip_m_lookup(DL_OTHER);
4154 4155 ASSERT(ipm != NULL);
4155 4156 }
4156 4157 ill->ill_media = ipm;
4157 4158
4158 4159 /*
4159 4160 * When the new DLPI stuff is ready we'll pull lengths
4160 4161 * from dlia.
4161 4162 */
4162 4163 if (dlia->dl_version == DL_VERSION_2) {
4163 4164 brdcst_addr_length = dlia->dl_brdcst_addr_length;
4164 4165 brdcst_addr = mi_offset_param(mp, dlia->dl_brdcst_addr_offset,
4165 4166 brdcst_addr_length);
4166 4167 if (brdcst_addr == NULL) {
4167 4168 brdcst_addr_length = 0;
4168 4169 }
4169 4170 sap_length = dlia->dl_sap_length;
4170 4171 phys_addr_length = dlia->dl_addr_length - ABS(sap_length);
4171 4172 ip1dbg(("ip: bcast_len %d, sap_len %d, phys_len %d\n",
4172 4173 brdcst_addr_length, sap_length, phys_addr_length));
4173 4174 } else {
4174 4175 brdcst_addr_length = 6;
4175 4176 brdcst_addr = ip_six_byte_all_ones;
4176 4177 sap_length = -2;
4177 4178 phys_addr_length = brdcst_addr_length;
4178 4179 }
4179 4180
4180 4181 ill->ill_bcast_addr_length = brdcst_addr_length;
4181 4182 ill->ill_phys_addr_length = phys_addr_length;
4182 4183 ill->ill_sap_length = sap_length;
4183 4184
4184 4185 /*
4185 4186 * Synthetic DLPI types such as SUNW_DL_IPMP specify a zero SDU,
4186 4187 * but we must ensure a minimum IP MTU is used since other bits of
4187 4188 * IP will fly apart otherwise.
4188 4189 */
4189 4190 min_mtu = ill->ill_isv6 ? IPV6_MIN_MTU : IP_MIN_MTU;
4190 4191 ill->ill_max_frag = MAX(min_mtu, dlia->dl_max_sdu);
4191 4192 ill->ill_current_frag = ill->ill_max_frag;
4192 4193 ill->ill_mtu = ill->ill_max_frag;
4193 4194 ill->ill_mc_mtu = ill->ill_mtu; /* Overridden by DL_NOTE_SDU_SIZE2 */
4194 4195
4195 4196 ill->ill_type = ipm->ip_m_type;
4196 4197
4197 4198 if (!ill->ill_dlpi_style_set) {
4198 4199 if (dlia->dl_provider_style == DL_STYLE2)
4199 4200 ill->ill_needs_attach = 1;
4200 4201
4201 4202 phyint_flags_init(ill->ill_phyint, ill->ill_mactype);
4202 4203
4203 4204 /*
4204 4205 * Allocate the first ipif on this ill. We don't delay it
4205 4206 * further as ioctl handling assumes at least one ipif exists.
4206 4207 *
4207 4208 * At this point we don't know whether the ill is v4 or v6.
4208 4209 * We will know this whan the SIOCSLIFNAME happens and
4209 4210 * the correct value for ill_isv6 will be assigned in
4210 4211 * ipif_set_values(). We need to hold the ill lock and
4211 4212 * clear the ILL_LL_SUBNET_PENDING flag and atomically do
4212 4213 * the wakeup.
4213 4214 */
4214 4215 (void) ipif_allocate(ill, 0, IRE_LOCAL,
4215 4216 dlia->dl_provider_style != DL_STYLE2, B_TRUE, NULL);
4216 4217 mutex_enter(&ill->ill_lock);
4217 4218 ASSERT(ill->ill_dlpi_style_set == 0);
4218 4219 ill->ill_dlpi_style_set = 1;
4219 4220 ill->ill_state_flags &= ~ILL_LL_SUBNET_PENDING;
4220 4221 cv_broadcast(&ill->ill_cv);
4221 4222 mutex_exit(&ill->ill_lock);
4222 4223 freemsg(mp);
4223 4224 return;
4224 4225 }
4225 4226 ASSERT(ill->ill_ipif != NULL);
4226 4227 /*
4227 4228 * We know whether it is IPv4 or IPv6 now, as this is the
4228 4229 * second DL_INFO_ACK we are recieving in response to the
4229 4230 * DL_INFO_REQ sent in ipif_set_values.
4230 4231 */
4231 4232 ill->ill_sap = (ill->ill_isv6) ? ipm->ip_m_ipv6sap : ipm->ip_m_ipv4sap;
4232 4233 /*
4233 4234 * Clear all the flags that were set based on ill_bcast_addr_length
4234 4235 * and ill_phys_addr_length (in ipif_set_values) as these could have
4235 4236 * changed now and we need to re-evaluate.
4236 4237 */
4237 4238 ill->ill_flags &= ~(ILLF_MULTICAST | ILLF_NONUD | ILLF_NOARP);
4238 4239 ill->ill_ipif->ipif_flags &= ~(IPIF_BROADCAST | IPIF_POINTOPOINT);
4239 4240
4240 4241 /*
4241 4242 * Free ill_bcast_mp as things could have changed now.
4242 4243 *
4243 4244 * NOTE: The IPMP meta-interface is special-cased because it starts
4244 4245 * with no underlying interfaces (and thus an unknown broadcast
4245 4246 * address length), but we enforce that an interface is broadcast-
4246 4247 * capable as part of allowing it to join a group.
4247 4248 */
4248 4249 if (ill->ill_bcast_addr_length == 0 && !IS_IPMP(ill)) {
4249 4250 if (ill->ill_bcast_mp != NULL)
4250 4251 freemsg(ill->ill_bcast_mp);
4251 4252 ill->ill_net_type = IRE_IF_NORESOLVER;
4252 4253
4253 4254 ill->ill_bcast_mp = ill_dlur_gen(NULL,
4254 4255 ill->ill_phys_addr_length,
4255 4256 ill->ill_sap,
4256 4257 ill->ill_sap_length);
4257 4258
4258 4259 if (ill->ill_isv6)
4259 4260 /*
4260 4261 * Note: xresolv interfaces will eventually need NOARP
4261 4262 * set here as well, but that will require those
4262 4263 * external resolvers to have some knowledge of
4263 4264 * that flag and act appropriately. Not to be changed
4264 4265 * at present.
4265 4266 */
4266 4267 ill->ill_flags |= ILLF_NONUD;
4267 4268 else
4268 4269 ill->ill_flags |= ILLF_NOARP;
4269 4270
4270 4271 if (ill->ill_mactype == SUNW_DL_VNI) {
4271 4272 ill->ill_ipif->ipif_flags |= IPIF_NOXMIT;
4272 4273 } else if (ill->ill_phys_addr_length == 0 ||
4273 4274 ill->ill_mactype == DL_IPV4 ||
4274 4275 ill->ill_mactype == DL_IPV6) {
4275 4276 /*
4276 4277 * The underying link is point-to-point, so mark the
4277 4278 * interface as such. We can do IP multicast over
4278 4279 * such a link since it transmits all network-layer
4279 4280 * packets to the remote side the same way.
4280 4281 */
4281 4282 ill->ill_flags |= ILLF_MULTICAST;
4282 4283 ill->ill_ipif->ipif_flags |= IPIF_POINTOPOINT;
4283 4284 }
4284 4285 } else {
4285 4286 ill->ill_net_type = IRE_IF_RESOLVER;
4286 4287 if (ill->ill_bcast_mp != NULL)
4287 4288 freemsg(ill->ill_bcast_mp);
4288 4289 ill->ill_bcast_mp = ill_dlur_gen(brdcst_addr,
4289 4290 ill->ill_bcast_addr_length, ill->ill_sap,
4290 4291 ill->ill_sap_length);
4291 4292 /*
4292 4293 * Later detect lack of DLPI driver multicast
4293 4294 * capability by catching DL_ENABMULTI errors in
4294 4295 * ip_rput_dlpi.
4295 4296 */
4296 4297 ill->ill_flags |= ILLF_MULTICAST;
4297 4298 if (!ill->ill_isv6)
4298 4299 ill->ill_ipif->ipif_flags |= IPIF_BROADCAST;
4299 4300 }
4300 4301
4301 4302 /* For IPMP, PHYI_IPMP should already be set by phyint_flags_init() */
4302 4303 if (ill->ill_mactype == SUNW_DL_IPMP)
4303 4304 ASSERT(ill->ill_phyint->phyint_flags & PHYI_IPMP);
4304 4305
4305 4306 /* By default an interface does not support any CoS marking */
4306 4307 ill->ill_flags &= ~ILLF_COS_ENABLED;
4307 4308
4308 4309 /*
4309 4310 * If we get QoS information in DL_INFO_ACK, the device supports
4310 4311 * some form of CoS marking, set ILLF_COS_ENABLED.
4311 4312 */
4312 4313 sel1 = (dl_qos_cl_sel1_t *)mi_offset_param(mp, dlia->dl_qos_offset,
4313 4314 dlia->dl_qos_length);
4314 4315 if ((sel1 != NULL) && (sel1->dl_qos_type == DL_QOS_CL_SEL1)) {
4315 4316 ill->ill_flags |= ILLF_COS_ENABLED;
4316 4317 }
4317 4318
4318 4319 /* Clear any previous error indication. */
4319 4320 ill->ill_error = 0;
4320 4321 freemsg(mp);
4321 4322 }
4322 4323
4323 4324 /*
4324 4325 * Perform various checks to verify that an address would make sense as a
4325 4326 * local, remote, or subnet interface address.
4326 4327 */
4327 4328 static boolean_t
4328 4329 ip_addr_ok_v4(ipaddr_t addr, ipaddr_t subnet_mask)
4329 4330 {
4330 4331 ipaddr_t net_mask;
4331 4332
4332 4333 /*
4333 4334 * Don't allow all zeroes, or all ones, but allow
4334 4335 * all ones netmask.
4335 4336 */
4336 4337 if ((net_mask = ip_net_mask(addr)) == 0)
4337 4338 return (B_FALSE);
4338 4339 /* A given netmask overrides the "guess" netmask */
4339 4340 if (subnet_mask != 0)
4340 4341 net_mask = subnet_mask;
4341 4342 if ((net_mask != ~(ipaddr_t)0) && ((addr == (addr & net_mask)) ||
4342 4343 (addr == (addr | ~net_mask)))) {
4343 4344 return (B_FALSE);
4344 4345 }
4345 4346
4346 4347 /*
4347 4348 * Even if the netmask is all ones, we do not allow address to be
4348 4349 * 255.255.255.255
4349 4350 */
4350 4351 if (addr == INADDR_BROADCAST)
4351 4352 return (B_FALSE);
4352 4353
4353 4354 if (CLASSD(addr))
4354 4355 return (B_FALSE);
4355 4356
4356 4357 return (B_TRUE);
4357 4358 }
4358 4359
4359 4360 #define V6_IPIF_LINKLOCAL(p) \
4360 4361 IN6_IS_ADDR_LINKLOCAL(&(p)->ipif_v6lcl_addr)
4361 4362
4362 4363 /*
4363 4364 * Compare two given ipifs and check if the second one is better than
4364 4365 * the first one using the order of preference (not taking deprecated
4365 4366 * into acount) specified in ipif_lookup_multicast().
4366 4367 */
4367 4368 static boolean_t
4368 4369 ipif_comp_multi(ipif_t *old_ipif, ipif_t *new_ipif, boolean_t isv6)
4369 4370 {
4370 4371 /* Check the least preferred first. */
4371 4372 if (IS_LOOPBACK(old_ipif->ipif_ill)) {
4372 4373 /* If both ipifs are the same, use the first one. */
4373 4374 if (IS_LOOPBACK(new_ipif->ipif_ill))
4374 4375 return (B_FALSE);
4375 4376 else
4376 4377 return (B_TRUE);
4377 4378 }
4378 4379
4379 4380 /* For IPv6, check for link local address. */
4380 4381 if (isv6 && V6_IPIF_LINKLOCAL(old_ipif)) {
4381 4382 if (IS_LOOPBACK(new_ipif->ipif_ill) ||
4382 4383 V6_IPIF_LINKLOCAL(new_ipif)) {
4383 4384 /* The second one is equal or less preferred. */
4384 4385 return (B_FALSE);
4385 4386 } else {
4386 4387 return (B_TRUE);
4387 4388 }
4388 4389 }
4389 4390
4390 4391 /* Then check for point to point interface. */
4391 4392 if (old_ipif->ipif_flags & IPIF_POINTOPOINT) {
4392 4393 if (IS_LOOPBACK(new_ipif->ipif_ill) ||
4393 4394 (isv6 && V6_IPIF_LINKLOCAL(new_ipif)) ||
4394 4395 (new_ipif->ipif_flags & IPIF_POINTOPOINT)) {
4395 4396 return (B_FALSE);
4396 4397 } else {
4397 4398 return (B_TRUE);
4398 4399 }
4399 4400 }
4400 4401
4401 4402 /* old_ipif is a normal interface, so no need to use the new one. */
4402 4403 return (B_FALSE);
4403 4404 }
4404 4405
4405 4406 /*
4406 4407 * Find a mulitcast-capable ipif given an IP instance and zoneid.
4407 4408 * The ipif must be up, and its ill must multicast-capable, not
4408 4409 * condemned, not an underlying interface in an IPMP group, and
4409 4410 * not a VNI interface. Order of preference:
4410 4411 *
4411 4412 * 1a. normal
4412 4413 * 1b. normal, but deprecated
4413 4414 * 2a. point to point
4414 4415 * 2b. point to point, but deprecated
4415 4416 * 3a. link local
4416 4417 * 3b. link local, but deprecated
4417 4418 * 4. loopback.
4418 4419 */
4419 4420 static ipif_t *
4420 4421 ipif_lookup_multicast(ip_stack_t *ipst, zoneid_t zoneid, boolean_t isv6)
4421 4422 {
4422 4423 ill_t *ill;
4423 4424 ill_walk_context_t ctx;
4424 4425 ipif_t *ipif;
4425 4426 ipif_t *saved_ipif = NULL;
4426 4427 ipif_t *dep_ipif = NULL;
4427 4428
4428 4429 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
4429 4430 if (isv6)
4430 4431 ill = ILL_START_WALK_V6(&ctx, ipst);
4431 4432 else
4432 4433 ill = ILL_START_WALK_V4(&ctx, ipst);
4433 4434
4434 4435 for (; ill != NULL; ill = ill_next(&ctx, ill)) {
4435 4436 mutex_enter(&ill->ill_lock);
4436 4437 if (IS_VNI(ill) || IS_UNDER_IPMP(ill) ||
4437 4438 ILL_IS_CONDEMNED(ill) ||
4438 4439 !(ill->ill_flags & ILLF_MULTICAST)) {
4439 4440 mutex_exit(&ill->ill_lock);
4440 4441 continue;
4441 4442 }
4442 4443 for (ipif = ill->ill_ipif; ipif != NULL;
4443 4444 ipif = ipif->ipif_next) {
4444 4445 if (zoneid != ipif->ipif_zoneid &&
4445 4446 zoneid != ALL_ZONES &&
4446 4447 ipif->ipif_zoneid != ALL_ZONES) {
4447 4448 continue;
4448 4449 }
4449 4450 if (!(ipif->ipif_flags & IPIF_UP) ||
4450 4451 IPIF_IS_CONDEMNED(ipif)) {
4451 4452 continue;
4452 4453 }
4453 4454
4454 4455 /*
4455 4456 * Found one candidate. If it is deprecated,
4456 4457 * remember it in dep_ipif. If it is not deprecated,
4457 4458 * remember it in saved_ipif.
4458 4459 */
4459 4460 if (ipif->ipif_flags & IPIF_DEPRECATED) {
4460 4461 if (dep_ipif == NULL) {
4461 4462 dep_ipif = ipif;
4462 4463 } else if (ipif_comp_multi(dep_ipif, ipif,
4463 4464 isv6)) {
4464 4465 /*
4465 4466 * If the previous dep_ipif does not
4466 4467 * belong to the same ill, we've done
4467 4468 * a ipif_refhold() on it. So we need
4468 4469 * to release it.
4469 4470 */
4470 4471 if (dep_ipif->ipif_ill != ill)
4471 4472 ipif_refrele(dep_ipif);
4472 4473 dep_ipif = ipif;
4473 4474 }
4474 4475 continue;
4475 4476 }
4476 4477 if (saved_ipif == NULL) {
4477 4478 saved_ipif = ipif;
4478 4479 } else {
4479 4480 if (ipif_comp_multi(saved_ipif, ipif, isv6)) {
4480 4481 if (saved_ipif->ipif_ill != ill)
4481 4482 ipif_refrele(saved_ipif);
4482 4483 saved_ipif = ipif;
4483 4484 }
4484 4485 }
4485 4486 }
4486 4487 /*
4487 4488 * Before going to the next ill, do a ipif_refhold() on the
4488 4489 * saved ones.
4489 4490 */
4490 4491 if (saved_ipif != NULL && saved_ipif->ipif_ill == ill)
4491 4492 ipif_refhold_locked(saved_ipif);
4492 4493 if (dep_ipif != NULL && dep_ipif->ipif_ill == ill)
4493 4494 ipif_refhold_locked(dep_ipif);
4494 4495 mutex_exit(&ill->ill_lock);
4495 4496 }
4496 4497 rw_exit(&ipst->ips_ill_g_lock);
4497 4498
4498 4499 /*
4499 4500 * If we have only the saved_ipif, return it. But if we have both
4500 4501 * saved_ipif and dep_ipif, check to see which one is better.
4501 4502 */
4502 4503 if (saved_ipif != NULL) {
4503 4504 if (dep_ipif != NULL) {
4504 4505 if (ipif_comp_multi(saved_ipif, dep_ipif, isv6)) {
4505 4506 ipif_refrele(saved_ipif);
4506 4507 return (dep_ipif);
4507 4508 } else {
4508 4509 ipif_refrele(dep_ipif);
4509 4510 return (saved_ipif);
4510 4511 }
4511 4512 }
4512 4513 return (saved_ipif);
4513 4514 } else {
4514 4515 return (dep_ipif);
4515 4516 }
4516 4517 }
4517 4518
4518 4519 ill_t *
4519 4520 ill_lookup_multicast(ip_stack_t *ipst, zoneid_t zoneid, boolean_t isv6)
4520 4521 {
4521 4522 ipif_t *ipif;
4522 4523 ill_t *ill;
4523 4524
4524 4525 ipif = ipif_lookup_multicast(ipst, zoneid, isv6);
4525 4526 if (ipif == NULL)
4526 4527 return (NULL);
4527 4528
4528 4529 ill = ipif->ipif_ill;
4529 4530 ill_refhold(ill);
4530 4531 ipif_refrele(ipif);
4531 4532 return (ill);
4532 4533 }
4533 4534
4534 4535 /*
4535 4536 * This function is called when an application does not specify an interface
4536 4537 * to be used for multicast traffic (joining a group/sending data). It
4537 4538 * calls ire_lookup_multi() to look for an interface route for the
4538 4539 * specified multicast group. Doing this allows the administrator to add
4539 4540 * prefix routes for multicast to indicate which interface to be used for
4540 4541 * multicast traffic in the above scenario. The route could be for all
4541 4542 * multicast (224.0/4), for a single multicast group (a /32 route) or
4542 4543 * anything in between. If there is no such multicast route, we just find
4543 4544 * any multicast capable interface and return it. The returned ipif
4544 4545 * is refhold'ed.
4545 4546 *
4546 4547 * We support MULTIRT and RTF_SETSRC on the multicast routes added to the
4547 4548 * unicast table. This is used by CGTP.
4548 4549 */
4549 4550 ill_t *
4550 4551 ill_lookup_group_v4(ipaddr_t group, zoneid_t zoneid, ip_stack_t *ipst,
4551 4552 boolean_t *multirtp, ipaddr_t *setsrcp)
4552 4553 {
4553 4554 ill_t *ill;
4554 4555
4555 4556 ill = ire_lookup_multi_ill_v4(group, zoneid, ipst, multirtp, setsrcp);
4556 4557 if (ill != NULL)
4557 4558 return (ill);
4558 4559
4559 4560 return (ill_lookup_multicast(ipst, zoneid, B_FALSE));
4560 4561 }
4561 4562
4562 4563 /*
4563 4564 * Look for an ipif with the specified interface address and destination.
4564 4565 * The destination address is used only for matching point-to-point interfaces.
4565 4566 */
4566 4567 ipif_t *
4567 4568 ipif_lookup_interface(ipaddr_t if_addr, ipaddr_t dst, ip_stack_t *ipst)
4568 4569 {
4569 4570 ipif_t *ipif;
4570 4571 ill_t *ill;
4571 4572 ill_walk_context_t ctx;
4572 4573
4573 4574 /*
4574 4575 * First match all the point-to-point interfaces
4575 4576 * before looking at non-point-to-point interfaces.
4576 4577 * This is done to avoid returning non-point-to-point
4577 4578 * ipif instead of unnumbered point-to-point ipif.
4578 4579 */
4579 4580 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
4580 4581 ill = ILL_START_WALK_V4(&ctx, ipst);
4581 4582 for (; ill != NULL; ill = ill_next(&ctx, ill)) {
4582 4583 mutex_enter(&ill->ill_lock);
4583 4584 for (ipif = ill->ill_ipif; ipif != NULL;
4584 4585 ipif = ipif->ipif_next) {
4585 4586 /* Allow the ipif to be down */
4586 4587 if ((ipif->ipif_flags & IPIF_POINTOPOINT) &&
4587 4588 (ipif->ipif_lcl_addr == if_addr) &&
4588 4589 (ipif->ipif_pp_dst_addr == dst)) {
4589 4590 if (!IPIF_IS_CONDEMNED(ipif)) {
4590 4591 ipif_refhold_locked(ipif);
4591 4592 mutex_exit(&ill->ill_lock);
4592 4593 rw_exit(&ipst->ips_ill_g_lock);
4593 4594 return (ipif);
4594 4595 }
4595 4596 }
4596 4597 }
4597 4598 mutex_exit(&ill->ill_lock);
4598 4599 }
4599 4600 rw_exit(&ipst->ips_ill_g_lock);
4600 4601
4601 4602 /* lookup the ipif based on interface address */
4602 4603 ipif = ipif_lookup_addr(if_addr, NULL, ALL_ZONES, ipst);
4603 4604 ASSERT(ipif == NULL || !ipif->ipif_isv6);
4604 4605 return (ipif);
4605 4606 }
4606 4607
4607 4608 /*
4608 4609 * Common function for ipif_lookup_addr() and ipif_lookup_addr_exact().
4609 4610 */
4610 4611 static ipif_t *
4611 4612 ipif_lookup_addr_common(ipaddr_t addr, ill_t *match_ill, uint32_t match_flags,
4612 4613 zoneid_t zoneid, ip_stack_t *ipst)
4613 4614 {
4614 4615 ipif_t *ipif;
4615 4616 ill_t *ill;
4616 4617 boolean_t ptp = B_FALSE;
4617 4618 ill_walk_context_t ctx;
4618 4619 boolean_t match_illgrp = (match_flags & IPIF_MATCH_ILLGRP);
4619 4620 boolean_t no_duplicate = (match_flags & IPIF_MATCH_NONDUP);
4620 4621
4621 4622 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
4622 4623 /*
4623 4624 * Repeat twice, first based on local addresses and
4624 4625 * next time for pointopoint.
4625 4626 */
4626 4627 repeat:
4627 4628 ill = ILL_START_WALK_V4(&ctx, ipst);
4628 4629 for (; ill != NULL; ill = ill_next(&ctx, ill)) {
4629 4630 if (match_ill != NULL && ill != match_ill &&
4630 4631 (!match_illgrp || !IS_IN_SAME_ILLGRP(ill, match_ill))) {
4631 4632 continue;
4632 4633 }
4633 4634 mutex_enter(&ill->ill_lock);
4634 4635 for (ipif = ill->ill_ipif; ipif != NULL;
4635 4636 ipif = ipif->ipif_next) {
4636 4637 if (zoneid != ALL_ZONES &&
4637 4638 zoneid != ipif->ipif_zoneid &&
4638 4639 ipif->ipif_zoneid != ALL_ZONES)
4639 4640 continue;
4640 4641
4641 4642 if (no_duplicate && !(ipif->ipif_flags & IPIF_UP))
4642 4643 continue;
4643 4644
4644 4645 /* Allow the ipif to be down */
4645 4646 if ((!ptp && (ipif->ipif_lcl_addr == addr) &&
4646 4647 ((ipif->ipif_flags & IPIF_UNNUMBERED) == 0)) ||
4647 4648 (ptp && (ipif->ipif_flags & IPIF_POINTOPOINT) &&
4648 4649 (ipif->ipif_pp_dst_addr == addr))) {
4649 4650 if (!IPIF_IS_CONDEMNED(ipif)) {
4650 4651 ipif_refhold_locked(ipif);
4651 4652 mutex_exit(&ill->ill_lock);
4652 4653 rw_exit(&ipst->ips_ill_g_lock);
4653 4654 return (ipif);
4654 4655 }
4655 4656 }
4656 4657 }
4657 4658 mutex_exit(&ill->ill_lock);
4658 4659 }
4659 4660
4660 4661 /* If we already did the ptp case, then we are done */
4661 4662 if (ptp) {
4662 4663 rw_exit(&ipst->ips_ill_g_lock);
4663 4664 return (NULL);
4664 4665 }
4665 4666 ptp = B_TRUE;
4666 4667 goto repeat;
4667 4668 }
4668 4669
4669 4670 /*
4670 4671 * Lookup an ipif with the specified address. For point-to-point links we
4671 4672 * look for matches on either the destination address or the local address,
4672 4673 * but we skip the local address check if IPIF_UNNUMBERED is set. If the
4673 4674 * `match_ill' argument is non-NULL, the lookup is restricted to that ill
4674 4675 * (or illgrp if `match_ill' is in an IPMP group).
4675 4676 */
4676 4677 ipif_t *
4677 4678 ipif_lookup_addr(ipaddr_t addr, ill_t *match_ill, zoneid_t zoneid,
4678 4679 ip_stack_t *ipst)
4679 4680 {
4680 4681 return (ipif_lookup_addr_common(addr, match_ill, IPIF_MATCH_ILLGRP,
4681 4682 zoneid, ipst));
4682 4683 }
4683 4684
4684 4685 /*
4685 4686 * Lookup an ipif with the specified address. Similar to ipif_lookup_addr,
4686 4687 * except that we will only return an address if it is not marked as
4687 4688 * IPIF_DUPLICATE
4688 4689 */
4689 4690 ipif_t *
4690 4691 ipif_lookup_addr_nondup(ipaddr_t addr, ill_t *match_ill, zoneid_t zoneid,
4691 4692 ip_stack_t *ipst)
4692 4693 {
4693 4694 return (ipif_lookup_addr_common(addr, match_ill,
4694 4695 (IPIF_MATCH_ILLGRP | IPIF_MATCH_NONDUP),
4695 4696 zoneid, ipst));
4696 4697 }
4697 4698
4698 4699 /*
4699 4700 * Special abbreviated version of ipif_lookup_addr() that doesn't match
4700 4701 * `match_ill' across the IPMP group. This function is only needed in some
4701 4702 * corner-cases; almost everything should use ipif_lookup_addr().
4702 4703 */
4703 4704 ipif_t *
4704 4705 ipif_lookup_addr_exact(ipaddr_t addr, ill_t *match_ill, ip_stack_t *ipst)
4705 4706 {
4706 4707 ASSERT(match_ill != NULL);
4707 4708 return (ipif_lookup_addr_common(addr, match_ill, 0, ALL_ZONES,
4708 4709 ipst));
4709 4710 }
4710 4711
4711 4712 /*
4712 4713 * Look for an ipif with the specified address. For point-point links
4713 4714 * we look for matches on either the destination address and the local
4714 4715 * address, but we ignore the check on the local address if IPIF_UNNUMBERED
4715 4716 * is set.
4716 4717 * If the `match_ill' argument is non-NULL, the lookup is restricted to that
4717 4718 * ill (or illgrp if `match_ill' is in an IPMP group).
4718 4719 * Return the zoneid for the ipif which matches. ALL_ZONES if no match.
4719 4720 */
4720 4721 zoneid_t
4721 4722 ipif_lookup_addr_zoneid(ipaddr_t addr, ill_t *match_ill, ip_stack_t *ipst)
4722 4723 {
4723 4724 zoneid_t zoneid;
4724 4725 ipif_t *ipif;
4725 4726 ill_t *ill;
4726 4727 boolean_t ptp = B_FALSE;
4727 4728 ill_walk_context_t ctx;
4728 4729
4729 4730 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
4730 4731 /*
4731 4732 * Repeat twice, first based on local addresses and
4732 4733 * next time for pointopoint.
4733 4734 */
4734 4735 repeat:
4735 4736 ill = ILL_START_WALK_V4(&ctx, ipst);
4736 4737 for (; ill != NULL; ill = ill_next(&ctx, ill)) {
4737 4738 if (match_ill != NULL && ill != match_ill &&
4738 4739 !IS_IN_SAME_ILLGRP(ill, match_ill)) {
4739 4740 continue;
4740 4741 }
4741 4742 mutex_enter(&ill->ill_lock);
4742 4743 for (ipif = ill->ill_ipif; ipif != NULL;
4743 4744 ipif = ipif->ipif_next) {
4744 4745 /* Allow the ipif to be down */
4745 4746 if ((!ptp && (ipif->ipif_lcl_addr == addr) &&
4746 4747 ((ipif->ipif_flags & IPIF_UNNUMBERED) == 0)) ||
4747 4748 (ptp && (ipif->ipif_flags & IPIF_POINTOPOINT) &&
4748 4749 (ipif->ipif_pp_dst_addr == addr)) &&
4749 4750 !(ipif->ipif_state_flags & IPIF_CONDEMNED)) {
4750 4751 zoneid = ipif->ipif_zoneid;
4751 4752 mutex_exit(&ill->ill_lock);
4752 4753 rw_exit(&ipst->ips_ill_g_lock);
4753 4754 /*
4754 4755 * If ipif_zoneid was ALL_ZONES then we have
4755 4756 * a trusted extensions shared IP address.
4756 4757 * In that case GLOBAL_ZONEID works to send.
4757 4758 */
4758 4759 if (zoneid == ALL_ZONES)
4759 4760 zoneid = GLOBAL_ZONEID;
4760 4761 return (zoneid);
4761 4762 }
4762 4763 }
4763 4764 mutex_exit(&ill->ill_lock);
4764 4765 }
4765 4766
4766 4767 /* If we already did the ptp case, then we are done */
4767 4768 if (ptp) {
4768 4769 rw_exit(&ipst->ips_ill_g_lock);
4769 4770 return (ALL_ZONES);
4770 4771 }
4771 4772 ptp = B_TRUE;
4772 4773 goto repeat;
4773 4774 }
4774 4775
4775 4776 /*
4776 4777 * Look for an ipif that matches the specified remote address i.e. the
4777 4778 * ipif that would receive the specified packet.
4778 4779 * First look for directly connected interfaces and then do a recursive
4779 4780 * IRE lookup and pick the first ipif corresponding to the source address in the
4780 4781 * ire.
4781 4782 * Returns: held ipif
4782 4783 *
4783 4784 * This is only used for ICMP_ADDRESS_MASK_REQUESTs
4784 4785 */
4785 4786 ipif_t *
4786 4787 ipif_lookup_remote(ill_t *ill, ipaddr_t addr, zoneid_t zoneid)
4787 4788 {
4788 4789 ipif_t *ipif;
4789 4790
4790 4791 ASSERT(!ill->ill_isv6);
4791 4792
4792 4793 /*
4793 4794 * Someone could be changing this ipif currently or change it
4794 4795 * after we return this. Thus a few packets could use the old
4795 4796 * old values. However structure updates/creates (ire, ilg, ilm etc)
4796 4797 * will atomically be updated or cleaned up with the new value
4797 4798 * Thus we don't need a lock to check the flags or other attrs below.
4798 4799 */
4799 4800 mutex_enter(&ill->ill_lock);
4800 4801 for (ipif = ill->ill_ipif; ipif != NULL; ipif = ipif->ipif_next) {
4801 4802 if (IPIF_IS_CONDEMNED(ipif))
4802 4803 continue;
4803 4804 if (zoneid != ALL_ZONES && zoneid != ipif->ipif_zoneid &&
4804 4805 ipif->ipif_zoneid != ALL_ZONES)
4805 4806 continue;
4806 4807 /* Allow the ipif to be down */
4807 4808 if (ipif->ipif_flags & IPIF_POINTOPOINT) {
4808 4809 if ((ipif->ipif_pp_dst_addr == addr) ||
4809 4810 (!(ipif->ipif_flags & IPIF_UNNUMBERED) &&
4810 4811 ipif->ipif_lcl_addr == addr)) {
4811 4812 ipif_refhold_locked(ipif);
4812 4813 mutex_exit(&ill->ill_lock);
4813 4814 return (ipif);
4814 4815 }
4815 4816 } else if (ipif->ipif_subnet == (addr & ipif->ipif_net_mask)) {
4816 4817 ipif_refhold_locked(ipif);
4817 4818 mutex_exit(&ill->ill_lock);
4818 4819 return (ipif);
4819 4820 }
4820 4821 }
4821 4822 mutex_exit(&ill->ill_lock);
4822 4823 /*
4823 4824 * For a remote destination it isn't possible to nail down a particular
4824 4825 * ipif.
4825 4826 */
4826 4827
4827 4828 /* Pick the first interface */
4828 4829 ipif = ipif_get_next_ipif(NULL, ill);
4829 4830 return (ipif);
4830 4831 }
4831 4832
4832 4833 /*
4833 4834 * This func does not prevent refcnt from increasing. But if
4834 4835 * the caller has taken steps to that effect, then this func
4835 4836 * can be used to determine whether the ill has become quiescent
4836 4837 */
4837 4838 static boolean_t
4838 4839 ill_is_quiescent(ill_t *ill)
4839 4840 {
4840 4841 ipif_t *ipif;
4841 4842
4842 4843 ASSERT(MUTEX_HELD(&ill->ill_lock));
4843 4844
4844 4845 for (ipif = ill->ill_ipif; ipif != NULL; ipif = ipif->ipif_next) {
4845 4846 if (ipif->ipif_refcnt != 0)
4846 4847 return (B_FALSE);
4847 4848 }
4848 4849 if (!ILL_DOWN_OK(ill) || ill->ill_refcnt != 0) {
4849 4850 return (B_FALSE);
4850 4851 }
4851 4852 return (B_TRUE);
4852 4853 }
4853 4854
4854 4855 boolean_t
4855 4856 ill_is_freeable(ill_t *ill)
4856 4857 {
4857 4858 ipif_t *ipif;
4858 4859
4859 4860 ASSERT(MUTEX_HELD(&ill->ill_lock));
4860 4861
4861 4862 for (ipif = ill->ill_ipif; ipif != NULL; ipif = ipif->ipif_next) {
4862 4863 if (ipif->ipif_refcnt != 0) {
4863 4864 return (B_FALSE);
4864 4865 }
4865 4866 }
4866 4867 if (!ILL_FREE_OK(ill) || ill->ill_refcnt != 0) {
4867 4868 return (B_FALSE);
4868 4869 }
4869 4870 return (B_TRUE);
4870 4871 }
4871 4872
4872 4873 /*
4873 4874 * This func does not prevent refcnt from increasing. But if
4874 4875 * the caller has taken steps to that effect, then this func
4875 4876 * can be used to determine whether the ipif has become quiescent
4876 4877 */
4877 4878 static boolean_t
4878 4879 ipif_is_quiescent(ipif_t *ipif)
4879 4880 {
4880 4881 ill_t *ill;
4881 4882
4882 4883 ASSERT(MUTEX_HELD(&ipif->ipif_ill->ill_lock));
4883 4884
4884 4885 if (ipif->ipif_refcnt != 0)
4885 4886 return (B_FALSE);
4886 4887
4887 4888 ill = ipif->ipif_ill;
4888 4889 if (ill->ill_ipif_up_count != 0 || ill->ill_ipif_dup_count != 0 ||
4889 4890 ill->ill_logical_down) {
4890 4891 return (B_TRUE);
4891 4892 }
4892 4893
4893 4894 /* This is the last ipif going down or being deleted on this ill */
4894 4895 if (ill->ill_ire_cnt != 0 || ill->ill_refcnt != 0) {
4895 4896 return (B_FALSE);
4896 4897 }
4897 4898
4898 4899 return (B_TRUE);
4899 4900 }
4900 4901
4901 4902 /*
4902 4903 * return true if the ipif can be destroyed: the ipif has to be quiescent
4903 4904 * with zero references from ire/ilm to it.
4904 4905 */
4905 4906 static boolean_t
4906 4907 ipif_is_freeable(ipif_t *ipif)
4907 4908 {
4908 4909 ASSERT(MUTEX_HELD(&ipif->ipif_ill->ill_lock));
4909 4910 ASSERT(ipif->ipif_id != 0);
4910 4911 return (ipif->ipif_refcnt == 0);
4911 4912 }
4912 4913
4913 4914 /*
4914 4915 * The ipif/ill/ire has been refreled. Do the tail processing.
4915 4916 * Determine if the ipif or ill in question has become quiescent and if so
4916 4917 * wakeup close and/or restart any queued pending ioctl that is waiting
4917 4918 * for the ipif_down (or ill_down)
4918 4919 */
4919 4920 void
4920 4921 ipif_ill_refrele_tail(ill_t *ill)
4921 4922 {
4922 4923 mblk_t *mp;
4923 4924 conn_t *connp;
4924 4925 ipsq_t *ipsq;
4925 4926 ipxop_t *ipx;
4926 4927 ipif_t *ipif;
4927 4928 dl_notify_ind_t *dlindp;
4928 4929
4929 4930 ASSERT(MUTEX_HELD(&ill->ill_lock));
4930 4931
4931 4932 if ((ill->ill_state_flags & ILL_CONDEMNED) && ill_is_freeable(ill)) {
4932 4933 /* ip_modclose() may be waiting */
4933 4934 cv_broadcast(&ill->ill_cv);
4934 4935 }
4935 4936
4936 4937 ipsq = ill->ill_phyint->phyint_ipsq;
4937 4938 mutex_enter(&ipsq->ipsq_lock);
4938 4939 ipx = ipsq->ipsq_xop;
4939 4940 mutex_enter(&ipx->ipx_lock);
4940 4941 if (ipx->ipx_waitfor == 0) /* no one's waiting; bail */
4941 4942 goto unlock;
4942 4943
4943 4944 ASSERT(ipx->ipx_pending_mp != NULL && ipx->ipx_pending_ipif != NULL);
4944 4945
4945 4946 ipif = ipx->ipx_pending_ipif;
4946 4947 if (ipif->ipif_ill != ill) /* wait is for another ill; bail */
4947 4948 goto unlock;
4948 4949
4949 4950 switch (ipx->ipx_waitfor) {
4950 4951 case IPIF_DOWN:
4951 4952 if (!ipif_is_quiescent(ipif))
4952 4953 goto unlock;
4953 4954 break;
4954 4955 case IPIF_FREE:
4955 4956 if (!ipif_is_freeable(ipif))
4956 4957 goto unlock;
4957 4958 break;
4958 4959 case ILL_DOWN:
4959 4960 if (!ill_is_quiescent(ill))
4960 4961 goto unlock;
4961 4962 break;
4962 4963 case ILL_FREE:
4963 4964 /*
4964 4965 * ILL_FREE is only for loopback; normal ill teardown waits
4965 4966 * synchronously in ip_modclose() without using ipx_waitfor,
4966 4967 * handled by the cv_broadcast() at the top of this function.
4967 4968 */
4968 4969 if (!ill_is_freeable(ill))
4969 4970 goto unlock;
4970 4971 break;
4971 4972 default:
4972 4973 cmn_err(CE_PANIC, "ipsq: %p unknown ipx_waitfor %d\n",
4973 4974 (void *)ipsq, ipx->ipx_waitfor);
4974 4975 }
4975 4976
4976 4977 ill_refhold_locked(ill); /* for qwriter_ip() call below */
4977 4978 mutex_exit(&ipx->ipx_lock);
4978 4979 mp = ipsq_pending_mp_get(ipsq, &connp);
4979 4980 mutex_exit(&ipsq->ipsq_lock);
4980 4981 mutex_exit(&ill->ill_lock);
4981 4982
4982 4983 ASSERT(mp != NULL);
4983 4984 /*
4984 4985 * NOTE: all of the qwriter_ip() calls below use CUR_OP since
4985 4986 * we can only get here when the current operation decides it
4986 4987 * it needs to quiesce via ipsq_pending_mp_add().
4987 4988 */
4988 4989 switch (mp->b_datap->db_type) {
4989 4990 case M_PCPROTO:
4990 4991 case M_PROTO:
4991 4992 /*
4992 4993 * For now, only DL_NOTIFY_IND messages can use this facility.
4993 4994 */
4994 4995 dlindp = (dl_notify_ind_t *)mp->b_rptr;
4995 4996 ASSERT(dlindp->dl_primitive == DL_NOTIFY_IND);
4996 4997
4997 4998 switch (dlindp->dl_notification) {
4998 4999 case DL_NOTE_PHYS_ADDR:
4999 5000 qwriter_ip(ill, ill->ill_rq, mp,
5000 5001 ill_set_phys_addr_tail, CUR_OP, B_TRUE);
5001 5002 return;
5002 5003 case DL_NOTE_REPLUMB:
5003 5004 qwriter_ip(ill, ill->ill_rq, mp,
5004 5005 ill_replumb_tail, CUR_OP, B_TRUE);
5005 5006 return;
5006 5007 default:
5007 5008 ASSERT(0);
5008 5009 ill_refrele(ill);
5009 5010 }
5010 5011 break;
5011 5012
5012 5013 case M_ERROR:
5013 5014 case M_HANGUP:
5014 5015 qwriter_ip(ill, ill->ill_rq, mp, ipif_all_down_tail, CUR_OP,
5015 5016 B_TRUE);
5016 5017 return;
5017 5018
5018 5019 case M_IOCTL:
5019 5020 case M_IOCDATA:
5020 5021 qwriter_ip(ill, (connp != NULL ? CONNP_TO_WQ(connp) :
5021 5022 ill->ill_wq), mp, ip_reprocess_ioctl, CUR_OP, B_TRUE);
5022 5023 return;
5023 5024
5024 5025 default:
5025 5026 cmn_err(CE_PANIC, "ipif_ill_refrele_tail mp %p "
5026 5027 "db_type %d\n", (void *)mp, mp->b_datap->db_type);
5027 5028 }
5028 5029 return;
5029 5030 unlock:
5030 5031 mutex_exit(&ipsq->ipsq_lock);
5031 5032 mutex_exit(&ipx->ipx_lock);
5032 5033 mutex_exit(&ill->ill_lock);
5033 5034 }
5034 5035
5035 5036 #ifdef DEBUG
5036 5037 /* Reuse trace buffer from beginning (if reached the end) and record trace */
5037 5038 static void
5038 5039 th_trace_rrecord(th_trace_t *th_trace)
5039 5040 {
5040 5041 tr_buf_t *tr_buf;
5041 5042 uint_t lastref;
5042 5043
5043 5044 lastref = th_trace->th_trace_lastref;
5044 5045 lastref++;
5045 5046 if (lastref == TR_BUF_MAX)
5046 5047 lastref = 0;
5047 5048 th_trace->th_trace_lastref = lastref;
5048 5049 tr_buf = &th_trace->th_trbuf[lastref];
5049 5050 tr_buf->tr_time = ddi_get_lbolt();
5050 5051 tr_buf->tr_depth = getpcstack(tr_buf->tr_stack, TR_STACK_DEPTH);
5051 5052 }
5052 5053
5053 5054 static void
5054 5055 th_trace_free(void *value)
5055 5056 {
5056 5057 th_trace_t *th_trace = value;
5057 5058
5058 5059 ASSERT(th_trace->th_refcnt == 0);
5059 5060 kmem_free(th_trace, sizeof (*th_trace));
5060 5061 }
5061 5062
5062 5063 /*
5063 5064 * Find or create the per-thread hash table used to track object references.
5064 5065 * The ipst argument is NULL if we shouldn't allocate.
5065 5066 *
5066 5067 * Accesses per-thread data, so there's no need to lock here.
5067 5068 */
5068 5069 static mod_hash_t *
5069 5070 th_trace_gethash(ip_stack_t *ipst)
5070 5071 {
5071 5072 th_hash_t *thh;
5072 5073
5073 5074 if ((thh = tsd_get(ip_thread_data)) == NULL && ipst != NULL) {
5074 5075 mod_hash_t *mh;
5075 5076 char name[256];
5076 5077 size_t objsize, rshift;
5077 5078 int retv;
5078 5079
5079 5080 if ((thh = kmem_alloc(sizeof (*thh), KM_NOSLEEP)) == NULL)
5080 5081 return (NULL);
5081 5082 (void) snprintf(name, sizeof (name), "th_trace_%p",
5082 5083 (void *)curthread);
5083 5084
5084 5085 /*
5085 5086 * We use mod_hash_create_extended here rather than the more
5086 5087 * obvious mod_hash_create_ptrhash because the latter has a
5087 5088 * hard-coded KM_SLEEP, and we'd prefer to fail rather than
5088 5089 * block.
5089 5090 */
5090 5091 objsize = MAX(MAX(sizeof (ill_t), sizeof (ipif_t)),
5091 5092 MAX(sizeof (ire_t), sizeof (ncec_t)));
5092 5093 rshift = highbit(objsize);
5093 5094 mh = mod_hash_create_extended(name, 64, mod_hash_null_keydtor,
5094 5095 th_trace_free, mod_hash_byptr, (void *)rshift,
5095 5096 mod_hash_ptrkey_cmp, KM_NOSLEEP);
5096 5097 if (mh == NULL) {
5097 5098 kmem_free(thh, sizeof (*thh));
5098 5099 return (NULL);
5099 5100 }
5100 5101 thh->thh_hash = mh;
5101 5102 thh->thh_ipst = ipst;
5102 5103 /*
5103 5104 * We trace ills, ipifs, ires, and nces. All of these are
5104 5105 * per-IP-stack, so the lock on the thread list is as well.
5105 5106 */
5106 5107 rw_enter(&ip_thread_rwlock, RW_WRITER);
5107 5108 list_insert_tail(&ip_thread_list, thh);
5108 5109 rw_exit(&ip_thread_rwlock);
5109 5110 retv = tsd_set(ip_thread_data, thh);
5110 5111 ASSERT(retv == 0);
5111 5112 }
5112 5113 return (thh != NULL ? thh->thh_hash : NULL);
5113 5114 }
5114 5115
5115 5116 boolean_t
5116 5117 th_trace_ref(const void *obj, ip_stack_t *ipst)
5117 5118 {
5118 5119 th_trace_t *th_trace;
5119 5120 mod_hash_t *mh;
5120 5121 mod_hash_val_t val;
5121 5122
5122 5123 if ((mh = th_trace_gethash(ipst)) == NULL)
5123 5124 return (B_FALSE);
5124 5125
5125 5126 /*
5126 5127 * Attempt to locate the trace buffer for this obj and thread.
5127 5128 * If it does not exist, then allocate a new trace buffer and
5128 5129 * insert into the hash.
5129 5130 */
5130 5131 if (mod_hash_find(mh, (mod_hash_key_t)obj, &val) == MH_ERR_NOTFOUND) {
5131 5132 th_trace = kmem_zalloc(sizeof (th_trace_t), KM_NOSLEEP);
5132 5133 if (th_trace == NULL)
5133 5134 return (B_FALSE);
5134 5135
5135 5136 th_trace->th_id = curthread;
5136 5137 if (mod_hash_insert(mh, (mod_hash_key_t)obj,
5137 5138 (mod_hash_val_t)th_trace) != 0) {
5138 5139 kmem_free(th_trace, sizeof (th_trace_t));
5139 5140 return (B_FALSE);
5140 5141 }
5141 5142 } else {
5142 5143 th_trace = (th_trace_t *)val;
5143 5144 }
5144 5145
5145 5146 ASSERT(th_trace->th_refcnt >= 0 &&
5146 5147 th_trace->th_refcnt < TR_BUF_MAX - 1);
5147 5148
5148 5149 th_trace->th_refcnt++;
5149 5150 th_trace_rrecord(th_trace);
5150 5151 return (B_TRUE);
5151 5152 }
5152 5153
5153 5154 /*
5154 5155 * For the purpose of tracing a reference release, we assume that global
5155 5156 * tracing is always on and that the same thread initiated the reference hold
5156 5157 * is releasing.
5157 5158 */
5158 5159 void
5159 5160 th_trace_unref(const void *obj)
5160 5161 {
5161 5162 int retv;
5162 5163 mod_hash_t *mh;
5163 5164 th_trace_t *th_trace;
5164 5165 mod_hash_val_t val;
5165 5166
5166 5167 mh = th_trace_gethash(NULL);
5167 5168 retv = mod_hash_find(mh, (mod_hash_key_t)obj, &val);
5168 5169 ASSERT(retv == 0);
5169 5170 th_trace = (th_trace_t *)val;
5170 5171
5171 5172 ASSERT(th_trace->th_refcnt > 0);
5172 5173 th_trace->th_refcnt--;
5173 5174 th_trace_rrecord(th_trace);
5174 5175 }
5175 5176
5176 5177 /*
5177 5178 * If tracing has been disabled, then we assume that the reference counts are
5178 5179 * now useless, and we clear them out before destroying the entries.
5179 5180 */
5180 5181 void
5181 5182 th_trace_cleanup(const void *obj, boolean_t trace_disable)
5182 5183 {
5183 5184 th_hash_t *thh;
5184 5185 mod_hash_t *mh;
5185 5186 mod_hash_val_t val;
5186 5187 th_trace_t *th_trace;
5187 5188 int retv;
5188 5189
5189 5190 rw_enter(&ip_thread_rwlock, RW_READER);
5190 5191 for (thh = list_head(&ip_thread_list); thh != NULL;
5191 5192 thh = list_next(&ip_thread_list, thh)) {
5192 5193 if (mod_hash_find(mh = thh->thh_hash, (mod_hash_key_t)obj,
5193 5194 &val) == 0) {
5194 5195 th_trace = (th_trace_t *)val;
5195 5196 if (trace_disable)
5196 5197 th_trace->th_refcnt = 0;
5197 5198 retv = mod_hash_destroy(mh, (mod_hash_key_t)obj);
5198 5199 ASSERT(retv == 0);
5199 5200 }
5200 5201 }
5201 5202 rw_exit(&ip_thread_rwlock);
5202 5203 }
5203 5204
5204 5205 void
5205 5206 ipif_trace_ref(ipif_t *ipif)
5206 5207 {
5207 5208 ASSERT(MUTEX_HELD(&ipif->ipif_ill->ill_lock));
5208 5209
5209 5210 if (ipif->ipif_trace_disable)
5210 5211 return;
5211 5212
5212 5213 if (!th_trace_ref(ipif, ipif->ipif_ill->ill_ipst)) {
5213 5214 ipif->ipif_trace_disable = B_TRUE;
5214 5215 ipif_trace_cleanup(ipif);
5215 5216 }
5216 5217 }
5217 5218
5218 5219 void
5219 5220 ipif_untrace_ref(ipif_t *ipif)
5220 5221 {
5221 5222 ASSERT(MUTEX_HELD(&ipif->ipif_ill->ill_lock));
5222 5223
5223 5224 if (!ipif->ipif_trace_disable)
5224 5225 th_trace_unref(ipif);
5225 5226 }
5226 5227
5227 5228 void
5228 5229 ill_trace_ref(ill_t *ill)
5229 5230 {
5230 5231 ASSERT(MUTEX_HELD(&ill->ill_lock));
5231 5232
5232 5233 if (ill->ill_trace_disable)
5233 5234 return;
5234 5235
5235 5236 if (!th_trace_ref(ill, ill->ill_ipst)) {
5236 5237 ill->ill_trace_disable = B_TRUE;
5237 5238 ill_trace_cleanup(ill);
5238 5239 }
5239 5240 }
5240 5241
5241 5242 void
5242 5243 ill_untrace_ref(ill_t *ill)
5243 5244 {
5244 5245 ASSERT(MUTEX_HELD(&ill->ill_lock));
5245 5246
5246 5247 if (!ill->ill_trace_disable)
5247 5248 th_trace_unref(ill);
5248 5249 }
5249 5250
5250 5251 /*
5251 5252 * Called when ipif is unplumbed or when memory alloc fails. Note that on
5252 5253 * failure, ipif_trace_disable is set.
5253 5254 */
5254 5255 static void
5255 5256 ipif_trace_cleanup(const ipif_t *ipif)
5256 5257 {
5257 5258 th_trace_cleanup(ipif, ipif->ipif_trace_disable);
5258 5259 }
5259 5260
5260 5261 /*
5261 5262 * Called when ill is unplumbed or when memory alloc fails. Note that on
5262 5263 * failure, ill_trace_disable is set.
5263 5264 */
5264 5265 static void
5265 5266 ill_trace_cleanup(const ill_t *ill)
5266 5267 {
5267 5268 th_trace_cleanup(ill, ill->ill_trace_disable);
5268 5269 }
5269 5270 #endif /* DEBUG */
5270 5271
5271 5272 void
5272 5273 ipif_refhold_locked(ipif_t *ipif)
5273 5274 {
5274 5275 ASSERT(MUTEX_HELD(&ipif->ipif_ill->ill_lock));
5275 5276 ipif->ipif_refcnt++;
5276 5277 IPIF_TRACE_REF(ipif);
5277 5278 }
5278 5279
5279 5280 void
5280 5281 ipif_refhold(ipif_t *ipif)
5281 5282 {
5282 5283 ill_t *ill;
5283 5284
5284 5285 ill = ipif->ipif_ill;
5285 5286 mutex_enter(&ill->ill_lock);
5286 5287 ipif->ipif_refcnt++;
5287 5288 IPIF_TRACE_REF(ipif);
5288 5289 mutex_exit(&ill->ill_lock);
5289 5290 }
5290 5291
5291 5292 /*
5292 5293 * Must not be called while holding any locks. Otherwise if this is
5293 5294 * the last reference to be released there is a chance of recursive mutex
5294 5295 * panic due to ipif_refrele -> ipif_ill_refrele_tail -> qwriter_ip trying
5295 5296 * to restart an ioctl.
5296 5297 */
5297 5298 void
5298 5299 ipif_refrele(ipif_t *ipif)
5299 5300 {
5300 5301 ill_t *ill;
5301 5302
5302 5303 ill = ipif->ipif_ill;
5303 5304
5304 5305 mutex_enter(&ill->ill_lock);
5305 5306 ASSERT(ipif->ipif_refcnt != 0);
5306 5307 ipif->ipif_refcnt--;
5307 5308 IPIF_UNTRACE_REF(ipif);
5308 5309 if (ipif->ipif_refcnt != 0) {
5309 5310 mutex_exit(&ill->ill_lock);
5310 5311 return;
5311 5312 }
5312 5313
5313 5314 /* Drops the ill_lock */
5314 5315 ipif_ill_refrele_tail(ill);
5315 5316 }
5316 5317
5317 5318 ipif_t *
5318 5319 ipif_get_next_ipif(ipif_t *curr, ill_t *ill)
5319 5320 {
5320 5321 ipif_t *ipif;
5321 5322
5322 5323 mutex_enter(&ill->ill_lock);
5323 5324 for (ipif = (curr == NULL ? ill->ill_ipif : curr->ipif_next);
5324 5325 ipif != NULL; ipif = ipif->ipif_next) {
5325 5326 if (IPIF_IS_CONDEMNED(ipif))
5326 5327 continue;
5327 5328 ipif_refhold_locked(ipif);
5328 5329 mutex_exit(&ill->ill_lock);
5329 5330 return (ipif);
5330 5331 }
5331 5332 mutex_exit(&ill->ill_lock);
5332 5333 return (NULL);
5333 5334 }
5334 5335
5335 5336 /*
5336 5337 * TODO: make this table extendible at run time
5337 5338 * Return a pointer to the mac type info for 'mac_type'
5338 5339 */
5339 5340 static ip_m_t *
5340 5341 ip_m_lookup(t_uscalar_t mac_type)
5341 5342 {
5342 5343 ip_m_t *ipm;
5343 5344
5344 5345 for (ipm = ip_m_tbl; ipm < A_END(ip_m_tbl); ipm++)
5345 5346 if (ipm->ip_m_mac_type == mac_type)
5346 5347 return (ipm);
5347 5348 return (NULL);
5348 5349 }
5349 5350
5350 5351 /*
5351 5352 * Make a link layer address from the multicast IP address *addr.
5352 5353 * To form the link layer address, invoke the ip_m_v*mapping function
5353 5354 * associated with the link-layer type.
5354 5355 */
5355 5356 void
5356 5357 ip_mcast_mapping(ill_t *ill, uchar_t *addr, uchar_t *hwaddr)
5357 5358 {
5358 5359 ip_m_t *ipm;
5359 5360
5360 5361 if (ill->ill_net_type == IRE_IF_NORESOLVER)
5361 5362 return;
5362 5363
5363 5364 ASSERT(addr != NULL);
5364 5365
5365 5366 ipm = ip_m_lookup(ill->ill_mactype);
5366 5367 if (ipm == NULL ||
5367 5368 (ill->ill_isv6 && ipm->ip_m_v6mapping == NULL) ||
5368 5369 (!ill->ill_isv6 && ipm->ip_m_v4mapping == NULL)) {
5369 5370 ip0dbg(("no mapping for ill %s mactype 0x%x\n",
5370 5371 ill->ill_name, ill->ill_mactype));
5371 5372 return;
5372 5373 }
5373 5374 if (ill->ill_isv6)
5374 5375 (*ipm->ip_m_v6mapping)(ill, addr, hwaddr);
5375 5376 else
5376 5377 (*ipm->ip_m_v4mapping)(ill, addr, hwaddr);
5377 5378 }
5378 5379
5379 5380 /*
5380 5381 * Returns B_FALSE if the IPv4 netmask pointed by `mask' is non-contiguous.
5381 5382 * Otherwise returns B_TRUE.
5382 5383 *
5383 5384 * The netmask can be verified to be contiguous with 32 shifts and or
5384 5385 * operations. Take the contiguous mask (in host byte order) and compute
5385 5386 * mask | mask << 1 | mask << 2 | ... | mask << 31
5386 5387 * the result will be the same as the 'mask' for contiguous mask.
5387 5388 */
5388 5389 static boolean_t
5389 5390 ip_contiguous_mask(uint32_t mask)
5390 5391 {
5391 5392 uint32_t m = mask;
5392 5393 int i;
5393 5394
5394 5395 for (i = 1; i < 32; i++)
5395 5396 m |= (mask << i);
5396 5397
5397 5398 return (m == mask);
5398 5399 }
5399 5400
5400 5401 /*
5401 5402 * ip_rt_add is called to add an IPv4 route to the forwarding table.
5402 5403 * ill is passed in to associate it with the correct interface.
5403 5404 * If ire_arg is set, then we return the held IRE in that location.
5404 5405 */
5405 5406 int
5406 5407 ip_rt_add(ipaddr_t dst_addr, ipaddr_t mask, ipaddr_t gw_addr,
5407 5408 ipaddr_t src_addr, int flags, ill_t *ill, ire_t **ire_arg,
5408 5409 boolean_t ioctl_msg, struct rtsa_s *sp, ip_stack_t *ipst, zoneid_t zoneid)
5409 5410 {
5410 5411 ire_t *ire, *nire;
5411 5412 ire_t *gw_ire = NULL;
5412 5413 ipif_t *ipif = NULL;
5413 5414 uint_t type;
5414 5415 int match_flags = MATCH_IRE_TYPE;
5415 5416 tsol_gc_t *gc = NULL;
5416 5417 tsol_gcgrp_t *gcgrp = NULL;
5417 5418 boolean_t gcgrp_xtraref = B_FALSE;
5418 5419 boolean_t cgtp_broadcast;
5419 5420 boolean_t unbound = B_FALSE;
5420 5421
5421 5422 ip1dbg(("ip_rt_add:"));
5422 5423
5423 5424 if (ire_arg != NULL)
5424 5425 *ire_arg = NULL;
5425 5426
5426 5427 /* disallow non-contiguous netmasks */
5427 5428 if (!ip_contiguous_mask(ntohl(mask)))
5428 5429 return (ENOTSUP);
5429 5430
5430 5431 /*
5431 5432 * If this is the case of RTF_HOST being set, then we set the netmask
5432 5433 * to all ones (regardless if one was supplied).
5433 5434 */
5434 5435 if (flags & RTF_HOST)
5435 5436 mask = IP_HOST_MASK;
5436 5437
5437 5438 /*
5438 5439 * Prevent routes with a zero gateway from being created (since
5439 5440 * interfaces can currently be plumbed and brought up no assigned
5440 5441 * address).
5441 5442 */
5442 5443 if (gw_addr == 0)
5443 5444 return (ENETUNREACH);
5444 5445 /*
5445 5446 * Get the ipif, if any, corresponding to the gw_addr
5446 5447 * If -ifp was specified we restrict ourselves to the ill, otherwise
5447 5448 * we match on the gatway and destination to handle unnumbered pt-pt
5448 5449 * interfaces.
5449 5450 */
5450 5451 if (ill != NULL)
5451 5452 ipif = ipif_lookup_addr(gw_addr, ill, ALL_ZONES, ipst);
5452 5453 else
5453 5454 ipif = ipif_lookup_interface(gw_addr, dst_addr, ipst);
5454 5455 if (ipif != NULL) {
5455 5456 if (IS_VNI(ipif->ipif_ill)) {
5456 5457 ipif_refrele(ipif);
5457 5458 return (EINVAL);
5458 5459 }
5459 5460 }
5460 5461
5461 5462 /*
5462 5463 * GateD will attempt to create routes with a loopback interface
5463 5464 * address as the gateway and with RTF_GATEWAY set. We allow
5464 5465 * these routes to be added, but create them as interface routes
5465 5466 * since the gateway is an interface address.
5466 5467 */
5467 5468 if ((ipif != NULL) && (ipif->ipif_ire_type == IRE_LOOPBACK)) {
5468 5469 flags &= ~RTF_GATEWAY;
5469 5470 if (gw_addr == INADDR_LOOPBACK && dst_addr == INADDR_LOOPBACK &&
5470 5471 mask == IP_HOST_MASK) {
5471 5472 ire = ire_ftable_lookup_v4(dst_addr, 0, 0, IRE_LOOPBACK,
5472 5473 NULL, ALL_ZONES, NULL, MATCH_IRE_TYPE, 0, ipst,
5473 5474 NULL);
5474 5475 if (ire != NULL) {
5475 5476 ire_refrele(ire);
5476 5477 ipif_refrele(ipif);
5477 5478 return (EEXIST);
5478 5479 }
5479 5480 ip1dbg(("ip_rt_add: 0x%p creating IRE 0x%x"
5480 5481 "for 0x%x\n", (void *)ipif,
5481 5482 ipif->ipif_ire_type,
5482 5483 ntohl(ipif->ipif_lcl_addr)));
5483 5484 ire = ire_create(
5484 5485 (uchar_t *)&dst_addr, /* dest address */
5485 5486 (uchar_t *)&mask, /* mask */
5486 5487 NULL, /* no gateway */
5487 5488 ipif->ipif_ire_type, /* LOOPBACK */
5488 5489 ipif->ipif_ill,
5489 5490 zoneid,
5490 5491 (ipif->ipif_flags & IPIF_PRIVATE) ? RTF_PRIVATE : 0,
5491 5492 NULL,
5492 5493 ipst);
5493 5494
5494 5495 if (ire == NULL) {
5495 5496 ipif_refrele(ipif);
5496 5497 return (ENOMEM);
5497 5498 }
5498 5499 /* src address assigned by the caller? */
5499 5500 if ((src_addr != INADDR_ANY) && (flags & RTF_SETSRC))
5500 5501 ire->ire_setsrc_addr = src_addr;
5501 5502
5502 5503 nire = ire_add(ire);
5503 5504 if (nire == NULL) {
5504 5505 /*
5505 5506 * In the result of failure, ire_add() will have
5506 5507 * already deleted the ire in question, so there
5507 5508 * is no need to do that here.
5508 5509 */
5509 5510 ipif_refrele(ipif);
5510 5511 return (ENOMEM);
5511 5512 }
5512 5513 /*
5513 5514 * Check if it was a duplicate entry. This handles
5514 5515 * the case of two racing route adds for the same route
5515 5516 */
5516 5517 if (nire != ire) {
5517 5518 ASSERT(nire->ire_identical_ref > 1);
5518 5519 ire_delete(nire);
5519 5520 ire_refrele(nire);
5520 5521 ipif_refrele(ipif);
5521 5522 return (EEXIST);
5522 5523 }
5523 5524 ire = nire;
5524 5525 goto save_ire;
5525 5526 }
5526 5527 }
5527 5528
5528 5529 /*
5529 5530 * The routes for multicast with CGTP are quite special in that
5530 5531 * the gateway is the local interface address, yet RTF_GATEWAY
5531 5532 * is set. We turn off RTF_GATEWAY to provide compatibility with
5532 5533 * this undocumented and unusual use of multicast routes.
5533 5534 */
5534 5535 if ((flags & RTF_MULTIRT) && ipif != NULL)
5535 5536 flags &= ~RTF_GATEWAY;
5536 5537
5537 5538 /*
5538 5539 * Traditionally, interface routes are ones where RTF_GATEWAY isn't set
5539 5540 * and the gateway address provided is one of the system's interface
5540 5541 * addresses. By using the routing socket interface and supplying an
5541 5542 * RTA_IFP sockaddr with an interface index, an alternate method of
5542 5543 * specifying an interface route to be created is available which uses
5543 5544 * the interface index that specifies the outgoing interface rather than
5544 5545 * the address of an outgoing interface (which may not be able to
5545 5546 * uniquely identify an interface). When coupled with the RTF_GATEWAY
5546 5547 * flag, routes can be specified which not only specify the next-hop to
5547 5548 * be used when routing to a certain prefix, but also which outgoing
5548 5549 * interface should be used.
5549 5550 *
5550 5551 * Previously, interfaces would have unique addresses assigned to them
5551 5552 * and so the address assigned to a particular interface could be used
5552 5553 * to identify a particular interface. One exception to this was the
5553 5554 * case of an unnumbered interface (where IPIF_UNNUMBERED was set).
5554 5555 *
5555 5556 * With the advent of IPv6 and its link-local addresses, this
5556 5557 * restriction was relaxed and interfaces could share addresses between
5557 5558 * themselves. In fact, typically all of the link-local interfaces on
5558 5559 * an IPv6 node or router will have the same link-local address. In
5559 5560 * order to differentiate between these interfaces, the use of an
5560 5561 * interface index is necessary and this index can be carried inside a
5561 5562 * RTA_IFP sockaddr (which is actually a sockaddr_dl). One restriction
5562 5563 * of using the interface index, however, is that all of the ipif's that
5563 5564 * are part of an ill have the same index and so the RTA_IFP sockaddr
5564 5565 * cannot be used to differentiate between ipif's (or logical
5565 5566 * interfaces) that belong to the same ill (physical interface).
5566 5567 *
5567 5568 * For example, in the following case involving IPv4 interfaces and
5568 5569 * logical interfaces
5569 5570 *
5570 5571 * 192.0.2.32 255.255.255.224 192.0.2.33 U if0
5571 5572 * 192.0.2.32 255.255.255.224 192.0.2.34 U if0
5572 5573 * 192.0.2.32 255.255.255.224 192.0.2.35 U if0
5573 5574 *
5574 5575 * the ipif's corresponding to each of these interface routes can be
5575 5576 * uniquely identified by the "gateway" (actually interface address).
5576 5577 *
5577 5578 * In this case involving multiple IPv6 default routes to a particular
5578 5579 * link-local gateway, the use of RTA_IFP is necessary to specify which
5579 5580 * default route is of interest:
5580 5581 *
5581 5582 * default fe80::123:4567:89ab:cdef U if0
5582 5583 * default fe80::123:4567:89ab:cdef U if1
5583 5584 */
5584 5585
5585 5586 /* RTF_GATEWAY not set */
5586 5587 if (!(flags & RTF_GATEWAY)) {
5587 5588 if (sp != NULL) {
5588 5589 ip2dbg(("ip_rt_add: gateway security attributes "
5589 5590 "cannot be set with interface route\n"));
5590 5591 if (ipif != NULL)
5591 5592 ipif_refrele(ipif);
5592 5593 return (EINVAL);
5593 5594 }
5594 5595
5595 5596 /*
5596 5597 * Whether or not ill (RTA_IFP) is set, we require that
5597 5598 * the gateway is one of our local addresses.
5598 5599 */
5599 5600 if (ipif == NULL)
5600 5601 return (ENETUNREACH);
5601 5602
5602 5603 /*
5603 5604 * We use MATCH_IRE_ILL here. If the caller specified an
5604 5605 * interface (from the RTA_IFP sockaddr) we use it, otherwise
5605 5606 * we use the ill derived from the gateway address.
5606 5607 * We can always match the gateway address since we record it
5607 5608 * in ire_gateway_addr.
5608 5609 * We don't allow RTA_IFP to specify a different ill than the
5609 5610 * one matching the ipif to make sure we can delete the route.
5610 5611 */
5611 5612 match_flags |= MATCH_IRE_GW | MATCH_IRE_ILL;
5612 5613 if (ill == NULL) {
5613 5614 ill = ipif->ipif_ill;
5614 5615 } else if (ill != ipif->ipif_ill) {
5615 5616 ipif_refrele(ipif);
5616 5617 return (EINVAL);
5617 5618 }
5618 5619
5619 5620 /*
5620 5621 * We check for an existing entry at this point.
5621 5622 *
5622 5623 * Since a netmask isn't passed in via the ioctl interface
5623 5624 * (SIOCADDRT), we don't check for a matching netmask in that
5624 5625 * case.
5625 5626 */
5626 5627 if (!ioctl_msg)
5627 5628 match_flags |= MATCH_IRE_MASK;
5628 5629 ire = ire_ftable_lookup_v4(dst_addr, mask, gw_addr,
5629 5630 IRE_INTERFACE, ill, ALL_ZONES, NULL, match_flags, 0, ipst,
5630 5631 NULL);
5631 5632 if (ire != NULL) {
5632 5633 ire_refrele(ire);
5633 5634 ipif_refrele(ipif);
5634 5635 return (EEXIST);
5635 5636 }
5636 5637
5637 5638 /*
5638 5639 * Some software (for example, GateD and Sun Cluster) attempts
5639 5640 * to create (what amount to) IRE_PREFIX routes with the
5640 5641 * loopback address as the gateway. This is primarily done to
5641 5642 * set up prefixes with the RTF_REJECT flag set (for example,
5642 5643 * when generating aggregate routes.)
5643 5644 *
5644 5645 * If the IRE type (as defined by ill->ill_net_type) would be
5645 5646 * IRE_LOOPBACK, then we map the request into a
5646 5647 * IRE_IF_NORESOLVER. We also OR in the RTF_BLACKHOLE flag as
5647 5648 * these interface routes, by definition, can only be that.
5648 5649 *
5649 5650 * Needless to say, the real IRE_LOOPBACK is NOT created by this
5650 5651 * routine, but rather using ire_create() directly.
5651 5652 *
5652 5653 */
5653 5654 type = ill->ill_net_type;
5654 5655 if (type == IRE_LOOPBACK) {
5655 5656 type = IRE_IF_NORESOLVER;
5656 5657 flags |= RTF_BLACKHOLE;
5657 5658 }
5658 5659
5659 5660 /*
5660 5661 * Create a copy of the IRE_IF_NORESOLVER or
5661 5662 * IRE_IF_RESOLVER with the modified address, netmask, and
5662 5663 * gateway.
5663 5664 */
5664 5665 ire = ire_create(
5665 5666 (uchar_t *)&dst_addr,
5666 5667 (uint8_t *)&mask,
5667 5668 (uint8_t *)&gw_addr,
5668 5669 type,
5669 5670 ill,
5670 5671 zoneid,
5671 5672 flags,
5672 5673 NULL,
5673 5674 ipst);
5674 5675 if (ire == NULL) {
5675 5676 ipif_refrele(ipif);
5676 5677 return (ENOMEM);
5677 5678 }
5678 5679
5679 5680 /* src address assigned by the caller? */
5680 5681 if ((src_addr != INADDR_ANY) && (flags & RTF_SETSRC))
5681 5682 ire->ire_setsrc_addr = src_addr;
5682 5683
5683 5684 nire = ire_add(ire);
5684 5685 if (nire == NULL) {
5685 5686 /*
5686 5687 * In the result of failure, ire_add() will have
5687 5688 * already deleted the ire in question, so there
5688 5689 * is no need to do that here.
5689 5690 */
5690 5691 ipif_refrele(ipif);
5691 5692 return (ENOMEM);
5692 5693 }
5693 5694 /*
5694 5695 * Check if it was a duplicate entry. This handles
5695 5696 * the case of two racing route adds for the same route
5696 5697 */
5697 5698 if (nire != ire) {
5698 5699 ire_delete(nire);
5699 5700 ire_refrele(nire);
5700 5701 ipif_refrele(ipif);
5701 5702 return (EEXIST);
5702 5703 }
5703 5704 ire = nire;
5704 5705 goto save_ire;
5705 5706 }
5706 5707
5707 5708 /*
5708 5709 * Get an interface IRE for the specified gateway.
5709 5710 * If we don't have an IRE_IF_NORESOLVER or IRE_IF_RESOLVER for the
5710 5711 * gateway, it is currently unreachable and we fail the request
5711 5712 * accordingly. We reject any RTF_GATEWAY routes where the gateway
5712 5713 * is an IRE_LOCAL or IRE_LOOPBACK.
5713 5714 * If RTA_IFP was specified we look on that particular ill.
5714 5715 */
5715 5716 if (ill != NULL)
5716 5717 match_flags |= MATCH_IRE_ILL;
5717 5718
5718 5719 /* Check whether the gateway is reachable. */
5719 5720 again:
5720 5721 type = IRE_INTERFACE | IRE_LOCAL | IRE_LOOPBACK;
5721 5722 if (flags & RTF_INDIRECT)
5722 5723 type |= IRE_OFFLINK;
5723 5724
5724 5725 gw_ire = ire_ftable_lookup_v4(gw_addr, 0, 0, type, ill,
5725 5726 ALL_ZONES, NULL, match_flags, 0, ipst, NULL);
5726 5727 if (gw_ire == NULL) {
5727 5728 /*
5728 5729 * With IPMP, we allow host routes to influence in.mpathd's
5729 5730 * target selection. However, if the test addresses are on
5730 5731 * their own network, the above lookup will fail since the
5731 5732 * underlying IRE_INTERFACEs are marked hidden. So allow
5732 5733 * hidden test IREs to be found and try again.
5733 5734 */
5734 5735 if (!(match_flags & MATCH_IRE_TESTHIDDEN)) {
5735 5736 match_flags |= MATCH_IRE_TESTHIDDEN;
5736 5737 goto again;
5737 5738 }
5738 5739 if (ipif != NULL)
5739 5740 ipif_refrele(ipif);
5740 5741 return (ENETUNREACH);
5741 5742 }
5742 5743 if (gw_ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK)) {
5743 5744 ire_refrele(gw_ire);
5744 5745 if (ipif != NULL)
5745 5746 ipif_refrele(ipif);
5746 5747 return (ENETUNREACH);
5747 5748 }
5748 5749
5749 5750 if (ill == NULL && !(flags & RTF_INDIRECT)) {
5750 5751 unbound = B_TRUE;
5751 5752 if (ipst->ips_ip_strict_src_multihoming > 0)
5752 5753 ill = gw_ire->ire_ill;
5753 5754 }
5754 5755
5755 5756 /*
5756 5757 * We create one of three types of IREs as a result of this request
5757 5758 * based on the netmask. A netmask of all ones (which is automatically
5758 5759 * assumed when RTF_HOST is set) results in an IRE_HOST being created.
5759 5760 * An all zeroes netmask implies a default route so an IRE_DEFAULT is
5760 5761 * created. Otherwise, an IRE_PREFIX route is created for the
5761 5762 * destination prefix.
5762 5763 */
5763 5764 if (mask == IP_HOST_MASK)
5764 5765 type = IRE_HOST;
5765 5766 else if (mask == 0)
5766 5767 type = IRE_DEFAULT;
5767 5768 else
5768 5769 type = IRE_PREFIX;
5769 5770
5770 5771 /* check for a duplicate entry */
5771 5772 ire = ire_ftable_lookup_v4(dst_addr, mask, gw_addr, type, ill,
5772 5773 ALL_ZONES, NULL, match_flags | MATCH_IRE_MASK | MATCH_IRE_GW,
5773 5774 0, ipst, NULL);
5774 5775 if (ire != NULL) {
5775 5776 if (ipif != NULL)
5776 5777 ipif_refrele(ipif);
5777 5778 ire_refrele(gw_ire);
5778 5779 ire_refrele(ire);
5779 5780 return (EEXIST);
5780 5781 }
5781 5782
5782 5783 /* Security attribute exists */
5783 5784 if (sp != NULL) {
5784 5785 tsol_gcgrp_addr_t ga;
5785 5786
5786 5787 /* find or create the gateway credentials group */
5787 5788 ga.ga_af = AF_INET;
5788 5789 IN6_IPADDR_TO_V4MAPPED(gw_addr, &ga.ga_addr);
5789 5790
5790 5791 /* we hold reference to it upon success */
5791 5792 gcgrp = gcgrp_lookup(&ga, B_TRUE);
5792 5793 if (gcgrp == NULL) {
5793 5794 if (ipif != NULL)
5794 5795 ipif_refrele(ipif);
5795 5796 ire_refrele(gw_ire);
5796 5797 return (ENOMEM);
5797 5798 }
5798 5799
5799 5800 /*
5800 5801 * Create and add the security attribute to the group; a
5801 5802 * reference to the group is made upon allocating a new
5802 5803 * entry successfully. If it finds an already-existing
5803 5804 * entry for the security attribute in the group, it simply
5804 5805 * returns it and no new reference is made to the group.
5805 5806 */
5806 5807 gc = gc_create(sp, gcgrp, &gcgrp_xtraref);
5807 5808 if (gc == NULL) {
5808 5809 if (ipif != NULL)
5809 5810 ipif_refrele(ipif);
5810 5811 /* release reference held by gcgrp_lookup */
5811 5812 GCGRP_REFRELE(gcgrp);
5812 5813 ire_refrele(gw_ire);
5813 5814 return (ENOMEM);
5814 5815 }
5815 5816 }
5816 5817
5817 5818 /* Create the IRE. */
5818 5819 ire = ire_create(
5819 5820 (uchar_t *)&dst_addr, /* dest address */
5820 5821 (uchar_t *)&mask, /* mask */
5821 5822 (uchar_t *)&gw_addr, /* gateway address */
5822 5823 (ushort_t)type, /* IRE type */
5823 5824 ill,
5824 5825 zoneid,
5825 5826 flags,
5826 5827 gc, /* security attribute */
5827 5828 ipst);
5828 5829
5829 5830 /*
5830 5831 * The ire holds a reference to the 'gc' and the 'gc' holds a
5831 5832 * reference to the 'gcgrp'. We can now release the extra reference
5832 5833 * the 'gcgrp' acquired in the gcgrp_lookup, if it was not used.
5833 5834 */
5834 5835 if (gcgrp_xtraref)
5835 5836 GCGRP_REFRELE(gcgrp);
5836 5837 if (ire == NULL) {
5837 5838 if (gc != NULL)
5838 5839 GC_REFRELE(gc);
5839 5840 if (ipif != NULL)
5840 5841 ipif_refrele(ipif);
5841 5842 ire_refrele(gw_ire);
5842 5843 return (ENOMEM);
5843 5844 }
5844 5845
5845 5846 /* Before we add, check if an extra CGTP broadcast is needed */
5846 5847 cgtp_broadcast = ((flags & RTF_MULTIRT) &&
5847 5848 ip_type_v4(ire->ire_addr, ipst) == IRE_BROADCAST);
5848 5849
5849 5850 /* src address assigned by the caller? */
5850 5851 if ((src_addr != INADDR_ANY) && (flags & RTF_SETSRC))
5851 5852 ire->ire_setsrc_addr = src_addr;
5852 5853
5853 5854 ire->ire_unbound = unbound;
5854 5855
5855 5856 /*
5856 5857 * POLICY: should we allow an RTF_HOST with address INADDR_ANY?
5857 5858 * SUN/OS socket stuff does but do we really want to allow 0.0.0.0?
5858 5859 */
5859 5860
5860 5861 /* Add the new IRE. */
5861 5862 nire = ire_add(ire);
5862 5863 if (nire == NULL) {
5863 5864 /*
5864 5865 * In the result of failure, ire_add() will have
5865 5866 * already deleted the ire in question, so there
5866 5867 * is no need to do that here.
5867 5868 */
5868 5869 if (ipif != NULL)
5869 5870 ipif_refrele(ipif);
5870 5871 ire_refrele(gw_ire);
5871 5872 return (ENOMEM);
5872 5873 }
5873 5874 /*
5874 5875 * Check if it was a duplicate entry. This handles
5875 5876 * the case of two racing route adds for the same route
5876 5877 */
5877 5878 if (nire != ire) {
5878 5879 ire_delete(nire);
5879 5880 ire_refrele(nire);
5880 5881 if (ipif != NULL)
5881 5882 ipif_refrele(ipif);
5882 5883 ire_refrele(gw_ire);
5883 5884 return (EEXIST);
5884 5885 }
5885 5886 ire = nire;
5886 5887
5887 5888 if (flags & RTF_MULTIRT) {
5888 5889 /*
5889 5890 * Invoke the CGTP (multirouting) filtering module
5890 5891 * to add the dst address in the filtering database.
5891 5892 * Replicated inbound packets coming from that address
5892 5893 * will be filtered to discard the duplicates.
5893 5894 * It is not necessary to call the CGTP filter hook
5894 5895 * when the dst address is a broadcast or multicast,
5895 5896 * because an IP source address cannot be a broadcast
5896 5897 * or a multicast.
5897 5898 */
5898 5899 if (cgtp_broadcast) {
5899 5900 ip_cgtp_bcast_add(ire, ipst);
5900 5901 goto save_ire;
5901 5902 }
5902 5903 if (ipst->ips_ip_cgtp_filter_ops != NULL &&
5903 5904 !CLASSD(ire->ire_addr)) {
5904 5905 int res;
5905 5906 ipif_t *src_ipif;
5906 5907
5907 5908 /* Find the source address corresponding to gw_ire */
5908 5909 src_ipif = ipif_lookup_addr(gw_ire->ire_gateway_addr,
5909 5910 NULL, zoneid, ipst);
5910 5911 if (src_ipif != NULL) {
5911 5912 res = ipst->ips_ip_cgtp_filter_ops->
5912 5913 cfo_add_dest_v4(
5913 5914 ipst->ips_netstack->netstack_stackid,
5914 5915 ire->ire_addr,
5915 5916 ire->ire_gateway_addr,
5916 5917 ire->ire_setsrc_addr,
5917 5918 src_ipif->ipif_lcl_addr);
5918 5919 ipif_refrele(src_ipif);
5919 5920 } else {
5920 5921 res = EADDRNOTAVAIL;
5921 5922 }
5922 5923 if (res != 0) {
5923 5924 if (ipif != NULL)
5924 5925 ipif_refrele(ipif);
5925 5926 ire_refrele(gw_ire);
5926 5927 ire_delete(ire);
5927 5928 ire_refrele(ire); /* Held in ire_add */
5928 5929 return (res);
5929 5930 }
5930 5931 }
5931 5932 }
5932 5933
5933 5934 save_ire:
5934 5935 if (gw_ire != NULL) {
5935 5936 ire_refrele(gw_ire);
5936 5937 gw_ire = NULL;
5937 5938 }
5938 5939 if (ill != NULL) {
5939 5940 /*
5940 5941 * Save enough information so that we can recreate the IRE if
5941 5942 * the interface goes down and then up. The metrics associated
5942 5943 * with the route will be saved as well when rts_setmetrics() is
5943 5944 * called after the IRE has been created. In the case where
5944 5945 * memory cannot be allocated, none of this information will be
5945 5946 * saved.
5946 5947 */
5947 5948 ill_save_ire(ill, ire);
5948 5949 }
5949 5950 if (ioctl_msg)
5950 5951 ip_rts_rtmsg(RTM_OLDADD, ire, 0, ipst);
5951 5952 if (ire_arg != NULL) {
5952 5953 /*
5953 5954 * Store the ire that was successfully added into where ire_arg
5954 5955 * points to so that callers don't have to look it up
5955 5956 * themselves (but they are responsible for ire_refrele()ing
5956 5957 * the ire when they are finished with it).
5957 5958 */
5958 5959 *ire_arg = ire;
5959 5960 } else {
5960 5961 ire_refrele(ire); /* Held in ire_add */
5961 5962 }
5962 5963 if (ipif != NULL)
5963 5964 ipif_refrele(ipif);
5964 5965 return (0);
5965 5966 }
5966 5967
5967 5968 /*
5968 5969 * ip_rt_delete is called to delete an IPv4 route.
5969 5970 * ill is passed in to associate it with the correct interface.
5970 5971 */
5971 5972 /* ARGSUSED4 */
5972 5973 int
5973 5974 ip_rt_delete(ipaddr_t dst_addr, ipaddr_t mask, ipaddr_t gw_addr,
5974 5975 uint_t rtm_addrs, int flags, ill_t *ill, boolean_t ioctl_msg,
5975 5976 ip_stack_t *ipst, zoneid_t zoneid)
5976 5977 {
5977 5978 ire_t *ire = NULL;
5978 5979 ipif_t *ipif;
5979 5980 uint_t type;
5980 5981 uint_t match_flags = MATCH_IRE_TYPE;
5981 5982 int err = 0;
5982 5983
5983 5984 ip1dbg(("ip_rt_delete:"));
5984 5985 /*
5985 5986 * If this is the case of RTF_HOST being set, then we set the netmask
5986 5987 * to all ones. Otherwise, we use the netmask if one was supplied.
5987 5988 */
5988 5989 if (flags & RTF_HOST) {
5989 5990 mask = IP_HOST_MASK;
5990 5991 match_flags |= MATCH_IRE_MASK;
5991 5992 } else if (rtm_addrs & RTA_NETMASK) {
5992 5993 match_flags |= MATCH_IRE_MASK;
5993 5994 }
5994 5995
5995 5996 /*
5996 5997 * Note that RTF_GATEWAY is never set on a delete, therefore
5997 5998 * we check if the gateway address is one of our interfaces first,
5998 5999 * and fall back on RTF_GATEWAY routes.
5999 6000 *
6000 6001 * This makes it possible to delete an original
6001 6002 * IRE_IF_NORESOLVER/IRE_IF_RESOLVER - consistent with SunOS 4.1.
6002 6003 * However, we have RTF_KERNEL set on the ones created by ipif_up
6003 6004 * and those can not be deleted here.
6004 6005 *
6005 6006 * We use MATCH_IRE_ILL if we know the interface. If the caller
6006 6007 * specified an interface (from the RTA_IFP sockaddr) we use it,
6007 6008 * otherwise we use the ill derived from the gateway address.
6008 6009 * We can always match the gateway address since we record it
6009 6010 * in ire_gateway_addr.
6010 6011 *
6011 6012 * For more detail on specifying routes by gateway address and by
6012 6013 * interface index, see the comments in ip_rt_add().
6013 6014 */
6014 6015 ipif = ipif_lookup_interface(gw_addr, dst_addr, ipst);
6015 6016 if (ipif != NULL) {
6016 6017 ill_t *ill_match;
6017 6018
6018 6019 if (ill != NULL)
6019 6020 ill_match = ill;
6020 6021 else
6021 6022 ill_match = ipif->ipif_ill;
6022 6023
6023 6024 match_flags |= MATCH_IRE_ILL;
6024 6025 if (ipif->ipif_ire_type == IRE_LOOPBACK) {
6025 6026 ire = ire_ftable_lookup_v4(dst_addr, mask, 0,
6026 6027 IRE_LOOPBACK, ill_match, ALL_ZONES, NULL,
6027 6028 match_flags, 0, ipst, NULL);
6028 6029 }
6029 6030 if (ire == NULL) {
6030 6031 match_flags |= MATCH_IRE_GW;
6031 6032 ire = ire_ftable_lookup_v4(dst_addr, mask, gw_addr,
6032 6033 IRE_INTERFACE, ill_match, ALL_ZONES, NULL,
6033 6034 match_flags, 0, ipst, NULL);
6034 6035 }
6035 6036 /* Avoid deleting routes created by kernel from an ipif */
6036 6037 if (ire != NULL && (ire->ire_flags & RTF_KERNEL)) {
6037 6038 ire_refrele(ire);
6038 6039 ire = NULL;
6039 6040 }
6040 6041
6041 6042 /* Restore in case we didn't find a match */
6042 6043 match_flags &= ~(MATCH_IRE_GW|MATCH_IRE_ILL);
6043 6044 }
6044 6045
6045 6046 if (ire == NULL) {
6046 6047 /*
6047 6048 * At this point, the gateway address is not one of our own
6048 6049 * addresses or a matching interface route was not found. We
6049 6050 * set the IRE type to lookup based on whether
6050 6051 * this is a host route, a default route or just a prefix.
6051 6052 *
6052 6053 * If an ill was passed in, then the lookup is based on an
6053 6054 * interface index so MATCH_IRE_ILL is added to match_flags.
6054 6055 */
6055 6056 match_flags |= MATCH_IRE_GW;
6056 6057 if (ill != NULL)
6057 6058 match_flags |= MATCH_IRE_ILL;
6058 6059 if (mask == IP_HOST_MASK)
6059 6060 type = IRE_HOST;
6060 6061 else if (mask == 0)
6061 6062 type = IRE_DEFAULT;
6062 6063 else
6063 6064 type = IRE_PREFIX;
6064 6065 ire = ire_ftable_lookup_v4(dst_addr, mask, gw_addr, type, ill,
6065 6066 ALL_ZONES, NULL, match_flags, 0, ipst, NULL);
6066 6067 }
6067 6068
6068 6069 if (ipif != NULL) {
6069 6070 ipif_refrele(ipif);
6070 6071 ipif = NULL;
6071 6072 }
6072 6073
6073 6074 if (ire == NULL)
6074 6075 return (ESRCH);
6075 6076
6076 6077 if (ire->ire_flags & RTF_MULTIRT) {
6077 6078 /*
6078 6079 * Invoke the CGTP (multirouting) filtering module
6079 6080 * to remove the dst address from the filtering database.
6080 6081 * Packets coming from that address will no longer be
6081 6082 * filtered to remove duplicates.
6082 6083 */
6083 6084 if (ipst->ips_ip_cgtp_filter_ops != NULL) {
6084 6085 err = ipst->ips_ip_cgtp_filter_ops->cfo_del_dest_v4(
6085 6086 ipst->ips_netstack->netstack_stackid,
6086 6087 ire->ire_addr, ire->ire_gateway_addr);
6087 6088 }
6088 6089 ip_cgtp_bcast_delete(ire, ipst);
6089 6090 }
6090 6091
6091 6092 ill = ire->ire_ill;
6092 6093 if (ill != NULL)
6093 6094 ill_remove_saved_ire(ill, ire);
6094 6095 if (ioctl_msg)
6095 6096 ip_rts_rtmsg(RTM_OLDDEL, ire, 0, ipst);
6096 6097 ire_delete(ire);
6097 6098 ire_refrele(ire);
6098 6099 return (err);
6099 6100 }
6100 6101
6101 6102 /*
6102 6103 * ip_siocaddrt is called to complete processing of an SIOCADDRT IOCTL.
6103 6104 */
6104 6105 /* ARGSUSED */
6105 6106 int
6106 6107 ip_siocaddrt(ipif_t *dummy_ipif, sin_t *dummy_sin, queue_t *q, mblk_t *mp,
6107 6108 ip_ioctl_cmd_t *ipip, void *dummy_if_req)
6108 6109 {
6109 6110 ipaddr_t dst_addr;
6110 6111 ipaddr_t gw_addr;
6111 6112 ipaddr_t mask;
6112 6113 int error = 0;
6113 6114 mblk_t *mp1;
6114 6115 struct rtentry *rt;
6115 6116 ipif_t *ipif = NULL;
6116 6117 ip_stack_t *ipst;
6117 6118
6118 6119 ASSERT(q->q_next == NULL);
6119 6120 ipst = CONNQ_TO_IPST(q);
6120 6121
6121 6122 ip1dbg(("ip_siocaddrt:"));
6122 6123 /* Existence of mp1 verified in ip_wput_nondata */
6123 6124 mp1 = mp->b_cont->b_cont;
6124 6125 rt = (struct rtentry *)mp1->b_rptr;
6125 6126
6126 6127 dst_addr = ((sin_t *)&rt->rt_dst)->sin_addr.s_addr;
6127 6128 gw_addr = ((sin_t *)&rt->rt_gateway)->sin_addr.s_addr;
6128 6129
6129 6130 /*
6130 6131 * If the RTF_HOST flag is on, this is a request to assign a gateway
6131 6132 * to a particular host address. In this case, we set the netmask to
6132 6133 * all ones for the particular destination address. Otherwise,
6133 6134 * determine the netmask to be used based on dst_addr and the interfaces
6134 6135 * in use.
6135 6136 */
6136 6137 if (rt->rt_flags & RTF_HOST) {
6137 6138 mask = IP_HOST_MASK;
6138 6139 } else {
6139 6140 /*
6140 6141 * Note that ip_subnet_mask returns a zero mask in the case of
6141 6142 * default (an all-zeroes address).
6142 6143 */
6143 6144 mask = ip_subnet_mask(dst_addr, &ipif, ipst);
6144 6145 }
6145 6146
6146 6147 error = ip_rt_add(dst_addr, mask, gw_addr, 0, rt->rt_flags, NULL, NULL,
6147 6148 B_TRUE, NULL, ipst, ALL_ZONES);
6148 6149 if (ipif != NULL)
6149 6150 ipif_refrele(ipif);
6150 6151 return (error);
6151 6152 }
6152 6153
6153 6154 /*
6154 6155 * ip_siocdelrt is called to complete processing of an SIOCDELRT IOCTL.
6155 6156 */
6156 6157 /* ARGSUSED */
6157 6158 int
6158 6159 ip_siocdelrt(ipif_t *dummy_ipif, sin_t *dummy_sin, queue_t *q, mblk_t *mp,
6159 6160 ip_ioctl_cmd_t *ipip, void *dummy_if_req)
6160 6161 {
6161 6162 ipaddr_t dst_addr;
6162 6163 ipaddr_t gw_addr;
6163 6164 ipaddr_t mask;
6164 6165 int error;
6165 6166 mblk_t *mp1;
6166 6167 struct rtentry *rt;
6167 6168 ipif_t *ipif = NULL;
6168 6169 ip_stack_t *ipst;
6169 6170
6170 6171 ASSERT(q->q_next == NULL);
6171 6172 ipst = CONNQ_TO_IPST(q);
6172 6173
6173 6174 ip1dbg(("ip_siocdelrt:"));
6174 6175 /* Existence of mp1 verified in ip_wput_nondata */
6175 6176 mp1 = mp->b_cont->b_cont;
6176 6177 rt = (struct rtentry *)mp1->b_rptr;
6177 6178
6178 6179 dst_addr = ((sin_t *)&rt->rt_dst)->sin_addr.s_addr;
6179 6180 gw_addr = ((sin_t *)&rt->rt_gateway)->sin_addr.s_addr;
6180 6181
6181 6182 /*
6182 6183 * If the RTF_HOST flag is on, this is a request to delete a gateway
6183 6184 * to a particular host address. In this case, we set the netmask to
6184 6185 * all ones for the particular destination address. Otherwise,
6185 6186 * determine the netmask to be used based on dst_addr and the interfaces
6186 6187 * in use.
6187 6188 */
6188 6189 if (rt->rt_flags & RTF_HOST) {
6189 6190 mask = IP_HOST_MASK;
6190 6191 } else {
6191 6192 /*
6192 6193 * Note that ip_subnet_mask returns a zero mask in the case of
6193 6194 * default (an all-zeroes address).
6194 6195 */
6195 6196 mask = ip_subnet_mask(dst_addr, &ipif, ipst);
6196 6197 }
6197 6198
6198 6199 error = ip_rt_delete(dst_addr, mask, gw_addr,
6199 6200 RTA_DST | RTA_GATEWAY | RTA_NETMASK, rt->rt_flags, NULL, B_TRUE,
6200 6201 ipst, ALL_ZONES);
6201 6202 if (ipif != NULL)
6202 6203 ipif_refrele(ipif);
6203 6204 return (error);
6204 6205 }
6205 6206
6206 6207 /*
6207 6208 * Enqueue the mp onto the ipsq, chained by b_next.
6208 6209 * b_prev stores the function to be executed later, and b_queue the queue
6209 6210 * where this mp originated.
6210 6211 */
6211 6212 void
6212 6213 ipsq_enq(ipsq_t *ipsq, queue_t *q, mblk_t *mp, ipsq_func_t func, int type,
6213 6214 ill_t *pending_ill)
6214 6215 {
6215 6216 conn_t *connp;
6216 6217 ipxop_t *ipx = ipsq->ipsq_xop;
6217 6218
6218 6219 ASSERT(MUTEX_HELD(&ipsq->ipsq_lock));
6219 6220 ASSERT(MUTEX_HELD(&ipx->ipx_lock));
6220 6221 ASSERT(func != NULL);
6221 6222
6222 6223 mp->b_queue = q;
6223 6224 mp->b_prev = (void *)func;
6224 6225 mp->b_next = NULL;
6225 6226
6226 6227 switch (type) {
6227 6228 case CUR_OP:
6228 6229 if (ipx->ipx_mptail != NULL) {
6229 6230 ASSERT(ipx->ipx_mphead != NULL);
6230 6231 ipx->ipx_mptail->b_next = mp;
6231 6232 } else {
6232 6233 ASSERT(ipx->ipx_mphead == NULL);
6233 6234 ipx->ipx_mphead = mp;
6234 6235 }
6235 6236 ipx->ipx_mptail = mp;
6236 6237 break;
6237 6238
6238 6239 case NEW_OP:
6239 6240 if (ipsq->ipsq_xopq_mptail != NULL) {
6240 6241 ASSERT(ipsq->ipsq_xopq_mphead != NULL);
6241 6242 ipsq->ipsq_xopq_mptail->b_next = mp;
6242 6243 } else {
6243 6244 ASSERT(ipsq->ipsq_xopq_mphead == NULL);
6244 6245 ipsq->ipsq_xopq_mphead = mp;
6245 6246 }
6246 6247 ipsq->ipsq_xopq_mptail = mp;
6247 6248 ipx->ipx_ipsq_queued = B_TRUE;
6248 6249 break;
6249 6250
6250 6251 case SWITCH_OP:
6251 6252 ASSERT(ipsq->ipsq_swxop != NULL);
6252 6253 /* only one switch operation is currently allowed */
6253 6254 ASSERT(ipsq->ipsq_switch_mp == NULL);
6254 6255 ipsq->ipsq_switch_mp = mp;
6255 6256 ipx->ipx_ipsq_queued = B_TRUE;
6256 6257 break;
6257 6258 default:
6258 6259 cmn_err(CE_PANIC, "ipsq_enq %d type \n", type);
6259 6260 }
6260 6261
6261 6262 if (CONN_Q(q) && pending_ill != NULL) {
6262 6263 connp = Q_TO_CONN(q);
6263 6264 ASSERT(MUTEX_HELD(&connp->conn_lock));
6264 6265 connp->conn_oper_pending_ill = pending_ill;
6265 6266 }
6266 6267 }
6267 6268
6268 6269 /*
6269 6270 * Dequeue the next message that requested exclusive access to this IPSQ's
6270 6271 * xop. Specifically:
6271 6272 *
6272 6273 * 1. If we're still processing the current operation on `ipsq', then
6273 6274 * dequeue the next message for the operation (from ipx_mphead), or
6274 6275 * return NULL if there are no queued messages for the operation.
6275 6276 * These messages are queued via CUR_OP to qwriter_ip() and friends.
6276 6277 *
6277 6278 * 2. If the current operation on `ipsq' has completed (ipx_current_ipif is
6278 6279 * not set) see if the ipsq has requested an xop switch. If so, switch
6279 6280 * `ipsq' to a different xop. Xop switches only happen when joining or
6280 6281 * leaving IPMP groups and require a careful dance -- see the comments
6281 6282 * in-line below for details. If we're leaving a group xop or if we're
6282 6283 * joining a group xop and become writer on it, then we proceed to (3).
6283 6284 * Otherwise, we return NULL and exit the xop.
6284 6285 *
6285 6286 * 3. For each IPSQ in the xop, return any switch operation stored on
6286 6287 * ipsq_switch_mp (set via SWITCH_OP); these must be processed before
6287 6288 * any other messages queued on the IPSQ. Otherwise, dequeue the next
6288 6289 * exclusive operation (queued via NEW_OP) stored on ipsq_xopq_mphead.
6289 6290 * Note that if the phyint tied to `ipsq' is not using IPMP there will
6290 6291 * only be one IPSQ in the xop. Otherwise, there will be one IPSQ for
6291 6292 * each phyint in the group, including the IPMP meta-interface phyint.
6292 6293 */
6293 6294 static mblk_t *
6294 6295 ipsq_dq(ipsq_t *ipsq)
6295 6296 {
6296 6297 ill_t *illv4, *illv6;
6297 6298 mblk_t *mp;
6298 6299 ipsq_t *xopipsq;
6299 6300 ipsq_t *leftipsq = NULL;
6300 6301 ipxop_t *ipx;
6301 6302 phyint_t *phyi = ipsq->ipsq_phyint;
6302 6303 ip_stack_t *ipst = ipsq->ipsq_ipst;
6303 6304 boolean_t emptied = B_FALSE;
6304 6305
6305 6306 /*
6306 6307 * Grab all the locks we need in the defined order (ill_g_lock ->
6307 6308 * ipsq_lock -> ipx_lock); ill_g_lock is needed to use ipsq_next.
6308 6309 */
6309 6310 rw_enter(&ipst->ips_ill_g_lock,
6310 6311 ipsq->ipsq_swxop != NULL ? RW_WRITER : RW_READER);
6311 6312 mutex_enter(&ipsq->ipsq_lock);
6312 6313 ipx = ipsq->ipsq_xop;
6313 6314 mutex_enter(&ipx->ipx_lock);
6314 6315
6315 6316 /*
6316 6317 * Dequeue the next message associated with the current exclusive
6317 6318 * operation, if any.
6318 6319 */
6319 6320 if ((mp = ipx->ipx_mphead) != NULL) {
6320 6321 ipx->ipx_mphead = mp->b_next;
6321 6322 if (ipx->ipx_mphead == NULL)
6322 6323 ipx->ipx_mptail = NULL;
6323 6324 mp->b_next = (void *)ipsq;
6324 6325 goto out;
6325 6326 }
6326 6327
6327 6328 if (ipx->ipx_current_ipif != NULL)
6328 6329 goto empty;
6329 6330
6330 6331 if (ipsq->ipsq_swxop != NULL) {
6331 6332 /*
6332 6333 * The exclusive operation that is now being completed has
6333 6334 * requested a switch to a different xop. This happens
6334 6335 * when an interface joins or leaves an IPMP group. Joins
6335 6336 * happen through SIOCSLIFGROUPNAME (ip_sioctl_groupname()).
6336 6337 * Leaves happen via SIOCSLIFGROUPNAME, interface unplumb
6337 6338 * (phyint_free()), or interface plumb for an ill type
6338 6339 * not in the IPMP group (ip_rput_dlpi_writer()).
6339 6340 *
6340 6341 * Xop switches are not allowed on the IPMP meta-interface.
6341 6342 */
6342 6343 ASSERT(phyi == NULL || !(phyi->phyint_flags & PHYI_IPMP));
6343 6344 ASSERT(RW_WRITE_HELD(&ipst->ips_ill_g_lock));
6344 6345 DTRACE_PROBE1(ipsq__switch, (ipsq_t *), ipsq);
6345 6346
6346 6347 if (ipsq->ipsq_swxop == &ipsq->ipsq_ownxop) {
6347 6348 /*
6348 6349 * We're switching back to our own xop, so we have two
6349 6350 * xop's to drain/exit: our own, and the group xop
6350 6351 * that we are leaving.
6351 6352 *
6352 6353 * First, pull ourselves out of the group ipsq list.
6353 6354 * This is safe since we're writer on ill_g_lock.
6354 6355 */
6355 6356 ASSERT(ipsq->ipsq_xop != &ipsq->ipsq_ownxop);
6356 6357
6357 6358 xopipsq = ipx->ipx_ipsq;
6358 6359 while (xopipsq->ipsq_next != ipsq)
6359 6360 xopipsq = xopipsq->ipsq_next;
6360 6361
6361 6362 xopipsq->ipsq_next = ipsq->ipsq_next;
6362 6363 ipsq->ipsq_next = ipsq;
6363 6364 ipsq->ipsq_xop = ipsq->ipsq_swxop;
6364 6365 ipsq->ipsq_swxop = NULL;
6365 6366
6366 6367 /*
6367 6368 * Second, prepare to exit the group xop. The actual
6368 6369 * ipsq_exit() is done at the end of this function
6369 6370 * since we cannot hold any locks across ipsq_exit().
6370 6371 * Note that although we drop the group's ipx_lock, no
6371 6372 * threads can proceed since we're still ipx_writer.
6372 6373 */
6373 6374 leftipsq = xopipsq;
6374 6375 mutex_exit(&ipx->ipx_lock);
6375 6376
6376 6377 /*
6377 6378 * Third, set ipx to point to our own xop (which was
6378 6379 * inactive and therefore can be entered).
6379 6380 */
6380 6381 ipx = ipsq->ipsq_xop;
6381 6382 mutex_enter(&ipx->ipx_lock);
6382 6383 ASSERT(ipx->ipx_writer == NULL);
6383 6384 ASSERT(ipx->ipx_current_ipif == NULL);
6384 6385 } else {
6385 6386 /*
6386 6387 * We're switching from our own xop to a group xop.
6387 6388 * The requestor of the switch must ensure that the
6388 6389 * group xop cannot go away (e.g. by ensuring the
6389 6390 * phyint associated with the xop cannot go away).
6390 6391 *
6391 6392 * If we can become writer on our new xop, then we'll
6392 6393 * do the drain. Otherwise, the current writer of our
6393 6394 * new xop will do the drain when it exits.
6394 6395 *
6395 6396 * First, splice ourselves into the group IPSQ list.
6396 6397 * This is safe since we're writer on ill_g_lock.
6397 6398 */
6398 6399 ASSERT(ipsq->ipsq_xop == &ipsq->ipsq_ownxop);
6399 6400
6400 6401 xopipsq = ipsq->ipsq_swxop->ipx_ipsq;
6401 6402 while (xopipsq->ipsq_next != ipsq->ipsq_swxop->ipx_ipsq)
6402 6403 xopipsq = xopipsq->ipsq_next;
6403 6404
6404 6405 xopipsq->ipsq_next = ipsq;
6405 6406 ipsq->ipsq_next = ipsq->ipsq_swxop->ipx_ipsq;
6406 6407 ipsq->ipsq_xop = ipsq->ipsq_swxop;
6407 6408 ipsq->ipsq_swxop = NULL;
6408 6409
6409 6410 /*
6410 6411 * Second, exit our own xop, since it's now unused.
6411 6412 * This is safe since we've got the only reference.
6412 6413 */
6413 6414 ASSERT(ipx->ipx_writer == curthread);
6414 6415 ipx->ipx_writer = NULL;
6415 6416 VERIFY(--ipx->ipx_reentry_cnt == 0);
6416 6417 ipx->ipx_ipsq_queued = B_FALSE;
6417 6418 mutex_exit(&ipx->ipx_lock);
6418 6419
6419 6420 /*
6420 6421 * Third, set ipx to point to our new xop, and check
6421 6422 * if we can become writer on it. If we cannot, then
6422 6423 * the current writer will drain the IPSQ group when
6423 6424 * it exits. Our ipsq_xop is guaranteed to be stable
6424 6425 * because we're still holding ipsq_lock.
6425 6426 */
6426 6427 ipx = ipsq->ipsq_xop;
6427 6428 mutex_enter(&ipx->ipx_lock);
6428 6429 if (ipx->ipx_writer != NULL ||
6429 6430 ipx->ipx_current_ipif != NULL) {
6430 6431 goto out;
6431 6432 }
6432 6433 }
6433 6434
6434 6435 /*
6435 6436 * Fourth, become writer on our new ipx before we continue
6436 6437 * with the drain. Note that we never dropped ipsq_lock
6437 6438 * above, so no other thread could've raced with us to
6438 6439 * become writer first. Also, we're holding ipx_lock, so
6439 6440 * no other thread can examine the ipx right now.
6440 6441 */
6441 6442 ASSERT(ipx->ipx_current_ipif == NULL);
6442 6443 ASSERT(ipx->ipx_mphead == NULL && ipx->ipx_mptail == NULL);
6443 6444 VERIFY(ipx->ipx_reentry_cnt++ == 0);
6444 6445 ipx->ipx_writer = curthread;
6445 6446 ipx->ipx_forced = B_FALSE;
6446 6447 #ifdef DEBUG
6447 6448 ipx->ipx_depth = getpcstack(ipx->ipx_stack, IPX_STACK_DEPTH);
6448 6449 #endif
6449 6450 }
6450 6451
6451 6452 xopipsq = ipsq;
6452 6453 do {
6453 6454 /*
6454 6455 * So that other operations operate on a consistent and
6455 6456 * complete phyint, a switch message on an IPSQ must be
6456 6457 * handled prior to any other operations on that IPSQ.
6457 6458 */
6458 6459 if ((mp = xopipsq->ipsq_switch_mp) != NULL) {
6459 6460 xopipsq->ipsq_switch_mp = NULL;
6460 6461 ASSERT(mp->b_next == NULL);
6461 6462 mp->b_next = (void *)xopipsq;
6462 6463 goto out;
6463 6464 }
6464 6465
6465 6466 if ((mp = xopipsq->ipsq_xopq_mphead) != NULL) {
6466 6467 xopipsq->ipsq_xopq_mphead = mp->b_next;
6467 6468 if (xopipsq->ipsq_xopq_mphead == NULL)
6468 6469 xopipsq->ipsq_xopq_mptail = NULL;
6469 6470 mp->b_next = (void *)xopipsq;
6470 6471 goto out;
6471 6472 }
6472 6473 } while ((xopipsq = xopipsq->ipsq_next) != ipsq);
6473 6474 empty:
6474 6475 /*
6475 6476 * There are no messages. Further, we are holding ipx_lock, hence no
6476 6477 * new messages can end up on any IPSQ in the xop.
6477 6478 */
6478 6479 ipx->ipx_writer = NULL;
6479 6480 ipx->ipx_forced = B_FALSE;
6480 6481 VERIFY(--ipx->ipx_reentry_cnt == 0);
6481 6482 ipx->ipx_ipsq_queued = B_FALSE;
6482 6483 emptied = B_TRUE;
6483 6484 #ifdef DEBUG
6484 6485 ipx->ipx_depth = 0;
6485 6486 #endif
6486 6487 out:
6487 6488 mutex_exit(&ipx->ipx_lock);
6488 6489 mutex_exit(&ipsq->ipsq_lock);
6489 6490
6490 6491 /*
6491 6492 * If we completely emptied the xop, then wake up any threads waiting
6492 6493 * to enter any of the IPSQ's associated with it.
6493 6494 */
6494 6495 if (emptied) {
6495 6496 xopipsq = ipsq;
6496 6497 do {
6497 6498 if ((phyi = xopipsq->ipsq_phyint) == NULL)
6498 6499 continue;
6499 6500
6500 6501 illv4 = phyi->phyint_illv4;
6501 6502 illv6 = phyi->phyint_illv6;
6502 6503
6503 6504 GRAB_ILL_LOCKS(illv4, illv6);
6504 6505 if (illv4 != NULL)
6505 6506 cv_broadcast(&illv4->ill_cv);
6506 6507 if (illv6 != NULL)
6507 6508 cv_broadcast(&illv6->ill_cv);
6508 6509 RELEASE_ILL_LOCKS(illv4, illv6);
6509 6510 } while ((xopipsq = xopipsq->ipsq_next) != ipsq);
6510 6511 }
6511 6512 rw_exit(&ipst->ips_ill_g_lock);
6512 6513
6513 6514 /*
6514 6515 * Now that all locks are dropped, exit the IPSQ we left.
6515 6516 */
6516 6517 if (leftipsq != NULL)
6517 6518 ipsq_exit(leftipsq);
6518 6519
6519 6520 return (mp);
6520 6521 }
6521 6522
6522 6523 /*
6523 6524 * Return completion status of previously initiated DLPI operations on
6524 6525 * ills in the purview of an ipsq.
6525 6526 */
6526 6527 static boolean_t
6527 6528 ipsq_dlpi_done(ipsq_t *ipsq)
6528 6529 {
6529 6530 ipsq_t *ipsq_start;
6530 6531 phyint_t *phyi;
6531 6532 ill_t *ill;
6532 6533
6533 6534 ASSERT(RW_LOCK_HELD(&ipsq->ipsq_ipst->ips_ill_g_lock));
6534 6535 ipsq_start = ipsq;
6535 6536
6536 6537 do {
6537 6538 /*
6538 6539 * The only current users of this function are ipsq_try_enter
6539 6540 * and ipsq_enter which have made sure that ipsq_writer is
6540 6541 * NULL before we reach here. ill_dlpi_pending is modified
6541 6542 * only by an ipsq writer
6542 6543 */
6543 6544 ASSERT(ipsq->ipsq_xop->ipx_writer == NULL);
6544 6545 phyi = ipsq->ipsq_phyint;
6545 6546 /*
6546 6547 * phyi could be NULL if a phyint that is part of an
6547 6548 * IPMP group is being unplumbed. A more detailed
6548 6549 * comment is in ipmp_grp_update_kstats()
6549 6550 */
6550 6551 if (phyi != NULL) {
6551 6552 ill = phyi->phyint_illv4;
6552 6553 if (ill != NULL &&
6553 6554 (ill->ill_dlpi_pending != DL_PRIM_INVAL ||
6554 6555 ill->ill_arl_dlpi_pending))
6555 6556 return (B_FALSE);
6556 6557
6557 6558 ill = phyi->phyint_illv6;
6558 6559 if (ill != NULL &&
6559 6560 ill->ill_dlpi_pending != DL_PRIM_INVAL)
6560 6561 return (B_FALSE);
6561 6562 }
6562 6563
6563 6564 } while ((ipsq = ipsq->ipsq_next) != ipsq_start);
6564 6565
6565 6566 return (B_TRUE);
6566 6567 }
6567 6568
6568 6569 /*
6569 6570 * Enter the ipsq corresponding to ill, by waiting synchronously till
6570 6571 * we can enter the ipsq exclusively. Unless 'force' is used, the ipsq
6571 6572 * will have to drain completely before ipsq_enter returns success.
6572 6573 * ipx_current_ipif will be set if some exclusive op is in progress,
6573 6574 * and the ipsq_exit logic will start the next enqueued op after
6574 6575 * completion of the current op. If 'force' is used, we don't wait
6575 6576 * for the enqueued ops. This is needed when a conn_close wants to
6576 6577 * enter the ipsq and abort an ioctl that is somehow stuck. Unplumb
6577 6578 * of an ill can also use this option. But we dont' use it currently.
6578 6579 */
6579 6580 #define ENTER_SQ_WAIT_TICKS 100
6580 6581 boolean_t
6581 6582 ipsq_enter(ill_t *ill, boolean_t force, int type)
6582 6583 {
6583 6584 ipsq_t *ipsq;
6584 6585 ipxop_t *ipx;
6585 6586 boolean_t waited_enough = B_FALSE;
6586 6587 ip_stack_t *ipst = ill->ill_ipst;
6587 6588
6588 6589 /*
6589 6590 * Note that the relationship between ill and ipsq is fixed as long as
6590 6591 * the ill is not ILL_CONDEMNED. Holding ipsq_lock ensures the
6591 6592 * relationship between the IPSQ and xop cannot change. However,
6592 6593 * since we cannot hold ipsq_lock across the cv_wait(), it may change
6593 6594 * while we're waiting. We wait on ill_cv and rely on ipsq_exit()
6594 6595 * waking up all ills in the xop when it becomes available.
6595 6596 */
6596 6597 for (;;) {
6597 6598 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
6598 6599 mutex_enter(&ill->ill_lock);
6599 6600 if (ill->ill_state_flags & ILL_CONDEMNED) {
6600 6601 mutex_exit(&ill->ill_lock);
6601 6602 rw_exit(&ipst->ips_ill_g_lock);
6602 6603 return (B_FALSE);
6603 6604 }
6604 6605
6605 6606 ipsq = ill->ill_phyint->phyint_ipsq;
6606 6607 mutex_enter(&ipsq->ipsq_lock);
6607 6608 ipx = ipsq->ipsq_xop;
6608 6609 mutex_enter(&ipx->ipx_lock);
6609 6610
6610 6611 if (ipx->ipx_writer == NULL && (type == CUR_OP ||
6611 6612 (ipx->ipx_current_ipif == NULL && ipsq_dlpi_done(ipsq)) ||
6612 6613 waited_enough))
6613 6614 break;
6614 6615
6615 6616 rw_exit(&ipst->ips_ill_g_lock);
6616 6617
6617 6618 if (!force || ipx->ipx_writer != NULL) {
6618 6619 mutex_exit(&ipx->ipx_lock);
6619 6620 mutex_exit(&ipsq->ipsq_lock);
6620 6621 cv_wait(&ill->ill_cv, &ill->ill_lock);
6621 6622 } else {
6622 6623 mutex_exit(&ipx->ipx_lock);
6623 6624 mutex_exit(&ipsq->ipsq_lock);
6624 6625 (void) cv_reltimedwait(&ill->ill_cv,
6625 6626 &ill->ill_lock, ENTER_SQ_WAIT_TICKS, TR_CLOCK_TICK);
6626 6627 waited_enough = B_TRUE;
6627 6628 }
6628 6629 mutex_exit(&ill->ill_lock);
6629 6630 }
6630 6631
6631 6632 ASSERT(ipx->ipx_mphead == NULL && ipx->ipx_mptail == NULL);
6632 6633 ASSERT(ipx->ipx_reentry_cnt == 0);
6633 6634 ipx->ipx_writer = curthread;
6634 6635 ipx->ipx_forced = (ipx->ipx_current_ipif != NULL);
6635 6636 ipx->ipx_reentry_cnt++;
6636 6637 #ifdef DEBUG
6637 6638 ipx->ipx_depth = getpcstack(ipx->ipx_stack, IPX_STACK_DEPTH);
6638 6639 #endif
6639 6640 mutex_exit(&ipx->ipx_lock);
6640 6641 mutex_exit(&ipsq->ipsq_lock);
6641 6642 mutex_exit(&ill->ill_lock);
6642 6643 rw_exit(&ipst->ips_ill_g_lock);
6643 6644
6644 6645 return (B_TRUE);
6645 6646 }
6646 6647
6647 6648 /*
6648 6649 * ipif_set_values() has a constraint that it cannot drop the ips_ill_g_lock
6649 6650 * across the call to the core interface ipsq_try_enter() and hence calls this
6650 6651 * function directly. This is explained more fully in ipif_set_values().
6651 6652 * In order to support the above constraint, ipsq_try_enter is implemented as
6652 6653 * a wrapper that grabs the ips_ill_g_lock and calls this function subsequently
6653 6654 */
6654 6655 static ipsq_t *
6655 6656 ipsq_try_enter_internal(ill_t *ill, queue_t *q, mblk_t *mp, ipsq_func_t func,
6656 6657 int type, boolean_t reentry_ok)
6657 6658 {
6658 6659 ipsq_t *ipsq;
6659 6660 ipxop_t *ipx;
6660 6661 ip_stack_t *ipst = ill->ill_ipst;
6661 6662
6662 6663 /*
6663 6664 * lock ordering:
6664 6665 * ill_g_lock -> conn_lock -> ill_lock -> ipsq_lock -> ipx_lock.
6665 6666 *
6666 6667 * ipx of an ipsq can't change when ipsq_lock is held.
6667 6668 */
6668 6669 ASSERT(RW_LOCK_HELD(&ipst->ips_ill_g_lock));
6669 6670 GRAB_CONN_LOCK(q);
6670 6671 mutex_enter(&ill->ill_lock);
6671 6672 ipsq = ill->ill_phyint->phyint_ipsq;
6672 6673 mutex_enter(&ipsq->ipsq_lock);
6673 6674 ipx = ipsq->ipsq_xop;
6674 6675 mutex_enter(&ipx->ipx_lock);
6675 6676
6676 6677 /*
6677 6678 * 1. Enter the ipsq if we are already writer and reentry is ok.
6678 6679 * (Note: If the caller does not specify reentry_ok then neither
6679 6680 * 'func' nor any of its callees must ever attempt to enter the ipsq
6680 6681 * again. Otherwise it can lead to an infinite loop
6681 6682 * 2. Enter the ipsq if there is no current writer and this attempted
6682 6683 * entry is part of the current operation
6683 6684 * 3. Enter the ipsq if there is no current writer and this is a new
6684 6685 * operation and the operation queue is empty and there is no
6685 6686 * operation currently in progress and if all previously initiated
6686 6687 * DLPI operations have completed.
6687 6688 */
6688 6689 if ((ipx->ipx_writer == curthread && reentry_ok) ||
6689 6690 (ipx->ipx_writer == NULL && (type == CUR_OP || (type == NEW_OP &&
6690 6691 !ipx->ipx_ipsq_queued && ipx->ipx_current_ipif == NULL &&
6691 6692 ipsq_dlpi_done(ipsq))))) {
6692 6693 /* Success. */
6693 6694 ipx->ipx_reentry_cnt++;
6694 6695 ipx->ipx_writer = curthread;
6695 6696 ipx->ipx_forced = B_FALSE;
6696 6697 mutex_exit(&ipx->ipx_lock);
6697 6698 mutex_exit(&ipsq->ipsq_lock);
6698 6699 mutex_exit(&ill->ill_lock);
6699 6700 RELEASE_CONN_LOCK(q);
6700 6701 #ifdef DEBUG
6701 6702 ipx->ipx_depth = getpcstack(ipx->ipx_stack, IPX_STACK_DEPTH);
6702 6703 #endif
6703 6704 return (ipsq);
6704 6705 }
6705 6706
6706 6707 if (func != NULL)
6707 6708 ipsq_enq(ipsq, q, mp, func, type, ill);
6708 6709
6709 6710 mutex_exit(&ipx->ipx_lock);
6710 6711 mutex_exit(&ipsq->ipsq_lock);
6711 6712 mutex_exit(&ill->ill_lock);
6712 6713 RELEASE_CONN_LOCK(q);
6713 6714 return (NULL);
6714 6715 }
6715 6716
6716 6717 /*
6717 6718 * The ipsq_t (ipsq) is the synchronization data structure used to serialize
6718 6719 * certain critical operations like plumbing (i.e. most set ioctls), etc.
6719 6720 * There is one ipsq per phyint. The ipsq
6720 6721 * serializes exclusive ioctls issued by applications on a per ipsq basis in
6721 6722 * ipsq_xopq_mphead. It also protects against multiple threads executing in
6722 6723 * the ipsq. Responses from the driver pertain to the current ioctl (say a
6723 6724 * DL_BIND_ACK in response to a DL_BIND_REQ initiated as part of bringing
6724 6725 * up the interface) and are enqueued in ipx_mphead.
6725 6726 *
6726 6727 * If a thread does not want to reenter the ipsq when it is already writer,
6727 6728 * it must make sure that the specified reentry point to be called later
6728 6729 * when the ipsq is empty, nor any code path starting from the specified reentry
6729 6730 * point must never ever try to enter the ipsq again. Otherwise it can lead
6730 6731 * to an infinite loop. The reentry point ip_rput_dlpi_writer is an example.
6731 6732 * When the thread that is currently exclusive finishes, it (ipsq_exit)
6732 6733 * dequeues the requests waiting to become exclusive in ipx_mphead and calls
6733 6734 * the reentry point. When the list at ipx_mphead becomes empty ipsq_exit
6734 6735 * proceeds to dequeue the next ioctl in ipsq_xopq_mphead and start the next
6735 6736 * ioctl if the current ioctl has completed. If the current ioctl is still
6736 6737 * in progress it simply returns. The current ioctl could be waiting for
6737 6738 * a response from another module (the driver or could be waiting for
6738 6739 * the ipif/ill/ire refcnts to drop to zero. In such a case the ipx_pending_mp
6739 6740 * and ipx_pending_ipif are set. ipx_current_ipif is set throughout the
6740 6741 * execution of the ioctl and ipsq_exit does not start the next ioctl unless
6741 6742 * ipx_current_ipif is NULL which happens only once the ioctl is complete and
6742 6743 * all associated DLPI operations have completed.
6743 6744 */
6744 6745
6745 6746 /*
6746 6747 * Try to enter the IPSQ corresponding to `ipif' or `ill' exclusively (`ipif'
6747 6748 * and `ill' cannot both be specified). Returns a pointer to the entered IPSQ
6748 6749 * on success, or NULL on failure. The caller ensures ipif/ill is valid by
6749 6750 * refholding it as necessary. If the IPSQ cannot be entered and `func' is
6750 6751 * non-NULL, then `func' will be called back with `q' and `mp' once the IPSQ
6751 6752 * can be entered. If `func' is NULL, then `q' and `mp' are ignored.
6752 6753 */
6753 6754 ipsq_t *
6754 6755 ipsq_try_enter(ipif_t *ipif, ill_t *ill, queue_t *q, mblk_t *mp,
6755 6756 ipsq_func_t func, int type, boolean_t reentry_ok)
6756 6757 {
6757 6758 ip_stack_t *ipst;
6758 6759 ipsq_t *ipsq;
6759 6760
6760 6761 /* Only 1 of ipif or ill can be specified */
6761 6762 ASSERT((ipif != NULL) ^ (ill != NULL));
6762 6763
6763 6764 if (ipif != NULL)
6764 6765 ill = ipif->ipif_ill;
6765 6766 ipst = ill->ill_ipst;
6766 6767
6767 6768 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
6768 6769 ipsq = ipsq_try_enter_internal(ill, q, mp, func, type, reentry_ok);
6769 6770 rw_exit(&ipst->ips_ill_g_lock);
6770 6771
6771 6772 return (ipsq);
6772 6773 }
6773 6774
6774 6775 /*
6775 6776 * Try to enter the IPSQ corresponding to `ill' as writer. The caller ensures
6776 6777 * ill is valid by refholding it if necessary; we will refrele. If the IPSQ
6777 6778 * cannot be entered, the mp is queued for completion.
6778 6779 */
6779 6780 void
6780 6781 qwriter_ip(ill_t *ill, queue_t *q, mblk_t *mp, ipsq_func_t func, int type,
6781 6782 boolean_t reentry_ok)
6782 6783 {
6783 6784 ipsq_t *ipsq;
6784 6785
6785 6786 ipsq = ipsq_try_enter(NULL, ill, q, mp, func, type, reentry_ok);
6786 6787
6787 6788 /*
6788 6789 * Drop the caller's refhold on the ill. This is safe since we either
6789 6790 * entered the IPSQ (and thus are exclusive), or failed to enter the
6790 6791 * IPSQ, in which case we return without accessing ill anymore. This
6791 6792 * is needed because func needs to see the correct refcount.
6792 6793 * e.g. removeif can work only then.
6793 6794 */
6794 6795 ill_refrele(ill);
6795 6796 if (ipsq != NULL) {
6796 6797 (*func)(ipsq, q, mp, NULL);
6797 6798 ipsq_exit(ipsq);
6798 6799 }
6799 6800 }
6800 6801
6801 6802 /*
6802 6803 * Exit the specified IPSQ. If this is the final exit on it then drain it
6803 6804 * prior to exiting. Caller must be writer on the specified IPSQ.
6804 6805 */
6805 6806 void
6806 6807 ipsq_exit(ipsq_t *ipsq)
6807 6808 {
6808 6809 mblk_t *mp;
6809 6810 ipsq_t *mp_ipsq;
6810 6811 queue_t *q;
6811 6812 phyint_t *phyi;
6812 6813 ipsq_func_t func;
6813 6814
6814 6815 ASSERT(IAM_WRITER_IPSQ(ipsq));
6815 6816
6816 6817 ASSERT(ipsq->ipsq_xop->ipx_reentry_cnt >= 1);
6817 6818 if (ipsq->ipsq_xop->ipx_reentry_cnt != 1) {
6818 6819 ipsq->ipsq_xop->ipx_reentry_cnt--;
6819 6820 return;
6820 6821 }
6821 6822
6822 6823 for (;;) {
6823 6824 phyi = ipsq->ipsq_phyint;
6824 6825 mp = ipsq_dq(ipsq);
6825 6826 mp_ipsq = (mp == NULL) ? NULL : (ipsq_t *)mp->b_next;
6826 6827
6827 6828 /*
6828 6829 * If we've changed to a new IPSQ, and the phyint associated
6829 6830 * with the old one has gone away, free the old IPSQ. Note
6830 6831 * that this cannot happen while the IPSQ is in a group.
6831 6832 */
6832 6833 if (mp_ipsq != ipsq && phyi == NULL) {
6833 6834 ASSERT(ipsq->ipsq_next == ipsq);
6834 6835 ASSERT(ipsq->ipsq_xop == &ipsq->ipsq_ownxop);
6835 6836 ipsq_delete(ipsq);
6836 6837 }
6837 6838
6838 6839 if (mp == NULL)
6839 6840 break;
6840 6841
6841 6842 q = mp->b_queue;
6842 6843 func = (ipsq_func_t)mp->b_prev;
6843 6844 ipsq = mp_ipsq;
6844 6845 mp->b_next = mp->b_prev = NULL;
6845 6846 mp->b_queue = NULL;
6846 6847
6847 6848 /*
6848 6849 * If 'q' is an conn queue, it is valid, since we did a
6849 6850 * a refhold on the conn at the start of the ioctl.
6850 6851 * If 'q' is an ill queue, it is valid, since close of an
6851 6852 * ill will clean up its IPSQ.
6852 6853 */
6853 6854 (*func)(ipsq, q, mp, NULL);
6854 6855 }
6855 6856 }
6856 6857
6857 6858 /*
6858 6859 * Used to start any igmp or mld timers that could not be started
6859 6860 * while holding ill_mcast_lock. The timers can't be started while holding
6860 6861 * the lock, since mld/igmp_start_timers may need to call untimeout()
6861 6862 * which can't be done while holding the lock which the timeout handler
6862 6863 * acquires. Otherwise
6863 6864 * there could be a deadlock since the timeout handlers
6864 6865 * mld_timeout_handler_per_ill/igmp_timeout_handler_per_ill also acquire
6865 6866 * ill_mcast_lock.
6866 6867 */
6867 6868 void
6868 6869 ill_mcast_timer_start(ip_stack_t *ipst)
6869 6870 {
6870 6871 int next;
6871 6872
6872 6873 mutex_enter(&ipst->ips_igmp_timer_lock);
6873 6874 next = ipst->ips_igmp_deferred_next;
6874 6875 ipst->ips_igmp_deferred_next = INFINITY;
6875 6876 mutex_exit(&ipst->ips_igmp_timer_lock);
6876 6877
6877 6878 if (next != INFINITY)
6878 6879 igmp_start_timers(next, ipst);
6879 6880
6880 6881 mutex_enter(&ipst->ips_mld_timer_lock);
6881 6882 next = ipst->ips_mld_deferred_next;
6882 6883 ipst->ips_mld_deferred_next = INFINITY;
6883 6884 mutex_exit(&ipst->ips_mld_timer_lock);
6884 6885
6885 6886 if (next != INFINITY)
6886 6887 mld_start_timers(next, ipst);
6887 6888 }
6888 6889
6889 6890 /*
6890 6891 * Start the current exclusive operation on `ipsq'; associate it with `ipif'
6891 6892 * and `ioccmd'.
6892 6893 */
6893 6894 void
6894 6895 ipsq_current_start(ipsq_t *ipsq, ipif_t *ipif, int ioccmd)
6895 6896 {
6896 6897 ill_t *ill = ipif->ipif_ill;
6897 6898 ipxop_t *ipx = ipsq->ipsq_xop;
6898 6899
6899 6900 ASSERT(IAM_WRITER_IPSQ(ipsq));
6900 6901 ASSERT(ipx->ipx_current_ipif == NULL);
6901 6902 ASSERT(ipx->ipx_current_ioctl == 0);
6902 6903
6903 6904 ipx->ipx_current_done = B_FALSE;
6904 6905 ipx->ipx_current_ioctl = ioccmd;
6905 6906 mutex_enter(&ipx->ipx_lock);
6906 6907 ipx->ipx_current_ipif = ipif;
6907 6908 mutex_exit(&ipx->ipx_lock);
6908 6909
6909 6910 /*
6910 6911 * Set IPIF_CHANGING on one or more ipifs associated with the
6911 6912 * current exclusive operation. IPIF_CHANGING prevents any new
6912 6913 * references to the ipif (so that the references will eventually
6913 6914 * drop to zero) and also prevents any "get" operations (e.g.,
6914 6915 * SIOCGLIFFLAGS) from being able to access the ipif until the
6915 6916 * operation has completed and the ipif is again in a stable state.
6916 6917 *
6917 6918 * For ioctls, IPIF_CHANGING is set on the ipif associated with the
6918 6919 * ioctl. For internal operations (where ioccmd is zero), all ipifs
6919 6920 * on the ill are marked with IPIF_CHANGING since it's unclear which
6920 6921 * ipifs will be affected.
6921 6922 *
6922 6923 * Note that SIOCLIFREMOVEIF is a special case as it sets
6923 6924 * IPIF_CONDEMNED internally after identifying the right ipif to
6924 6925 * operate on.
6925 6926 */
6926 6927 switch (ioccmd) {
6927 6928 case SIOCLIFREMOVEIF:
6928 6929 break;
6929 6930 case 0:
6930 6931 mutex_enter(&ill->ill_lock);
6931 6932 ipif = ipif->ipif_ill->ill_ipif;
6932 6933 for (; ipif != NULL; ipif = ipif->ipif_next)
6933 6934 ipif->ipif_state_flags |= IPIF_CHANGING;
6934 6935 mutex_exit(&ill->ill_lock);
6935 6936 break;
6936 6937 default:
6937 6938 mutex_enter(&ill->ill_lock);
6938 6939 ipif->ipif_state_flags |= IPIF_CHANGING;
6939 6940 mutex_exit(&ill->ill_lock);
6940 6941 }
6941 6942 }
6942 6943
6943 6944 /*
6944 6945 * Finish the current exclusive operation on `ipsq'. Usually, this will allow
6945 6946 * the next exclusive operation to begin once we ipsq_exit(). However, if
6946 6947 * pending DLPI operations remain, then we will wait for the queue to drain
6947 6948 * before allowing the next exclusive operation to begin. This ensures that
6948 6949 * DLPI operations from one exclusive operation are never improperly processed
6949 6950 * as part of a subsequent exclusive operation.
6950 6951 */
6951 6952 void
6952 6953 ipsq_current_finish(ipsq_t *ipsq)
6953 6954 {
6954 6955 ipxop_t *ipx = ipsq->ipsq_xop;
6955 6956 t_uscalar_t dlpi_pending = DL_PRIM_INVAL;
6956 6957 ipif_t *ipif = ipx->ipx_current_ipif;
6957 6958
6958 6959 ASSERT(IAM_WRITER_IPSQ(ipsq));
6959 6960
6960 6961 /*
6961 6962 * For SIOCLIFREMOVEIF, the ipif has been already been blown away
6962 6963 * (but in that case, IPIF_CHANGING will already be clear and no
6963 6964 * pending DLPI messages can remain).
6964 6965 */
6965 6966 if (ipx->ipx_current_ioctl != SIOCLIFREMOVEIF) {
6966 6967 ill_t *ill = ipif->ipif_ill;
6967 6968
6968 6969 mutex_enter(&ill->ill_lock);
6969 6970 dlpi_pending = ill->ill_dlpi_pending;
6970 6971 if (ipx->ipx_current_ioctl == 0) {
6971 6972 ipif = ill->ill_ipif;
6972 6973 for (; ipif != NULL; ipif = ipif->ipif_next)
6973 6974 ipif->ipif_state_flags &= ~IPIF_CHANGING;
6974 6975 } else {
6975 6976 ipif->ipif_state_flags &= ~IPIF_CHANGING;
6976 6977 }
6977 6978 mutex_exit(&ill->ill_lock);
6978 6979 }
6979 6980
6980 6981 ASSERT(!ipx->ipx_current_done);
6981 6982 ipx->ipx_current_done = B_TRUE;
6982 6983 ipx->ipx_current_ioctl = 0;
6983 6984 if (dlpi_pending == DL_PRIM_INVAL) {
6984 6985 mutex_enter(&ipx->ipx_lock);
6985 6986 ipx->ipx_current_ipif = NULL;
6986 6987 mutex_exit(&ipx->ipx_lock);
6987 6988 }
6988 6989 }
6989 6990
6990 6991 /*
6991 6992 * The ill is closing. Flush all messages on the ipsq that originated
6992 6993 * from this ill. Usually there wont' be any messages on the ipsq_xopq_mphead
6993 6994 * for this ill since ipsq_enter could not have entered until then.
6994 6995 * New messages can't be queued since the CONDEMNED flag is set.
6995 6996 */
6996 6997 static void
6997 6998 ipsq_flush(ill_t *ill)
6998 6999 {
6999 7000 queue_t *q;
7000 7001 mblk_t *prev;
7001 7002 mblk_t *mp;
7002 7003 mblk_t *mp_next;
7003 7004 ipxop_t *ipx = ill->ill_phyint->phyint_ipsq->ipsq_xop;
7004 7005
7005 7006 ASSERT(IAM_WRITER_ILL(ill));
7006 7007
7007 7008 /*
7008 7009 * Flush any messages sent up by the driver.
7009 7010 */
7010 7011 mutex_enter(&ipx->ipx_lock);
7011 7012 for (prev = NULL, mp = ipx->ipx_mphead; mp != NULL; mp = mp_next) {
7012 7013 mp_next = mp->b_next;
7013 7014 q = mp->b_queue;
7014 7015 if (q == ill->ill_rq || q == ill->ill_wq) {
7015 7016 /* dequeue mp */
7016 7017 if (prev == NULL)
7017 7018 ipx->ipx_mphead = mp->b_next;
7018 7019 else
7019 7020 prev->b_next = mp->b_next;
7020 7021 if (ipx->ipx_mptail == mp) {
7021 7022 ASSERT(mp_next == NULL);
7022 7023 ipx->ipx_mptail = prev;
7023 7024 }
7024 7025 inet_freemsg(mp);
7025 7026 } else {
7026 7027 prev = mp;
7027 7028 }
7028 7029 }
7029 7030 mutex_exit(&ipx->ipx_lock);
7030 7031 (void) ipsq_pending_mp_cleanup(ill, NULL);
7031 7032 ipsq_xopq_mp_cleanup(ill, NULL);
7032 7033 }
7033 7034
7034 7035 /*
7035 7036 * Parse an ifreq or lifreq struct coming down ioctls and refhold
7036 7037 * and return the associated ipif.
7037 7038 * Return value:
7038 7039 * Non zero: An error has occurred. ci may not be filled out.
7039 7040 * zero : ci is filled out with the ioctl cmd in ci.ci_name, and
7040 7041 * a held ipif in ci.ci_ipif.
7041 7042 */
7042 7043 int
7043 7044 ip_extract_lifreq(queue_t *q, mblk_t *mp, const ip_ioctl_cmd_t *ipip,
7044 7045 cmd_info_t *ci)
7045 7046 {
7046 7047 char *name;
7047 7048 struct ifreq *ifr;
7048 7049 struct lifreq *lifr;
7049 7050 ipif_t *ipif = NULL;
7050 7051 ill_t *ill;
7051 7052 conn_t *connp;
7052 7053 boolean_t isv6;
7053 7054 int err;
7054 7055 mblk_t *mp1;
7055 7056 zoneid_t zoneid;
7056 7057 ip_stack_t *ipst;
7057 7058
7058 7059 if (q->q_next != NULL) {
7059 7060 ill = (ill_t *)q->q_ptr;
7060 7061 isv6 = ill->ill_isv6;
7061 7062 connp = NULL;
7062 7063 zoneid = ALL_ZONES;
7063 7064 ipst = ill->ill_ipst;
7064 7065 } else {
7065 7066 ill = NULL;
7066 7067 connp = Q_TO_CONN(q);
7067 7068 isv6 = (connp->conn_family == AF_INET6);
7068 7069 zoneid = connp->conn_zoneid;
7069 7070 if (zoneid == GLOBAL_ZONEID) {
7070 7071 /* global zone can access ipifs in all zones */
7071 7072 zoneid = ALL_ZONES;
7072 7073 }
7073 7074 ipst = connp->conn_netstack->netstack_ip;
7074 7075 }
7075 7076
7076 7077 /* Has been checked in ip_wput_nondata */
7077 7078 mp1 = mp->b_cont->b_cont;
7078 7079
7079 7080 if (ipip->ipi_cmd_type == IF_CMD) {
7080 7081 /* This a old style SIOC[GS]IF* command */
7081 7082 ifr = (struct ifreq *)mp1->b_rptr;
7082 7083 /*
7083 7084 * Null terminate the string to protect against buffer
7084 7085 * overrun. String was generated by user code and may not
7085 7086 * be trusted.
7086 7087 */
7087 7088 ifr->ifr_name[IFNAMSIZ - 1] = '\0';
7088 7089 name = ifr->ifr_name;
7089 7090 ci->ci_sin = (sin_t *)&ifr->ifr_addr;
7090 7091 ci->ci_sin6 = NULL;
7091 7092 ci->ci_lifr = (struct lifreq *)ifr;
7092 7093 } else {
7093 7094 /* This a new style SIOC[GS]LIF* command */
7094 7095 ASSERT(ipip->ipi_cmd_type == LIF_CMD);
7095 7096 lifr = (struct lifreq *)mp1->b_rptr;
7096 7097 /*
7097 7098 * Null terminate the string to protect against buffer
7098 7099 * overrun. String was generated by user code and may not
7099 7100 * be trusted.
7100 7101 */
7101 7102 lifr->lifr_name[LIFNAMSIZ - 1] = '\0';
7102 7103 name = lifr->lifr_name;
7103 7104 ci->ci_sin = (sin_t *)&lifr->lifr_addr;
7104 7105 ci->ci_sin6 = (sin6_t *)&lifr->lifr_addr;
7105 7106 ci->ci_lifr = lifr;
7106 7107 }
7107 7108
7108 7109 if (ipip->ipi_cmd == SIOCSLIFNAME) {
7109 7110 /*
7110 7111 * The ioctl will be failed if the ioctl comes down
7111 7112 * an conn stream
7112 7113 */
7113 7114 if (ill == NULL) {
7114 7115 /*
7115 7116 * Not an ill queue, return EINVAL same as the
7116 7117 * old error code.
7117 7118 */
7118 7119 return (ENXIO);
7119 7120 }
7120 7121 ipif = ill->ill_ipif;
7121 7122 ipif_refhold(ipif);
7122 7123 } else {
7123 7124 /*
7124 7125 * Ensure that ioctls don't see any internal state changes
7125 7126 * caused by set ioctls by deferring them if IPIF_CHANGING is
7126 7127 * set.
7127 7128 */
7128 7129 ipif = ipif_lookup_on_name_async(name, mi_strlen(name),
7129 7130 isv6, zoneid, q, mp, ip_process_ioctl, &err, ipst);
7130 7131 if (ipif == NULL) {
7131 7132 if (err == EINPROGRESS)
7132 7133 return (err);
7133 7134 err = 0; /* Ensure we don't use it below */
7134 7135 }
7135 7136 }
7136 7137
7137 7138 /*
7138 7139 * Old style [GS]IFCMD does not admit IPv6 ipif
7139 7140 */
7140 7141 if (ipif != NULL && ipif->ipif_isv6 && ipip->ipi_cmd_type == IF_CMD) {
7141 7142 ipif_refrele(ipif);
7142 7143 return (ENXIO);
7143 7144 }
7144 7145
7145 7146 if (ipif == NULL && ill != NULL && ill->ill_ipif != NULL &&
7146 7147 name[0] == '\0') {
7147 7148 /*
7148 7149 * Handle a or a SIOC?IF* with a null name
7149 7150 * during plumb (on the ill queue before the I_PLINK).
7150 7151 */
7151 7152 ipif = ill->ill_ipif;
7152 7153 ipif_refhold(ipif);
7153 7154 }
7154 7155
7155 7156 if (ipif == NULL)
7156 7157 return (ENXIO);
7157 7158
7158 7159 DTRACE_PROBE4(ipif__ioctl, char *, "ip_extract_lifreq",
7159 7160 int, ipip->ipi_cmd, ill_t *, ipif->ipif_ill, ipif_t *, ipif);
7160 7161
7161 7162 ci->ci_ipif = ipif;
7162 7163 return (0);
7163 7164 }
7164 7165
7165 7166 /*
7166 7167 * Return the total number of ipifs.
7167 7168 */
7168 7169 static uint_t
7169 7170 ip_get_numifs(zoneid_t zoneid, ip_stack_t *ipst)
7170 7171 {
7171 7172 uint_t numifs = 0;
7172 7173 ill_t *ill;
7173 7174 ill_walk_context_t ctx;
7174 7175 ipif_t *ipif;
7175 7176
7176 7177 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
7177 7178 ill = ILL_START_WALK_V4(&ctx, ipst);
7178 7179 for (; ill != NULL; ill = ill_next(&ctx, ill)) {
7179 7180 if (IS_UNDER_IPMP(ill))
7180 7181 continue;
7181 7182 for (ipif = ill->ill_ipif; ipif != NULL;
7182 7183 ipif = ipif->ipif_next) {
7183 7184 if (ipif->ipif_zoneid == zoneid ||
7184 7185 ipif->ipif_zoneid == ALL_ZONES)
7185 7186 numifs++;
7186 7187 }
7187 7188 }
7188 7189 rw_exit(&ipst->ips_ill_g_lock);
7189 7190 return (numifs);
7190 7191 }
7191 7192
7192 7193 /*
7193 7194 * Return the total number of ipifs.
7194 7195 */
7195 7196 static uint_t
7196 7197 ip_get_numlifs(int family, int lifn_flags, zoneid_t zoneid, ip_stack_t *ipst)
7197 7198 {
7198 7199 uint_t numifs = 0;
7199 7200 ill_t *ill;
7200 7201 ipif_t *ipif;
7201 7202 ill_walk_context_t ctx;
7202 7203
7203 7204 ip1dbg(("ip_get_numlifs(%d %u %d)\n", family, lifn_flags, (int)zoneid));
7204 7205
7205 7206 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
7206 7207 if (family == AF_INET)
7207 7208 ill = ILL_START_WALK_V4(&ctx, ipst);
7208 7209 else if (family == AF_INET6)
7209 7210 ill = ILL_START_WALK_V6(&ctx, ipst);
7210 7211 else
7211 7212 ill = ILL_START_WALK_ALL(&ctx, ipst);
7212 7213
7213 7214 for (; ill != NULL; ill = ill_next(&ctx, ill)) {
7214 7215 if (IS_UNDER_IPMP(ill) && !(lifn_flags & LIFC_UNDER_IPMP))
7215 7216 continue;
7216 7217
7217 7218 for (ipif = ill->ill_ipif; ipif != NULL;
7218 7219 ipif = ipif->ipif_next) {
7219 7220 if ((ipif->ipif_flags & IPIF_NOXMIT) &&
7220 7221 !(lifn_flags & LIFC_NOXMIT))
7221 7222 continue;
7222 7223 if ((ipif->ipif_flags & IPIF_TEMPORARY) &&
7223 7224 !(lifn_flags & LIFC_TEMPORARY))
7224 7225 continue;
7225 7226 if (((ipif->ipif_flags &
7226 7227 (IPIF_NOXMIT|IPIF_NOLOCAL|
7227 7228 IPIF_DEPRECATED)) ||
7228 7229 IS_LOOPBACK(ill) ||
7229 7230 !(ipif->ipif_flags & IPIF_UP)) &&
7230 7231 (lifn_flags & LIFC_EXTERNAL_SOURCE))
7231 7232 continue;
7232 7233
7233 7234 if (zoneid != ipif->ipif_zoneid &&
7234 7235 ipif->ipif_zoneid != ALL_ZONES &&
7235 7236 (zoneid != GLOBAL_ZONEID ||
7236 7237 !(lifn_flags & LIFC_ALLZONES)))
7237 7238 continue;
7238 7239
7239 7240 numifs++;
7240 7241 }
7241 7242 }
7242 7243 rw_exit(&ipst->ips_ill_g_lock);
7243 7244 return (numifs);
7244 7245 }
7245 7246
7246 7247 uint_t
7247 7248 ip_get_lifsrcofnum(ill_t *ill)
7248 7249 {
7249 7250 uint_t numifs = 0;
7250 7251 ill_t *ill_head = ill;
7251 7252 ip_stack_t *ipst = ill->ill_ipst;
7252 7253
7253 7254 /*
7254 7255 * ill_g_usesrc_lock protects ill_usesrc_grp_next, for example, some
7255 7256 * other thread may be trying to relink the ILLs in this usesrc group
7256 7257 * and adjusting the ill_usesrc_grp_next pointers
7257 7258 */
7258 7259 rw_enter(&ipst->ips_ill_g_usesrc_lock, RW_READER);
7259 7260 if ((ill->ill_usesrc_ifindex == 0) &&
7260 7261 (ill->ill_usesrc_grp_next != NULL)) {
7261 7262 for (; (ill != NULL) && (ill->ill_usesrc_grp_next != ill_head);
7262 7263 ill = ill->ill_usesrc_grp_next)
7263 7264 numifs++;
7264 7265 }
7265 7266 rw_exit(&ipst->ips_ill_g_usesrc_lock);
7266 7267
7267 7268 return (numifs);
7268 7269 }
7269 7270
7270 7271 /* Null values are passed in for ipif, sin, and ifreq */
7271 7272 /* ARGSUSED */
7272 7273 int
7273 7274 ip_sioctl_get_ifnum(ipif_t *dummy_ipif, sin_t *dummy_sin, queue_t *q,
7274 7275 mblk_t *mp, ip_ioctl_cmd_t *ipip, void *ifreq)
7275 7276 {
7276 7277 int *nump;
7277 7278 conn_t *connp = Q_TO_CONN(q);
7278 7279
7279 7280 ASSERT(q->q_next == NULL); /* not a valid ioctl for ip as a module */
7280 7281
7281 7282 /* Existence of b_cont->b_cont checked in ip_wput_nondata */
7282 7283 nump = (int *)mp->b_cont->b_cont->b_rptr;
7283 7284
7284 7285 *nump = ip_get_numifs(connp->conn_zoneid,
7285 7286 connp->conn_netstack->netstack_ip);
7286 7287 ip1dbg(("ip_sioctl_get_ifnum numifs %d", *nump));
7287 7288 return (0);
7288 7289 }
7289 7290
7290 7291 /* Null values are passed in for ipif, sin, and ifreq */
7291 7292 /* ARGSUSED */
7292 7293 int
7293 7294 ip_sioctl_get_lifnum(ipif_t *dummy_ipif, sin_t *dummy_sin,
7294 7295 queue_t *q, mblk_t *mp, ip_ioctl_cmd_t *ipip, void *ifreq)
7295 7296 {
7296 7297 struct lifnum *lifn;
7297 7298 mblk_t *mp1;
7298 7299 conn_t *connp = Q_TO_CONN(q);
7299 7300
7300 7301 ASSERT(q->q_next == NULL); /* not a valid ioctl for ip as a module */
7301 7302
7302 7303 /* Existence checked in ip_wput_nondata */
7303 7304 mp1 = mp->b_cont->b_cont;
7304 7305
7305 7306 lifn = (struct lifnum *)mp1->b_rptr;
7306 7307 switch (lifn->lifn_family) {
7307 7308 case AF_UNSPEC:
7308 7309 case AF_INET:
7309 7310 case AF_INET6:
7310 7311 break;
7311 7312 default:
7312 7313 return (EAFNOSUPPORT);
7313 7314 }
7314 7315
7315 7316 lifn->lifn_count = ip_get_numlifs(lifn->lifn_family, lifn->lifn_flags,
7316 7317 connp->conn_zoneid, connp->conn_netstack->netstack_ip);
7317 7318 ip1dbg(("ip_sioctl_get_lifnum numifs %d", lifn->lifn_count));
7318 7319 return (0);
7319 7320 }
7320 7321
7321 7322 /* ARGSUSED */
7322 7323 int
7323 7324 ip_sioctl_get_ifconf(ipif_t *dummy_ipif, sin_t *dummy_sin, queue_t *q,
7324 7325 mblk_t *mp, ip_ioctl_cmd_t *ipip, void *ifreq)
7325 7326 {
7326 7327 STRUCT_HANDLE(ifconf, ifc);
7327 7328 mblk_t *mp1;
7328 7329 struct iocblk *iocp;
7329 7330 struct ifreq *ifr;
7330 7331 ill_walk_context_t ctx;
7331 7332 ill_t *ill;
7332 7333 ipif_t *ipif;
7333 7334 struct sockaddr_in *sin;
7334 7335 int32_t ifclen;
7335 7336 zoneid_t zoneid;
7336 7337 ip_stack_t *ipst = CONNQ_TO_IPST(q);
7337 7338
7338 7339 ASSERT(q->q_next == NULL); /* not valid ioctls for ip as a module */
7339 7340
7340 7341 ip1dbg(("ip_sioctl_get_ifconf"));
7341 7342 /* Existence verified in ip_wput_nondata */
7342 7343 mp1 = mp->b_cont->b_cont;
7343 7344 iocp = (struct iocblk *)mp->b_rptr;
7344 7345 zoneid = Q_TO_CONN(q)->conn_zoneid;
7345 7346
7346 7347 /*
7347 7348 * The original SIOCGIFCONF passed in a struct ifconf which specified
7348 7349 * the user buffer address and length into which the list of struct
7349 7350 * ifreqs was to be copied. Since AT&T Streams does not seem to
7350 7351 * allow M_COPYOUT to be used in conjunction with I_STR IOCTLS,
7351 7352 * the SIOCGIFCONF operation was redefined to simply provide
7352 7353 * a large output buffer into which we are supposed to jam the ifreq
7353 7354 * array. The same ioctl command code was used, despite the fact that
7354 7355 * both the applications and the kernel code had to change, thus making
7355 7356 * it impossible to support both interfaces.
7356 7357 *
7357 7358 * For reasons not good enough to try to explain, the following
7358 7359 * algorithm is used for deciding what to do with one of these:
7359 7360 * If the IOCTL comes in as an I_STR, it is assumed to be of the new
7360 7361 * form with the output buffer coming down as the continuation message.
7361 7362 * If it arrives as a TRANSPARENT IOCTL, it is assumed to be old style,
7362 7363 * and we have to copy in the ifconf structure to find out how big the
7363 7364 * output buffer is and where to copy out to. Sure no problem...
7364 7365 *
7365 7366 */
7366 7367 STRUCT_SET_HANDLE(ifc, iocp->ioc_flag, NULL);
7367 7368 if ((mp1->b_wptr - mp1->b_rptr) == STRUCT_SIZE(ifc)) {
7368 7369 int numifs = 0;
7369 7370 size_t ifc_bufsize;
7370 7371
7371 7372 /*
7372 7373 * Must be (better be!) continuation of a TRANSPARENT
7373 7374 * IOCTL. We just copied in the ifconf structure.
7374 7375 */
7375 7376 STRUCT_SET_HANDLE(ifc, iocp->ioc_flag,
7376 7377 (struct ifconf *)mp1->b_rptr);
7377 7378
7378 7379 /*
7379 7380 * Allocate a buffer to hold requested information.
7380 7381 *
7381 7382 * If ifc_len is larger than what is needed, we only
7382 7383 * allocate what we will use.
7383 7384 *
7384 7385 * If ifc_len is smaller than what is needed, return
7385 7386 * EINVAL.
7386 7387 *
7387 7388 * XXX: the ill_t structure can hava 2 counters, for
7388 7389 * v4 and v6 (not just ill_ipif_up_count) to store the
7389 7390 * number of interfaces for a device, so we don't need
7390 7391 * to count them here...
7391 7392 */
7392 7393 numifs = ip_get_numifs(zoneid, ipst);
7393 7394
7394 7395 ifclen = STRUCT_FGET(ifc, ifc_len);
7395 7396 ifc_bufsize = numifs * sizeof (struct ifreq);
7396 7397 if (ifc_bufsize > ifclen) {
7397 7398 if (iocp->ioc_cmd == O_SIOCGIFCONF) {
7398 7399 /* old behaviour */
7399 7400 return (EINVAL);
7400 7401 } else {
7401 7402 ifc_bufsize = ifclen;
7402 7403 }
7403 7404 }
7404 7405
7405 7406 mp1 = mi_copyout_alloc(q, mp,
7406 7407 STRUCT_FGETP(ifc, ifc_buf), ifc_bufsize, B_FALSE);
7407 7408 if (mp1 == NULL)
7408 7409 return (ENOMEM);
7409 7410
7410 7411 mp1->b_wptr = mp1->b_rptr + ifc_bufsize;
7411 7412 }
7412 7413 bzero(mp1->b_rptr, mp1->b_wptr - mp1->b_rptr);
7413 7414 /*
7414 7415 * the SIOCGIFCONF ioctl only knows about
7415 7416 * IPv4 addresses, so don't try to tell
7416 7417 * it about interfaces with IPv6-only
7417 7418 * addresses. (Last parm 'isv6' is B_FALSE)
7418 7419 */
7419 7420
7420 7421 ifr = (struct ifreq *)mp1->b_rptr;
7421 7422
7422 7423 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
7423 7424 ill = ILL_START_WALK_V4(&ctx, ipst);
7424 7425 for (; ill != NULL; ill = ill_next(&ctx, ill)) {
7425 7426 if (IS_UNDER_IPMP(ill))
7426 7427 continue;
7427 7428 for (ipif = ill->ill_ipif; ipif != NULL;
7428 7429 ipif = ipif->ipif_next) {
7429 7430 if (zoneid != ipif->ipif_zoneid &&
7430 7431 ipif->ipif_zoneid != ALL_ZONES)
7431 7432 continue;
7432 7433 if ((uchar_t *)&ifr[1] > mp1->b_wptr) {
7433 7434 if (iocp->ioc_cmd == O_SIOCGIFCONF) {
7434 7435 /* old behaviour */
7435 7436 rw_exit(&ipst->ips_ill_g_lock);
7436 7437 return (EINVAL);
7437 7438 } else {
7438 7439 goto if_copydone;
7439 7440 }
7440 7441 }
7441 7442 ipif_get_name(ipif, ifr->ifr_name,
7442 7443 sizeof (ifr->ifr_name));
7443 7444 sin = (sin_t *)&ifr->ifr_addr;
7444 7445 *sin = sin_null;
7445 7446 sin->sin_family = AF_INET;
7446 7447 sin->sin_addr.s_addr = ipif->ipif_lcl_addr;
7447 7448 ifr++;
7448 7449 }
7449 7450 }
7450 7451 if_copydone:
7451 7452 rw_exit(&ipst->ips_ill_g_lock);
7452 7453 mp1->b_wptr = (uchar_t *)ifr;
7453 7454
7454 7455 if (STRUCT_BUF(ifc) != NULL) {
7455 7456 STRUCT_FSET(ifc, ifc_len,
7456 7457 (int)((uchar_t *)ifr - mp1->b_rptr));
7457 7458 }
7458 7459 return (0);
7459 7460 }
7460 7461
7461 7462 /*
7462 7463 * Get the interfaces using the address hosted on the interface passed in,
7463 7464 * as a source adddress
7464 7465 */
7465 7466 /* ARGSUSED */
7466 7467 int
7467 7468 ip_sioctl_get_lifsrcof(ipif_t *dummy_ipif, sin_t *dummy_sin, queue_t *q,
7468 7469 mblk_t *mp, ip_ioctl_cmd_t *ipip, void *ifreq)
7469 7470 {
7470 7471 mblk_t *mp1;
7471 7472 ill_t *ill, *ill_head;
7472 7473 ipif_t *ipif, *orig_ipif;
7473 7474 int numlifs = 0;
7474 7475 size_t lifs_bufsize, lifsmaxlen;
7475 7476 struct lifreq *lifr;
7476 7477 struct iocblk *iocp = (struct iocblk *)mp->b_rptr;
7477 7478 uint_t ifindex;
7478 7479 zoneid_t zoneid;
7479 7480 boolean_t isv6 = B_FALSE;
7480 7481 struct sockaddr_in *sin;
7481 7482 struct sockaddr_in6 *sin6;
7482 7483 STRUCT_HANDLE(lifsrcof, lifs);
7483 7484 ip_stack_t *ipst;
7484 7485
7485 7486 ipst = CONNQ_TO_IPST(q);
7486 7487
7487 7488 ASSERT(q->q_next == NULL);
7488 7489
7489 7490 zoneid = Q_TO_CONN(q)->conn_zoneid;
7490 7491
7491 7492 /* Existence verified in ip_wput_nondata */
7492 7493 mp1 = mp->b_cont->b_cont;
7493 7494
7494 7495 /*
7495 7496 * Must be (better be!) continuation of a TRANSPARENT
7496 7497 * IOCTL. We just copied in the lifsrcof structure.
7497 7498 */
7498 7499 STRUCT_SET_HANDLE(lifs, iocp->ioc_flag,
7499 7500 (struct lifsrcof *)mp1->b_rptr);
7500 7501
7501 7502 if (MBLKL(mp1) != STRUCT_SIZE(lifs))
7502 7503 return (EINVAL);
7503 7504
7504 7505 ifindex = STRUCT_FGET(lifs, lifs_ifindex);
7505 7506 isv6 = (Q_TO_CONN(q))->conn_family == AF_INET6;
7506 7507 ipif = ipif_lookup_on_ifindex(ifindex, isv6, zoneid, ipst);
7507 7508 if (ipif == NULL) {
7508 7509 ip1dbg(("ip_sioctl_get_lifsrcof: no ipif for ifindex %d\n",
7509 7510 ifindex));
7510 7511 return (ENXIO);
7511 7512 }
7512 7513
7513 7514 /* Allocate a buffer to hold requested information */
7514 7515 numlifs = ip_get_lifsrcofnum(ipif->ipif_ill);
7515 7516 lifs_bufsize = numlifs * sizeof (struct lifreq);
7516 7517 lifsmaxlen = STRUCT_FGET(lifs, lifs_maxlen);
7517 7518 /* The actual size needed is always returned in lifs_len */
7518 7519 STRUCT_FSET(lifs, lifs_len, lifs_bufsize);
7519 7520
7520 7521 /* If the amount we need is more than what is passed in, abort */
7521 7522 if (lifs_bufsize > lifsmaxlen || lifs_bufsize == 0) {
7522 7523 ipif_refrele(ipif);
7523 7524 return (0);
7524 7525 }
7525 7526
7526 7527 mp1 = mi_copyout_alloc(q, mp,
7527 7528 STRUCT_FGETP(lifs, lifs_buf), lifs_bufsize, B_FALSE);
7528 7529 if (mp1 == NULL) {
7529 7530 ipif_refrele(ipif);
7530 7531 return (ENOMEM);
7531 7532 }
7532 7533
7533 7534 mp1->b_wptr = mp1->b_rptr + lifs_bufsize;
7534 7535 bzero(mp1->b_rptr, lifs_bufsize);
7535 7536
7536 7537 lifr = (struct lifreq *)mp1->b_rptr;
7537 7538
7538 7539 ill = ill_head = ipif->ipif_ill;
7539 7540 orig_ipif = ipif;
7540 7541
7541 7542 /* ill_g_usesrc_lock protects ill_usesrc_grp_next */
7542 7543 rw_enter(&ipst->ips_ill_g_usesrc_lock, RW_READER);
7543 7544 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
7544 7545
7545 7546 ill = ill->ill_usesrc_grp_next; /* start from next ill */
7546 7547 for (; (ill != NULL) && (ill != ill_head);
7547 7548 ill = ill->ill_usesrc_grp_next) {
7548 7549
7549 7550 if ((uchar_t *)&lifr[1] > mp1->b_wptr)
7550 7551 break;
7551 7552
7552 7553 ipif = ill->ill_ipif;
7553 7554 ipif_get_name(ipif, lifr->lifr_name, sizeof (lifr->lifr_name));
7554 7555 if (ipif->ipif_isv6) {
7555 7556 sin6 = (sin6_t *)&lifr->lifr_addr;
7556 7557 *sin6 = sin6_null;
7557 7558 sin6->sin6_family = AF_INET6;
7558 7559 sin6->sin6_addr = ipif->ipif_v6lcl_addr;
7559 7560 lifr->lifr_addrlen = ip_mask_to_plen_v6(
7560 7561 &ipif->ipif_v6net_mask);
7561 7562 } else {
7562 7563 sin = (sin_t *)&lifr->lifr_addr;
7563 7564 *sin = sin_null;
7564 7565 sin->sin_family = AF_INET;
7565 7566 sin->sin_addr.s_addr = ipif->ipif_lcl_addr;
7566 7567 lifr->lifr_addrlen = ip_mask_to_plen(
7567 7568 ipif->ipif_net_mask);
7568 7569 }
7569 7570 lifr++;
7570 7571 }
7571 7572 rw_exit(&ipst->ips_ill_g_lock);
7572 7573 rw_exit(&ipst->ips_ill_g_usesrc_lock);
7573 7574 ipif_refrele(orig_ipif);
7574 7575 mp1->b_wptr = (uchar_t *)lifr;
7575 7576 STRUCT_FSET(lifs, lifs_len, (int)((uchar_t *)lifr - mp1->b_rptr));
7576 7577
7577 7578 return (0);
7578 7579 }
7579 7580
7580 7581 /* ARGSUSED */
7581 7582 int
7582 7583 ip_sioctl_get_lifconf(ipif_t *dummy_ipif, sin_t *dummy_sin, queue_t *q,
7583 7584 mblk_t *mp, ip_ioctl_cmd_t *ipip, void *ifreq)
7584 7585 {
7585 7586 mblk_t *mp1;
7586 7587 int list;
7587 7588 ill_t *ill;
7588 7589 ipif_t *ipif;
7589 7590 int flags;
7590 7591 int numlifs = 0;
7591 7592 size_t lifc_bufsize;
7592 7593 struct lifreq *lifr;
7593 7594 sa_family_t family;
7594 7595 struct sockaddr_in *sin;
7595 7596 struct sockaddr_in6 *sin6;
7596 7597 ill_walk_context_t ctx;
7597 7598 struct iocblk *iocp = (struct iocblk *)mp->b_rptr;
7598 7599 int32_t lifclen;
7599 7600 zoneid_t zoneid;
7600 7601 STRUCT_HANDLE(lifconf, lifc);
7601 7602 ip_stack_t *ipst = CONNQ_TO_IPST(q);
7602 7603
7603 7604 ip1dbg(("ip_sioctl_get_lifconf"));
7604 7605
7605 7606 ASSERT(q->q_next == NULL);
7606 7607
7607 7608 zoneid = Q_TO_CONN(q)->conn_zoneid;
7608 7609
7609 7610 /* Existence verified in ip_wput_nondata */
7610 7611 mp1 = mp->b_cont->b_cont;
7611 7612
7612 7613 /*
7613 7614 * An extended version of SIOCGIFCONF that takes an
7614 7615 * additional address family and flags field.
7615 7616 * AF_UNSPEC retrieve both IPv4 and IPv6.
7616 7617 * Unless LIFC_NOXMIT is specified the IPIF_NOXMIT
7617 7618 * interfaces are omitted.
7618 7619 * Similarly, IPIF_TEMPORARY interfaces are omitted
7619 7620 * unless LIFC_TEMPORARY is specified.
7620 7621 * If LIFC_EXTERNAL_SOURCE is specified, IPIF_NOXMIT,
7621 7622 * IPIF_NOLOCAL, PHYI_LOOPBACK, IPIF_DEPRECATED and
7622 7623 * not IPIF_UP interfaces are omitted. LIFC_EXTERNAL_SOURCE
7623 7624 * has priority over LIFC_NOXMIT.
7624 7625 */
7625 7626 STRUCT_SET_HANDLE(lifc, iocp->ioc_flag, NULL);
7626 7627
7627 7628 if ((mp1->b_wptr - mp1->b_rptr) != STRUCT_SIZE(lifc))
7628 7629 return (EINVAL);
7629 7630
7630 7631 /*
7631 7632 * Must be (better be!) continuation of a TRANSPARENT
7632 7633 * IOCTL. We just copied in the lifconf structure.
7633 7634 */
7634 7635 STRUCT_SET_HANDLE(lifc, iocp->ioc_flag, (struct lifconf *)mp1->b_rptr);
7635 7636
7636 7637 family = STRUCT_FGET(lifc, lifc_family);
7637 7638 flags = STRUCT_FGET(lifc, lifc_flags);
7638 7639
7639 7640 switch (family) {
7640 7641 case AF_UNSPEC:
7641 7642 /*
7642 7643 * walk all ILL's.
7643 7644 */
7644 7645 list = MAX_G_HEADS;
7645 7646 break;
7646 7647 case AF_INET:
7647 7648 /*
7648 7649 * walk only IPV4 ILL's.
7649 7650 */
7650 7651 list = IP_V4_G_HEAD;
7651 7652 break;
7652 7653 case AF_INET6:
7653 7654 /*
7654 7655 * walk only IPV6 ILL's.
7655 7656 */
7656 7657 list = IP_V6_G_HEAD;
7657 7658 break;
7658 7659 default:
7659 7660 return (EAFNOSUPPORT);
7660 7661 }
7661 7662
7662 7663 /*
7663 7664 * Allocate a buffer to hold requested information.
7664 7665 *
7665 7666 * If lifc_len is larger than what is needed, we only
7666 7667 * allocate what we will use.
7667 7668 *
7668 7669 * If lifc_len is smaller than what is needed, return
7669 7670 * EINVAL.
7670 7671 */
7671 7672 numlifs = ip_get_numlifs(family, flags, zoneid, ipst);
7672 7673 lifc_bufsize = numlifs * sizeof (struct lifreq);
7673 7674 lifclen = STRUCT_FGET(lifc, lifc_len);
7674 7675 if (lifc_bufsize > lifclen) {
7675 7676 if (iocp->ioc_cmd == O_SIOCGLIFCONF)
7676 7677 return (EINVAL);
7677 7678 else
7678 7679 lifc_bufsize = lifclen;
7679 7680 }
7680 7681
7681 7682 mp1 = mi_copyout_alloc(q, mp,
7682 7683 STRUCT_FGETP(lifc, lifc_buf), lifc_bufsize, B_FALSE);
7683 7684 if (mp1 == NULL)
7684 7685 return (ENOMEM);
7685 7686
7686 7687 mp1->b_wptr = mp1->b_rptr + lifc_bufsize;
7687 7688 bzero(mp1->b_rptr, mp1->b_wptr - mp1->b_rptr);
7688 7689
7689 7690 lifr = (struct lifreq *)mp1->b_rptr;
7690 7691
7691 7692 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
7692 7693 ill = ill_first(list, list, &ctx, ipst);
7693 7694 for (; ill != NULL; ill = ill_next(&ctx, ill)) {
7694 7695 if (IS_UNDER_IPMP(ill) && !(flags & LIFC_UNDER_IPMP))
7695 7696 continue;
7696 7697
7697 7698 for (ipif = ill->ill_ipif; ipif != NULL;
7698 7699 ipif = ipif->ipif_next) {
7699 7700 if ((ipif->ipif_flags & IPIF_NOXMIT) &&
7700 7701 !(flags & LIFC_NOXMIT))
7701 7702 continue;
7702 7703
7703 7704 if ((ipif->ipif_flags & IPIF_TEMPORARY) &&
7704 7705 !(flags & LIFC_TEMPORARY))
7705 7706 continue;
7706 7707
7707 7708 if (((ipif->ipif_flags &
7708 7709 (IPIF_NOXMIT|IPIF_NOLOCAL|
7709 7710 IPIF_DEPRECATED)) ||
7710 7711 IS_LOOPBACK(ill) ||
7711 7712 !(ipif->ipif_flags & IPIF_UP)) &&
7712 7713 (flags & LIFC_EXTERNAL_SOURCE))
7713 7714 continue;
7714 7715
7715 7716 if (zoneid != ipif->ipif_zoneid &&
7716 7717 ipif->ipif_zoneid != ALL_ZONES &&
7717 7718 (zoneid != GLOBAL_ZONEID ||
7718 7719 !(flags & LIFC_ALLZONES)))
7719 7720 continue;
7720 7721
7721 7722 if ((uchar_t *)&lifr[1] > mp1->b_wptr) {
7722 7723 if (iocp->ioc_cmd == O_SIOCGLIFCONF) {
7723 7724 rw_exit(&ipst->ips_ill_g_lock);
7724 7725 return (EINVAL);
7725 7726 } else {
7726 7727 goto lif_copydone;
7727 7728 }
7728 7729 }
7729 7730
7730 7731 ipif_get_name(ipif, lifr->lifr_name,
7731 7732 sizeof (lifr->lifr_name));
7732 7733 lifr->lifr_type = ill->ill_type;
7733 7734 if (ipif->ipif_isv6) {
7734 7735 sin6 = (sin6_t *)&lifr->lifr_addr;
7735 7736 *sin6 = sin6_null;
7736 7737 sin6->sin6_family = AF_INET6;
7737 7738 sin6->sin6_addr =
7738 7739 ipif->ipif_v6lcl_addr;
7739 7740 lifr->lifr_addrlen =
7740 7741 ip_mask_to_plen_v6(
7741 7742 &ipif->ipif_v6net_mask);
7742 7743 } else {
7743 7744 sin = (sin_t *)&lifr->lifr_addr;
7744 7745 *sin = sin_null;
7745 7746 sin->sin_family = AF_INET;
7746 7747 sin->sin_addr.s_addr =
7747 7748 ipif->ipif_lcl_addr;
7748 7749 lifr->lifr_addrlen =
7749 7750 ip_mask_to_plen(
7750 7751 ipif->ipif_net_mask);
7751 7752 }
7752 7753 lifr++;
7753 7754 }
7754 7755 }
7755 7756 lif_copydone:
7756 7757 rw_exit(&ipst->ips_ill_g_lock);
7757 7758
7758 7759 mp1->b_wptr = (uchar_t *)lifr;
7759 7760 if (STRUCT_BUF(lifc) != NULL) {
7760 7761 STRUCT_FSET(lifc, lifc_len,
7761 7762 (int)((uchar_t *)lifr - mp1->b_rptr));
7762 7763 }
7763 7764 return (0);
7764 7765 }
7765 7766
7766 7767 static void
7767 7768 ip_sioctl_ip6addrpolicy(queue_t *q, mblk_t *mp)
7768 7769 {
7769 7770 ip6_asp_t *table;
7770 7771 size_t table_size;
7771 7772 mblk_t *data_mp;
7772 7773 struct iocblk *iocp = (struct iocblk *)mp->b_rptr;
7773 7774 ip_stack_t *ipst;
7774 7775
7775 7776 if (q->q_next == NULL)
7776 7777 ipst = CONNQ_TO_IPST(q);
7777 7778 else
7778 7779 ipst = ILLQ_TO_IPST(q);
7779 7780
7780 7781 /* These two ioctls are I_STR only */
7781 7782 if (iocp->ioc_count == TRANSPARENT) {
7782 7783 miocnak(q, mp, 0, EINVAL);
7783 7784 return;
7784 7785 }
7785 7786
7786 7787 data_mp = mp->b_cont;
7787 7788 if (data_mp == NULL) {
7788 7789 /* The user passed us a NULL argument */
7789 7790 table = NULL;
7790 7791 table_size = iocp->ioc_count;
7791 7792 } else {
7792 7793 /*
7793 7794 * The user provided a table. The stream head
7794 7795 * may have copied in the user data in chunks,
7795 7796 * so make sure everything is pulled up
7796 7797 * properly.
7797 7798 */
7798 7799 if (MBLKL(data_mp) < iocp->ioc_count) {
7799 7800 mblk_t *new_data_mp;
7800 7801 if ((new_data_mp = msgpullup(data_mp, -1)) ==
7801 7802 NULL) {
7802 7803 miocnak(q, mp, 0, ENOMEM);
7803 7804 return;
7804 7805 }
7805 7806 freemsg(data_mp);
7806 7807 data_mp = new_data_mp;
7807 7808 mp->b_cont = data_mp;
7808 7809 }
7809 7810 table = (ip6_asp_t *)data_mp->b_rptr;
7810 7811 table_size = iocp->ioc_count;
7811 7812 }
7812 7813
7813 7814 switch (iocp->ioc_cmd) {
7814 7815 case SIOCGIP6ADDRPOLICY:
7815 7816 iocp->ioc_rval = ip6_asp_get(table, table_size, ipst);
7816 7817 if (iocp->ioc_rval == -1)
7817 7818 iocp->ioc_error = EINVAL;
7818 7819 #if defined(_SYSCALL32_IMPL) && _LONG_LONG_ALIGNMENT_32 == 4
7819 7820 else if (table != NULL &&
7820 7821 (iocp->ioc_flag & IOC_MODELS) == IOC_ILP32) {
7821 7822 ip6_asp_t *src = table;
7822 7823 ip6_asp32_t *dst = (void *)table;
7823 7824 int count = table_size / sizeof (ip6_asp_t);
7824 7825 int i;
7825 7826
7826 7827 /*
7827 7828 * We need to do an in-place shrink of the array
7828 7829 * to match the alignment attributes of the
7829 7830 * 32-bit ABI looking at it.
7830 7831 */
7831 7832 /* LINTED: logical expression always true: op "||" */
7832 7833 ASSERT(sizeof (*src) > sizeof (*dst));
7833 7834 for (i = 1; i < count; i++)
7834 7835 bcopy(src + i, dst + i, sizeof (*dst));
7835 7836 }
7836 7837 #endif
7837 7838 break;
7838 7839
7839 7840 case SIOCSIP6ADDRPOLICY:
7840 7841 ASSERT(mp->b_prev == NULL);
7841 7842 mp->b_prev = (void *)q;
7842 7843 #if defined(_SYSCALL32_IMPL) && _LONG_LONG_ALIGNMENT_32 == 4
7843 7844 /*
7844 7845 * We pass in the datamodel here so that the ip6_asp_replace()
7845 7846 * routine can handle converting from 32-bit to native formats
7846 7847 * where necessary.
7847 7848 *
7848 7849 * A better way to handle this might be to convert the inbound
7849 7850 * data structure here, and hang it off a new 'mp'; thus the
7850 7851 * ip6_asp_replace() logic would always be dealing with native
7851 7852 * format data structures..
7852 7853 *
7853 7854 * (An even simpler way to handle these ioctls is to just
7854 7855 * add a 32-bit trailing 'pad' field to the ip6_asp_t structure
7855 7856 * and just recompile everything that depends on it.)
7856 7857 */
7857 7858 #endif
7858 7859 ip6_asp_replace(mp, table, table_size, B_FALSE, ipst,
7859 7860 iocp->ioc_flag & IOC_MODELS);
7860 7861 return;
7861 7862 }
7862 7863
7863 7864 DB_TYPE(mp) = (iocp->ioc_error == 0) ? M_IOCACK : M_IOCNAK;
7864 7865 qreply(q, mp);
7865 7866 }
7866 7867
7867 7868 static void
7868 7869 ip_sioctl_dstinfo(queue_t *q, mblk_t *mp)
7869 7870 {
7870 7871 mblk_t *data_mp;
7871 7872 struct dstinforeq *dir;
7872 7873 uint8_t *end, *cur;
7873 7874 in6_addr_t *daddr, *saddr;
7874 7875 ipaddr_t v4daddr;
7875 7876 ire_t *ire;
7876 7877 ipaddr_t v4setsrc;
7877 7878 in6_addr_t v6setsrc;
7878 7879 char *slabel, *dlabel;
7879 7880 boolean_t isipv4;
7880 7881 int match_ire;
7881 7882 ill_t *dst_ill;
7882 7883 struct iocblk *iocp = (struct iocblk *)mp->b_rptr;
7883 7884 conn_t *connp = Q_TO_CONN(q);
7884 7885 zoneid_t zoneid = IPCL_ZONEID(connp);
7885 7886 ip_stack_t *ipst = connp->conn_netstack->netstack_ip;
7886 7887 uint64_t ipif_flags;
7887 7888
7888 7889 ASSERT(q->q_next == NULL); /* this ioctl not allowed if ip is module */
7889 7890
7890 7891 /*
7891 7892 * This ioctl is I_STR only, and must have a
7892 7893 * data mblk following the M_IOCTL mblk.
7893 7894 */
7894 7895 data_mp = mp->b_cont;
7895 7896 if (iocp->ioc_count == TRANSPARENT || data_mp == NULL) {
7896 7897 miocnak(q, mp, 0, EINVAL);
7897 7898 return;
7898 7899 }
7899 7900
7900 7901 if (MBLKL(data_mp) < iocp->ioc_count) {
7901 7902 mblk_t *new_data_mp;
7902 7903
7903 7904 if ((new_data_mp = msgpullup(data_mp, -1)) == NULL) {
7904 7905 miocnak(q, mp, 0, ENOMEM);
7905 7906 return;
7906 7907 }
7907 7908 freemsg(data_mp);
7908 7909 data_mp = new_data_mp;
7909 7910 mp->b_cont = data_mp;
7910 7911 }
7911 7912 match_ire = MATCH_IRE_DSTONLY;
7912 7913
7913 7914 for (cur = data_mp->b_rptr, end = data_mp->b_wptr;
7914 7915 end - cur >= sizeof (struct dstinforeq);
7915 7916 cur += sizeof (struct dstinforeq)) {
7916 7917 dir = (struct dstinforeq *)cur;
7917 7918 daddr = &dir->dir_daddr;
7918 7919 saddr = &dir->dir_saddr;
7919 7920
7920 7921 /*
7921 7922 * ip_addr_scope_v6() and ip6_asp_lookup() handle
7922 7923 * v4 mapped addresses; ire_ftable_lookup_v6()
7923 7924 * and ip_select_source_v6() do not.
7924 7925 */
7925 7926 dir->dir_dscope = ip_addr_scope_v6(daddr);
7926 7927 dlabel = ip6_asp_lookup(daddr, &dir->dir_precedence, ipst);
7927 7928
7928 7929 isipv4 = IN6_IS_ADDR_V4MAPPED(daddr);
7929 7930 if (isipv4) {
7930 7931 IN6_V4MAPPED_TO_IPADDR(daddr, v4daddr);
7931 7932 v4setsrc = INADDR_ANY;
7932 7933 ire = ire_route_recursive_v4(v4daddr, 0, NULL, zoneid,
7933 7934 NULL, match_ire, IRR_ALLOCATE, 0, ipst, &v4setsrc,
7934 7935 NULL, NULL);
7935 7936 } else {
7936 7937 v6setsrc = ipv6_all_zeros;
7937 7938 ire = ire_route_recursive_v6(daddr, 0, NULL, zoneid,
7938 7939 NULL, match_ire, IRR_ALLOCATE, 0, ipst, &v6setsrc,
7939 7940 NULL, NULL);
7940 7941 }
7941 7942 ASSERT(ire != NULL);
7942 7943 if (ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
7943 7944 ire_refrele(ire);
7944 7945 dir->dir_dreachable = 0;
7945 7946
7946 7947 /* move on to next dst addr */
7947 7948 continue;
7948 7949 }
7949 7950 dir->dir_dreachable = 1;
7950 7951
7951 7952 dst_ill = ire_nexthop_ill(ire);
7952 7953 if (dst_ill == NULL) {
7953 7954 ire_refrele(ire);
7954 7955 continue;
7955 7956 }
7956 7957
7957 7958 /* With ipmp we most likely look at the ipmp ill here */
7958 7959 dir->dir_dmactype = dst_ill->ill_mactype;
7959 7960
7960 7961 if (isipv4) {
7961 7962 ipaddr_t v4saddr;
7962 7963
7963 7964 if (ip_select_source_v4(dst_ill, v4setsrc, v4daddr,
7964 7965 connp->conn_ixa->ixa_multicast_ifaddr, zoneid, ipst,
7965 7966 &v4saddr, NULL, &ipif_flags) != 0) {
7966 7967 v4saddr = INADDR_ANY;
7967 7968 ipif_flags = 0;
7968 7969 }
7969 7970 IN6_IPADDR_TO_V4MAPPED(v4saddr, saddr);
7970 7971 } else {
7971 7972 if (ip_select_source_v6(dst_ill, &v6setsrc, daddr,
7972 7973 zoneid, ipst, B_FALSE, IPV6_PREFER_SRC_DEFAULT,
7973 7974 saddr, NULL, &ipif_flags) != 0) {
7974 7975 *saddr = ipv6_all_zeros;
7975 7976 ipif_flags = 0;
7976 7977 }
7977 7978 }
7978 7979
7979 7980 dir->dir_sscope = ip_addr_scope_v6(saddr);
7980 7981 slabel = ip6_asp_lookup(saddr, NULL, ipst);
7981 7982 dir->dir_labelmatch = ip6_asp_labelcmp(dlabel, slabel);
7982 7983 dir->dir_sdeprecated = (ipif_flags & IPIF_DEPRECATED) ? 1 : 0;
7983 7984 ire_refrele(ire);
7984 7985 ill_refrele(dst_ill);
7985 7986 }
7986 7987 miocack(q, mp, iocp->ioc_count, 0);
7987 7988 }
7988 7989
7989 7990 /*
7990 7991 * Check if this is an address assigned to this machine.
7991 7992 * Skips interfaces that are down by using ire checks.
7992 7993 * Translates mapped addresses to v4 addresses and then
7993 7994 * treats them as such, returning true if the v4 address
7994 7995 * associated with this mapped address is configured.
7995 7996 * Note: Applications will have to be careful what they do
7996 7997 * with the response; use of mapped addresses limits
7997 7998 * what can be done with the socket, especially with
7998 7999 * respect to socket options and ioctls - neither IPv4
7999 8000 * options nor IPv6 sticky options/ancillary data options
8000 8001 * may be used.
8001 8002 */
8002 8003 /* ARGSUSED */
8003 8004 int
8004 8005 ip_sioctl_tmyaddr(ipif_t *dummy_ipif, sin_t *dummy_sin, queue_t *q, mblk_t *mp,
8005 8006 ip_ioctl_cmd_t *ipip, void *dummy_ifreq)
8006 8007 {
8007 8008 struct sioc_addrreq *sia;
8008 8009 sin_t *sin;
8009 8010 ire_t *ire;
8010 8011 mblk_t *mp1;
8011 8012 zoneid_t zoneid;
8012 8013 ip_stack_t *ipst;
8013 8014
8014 8015 ip1dbg(("ip_sioctl_tmyaddr"));
8015 8016
8016 8017 ASSERT(q->q_next == NULL); /* this ioctl not allowed if ip is module */
8017 8018 zoneid = Q_TO_CONN(q)->conn_zoneid;
8018 8019 ipst = CONNQ_TO_IPST(q);
8019 8020
8020 8021 /* Existence verified in ip_wput_nondata */
8021 8022 mp1 = mp->b_cont->b_cont;
8022 8023 sia = (struct sioc_addrreq *)mp1->b_rptr;
8023 8024 sin = (sin_t *)&sia->sa_addr;
8024 8025 switch (sin->sin_family) {
8025 8026 case AF_INET6: {
8026 8027 sin6_t *sin6 = (sin6_t *)sin;
8027 8028
8028 8029 if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
8029 8030 ipaddr_t v4_addr;
8030 8031
8031 8032 IN6_V4MAPPED_TO_IPADDR(&sin6->sin6_addr,
8032 8033 v4_addr);
8033 8034 ire = ire_ftable_lookup_v4(v4_addr, 0, 0,
8034 8035 IRE_LOCAL|IRE_LOOPBACK, NULL, zoneid, NULL,
8035 8036 MATCH_IRE_TYPE | MATCH_IRE_ZONEONLY, 0, ipst, NULL);
8036 8037 } else {
8037 8038 in6_addr_t v6addr;
8038 8039
8039 8040 v6addr = sin6->sin6_addr;
8040 8041 ire = ire_ftable_lookup_v6(&v6addr, 0, 0,
8041 8042 IRE_LOCAL|IRE_LOOPBACK, NULL, zoneid, NULL,
8042 8043 MATCH_IRE_TYPE | MATCH_IRE_ZONEONLY, 0, ipst, NULL);
8043 8044 }
8044 8045 break;
8045 8046 }
8046 8047 case AF_INET: {
8047 8048 ipaddr_t v4addr;
8048 8049
8049 8050 v4addr = sin->sin_addr.s_addr;
8050 8051 ire = ire_ftable_lookup_v4(v4addr, 0, 0,
8051 8052 IRE_LOCAL|IRE_LOOPBACK, NULL, zoneid,
8052 8053 NULL, MATCH_IRE_TYPE | MATCH_IRE_ZONEONLY, 0, ipst, NULL);
8053 8054 break;
8054 8055 }
8055 8056 default:
8056 8057 return (EAFNOSUPPORT);
8057 8058 }
8058 8059 if (ire != NULL) {
8059 8060 sia->sa_res = 1;
8060 8061 ire_refrele(ire);
8061 8062 } else {
8062 8063 sia->sa_res = 0;
8063 8064 }
8064 8065 return (0);
8065 8066 }
8066 8067
8067 8068 /*
8068 8069 * Check if this is an address assigned on-link i.e. neighbor,
8069 8070 * and makes sure it's reachable from the current zone.
8070 8071 * Returns true for my addresses as well.
8071 8072 * Translates mapped addresses to v4 addresses and then
8072 8073 * treats them as such, returning true if the v4 address
8073 8074 * associated with this mapped address is configured.
8074 8075 * Note: Applications will have to be careful what they do
8075 8076 * with the response; use of mapped addresses limits
8076 8077 * what can be done with the socket, especially with
8077 8078 * respect to socket options and ioctls - neither IPv4
8078 8079 * options nor IPv6 sticky options/ancillary data options
8079 8080 * may be used.
8080 8081 */
8081 8082 /* ARGSUSED */
8082 8083 int
8083 8084 ip_sioctl_tonlink(ipif_t *dummy_ipif, sin_t *dummy_sin, queue_t *q, mblk_t *mp,
8084 8085 ip_ioctl_cmd_t *ipip, void *duymmy_ifreq)
8085 8086 {
8086 8087 struct sioc_addrreq *sia;
8087 8088 sin_t *sin;
8088 8089 mblk_t *mp1;
8089 8090 ire_t *ire = NULL;
8090 8091 zoneid_t zoneid;
8091 8092 ip_stack_t *ipst;
8092 8093
8093 8094 ip1dbg(("ip_sioctl_tonlink"));
8094 8095
8095 8096 ASSERT(q->q_next == NULL); /* this ioctl not allowed if ip is module */
8096 8097 zoneid = Q_TO_CONN(q)->conn_zoneid;
8097 8098 ipst = CONNQ_TO_IPST(q);
8098 8099
8099 8100 /* Existence verified in ip_wput_nondata */
8100 8101 mp1 = mp->b_cont->b_cont;
8101 8102 sia = (struct sioc_addrreq *)mp1->b_rptr;
8102 8103 sin = (sin_t *)&sia->sa_addr;
8103 8104
8104 8105 /*
8105 8106 * We check for IRE_ONLINK and exclude IRE_BROADCAST|IRE_MULTICAST
8106 8107 * to make sure we only look at on-link unicast address.
8107 8108 */
8108 8109 switch (sin->sin_family) {
8109 8110 case AF_INET6: {
8110 8111 sin6_t *sin6 = (sin6_t *)sin;
8111 8112
8112 8113 if (IN6_IS_ADDR_V4MAPPED(&sin6->sin6_addr)) {
8113 8114 ipaddr_t v4_addr;
8114 8115
8115 8116 IN6_V4MAPPED_TO_IPADDR(&sin6->sin6_addr,
8116 8117 v4_addr);
8117 8118 if (!CLASSD(v4_addr)) {
8118 8119 ire = ire_ftable_lookup_v4(v4_addr, 0, 0, 0,
8119 8120 NULL, zoneid, NULL, MATCH_IRE_DSTONLY,
8120 8121 0, ipst, NULL);
8121 8122 }
8122 8123 } else {
8123 8124 in6_addr_t v6addr;
8124 8125
8125 8126 v6addr = sin6->sin6_addr;
8126 8127 if (!IN6_IS_ADDR_MULTICAST(&v6addr)) {
8127 8128 ire = ire_ftable_lookup_v6(&v6addr, 0, 0, 0,
8128 8129 NULL, zoneid, NULL, MATCH_IRE_DSTONLY, 0,
8129 8130 ipst, NULL);
8130 8131 }
8131 8132 }
8132 8133 break;
8133 8134 }
8134 8135 case AF_INET: {
8135 8136 ipaddr_t v4addr;
8136 8137
8137 8138 v4addr = sin->sin_addr.s_addr;
8138 8139 if (!CLASSD(v4addr)) {
8139 8140 ire = ire_ftable_lookup_v4(v4addr, 0, 0, 0, NULL,
8140 8141 zoneid, NULL, MATCH_IRE_DSTONLY, 0, ipst, NULL);
8141 8142 }
8142 8143 break;
8143 8144 }
8144 8145 default:
8145 8146 return (EAFNOSUPPORT);
8146 8147 }
8147 8148 sia->sa_res = 0;
8148 8149 if (ire != NULL) {
8149 8150 ASSERT(!(ire->ire_type & IRE_MULTICAST));
8150 8151
8151 8152 if ((ire->ire_type & IRE_ONLINK) &&
8152 8153 !(ire->ire_type & IRE_BROADCAST))
8153 8154 sia->sa_res = 1;
8154 8155 ire_refrele(ire);
8155 8156 }
8156 8157 return (0);
8157 8158 }
8158 8159
8159 8160 /*
8160 8161 * TBD: implement when kernel maintaines a list of site prefixes.
8161 8162 */
8162 8163 /* ARGSUSED */
8163 8164 int
8164 8165 ip_sioctl_tmysite(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
8165 8166 ip_ioctl_cmd_t *ipip, void *ifreq)
8166 8167 {
8167 8168 return (ENXIO);
8168 8169 }
8169 8170
8170 8171 /* ARP IOCTLs. */
8171 8172 /* ARGSUSED */
8172 8173 int
8173 8174 ip_sioctl_arp(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
8174 8175 ip_ioctl_cmd_t *ipip, void *dummy_ifreq)
8175 8176 {
8176 8177 int err;
8177 8178 ipaddr_t ipaddr;
8178 8179 struct iocblk *iocp;
8179 8180 conn_t *connp;
8180 8181 struct arpreq *ar;
8181 8182 struct xarpreq *xar;
8182 8183 int arp_flags, flags, alength;
8183 8184 uchar_t *lladdr;
8184 8185 ip_stack_t *ipst;
8185 8186 ill_t *ill = ipif->ipif_ill;
8186 8187 ill_t *proxy_ill = NULL;
8187 8188 ipmp_arpent_t *entp = NULL;
8188 8189 boolean_t proxyarp = B_FALSE;
8189 8190 boolean_t if_arp_ioctl = B_FALSE;
8190 8191 ncec_t *ncec = NULL;
8191 8192 nce_t *nce;
8192 8193
8193 8194 ASSERT(!(q->q_flag & QREADR) && q->q_next == NULL);
8194 8195 connp = Q_TO_CONN(q);
8195 8196 ipst = connp->conn_netstack->netstack_ip;
8196 8197 iocp = (struct iocblk *)mp->b_rptr;
8197 8198
8198 8199 if (ipip->ipi_cmd_type == XARP_CMD) {
8199 8200 /* We have a chain - M_IOCTL-->MI_COPY_MBLK-->XARPREQ_MBLK */
8200 8201 xar = (struct xarpreq *)mp->b_cont->b_cont->b_rptr;
8201 8202 ar = NULL;
8202 8203
8203 8204 arp_flags = xar->xarp_flags;
8204 8205 lladdr = (uchar_t *)LLADDR(&xar->xarp_ha);
8205 8206 if_arp_ioctl = (xar->xarp_ha.sdl_nlen != 0);
8206 8207 /*
8207 8208 * Validate against user's link layer address length
8208 8209 * input and name and addr length limits.
8209 8210 */
8210 8211 alength = ill->ill_phys_addr_length;
8211 8212 if (ipip->ipi_cmd == SIOCSXARP) {
8212 8213 if (alength != xar->xarp_ha.sdl_alen ||
8213 8214 (alength + xar->xarp_ha.sdl_nlen >
8214 8215 sizeof (xar->xarp_ha.sdl_data)))
8215 8216 return (EINVAL);
8216 8217 }
8217 8218 } else {
8218 8219 /* We have a chain - M_IOCTL-->MI_COPY_MBLK-->ARPREQ_MBLK */
8219 8220 ar = (struct arpreq *)mp->b_cont->b_cont->b_rptr;
8220 8221 xar = NULL;
8221 8222
8222 8223 arp_flags = ar->arp_flags;
8223 8224 lladdr = (uchar_t *)ar->arp_ha.sa_data;
8224 8225 /*
8225 8226 * Theoretically, the sa_family could tell us what link
8226 8227 * layer type this operation is trying to deal with. By
8227 8228 * common usage AF_UNSPEC means ethernet. We'll assume
8228 8229 * any attempt to use the SIOC?ARP ioctls is for ethernet,
8229 8230 * for now. Our new SIOC*XARP ioctls can be used more
8230 8231 * generally.
8231 8232 *
8232 8233 * If the underlying media happens to have a non 6 byte
8233 8234 * address, arp module will fail set/get, but the del
8234 8235 * operation will succeed.
8235 8236 */
8236 8237 alength = 6;
8237 8238 if ((ipip->ipi_cmd != SIOCDARP) &&
8238 8239 (alength != ill->ill_phys_addr_length)) {
8239 8240 return (EINVAL);
8240 8241 }
8241 8242 }
8242 8243
8243 8244 /* Translate ATF* flags to NCE* flags */
8244 8245 flags = 0;
8245 8246 if (arp_flags & ATF_AUTHORITY)
8246 8247 flags |= NCE_F_AUTHORITY;
8247 8248 if (arp_flags & ATF_PERM)
8248 8249 flags |= NCE_F_NONUD; /* not subject to aging */
8249 8250 if (arp_flags & ATF_PUBL)
8250 8251 flags |= NCE_F_PUBLISH;
8251 8252
8252 8253 /*
8253 8254 * IPMP ARP special handling:
8254 8255 *
8255 8256 * 1. Since ARP mappings must appear consistent across the group,
8256 8257 * prohibit changing ARP mappings on the underlying interfaces.
8257 8258 *
8258 8259 * 2. Since ARP mappings for IPMP data addresses are maintained by
8259 8260 * IP itself, prohibit changing them.
8260 8261 *
8261 8262 * 3. For proxy ARP, use a functioning hardware address in the group,
8262 8263 * provided one exists. If one doesn't, just add the entry as-is;
8263 8264 * ipmp_illgrp_refresh_arpent() will refresh it if things change.
8264 8265 */
8265 8266 if (IS_UNDER_IPMP(ill)) {
8266 8267 if (ipip->ipi_cmd != SIOCGARP && ipip->ipi_cmd != SIOCGXARP)
8267 8268 return (EPERM);
8268 8269 }
8269 8270 if (IS_IPMP(ill)) {
8270 8271 ipmp_illgrp_t *illg = ill->ill_grp;
8271 8272
8272 8273 switch (ipip->ipi_cmd) {
8273 8274 case SIOCSARP:
8274 8275 case SIOCSXARP:
8275 8276 proxy_ill = ipmp_illgrp_find_ill(illg, lladdr, alength);
8276 8277 if (proxy_ill != NULL) {
8277 8278 proxyarp = B_TRUE;
8278 8279 if (!ipmp_ill_is_active(proxy_ill))
8279 8280 proxy_ill = ipmp_illgrp_next_ill(illg);
8280 8281 if (proxy_ill != NULL)
8281 8282 lladdr = proxy_ill->ill_phys_addr;
8282 8283 }
8283 8284 /* FALLTHRU */
8284 8285 }
8285 8286 }
8286 8287
8287 8288 ipaddr = sin->sin_addr.s_addr;
8288 8289 /*
8289 8290 * don't match across illgrp per case (1) and (2).
8290 8291 * XXX use IS_IPMP(ill) like ndp_sioc_update?
8291 8292 */
8292 8293 nce = nce_lookup_v4(ill, &ipaddr);
8293 8294 if (nce != NULL)
8294 8295 ncec = nce->nce_common;
8295 8296
8296 8297 switch (iocp->ioc_cmd) {
8297 8298 case SIOCDARP:
8298 8299 case SIOCDXARP: {
8299 8300 /*
8300 8301 * Delete the NCE if any.
8301 8302 */
8302 8303 if (ncec == NULL) {
8303 8304 iocp->ioc_error = ENXIO;
8304 8305 break;
8305 8306 }
8306 8307 /* Don't allow changes to arp mappings of local addresses. */
8307 8308 if (NCE_MYADDR(ncec)) {
8308 8309 nce_refrele(nce);
8309 8310 return (ENOTSUP);
8310 8311 }
8311 8312 iocp->ioc_error = 0;
8312 8313
8313 8314 /*
8314 8315 * Delete the nce_common which has ncec_ill set to ipmp_ill.
8315 8316 * This will delete all the nce entries on the under_ills.
8316 8317 */
8317 8318 ncec_delete(ncec);
8318 8319 /*
8319 8320 * Once the NCE has been deleted, then the ire_dep* consistency
8320 8321 * mechanism will find any IRE which depended on the now
8321 8322 * condemned NCE (as part of sending packets).
8322 8323 * That mechanism handles redirects by deleting redirects
8323 8324 * that refer to UNREACHABLE nces.
8324 8325 */
8325 8326 break;
8326 8327 }
8327 8328 case SIOCGARP:
8328 8329 case SIOCGXARP:
8329 8330 if (ncec != NULL) {
8330 8331 lladdr = ncec->ncec_lladdr;
8331 8332 flags = ncec->ncec_flags;
8332 8333 iocp->ioc_error = 0;
8333 8334 ip_sioctl_garp_reply(mp, ncec->ncec_ill, lladdr, flags);
8334 8335 } else {
8335 8336 iocp->ioc_error = ENXIO;
8336 8337 }
8337 8338 break;
8338 8339 case SIOCSARP:
8339 8340 case SIOCSXARP:
8340 8341 /* Don't allow changes to arp mappings of local addresses. */
8341 8342 if (ncec != NULL && NCE_MYADDR(ncec)) {
8342 8343 nce_refrele(nce);
8343 8344 return (ENOTSUP);
8344 8345 }
8345 8346
8346 8347 /* static arp entries will undergo NUD if ATF_PERM is not set */
8347 8348 flags |= NCE_F_STATIC;
8348 8349 if (!if_arp_ioctl) {
8349 8350 ip_nce_lookup_and_update(&ipaddr, NULL, ipst,
8350 8351 lladdr, alength, flags);
8351 8352 } else {
8352 8353 ipif_t *ipif = ipif_get_next_ipif(NULL, ill);
8353 8354 if (ipif != NULL) {
8354 8355 ip_nce_lookup_and_update(&ipaddr, ipif, ipst,
8355 8356 lladdr, alength, flags);
8356 8357 ipif_refrele(ipif);
8357 8358 }
8358 8359 }
8359 8360 if (nce != NULL) {
8360 8361 nce_refrele(nce);
8361 8362 nce = NULL;
8362 8363 }
8363 8364 /*
8364 8365 * NCE_F_STATIC entries will be added in state ND_REACHABLE
8365 8366 * by nce_add_common()
8366 8367 */
8367 8368 err = nce_lookup_then_add_v4(ill, lladdr,
8368 8369 ill->ill_phys_addr_length, &ipaddr, flags, ND_UNCHANGED,
8369 8370 &nce);
8370 8371 if (err == EEXIST) {
8371 8372 ncec = nce->nce_common;
8372 8373 mutex_enter(&ncec->ncec_lock);
8373 8374 ncec->ncec_state = ND_REACHABLE;
8374 8375 ncec->ncec_flags = flags;
8375 8376 nce_update(ncec, ND_UNCHANGED, lladdr);
8376 8377 mutex_exit(&ncec->ncec_lock);
8377 8378 err = 0;
8378 8379 }
8379 8380 if (nce != NULL) {
8380 8381 nce_refrele(nce);
8381 8382 nce = NULL;
8382 8383 }
8383 8384 if (IS_IPMP(ill) && err == 0) {
8384 8385 entp = ipmp_illgrp_create_arpent(ill->ill_grp,
8385 8386 proxyarp, ipaddr, lladdr, ill->ill_phys_addr_length,
8386 8387 flags);
8387 8388 if (entp == NULL || (proxyarp && proxy_ill == NULL)) {
8388 8389 iocp->ioc_error = (entp == NULL ? ENOMEM : 0);
8389 8390 break;
8390 8391 }
8391 8392 }
8392 8393 iocp->ioc_error = err;
8393 8394 }
8394 8395
8395 8396 if (nce != NULL) {
8396 8397 nce_refrele(nce);
8397 8398 }
8398 8399
8399 8400 /*
8400 8401 * If we created an IPMP ARP entry, mark that we've notified ARP.
8401 8402 */
8402 8403 if (entp != NULL)
8403 8404 ipmp_illgrp_mark_arpent(ill->ill_grp, entp);
8404 8405
8405 8406 return (iocp->ioc_error);
8406 8407 }
8407 8408
8408 8409 /*
8409 8410 * Parse an [x]arpreq structure coming down SIOC[GSD][X]ARP ioctls, identify
8410 8411 * the associated sin and refhold and return the associated ipif via `ci'.
8411 8412 */
8412 8413 int
8413 8414 ip_extract_arpreq(queue_t *q, mblk_t *mp, const ip_ioctl_cmd_t *ipip,
8414 8415 cmd_info_t *ci)
8415 8416 {
8416 8417 mblk_t *mp1;
8417 8418 sin_t *sin;
8418 8419 conn_t *connp;
8419 8420 ipif_t *ipif;
8420 8421 ire_t *ire = NULL;
8421 8422 ill_t *ill = NULL;
8422 8423 boolean_t exists;
8423 8424 ip_stack_t *ipst;
8424 8425 struct arpreq *ar;
8425 8426 struct xarpreq *xar;
8426 8427 struct sockaddr_dl *sdl;
8427 8428
8428 8429 /* ioctl comes down on a conn */
8429 8430 ASSERT(!(q->q_flag & QREADR) && q->q_next == NULL);
8430 8431 connp = Q_TO_CONN(q);
8431 8432 if (connp->conn_family == AF_INET6)
8432 8433 return (ENXIO);
8433 8434
8434 8435 ipst = connp->conn_netstack->netstack_ip;
8435 8436
8436 8437 /* Verified in ip_wput_nondata */
8437 8438 mp1 = mp->b_cont->b_cont;
8438 8439
8439 8440 if (ipip->ipi_cmd_type == XARP_CMD) {
8440 8441 ASSERT(MBLKL(mp1) >= sizeof (struct xarpreq));
8441 8442 xar = (struct xarpreq *)mp1->b_rptr;
8442 8443 sin = (sin_t *)&xar->xarp_pa;
8443 8444 sdl = &xar->xarp_ha;
8444 8445
8445 8446 if (sdl->sdl_family != AF_LINK || sin->sin_family != AF_INET)
8446 8447 return (ENXIO);
8447 8448 if (sdl->sdl_nlen >= LIFNAMSIZ)
8448 8449 return (EINVAL);
8449 8450 } else {
8450 8451 ASSERT(ipip->ipi_cmd_type == ARP_CMD);
8451 8452 ASSERT(MBLKL(mp1) >= sizeof (struct arpreq));
8452 8453 ar = (struct arpreq *)mp1->b_rptr;
8453 8454 sin = (sin_t *)&ar->arp_pa;
8454 8455 }
8455 8456
8456 8457 if (ipip->ipi_cmd_type == XARP_CMD && sdl->sdl_nlen != 0) {
8457 8458 ipif = ipif_lookup_on_name(sdl->sdl_data, sdl->sdl_nlen,
8458 8459 B_FALSE, &exists, B_FALSE, ALL_ZONES, ipst);
8459 8460 if (ipif == NULL)
8460 8461 return (ENXIO);
8461 8462 if (ipif->ipif_id != 0) {
8462 8463 ipif_refrele(ipif);
8463 8464 return (ENXIO);
8464 8465 }
8465 8466 } else {
8466 8467 /*
8467 8468 * Either an SIOC[DGS]ARP or an SIOC[DGS]XARP with an sdl_nlen
8468 8469 * of 0: use the IP address to find the ipif. If the IP
8469 8470 * address is an IPMP test address, ire_ftable_lookup() will
8470 8471 * find the wrong ill, so we first do an ipif_lookup_addr().
8471 8472 */
8472 8473 ipif = ipif_lookup_addr(sin->sin_addr.s_addr, NULL, ALL_ZONES,
8473 8474 ipst);
8474 8475 if (ipif == NULL) {
8475 8476 ire = ire_ftable_lookup_v4(sin->sin_addr.s_addr,
8476 8477 0, 0, IRE_IF_RESOLVER, NULL, ALL_ZONES,
8477 8478 NULL, MATCH_IRE_TYPE, 0, ipst, NULL);
8478 8479 if (ire == NULL || ((ill = ire->ire_ill) == NULL)) {
8479 8480 if (ire != NULL)
8480 8481 ire_refrele(ire);
8481 8482 return (ENXIO);
8482 8483 }
8483 8484 ASSERT(ire != NULL && ill != NULL);
8484 8485 ipif = ill->ill_ipif;
8485 8486 ipif_refhold(ipif);
8486 8487 ire_refrele(ire);
8487 8488 }
8488 8489 }
8489 8490
8490 8491 if (ipif->ipif_ill->ill_net_type != IRE_IF_RESOLVER) {
8491 8492 ipif_refrele(ipif);
8492 8493 return (ENXIO);
8493 8494 }
8494 8495
8495 8496 ci->ci_sin = sin;
8496 8497 ci->ci_ipif = ipif;
8497 8498 return (0);
8498 8499 }
8499 8500
8500 8501 /*
8501 8502 * Link or unlink the illgrp on IPMP meta-interface `ill' depending on the
8502 8503 * value of `ioccmd'. While an illgrp is linked to an ipmp_grp_t, it is
8503 8504 * accessible from that ipmp_grp_t, which means SIOCSLIFGROUPNAME can look it
8504 8505 * up and thus an ill can join that illgrp.
8505 8506 *
8506 8507 * We use I_PLINK/I_PUNLINK to do the link/unlink operations rather than
8507 8508 * open()/close() primarily because close() is not allowed to fail or block
8508 8509 * forever. On the other hand, I_PUNLINK *can* fail, and there's no reason
8509 8510 * why anyone should ever need to I_PUNLINK an in-use IPMP stream. To ensure
8510 8511 * symmetric behavior (e.g., doing an I_PLINK after and I_PUNLINK undoes the
8511 8512 * I_PUNLINK) we defer linking to I_PLINK. Separately, we also fail attempts
8512 8513 * to I_LINK since I_UNLINK is optional and we'd end up in an inconsistent
8513 8514 * state if I_UNLINK didn't occur.
8514 8515 *
8515 8516 * Note that for each plumb/unplumb operation, we may end up here more than
8516 8517 * once because of the way ifconfig works. However, it's OK to link the same
8517 8518 * illgrp more than once, or unlink an illgrp that's already unlinked.
8518 8519 */
8519 8520 static int
8520 8521 ip_sioctl_plink_ipmp(ill_t *ill, int ioccmd)
8521 8522 {
8522 8523 int err;
8523 8524 ip_stack_t *ipst = ill->ill_ipst;
8524 8525
8525 8526 ASSERT(IS_IPMP(ill));
8526 8527 ASSERT(IAM_WRITER_ILL(ill));
8527 8528
8528 8529 switch (ioccmd) {
8529 8530 case I_LINK:
8530 8531 return (ENOTSUP);
8531 8532
8532 8533 case I_PLINK:
8533 8534 rw_enter(&ipst->ips_ipmp_lock, RW_WRITER);
8534 8535 ipmp_illgrp_link_grp(ill->ill_grp, ill->ill_phyint->phyint_grp);
8535 8536 rw_exit(&ipst->ips_ipmp_lock);
8536 8537 break;
8537 8538
8538 8539 case I_PUNLINK:
8539 8540 /*
8540 8541 * Require all UP ipifs be brought down prior to unlinking the
8541 8542 * illgrp so any associated IREs (and other state) is torched.
8542 8543 */
8543 8544 if (ill->ill_ipif_up_count + ill->ill_ipif_dup_count > 0)
8544 8545 return (EBUSY);
8545 8546
8546 8547 /*
8547 8548 * NOTE: We hold ipmp_lock across the unlink to prevent a race
8548 8549 * with an SIOCSLIFGROUPNAME request from an ill trying to
8549 8550 * join this group. Specifically: ills trying to join grab
8550 8551 * ipmp_lock and bump a "pending join" counter checked by
8551 8552 * ipmp_illgrp_unlink_grp(). During the unlink no new pending
8552 8553 * joins can occur (since we have ipmp_lock). Once we drop
8553 8554 * ipmp_lock, subsequent SIOCSLIFGROUPNAME requests will not
8554 8555 * find the illgrp (since we unlinked it) and will return
8555 8556 * EAFNOSUPPORT. This will then take them back through the
8556 8557 * IPMP meta-interface plumbing logic in ifconfig, and thus
8557 8558 * back through I_PLINK above.
8558 8559 */
8559 8560 rw_enter(&ipst->ips_ipmp_lock, RW_WRITER);
8560 8561 err = ipmp_illgrp_unlink_grp(ill->ill_grp);
8561 8562 rw_exit(&ipst->ips_ipmp_lock);
8562 8563 return (err);
8563 8564 default:
8564 8565 break;
8565 8566 }
8566 8567 return (0);
8567 8568 }
8568 8569
8569 8570 /*
8570 8571 * Do I_PLINK/I_LINK or I_PUNLINK/I_UNLINK with consistency checks and also
8571 8572 * atomically set/clear the muxids. Also complete the ioctl by acking or
8572 8573 * naking it. Note that the code is structured such that the link type,
8573 8574 * whether it's persistent or not, is treated equally. ifconfig(1M) and
8574 8575 * its clones use the persistent link, while pppd(1M) and perhaps many
8575 8576 * other daemons may use non-persistent link. When combined with some
8576 8577 * ill_t states, linking and unlinking lower streams may be used as
8577 8578 * indicators of dynamic re-plumbing events [see PSARC/1999/348].
8578 8579 */
8579 8580 /* ARGSUSED */
8580 8581 void
8581 8582 ip_sioctl_plink(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *dummy_arg)
8582 8583 {
8583 8584 mblk_t *mp1;
8584 8585 struct linkblk *li;
8585 8586 int ioccmd = ((struct iocblk *)mp->b_rptr)->ioc_cmd;
8586 8587 int err = 0;
8587 8588
8588 8589 ASSERT(ioccmd == I_PLINK || ioccmd == I_PUNLINK ||
8589 8590 ioccmd == I_LINK || ioccmd == I_UNLINK);
8590 8591
8591 8592 mp1 = mp->b_cont; /* This is the linkblk info */
8592 8593 li = (struct linkblk *)mp1->b_rptr;
8593 8594
8594 8595 err = ip_sioctl_plink_ipmod(ipsq, q, mp, ioccmd, li);
8595 8596 if (err == EINPROGRESS)
8596 8597 return;
8597 8598 if (err == 0)
8598 8599 miocack(q, mp, 0, 0);
8599 8600 else
8600 8601 miocnak(q, mp, 0, err);
8601 8602
8602 8603 /* Conn was refheld in ip_sioctl_copyin_setup */
8603 8604 if (CONN_Q(q)) {
8604 8605 CONN_DEC_IOCTLREF(Q_TO_CONN(q));
8605 8606 CONN_OPER_PENDING_DONE(Q_TO_CONN(q));
8606 8607 }
8607 8608 }
8608 8609
8609 8610 /*
8610 8611 * Process I_{P}LINK and I_{P}UNLINK requests named by `ioccmd' and pointed to
8611 8612 * by `mp' and `li' for the IP module stream (if li->q_bot is in fact an IP
8612 8613 * module stream).
8613 8614 * Returns zero on success, EINPROGRESS if the operation is still pending, or
8614 8615 * an error code on failure.
8615 8616 */
8616 8617 static int
8617 8618 ip_sioctl_plink_ipmod(ipsq_t *ipsq, queue_t *q, mblk_t *mp, int ioccmd,
8618 8619 struct linkblk *li)
8619 8620 {
8620 8621 int err = 0;
8621 8622 ill_t *ill;
8622 8623 queue_t *ipwq, *dwq;
8623 8624 const char *name;
8624 8625 struct qinit *qinfo;
8625 8626 boolean_t islink = (ioccmd == I_PLINK || ioccmd == I_LINK);
8626 8627 boolean_t entered_ipsq = B_FALSE;
8627 8628 boolean_t is_ip = B_FALSE;
8628 8629 arl_t *arl;
8629 8630
8630 8631 /*
8631 8632 * Walk the lower stream to verify it's the IP module stream.
8632 8633 * The IP module is identified by its name, wput function,
8633 8634 * and non-NULL q_next. STREAMS ensures that the lower stream
8634 8635 * (li->l_qbot) will not vanish until this ioctl completes.
8635 8636 */
8636 8637 for (ipwq = li->l_qbot; ipwq != NULL; ipwq = ipwq->q_next) {
8637 8638 qinfo = ipwq->q_qinfo;
8638 8639 name = qinfo->qi_minfo->mi_idname;
8639 8640 if (name != NULL && strcmp(name, ip_mod_info.mi_idname) == 0 &&
8640 8641 qinfo->qi_putp != (pfi_t)ip_lwput && ipwq->q_next != NULL) {
8641 8642 is_ip = B_TRUE;
8642 8643 break;
8643 8644 }
8644 8645 if (name != NULL && strcmp(name, arp_mod_info.mi_idname) == 0 &&
8645 8646 qinfo->qi_putp != (pfi_t)ip_lwput && ipwq->q_next != NULL) {
8646 8647 break;
8647 8648 }
8648 8649 }
8649 8650
8650 8651 /*
8651 8652 * If this isn't an IP module stream, bail.
8652 8653 */
8653 8654 if (ipwq == NULL)
8654 8655 return (0);
8655 8656
8656 8657 if (!is_ip) {
8657 8658 arl = (arl_t *)ipwq->q_ptr;
8658 8659 ill = arl_to_ill(arl);
8659 8660 if (ill == NULL)
8660 8661 return (0);
8661 8662 } else {
8662 8663 ill = ipwq->q_ptr;
8663 8664 }
8664 8665 ASSERT(ill != NULL);
8665 8666
8666 8667 if (ipsq == NULL) {
8667 8668 ipsq = ipsq_try_enter(NULL, ill, q, mp, ip_sioctl_plink,
8668 8669 NEW_OP, B_FALSE);
8669 8670 if (ipsq == NULL) {
8670 8671 if (!is_ip)
8671 8672 ill_refrele(ill);
8672 8673 return (EINPROGRESS);
8673 8674 }
8674 8675 entered_ipsq = B_TRUE;
8675 8676 }
8676 8677 ASSERT(IAM_WRITER_ILL(ill));
8677 8678 mutex_enter(&ill->ill_lock);
8678 8679 if (!is_ip) {
8679 8680 if (islink && ill->ill_muxid == 0) {
8680 8681 /*
8681 8682 * Plumbing has to be done with IP plumbed first, arp
8682 8683 * second, but here we have arp being plumbed first.
8683 8684 */
8684 8685 mutex_exit(&ill->ill_lock);
8685 8686 if (entered_ipsq)
8686 8687 ipsq_exit(ipsq);
8687 8688 ill_refrele(ill);
8688 8689 return (EINVAL);
8689 8690 }
8690 8691 }
8691 8692 mutex_exit(&ill->ill_lock);
8692 8693 if (!is_ip) {
8693 8694 arl->arl_muxid = islink ? li->l_index : 0;
8694 8695 ill_refrele(ill);
8695 8696 goto done;
8696 8697 }
8697 8698
8698 8699 if (IS_IPMP(ill) && (err = ip_sioctl_plink_ipmp(ill, ioccmd)) != 0)
8699 8700 goto done;
8700 8701
8701 8702 /*
8702 8703 * As part of I_{P}LINKing, stash the number of downstream modules and
8703 8704 * the read queue of the module immediately below IP in the ill.
8704 8705 * These are used during the capability negotiation below.
8705 8706 */
8706 8707 ill->ill_lmod_rq = NULL;
8707 8708 ill->ill_lmod_cnt = 0;
8708 8709 if (islink && ((dwq = ipwq->q_next) != NULL)) {
8709 8710 ill->ill_lmod_rq = RD(dwq);
8710 8711 for (; dwq != NULL; dwq = dwq->q_next)
8711 8712 ill->ill_lmod_cnt++;
8712 8713 }
8713 8714
8714 8715 ill->ill_muxid = islink ? li->l_index : 0;
8715 8716
8716 8717 /*
8717 8718 * Mark the ipsq busy until the capability operations initiated below
8718 8719 * complete. The PLINK/UNLINK ioctl itself completes when our caller
8719 8720 * returns, but the capability operation may complete asynchronously
8720 8721 * much later.
8721 8722 */
8722 8723 ipsq_current_start(ipsq, ill->ill_ipif, ioccmd);
8723 8724 /*
8724 8725 * If there's at least one up ipif on this ill, then we're bound to
8725 8726 * the underlying driver via DLPI. In that case, renegotiate
8726 8727 * capabilities to account for any possible change in modules
8727 8728 * interposed between IP and the driver.
8728 8729 */
8729 8730 if (ill->ill_ipif_up_count > 0) {
8730 8731 if (islink)
8731 8732 ill_capability_probe(ill);
8732 8733 else
8733 8734 ill_capability_reset(ill, B_FALSE);
8734 8735 }
8735 8736 ipsq_current_finish(ipsq);
8736 8737 done:
8737 8738 if (entered_ipsq)
8738 8739 ipsq_exit(ipsq);
8739 8740
8740 8741 return (err);
8741 8742 }
8742 8743
8743 8744 /*
8744 8745 * Search the ioctl command in the ioctl tables and return a pointer
8745 8746 * to the ioctl command information. The ioctl command tables are
8746 8747 * static and fully populated at compile time.
8747 8748 */
8748 8749 ip_ioctl_cmd_t *
8749 8750 ip_sioctl_lookup(int ioc_cmd)
8750 8751 {
8751 8752 int index;
8752 8753 ip_ioctl_cmd_t *ipip;
8753 8754 ip_ioctl_cmd_t *ipip_end;
8754 8755
8755 8756 if (ioc_cmd == IPI_DONTCARE)
8756 8757 return (NULL);
8757 8758
8758 8759 /*
8759 8760 * Do a 2 step search. First search the indexed table
8760 8761 * based on the least significant byte of the ioctl cmd.
8761 8762 * If we don't find a match, then search the misc table
8762 8763 * serially.
8763 8764 */
8764 8765 index = ioc_cmd & 0xFF;
8765 8766 if (index < ip_ndx_ioctl_count) {
8766 8767 ipip = &ip_ndx_ioctl_table[index];
8767 8768 if (ipip->ipi_cmd == ioc_cmd) {
8768 8769 /* Found a match in the ndx table */
8769 8770 return (ipip);
8770 8771 }
8771 8772 }
8772 8773
8773 8774 /* Search the misc table */
8774 8775 ipip_end = &ip_misc_ioctl_table[ip_misc_ioctl_count];
8775 8776 for (ipip = ip_misc_ioctl_table; ipip < ipip_end; ipip++) {
8776 8777 if (ipip->ipi_cmd == ioc_cmd)
8777 8778 /* Found a match in the misc table */
8778 8779 return (ipip);
8779 8780 }
8780 8781
8781 8782 return (NULL);
8782 8783 }
8783 8784
8784 8785 /*
8785 8786 * helper function for ip_sioctl_getsetprop(), which does some sanity checks
8786 8787 */
8787 8788 static boolean_t
8788 8789 getset_ioctl_checks(mblk_t *mp)
8789 8790 {
8790 8791 struct iocblk *iocp = (struct iocblk *)mp->b_rptr;
8791 8792 mblk_t *mp1 = mp->b_cont;
8792 8793 mod_ioc_prop_t *pioc;
8793 8794 uint_t flags;
8794 8795 uint_t pioc_size;
8795 8796
8796 8797 /* do sanity checks on various arguments */
8797 8798 if (mp1 == NULL || iocp->ioc_count == 0 ||
8798 8799 iocp->ioc_count == TRANSPARENT) {
8799 8800 return (B_FALSE);
8800 8801 }
8801 8802 if (msgdsize(mp1) < iocp->ioc_count) {
8802 8803 if (!pullupmsg(mp1, iocp->ioc_count))
8803 8804 return (B_FALSE);
8804 8805 }
8805 8806
8806 8807 pioc = (mod_ioc_prop_t *)mp1->b_rptr;
8807 8808
8808 8809 /* sanity checks on mpr_valsize */
8809 8810 pioc_size = sizeof (mod_ioc_prop_t);
8810 8811 if (pioc->mpr_valsize != 0)
8811 8812 pioc_size += pioc->mpr_valsize - 1;
8812 8813
8813 8814 if (iocp->ioc_count != pioc_size)
8814 8815 return (B_FALSE);
8815 8816
8816 8817 flags = pioc->mpr_flags;
8817 8818 if (iocp->ioc_cmd == SIOCSETPROP) {
8818 8819 /*
8819 8820 * One can either reset the value to it's default value or
8820 8821 * change the current value or append/remove the value from
8821 8822 * a multi-valued properties.
8822 8823 */
8823 8824 if ((flags & MOD_PROP_DEFAULT) != MOD_PROP_DEFAULT &&
8824 8825 flags != MOD_PROP_ACTIVE &&
8825 8826 flags != (MOD_PROP_ACTIVE|MOD_PROP_APPEND) &&
8826 8827 flags != (MOD_PROP_ACTIVE|MOD_PROP_REMOVE))
8827 8828 return (B_FALSE);
8828 8829 } else {
8829 8830 ASSERT(iocp->ioc_cmd == SIOCGETPROP);
8830 8831
8831 8832 /*
8832 8833 * One can retrieve only one kind of property information
8833 8834 * at a time.
8834 8835 */
8835 8836 if ((flags & MOD_PROP_ACTIVE) != MOD_PROP_ACTIVE &&
8836 8837 (flags & MOD_PROP_DEFAULT) != MOD_PROP_DEFAULT &&
8837 8838 (flags & MOD_PROP_POSSIBLE) != MOD_PROP_POSSIBLE &&
8838 8839 (flags & MOD_PROP_PERM) != MOD_PROP_PERM)
8839 8840 return (B_FALSE);
8840 8841 }
8841 8842
8842 8843 return (B_TRUE);
8843 8844 }
8844 8845
8845 8846 /*
8846 8847 * process the SIOC{SET|GET}PROP ioctl's
8847 8848 */
8848 8849 /* ARGSUSED */
8849 8850 static void
|
↓ open down ↓ |
8733 lines elided |
↑ open up ↑ |
8850 8851 ip_sioctl_getsetprop(queue_t *q, mblk_t *mp)
8851 8852 {
8852 8853 struct iocblk *iocp = (struct iocblk *)mp->b_rptr;
8853 8854 mblk_t *mp1 = mp->b_cont;
8854 8855 mod_ioc_prop_t *pioc;
8855 8856 mod_prop_info_t *ptbl = NULL, *pinfo = NULL;
8856 8857 ip_stack_t *ipst;
8857 8858 icmp_stack_t *is;
8858 8859 tcp_stack_t *tcps;
8859 8860 sctp_stack_t *sctps;
8861 + dccp_stack_t *dccps;
8860 8862 udp_stack_t *us;
8861 8863 netstack_t *stack;
8862 8864 void *cbarg;
8863 8865 cred_t *cr;
8864 8866 boolean_t set;
8865 8867 int err;
8866 8868
8867 8869 ASSERT(q->q_next == NULL);
8868 8870 ASSERT(CONN_Q(q));
8869 8871
8870 8872 if (!getset_ioctl_checks(mp)) {
8871 8873 miocnak(q, mp, 0, EINVAL);
8872 8874 return;
8873 8875 }
8874 8876 ipst = CONNQ_TO_IPST(q);
8875 8877 stack = ipst->ips_netstack;
8876 8878 pioc = (mod_ioc_prop_t *)mp1->b_rptr;
8877 8879
8878 8880 switch (pioc->mpr_proto) {
8879 8881 case MOD_PROTO_IP:
8880 8882 case MOD_PROTO_IPV4:
8881 8883 case MOD_PROTO_IPV6:
8882 8884 ptbl = ipst->ips_propinfo_tbl;
8883 8885 cbarg = ipst;
8884 8886 break;
8885 8887 case MOD_PROTO_RAWIP:
8886 8888 is = stack->netstack_icmp;
8887 8889 ptbl = is->is_propinfo_tbl;
8888 8890 cbarg = is;
8889 8891 break;
8890 8892 case MOD_PROTO_TCP:
8891 8893 tcps = stack->netstack_tcp;
8892 8894 ptbl = tcps->tcps_propinfo_tbl;
8893 8895 cbarg = tcps;
8894 8896 break;
|
↓ open down ↓ |
25 lines elided |
↑ open up ↑ |
8895 8897 case MOD_PROTO_UDP:
8896 8898 us = stack->netstack_udp;
8897 8899 ptbl = us->us_propinfo_tbl;
8898 8900 cbarg = us;
8899 8901 break;
8900 8902 case MOD_PROTO_SCTP:
8901 8903 sctps = stack->netstack_sctp;
8902 8904 ptbl = sctps->sctps_propinfo_tbl;
8903 8905 cbarg = sctps;
8904 8906 break;
8907 + case MOD_PROTO_DCCP:
8908 + dccps = stack->netstack_dccp;
8909 + ptbl = dccps->dccps_propinfo_tbl;
8910 + cbarg = dccps;
8905 8911 default:
8906 8912 miocnak(q, mp, 0, EINVAL);
8907 8913 return;
8908 8914 }
8909 8915
8910 8916 /* search for given property in respective protocol propinfo table */
8911 8917 for (pinfo = ptbl; pinfo->mpi_name != NULL; pinfo++) {
8912 8918 if (strcmp(pinfo->mpi_name, pioc->mpr_name) == 0 &&
8913 8919 pinfo->mpi_proto == pioc->mpr_proto)
8914 8920 break;
8915 8921 }
8916 8922 if (pinfo->mpi_name == NULL) {
8917 8923 miocnak(q, mp, 0, ENOENT);
8918 8924 return;
8919 8925 }
8920 8926
8921 8927 set = (iocp->ioc_cmd == SIOCSETPROP) ? B_TRUE : B_FALSE;
8922 8928 if (set && pinfo->mpi_setf != NULL) {
8923 8929 cr = msg_getcred(mp, NULL);
8924 8930 if (cr == NULL)
8925 8931 cr = iocp->ioc_cr;
8926 8932 err = pinfo->mpi_setf(cbarg, cr, pinfo, pioc->mpr_ifname,
8927 8933 pioc->mpr_val, pioc->mpr_flags);
8928 8934 } else if (!set && pinfo->mpi_getf != NULL) {
8929 8935 err = pinfo->mpi_getf(cbarg, pinfo, pioc->mpr_ifname,
8930 8936 pioc->mpr_val, pioc->mpr_valsize, pioc->mpr_flags);
8931 8937 } else {
8932 8938 err = EPERM;
8933 8939 }
8934 8940
8935 8941 if (err != 0) {
8936 8942 miocnak(q, mp, 0, err);
8937 8943 } else {
8938 8944 if (set)
8939 8945 miocack(q, mp, 0, 0);
8940 8946 else /* For get, we need to return back the data */
8941 8947 miocack(q, mp, iocp->ioc_count, 0);
8942 8948 }
8943 8949 }
8944 8950
8945 8951 /*
8946 8952 * process the legacy ND_GET, ND_SET ioctl just for {ip|ip6}_forwarding
8947 8953 * as several routing daemons have unfortunately used this 'unpublished'
8948 8954 * but well-known ioctls.
8949 8955 */
8950 8956 /* ARGSUSED */
8951 8957 static void
8952 8958 ip_process_legacy_nddprop(queue_t *q, mblk_t *mp)
8953 8959 {
8954 8960 struct iocblk *iocp = (struct iocblk *)mp->b_rptr;
8955 8961 mblk_t *mp1 = mp->b_cont;
8956 8962 char *pname, *pval, *buf;
8957 8963 uint_t bufsize, proto;
8958 8964 mod_prop_info_t *ptbl = NULL, *pinfo = NULL;
8959 8965 ip_stack_t *ipst;
8960 8966 int err = 0;
8961 8967
8962 8968 ASSERT(CONN_Q(q));
8963 8969 ipst = CONNQ_TO_IPST(q);
8964 8970
8965 8971 if (iocp->ioc_count == 0 || mp1 == NULL) {
8966 8972 miocnak(q, mp, 0, EINVAL);
8967 8973 return;
8968 8974 }
8969 8975
8970 8976 mp1->b_datap->db_lim[-1] = '\0'; /* Force null termination */
8971 8977 pval = buf = pname = (char *)mp1->b_rptr;
8972 8978 bufsize = MBLKL(mp1);
8973 8979
8974 8980 if (strcmp(pname, "ip_forwarding") == 0) {
8975 8981 pname = "forwarding";
8976 8982 proto = MOD_PROTO_IPV4;
8977 8983 } else if (strcmp(pname, "ip6_forwarding") == 0) {
8978 8984 pname = "forwarding";
8979 8985 proto = MOD_PROTO_IPV6;
8980 8986 } else {
8981 8987 miocnak(q, mp, 0, EINVAL);
8982 8988 return;
8983 8989 }
8984 8990
8985 8991 ptbl = ipst->ips_propinfo_tbl;
8986 8992 for (pinfo = ptbl; pinfo->mpi_name != NULL; pinfo++) {
8987 8993 if (strcmp(pinfo->mpi_name, pname) == 0 &&
8988 8994 pinfo->mpi_proto == proto)
8989 8995 break;
8990 8996 }
8991 8997
8992 8998 ASSERT(pinfo->mpi_name != NULL);
8993 8999
8994 9000 switch (iocp->ioc_cmd) {
8995 9001 case ND_GET:
8996 9002 if ((err = pinfo->mpi_getf(ipst, pinfo, NULL, buf, bufsize,
8997 9003 0)) == 0) {
8998 9004 miocack(q, mp, iocp->ioc_count, 0);
8999 9005 return;
9000 9006 }
9001 9007 break;
9002 9008 case ND_SET:
9003 9009 /*
9004 9010 * buffer will have property name and value in the following
9005 9011 * format,
9006 9012 * <property name>'\0'<property value>'\0', extract them;
9007 9013 */
9008 9014 while (*pval++)
9009 9015 noop;
9010 9016
9011 9017 if (!*pval || pval >= (char *)mp1->b_wptr) {
9012 9018 err = EINVAL;
9013 9019 } else if ((err = pinfo->mpi_setf(ipst, NULL, pinfo, NULL,
9014 9020 pval, 0)) == 0) {
9015 9021 miocack(q, mp, 0, 0);
9016 9022 return;
9017 9023 }
9018 9024 break;
9019 9025 default:
9020 9026 err = EINVAL;
9021 9027 break;
9022 9028 }
9023 9029 miocnak(q, mp, 0, err);
9024 9030 }
9025 9031
9026 9032 /*
9027 9033 * Wrapper function for resuming deferred ioctl processing
9028 9034 * Used for SIOCGDSTINFO, SIOCGIP6ADDRPOLICY, SIOCGMSFILTER,
9029 9035 * SIOCSMSFILTER, SIOCGIPMSFILTER, and SIOCSIPMSFILTER currently.
9030 9036 */
9031 9037 /* ARGSUSED */
9032 9038 void
9033 9039 ip_sioctl_copyin_resume(ipsq_t *dummy_ipsq, queue_t *q, mblk_t *mp,
9034 9040 void *dummy_arg)
9035 9041 {
9036 9042 ip_sioctl_copyin_setup(q, mp);
9037 9043 }
9038 9044
9039 9045 /*
9040 9046 * ip_sioctl_copyin_setup is called by ip_wput_nondata with any M_IOCTL message
9041 9047 * that arrives. Most of the IOCTLs are "socket" IOCTLs which we handle
9042 9048 * in either I_STR or TRANSPARENT form, using the mi_copy facility.
9043 9049 * We establish here the size of the block to be copied in. mi_copyin
9044 9050 * arranges for this to happen, an processing continues in ip_wput_nondata with
9045 9051 * an M_IOCDATA message.
9046 9052 */
9047 9053 void
9048 9054 ip_sioctl_copyin_setup(queue_t *q, mblk_t *mp)
9049 9055 {
9050 9056 int copyin_size;
9051 9057 struct iocblk *iocp = (struct iocblk *)mp->b_rptr;
9052 9058 ip_ioctl_cmd_t *ipip;
9053 9059 cred_t *cr;
9054 9060 ip_stack_t *ipst;
9055 9061
9056 9062 if (CONN_Q(q))
9057 9063 ipst = CONNQ_TO_IPST(q);
9058 9064 else
9059 9065 ipst = ILLQ_TO_IPST(q);
9060 9066
9061 9067 ipip = ip_sioctl_lookup(iocp->ioc_cmd);
9062 9068 if (ipip == NULL) {
9063 9069 /*
9064 9070 * The ioctl is not one we understand or own.
9065 9071 * Pass it along to be processed down stream,
9066 9072 * if this is a module instance of IP, else nak
9067 9073 * the ioctl.
9068 9074 */
9069 9075 if (q->q_next == NULL) {
9070 9076 goto nak;
9071 9077 } else {
9072 9078 putnext(q, mp);
9073 9079 return;
9074 9080 }
9075 9081 }
9076 9082
9077 9083 /*
9078 9084 * If this is deferred, then we will do all the checks when we
9079 9085 * come back.
9080 9086 */
9081 9087 if ((iocp->ioc_cmd == SIOCGDSTINFO ||
9082 9088 iocp->ioc_cmd == SIOCGIP6ADDRPOLICY) && !ip6_asp_can_lookup(ipst)) {
9083 9089 ip6_asp_pending_op(q, mp, ip_sioctl_copyin_resume);
9084 9090 return;
9085 9091 }
9086 9092
9087 9093 /*
9088 9094 * Only allow a very small subset of IP ioctls on this stream if
9089 9095 * IP is a module and not a driver. Allowing ioctls to be processed
9090 9096 * in this case may cause assert failures or data corruption.
9091 9097 * Typically G[L]IFFLAGS, SLIFNAME/IF_UNITSEL are the only few
9092 9098 * ioctls allowed on an IP module stream, after which this stream
9093 9099 * normally becomes a multiplexor (at which time the stream head
9094 9100 * will fail all ioctls).
9095 9101 */
9096 9102 if ((q->q_next != NULL) && !(ipip->ipi_flags & IPI_MODOK)) {
9097 9103 goto nak;
9098 9104 }
9099 9105
9100 9106 /* Make sure we have ioctl data to process. */
9101 9107 if (mp->b_cont == NULL && !(ipip->ipi_flags & IPI_NULL_BCONT))
9102 9108 goto nak;
9103 9109
9104 9110 /*
9105 9111 * Prefer dblk credential over ioctl credential; some synthesized
9106 9112 * ioctls have kcred set because there's no way to crhold()
9107 9113 * a credential in some contexts. (ioc_cr is not crfree() by
9108 9114 * the framework; the caller of ioctl needs to hold the reference
9109 9115 * for the duration of the call).
9110 9116 */
9111 9117 cr = msg_getcred(mp, NULL);
9112 9118 if (cr == NULL)
9113 9119 cr = iocp->ioc_cr;
9114 9120
9115 9121 /* Make sure normal users don't send down privileged ioctls */
9116 9122 if ((ipip->ipi_flags & IPI_PRIV) &&
9117 9123 (cr != NULL) && secpolicy_ip_config(cr, B_TRUE) != 0) {
9118 9124 /* We checked the privilege earlier but log it here */
9119 9125 miocnak(q, mp, 0, secpolicy_ip_config(cr, B_FALSE));
9120 9126 return;
9121 9127 }
9122 9128
9123 9129 /*
9124 9130 * The ioctl command tables can only encode fixed length
9125 9131 * ioctl data. If the length is variable, the table will
9126 9132 * encode the length as zero. Such special cases are handled
9127 9133 * below in the switch.
9128 9134 */
9129 9135 if (ipip->ipi_copyin_size != 0) {
9130 9136 mi_copyin(q, mp, NULL, ipip->ipi_copyin_size);
9131 9137 return;
9132 9138 }
9133 9139
9134 9140 switch (iocp->ioc_cmd) {
9135 9141 case O_SIOCGIFCONF:
9136 9142 case SIOCGIFCONF:
9137 9143 /*
9138 9144 * This IOCTL is hilarious. See comments in
9139 9145 * ip_sioctl_get_ifconf for the story.
9140 9146 */
9141 9147 if (iocp->ioc_count == TRANSPARENT)
9142 9148 copyin_size = SIZEOF_STRUCT(ifconf,
9143 9149 iocp->ioc_flag);
9144 9150 else
9145 9151 copyin_size = iocp->ioc_count;
9146 9152 mi_copyin(q, mp, NULL, copyin_size);
9147 9153 return;
9148 9154
9149 9155 case O_SIOCGLIFCONF:
9150 9156 case SIOCGLIFCONF:
9151 9157 copyin_size = SIZEOF_STRUCT(lifconf, iocp->ioc_flag);
9152 9158 mi_copyin(q, mp, NULL, copyin_size);
9153 9159 return;
9154 9160
9155 9161 case SIOCGLIFSRCOF:
9156 9162 copyin_size = SIZEOF_STRUCT(lifsrcof, iocp->ioc_flag);
9157 9163 mi_copyin(q, mp, NULL, copyin_size);
9158 9164 return;
9159 9165
9160 9166 case SIOCGIP6ADDRPOLICY:
9161 9167 ip_sioctl_ip6addrpolicy(q, mp);
9162 9168 ip6_asp_table_refrele(ipst);
9163 9169 return;
9164 9170
9165 9171 case SIOCSIP6ADDRPOLICY:
9166 9172 ip_sioctl_ip6addrpolicy(q, mp);
9167 9173 return;
9168 9174
9169 9175 case SIOCGDSTINFO:
9170 9176 ip_sioctl_dstinfo(q, mp);
9171 9177 ip6_asp_table_refrele(ipst);
9172 9178 return;
9173 9179
9174 9180 case ND_SET:
9175 9181 case ND_GET:
9176 9182 ip_process_legacy_nddprop(q, mp);
9177 9183 return;
9178 9184
9179 9185 case SIOCSETPROP:
9180 9186 case SIOCGETPROP:
9181 9187 ip_sioctl_getsetprop(q, mp);
9182 9188 return;
9183 9189
9184 9190 case I_PLINK:
9185 9191 case I_PUNLINK:
9186 9192 case I_LINK:
9187 9193 case I_UNLINK:
9188 9194 /*
9189 9195 * We treat non-persistent link similarly as the persistent
9190 9196 * link case, in terms of plumbing/unplumbing, as well as
9191 9197 * dynamic re-plumbing events indicator. See comments
9192 9198 * in ip_sioctl_plink() for more.
9193 9199 *
9194 9200 * Request can be enqueued in the 'ipsq' while waiting
9195 9201 * to become exclusive. So bump up the conn ref.
9196 9202 */
9197 9203 if (CONN_Q(q)) {
9198 9204 CONN_INC_REF(Q_TO_CONN(q));
9199 9205 CONN_INC_IOCTLREF(Q_TO_CONN(q))
9200 9206 }
9201 9207 ip_sioctl_plink(NULL, q, mp, NULL);
9202 9208 return;
9203 9209
9204 9210 case IP_IOCTL:
9205 9211 ip_wput_ioctl(q, mp);
9206 9212 return;
9207 9213
9208 9214 case SIOCILB:
9209 9215 /* The ioctl length varies depending on the ILB command. */
9210 9216 copyin_size = iocp->ioc_count;
9211 9217 if (copyin_size < sizeof (ilb_cmd_t))
9212 9218 goto nak;
9213 9219 mi_copyin(q, mp, NULL, copyin_size);
9214 9220 return;
9215 9221
9216 9222 default:
9217 9223 cmn_err(CE_PANIC, "should not happen ");
9218 9224 }
9219 9225 nak:
9220 9226 if (mp->b_cont != NULL) {
9221 9227 freemsg(mp->b_cont);
9222 9228 mp->b_cont = NULL;
9223 9229 }
9224 9230 iocp->ioc_error = EINVAL;
9225 9231 mp->b_datap->db_type = M_IOCNAK;
9226 9232 iocp->ioc_count = 0;
9227 9233 qreply(q, mp);
9228 9234 }
9229 9235
9230 9236 static void
9231 9237 ip_sioctl_garp_reply(mblk_t *mp, ill_t *ill, void *hwaddr, int flags)
9232 9238 {
9233 9239 struct arpreq *ar;
9234 9240 struct xarpreq *xar;
9235 9241 mblk_t *tmp;
9236 9242 struct iocblk *iocp;
9237 9243 int x_arp_ioctl = B_FALSE;
9238 9244 int *flagsp;
9239 9245 char *storage = NULL;
9240 9246
9241 9247 ASSERT(ill != NULL);
9242 9248
9243 9249 iocp = (struct iocblk *)mp->b_rptr;
9244 9250 ASSERT(iocp->ioc_cmd == SIOCGXARP || iocp->ioc_cmd == SIOCGARP);
9245 9251
9246 9252 tmp = (mp->b_cont)->b_cont; /* xarpreq/arpreq */
9247 9253 if ((iocp->ioc_cmd == SIOCGXARP) ||
9248 9254 (iocp->ioc_cmd == SIOCSXARP)) {
9249 9255 x_arp_ioctl = B_TRUE;
9250 9256 xar = (struct xarpreq *)tmp->b_rptr;
9251 9257 flagsp = &xar->xarp_flags;
9252 9258 storage = xar->xarp_ha.sdl_data;
9253 9259 } else {
9254 9260 ar = (struct arpreq *)tmp->b_rptr;
9255 9261 flagsp = &ar->arp_flags;
9256 9262 storage = ar->arp_ha.sa_data;
9257 9263 }
9258 9264
9259 9265 /*
9260 9266 * We're done if this is not an SIOCG{X}ARP
9261 9267 */
9262 9268 if (x_arp_ioctl) {
9263 9269 storage += ill_xarp_info(&xar->xarp_ha, ill);
9264 9270 if ((ill->ill_phys_addr_length + ill->ill_name_length) >
9265 9271 sizeof (xar->xarp_ha.sdl_data)) {
9266 9272 iocp->ioc_error = EINVAL;
9267 9273 return;
9268 9274 }
9269 9275 }
9270 9276 *flagsp = ATF_INUSE;
9271 9277 /*
9272 9278 * If /sbin/arp told us we are the authority using the "permanent"
9273 9279 * flag, or if this is one of my addresses print "permanent"
9274 9280 * in the /sbin/arp output.
9275 9281 */
9276 9282 if ((flags & NCE_F_MYADDR) || (flags & NCE_F_AUTHORITY))
9277 9283 *flagsp |= ATF_AUTHORITY;
9278 9284 if (flags & NCE_F_NONUD)
9279 9285 *flagsp |= ATF_PERM; /* not subject to aging */
9280 9286 if (flags & NCE_F_PUBLISH)
9281 9287 *flagsp |= ATF_PUBL;
9282 9288 if (hwaddr != NULL) {
9283 9289 *flagsp |= ATF_COM;
9284 9290 bcopy((char *)hwaddr, storage, ill->ill_phys_addr_length);
9285 9291 }
9286 9292 }
9287 9293
9288 9294 /*
9289 9295 * Create a new logical interface. If ipif_id is zero (i.e. not a logical
9290 9296 * interface) create the next available logical interface for this
9291 9297 * physical interface.
9292 9298 * If ipif is NULL (i.e. the lookup didn't find one) attempt to create an
9293 9299 * ipif with the specified name.
9294 9300 *
9295 9301 * If the address family is not AF_UNSPEC then set the address as well.
9296 9302 *
9297 9303 * If ip_sioctl_addr returns EINPROGRESS then the ioctl (the copyout)
9298 9304 * is completed when the DL_BIND_ACK arrive in ip_rput_dlpi_writer.
9299 9305 *
9300 9306 * Executed as a writer on the ill.
9301 9307 * So no lock is needed to traverse the ipif chain, or examine the
9302 9308 * phyint flags.
9303 9309 */
9304 9310 /* ARGSUSED */
9305 9311 int
9306 9312 ip_sioctl_addif(ipif_t *dummy_ipif, sin_t *dummy_sin, queue_t *q, mblk_t *mp,
9307 9313 ip_ioctl_cmd_t *dummy_ipip, void *dummy_ifreq)
9308 9314 {
9309 9315 mblk_t *mp1;
9310 9316 struct lifreq *lifr;
9311 9317 boolean_t isv6;
9312 9318 boolean_t exists;
9313 9319 char *name;
9314 9320 char *endp;
9315 9321 char *cp;
9316 9322 int namelen;
9317 9323 ipif_t *ipif;
9318 9324 long id;
9319 9325 ipsq_t *ipsq;
9320 9326 ill_t *ill;
9321 9327 sin_t *sin;
9322 9328 int err = 0;
9323 9329 boolean_t found_sep = B_FALSE;
9324 9330 conn_t *connp;
9325 9331 zoneid_t zoneid;
9326 9332 ip_stack_t *ipst = CONNQ_TO_IPST(q);
9327 9333
9328 9334 ASSERT(q->q_next == NULL);
9329 9335 ip1dbg(("ip_sioctl_addif\n"));
9330 9336 /* Existence of mp1 has been checked in ip_wput_nondata */
9331 9337 mp1 = mp->b_cont->b_cont;
9332 9338 /*
9333 9339 * Null terminate the string to protect against buffer
9334 9340 * overrun. String was generated by user code and may not
9335 9341 * be trusted.
9336 9342 */
9337 9343 lifr = (struct lifreq *)mp1->b_rptr;
9338 9344 lifr->lifr_name[LIFNAMSIZ - 1] = '\0';
9339 9345 name = lifr->lifr_name;
9340 9346 ASSERT(CONN_Q(q));
9341 9347 connp = Q_TO_CONN(q);
9342 9348 isv6 = (connp->conn_family == AF_INET6);
9343 9349 zoneid = connp->conn_zoneid;
9344 9350 namelen = mi_strlen(name);
9345 9351 if (namelen == 0)
9346 9352 return (EINVAL);
9347 9353
9348 9354 exists = B_FALSE;
9349 9355 if ((namelen + 1 == sizeof (ipif_loopback_name)) &&
9350 9356 (mi_strcmp(name, ipif_loopback_name) == 0)) {
9351 9357 /*
9352 9358 * Allow creating lo0 using SIOCLIFADDIF.
9353 9359 * can't be any other writer thread. So can pass null below
9354 9360 * for the last 4 args to ipif_lookup_name.
9355 9361 */
9356 9362 ipif = ipif_lookup_on_name(lifr->lifr_name, namelen, B_TRUE,
9357 9363 &exists, isv6, zoneid, ipst);
9358 9364 /* Prevent any further action */
9359 9365 if (ipif == NULL) {
9360 9366 return (ENOBUFS);
9361 9367 } else if (!exists) {
9362 9368 /* We created the ipif now and as writer */
9363 9369 ipif_refrele(ipif);
9364 9370 return (0);
9365 9371 } else {
9366 9372 ill = ipif->ipif_ill;
9367 9373 ill_refhold(ill);
9368 9374 ipif_refrele(ipif);
9369 9375 }
9370 9376 } else {
9371 9377 /* Look for a colon in the name. */
9372 9378 endp = &name[namelen];
9373 9379 for (cp = endp; --cp > name; ) {
9374 9380 if (*cp == IPIF_SEPARATOR_CHAR) {
9375 9381 found_sep = B_TRUE;
9376 9382 /*
9377 9383 * Reject any non-decimal aliases for plumbing
9378 9384 * of logical interfaces. Aliases with leading
9379 9385 * zeroes are also rejected as they introduce
9380 9386 * ambiguity in the naming of the interfaces.
9381 9387 * Comparing with "0" takes care of all such
9382 9388 * cases.
9383 9389 */
9384 9390 if ((strncmp("0", cp+1, 1)) == 0)
9385 9391 return (EINVAL);
9386 9392
9387 9393 if (ddi_strtol(cp+1, &endp, 10, &id) != 0 ||
9388 9394 id <= 0 || *endp != '\0') {
9389 9395 return (EINVAL);
9390 9396 }
9391 9397 *cp = '\0';
9392 9398 break;
9393 9399 }
9394 9400 }
9395 9401 ill = ill_lookup_on_name(name, B_FALSE, isv6, NULL, ipst);
9396 9402 if (found_sep)
9397 9403 *cp = IPIF_SEPARATOR_CHAR;
9398 9404 if (ill == NULL)
9399 9405 return (ENXIO);
9400 9406 }
9401 9407
9402 9408 ipsq = ipsq_try_enter(NULL, ill, q, mp, ip_process_ioctl, NEW_OP,
9403 9409 B_TRUE);
9404 9410
9405 9411 /*
9406 9412 * Release the refhold due to the lookup, now that we are excl
9407 9413 * or we are just returning
9408 9414 */
9409 9415 ill_refrele(ill);
9410 9416
9411 9417 if (ipsq == NULL)
9412 9418 return (EINPROGRESS);
9413 9419
9414 9420 /* We are now exclusive on the IPSQ */
9415 9421 ASSERT(IAM_WRITER_ILL(ill));
9416 9422
9417 9423 if (found_sep) {
9418 9424 /* Now see if there is an IPIF with this unit number. */
9419 9425 for (ipif = ill->ill_ipif; ipif != NULL;
9420 9426 ipif = ipif->ipif_next) {
9421 9427 if (ipif->ipif_id == id) {
9422 9428 err = EEXIST;
9423 9429 goto done;
9424 9430 }
9425 9431 }
9426 9432 }
9427 9433
9428 9434 /*
9429 9435 * We use IRE_LOCAL for lo0:1 etc. for "receive only" use
9430 9436 * of lo0. Plumbing for lo0:0 happens in ipif_lookup_on_name()
9431 9437 * instead.
9432 9438 */
9433 9439 if ((ipif = ipif_allocate(ill, found_sep ? id : -1, IRE_LOCAL,
9434 9440 B_TRUE, B_TRUE, &err)) == NULL) {
9435 9441 goto done;
9436 9442 }
9437 9443
9438 9444 /* Return created name with ioctl */
9439 9445 (void) sprintf(lifr->lifr_name, "%s%c%d", ill->ill_name,
9440 9446 IPIF_SEPARATOR_CHAR, ipif->ipif_id);
9441 9447 ip1dbg(("created %s\n", lifr->lifr_name));
9442 9448
9443 9449 /* Set address */
9444 9450 sin = (sin_t *)&lifr->lifr_addr;
9445 9451 if (sin->sin_family != AF_UNSPEC) {
9446 9452 err = ip_sioctl_addr(ipif, sin, q, mp,
9447 9453 &ip_ndx_ioctl_table[SIOCLIFADDR_NDX], lifr);
9448 9454 }
9449 9455
9450 9456 done:
9451 9457 ipsq_exit(ipsq);
9452 9458 return (err);
9453 9459 }
9454 9460
9455 9461 /*
9456 9462 * Remove an existing logical interface. If ipif_id is zero (i.e. not a logical
9457 9463 * interface) delete it based on the IP address (on this physical interface).
9458 9464 * Otherwise delete it based on the ipif_id.
9459 9465 * Also, special handling to allow a removeif of lo0.
9460 9466 */
9461 9467 /* ARGSUSED */
9462 9468 int
9463 9469 ip_sioctl_removeif(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
9464 9470 ip_ioctl_cmd_t *ipip, void *dummy_if_req)
9465 9471 {
9466 9472 conn_t *connp;
9467 9473 ill_t *ill = ipif->ipif_ill;
9468 9474 boolean_t success;
9469 9475 ip_stack_t *ipst;
9470 9476
9471 9477 ipst = CONNQ_TO_IPST(q);
9472 9478
9473 9479 ASSERT(q->q_next == NULL);
9474 9480 ip1dbg(("ip_sioctl_remove_if(%s:%u %p)\n",
9475 9481 ill->ill_name, ipif->ipif_id, (void *)ipif));
9476 9482 ASSERT(IAM_WRITER_IPIF(ipif));
9477 9483
9478 9484 connp = Q_TO_CONN(q);
9479 9485 /*
9480 9486 * Special case for unplumbing lo0 (the loopback physical interface).
9481 9487 * If unplumbing lo0, the incoming address structure has been
9482 9488 * initialized to all zeros. When unplumbing lo0, all its logical
9483 9489 * interfaces must be removed too.
9484 9490 *
9485 9491 * Note that this interface may be called to remove a specific
9486 9492 * loopback logical interface (eg, lo0:1). But in that case
9487 9493 * ipif->ipif_id != 0 so that the code path for that case is the
9488 9494 * same as any other interface (meaning it skips the code directly
9489 9495 * below).
9490 9496 */
9491 9497 if (ipif->ipif_id == 0 && ill->ill_net_type == IRE_LOOPBACK) {
9492 9498 if (sin->sin_family == AF_UNSPEC &&
9493 9499 (IN6_IS_ADDR_UNSPECIFIED(&((sin6_t *)sin)->sin6_addr))) {
9494 9500 /*
9495 9501 * Mark it condemned. No new ref. will be made to ill.
9496 9502 */
9497 9503 mutex_enter(&ill->ill_lock);
9498 9504 ill->ill_state_flags |= ILL_CONDEMNED;
9499 9505 for (ipif = ill->ill_ipif; ipif != NULL;
9500 9506 ipif = ipif->ipif_next) {
9501 9507 ipif->ipif_state_flags |= IPIF_CONDEMNED;
9502 9508 }
9503 9509 mutex_exit(&ill->ill_lock);
9504 9510
9505 9511 ipif = ill->ill_ipif;
9506 9512 /* unplumb the loopback interface */
9507 9513 ill_delete(ill);
9508 9514 mutex_enter(&connp->conn_lock);
9509 9515 mutex_enter(&ill->ill_lock);
9510 9516
9511 9517 /* Are any references to this ill active */
9512 9518 if (ill_is_freeable(ill)) {
9513 9519 mutex_exit(&ill->ill_lock);
9514 9520 mutex_exit(&connp->conn_lock);
9515 9521 ill_delete_tail(ill);
9516 9522 mi_free(ill);
9517 9523 return (0);
9518 9524 }
9519 9525 success = ipsq_pending_mp_add(connp, ipif,
9520 9526 CONNP_TO_WQ(connp), mp, ILL_FREE);
9521 9527 mutex_exit(&connp->conn_lock);
9522 9528 mutex_exit(&ill->ill_lock);
9523 9529 if (success)
9524 9530 return (EINPROGRESS);
9525 9531 else
9526 9532 return (EINTR);
9527 9533 }
9528 9534 }
9529 9535
9530 9536 if (ipif->ipif_id == 0) {
9531 9537 ipsq_t *ipsq;
9532 9538
9533 9539 /* Find based on address */
9534 9540 if (ipif->ipif_isv6) {
9535 9541 sin6_t *sin6;
9536 9542
9537 9543 if (sin->sin_family != AF_INET6)
9538 9544 return (EAFNOSUPPORT);
9539 9545
9540 9546 sin6 = (sin6_t *)sin;
9541 9547 /* We are a writer, so we should be able to lookup */
9542 9548 ipif = ipif_lookup_addr_exact_v6(&sin6->sin6_addr, ill,
9543 9549 ipst);
9544 9550 } else {
9545 9551 if (sin->sin_family != AF_INET)
9546 9552 return (EAFNOSUPPORT);
9547 9553
9548 9554 /* We are a writer, so we should be able to lookup */
9549 9555 ipif = ipif_lookup_addr_exact(sin->sin_addr.s_addr, ill,
9550 9556 ipst);
9551 9557 }
9552 9558 if (ipif == NULL) {
9553 9559 return (EADDRNOTAVAIL);
9554 9560 }
9555 9561
9556 9562 /*
9557 9563 * It is possible for a user to send an SIOCLIFREMOVEIF with
9558 9564 * lifr_name of the physical interface but with an ip address
9559 9565 * lifr_addr of a logical interface plumbed over it.
9560 9566 * So update ipx_current_ipif now that ipif points to the
9561 9567 * correct one.
9562 9568 */
9563 9569 ipsq = ipif->ipif_ill->ill_phyint->phyint_ipsq;
9564 9570 ipsq->ipsq_xop->ipx_current_ipif = ipif;
9565 9571
9566 9572 /* This is a writer */
9567 9573 ipif_refrele(ipif);
9568 9574 }
9569 9575
9570 9576 /*
9571 9577 * Can not delete instance zero since it is tied to the ill.
9572 9578 */
9573 9579 if (ipif->ipif_id == 0)
9574 9580 return (EBUSY);
9575 9581
9576 9582 mutex_enter(&ill->ill_lock);
9577 9583 ipif->ipif_state_flags |= IPIF_CONDEMNED;
9578 9584 mutex_exit(&ill->ill_lock);
9579 9585
9580 9586 ipif_free(ipif);
9581 9587
9582 9588 mutex_enter(&connp->conn_lock);
9583 9589 mutex_enter(&ill->ill_lock);
9584 9590
9585 9591 /* Are any references to this ipif active */
9586 9592 if (ipif_is_freeable(ipif)) {
9587 9593 mutex_exit(&ill->ill_lock);
9588 9594 mutex_exit(&connp->conn_lock);
9589 9595 ipif_non_duplicate(ipif);
9590 9596 (void) ipif_down_tail(ipif);
9591 9597 ipif_free_tail(ipif); /* frees ipif */
9592 9598 return (0);
9593 9599 }
9594 9600 success = ipsq_pending_mp_add(connp, ipif, CONNP_TO_WQ(connp), mp,
9595 9601 IPIF_FREE);
9596 9602 mutex_exit(&ill->ill_lock);
9597 9603 mutex_exit(&connp->conn_lock);
9598 9604 if (success)
9599 9605 return (EINPROGRESS);
9600 9606 else
9601 9607 return (EINTR);
9602 9608 }
9603 9609
9604 9610 /*
9605 9611 * Restart the removeif ioctl. The refcnt has gone down to 0.
9606 9612 * The ipif is already condemned. So can't find it thru lookups.
9607 9613 */
9608 9614 /* ARGSUSED */
9609 9615 int
9610 9616 ip_sioctl_removeif_restart(ipif_t *ipif, sin_t *dummy_sin, queue_t *q,
9611 9617 mblk_t *mp, ip_ioctl_cmd_t *ipip, void *dummy_if_req)
9612 9618 {
9613 9619 ill_t *ill = ipif->ipif_ill;
9614 9620
9615 9621 ASSERT(IAM_WRITER_IPIF(ipif));
9616 9622 ASSERT(ipif->ipif_state_flags & IPIF_CONDEMNED);
9617 9623
9618 9624 ip1dbg(("ip_sioctl_removeif_restart(%s:%u %p)\n",
9619 9625 ill->ill_name, ipif->ipif_id, (void *)ipif));
9620 9626
9621 9627 if (ipif->ipif_id == 0 && ill->ill_net_type == IRE_LOOPBACK) {
9622 9628 ASSERT(ill->ill_state_flags & ILL_CONDEMNED);
9623 9629 ill_delete_tail(ill);
9624 9630 mi_free(ill);
9625 9631 return (0);
9626 9632 }
9627 9633
9628 9634 ipif_non_duplicate(ipif);
9629 9635 (void) ipif_down_tail(ipif);
9630 9636 ipif_free_tail(ipif);
9631 9637
9632 9638 return (0);
9633 9639 }
9634 9640
9635 9641 /*
9636 9642 * Set the local interface address using the given prefix and ill_token.
9637 9643 */
9638 9644 /* ARGSUSED */
9639 9645 int
9640 9646 ip_sioctl_prefix(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
9641 9647 ip_ioctl_cmd_t *dummy_ipip, void *dummy_ifreq)
9642 9648 {
9643 9649 int err;
9644 9650 in6_addr_t v6addr;
9645 9651 sin6_t *sin6;
9646 9652 ill_t *ill;
9647 9653 int i;
9648 9654
9649 9655 ip1dbg(("ip_sioctl_prefix(%s:%u %p)\n",
9650 9656 ipif->ipif_ill->ill_name, ipif->ipif_id, (void *)ipif));
9651 9657
9652 9658 ASSERT(IAM_WRITER_IPIF(ipif));
9653 9659
9654 9660 if (!ipif->ipif_isv6)
9655 9661 return (EINVAL);
9656 9662
9657 9663 if (sin->sin_family != AF_INET6)
9658 9664 return (EAFNOSUPPORT);
9659 9665
9660 9666 sin6 = (sin6_t *)sin;
9661 9667 v6addr = sin6->sin6_addr;
9662 9668 ill = ipif->ipif_ill;
9663 9669
9664 9670 if (IN6_IS_ADDR_UNSPECIFIED(&v6addr) ||
9665 9671 IN6_IS_ADDR_UNSPECIFIED(&ill->ill_token))
9666 9672 return (EADDRNOTAVAIL);
9667 9673
9668 9674 for (i = 0; i < 4; i++)
9669 9675 sin6->sin6_addr.s6_addr32[i] |= ill->ill_token.s6_addr32[i];
9670 9676
9671 9677 err = ip_sioctl_addr(ipif, sin, q, mp,
9672 9678 &ip_ndx_ioctl_table[SIOCLIFADDR_NDX], dummy_ifreq);
9673 9679 return (err);
9674 9680 }
9675 9681
9676 9682 /*
9677 9683 * Restart entry point to restart the address set operation after the
9678 9684 * refcounts have dropped to zero.
9679 9685 */
9680 9686 /* ARGSUSED */
9681 9687 int
9682 9688 ip_sioctl_prefix_restart(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
9683 9689 ip_ioctl_cmd_t *ipip, void *ifreq)
9684 9690 {
9685 9691 ip1dbg(("ip_sioctl_prefix_restart(%s:%u %p)\n",
9686 9692 ipif->ipif_ill->ill_name, ipif->ipif_id, (void *)ipif));
9687 9693 return (ip_sioctl_addr_restart(ipif, sin, q, mp, ipip, ifreq));
9688 9694 }
9689 9695
9690 9696 /*
9691 9697 * Set the local interface address.
9692 9698 * Allow an address of all zero when the interface is down.
9693 9699 */
9694 9700 /* ARGSUSED */
9695 9701 int
9696 9702 ip_sioctl_addr(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
9697 9703 ip_ioctl_cmd_t *dummy_ipip, void *dummy_ifreq)
9698 9704 {
9699 9705 int err = 0;
9700 9706 in6_addr_t v6addr;
9701 9707 boolean_t need_up = B_FALSE;
9702 9708 ill_t *ill;
9703 9709 int i;
9704 9710
9705 9711 ip1dbg(("ip_sioctl_addr(%s:%u %p)\n",
9706 9712 ipif->ipif_ill->ill_name, ipif->ipif_id, (void *)ipif));
9707 9713
9708 9714 ASSERT(IAM_WRITER_IPIF(ipif));
9709 9715
9710 9716 ill = ipif->ipif_ill;
9711 9717 if (ipif->ipif_isv6) {
9712 9718 sin6_t *sin6;
9713 9719 phyint_t *phyi;
9714 9720
9715 9721 if (sin->sin_family != AF_INET6)
9716 9722 return (EAFNOSUPPORT);
9717 9723
9718 9724 sin6 = (sin6_t *)sin;
9719 9725 v6addr = sin6->sin6_addr;
9720 9726 phyi = ill->ill_phyint;
9721 9727
9722 9728 /*
9723 9729 * Enforce that true multicast interfaces have a link-local
9724 9730 * address for logical unit 0.
9725 9731 *
9726 9732 * However for those ipif's for which link-local address was
9727 9733 * not created by default, also allow setting :: as the address.
9728 9734 * This scenario would arise, when we delete an address on ipif
9729 9735 * with logical unit 0, we would want to set :: as the address.
9730 9736 */
9731 9737 if (ipif->ipif_id == 0 &&
9732 9738 (ill->ill_flags & ILLF_MULTICAST) &&
9733 9739 !(ipif->ipif_flags & (IPIF_POINTOPOINT)) &&
9734 9740 !(phyi->phyint_flags & (PHYI_LOOPBACK)) &&
9735 9741 !IN6_IS_ADDR_LINKLOCAL(&v6addr)) {
9736 9742
9737 9743 /*
9738 9744 * if default link-local was not created by kernel for
9739 9745 * this ill, allow setting :: as the address on ipif:0.
9740 9746 */
9741 9747 if (ill->ill_flags & ILLF_NOLINKLOCAL) {
9742 9748 if (!IN6_IS_ADDR_UNSPECIFIED(&v6addr))
9743 9749 return (EADDRNOTAVAIL);
9744 9750 } else {
9745 9751 return (EADDRNOTAVAIL);
9746 9752 }
9747 9753 }
9748 9754
9749 9755 /*
9750 9756 * up interfaces shouldn't have the unspecified address
9751 9757 * unless they also have the IPIF_NOLOCAL flags set and
9752 9758 * have a subnet assigned.
9753 9759 */
9754 9760 if ((ipif->ipif_flags & IPIF_UP) &&
9755 9761 IN6_IS_ADDR_UNSPECIFIED(&v6addr) &&
9756 9762 (!(ipif->ipif_flags & IPIF_NOLOCAL) ||
9757 9763 IN6_IS_ADDR_UNSPECIFIED(&ipif->ipif_v6subnet))) {
9758 9764 return (EADDRNOTAVAIL);
9759 9765 }
9760 9766
9761 9767 if (!ip_local_addr_ok_v6(&v6addr, &ipif->ipif_v6net_mask))
9762 9768 return (EADDRNOTAVAIL);
9763 9769 } else {
9764 9770 ipaddr_t addr;
9765 9771
9766 9772 if (sin->sin_family != AF_INET)
9767 9773 return (EAFNOSUPPORT);
9768 9774
9769 9775 addr = sin->sin_addr.s_addr;
9770 9776
9771 9777 /* Allow INADDR_ANY as the local address. */
9772 9778 if (addr != INADDR_ANY &&
9773 9779 !ip_addr_ok_v4(addr, ipif->ipif_net_mask))
9774 9780 return (EADDRNOTAVAIL);
9775 9781
9776 9782 IN6_IPADDR_TO_V4MAPPED(addr, &v6addr);
9777 9783 }
9778 9784 /*
9779 9785 * verify that the address being configured is permitted by the
9780 9786 * ill_allowed_ips[] for the interface.
9781 9787 */
9782 9788 if (ill->ill_allowed_ips_cnt > 0) {
9783 9789 for (i = 0; i < ill->ill_allowed_ips_cnt; i++) {
9784 9790 if (IN6_ARE_ADDR_EQUAL(&ill->ill_allowed_ips[i],
9785 9791 &v6addr))
9786 9792 break;
9787 9793 }
9788 9794 if (i == ill->ill_allowed_ips_cnt) {
9789 9795 pr_addr_dbg("!allowed addr %s\n", AF_INET6, &v6addr);
9790 9796 return (EPERM);
9791 9797 }
9792 9798 }
9793 9799 /*
9794 9800 * Even if there is no change we redo things just to rerun
9795 9801 * ipif_set_default.
9796 9802 */
9797 9803 if (ipif->ipif_flags & IPIF_UP) {
9798 9804 /*
9799 9805 * Setting a new local address, make sure
9800 9806 * we have net and subnet bcast ire's for
9801 9807 * the old address if we need them.
9802 9808 */
9803 9809 /*
9804 9810 * If the interface is already marked up,
9805 9811 * we call ipif_down which will take care
9806 9812 * of ditching any IREs that have been set
9807 9813 * up based on the old interface address.
9808 9814 */
9809 9815 err = ipif_logical_down(ipif, q, mp);
9810 9816 if (err == EINPROGRESS)
9811 9817 return (err);
9812 9818 (void) ipif_down_tail(ipif);
9813 9819 need_up = 1;
9814 9820 }
9815 9821
9816 9822 err = ip_sioctl_addr_tail(ipif, sin, q, mp, need_up);
9817 9823 return (err);
9818 9824 }
9819 9825
9820 9826 int
9821 9827 ip_sioctl_addr_tail(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
9822 9828 boolean_t need_up)
9823 9829 {
9824 9830 in6_addr_t v6addr;
9825 9831 in6_addr_t ov6addr;
9826 9832 ipaddr_t addr;
9827 9833 sin6_t *sin6;
9828 9834 int sinlen;
9829 9835 int err = 0;
9830 9836 ill_t *ill = ipif->ipif_ill;
9831 9837 boolean_t need_dl_down;
9832 9838 boolean_t need_arp_down;
9833 9839 struct iocblk *iocp;
9834 9840
9835 9841 iocp = (mp != NULL) ? (struct iocblk *)mp->b_rptr : NULL;
9836 9842
9837 9843 ip1dbg(("ip_sioctl_addr_tail(%s:%u %p)\n",
9838 9844 ill->ill_name, ipif->ipif_id, (void *)ipif));
9839 9845 ASSERT(IAM_WRITER_IPIF(ipif));
9840 9846
9841 9847 /* Must cancel any pending timer before taking the ill_lock */
9842 9848 if (ipif->ipif_recovery_id != 0)
9843 9849 (void) untimeout(ipif->ipif_recovery_id);
9844 9850 ipif->ipif_recovery_id = 0;
9845 9851
9846 9852 if (ipif->ipif_isv6) {
9847 9853 sin6 = (sin6_t *)sin;
9848 9854 v6addr = sin6->sin6_addr;
9849 9855 sinlen = sizeof (struct sockaddr_in6);
9850 9856 } else {
9851 9857 addr = sin->sin_addr.s_addr;
9852 9858 IN6_IPADDR_TO_V4MAPPED(addr, &v6addr);
9853 9859 sinlen = sizeof (struct sockaddr_in);
9854 9860 }
9855 9861 mutex_enter(&ill->ill_lock);
9856 9862 ov6addr = ipif->ipif_v6lcl_addr;
9857 9863 ipif->ipif_v6lcl_addr = v6addr;
9858 9864 sctp_update_ipif_addr(ipif, ov6addr);
9859 9865 ipif->ipif_addr_ready = 0;
9860 9866
9861 9867 ip_rts_newaddrmsg(RTM_CHGADDR, 0, ipif, RTSQ_DEFAULT);
9862 9868
9863 9869 /*
9864 9870 * If the interface was previously marked as a duplicate, then since
9865 9871 * we've now got a "new" address, it should no longer be considered a
9866 9872 * duplicate -- even if the "new" address is the same as the old one.
9867 9873 * Note that if all ipifs are down, we may have a pending ARP down
9868 9874 * event to handle. This is because we want to recover from duplicates
9869 9875 * and thus delay tearing down ARP until the duplicates have been
9870 9876 * removed or disabled.
9871 9877 */
9872 9878 need_dl_down = need_arp_down = B_FALSE;
9873 9879 if (ipif->ipif_flags & IPIF_DUPLICATE) {
9874 9880 need_arp_down = !need_up;
9875 9881 ipif->ipif_flags &= ~IPIF_DUPLICATE;
9876 9882 if (--ill->ill_ipif_dup_count == 0 && !need_up &&
9877 9883 ill->ill_ipif_up_count == 0 && ill->ill_dl_up) {
9878 9884 need_dl_down = B_TRUE;
9879 9885 }
9880 9886 }
9881 9887
9882 9888 ipif_set_default(ipif);
9883 9889
9884 9890 /*
9885 9891 * If we've just manually set the IPv6 link-local address (0th ipif),
9886 9892 * tag the ill so that future updates to the interface ID don't result
9887 9893 * in this address getting automatically reconfigured from under the
9888 9894 * administrator.
9889 9895 */
9890 9896 if (ipif->ipif_isv6 && ipif->ipif_id == 0) {
9891 9897 if (iocp == NULL || (iocp->ioc_cmd == SIOCSLIFADDR &&
9892 9898 !IN6_IS_ADDR_UNSPECIFIED(&v6addr)))
9893 9899 ill->ill_manual_linklocal = 1;
9894 9900 }
9895 9901
9896 9902 /*
9897 9903 * When publishing an interface address change event, we only notify
9898 9904 * the event listeners of the new address. It is assumed that if they
9899 9905 * actively care about the addresses assigned that they will have
9900 9906 * already discovered the previous address assigned (if there was one.)
9901 9907 *
9902 9908 * Don't attach nic event message for SIOCLIFADDIF ioctl.
9903 9909 */
9904 9910 if (iocp != NULL && iocp->ioc_cmd != SIOCLIFADDIF) {
9905 9911 ill_nic_event_dispatch(ill, MAP_IPIF_ID(ipif->ipif_id),
9906 9912 NE_ADDRESS_CHANGE, sin, sinlen);
9907 9913 }
9908 9914
9909 9915 mutex_exit(&ill->ill_lock);
9910 9916
9911 9917 if (need_up) {
9912 9918 /*
9913 9919 * Now bring the interface back up. If this
9914 9920 * is the only IPIF for the ILL, ipif_up
9915 9921 * will have to re-bind to the device, so
9916 9922 * we may get back EINPROGRESS, in which
9917 9923 * case, this IOCTL will get completed in
9918 9924 * ip_rput_dlpi when we see the DL_BIND_ACK.
9919 9925 */
9920 9926 err = ipif_up(ipif, q, mp);
9921 9927 } else {
9922 9928 /* Perhaps ilgs should use this ill */
9923 9929 update_conn_ill(NULL, ill->ill_ipst);
9924 9930 }
9925 9931
9926 9932 if (need_dl_down)
9927 9933 ill_dl_down(ill);
9928 9934
9929 9935 if (need_arp_down && !ill->ill_isv6)
9930 9936 (void) ipif_arp_down(ipif);
9931 9937
9932 9938 /*
9933 9939 * The default multicast interface might have changed (for
9934 9940 * instance if the IPv6 scope of the address changed)
9935 9941 */
9936 9942 ire_increment_multicast_generation(ill->ill_ipst, ill->ill_isv6);
9937 9943
9938 9944 return (err);
9939 9945 }
9940 9946
9941 9947 /*
9942 9948 * Restart entry point to restart the address set operation after the
9943 9949 * refcounts have dropped to zero.
9944 9950 */
9945 9951 /* ARGSUSED */
9946 9952 int
9947 9953 ip_sioctl_addr_restart(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
9948 9954 ip_ioctl_cmd_t *ipip, void *ifreq)
9949 9955 {
9950 9956 ip1dbg(("ip_sioctl_addr_restart(%s:%u %p)\n",
9951 9957 ipif->ipif_ill->ill_name, ipif->ipif_id, (void *)ipif));
9952 9958 ASSERT(IAM_WRITER_IPIF(ipif));
9953 9959 (void) ipif_down_tail(ipif);
9954 9960 return (ip_sioctl_addr_tail(ipif, sin, q, mp, B_TRUE));
9955 9961 }
9956 9962
9957 9963 /* ARGSUSED */
9958 9964 int
9959 9965 ip_sioctl_get_addr(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
9960 9966 ip_ioctl_cmd_t *ipip, void *if_req)
9961 9967 {
9962 9968 sin6_t *sin6 = (struct sockaddr_in6 *)sin;
9963 9969 struct lifreq *lifr = (struct lifreq *)if_req;
9964 9970
9965 9971 ip1dbg(("ip_sioctl_get_addr(%s:%u %p)\n",
9966 9972 ipif->ipif_ill->ill_name, ipif->ipif_id, (void *)ipif));
9967 9973 /*
9968 9974 * The net mask and address can't change since we have a
9969 9975 * reference to the ipif. So no lock is necessary.
9970 9976 */
9971 9977 if (ipif->ipif_isv6) {
9972 9978 *sin6 = sin6_null;
9973 9979 sin6->sin6_family = AF_INET6;
9974 9980 sin6->sin6_addr = ipif->ipif_v6lcl_addr;
9975 9981 ASSERT(ipip->ipi_cmd_type == LIF_CMD);
9976 9982 lifr->lifr_addrlen =
9977 9983 ip_mask_to_plen_v6(&ipif->ipif_v6net_mask);
9978 9984 } else {
9979 9985 *sin = sin_null;
9980 9986 sin->sin_family = AF_INET;
9981 9987 sin->sin_addr.s_addr = ipif->ipif_lcl_addr;
9982 9988 if (ipip->ipi_cmd_type == LIF_CMD) {
9983 9989 lifr->lifr_addrlen =
9984 9990 ip_mask_to_plen(ipif->ipif_net_mask);
9985 9991 }
9986 9992 }
9987 9993 return (0);
9988 9994 }
9989 9995
9990 9996 /*
9991 9997 * Set the destination address for a pt-pt interface.
9992 9998 */
9993 9999 /* ARGSUSED */
9994 10000 int
9995 10001 ip_sioctl_dstaddr(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
9996 10002 ip_ioctl_cmd_t *ipip, void *if_req)
9997 10003 {
9998 10004 int err = 0;
9999 10005 in6_addr_t v6addr;
10000 10006 boolean_t need_up = B_FALSE;
10001 10007
10002 10008 ip1dbg(("ip_sioctl_dstaddr(%s:%u %p)\n",
10003 10009 ipif->ipif_ill->ill_name, ipif->ipif_id, (void *)ipif));
10004 10010 ASSERT(IAM_WRITER_IPIF(ipif));
10005 10011
10006 10012 if (ipif->ipif_isv6) {
10007 10013 sin6_t *sin6;
10008 10014
10009 10015 if (sin->sin_family != AF_INET6)
10010 10016 return (EAFNOSUPPORT);
10011 10017
10012 10018 sin6 = (sin6_t *)sin;
10013 10019 v6addr = sin6->sin6_addr;
10014 10020
10015 10021 if (!ip_remote_addr_ok_v6(&v6addr, &ipif->ipif_v6net_mask))
10016 10022 return (EADDRNOTAVAIL);
10017 10023 } else {
10018 10024 ipaddr_t addr;
10019 10025
10020 10026 if (sin->sin_family != AF_INET)
10021 10027 return (EAFNOSUPPORT);
10022 10028
10023 10029 addr = sin->sin_addr.s_addr;
10024 10030 if (addr != INADDR_ANY &&
10025 10031 !ip_addr_ok_v4(addr, ipif->ipif_net_mask)) {
10026 10032 return (EADDRNOTAVAIL);
10027 10033 }
10028 10034
10029 10035 IN6_IPADDR_TO_V4MAPPED(addr, &v6addr);
10030 10036 }
10031 10037
10032 10038 if (IN6_ARE_ADDR_EQUAL(&ipif->ipif_v6pp_dst_addr, &v6addr))
10033 10039 return (0); /* No change */
10034 10040
10035 10041 if (ipif->ipif_flags & IPIF_UP) {
10036 10042 /*
10037 10043 * If the interface is already marked up,
10038 10044 * we call ipif_down which will take care
10039 10045 * of ditching any IREs that have been set
10040 10046 * up based on the old pp dst address.
10041 10047 */
10042 10048 err = ipif_logical_down(ipif, q, mp);
10043 10049 if (err == EINPROGRESS)
10044 10050 return (err);
10045 10051 (void) ipif_down_tail(ipif);
10046 10052 need_up = B_TRUE;
10047 10053 }
10048 10054 /*
10049 10055 * could return EINPROGRESS. If so ioctl will complete in
10050 10056 * ip_rput_dlpi_writer
10051 10057 */
10052 10058 err = ip_sioctl_dstaddr_tail(ipif, sin, q, mp, need_up);
10053 10059 return (err);
10054 10060 }
10055 10061
10056 10062 static int
10057 10063 ip_sioctl_dstaddr_tail(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
10058 10064 boolean_t need_up)
10059 10065 {
10060 10066 in6_addr_t v6addr;
10061 10067 ill_t *ill = ipif->ipif_ill;
10062 10068 int err = 0;
10063 10069 boolean_t need_dl_down;
10064 10070 boolean_t need_arp_down;
10065 10071
10066 10072 ip1dbg(("ip_sioctl_dstaddr_tail(%s:%u %p)\n", ill->ill_name,
10067 10073 ipif->ipif_id, (void *)ipif));
10068 10074
10069 10075 /* Must cancel any pending timer before taking the ill_lock */
10070 10076 if (ipif->ipif_recovery_id != 0)
10071 10077 (void) untimeout(ipif->ipif_recovery_id);
10072 10078 ipif->ipif_recovery_id = 0;
10073 10079
10074 10080 if (ipif->ipif_isv6) {
10075 10081 sin6_t *sin6;
10076 10082
10077 10083 sin6 = (sin6_t *)sin;
10078 10084 v6addr = sin6->sin6_addr;
10079 10085 } else {
10080 10086 ipaddr_t addr;
10081 10087
10082 10088 addr = sin->sin_addr.s_addr;
10083 10089 IN6_IPADDR_TO_V4MAPPED(addr, &v6addr);
10084 10090 }
10085 10091 mutex_enter(&ill->ill_lock);
10086 10092 /* Set point to point destination address. */
10087 10093 if ((ipif->ipif_flags & IPIF_POINTOPOINT) == 0) {
10088 10094 /*
10089 10095 * Allow this as a means of creating logical
10090 10096 * pt-pt interfaces on top of e.g. an Ethernet.
10091 10097 * XXX Undocumented HACK for testing.
10092 10098 * pt-pt interfaces are created with NUD disabled.
10093 10099 */
10094 10100 ipif->ipif_flags |= IPIF_POINTOPOINT;
10095 10101 ipif->ipif_flags &= ~IPIF_BROADCAST;
10096 10102 if (ipif->ipif_isv6)
10097 10103 ill->ill_flags |= ILLF_NONUD;
10098 10104 }
10099 10105
10100 10106 /*
10101 10107 * If the interface was previously marked as a duplicate, then since
10102 10108 * we've now got a "new" address, it should no longer be considered a
10103 10109 * duplicate -- even if the "new" address is the same as the old one.
10104 10110 * Note that if all ipifs are down, we may have a pending ARP down
10105 10111 * event to handle.
10106 10112 */
10107 10113 need_dl_down = need_arp_down = B_FALSE;
10108 10114 if (ipif->ipif_flags & IPIF_DUPLICATE) {
10109 10115 need_arp_down = !need_up;
10110 10116 ipif->ipif_flags &= ~IPIF_DUPLICATE;
10111 10117 if (--ill->ill_ipif_dup_count == 0 && !need_up &&
10112 10118 ill->ill_ipif_up_count == 0 && ill->ill_dl_up) {
10113 10119 need_dl_down = B_TRUE;
10114 10120 }
10115 10121 }
10116 10122
10117 10123 /*
10118 10124 * If we've just manually set the IPv6 destination link-local address
10119 10125 * (0th ipif), tag the ill so that future updates to the destination
10120 10126 * interface ID (as can happen with interfaces over IP tunnels) don't
10121 10127 * result in this address getting automatically reconfigured from
10122 10128 * under the administrator.
10123 10129 */
10124 10130 if (ipif->ipif_isv6 && ipif->ipif_id == 0)
10125 10131 ill->ill_manual_dst_linklocal = 1;
10126 10132
10127 10133 /* Set the new address. */
10128 10134 ipif->ipif_v6pp_dst_addr = v6addr;
10129 10135 /* Make sure subnet tracks pp_dst */
10130 10136 ipif->ipif_v6subnet = ipif->ipif_v6pp_dst_addr;
10131 10137 mutex_exit(&ill->ill_lock);
10132 10138
10133 10139 if (need_up) {
10134 10140 /*
10135 10141 * Now bring the interface back up. If this
10136 10142 * is the only IPIF for the ILL, ipif_up
10137 10143 * will have to re-bind to the device, so
10138 10144 * we may get back EINPROGRESS, in which
10139 10145 * case, this IOCTL will get completed in
10140 10146 * ip_rput_dlpi when we see the DL_BIND_ACK.
10141 10147 */
10142 10148 err = ipif_up(ipif, q, mp);
10143 10149 }
10144 10150
10145 10151 if (need_dl_down)
10146 10152 ill_dl_down(ill);
10147 10153 if (need_arp_down && !ipif->ipif_isv6)
10148 10154 (void) ipif_arp_down(ipif);
10149 10155
10150 10156 return (err);
10151 10157 }
10152 10158
10153 10159 /*
10154 10160 * Restart entry point to restart the dstaddress set operation after the
10155 10161 * refcounts have dropped to zero.
10156 10162 */
10157 10163 /* ARGSUSED */
10158 10164 int
10159 10165 ip_sioctl_dstaddr_restart(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
10160 10166 ip_ioctl_cmd_t *ipip, void *ifreq)
10161 10167 {
10162 10168 ip1dbg(("ip_sioctl_dstaddr_restart(%s:%u %p)\n",
10163 10169 ipif->ipif_ill->ill_name, ipif->ipif_id, (void *)ipif));
10164 10170 (void) ipif_down_tail(ipif);
10165 10171 return (ip_sioctl_dstaddr_tail(ipif, sin, q, mp, B_TRUE));
10166 10172 }
10167 10173
10168 10174 /* ARGSUSED */
10169 10175 int
10170 10176 ip_sioctl_get_dstaddr(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
10171 10177 ip_ioctl_cmd_t *ipip, void *if_req)
10172 10178 {
10173 10179 sin6_t *sin6 = (struct sockaddr_in6 *)sin;
10174 10180
10175 10181 ip1dbg(("ip_sioctl_get_dstaddr(%s:%u %p)\n",
10176 10182 ipif->ipif_ill->ill_name, ipif->ipif_id, (void *)ipif));
10177 10183 /*
10178 10184 * Get point to point destination address. The addresses can't
10179 10185 * change since we hold a reference to the ipif.
10180 10186 */
10181 10187 if ((ipif->ipif_flags & IPIF_POINTOPOINT) == 0)
10182 10188 return (EADDRNOTAVAIL);
10183 10189
10184 10190 if (ipif->ipif_isv6) {
10185 10191 ASSERT(ipip->ipi_cmd_type == LIF_CMD);
10186 10192 *sin6 = sin6_null;
10187 10193 sin6->sin6_family = AF_INET6;
10188 10194 sin6->sin6_addr = ipif->ipif_v6pp_dst_addr;
10189 10195 } else {
10190 10196 *sin = sin_null;
10191 10197 sin->sin_family = AF_INET;
10192 10198 sin->sin_addr.s_addr = ipif->ipif_pp_dst_addr;
10193 10199 }
10194 10200 return (0);
10195 10201 }
10196 10202
10197 10203 /*
10198 10204 * Check which flags will change by the given flags being set
10199 10205 * silently ignore flags which userland is not allowed to control.
10200 10206 * (Because these flags may change between SIOCGLIFFLAGS and
10201 10207 * SIOCSLIFFLAGS, and that's outside of userland's control,
10202 10208 * we need to silently ignore them rather than fail.)
10203 10209 */
10204 10210 static void
10205 10211 ip_sioctl_flags_onoff(ipif_t *ipif, uint64_t flags, uint64_t *onp,
10206 10212 uint64_t *offp)
10207 10213 {
10208 10214 ill_t *ill = ipif->ipif_ill;
10209 10215 phyint_t *phyi = ill->ill_phyint;
10210 10216 uint64_t cantchange_flags, intf_flags;
10211 10217 uint64_t turn_on, turn_off;
10212 10218
10213 10219 intf_flags = ipif->ipif_flags | ill->ill_flags | phyi->phyint_flags;
10214 10220 cantchange_flags = IFF_CANTCHANGE;
10215 10221 if (IS_IPMP(ill))
10216 10222 cantchange_flags |= IFF_IPMP_CANTCHANGE;
10217 10223 turn_on = (flags ^ intf_flags) & ~cantchange_flags;
10218 10224 turn_off = intf_flags & turn_on;
10219 10225 turn_on ^= turn_off;
10220 10226 *onp = turn_on;
10221 10227 *offp = turn_off;
10222 10228 }
10223 10229
10224 10230 /*
10225 10231 * Set interface flags. Many flags require special handling (e.g.,
10226 10232 * bringing the interface down); see below for details.
10227 10233 *
10228 10234 * NOTE : We really don't enforce that ipif_id zero should be used
10229 10235 * for setting any flags other than IFF_LOGINT_FLAGS. This
10230 10236 * is because applications generally does SICGLIFFLAGS and
10231 10237 * ORs in the new flags (that affects the logical) and does a
10232 10238 * SIOCSLIFFLAGS. Thus, "flags" below could contain bits other
10233 10239 * than IFF_LOGINT_FLAGS. One could check whether "turn_on" - the
10234 10240 * flags that will be turned on is correct with respect to
10235 10241 * ipif_id 0. For backward compatibility reasons, it is not done.
10236 10242 */
10237 10243 /* ARGSUSED */
10238 10244 int
10239 10245 ip_sioctl_flags(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
10240 10246 ip_ioctl_cmd_t *ipip, void *if_req)
10241 10247 {
10242 10248 uint64_t turn_on;
10243 10249 uint64_t turn_off;
10244 10250 int err = 0;
10245 10251 phyint_t *phyi;
10246 10252 ill_t *ill;
10247 10253 conn_t *connp;
10248 10254 uint64_t intf_flags;
10249 10255 boolean_t phyint_flags_modified = B_FALSE;
10250 10256 uint64_t flags;
10251 10257 struct ifreq *ifr;
10252 10258 struct lifreq *lifr;
10253 10259 boolean_t set_linklocal = B_FALSE;
10254 10260
10255 10261 ip1dbg(("ip_sioctl_flags(%s:%u %p)\n",
10256 10262 ipif->ipif_ill->ill_name, ipif->ipif_id, (void *)ipif));
10257 10263
10258 10264 ASSERT(IAM_WRITER_IPIF(ipif));
10259 10265
10260 10266 ill = ipif->ipif_ill;
10261 10267 phyi = ill->ill_phyint;
10262 10268
10263 10269 if (ipip->ipi_cmd_type == IF_CMD) {
10264 10270 ifr = (struct ifreq *)if_req;
10265 10271 flags = (uint64_t)(ifr->ifr_flags & 0x0000ffff);
10266 10272 } else {
10267 10273 lifr = (struct lifreq *)if_req;
10268 10274 flags = lifr->lifr_flags;
10269 10275 }
10270 10276
10271 10277 intf_flags = ipif->ipif_flags | ill->ill_flags | phyi->phyint_flags;
10272 10278
10273 10279 /*
10274 10280 * Have the flags been set correctly until now?
10275 10281 */
10276 10282 ASSERT((phyi->phyint_flags & ~(IFF_PHYINT_FLAGS)) == 0);
10277 10283 ASSERT((ill->ill_flags & ~(IFF_PHYINTINST_FLAGS)) == 0);
10278 10284 ASSERT((ipif->ipif_flags & ~(IFF_LOGINT_FLAGS)) == 0);
10279 10285 /*
10280 10286 * Compare the new flags to the old, and partition
10281 10287 * into those coming on and those going off.
10282 10288 * For the 16 bit command keep the bits above bit 16 unchanged.
10283 10289 */
10284 10290 if (ipip->ipi_cmd == SIOCSIFFLAGS)
10285 10291 flags |= intf_flags & ~0xFFFF;
10286 10292
10287 10293 /*
10288 10294 * Explicitly fail attempts to change flags that are always invalid on
10289 10295 * an IPMP meta-interface.
10290 10296 */
10291 10297 if (IS_IPMP(ill) && ((flags ^ intf_flags) & IFF_IPMP_INVALID))
10292 10298 return (EINVAL);
10293 10299
10294 10300 ip_sioctl_flags_onoff(ipif, flags, &turn_on, &turn_off);
10295 10301 if ((turn_on|turn_off) == 0)
10296 10302 return (0); /* No change */
10297 10303
10298 10304 /*
10299 10305 * All test addresses must be IFF_DEPRECATED (to ensure source address
10300 10306 * selection avoids them) -- so force IFF_DEPRECATED on, and do not
10301 10307 * allow it to be turned off.
10302 10308 */
10303 10309 if ((turn_off & (IFF_DEPRECATED|IFF_NOFAILOVER)) == IFF_DEPRECATED &&
10304 10310 (turn_on|intf_flags) & IFF_NOFAILOVER)
10305 10311 return (EINVAL);
10306 10312
10307 10313 if ((connp = Q_TO_CONN(q)) == NULL)
10308 10314 return (EINVAL);
10309 10315
10310 10316 /*
10311 10317 * Only vrrp control socket is allowed to change IFF_UP and
10312 10318 * IFF_NOACCEPT flags when IFF_VRRP is set.
10313 10319 */
10314 10320 if ((intf_flags & IFF_VRRP) && ((turn_off | turn_on) & IFF_UP)) {
10315 10321 if (!connp->conn_isvrrp)
10316 10322 return (EINVAL);
10317 10323 }
10318 10324
10319 10325 /*
10320 10326 * The IFF_NOACCEPT flag can only be set on an IFF_VRRP IP address by
10321 10327 * VRRP control socket.
10322 10328 */
10323 10329 if ((turn_off | turn_on) & IFF_NOACCEPT) {
10324 10330 if (!connp->conn_isvrrp || !(intf_flags & IFF_VRRP))
10325 10331 return (EINVAL);
10326 10332 }
10327 10333
10328 10334 if (turn_on & IFF_NOFAILOVER) {
10329 10335 turn_on |= IFF_DEPRECATED;
10330 10336 flags |= IFF_DEPRECATED;
10331 10337 }
10332 10338
10333 10339 /*
10334 10340 * On underlying interfaces, only allow applications to manage test
10335 10341 * addresses -- otherwise, they may get confused when the address
10336 10342 * moves as part of being brought up. Likewise, prevent an
10337 10343 * application-managed test address from being converted to a data
10338 10344 * address. To prevent migration of administratively up addresses in
10339 10345 * the kernel, we don't allow them to be converted either.
10340 10346 */
10341 10347 if (IS_UNDER_IPMP(ill)) {
10342 10348 const uint64_t appflags = IFF_DHCPRUNNING | IFF_ADDRCONF;
10343 10349
10344 10350 if ((turn_on & appflags) && !(flags & IFF_NOFAILOVER))
10345 10351 return (EINVAL);
10346 10352
10347 10353 if ((turn_off & IFF_NOFAILOVER) &&
10348 10354 (flags & (appflags | IFF_UP | IFF_DUPLICATE)))
10349 10355 return (EINVAL);
10350 10356 }
10351 10357
10352 10358 /*
10353 10359 * Only allow IFF_TEMPORARY flag to be set on
10354 10360 * IPv6 interfaces.
10355 10361 */
10356 10362 if ((turn_on & IFF_TEMPORARY) && !(ipif->ipif_isv6))
10357 10363 return (EINVAL);
10358 10364
10359 10365 /*
10360 10366 * cannot turn off IFF_NOXMIT on VNI interfaces.
10361 10367 */
10362 10368 if ((turn_off & IFF_NOXMIT) && IS_VNI(ipif->ipif_ill))
10363 10369 return (EINVAL);
10364 10370
10365 10371 /*
10366 10372 * Don't allow the IFF_ROUTER flag to be turned on on loopback
10367 10373 * interfaces. It makes no sense in that context.
10368 10374 */
10369 10375 if ((turn_on & IFF_ROUTER) && (phyi->phyint_flags & PHYI_LOOPBACK))
10370 10376 return (EINVAL);
10371 10377
10372 10378 /*
10373 10379 * For IPv6 ipif_id 0, don't allow the interface to be up without
10374 10380 * a link local address if IFF_NOLOCAL or IFF_ANYCAST are not set.
10375 10381 * If the link local address isn't set, and can be set, it will get
10376 10382 * set later on in this function.
10377 10383 */
10378 10384 if (ipif->ipif_id == 0 && ipif->ipif_isv6 &&
10379 10385 (flags & IFF_UP) && !(flags & (IFF_NOLOCAL|IFF_ANYCAST)) &&
10380 10386 IN6_IS_ADDR_UNSPECIFIED(&ipif->ipif_v6lcl_addr)) {
10381 10387 if (ipif_cant_setlinklocal(ipif))
10382 10388 return (EINVAL);
10383 10389 set_linklocal = B_TRUE;
10384 10390 }
10385 10391
10386 10392 /*
10387 10393 * If we modify physical interface flags, we'll potentially need to
10388 10394 * send up two routing socket messages for the changes (one for the
10389 10395 * IPv4 ill, and another for the IPv6 ill). Note that here.
10390 10396 */
10391 10397 if ((turn_on|turn_off) & IFF_PHYINT_FLAGS)
10392 10398 phyint_flags_modified = B_TRUE;
10393 10399
10394 10400 /*
10395 10401 * All functioning PHYI_STANDBY interfaces start life PHYI_INACTIVE
10396 10402 * (otherwise, we'd immediately use them, defeating standby). Also,
10397 10403 * since PHYI_INACTIVE has a separate meaning when PHYI_STANDBY is not
10398 10404 * set, don't allow PHYI_STANDBY to be set if PHYI_INACTIVE is already
10399 10405 * set, and clear PHYI_INACTIVE if PHYI_STANDBY is being cleared. We
10400 10406 * also don't allow PHYI_STANDBY if VNI is enabled since its semantics
10401 10407 * will not be honored.
10402 10408 */
10403 10409 if (turn_on & PHYI_STANDBY) {
10404 10410 /*
10405 10411 * No need to grab ill_g_usesrc_lock here; see the
10406 10412 * synchronization notes in ip.c.
10407 10413 */
10408 10414 if (ill->ill_usesrc_grp_next != NULL ||
10409 10415 intf_flags & PHYI_INACTIVE)
10410 10416 return (EINVAL);
10411 10417 if (!(flags & PHYI_FAILED)) {
10412 10418 flags |= PHYI_INACTIVE;
10413 10419 turn_on |= PHYI_INACTIVE;
10414 10420 }
10415 10421 }
10416 10422
10417 10423 if (turn_off & PHYI_STANDBY) {
10418 10424 flags &= ~PHYI_INACTIVE;
10419 10425 turn_off |= PHYI_INACTIVE;
10420 10426 }
10421 10427
10422 10428 /*
10423 10429 * PHYI_FAILED and PHYI_INACTIVE are mutually exclusive; fail if both
10424 10430 * would end up on.
10425 10431 */
10426 10432 if ((flags & (PHYI_FAILED | PHYI_INACTIVE)) ==
10427 10433 (PHYI_FAILED | PHYI_INACTIVE))
10428 10434 return (EINVAL);
10429 10435
10430 10436 /*
10431 10437 * If ILLF_ROUTER changes, we need to change the ip forwarding
10432 10438 * status of the interface.
10433 10439 */
10434 10440 if ((turn_on | turn_off) & ILLF_ROUTER) {
10435 10441 err = ill_forward_set(ill, ((turn_on & ILLF_ROUTER) != 0));
10436 10442 if (err != 0)
10437 10443 return (err);
10438 10444 }
10439 10445
10440 10446 /*
10441 10447 * If the interface is not UP and we are not going to
10442 10448 * bring it UP, record the flags and return. When the
10443 10449 * interface comes UP later, the right actions will be
10444 10450 * taken.
10445 10451 */
10446 10452 if (!(ipif->ipif_flags & IPIF_UP) &&
10447 10453 !(turn_on & IPIF_UP)) {
10448 10454 /* Record new flags in their respective places. */
10449 10455 mutex_enter(&ill->ill_lock);
10450 10456 mutex_enter(&ill->ill_phyint->phyint_lock);
10451 10457 ipif->ipif_flags |= (turn_on & IFF_LOGINT_FLAGS);
10452 10458 ipif->ipif_flags &= (~turn_off & IFF_LOGINT_FLAGS);
10453 10459 ill->ill_flags |= (turn_on & IFF_PHYINTINST_FLAGS);
10454 10460 ill->ill_flags &= (~turn_off & IFF_PHYINTINST_FLAGS);
10455 10461 phyi->phyint_flags |= (turn_on & IFF_PHYINT_FLAGS);
10456 10462 phyi->phyint_flags &= (~turn_off & IFF_PHYINT_FLAGS);
10457 10463 mutex_exit(&ill->ill_lock);
10458 10464 mutex_exit(&ill->ill_phyint->phyint_lock);
10459 10465
10460 10466 /*
10461 10467 * PHYI_FAILED, PHYI_INACTIVE, and PHYI_OFFLINE are all the
10462 10468 * same to the kernel: if any of them has been set by
10463 10469 * userland, the interface cannot be used for data traffic.
10464 10470 */
10465 10471 if ((turn_on|turn_off) &
10466 10472 (PHYI_FAILED | PHYI_INACTIVE | PHYI_OFFLINE)) {
10467 10473 ASSERT(!IS_IPMP(ill));
10468 10474 /*
10469 10475 * It's possible the ill is part of an "anonymous"
10470 10476 * IPMP group rather than a real group. In that case,
10471 10477 * there are no other interfaces in the group and thus
10472 10478 * no need to call ipmp_phyint_refresh_active().
10473 10479 */
10474 10480 if (IS_UNDER_IPMP(ill))
10475 10481 ipmp_phyint_refresh_active(phyi);
10476 10482 }
10477 10483
10478 10484 if (phyint_flags_modified) {
10479 10485 if (phyi->phyint_illv4 != NULL) {
10480 10486 ip_rts_ifmsg(phyi->phyint_illv4->
10481 10487 ill_ipif, RTSQ_DEFAULT);
10482 10488 }
10483 10489 if (phyi->phyint_illv6 != NULL) {
10484 10490 ip_rts_ifmsg(phyi->phyint_illv6->
10485 10491 ill_ipif, RTSQ_DEFAULT);
10486 10492 }
10487 10493 }
10488 10494 /* The default multicast interface might have changed */
10489 10495 ire_increment_multicast_generation(ill->ill_ipst,
10490 10496 ill->ill_isv6);
10491 10497
10492 10498 return (0);
10493 10499 } else if (set_linklocal) {
10494 10500 mutex_enter(&ill->ill_lock);
10495 10501 if (set_linklocal)
10496 10502 ipif->ipif_state_flags |= IPIF_SET_LINKLOCAL;
10497 10503 mutex_exit(&ill->ill_lock);
10498 10504 }
10499 10505
10500 10506 /*
10501 10507 * Disallow IPv6 interfaces coming up that have the unspecified address,
10502 10508 * or point-to-point interfaces with an unspecified destination. We do
10503 10509 * allow the address to be unspecified for IPIF_NOLOCAL interfaces that
10504 10510 * have a subnet assigned, which is how in.ndpd currently manages its
10505 10511 * onlink prefix list when no addresses are configured with those
10506 10512 * prefixes.
10507 10513 */
10508 10514 if (ipif->ipif_isv6 &&
10509 10515 ((IN6_IS_ADDR_UNSPECIFIED(&ipif->ipif_v6lcl_addr) &&
10510 10516 (!(ipif->ipif_flags & IPIF_NOLOCAL) && !(turn_on & IPIF_NOLOCAL) ||
10511 10517 IN6_IS_ADDR_UNSPECIFIED(&ipif->ipif_v6subnet))) ||
10512 10518 ((ipif->ipif_flags & IPIF_POINTOPOINT) &&
10513 10519 IN6_IS_ADDR_UNSPECIFIED(&ipif->ipif_v6pp_dst_addr)))) {
10514 10520 return (EINVAL);
10515 10521 }
10516 10522
10517 10523 /*
10518 10524 * Prevent IPv4 point-to-point interfaces with a 0.0.0.0 destination
10519 10525 * from being brought up.
10520 10526 */
10521 10527 if (!ipif->ipif_isv6 &&
10522 10528 ((ipif->ipif_flags & IPIF_POINTOPOINT) &&
10523 10529 ipif->ipif_pp_dst_addr == INADDR_ANY)) {
10524 10530 return (EINVAL);
10525 10531 }
10526 10532
10527 10533 /*
10528 10534 * If we are going to change one or more of the flags that are
10529 10535 * IPIF_UP, IPIF_DEPRECATED, IPIF_NOXMIT, IPIF_NOLOCAL, ILLF_NOARP,
10530 10536 * ILLF_NONUD, IPIF_PRIVATE, IPIF_ANYCAST, IPIF_PREFERRED, and
10531 10537 * IPIF_NOFAILOVER, we will take special action. This is
10532 10538 * done by bring the ipif down, changing the flags and bringing
10533 10539 * it back up again. For IPIF_NOFAILOVER, the act of bringing it
10534 10540 * back up will trigger the address to be moved.
10535 10541 *
10536 10542 * If we are going to change IFF_NOACCEPT, we need to bring
10537 10543 * all the ipifs down then bring them up again. The act of
10538 10544 * bringing all the ipifs back up will trigger the local
10539 10545 * ires being recreated with "no_accept" set/cleared.
10540 10546 *
10541 10547 * Note that ILLF_NOACCEPT is always set separately from the
10542 10548 * other flags.
10543 10549 */
10544 10550 if ((turn_on|turn_off) &
10545 10551 (IPIF_UP|IPIF_DEPRECATED|IPIF_NOXMIT|IPIF_NOLOCAL|ILLF_NOARP|
10546 10552 ILLF_NONUD|IPIF_PRIVATE|IPIF_ANYCAST|IPIF_PREFERRED|
10547 10553 IPIF_NOFAILOVER)) {
10548 10554 /*
10549 10555 * ipif_down() will ire_delete bcast ire's for the subnet,
10550 10556 * while the ire_identical_ref tracks the case of IRE_BROADCAST
10551 10557 * entries shared between multiple ipifs on the same subnet.
10552 10558 */
10553 10559 if (((ipif->ipif_flags | turn_on) & IPIF_UP) &&
10554 10560 !(turn_off & IPIF_UP)) {
10555 10561 if (ipif->ipif_flags & IPIF_UP)
10556 10562 ill->ill_logical_down = 1;
10557 10563 turn_on &= ~IPIF_UP;
10558 10564 }
10559 10565 err = ipif_down(ipif, q, mp);
10560 10566 ip1dbg(("ipif_down returns %d err ", err));
10561 10567 if (err == EINPROGRESS)
10562 10568 return (err);
10563 10569 (void) ipif_down_tail(ipif);
10564 10570 } else if ((turn_on|turn_off) & ILLF_NOACCEPT) {
10565 10571 /*
10566 10572 * If we can quiesce the ill, then continue. If not, then
10567 10573 * ip_sioctl_flags_tail() will be called from
10568 10574 * ipif_ill_refrele_tail().
10569 10575 */
10570 10576 ill_down_ipifs(ill, B_TRUE);
10571 10577
10572 10578 mutex_enter(&connp->conn_lock);
10573 10579 mutex_enter(&ill->ill_lock);
10574 10580 if (!ill_is_quiescent(ill)) {
10575 10581 boolean_t success;
10576 10582
10577 10583 success = ipsq_pending_mp_add(connp, ill->ill_ipif,
10578 10584 q, mp, ILL_DOWN);
10579 10585 mutex_exit(&ill->ill_lock);
10580 10586 mutex_exit(&connp->conn_lock);
10581 10587 return (success ? EINPROGRESS : EINTR);
10582 10588 }
10583 10589 mutex_exit(&ill->ill_lock);
10584 10590 mutex_exit(&connp->conn_lock);
10585 10591 }
10586 10592 return (ip_sioctl_flags_tail(ipif, flags, q, mp));
10587 10593 }
10588 10594
10589 10595 static int
10590 10596 ip_sioctl_flags_tail(ipif_t *ipif, uint64_t flags, queue_t *q, mblk_t *mp)
10591 10597 {
10592 10598 ill_t *ill;
10593 10599 phyint_t *phyi;
10594 10600 uint64_t turn_on, turn_off;
10595 10601 boolean_t phyint_flags_modified = B_FALSE;
10596 10602 int err = 0;
10597 10603 boolean_t set_linklocal = B_FALSE;
10598 10604
10599 10605 ip1dbg(("ip_sioctl_flags_tail(%s:%u)\n",
10600 10606 ipif->ipif_ill->ill_name, ipif->ipif_id));
10601 10607
10602 10608 ASSERT(IAM_WRITER_IPIF(ipif));
10603 10609
10604 10610 ill = ipif->ipif_ill;
10605 10611 phyi = ill->ill_phyint;
10606 10612
10607 10613 ip_sioctl_flags_onoff(ipif, flags, &turn_on, &turn_off);
10608 10614
10609 10615 /*
10610 10616 * IFF_UP is handled separately.
10611 10617 */
10612 10618 turn_on &= ~IFF_UP;
10613 10619 turn_off &= ~IFF_UP;
10614 10620
10615 10621 if ((turn_on|turn_off) & IFF_PHYINT_FLAGS)
10616 10622 phyint_flags_modified = B_TRUE;
10617 10623
10618 10624 /*
10619 10625 * Now we change the flags. Track current value of
10620 10626 * other flags in their respective places.
10621 10627 */
10622 10628 mutex_enter(&ill->ill_lock);
10623 10629 mutex_enter(&phyi->phyint_lock);
10624 10630 ipif->ipif_flags |= (turn_on & IFF_LOGINT_FLAGS);
10625 10631 ipif->ipif_flags &= (~turn_off & IFF_LOGINT_FLAGS);
10626 10632 ill->ill_flags |= (turn_on & IFF_PHYINTINST_FLAGS);
10627 10633 ill->ill_flags &= (~turn_off & IFF_PHYINTINST_FLAGS);
10628 10634 phyi->phyint_flags |= (turn_on & IFF_PHYINT_FLAGS);
10629 10635 phyi->phyint_flags &= (~turn_off & IFF_PHYINT_FLAGS);
10630 10636 if (ipif->ipif_state_flags & IPIF_SET_LINKLOCAL) {
10631 10637 set_linklocal = B_TRUE;
10632 10638 ipif->ipif_state_flags &= ~IPIF_SET_LINKLOCAL;
10633 10639 }
10634 10640
10635 10641 mutex_exit(&ill->ill_lock);
10636 10642 mutex_exit(&phyi->phyint_lock);
10637 10643
10638 10644 if (set_linklocal)
10639 10645 (void) ipif_setlinklocal(ipif);
10640 10646
10641 10647 /*
10642 10648 * PHYI_FAILED, PHYI_INACTIVE, and PHYI_OFFLINE are all the same to
10643 10649 * the kernel: if any of them has been set by userland, the interface
10644 10650 * cannot be used for data traffic.
10645 10651 */
10646 10652 if ((turn_on|turn_off) & (PHYI_FAILED | PHYI_INACTIVE | PHYI_OFFLINE)) {
10647 10653 ASSERT(!IS_IPMP(ill));
10648 10654 /*
10649 10655 * It's possible the ill is part of an "anonymous" IPMP group
10650 10656 * rather than a real group. In that case, there are no other
10651 10657 * interfaces in the group and thus no need for us to call
10652 10658 * ipmp_phyint_refresh_active().
10653 10659 */
10654 10660 if (IS_UNDER_IPMP(ill))
10655 10661 ipmp_phyint_refresh_active(phyi);
10656 10662 }
10657 10663
10658 10664 if ((turn_on|turn_off) & ILLF_NOACCEPT) {
10659 10665 /*
10660 10666 * If the ILLF_NOACCEPT flag is changed, bring up all the
10661 10667 * ipifs that were brought down.
10662 10668 *
10663 10669 * The routing sockets messages are sent as the result
10664 10670 * of ill_up_ipifs(), further, SCTP's IPIF list was updated
10665 10671 * as well.
10666 10672 */
10667 10673 err = ill_up_ipifs(ill, q, mp);
10668 10674 } else if ((flags & IFF_UP) && !(ipif->ipif_flags & IPIF_UP)) {
10669 10675 /*
10670 10676 * XXX ipif_up really does not know whether a phyint flags
10671 10677 * was modified or not. So, it sends up information on
10672 10678 * only one routing sockets message. As we don't bring up
10673 10679 * the interface and also set PHYI_ flags simultaneously
10674 10680 * it should be okay.
10675 10681 */
10676 10682 err = ipif_up(ipif, q, mp);
10677 10683 } else {
10678 10684 /*
10679 10685 * Make sure routing socket sees all changes to the flags.
10680 10686 * ipif_up_done* handles this when we use ipif_up.
10681 10687 */
10682 10688 if (phyint_flags_modified) {
10683 10689 if (phyi->phyint_illv4 != NULL) {
10684 10690 ip_rts_ifmsg(phyi->phyint_illv4->
10685 10691 ill_ipif, RTSQ_DEFAULT);
10686 10692 }
10687 10693 if (phyi->phyint_illv6 != NULL) {
10688 10694 ip_rts_ifmsg(phyi->phyint_illv6->
10689 10695 ill_ipif, RTSQ_DEFAULT);
10690 10696 }
10691 10697 } else {
10692 10698 ip_rts_ifmsg(ipif, RTSQ_DEFAULT);
10693 10699 }
10694 10700 /*
10695 10701 * Update the flags in SCTP's IPIF list, ipif_up() will do
10696 10702 * this in need_up case.
10697 10703 */
10698 10704 sctp_update_ipif(ipif, SCTP_IPIF_UPDATE);
10699 10705 }
10700 10706
10701 10707 /* The default multicast interface might have changed */
10702 10708 ire_increment_multicast_generation(ill->ill_ipst, ill->ill_isv6);
10703 10709 return (err);
10704 10710 }
10705 10711
10706 10712 /*
10707 10713 * Restart the flags operation now that the refcounts have dropped to zero.
10708 10714 */
10709 10715 /* ARGSUSED */
10710 10716 int
10711 10717 ip_sioctl_flags_restart(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
10712 10718 ip_ioctl_cmd_t *ipip, void *if_req)
10713 10719 {
10714 10720 uint64_t flags;
10715 10721 struct ifreq *ifr = if_req;
10716 10722 struct lifreq *lifr = if_req;
10717 10723 uint64_t turn_on, turn_off;
10718 10724
10719 10725 ip1dbg(("ip_sioctl_flags_restart(%s:%u %p)\n",
10720 10726 ipif->ipif_ill->ill_name, ipif->ipif_id, (void *)ipif));
10721 10727
10722 10728 if (ipip->ipi_cmd_type == IF_CMD) {
10723 10729 /* cast to uint16_t prevents unwanted sign extension */
10724 10730 flags = (uint16_t)ifr->ifr_flags;
10725 10731 } else {
10726 10732 flags = lifr->lifr_flags;
10727 10733 }
10728 10734
10729 10735 /*
10730 10736 * If this function call is a result of the ILLF_NOACCEPT flag
10731 10737 * change, do not call ipif_down_tail(). See ip_sioctl_flags().
10732 10738 */
10733 10739 ip_sioctl_flags_onoff(ipif, flags, &turn_on, &turn_off);
10734 10740 if (!((turn_on|turn_off) & ILLF_NOACCEPT))
10735 10741 (void) ipif_down_tail(ipif);
10736 10742
10737 10743 return (ip_sioctl_flags_tail(ipif, flags, q, mp));
10738 10744 }
10739 10745
10740 10746 /*
10741 10747 * Can operate on either a module or a driver queue.
10742 10748 */
10743 10749 /* ARGSUSED */
10744 10750 int
10745 10751 ip_sioctl_get_flags(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
10746 10752 ip_ioctl_cmd_t *ipip, void *if_req)
10747 10753 {
10748 10754 /*
10749 10755 * Has the flags been set correctly till now ?
10750 10756 */
10751 10757 ill_t *ill = ipif->ipif_ill;
10752 10758 phyint_t *phyi = ill->ill_phyint;
10753 10759
10754 10760 ip1dbg(("ip_sioctl_get_flags(%s:%u %p)\n",
10755 10761 ipif->ipif_ill->ill_name, ipif->ipif_id, (void *)ipif));
10756 10762 ASSERT((phyi->phyint_flags & ~(IFF_PHYINT_FLAGS)) == 0);
10757 10763 ASSERT((ill->ill_flags & ~(IFF_PHYINTINST_FLAGS)) == 0);
10758 10764 ASSERT((ipif->ipif_flags & ~(IFF_LOGINT_FLAGS)) == 0);
10759 10765
10760 10766 /*
10761 10767 * Need a lock since some flags can be set even when there are
10762 10768 * references to the ipif.
10763 10769 */
10764 10770 mutex_enter(&ill->ill_lock);
10765 10771 if (ipip->ipi_cmd_type == IF_CMD) {
10766 10772 struct ifreq *ifr = (struct ifreq *)if_req;
10767 10773
10768 10774 /* Get interface flags (low 16 only). */
10769 10775 ifr->ifr_flags = ((ipif->ipif_flags |
10770 10776 ill->ill_flags | phyi->phyint_flags) & 0xffff);
10771 10777 } else {
10772 10778 struct lifreq *lifr = (struct lifreq *)if_req;
10773 10779
10774 10780 /* Get interface flags. */
10775 10781 lifr->lifr_flags = ipif->ipif_flags |
10776 10782 ill->ill_flags | phyi->phyint_flags;
10777 10783 }
10778 10784 mutex_exit(&ill->ill_lock);
10779 10785 return (0);
10780 10786 }
10781 10787
10782 10788 /*
10783 10789 * We allow the MTU to be set on an ILL, but not have it be different
10784 10790 * for different IPIFs since we don't actually send packets on IPIFs.
10785 10791 */
10786 10792 /* ARGSUSED */
10787 10793 int
10788 10794 ip_sioctl_mtu(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
10789 10795 ip_ioctl_cmd_t *ipip, void *if_req)
10790 10796 {
10791 10797 int mtu;
10792 10798 int ip_min_mtu;
10793 10799 struct ifreq *ifr;
10794 10800 struct lifreq *lifr;
10795 10801 ill_t *ill;
10796 10802
10797 10803 ip1dbg(("ip_sioctl_mtu(%s:%u %p)\n", ipif->ipif_ill->ill_name,
10798 10804 ipif->ipif_id, (void *)ipif));
10799 10805 if (ipip->ipi_cmd_type == IF_CMD) {
10800 10806 ifr = (struct ifreq *)if_req;
10801 10807 mtu = ifr->ifr_metric;
10802 10808 } else {
10803 10809 lifr = (struct lifreq *)if_req;
10804 10810 mtu = lifr->lifr_mtu;
10805 10811 }
10806 10812 /* Only allow for logical unit zero i.e. not on "bge0:17" */
10807 10813 if (ipif->ipif_id != 0)
10808 10814 return (EINVAL);
10809 10815
10810 10816 ill = ipif->ipif_ill;
10811 10817 if (ipif->ipif_isv6)
10812 10818 ip_min_mtu = IPV6_MIN_MTU;
10813 10819 else
10814 10820 ip_min_mtu = IP_MIN_MTU;
10815 10821
10816 10822 mutex_enter(&ill->ill_lock);
10817 10823 if (mtu > ill->ill_max_frag || mtu < ip_min_mtu) {
10818 10824 mutex_exit(&ill->ill_lock);
10819 10825 return (EINVAL);
10820 10826 }
10821 10827 /* Avoid increasing ill_mc_mtu */
10822 10828 if (ill->ill_mc_mtu > mtu)
10823 10829 ill->ill_mc_mtu = mtu;
10824 10830
10825 10831 /*
10826 10832 * The dce and fragmentation code can handle changes to ill_mtu
10827 10833 * concurrent with sending/fragmenting packets.
10828 10834 */
10829 10835 ill->ill_mtu = mtu;
10830 10836 ill->ill_flags |= ILLF_FIXEDMTU;
10831 10837 mutex_exit(&ill->ill_lock);
10832 10838
10833 10839 /*
10834 10840 * Make sure all dce_generation checks find out
10835 10841 * that ill_mtu/ill_mc_mtu has changed.
10836 10842 */
10837 10843 dce_increment_all_generations(ill->ill_isv6, ill->ill_ipst);
10838 10844
10839 10845 /*
10840 10846 * Refresh IPMP meta-interface MTU if necessary.
10841 10847 */
10842 10848 if (IS_UNDER_IPMP(ill))
10843 10849 ipmp_illgrp_refresh_mtu(ill->ill_grp);
10844 10850
10845 10851 /* Update the MTU in SCTP's list */
10846 10852 sctp_update_ipif(ipif, SCTP_IPIF_UPDATE);
10847 10853 return (0);
10848 10854 }
10849 10855
10850 10856 /* Get interface MTU. */
10851 10857 /* ARGSUSED */
10852 10858 int
10853 10859 ip_sioctl_get_mtu(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
10854 10860 ip_ioctl_cmd_t *ipip, void *if_req)
10855 10861 {
10856 10862 struct ifreq *ifr;
10857 10863 struct lifreq *lifr;
10858 10864
10859 10865 ip1dbg(("ip_sioctl_get_mtu(%s:%u %p)\n",
10860 10866 ipif->ipif_ill->ill_name, ipif->ipif_id, (void *)ipif));
10861 10867
10862 10868 /*
10863 10869 * We allow a get on any logical interface even though the set
10864 10870 * can only be done on logical unit 0.
10865 10871 */
10866 10872 if (ipip->ipi_cmd_type == IF_CMD) {
10867 10873 ifr = (struct ifreq *)if_req;
10868 10874 ifr->ifr_metric = ipif->ipif_ill->ill_mtu;
10869 10875 } else {
10870 10876 lifr = (struct lifreq *)if_req;
10871 10877 lifr->lifr_mtu = ipif->ipif_ill->ill_mtu;
10872 10878 }
10873 10879 return (0);
10874 10880 }
10875 10881
10876 10882 /* Set interface broadcast address. */
10877 10883 /* ARGSUSED2 */
10878 10884 int
10879 10885 ip_sioctl_brdaddr(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
10880 10886 ip_ioctl_cmd_t *ipip, void *if_req)
10881 10887 {
10882 10888 ipaddr_t addr;
10883 10889 ire_t *ire;
10884 10890 ill_t *ill = ipif->ipif_ill;
10885 10891 ip_stack_t *ipst = ill->ill_ipst;
10886 10892
10887 10893 ip1dbg(("ip_sioctl_brdaddr(%s:%u)\n", ill->ill_name,
10888 10894 ipif->ipif_id));
10889 10895
10890 10896 ASSERT(IAM_WRITER_IPIF(ipif));
10891 10897 if (!(ipif->ipif_flags & IPIF_BROADCAST))
10892 10898 return (EADDRNOTAVAIL);
10893 10899
10894 10900 ASSERT(!(ipif->ipif_isv6)); /* No IPv6 broadcast */
10895 10901
10896 10902 if (sin->sin_family != AF_INET)
10897 10903 return (EAFNOSUPPORT);
10898 10904
10899 10905 addr = sin->sin_addr.s_addr;
10900 10906
10901 10907 if (ipif->ipif_flags & IPIF_UP) {
10902 10908 /*
10903 10909 * If we are already up, make sure the new
10904 10910 * broadcast address makes sense. If it does,
10905 10911 * there should be an IRE for it already.
10906 10912 */
10907 10913 ire = ire_ftable_lookup_v4(addr, 0, 0, IRE_BROADCAST,
10908 10914 ill, ipif->ipif_zoneid, NULL,
10909 10915 (MATCH_IRE_ILL | MATCH_IRE_TYPE), 0, ipst, NULL);
10910 10916 if (ire == NULL) {
10911 10917 return (EINVAL);
10912 10918 } else {
10913 10919 ire_refrele(ire);
10914 10920 }
10915 10921 }
10916 10922 /*
10917 10923 * Changing the broadcast addr for this ipif. Since the IRE_BROADCAST
10918 10924 * needs to already exist we never need to change the set of
10919 10925 * IRE_BROADCASTs when we are UP.
10920 10926 */
10921 10927 if (addr != ipif->ipif_brd_addr)
10922 10928 IN6_IPADDR_TO_V4MAPPED(addr, &ipif->ipif_v6brd_addr);
10923 10929
10924 10930 return (0);
10925 10931 }
10926 10932
10927 10933 /* Get interface broadcast address. */
10928 10934 /* ARGSUSED */
10929 10935 int
10930 10936 ip_sioctl_get_brdaddr(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
10931 10937 ip_ioctl_cmd_t *ipip, void *if_req)
10932 10938 {
10933 10939 ip1dbg(("ip_sioctl_get_brdaddr(%s:%u %p)\n",
10934 10940 ipif->ipif_ill->ill_name, ipif->ipif_id, (void *)ipif));
10935 10941 if (!(ipif->ipif_flags & IPIF_BROADCAST))
10936 10942 return (EADDRNOTAVAIL);
10937 10943
10938 10944 /* IPIF_BROADCAST not possible with IPv6 */
10939 10945 ASSERT(!ipif->ipif_isv6);
10940 10946 *sin = sin_null;
10941 10947 sin->sin_family = AF_INET;
10942 10948 sin->sin_addr.s_addr = ipif->ipif_brd_addr;
10943 10949 return (0);
10944 10950 }
10945 10951
10946 10952 /*
10947 10953 * This routine is called to handle the SIOCS*IFNETMASK IOCTL.
10948 10954 */
10949 10955 /* ARGSUSED */
10950 10956 int
10951 10957 ip_sioctl_netmask(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
10952 10958 ip_ioctl_cmd_t *ipip, void *if_req)
10953 10959 {
10954 10960 int err = 0;
10955 10961 in6_addr_t v6mask;
10956 10962
10957 10963 ip1dbg(("ip_sioctl_netmask(%s:%u %p)\n",
10958 10964 ipif->ipif_ill->ill_name, ipif->ipif_id, (void *)ipif));
10959 10965
10960 10966 ASSERT(IAM_WRITER_IPIF(ipif));
10961 10967
10962 10968 if (ipif->ipif_isv6) {
10963 10969 sin6_t *sin6;
10964 10970
10965 10971 if (sin->sin_family != AF_INET6)
10966 10972 return (EAFNOSUPPORT);
10967 10973
10968 10974 sin6 = (sin6_t *)sin;
10969 10975 v6mask = sin6->sin6_addr;
10970 10976 } else {
10971 10977 ipaddr_t mask;
10972 10978
10973 10979 if (sin->sin_family != AF_INET)
10974 10980 return (EAFNOSUPPORT);
10975 10981
10976 10982 mask = sin->sin_addr.s_addr;
10977 10983 if (!ip_contiguous_mask(ntohl(mask)))
10978 10984 return (ENOTSUP);
10979 10985 V4MASK_TO_V6(mask, v6mask);
10980 10986 }
10981 10987
10982 10988 /*
10983 10989 * No big deal if the interface isn't already up, or the mask
10984 10990 * isn't really changing, or this is pt-pt.
10985 10991 */
10986 10992 if (!(ipif->ipif_flags & IPIF_UP) ||
10987 10993 IN6_ARE_ADDR_EQUAL(&v6mask, &ipif->ipif_v6net_mask) ||
10988 10994 (ipif->ipif_flags & IPIF_POINTOPOINT)) {
10989 10995 ipif->ipif_v6net_mask = v6mask;
10990 10996 if ((ipif->ipif_flags & IPIF_POINTOPOINT) == 0) {
10991 10997 V6_MASK_COPY(ipif->ipif_v6lcl_addr,
10992 10998 ipif->ipif_v6net_mask,
10993 10999 ipif->ipif_v6subnet);
10994 11000 }
10995 11001 return (0);
10996 11002 }
10997 11003 /*
10998 11004 * Make sure we have valid net and subnet broadcast ire's
10999 11005 * for the old netmask, if needed by other logical interfaces.
11000 11006 */
11001 11007 err = ipif_logical_down(ipif, q, mp);
11002 11008 if (err == EINPROGRESS)
11003 11009 return (err);
11004 11010 (void) ipif_down_tail(ipif);
11005 11011 err = ip_sioctl_netmask_tail(ipif, sin, q, mp);
11006 11012 return (err);
11007 11013 }
11008 11014
11009 11015 static int
11010 11016 ip_sioctl_netmask_tail(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp)
11011 11017 {
11012 11018 in6_addr_t v6mask;
11013 11019 int err = 0;
11014 11020
11015 11021 ip1dbg(("ip_sioctl_netmask_tail(%s:%u %p)\n",
11016 11022 ipif->ipif_ill->ill_name, ipif->ipif_id, (void *)ipif));
11017 11023
11018 11024 if (ipif->ipif_isv6) {
11019 11025 sin6_t *sin6;
11020 11026
11021 11027 sin6 = (sin6_t *)sin;
11022 11028 v6mask = sin6->sin6_addr;
11023 11029 } else {
11024 11030 ipaddr_t mask;
11025 11031
11026 11032 mask = sin->sin_addr.s_addr;
11027 11033 V4MASK_TO_V6(mask, v6mask);
11028 11034 }
11029 11035
11030 11036 ipif->ipif_v6net_mask = v6mask;
11031 11037 if ((ipif->ipif_flags & IPIF_POINTOPOINT) == 0) {
11032 11038 V6_MASK_COPY(ipif->ipif_v6lcl_addr, ipif->ipif_v6net_mask,
11033 11039 ipif->ipif_v6subnet);
11034 11040 }
11035 11041 err = ipif_up(ipif, q, mp);
11036 11042
11037 11043 if (err == 0 || err == EINPROGRESS) {
11038 11044 /*
11039 11045 * The interface must be DL_BOUND if this packet has to
11040 11046 * go out on the wire. Since we only go through a logical
11041 11047 * down and are bound with the driver during an internal
11042 11048 * down/up that is satisfied.
11043 11049 */
11044 11050 if (!ipif->ipif_isv6 && ipif->ipif_ill->ill_wq != NULL) {
11045 11051 /* Potentially broadcast an address mask reply. */
11046 11052 ipif_mask_reply(ipif);
11047 11053 }
11048 11054 }
11049 11055 return (err);
11050 11056 }
11051 11057
11052 11058 /* ARGSUSED */
11053 11059 int
11054 11060 ip_sioctl_netmask_restart(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
11055 11061 ip_ioctl_cmd_t *ipip, void *if_req)
11056 11062 {
11057 11063 ip1dbg(("ip_sioctl_netmask_restart(%s:%u %p)\n",
11058 11064 ipif->ipif_ill->ill_name, ipif->ipif_id, (void *)ipif));
11059 11065 (void) ipif_down_tail(ipif);
11060 11066 return (ip_sioctl_netmask_tail(ipif, sin, q, mp));
11061 11067 }
11062 11068
11063 11069 /* Get interface net mask. */
11064 11070 /* ARGSUSED */
11065 11071 int
11066 11072 ip_sioctl_get_netmask(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
11067 11073 ip_ioctl_cmd_t *ipip, void *if_req)
11068 11074 {
11069 11075 struct lifreq *lifr = (struct lifreq *)if_req;
11070 11076 struct sockaddr_in6 *sin6 = (sin6_t *)sin;
11071 11077
11072 11078 ip1dbg(("ip_sioctl_get_netmask(%s:%u %p)\n",
11073 11079 ipif->ipif_ill->ill_name, ipif->ipif_id, (void *)ipif));
11074 11080
11075 11081 /*
11076 11082 * net mask can't change since we have a reference to the ipif.
11077 11083 */
11078 11084 if (ipif->ipif_isv6) {
11079 11085 ASSERT(ipip->ipi_cmd_type == LIF_CMD);
11080 11086 *sin6 = sin6_null;
11081 11087 sin6->sin6_family = AF_INET6;
11082 11088 sin6->sin6_addr = ipif->ipif_v6net_mask;
11083 11089 lifr->lifr_addrlen =
11084 11090 ip_mask_to_plen_v6(&ipif->ipif_v6net_mask);
11085 11091 } else {
11086 11092 *sin = sin_null;
11087 11093 sin->sin_family = AF_INET;
11088 11094 sin->sin_addr.s_addr = ipif->ipif_net_mask;
11089 11095 if (ipip->ipi_cmd_type == LIF_CMD) {
11090 11096 lifr->lifr_addrlen =
11091 11097 ip_mask_to_plen(ipif->ipif_net_mask);
11092 11098 }
11093 11099 }
11094 11100 return (0);
11095 11101 }
11096 11102
11097 11103 /* ARGSUSED */
11098 11104 int
11099 11105 ip_sioctl_metric(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
11100 11106 ip_ioctl_cmd_t *ipip, void *if_req)
11101 11107 {
11102 11108 ip1dbg(("ip_sioctl_metric(%s:%u %p)\n",
11103 11109 ipif->ipif_ill->ill_name, ipif->ipif_id, (void *)ipif));
11104 11110
11105 11111 /*
11106 11112 * Since no applications should ever be setting metrics on underlying
11107 11113 * interfaces, we explicitly fail to smoke 'em out.
11108 11114 */
11109 11115 if (IS_UNDER_IPMP(ipif->ipif_ill))
11110 11116 return (EINVAL);
11111 11117
11112 11118 /*
11113 11119 * Set interface metric. We don't use this for
11114 11120 * anything but we keep track of it in case it is
11115 11121 * important to routing applications or such.
11116 11122 */
11117 11123 if (ipip->ipi_cmd_type == IF_CMD) {
11118 11124 struct ifreq *ifr;
11119 11125
11120 11126 ifr = (struct ifreq *)if_req;
11121 11127 ipif->ipif_ill->ill_metric = ifr->ifr_metric;
11122 11128 } else {
11123 11129 struct lifreq *lifr;
11124 11130
11125 11131 lifr = (struct lifreq *)if_req;
11126 11132 ipif->ipif_ill->ill_metric = lifr->lifr_metric;
11127 11133 }
11128 11134 return (0);
11129 11135 }
11130 11136
11131 11137 /* ARGSUSED */
11132 11138 int
11133 11139 ip_sioctl_get_metric(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
11134 11140 ip_ioctl_cmd_t *ipip, void *if_req)
11135 11141 {
11136 11142 /* Get interface metric. */
11137 11143 ip1dbg(("ip_sioctl_get_metric(%s:%u %p)\n",
11138 11144 ipif->ipif_ill->ill_name, ipif->ipif_id, (void *)ipif));
11139 11145
11140 11146 if (ipip->ipi_cmd_type == IF_CMD) {
11141 11147 struct ifreq *ifr;
11142 11148
11143 11149 ifr = (struct ifreq *)if_req;
11144 11150 ifr->ifr_metric = ipif->ipif_ill->ill_metric;
11145 11151 } else {
11146 11152 struct lifreq *lifr;
11147 11153
11148 11154 lifr = (struct lifreq *)if_req;
11149 11155 lifr->lifr_metric = ipif->ipif_ill->ill_metric;
11150 11156 }
11151 11157
11152 11158 return (0);
11153 11159 }
11154 11160
11155 11161 /* ARGSUSED */
11156 11162 int
11157 11163 ip_sioctl_muxid(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
11158 11164 ip_ioctl_cmd_t *ipip, void *if_req)
11159 11165 {
11160 11166 int arp_muxid;
11161 11167
11162 11168 ip1dbg(("ip_sioctl_muxid(%s:%u %p)\n",
11163 11169 ipif->ipif_ill->ill_name, ipif->ipif_id, (void *)ipif));
11164 11170 /*
11165 11171 * Set the muxid returned from I_PLINK.
11166 11172 */
11167 11173 if (ipip->ipi_cmd_type == IF_CMD) {
11168 11174 struct ifreq *ifr = (struct ifreq *)if_req;
11169 11175
11170 11176 ipif->ipif_ill->ill_muxid = ifr->ifr_ip_muxid;
11171 11177 arp_muxid = ifr->ifr_arp_muxid;
11172 11178 } else {
11173 11179 struct lifreq *lifr = (struct lifreq *)if_req;
11174 11180
11175 11181 ipif->ipif_ill->ill_muxid = lifr->lifr_ip_muxid;
11176 11182 arp_muxid = lifr->lifr_arp_muxid;
11177 11183 }
11178 11184 arl_set_muxid(ipif->ipif_ill, arp_muxid);
11179 11185 return (0);
11180 11186 }
11181 11187
11182 11188 /* ARGSUSED */
11183 11189 int
11184 11190 ip_sioctl_get_muxid(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
11185 11191 ip_ioctl_cmd_t *ipip, void *if_req)
11186 11192 {
11187 11193 int arp_muxid = 0;
11188 11194
11189 11195 ip1dbg(("ip_sioctl_get_muxid(%s:%u %p)\n",
11190 11196 ipif->ipif_ill->ill_name, ipif->ipif_id, (void *)ipif));
11191 11197 /*
11192 11198 * Get the muxid saved in ill for I_PUNLINK.
11193 11199 */
11194 11200 arp_muxid = arl_get_muxid(ipif->ipif_ill);
11195 11201 if (ipip->ipi_cmd_type == IF_CMD) {
11196 11202 struct ifreq *ifr = (struct ifreq *)if_req;
11197 11203
11198 11204 ifr->ifr_ip_muxid = ipif->ipif_ill->ill_muxid;
11199 11205 ifr->ifr_arp_muxid = arp_muxid;
11200 11206 } else {
11201 11207 struct lifreq *lifr = (struct lifreq *)if_req;
11202 11208
11203 11209 lifr->lifr_ip_muxid = ipif->ipif_ill->ill_muxid;
11204 11210 lifr->lifr_arp_muxid = arp_muxid;
11205 11211 }
11206 11212 return (0);
11207 11213 }
11208 11214
11209 11215 /*
11210 11216 * Set the subnet prefix. Does not modify the broadcast address.
11211 11217 */
11212 11218 /* ARGSUSED */
11213 11219 int
11214 11220 ip_sioctl_subnet(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
11215 11221 ip_ioctl_cmd_t *ipip, void *if_req)
11216 11222 {
11217 11223 int err = 0;
11218 11224 in6_addr_t v6addr;
11219 11225 in6_addr_t v6mask;
11220 11226 boolean_t need_up = B_FALSE;
11221 11227 int addrlen;
11222 11228
11223 11229 ip1dbg(("ip_sioctl_subnet(%s:%u %p)\n",
11224 11230 ipif->ipif_ill->ill_name, ipif->ipif_id, (void *)ipif));
11225 11231
11226 11232 ASSERT(IAM_WRITER_IPIF(ipif));
11227 11233 addrlen = ((struct lifreq *)if_req)->lifr_addrlen;
11228 11234
11229 11235 if (ipif->ipif_isv6) {
11230 11236 sin6_t *sin6;
11231 11237
11232 11238 if (sin->sin_family != AF_INET6)
11233 11239 return (EAFNOSUPPORT);
11234 11240
11235 11241 sin6 = (sin6_t *)sin;
11236 11242 v6addr = sin6->sin6_addr;
11237 11243 if (!ip_remote_addr_ok_v6(&v6addr, &ipv6_all_ones))
11238 11244 return (EADDRNOTAVAIL);
11239 11245 } else {
11240 11246 ipaddr_t addr;
11241 11247
11242 11248 if (sin->sin_family != AF_INET)
11243 11249 return (EAFNOSUPPORT);
11244 11250
11245 11251 addr = sin->sin_addr.s_addr;
11246 11252 if (!ip_addr_ok_v4(addr, 0xFFFFFFFF))
11247 11253 return (EADDRNOTAVAIL);
11248 11254 IN6_IPADDR_TO_V4MAPPED(addr, &v6addr);
11249 11255 /* Add 96 bits */
11250 11256 addrlen += IPV6_ABITS - IP_ABITS;
11251 11257 }
11252 11258
11253 11259 if (ip_plen_to_mask_v6(addrlen, &v6mask) == NULL)
11254 11260 return (EINVAL);
11255 11261
11256 11262 /* Check if bits in the address is set past the mask */
11257 11263 if (!V6_MASK_EQ(v6addr, v6mask, v6addr))
11258 11264 return (EINVAL);
11259 11265
11260 11266 if (IN6_ARE_ADDR_EQUAL(&ipif->ipif_v6subnet, &v6addr) &&
11261 11267 IN6_ARE_ADDR_EQUAL(&ipif->ipif_v6net_mask, &v6mask))
11262 11268 return (0); /* No change */
11263 11269
11264 11270 if (ipif->ipif_flags & IPIF_UP) {
11265 11271 /*
11266 11272 * If the interface is already marked up,
11267 11273 * we call ipif_down which will take care
11268 11274 * of ditching any IREs that have been set
11269 11275 * up based on the old interface address.
11270 11276 */
11271 11277 err = ipif_logical_down(ipif, q, mp);
11272 11278 if (err == EINPROGRESS)
11273 11279 return (err);
11274 11280 (void) ipif_down_tail(ipif);
11275 11281 need_up = B_TRUE;
11276 11282 }
11277 11283
11278 11284 err = ip_sioctl_subnet_tail(ipif, v6addr, v6mask, q, mp, need_up);
11279 11285 return (err);
11280 11286 }
11281 11287
11282 11288 static int
11283 11289 ip_sioctl_subnet_tail(ipif_t *ipif, in6_addr_t v6addr, in6_addr_t v6mask,
11284 11290 queue_t *q, mblk_t *mp, boolean_t need_up)
11285 11291 {
11286 11292 ill_t *ill = ipif->ipif_ill;
11287 11293 int err = 0;
11288 11294
11289 11295 ip1dbg(("ip_sioctl_subnet_tail(%s:%u %p)\n",
11290 11296 ipif->ipif_ill->ill_name, ipif->ipif_id, (void *)ipif));
11291 11297
11292 11298 /* Set the new address. */
11293 11299 mutex_enter(&ill->ill_lock);
11294 11300 ipif->ipif_v6net_mask = v6mask;
11295 11301 if ((ipif->ipif_flags & IPIF_POINTOPOINT) == 0) {
11296 11302 V6_MASK_COPY(v6addr, ipif->ipif_v6net_mask,
11297 11303 ipif->ipif_v6subnet);
11298 11304 }
11299 11305 mutex_exit(&ill->ill_lock);
11300 11306
11301 11307 if (need_up) {
11302 11308 /*
11303 11309 * Now bring the interface back up. If this
11304 11310 * is the only IPIF for the ILL, ipif_up
11305 11311 * will have to re-bind to the device, so
11306 11312 * we may get back EINPROGRESS, in which
11307 11313 * case, this IOCTL will get completed in
11308 11314 * ip_rput_dlpi when we see the DL_BIND_ACK.
11309 11315 */
11310 11316 err = ipif_up(ipif, q, mp);
11311 11317 if (err == EINPROGRESS)
11312 11318 return (err);
11313 11319 }
11314 11320 return (err);
11315 11321 }
11316 11322
11317 11323 /* ARGSUSED */
11318 11324 int
11319 11325 ip_sioctl_subnet_restart(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
11320 11326 ip_ioctl_cmd_t *ipip, void *if_req)
11321 11327 {
11322 11328 int addrlen;
11323 11329 in6_addr_t v6addr;
11324 11330 in6_addr_t v6mask;
11325 11331 struct lifreq *lifr = (struct lifreq *)if_req;
11326 11332
11327 11333 ip1dbg(("ip_sioctl_subnet_restart(%s:%u %p)\n",
11328 11334 ipif->ipif_ill->ill_name, ipif->ipif_id, (void *)ipif));
11329 11335 (void) ipif_down_tail(ipif);
11330 11336
11331 11337 addrlen = lifr->lifr_addrlen;
11332 11338 if (ipif->ipif_isv6) {
11333 11339 sin6_t *sin6;
11334 11340
11335 11341 sin6 = (sin6_t *)sin;
11336 11342 v6addr = sin6->sin6_addr;
11337 11343 } else {
11338 11344 ipaddr_t addr;
11339 11345
11340 11346 addr = sin->sin_addr.s_addr;
11341 11347 IN6_IPADDR_TO_V4MAPPED(addr, &v6addr);
11342 11348 addrlen += IPV6_ABITS - IP_ABITS;
11343 11349 }
11344 11350 (void) ip_plen_to_mask_v6(addrlen, &v6mask);
11345 11351
11346 11352 return (ip_sioctl_subnet_tail(ipif, v6addr, v6mask, q, mp, B_TRUE));
11347 11353 }
11348 11354
11349 11355 /* ARGSUSED */
11350 11356 int
11351 11357 ip_sioctl_get_subnet(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
11352 11358 ip_ioctl_cmd_t *ipip, void *if_req)
11353 11359 {
11354 11360 struct lifreq *lifr = (struct lifreq *)if_req;
11355 11361 struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)sin;
11356 11362
11357 11363 ip1dbg(("ip_sioctl_get_subnet(%s:%u %p)\n",
11358 11364 ipif->ipif_ill->ill_name, ipif->ipif_id, (void *)ipif));
11359 11365 ASSERT(ipip->ipi_cmd_type == LIF_CMD);
11360 11366
11361 11367 if (ipif->ipif_isv6) {
11362 11368 *sin6 = sin6_null;
11363 11369 sin6->sin6_family = AF_INET6;
11364 11370 sin6->sin6_addr = ipif->ipif_v6subnet;
11365 11371 lifr->lifr_addrlen =
11366 11372 ip_mask_to_plen_v6(&ipif->ipif_v6net_mask);
11367 11373 } else {
11368 11374 *sin = sin_null;
11369 11375 sin->sin_family = AF_INET;
11370 11376 sin->sin_addr.s_addr = ipif->ipif_subnet;
11371 11377 lifr->lifr_addrlen = ip_mask_to_plen(ipif->ipif_net_mask);
11372 11378 }
11373 11379 return (0);
11374 11380 }
11375 11381
11376 11382 /*
11377 11383 * Set the IPv6 address token.
11378 11384 */
11379 11385 /* ARGSUSED */
11380 11386 int
11381 11387 ip_sioctl_token(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
11382 11388 ip_ioctl_cmd_t *ipi, void *if_req)
11383 11389 {
11384 11390 ill_t *ill = ipif->ipif_ill;
11385 11391 int err;
11386 11392 in6_addr_t v6addr;
11387 11393 in6_addr_t v6mask;
11388 11394 boolean_t need_up = B_FALSE;
11389 11395 int i;
11390 11396 sin6_t *sin6 = (sin6_t *)sin;
11391 11397 struct lifreq *lifr = (struct lifreq *)if_req;
11392 11398 int addrlen;
11393 11399
11394 11400 ip1dbg(("ip_sioctl_token(%s:%u %p)\n",
11395 11401 ipif->ipif_ill->ill_name, ipif->ipif_id, (void *)ipif));
11396 11402 ASSERT(IAM_WRITER_IPIF(ipif));
11397 11403
11398 11404 addrlen = lifr->lifr_addrlen;
11399 11405 /* Only allow for logical unit zero i.e. not on "le0:17" */
11400 11406 if (ipif->ipif_id != 0)
11401 11407 return (EINVAL);
11402 11408
11403 11409 if (!ipif->ipif_isv6)
11404 11410 return (EINVAL);
11405 11411
11406 11412 if (addrlen > IPV6_ABITS)
11407 11413 return (EINVAL);
11408 11414
11409 11415 v6addr = sin6->sin6_addr;
11410 11416
11411 11417 /*
11412 11418 * The length of the token is the length from the end. To get
11413 11419 * the proper mask for this, compute the mask of the bits not
11414 11420 * in the token; ie. the prefix, and then xor to get the mask.
11415 11421 */
11416 11422 if (ip_plen_to_mask_v6(IPV6_ABITS - addrlen, &v6mask) == NULL)
11417 11423 return (EINVAL);
11418 11424 for (i = 0; i < 4; i++) {
11419 11425 v6mask.s6_addr32[i] ^= (uint32_t)0xffffffff;
11420 11426 }
11421 11427
11422 11428 if (V6_MASK_EQ(v6addr, v6mask, ill->ill_token) &&
11423 11429 ill->ill_token_length == addrlen)
11424 11430 return (0); /* No change */
11425 11431
11426 11432 if (ipif->ipif_flags & IPIF_UP) {
11427 11433 err = ipif_logical_down(ipif, q, mp);
11428 11434 if (err == EINPROGRESS)
11429 11435 return (err);
11430 11436 (void) ipif_down_tail(ipif);
11431 11437 need_up = B_TRUE;
11432 11438 }
11433 11439 err = ip_sioctl_token_tail(ipif, sin6, addrlen, q, mp, need_up);
11434 11440 return (err);
11435 11441 }
11436 11442
11437 11443 static int
11438 11444 ip_sioctl_token_tail(ipif_t *ipif, sin6_t *sin6, int addrlen, queue_t *q,
11439 11445 mblk_t *mp, boolean_t need_up)
11440 11446 {
11441 11447 in6_addr_t v6addr;
11442 11448 in6_addr_t v6mask;
11443 11449 ill_t *ill = ipif->ipif_ill;
11444 11450 int i;
11445 11451 int err = 0;
11446 11452
11447 11453 ip1dbg(("ip_sioctl_token_tail(%s:%u %p)\n",
11448 11454 ipif->ipif_ill->ill_name, ipif->ipif_id, (void *)ipif));
11449 11455 v6addr = sin6->sin6_addr;
11450 11456 /*
11451 11457 * The length of the token is the length from the end. To get
11452 11458 * the proper mask for this, compute the mask of the bits not
11453 11459 * in the token; ie. the prefix, and then xor to get the mask.
11454 11460 */
11455 11461 (void) ip_plen_to_mask_v6(IPV6_ABITS - addrlen, &v6mask);
11456 11462 for (i = 0; i < 4; i++)
11457 11463 v6mask.s6_addr32[i] ^= (uint32_t)0xffffffff;
11458 11464
11459 11465 mutex_enter(&ill->ill_lock);
11460 11466 V6_MASK_COPY(v6addr, v6mask, ill->ill_token);
11461 11467 ill->ill_token_length = addrlen;
11462 11468 ill->ill_manual_token = 1;
11463 11469
11464 11470 /* Reconfigure the link-local address based on this new token */
11465 11471 ipif_setlinklocal(ill->ill_ipif);
11466 11472
11467 11473 mutex_exit(&ill->ill_lock);
11468 11474
11469 11475 if (need_up) {
11470 11476 /*
11471 11477 * Now bring the interface back up. If this
11472 11478 * is the only IPIF for the ILL, ipif_up
11473 11479 * will have to re-bind to the device, so
11474 11480 * we may get back EINPROGRESS, in which
11475 11481 * case, this IOCTL will get completed in
11476 11482 * ip_rput_dlpi when we see the DL_BIND_ACK.
11477 11483 */
11478 11484 err = ipif_up(ipif, q, mp);
11479 11485 if (err == EINPROGRESS)
11480 11486 return (err);
11481 11487 }
11482 11488 return (err);
11483 11489 }
11484 11490
11485 11491 /* ARGSUSED */
11486 11492 int
11487 11493 ip_sioctl_get_token(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
11488 11494 ip_ioctl_cmd_t *ipi, void *if_req)
11489 11495 {
11490 11496 ill_t *ill;
11491 11497 sin6_t *sin6 = (sin6_t *)sin;
11492 11498 struct lifreq *lifr = (struct lifreq *)if_req;
11493 11499
11494 11500 ip1dbg(("ip_sioctl_get_token(%s:%u %p)\n",
11495 11501 ipif->ipif_ill->ill_name, ipif->ipif_id, (void *)ipif));
11496 11502 if (ipif->ipif_id != 0)
11497 11503 return (EINVAL);
11498 11504
11499 11505 ill = ipif->ipif_ill;
11500 11506 if (!ill->ill_isv6)
11501 11507 return (ENXIO);
11502 11508
11503 11509 *sin6 = sin6_null;
11504 11510 sin6->sin6_family = AF_INET6;
11505 11511 ASSERT(!IN6_IS_ADDR_V4MAPPED(&ill->ill_token));
11506 11512 sin6->sin6_addr = ill->ill_token;
11507 11513 lifr->lifr_addrlen = ill->ill_token_length;
11508 11514 return (0);
11509 11515 }
11510 11516
11511 11517 /*
11512 11518 * Set (hardware) link specific information that might override
11513 11519 * what was acquired through the DL_INFO_ACK.
11514 11520 */
11515 11521 /* ARGSUSED */
11516 11522 int
11517 11523 ip_sioctl_lnkinfo(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
11518 11524 ip_ioctl_cmd_t *ipi, void *if_req)
11519 11525 {
11520 11526 ill_t *ill = ipif->ipif_ill;
11521 11527 int ip_min_mtu;
11522 11528 struct lifreq *lifr = (struct lifreq *)if_req;
11523 11529 lif_ifinfo_req_t *lir;
11524 11530
11525 11531 ip1dbg(("ip_sioctl_lnkinfo(%s:%u %p)\n",
11526 11532 ipif->ipif_ill->ill_name, ipif->ipif_id, (void *)ipif));
11527 11533 lir = &lifr->lifr_ifinfo;
11528 11534 ASSERT(IAM_WRITER_IPIF(ipif));
11529 11535
11530 11536 /* Only allow for logical unit zero i.e. not on "bge0:17" */
11531 11537 if (ipif->ipif_id != 0)
11532 11538 return (EINVAL);
11533 11539
11534 11540 /* Set interface MTU. */
11535 11541 if (ipif->ipif_isv6)
11536 11542 ip_min_mtu = IPV6_MIN_MTU;
11537 11543 else
11538 11544 ip_min_mtu = IP_MIN_MTU;
11539 11545
11540 11546 /*
11541 11547 * Verify values before we set anything. Allow zero to
11542 11548 * mean unspecified.
11543 11549 *
11544 11550 * XXX We should be able to set the user-defined lir_mtu to some value
11545 11551 * that is greater than ill_current_frag but less than ill_max_frag- the
11546 11552 * ill_max_frag value tells us the max MTU that can be handled by the
11547 11553 * datalink, whereas the ill_current_frag is dynamically computed for
11548 11554 * some link-types like tunnels, based on the tunnel PMTU. However,
11549 11555 * since there is currently no way of distinguishing between
11550 11556 * administratively fixed link mtu values (e.g., those set via
11551 11557 * /sbin/dladm) and dynamically discovered MTUs (e.g., those discovered
11552 11558 * for tunnels) we conservatively choose the ill_current_frag as the
11553 11559 * upper-bound.
11554 11560 */
11555 11561 if (lir->lir_maxmtu != 0 &&
11556 11562 (lir->lir_maxmtu > ill->ill_current_frag ||
11557 11563 lir->lir_maxmtu < ip_min_mtu))
11558 11564 return (EINVAL);
11559 11565 if (lir->lir_reachtime != 0 &&
11560 11566 lir->lir_reachtime > ND_MAX_REACHTIME)
11561 11567 return (EINVAL);
11562 11568 if (lir->lir_reachretrans != 0 &&
11563 11569 lir->lir_reachretrans > ND_MAX_REACHRETRANSTIME)
11564 11570 return (EINVAL);
11565 11571
11566 11572 mutex_enter(&ill->ill_lock);
11567 11573 /*
11568 11574 * The dce and fragmentation code can handle changes to ill_mtu
11569 11575 * concurrent with sending/fragmenting packets.
11570 11576 */
11571 11577 if (lir->lir_maxmtu != 0)
11572 11578 ill->ill_user_mtu = lir->lir_maxmtu;
11573 11579
11574 11580 if (lir->lir_reachtime != 0)
11575 11581 ill->ill_reachable_time = lir->lir_reachtime;
11576 11582
11577 11583 if (lir->lir_reachretrans != 0)
11578 11584 ill->ill_reachable_retrans_time = lir->lir_reachretrans;
11579 11585
11580 11586 ill->ill_max_hops = lir->lir_maxhops;
11581 11587 ill->ill_max_buf = ND_MAX_Q;
11582 11588 if (!(ill->ill_flags & ILLF_FIXEDMTU) && ill->ill_user_mtu != 0) {
11583 11589 /*
11584 11590 * ill_mtu is the actual interface MTU, obtained as the min
11585 11591 * of user-configured mtu and the value announced by the
11586 11592 * driver (via DL_NOTE_SDU_SIZE/DL_INFO_ACK). Note that since
11587 11593 * we have already made the choice of requiring
11588 11594 * ill_user_mtu < ill_current_frag by the time we get here,
11589 11595 * the ill_mtu effectively gets assigned to the ill_user_mtu
11590 11596 * here.
11591 11597 */
11592 11598 ill->ill_mtu = MIN(ill->ill_current_frag, ill->ill_user_mtu);
11593 11599 ill->ill_mc_mtu = MIN(ill->ill_mc_mtu, ill->ill_user_mtu);
11594 11600 }
11595 11601 mutex_exit(&ill->ill_lock);
11596 11602
11597 11603 /*
11598 11604 * Make sure all dce_generation checks find out
11599 11605 * that ill_mtu/ill_mc_mtu has changed.
11600 11606 */
11601 11607 if (!(ill->ill_flags & ILLF_FIXEDMTU) && (lir->lir_maxmtu != 0))
11602 11608 dce_increment_all_generations(ill->ill_isv6, ill->ill_ipst);
11603 11609
11604 11610 /*
11605 11611 * Refresh IPMP meta-interface MTU if necessary.
11606 11612 */
11607 11613 if (IS_UNDER_IPMP(ill))
11608 11614 ipmp_illgrp_refresh_mtu(ill->ill_grp);
11609 11615
11610 11616 return (0);
11611 11617 }
11612 11618
11613 11619 /* ARGSUSED */
11614 11620 int
11615 11621 ip_sioctl_get_lnkinfo(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
11616 11622 ip_ioctl_cmd_t *ipi, void *if_req)
11617 11623 {
11618 11624 struct lif_ifinfo_req *lir;
11619 11625 ill_t *ill = ipif->ipif_ill;
11620 11626
11621 11627 ip1dbg(("ip_sioctl_get_lnkinfo(%s:%u %p)\n",
11622 11628 ipif->ipif_ill->ill_name, ipif->ipif_id, (void *)ipif));
11623 11629 if (ipif->ipif_id != 0)
11624 11630 return (EINVAL);
11625 11631
11626 11632 lir = &((struct lifreq *)if_req)->lifr_ifinfo;
11627 11633 lir->lir_maxhops = ill->ill_max_hops;
11628 11634 lir->lir_reachtime = ill->ill_reachable_time;
11629 11635 lir->lir_reachretrans = ill->ill_reachable_retrans_time;
11630 11636 lir->lir_maxmtu = ill->ill_mtu;
11631 11637
11632 11638 return (0);
11633 11639 }
11634 11640
11635 11641 /*
11636 11642 * Return best guess as to the subnet mask for the specified address.
11637 11643 * Based on the subnet masks for all the configured interfaces.
11638 11644 *
11639 11645 * We end up returning a zero mask in the case of default, multicast or
11640 11646 * experimental.
11641 11647 */
11642 11648 static ipaddr_t
11643 11649 ip_subnet_mask(ipaddr_t addr, ipif_t **ipifp, ip_stack_t *ipst)
11644 11650 {
11645 11651 ipaddr_t net_mask;
11646 11652 ill_t *ill;
11647 11653 ipif_t *ipif;
11648 11654 ill_walk_context_t ctx;
11649 11655 ipif_t *fallback_ipif = NULL;
11650 11656
11651 11657 net_mask = ip_net_mask(addr);
11652 11658 if (net_mask == 0) {
11653 11659 *ipifp = NULL;
11654 11660 return (0);
11655 11661 }
11656 11662
11657 11663 /* Let's check to see if this is maybe a local subnet route. */
11658 11664 /* this function only applies to IPv4 interfaces */
11659 11665 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
11660 11666 ill = ILL_START_WALK_V4(&ctx, ipst);
11661 11667 for (; ill != NULL; ill = ill_next(&ctx, ill)) {
11662 11668 mutex_enter(&ill->ill_lock);
11663 11669 for (ipif = ill->ill_ipif; ipif != NULL;
11664 11670 ipif = ipif->ipif_next) {
11665 11671 if (IPIF_IS_CONDEMNED(ipif))
11666 11672 continue;
11667 11673 if (!(ipif->ipif_flags & IPIF_UP))
11668 11674 continue;
11669 11675 if ((ipif->ipif_subnet & net_mask) ==
11670 11676 (addr & net_mask)) {
11671 11677 /*
11672 11678 * Don't trust pt-pt interfaces if there are
11673 11679 * other interfaces.
11674 11680 */
11675 11681 if (ipif->ipif_flags & IPIF_POINTOPOINT) {
11676 11682 if (fallback_ipif == NULL) {
11677 11683 ipif_refhold_locked(ipif);
11678 11684 fallback_ipif = ipif;
11679 11685 }
11680 11686 continue;
11681 11687 }
11682 11688
11683 11689 /*
11684 11690 * Fine. Just assume the same net mask as the
11685 11691 * directly attached subnet interface is using.
11686 11692 */
11687 11693 ipif_refhold_locked(ipif);
11688 11694 mutex_exit(&ill->ill_lock);
11689 11695 rw_exit(&ipst->ips_ill_g_lock);
11690 11696 if (fallback_ipif != NULL)
11691 11697 ipif_refrele(fallback_ipif);
11692 11698 *ipifp = ipif;
11693 11699 return (ipif->ipif_net_mask);
11694 11700 }
11695 11701 }
11696 11702 mutex_exit(&ill->ill_lock);
11697 11703 }
11698 11704 rw_exit(&ipst->ips_ill_g_lock);
11699 11705
11700 11706 *ipifp = fallback_ipif;
11701 11707 return ((fallback_ipif != NULL) ?
11702 11708 fallback_ipif->ipif_net_mask : net_mask);
11703 11709 }
11704 11710
11705 11711 /*
11706 11712 * ip_sioctl_copyin_setup calls ip_wput_ioctl to process the IP_IOCTL ioctl.
11707 11713 */
11708 11714 static void
11709 11715 ip_wput_ioctl(queue_t *q, mblk_t *mp)
11710 11716 {
11711 11717 IOCP iocp;
11712 11718 ipft_t *ipft;
11713 11719 ipllc_t *ipllc;
11714 11720 mblk_t *mp1;
11715 11721 cred_t *cr;
11716 11722 int error = 0;
11717 11723 conn_t *connp;
11718 11724
11719 11725 ip1dbg(("ip_wput_ioctl"));
11720 11726 iocp = (IOCP)mp->b_rptr;
11721 11727 mp1 = mp->b_cont;
11722 11728 if (mp1 == NULL) {
11723 11729 iocp->ioc_error = EINVAL;
11724 11730 mp->b_datap->db_type = M_IOCNAK;
11725 11731 iocp->ioc_count = 0;
11726 11732 qreply(q, mp);
11727 11733 return;
11728 11734 }
11729 11735
11730 11736 /*
11731 11737 * These IOCTLs provide various control capabilities to
11732 11738 * upstream agents such as ULPs and processes. There
11733 11739 * are currently two such IOCTLs implemented. They
11734 11740 * are used by TCP to provide update information for
11735 11741 * existing IREs and to forcibly delete an IRE for a
11736 11742 * host that is not responding, thereby forcing an
11737 11743 * attempt at a new route.
11738 11744 */
11739 11745 iocp->ioc_error = EINVAL;
11740 11746 if (!pullupmsg(mp1, sizeof (ipllc->ipllc_cmd)))
11741 11747 goto done;
11742 11748
11743 11749 ipllc = (ipllc_t *)mp1->b_rptr;
11744 11750 for (ipft = ip_ioctl_ftbl; ipft->ipft_pfi; ipft++) {
11745 11751 if (ipllc->ipllc_cmd == ipft->ipft_cmd)
11746 11752 break;
11747 11753 }
11748 11754 /*
11749 11755 * prefer credential from mblk over ioctl;
11750 11756 * see ip_sioctl_copyin_setup
11751 11757 */
11752 11758 cr = msg_getcred(mp, NULL);
11753 11759 if (cr == NULL)
11754 11760 cr = iocp->ioc_cr;
11755 11761
11756 11762 /*
11757 11763 * Refhold the conn in case the request gets queued up in some lookup
11758 11764 */
11759 11765 ASSERT(CONN_Q(q));
11760 11766 connp = Q_TO_CONN(q);
11761 11767 CONN_INC_REF(connp);
11762 11768 CONN_INC_IOCTLREF(connp);
11763 11769 if (ipft->ipft_pfi &&
11764 11770 ((mp1->b_wptr - mp1->b_rptr) >= ipft->ipft_min_size ||
11765 11771 pullupmsg(mp1, ipft->ipft_min_size))) {
11766 11772 error = (*ipft->ipft_pfi)(q,
11767 11773 (ipft->ipft_flags & IPFT_F_SELF_REPLY) ? mp : mp1, cr);
11768 11774 }
11769 11775 if (ipft->ipft_flags & IPFT_F_SELF_REPLY) {
11770 11776 /*
11771 11777 * CONN_OPER_PENDING_DONE happens in the function called
11772 11778 * through ipft_pfi above.
11773 11779 */
11774 11780 return;
11775 11781 }
11776 11782
11777 11783 CONN_DEC_IOCTLREF(connp);
11778 11784 CONN_OPER_PENDING_DONE(connp);
11779 11785 if (ipft->ipft_flags & IPFT_F_NO_REPLY) {
11780 11786 freemsg(mp);
11781 11787 return;
11782 11788 }
11783 11789 iocp->ioc_error = error;
11784 11790
11785 11791 done:
11786 11792 mp->b_datap->db_type = M_IOCACK;
11787 11793 if (iocp->ioc_error)
11788 11794 iocp->ioc_count = 0;
11789 11795 qreply(q, mp);
11790 11796 }
11791 11797
11792 11798 /*
11793 11799 * Assign a unique id for the ipif. This is used by sctp_addr.c
11794 11800 * Note: remove if sctp_addr.c is redone to not shadow ill/ipif data structures.
11795 11801 */
11796 11802 static void
11797 11803 ipif_assign_seqid(ipif_t *ipif)
11798 11804 {
11799 11805 ip_stack_t *ipst = ipif->ipif_ill->ill_ipst;
11800 11806
11801 11807 ipif->ipif_seqid = atomic_add_64_nv(&ipst->ips_ipif_g_seqid, 1);
11802 11808 }
11803 11809
11804 11810 /*
11805 11811 * Clone the contents of `sipif' to `dipif'. Requires that both ipifs are
11806 11812 * administratively down (i.e., no DAD), of the same type, and locked. Note
11807 11813 * that the clone is complete -- including the seqid -- and the expectation is
11808 11814 * that the caller will either free or overwrite `sipif' before it's unlocked.
11809 11815 */
11810 11816 static void
11811 11817 ipif_clone(const ipif_t *sipif, ipif_t *dipif)
11812 11818 {
11813 11819 ASSERT(MUTEX_HELD(&sipif->ipif_ill->ill_lock));
11814 11820 ASSERT(MUTEX_HELD(&dipif->ipif_ill->ill_lock));
11815 11821 ASSERT(!(sipif->ipif_flags & (IPIF_UP|IPIF_DUPLICATE)));
11816 11822 ASSERT(!(dipif->ipif_flags & (IPIF_UP|IPIF_DUPLICATE)));
11817 11823 ASSERT(sipif->ipif_ire_type == dipif->ipif_ire_type);
11818 11824
11819 11825 dipif->ipif_flags = sipif->ipif_flags;
11820 11826 dipif->ipif_zoneid = sipif->ipif_zoneid;
11821 11827 dipif->ipif_v6subnet = sipif->ipif_v6subnet;
11822 11828 dipif->ipif_v6lcl_addr = sipif->ipif_v6lcl_addr;
11823 11829 dipif->ipif_v6net_mask = sipif->ipif_v6net_mask;
11824 11830 dipif->ipif_v6brd_addr = sipif->ipif_v6brd_addr;
11825 11831 dipif->ipif_v6pp_dst_addr = sipif->ipif_v6pp_dst_addr;
11826 11832
11827 11833 /*
11828 11834 * As per the comment atop the function, we assume that these sipif
11829 11835 * fields will be changed before sipif is unlocked.
11830 11836 */
11831 11837 dipif->ipif_seqid = sipif->ipif_seqid;
11832 11838 dipif->ipif_state_flags = sipif->ipif_state_flags;
11833 11839 }
11834 11840
11835 11841 /*
11836 11842 * Transfer the contents of `sipif' to `dipif', and then free (if `virgipif'
11837 11843 * is NULL) or overwrite `sipif' with `virgipif', which must be a virgin
11838 11844 * (unreferenced) ipif. Also, if `sipif' is used by the current xop, then
11839 11845 * transfer the xop to `dipif'. Requires that all ipifs are administratively
11840 11846 * down (i.e., no DAD), of the same type, and unlocked.
11841 11847 */
11842 11848 static void
11843 11849 ipif_transfer(ipif_t *sipif, ipif_t *dipif, ipif_t *virgipif)
11844 11850 {
11845 11851 ipsq_t *ipsq = sipif->ipif_ill->ill_phyint->phyint_ipsq;
11846 11852 ipxop_t *ipx = ipsq->ipsq_xop;
11847 11853
11848 11854 ASSERT(sipif != dipif);
11849 11855 ASSERT(sipif != virgipif);
11850 11856
11851 11857 /*
11852 11858 * Grab all of the locks that protect the ipif in a defined order.
11853 11859 */
11854 11860 GRAB_ILL_LOCKS(sipif->ipif_ill, dipif->ipif_ill);
11855 11861
11856 11862 ipif_clone(sipif, dipif);
11857 11863 if (virgipif != NULL) {
11858 11864 ipif_clone(virgipif, sipif);
11859 11865 mi_free(virgipif);
11860 11866 }
11861 11867
11862 11868 RELEASE_ILL_LOCKS(sipif->ipif_ill, dipif->ipif_ill);
11863 11869
11864 11870 /*
11865 11871 * Transfer ownership of the current xop, if necessary.
11866 11872 */
11867 11873 if (ipx->ipx_current_ipif == sipif) {
11868 11874 ASSERT(ipx->ipx_pending_ipif == NULL);
11869 11875 mutex_enter(&ipx->ipx_lock);
11870 11876 ipx->ipx_current_ipif = dipif;
11871 11877 mutex_exit(&ipx->ipx_lock);
11872 11878 }
11873 11879
11874 11880 if (virgipif == NULL)
11875 11881 mi_free(sipif);
11876 11882 }
11877 11883
11878 11884 /*
11879 11885 * checks if:
11880 11886 * - <ill_name>:<ipif_id> is at most LIFNAMSIZ - 1 and
11881 11887 * - logical interface is within the allowed range
11882 11888 */
11883 11889 static int
11884 11890 is_lifname_valid(ill_t *ill, unsigned int ipif_id)
11885 11891 {
11886 11892 if (snprintf(NULL, 0, "%s:%d", ill->ill_name, ipif_id) >= LIFNAMSIZ)
11887 11893 return (ENAMETOOLONG);
11888 11894
11889 11895 if (ipif_id >= ill->ill_ipst->ips_ip_addrs_per_if)
11890 11896 return (ERANGE);
11891 11897 return (0);
11892 11898 }
11893 11899
11894 11900 /*
11895 11901 * Insert the ipif, so that the list of ipifs on the ill will be sorted
11896 11902 * with respect to ipif_id. Note that an ipif with an ipif_id of -1 will
11897 11903 * be inserted into the first space available in the list. The value of
11898 11904 * ipif_id will then be set to the appropriate value for its position.
11899 11905 */
11900 11906 static int
11901 11907 ipif_insert(ipif_t *ipif, boolean_t acquire_g_lock)
11902 11908 {
11903 11909 ill_t *ill;
11904 11910 ipif_t *tipif;
11905 11911 ipif_t **tipifp;
11906 11912 int id, err;
11907 11913 ip_stack_t *ipst;
11908 11914
11909 11915 ASSERT(ipif->ipif_ill->ill_net_type == IRE_LOOPBACK ||
11910 11916 IAM_WRITER_IPIF(ipif));
11911 11917
11912 11918 ill = ipif->ipif_ill;
11913 11919 ASSERT(ill != NULL);
11914 11920 ipst = ill->ill_ipst;
11915 11921
11916 11922 /*
11917 11923 * In the case of lo0:0 we already hold the ill_g_lock.
11918 11924 * ill_lookup_on_name (acquires ill_g_lock) -> ipif_allocate ->
11919 11925 * ipif_insert.
11920 11926 */
11921 11927 if (acquire_g_lock)
11922 11928 rw_enter(&ipst->ips_ill_g_lock, RW_WRITER);
11923 11929 mutex_enter(&ill->ill_lock);
11924 11930 id = ipif->ipif_id;
11925 11931 tipifp = &(ill->ill_ipif);
11926 11932 if (id == -1) { /* need to find a real id */
11927 11933 id = 0;
11928 11934 while ((tipif = *tipifp) != NULL) {
11929 11935 ASSERT(tipif->ipif_id >= id);
11930 11936 if (tipif->ipif_id != id)
11931 11937 break; /* non-consecutive id */
11932 11938 id++;
11933 11939 tipifp = &(tipif->ipif_next);
11934 11940 }
11935 11941 if ((err = is_lifname_valid(ill, id)) != 0) {
11936 11942 mutex_exit(&ill->ill_lock);
11937 11943 if (acquire_g_lock)
11938 11944 rw_exit(&ipst->ips_ill_g_lock);
11939 11945 return (err);
11940 11946 }
11941 11947 ipif->ipif_id = id; /* assign new id */
11942 11948 } else if ((err = is_lifname_valid(ill, id)) == 0) {
11943 11949 /* we have a real id; insert ipif in the right place */
11944 11950 while ((tipif = *tipifp) != NULL) {
11945 11951 ASSERT(tipif->ipif_id != id);
11946 11952 if (tipif->ipif_id > id)
11947 11953 break; /* found correct location */
11948 11954 tipifp = &(tipif->ipif_next);
11949 11955 }
11950 11956 } else {
11951 11957 mutex_exit(&ill->ill_lock);
11952 11958 if (acquire_g_lock)
11953 11959 rw_exit(&ipst->ips_ill_g_lock);
11954 11960 return (err);
11955 11961 }
11956 11962
11957 11963 ASSERT(tipifp != &(ill->ill_ipif) || id == 0);
11958 11964
11959 11965 ipif->ipif_next = tipif;
11960 11966 *tipifp = ipif;
11961 11967 mutex_exit(&ill->ill_lock);
11962 11968 if (acquire_g_lock)
11963 11969 rw_exit(&ipst->ips_ill_g_lock);
11964 11970
11965 11971 return (0);
11966 11972 }
11967 11973
11968 11974 static void
11969 11975 ipif_remove(ipif_t *ipif)
11970 11976 {
11971 11977 ipif_t **ipifp;
11972 11978 ill_t *ill = ipif->ipif_ill;
11973 11979
11974 11980 ASSERT(RW_WRITE_HELD(&ill->ill_ipst->ips_ill_g_lock));
11975 11981
11976 11982 mutex_enter(&ill->ill_lock);
11977 11983 ipifp = &ill->ill_ipif;
11978 11984 for (; *ipifp != NULL; ipifp = &ipifp[0]->ipif_next) {
11979 11985 if (*ipifp == ipif) {
11980 11986 *ipifp = ipif->ipif_next;
11981 11987 break;
11982 11988 }
11983 11989 }
11984 11990 mutex_exit(&ill->ill_lock);
11985 11991 }
11986 11992
11987 11993 /*
11988 11994 * Allocate and initialize a new interface control structure. (Always
11989 11995 * called as writer.)
11990 11996 * When ipif_allocate() is called from ip_ll_subnet_defaults, the ill
11991 11997 * is not part of the global linked list of ills. ipif_seqid is unique
11992 11998 * in the system and to preserve the uniqueness, it is assigned only
11993 11999 * when ill becomes part of the global list. At that point ill will
11994 12000 * have a name. If it doesn't get assigned here, it will get assigned
11995 12001 * in ipif_set_values() as part of SIOCSLIFNAME processing.
11996 12002 * Aditionally, if we come here from ip_ll_subnet_defaults, we don't set
11997 12003 * the interface flags or any other information from the DL_INFO_ACK for
11998 12004 * DL_STYLE2 drivers (initialize == B_FALSE), since we won't have them at
11999 12005 * this point. The flags etc. will be set in ip_ll_subnet_defaults when the
12000 12006 * second DL_INFO_ACK comes in from the driver.
12001 12007 */
12002 12008 static ipif_t *
12003 12009 ipif_allocate(ill_t *ill, int id, uint_t ire_type, boolean_t initialize,
12004 12010 boolean_t insert, int *errorp)
12005 12011 {
12006 12012 int err;
12007 12013 ipif_t *ipif;
12008 12014 ip_stack_t *ipst = ill->ill_ipst;
12009 12015
12010 12016 ip1dbg(("ipif_allocate(%s:%d ill %p)\n",
12011 12017 ill->ill_name, id, (void *)ill));
12012 12018 ASSERT(ire_type == IRE_LOOPBACK || IAM_WRITER_ILL(ill));
12013 12019
12014 12020 if (errorp != NULL)
12015 12021 *errorp = 0;
12016 12022
12017 12023 if ((ipif = mi_alloc(sizeof (ipif_t), BPRI_MED)) == NULL) {
12018 12024 if (errorp != NULL)
12019 12025 *errorp = ENOMEM;
12020 12026 return (NULL);
12021 12027 }
12022 12028 *ipif = ipif_zero; /* start clean */
12023 12029
12024 12030 ipif->ipif_ill = ill;
12025 12031 ipif->ipif_id = id; /* could be -1 */
12026 12032 /*
12027 12033 * Inherit the zoneid from the ill; for the shared stack instance
12028 12034 * this is always the global zone
12029 12035 */
12030 12036 ipif->ipif_zoneid = ill->ill_zoneid;
12031 12037
12032 12038 ipif->ipif_refcnt = 0;
12033 12039
12034 12040 if (insert) {
12035 12041 if ((err = ipif_insert(ipif, ire_type != IRE_LOOPBACK)) != 0) {
12036 12042 mi_free(ipif);
12037 12043 if (errorp != NULL)
12038 12044 *errorp = err;
12039 12045 return (NULL);
12040 12046 }
12041 12047 /* -1 id should have been replaced by real id */
12042 12048 id = ipif->ipif_id;
12043 12049 ASSERT(id >= 0);
12044 12050 }
12045 12051
12046 12052 if (ill->ill_name[0] != '\0')
12047 12053 ipif_assign_seqid(ipif);
12048 12054
12049 12055 /*
12050 12056 * If this is the zeroth ipif on the IPMP ill, create the illgrp
12051 12057 * (which must not exist yet because the zeroth ipif is created once
12052 12058 * per ill). However, do not not link it to the ipmp_grp_t until
12053 12059 * I_PLINK is called; see ip_sioctl_plink_ipmp() for details.
12054 12060 */
12055 12061 if (id == 0 && IS_IPMP(ill)) {
12056 12062 if (ipmp_illgrp_create(ill) == NULL) {
12057 12063 if (insert) {
12058 12064 rw_enter(&ipst->ips_ill_g_lock, RW_WRITER);
12059 12065 ipif_remove(ipif);
12060 12066 rw_exit(&ipst->ips_ill_g_lock);
12061 12067 }
12062 12068 mi_free(ipif);
12063 12069 if (errorp != NULL)
12064 12070 *errorp = ENOMEM;
12065 12071 return (NULL);
12066 12072 }
12067 12073 }
12068 12074
12069 12075 /*
12070 12076 * We grab ill_lock to protect the flag changes. The ipif is still
12071 12077 * not up and can't be looked up until the ioctl completes and the
12072 12078 * IPIF_CHANGING flag is cleared.
12073 12079 */
12074 12080 mutex_enter(&ill->ill_lock);
12075 12081
12076 12082 ipif->ipif_ire_type = ire_type;
12077 12083
12078 12084 if (ipif->ipif_isv6) {
12079 12085 ill->ill_flags |= ILLF_IPV6;
12080 12086 } else {
12081 12087 ipaddr_t inaddr_any = INADDR_ANY;
12082 12088
12083 12089 ill->ill_flags |= ILLF_IPV4;
12084 12090
12085 12091 /* Keep the IN6_IS_ADDR_V4MAPPED assertions happy */
12086 12092 IN6_IPADDR_TO_V4MAPPED(inaddr_any,
12087 12093 &ipif->ipif_v6lcl_addr);
12088 12094 IN6_IPADDR_TO_V4MAPPED(inaddr_any,
12089 12095 &ipif->ipif_v6subnet);
12090 12096 IN6_IPADDR_TO_V4MAPPED(inaddr_any,
12091 12097 &ipif->ipif_v6net_mask);
12092 12098 IN6_IPADDR_TO_V4MAPPED(inaddr_any,
12093 12099 &ipif->ipif_v6brd_addr);
12094 12100 IN6_IPADDR_TO_V4MAPPED(inaddr_any,
12095 12101 &ipif->ipif_v6pp_dst_addr);
12096 12102 }
12097 12103
12098 12104 /*
12099 12105 * Don't set the interface flags etc. now, will do it in
12100 12106 * ip_ll_subnet_defaults.
12101 12107 */
12102 12108 if (!initialize)
12103 12109 goto out;
12104 12110
12105 12111 /*
12106 12112 * NOTE: The IPMP meta-interface is special-cased because it starts
12107 12113 * with no underlying interfaces (and thus an unknown broadcast
12108 12114 * address length), but all interfaces that can be placed into an IPMP
12109 12115 * group are required to be broadcast-capable.
12110 12116 */
12111 12117 if (ill->ill_bcast_addr_length != 0 || IS_IPMP(ill)) {
12112 12118 /*
12113 12119 * Later detect lack of DLPI driver multicast capability by
12114 12120 * catching DL_ENABMULTI_REQ errors in ip_rput_dlpi().
12115 12121 */
12116 12122 ill->ill_flags |= ILLF_MULTICAST;
12117 12123 if (!ipif->ipif_isv6)
12118 12124 ipif->ipif_flags |= IPIF_BROADCAST;
12119 12125 } else {
12120 12126 if (ill->ill_net_type != IRE_LOOPBACK) {
12121 12127 if (ipif->ipif_isv6)
12122 12128 /*
12123 12129 * Note: xresolv interfaces will eventually need
12124 12130 * NOARP set here as well, but that will require
12125 12131 * those external resolvers to have some
12126 12132 * knowledge of that flag and act appropriately.
12127 12133 * Not to be changed at present.
12128 12134 */
12129 12135 ill->ill_flags |= ILLF_NONUD;
12130 12136 else
12131 12137 ill->ill_flags |= ILLF_NOARP;
12132 12138 }
12133 12139 if (ill->ill_phys_addr_length == 0) {
12134 12140 if (IS_VNI(ill)) {
12135 12141 ipif->ipif_flags |= IPIF_NOXMIT;
12136 12142 } else {
12137 12143 /* pt-pt supports multicast. */
12138 12144 ill->ill_flags |= ILLF_MULTICAST;
12139 12145 if (ill->ill_net_type != IRE_LOOPBACK)
12140 12146 ipif->ipif_flags |= IPIF_POINTOPOINT;
12141 12147 }
12142 12148 }
12143 12149 }
12144 12150 out:
12145 12151 mutex_exit(&ill->ill_lock);
12146 12152 return (ipif);
12147 12153 }
12148 12154
12149 12155 /*
12150 12156 * Remove the neighbor cache entries associated with this logical
12151 12157 * interface.
12152 12158 */
12153 12159 int
12154 12160 ipif_arp_down(ipif_t *ipif)
12155 12161 {
12156 12162 ill_t *ill = ipif->ipif_ill;
12157 12163 int err = 0;
12158 12164
12159 12165 ip1dbg(("ipif_arp_down(%s:%u)\n", ill->ill_name, ipif->ipif_id));
12160 12166 ASSERT(IAM_WRITER_IPIF(ipif));
12161 12167
12162 12168 DTRACE_PROBE3(ipif__downup, char *, "ipif_arp_down",
12163 12169 ill_t *, ill, ipif_t *, ipif);
12164 12170 ipif_nce_down(ipif);
12165 12171
12166 12172 /*
12167 12173 * If this is the last ipif that is going down and there are no
12168 12174 * duplicate addresses we may yet attempt to re-probe, then we need to
12169 12175 * clean up ARP completely.
12170 12176 */
12171 12177 if (ill->ill_ipif_up_count == 0 && ill->ill_ipif_dup_count == 0 &&
12172 12178 !ill->ill_logical_down && ill->ill_net_type == IRE_IF_RESOLVER) {
12173 12179 /*
12174 12180 * If this was the last ipif on an IPMP interface, purge any
12175 12181 * static ARP entries associated with it.
12176 12182 */
12177 12183 if (IS_IPMP(ill))
12178 12184 ipmp_illgrp_refresh_arpent(ill->ill_grp);
12179 12185
12180 12186 /* UNBIND, DETACH */
12181 12187 err = arp_ll_down(ill);
12182 12188 }
12183 12189
12184 12190 return (err);
12185 12191 }
12186 12192
12187 12193 /*
12188 12194 * Get the resolver set up for a new IP address. (Always called as writer.)
12189 12195 * Called both for IPv4 and IPv6 interfaces, though it only does some
12190 12196 * basic DAD related initialization for IPv6. Honors ILLF_NOARP.
12191 12197 *
12192 12198 * The enumerated value res_act tunes the behavior:
12193 12199 * * Res_act_initial: set up all the resolver structures for a new
12194 12200 * IP address.
12195 12201 * * Res_act_defend: tell ARP that it needs to send a single gratuitous
12196 12202 * ARP message in defense of the address.
12197 12203 * * Res_act_rebind: tell ARP to change the hardware address for an IP
12198 12204 * address (and issue gratuitous ARPs). Used by ipmp_ill_bind_ipif().
12199 12205 *
12200 12206 * Returns zero on success, or an errno upon failure.
12201 12207 */
12202 12208 int
12203 12209 ipif_resolver_up(ipif_t *ipif, enum ip_resolver_action res_act)
12204 12210 {
12205 12211 ill_t *ill = ipif->ipif_ill;
12206 12212 int err;
12207 12213 boolean_t was_dup;
12208 12214
12209 12215 ip1dbg(("ipif_resolver_up(%s:%u) flags 0x%x\n",
12210 12216 ill->ill_name, ipif->ipif_id, (uint_t)ipif->ipif_flags));
12211 12217 ASSERT(IAM_WRITER_IPIF(ipif));
12212 12218
12213 12219 was_dup = B_FALSE;
12214 12220 if (res_act == Res_act_initial) {
12215 12221 ipif->ipif_addr_ready = 0;
12216 12222 /*
12217 12223 * We're bringing an interface up here. There's no way that we
12218 12224 * should need to shut down ARP now.
12219 12225 */
12220 12226 mutex_enter(&ill->ill_lock);
12221 12227 if (ipif->ipif_flags & IPIF_DUPLICATE) {
12222 12228 ipif->ipif_flags &= ~IPIF_DUPLICATE;
12223 12229 ill->ill_ipif_dup_count--;
12224 12230 was_dup = B_TRUE;
12225 12231 }
12226 12232 mutex_exit(&ill->ill_lock);
12227 12233 }
12228 12234 if (ipif->ipif_recovery_id != 0)
12229 12235 (void) untimeout(ipif->ipif_recovery_id);
12230 12236 ipif->ipif_recovery_id = 0;
12231 12237 if (ill->ill_net_type != IRE_IF_RESOLVER) {
12232 12238 ipif->ipif_addr_ready = 1;
12233 12239 return (0);
12234 12240 }
12235 12241 /* NDP will set the ipif_addr_ready flag when it's ready */
12236 12242 if (ill->ill_isv6)
12237 12243 return (0);
12238 12244
12239 12245 err = ipif_arp_up(ipif, res_act, was_dup);
12240 12246 return (err);
12241 12247 }
12242 12248
12243 12249 /*
12244 12250 * This routine restarts IPv4/IPv6 duplicate address detection (DAD)
12245 12251 * when a link has just gone back up.
12246 12252 */
12247 12253 static void
12248 12254 ipif_nce_start_dad(ipif_t *ipif)
12249 12255 {
12250 12256 ncec_t *ncec;
12251 12257 ill_t *ill = ipif->ipif_ill;
12252 12258 boolean_t isv6 = ill->ill_isv6;
12253 12259
12254 12260 if (isv6) {
12255 12261 ncec = ncec_lookup_illgrp_v6(ipif->ipif_ill,
12256 12262 &ipif->ipif_v6lcl_addr);
12257 12263 } else {
12258 12264 ipaddr_t v4addr;
12259 12265
12260 12266 if (ill->ill_net_type != IRE_IF_RESOLVER ||
12261 12267 (ipif->ipif_flags & IPIF_UNNUMBERED) ||
12262 12268 ipif->ipif_lcl_addr == INADDR_ANY) {
12263 12269 /*
12264 12270 * If we can't contact ARP for some reason,
12265 12271 * that's not really a problem. Just send
12266 12272 * out the routing socket notification that
12267 12273 * DAD completion would have done, and continue.
12268 12274 */
12269 12275 ipif_mask_reply(ipif);
12270 12276 ipif_up_notify(ipif);
12271 12277 ipif->ipif_addr_ready = 1;
12272 12278 return;
12273 12279 }
12274 12280
12275 12281 IN6_V4MAPPED_TO_IPADDR(&ipif->ipif_v6lcl_addr, v4addr);
12276 12282 ncec = ncec_lookup_illgrp_v4(ipif->ipif_ill, &v4addr);
12277 12283 }
12278 12284
12279 12285 if (ncec == NULL) {
12280 12286 ip1dbg(("couldn't find ncec for ipif %p leaving !ready\n",
12281 12287 (void *)ipif));
12282 12288 return;
12283 12289 }
12284 12290 if (!nce_restart_dad(ncec)) {
12285 12291 /*
12286 12292 * If we can't restart DAD for some reason, that's not really a
12287 12293 * problem. Just send out the routing socket notification that
12288 12294 * DAD completion would have done, and continue.
12289 12295 */
12290 12296 ipif_up_notify(ipif);
12291 12297 ipif->ipif_addr_ready = 1;
12292 12298 }
12293 12299 ncec_refrele(ncec);
12294 12300 }
12295 12301
12296 12302 /*
12297 12303 * Restart duplicate address detection on all interfaces on the given ill.
12298 12304 *
12299 12305 * This is called when an interface transitions from down to up
12300 12306 * (DL_NOTE_LINK_UP) or up to down (DL_NOTE_LINK_DOWN).
12301 12307 *
12302 12308 * Note that since the underlying physical link has transitioned, we must cause
12303 12309 * at least one routing socket message to be sent here, either via DAD
12304 12310 * completion or just by default on the first ipif. (If we don't do this, then
12305 12311 * in.mpathd will see long delays when doing link-based failure recovery.)
12306 12312 */
12307 12313 void
12308 12314 ill_restart_dad(ill_t *ill, boolean_t went_up)
12309 12315 {
12310 12316 ipif_t *ipif;
12311 12317
12312 12318 if (ill == NULL)
12313 12319 return;
12314 12320
12315 12321 /*
12316 12322 * If layer two doesn't support duplicate address detection, then just
12317 12323 * send the routing socket message now and be done with it.
12318 12324 */
12319 12325 if (!ill->ill_isv6 && arp_no_defense) {
12320 12326 ip_rts_ifmsg(ill->ill_ipif, RTSQ_DEFAULT);
12321 12327 return;
12322 12328 }
12323 12329
12324 12330 for (ipif = ill->ill_ipif; ipif != NULL; ipif = ipif->ipif_next) {
12325 12331 if (went_up) {
12326 12332
12327 12333 if (ipif->ipif_flags & IPIF_UP) {
12328 12334 ipif_nce_start_dad(ipif);
12329 12335 } else if (ipif->ipif_flags & IPIF_DUPLICATE) {
12330 12336 /*
12331 12337 * kick off the bring-up process now.
12332 12338 */
12333 12339 ipif_do_recovery(ipif);
12334 12340 } else {
12335 12341 /*
12336 12342 * Unfortunately, the first ipif is "special"
12337 12343 * and represents the underlying ill in the
12338 12344 * routing socket messages. Thus, when this
12339 12345 * one ipif is down, we must still notify so
12340 12346 * that the user knows the IFF_RUNNING status
12341 12347 * change. (If the first ipif is up, then
12342 12348 * we'll handle eventual routing socket
12343 12349 * notification via DAD completion.)
12344 12350 */
12345 12351 if (ipif == ill->ill_ipif) {
12346 12352 ip_rts_ifmsg(ill->ill_ipif,
12347 12353 RTSQ_DEFAULT);
12348 12354 }
12349 12355 }
12350 12356 } else {
12351 12357 /*
12352 12358 * After link down, we'll need to send a new routing
12353 12359 * message when the link comes back, so clear
12354 12360 * ipif_addr_ready.
12355 12361 */
12356 12362 ipif->ipif_addr_ready = 0;
12357 12363 }
12358 12364 }
12359 12365
12360 12366 /*
12361 12367 * If we've torn down links, then notify the user right away.
12362 12368 */
12363 12369 if (!went_up)
12364 12370 ip_rts_ifmsg(ill->ill_ipif, RTSQ_DEFAULT);
12365 12371 }
12366 12372
12367 12373 static void
12368 12374 ipsq_delete(ipsq_t *ipsq)
12369 12375 {
12370 12376 ipxop_t *ipx = ipsq->ipsq_xop;
12371 12377
12372 12378 ipsq->ipsq_ipst = NULL;
12373 12379 ASSERT(ipsq->ipsq_phyint == NULL);
12374 12380 ASSERT(ipsq->ipsq_xop != NULL);
12375 12381 ASSERT(ipsq->ipsq_xopq_mphead == NULL && ipx->ipx_mphead == NULL);
12376 12382 ASSERT(ipx->ipx_pending_mp == NULL);
12377 12383 kmem_free(ipsq, sizeof (ipsq_t));
12378 12384 }
12379 12385
12380 12386 static int
12381 12387 ill_up_ipifs_on_ill(ill_t *ill, queue_t *q, mblk_t *mp)
12382 12388 {
12383 12389 int err = 0;
12384 12390 ipif_t *ipif;
12385 12391
12386 12392 if (ill == NULL)
12387 12393 return (0);
12388 12394
12389 12395 ASSERT(IAM_WRITER_ILL(ill));
12390 12396 ill->ill_up_ipifs = B_TRUE;
12391 12397 for (ipif = ill->ill_ipif; ipif != NULL; ipif = ipif->ipif_next) {
12392 12398 if (ipif->ipif_was_up) {
12393 12399 if (!(ipif->ipif_flags & IPIF_UP))
12394 12400 err = ipif_up(ipif, q, mp);
12395 12401 ipif->ipif_was_up = B_FALSE;
12396 12402 if (err != 0) {
12397 12403 ASSERT(err == EINPROGRESS);
12398 12404 return (err);
12399 12405 }
12400 12406 }
12401 12407 }
12402 12408 ill->ill_up_ipifs = B_FALSE;
12403 12409 return (0);
12404 12410 }
12405 12411
12406 12412 /*
12407 12413 * This function is called to bring up all the ipifs that were up before
12408 12414 * bringing the ill down via ill_down_ipifs().
12409 12415 */
12410 12416 int
12411 12417 ill_up_ipifs(ill_t *ill, queue_t *q, mblk_t *mp)
12412 12418 {
12413 12419 int err;
12414 12420
12415 12421 ASSERT(IAM_WRITER_ILL(ill));
12416 12422
12417 12423 if (ill->ill_replumbing) {
12418 12424 ill->ill_replumbing = 0;
12419 12425 /*
12420 12426 * Send down REPLUMB_DONE notification followed by the
12421 12427 * BIND_REQ on the arp stream.
12422 12428 */
12423 12429 if (!ill->ill_isv6)
12424 12430 arp_send_replumb_conf(ill);
12425 12431 }
12426 12432 err = ill_up_ipifs_on_ill(ill->ill_phyint->phyint_illv4, q, mp);
12427 12433 if (err != 0)
12428 12434 return (err);
12429 12435
12430 12436 return (ill_up_ipifs_on_ill(ill->ill_phyint->phyint_illv6, q, mp));
12431 12437 }
12432 12438
12433 12439 /*
12434 12440 * Bring down any IPIF_UP ipifs on ill. If "logical" is B_TRUE, we bring
12435 12441 * down the ipifs without sending DL_UNBIND_REQ to the driver.
12436 12442 */
12437 12443 static void
12438 12444 ill_down_ipifs(ill_t *ill, boolean_t logical)
12439 12445 {
12440 12446 ipif_t *ipif;
12441 12447
12442 12448 ASSERT(IAM_WRITER_ILL(ill));
12443 12449
12444 12450 for (ipif = ill->ill_ipif; ipif != NULL; ipif = ipif->ipif_next) {
12445 12451 /*
12446 12452 * We go through the ipif_down logic even if the ipif
12447 12453 * is already down, since routes can be added based
12448 12454 * on down ipifs. Going through ipif_down once again
12449 12455 * will delete any IREs created based on these routes.
12450 12456 */
12451 12457 if (ipif->ipif_flags & IPIF_UP)
12452 12458 ipif->ipif_was_up = B_TRUE;
12453 12459
12454 12460 if (logical) {
12455 12461 (void) ipif_logical_down(ipif, NULL, NULL);
12456 12462 ipif_non_duplicate(ipif);
12457 12463 (void) ipif_down_tail(ipif);
12458 12464 } else {
12459 12465 (void) ipif_down(ipif, NULL, NULL);
12460 12466 }
12461 12467 }
12462 12468 }
12463 12469
12464 12470 /*
12465 12471 * Redo source address selection. This makes IXAF_VERIFY_SOURCE take
12466 12472 * a look again at valid source addresses.
12467 12473 * This should be called each time after the set of source addresses has been
12468 12474 * changed.
12469 12475 */
12470 12476 void
12471 12477 ip_update_source_selection(ip_stack_t *ipst)
12472 12478 {
12473 12479 /* We skip past SRC_GENERATION_VERIFY */
12474 12480 if (atomic_add_32_nv(&ipst->ips_src_generation, 1) ==
12475 12481 SRC_GENERATION_VERIFY)
12476 12482 atomic_add_32(&ipst->ips_src_generation, 1);
12477 12483 }
12478 12484
12479 12485 /*
12480 12486 * Finish the group join started in ip_sioctl_groupname().
12481 12487 */
12482 12488 /* ARGSUSED */
12483 12489 static void
12484 12490 ip_join_illgrps(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *dummy)
12485 12491 {
12486 12492 ill_t *ill = q->q_ptr;
12487 12493 phyint_t *phyi = ill->ill_phyint;
12488 12494 ipmp_grp_t *grp = phyi->phyint_grp;
12489 12495 ip_stack_t *ipst = ill->ill_ipst;
12490 12496
12491 12497 /* IS_UNDER_IPMP() won't work until ipmp_ill_join_illgrp() is called */
12492 12498 ASSERT(!IS_IPMP(ill) && grp != NULL);
12493 12499 ASSERT(IAM_WRITER_IPSQ(ipsq));
12494 12500
12495 12501 if (phyi->phyint_illv4 != NULL) {
12496 12502 rw_enter(&ipst->ips_ipmp_lock, RW_WRITER);
12497 12503 VERIFY(grp->gr_pendv4-- > 0);
12498 12504 rw_exit(&ipst->ips_ipmp_lock);
12499 12505 ipmp_ill_join_illgrp(phyi->phyint_illv4, grp->gr_v4);
12500 12506 }
12501 12507 if (phyi->phyint_illv6 != NULL) {
12502 12508 rw_enter(&ipst->ips_ipmp_lock, RW_WRITER);
12503 12509 VERIFY(grp->gr_pendv6-- > 0);
12504 12510 rw_exit(&ipst->ips_ipmp_lock);
12505 12511 ipmp_ill_join_illgrp(phyi->phyint_illv6, grp->gr_v6);
12506 12512 }
12507 12513 freemsg(mp);
12508 12514 }
12509 12515
12510 12516 /*
12511 12517 * Process an SIOCSLIFGROUPNAME request.
12512 12518 */
12513 12519 /* ARGSUSED */
12514 12520 int
12515 12521 ip_sioctl_groupname(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
12516 12522 ip_ioctl_cmd_t *ipip, void *ifreq)
12517 12523 {
12518 12524 struct lifreq *lifr = ifreq;
12519 12525 ill_t *ill = ipif->ipif_ill;
12520 12526 ip_stack_t *ipst = ill->ill_ipst;
12521 12527 phyint_t *phyi = ill->ill_phyint;
12522 12528 ipmp_grp_t *grp = phyi->phyint_grp;
12523 12529 mblk_t *ipsq_mp;
12524 12530 int err = 0;
12525 12531
12526 12532 /*
12527 12533 * Note that phyint_grp can only change here, where we're exclusive.
12528 12534 */
12529 12535 ASSERT(IAM_WRITER_ILL(ill));
12530 12536
12531 12537 if (ipif->ipif_id != 0 || ill->ill_usesrc_grp_next != NULL ||
12532 12538 (phyi->phyint_flags & PHYI_VIRTUAL))
12533 12539 return (EINVAL);
12534 12540
12535 12541 lifr->lifr_groupname[LIFGRNAMSIZ - 1] = '\0';
12536 12542
12537 12543 rw_enter(&ipst->ips_ipmp_lock, RW_WRITER);
12538 12544
12539 12545 /*
12540 12546 * If the name hasn't changed, there's nothing to do.
12541 12547 */
12542 12548 if (grp != NULL && strcmp(grp->gr_name, lifr->lifr_groupname) == 0)
12543 12549 goto unlock;
12544 12550
12545 12551 /*
12546 12552 * Handle requests to rename an IPMP meta-interface.
12547 12553 *
12548 12554 * Note that creation of the IPMP meta-interface is handled in
12549 12555 * userland through the standard plumbing sequence. As part of the
12550 12556 * plumbing the IPMP meta-interface, its initial groupname is set to
12551 12557 * the name of the interface (see ipif_set_values_tail()).
12552 12558 */
12553 12559 if (IS_IPMP(ill)) {
12554 12560 err = ipmp_grp_rename(grp, lifr->lifr_groupname);
12555 12561 goto unlock;
12556 12562 }
12557 12563
12558 12564 /*
12559 12565 * Handle requests to add or remove an IP interface from a group.
12560 12566 */
12561 12567 if (lifr->lifr_groupname[0] != '\0') { /* add */
12562 12568 /*
12563 12569 * Moves are handled by first removing the interface from
12564 12570 * its existing group, and then adding it to another group.
12565 12571 * So, fail if it's already in a group.
12566 12572 */
12567 12573 if (IS_UNDER_IPMP(ill)) {
12568 12574 err = EALREADY;
12569 12575 goto unlock;
12570 12576 }
12571 12577
12572 12578 grp = ipmp_grp_lookup(lifr->lifr_groupname, ipst);
12573 12579 if (grp == NULL) {
12574 12580 err = ENOENT;
12575 12581 goto unlock;
12576 12582 }
12577 12583
12578 12584 /*
12579 12585 * Check if the phyint and its ills are suitable for
12580 12586 * inclusion into the group.
12581 12587 */
12582 12588 if ((err = ipmp_grp_vet_phyint(grp, phyi)) != 0)
12583 12589 goto unlock;
12584 12590
12585 12591 /*
12586 12592 * Checks pass; join the group, and enqueue the remaining
12587 12593 * illgrp joins for when we've become part of the group xop
12588 12594 * and are exclusive across its IPSQs. Since qwriter_ip()
12589 12595 * requires an mblk_t to scribble on, and since `mp' will be
12590 12596 * freed as part of completing the ioctl, allocate another.
12591 12597 */
12592 12598 if ((ipsq_mp = allocb(0, BPRI_MED)) == NULL) {
12593 12599 err = ENOMEM;
12594 12600 goto unlock;
12595 12601 }
12596 12602
12597 12603 /*
12598 12604 * Before we drop ipmp_lock, bump gr_pend* to ensure that the
12599 12605 * IPMP meta-interface ills needed by `phyi' cannot go away
12600 12606 * before ip_join_illgrps() is called back. See the comments
12601 12607 * in ip_sioctl_plink_ipmp() for more.
12602 12608 */
12603 12609 if (phyi->phyint_illv4 != NULL)
12604 12610 grp->gr_pendv4++;
12605 12611 if (phyi->phyint_illv6 != NULL)
12606 12612 grp->gr_pendv6++;
12607 12613
12608 12614 rw_exit(&ipst->ips_ipmp_lock);
12609 12615
12610 12616 ipmp_phyint_join_grp(phyi, grp);
12611 12617 ill_refhold(ill);
12612 12618 qwriter_ip(ill, ill->ill_rq, ipsq_mp, ip_join_illgrps,
12613 12619 SWITCH_OP, B_FALSE);
12614 12620 return (0);
12615 12621 } else {
12616 12622 /*
12617 12623 * Request to remove the interface from a group. If the
12618 12624 * interface is not in a group, this trivially succeeds.
12619 12625 */
12620 12626 rw_exit(&ipst->ips_ipmp_lock);
12621 12627 if (IS_UNDER_IPMP(ill))
12622 12628 ipmp_phyint_leave_grp(phyi);
12623 12629 return (0);
12624 12630 }
12625 12631 unlock:
12626 12632 rw_exit(&ipst->ips_ipmp_lock);
12627 12633 return (err);
12628 12634 }
12629 12635
12630 12636 /*
12631 12637 * Process an SIOCGLIFBINDING request.
12632 12638 */
12633 12639 /* ARGSUSED */
12634 12640 int
12635 12641 ip_sioctl_get_binding(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
12636 12642 ip_ioctl_cmd_t *ipip, void *ifreq)
12637 12643 {
12638 12644 ill_t *ill;
12639 12645 struct lifreq *lifr = ifreq;
12640 12646 ip_stack_t *ipst = ipif->ipif_ill->ill_ipst;
12641 12647
12642 12648 if (!IS_IPMP(ipif->ipif_ill))
12643 12649 return (EINVAL);
12644 12650
12645 12651 rw_enter(&ipst->ips_ipmp_lock, RW_READER);
12646 12652 if ((ill = ipif->ipif_bound_ill) == NULL)
12647 12653 lifr->lifr_binding[0] = '\0';
12648 12654 else
12649 12655 (void) strlcpy(lifr->lifr_binding, ill->ill_name, LIFNAMSIZ);
12650 12656 rw_exit(&ipst->ips_ipmp_lock);
12651 12657 return (0);
12652 12658 }
12653 12659
12654 12660 /*
12655 12661 * Process an SIOCGLIFGROUPNAME request.
12656 12662 */
12657 12663 /* ARGSUSED */
12658 12664 int
12659 12665 ip_sioctl_get_groupname(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
12660 12666 ip_ioctl_cmd_t *ipip, void *ifreq)
12661 12667 {
12662 12668 ipmp_grp_t *grp;
12663 12669 struct lifreq *lifr = ifreq;
12664 12670 ip_stack_t *ipst = ipif->ipif_ill->ill_ipst;
12665 12671
12666 12672 rw_enter(&ipst->ips_ipmp_lock, RW_READER);
12667 12673 if ((grp = ipif->ipif_ill->ill_phyint->phyint_grp) == NULL)
12668 12674 lifr->lifr_groupname[0] = '\0';
12669 12675 else
12670 12676 (void) strlcpy(lifr->lifr_groupname, grp->gr_name, LIFGRNAMSIZ);
12671 12677 rw_exit(&ipst->ips_ipmp_lock);
12672 12678 return (0);
12673 12679 }
12674 12680
12675 12681 /*
12676 12682 * Process an SIOCGLIFGROUPINFO request.
12677 12683 */
12678 12684 /* ARGSUSED */
12679 12685 int
12680 12686 ip_sioctl_groupinfo(ipif_t *dummy_ipif, sin_t *sin, queue_t *q, mblk_t *mp,
12681 12687 ip_ioctl_cmd_t *ipip, void *dummy)
12682 12688 {
12683 12689 ipmp_grp_t *grp;
12684 12690 lifgroupinfo_t *lifgr;
12685 12691 ip_stack_t *ipst = CONNQ_TO_IPST(q);
12686 12692
12687 12693 /* ip_wput_nondata() verified mp->b_cont->b_cont */
12688 12694 lifgr = (lifgroupinfo_t *)mp->b_cont->b_cont->b_rptr;
12689 12695 lifgr->gi_grname[LIFGRNAMSIZ - 1] = '\0';
12690 12696
12691 12697 rw_enter(&ipst->ips_ipmp_lock, RW_READER);
12692 12698 if ((grp = ipmp_grp_lookup(lifgr->gi_grname, ipst)) == NULL) {
12693 12699 rw_exit(&ipst->ips_ipmp_lock);
12694 12700 return (ENOENT);
12695 12701 }
12696 12702 ipmp_grp_info(grp, lifgr);
12697 12703 rw_exit(&ipst->ips_ipmp_lock);
12698 12704 return (0);
12699 12705 }
12700 12706
12701 12707 static void
12702 12708 ill_dl_down(ill_t *ill)
12703 12709 {
12704 12710 DTRACE_PROBE2(ill__downup, char *, "ill_dl_down", ill_t *, ill);
12705 12711
12706 12712 /*
12707 12713 * The ill is down; unbind but stay attached since we're still
12708 12714 * associated with a PPA. If we have negotiated DLPI capabilites
12709 12715 * with the data link service provider (IDS_OK) then reset them.
12710 12716 * The interval between unbinding and rebinding is potentially
12711 12717 * unbounded hence we cannot assume things will be the same.
12712 12718 * The DLPI capabilities will be probed again when the data link
12713 12719 * is brought up.
12714 12720 */
12715 12721 mblk_t *mp = ill->ill_unbind_mp;
12716 12722
12717 12723 ip1dbg(("ill_dl_down(%s)\n", ill->ill_name));
12718 12724
12719 12725 if (!ill->ill_replumbing) {
12720 12726 /* Free all ilms for this ill */
12721 12727 update_conn_ill(ill, ill->ill_ipst);
12722 12728 } else {
12723 12729 ill_leave_multicast(ill);
12724 12730 }
12725 12731
12726 12732 ill->ill_unbind_mp = NULL;
12727 12733 if (mp != NULL) {
12728 12734 ip1dbg(("ill_dl_down: %s (%u) for %s\n",
12729 12735 dl_primstr(*(int *)mp->b_rptr), *(int *)mp->b_rptr,
12730 12736 ill->ill_name));
12731 12737 mutex_enter(&ill->ill_lock);
12732 12738 ill->ill_state_flags |= ILL_DL_UNBIND_IN_PROGRESS;
12733 12739 mutex_exit(&ill->ill_lock);
12734 12740 /*
12735 12741 * ip_rput does not pass up normal (M_PROTO) DLPI messages
12736 12742 * after ILL_CONDEMNED is set. So in the unplumb case, we call
12737 12743 * ill_capability_dld_disable disable rightaway. If this is not
12738 12744 * an unplumb operation then the disable happens on receipt of
12739 12745 * the capab ack via ip_rput_dlpi_writer ->
12740 12746 * ill_capability_ack_thr. In both cases the order of
12741 12747 * the operations seen by DLD is capability disable followed
12742 12748 * by DL_UNBIND. Also the DLD capability disable needs a
12743 12749 * cv_wait'able context.
12744 12750 */
12745 12751 if (ill->ill_state_flags & ILL_CONDEMNED)
12746 12752 ill_capability_dld_disable(ill);
12747 12753 ill_capability_reset(ill, B_FALSE);
12748 12754 ill_dlpi_send(ill, mp);
12749 12755 }
12750 12756 mutex_enter(&ill->ill_lock);
12751 12757 ill->ill_dl_up = 0;
12752 12758 ill_nic_event_dispatch(ill, 0, NE_DOWN, NULL, 0);
12753 12759 mutex_exit(&ill->ill_lock);
12754 12760 }
12755 12761
12756 12762 void
12757 12763 ill_dlpi_dispatch(ill_t *ill, mblk_t *mp)
12758 12764 {
12759 12765 union DL_primitives *dlp;
12760 12766 t_uscalar_t prim;
12761 12767 boolean_t waitack = B_FALSE;
12762 12768
12763 12769 ASSERT(DB_TYPE(mp) == M_PROTO || DB_TYPE(mp) == M_PCPROTO);
12764 12770
12765 12771 dlp = (union DL_primitives *)mp->b_rptr;
12766 12772 prim = dlp->dl_primitive;
12767 12773
12768 12774 ip1dbg(("ill_dlpi_dispatch: sending %s (%u) to %s\n",
12769 12775 dl_primstr(prim), prim, ill->ill_name));
12770 12776
12771 12777 switch (prim) {
12772 12778 case DL_PHYS_ADDR_REQ:
12773 12779 {
12774 12780 dl_phys_addr_req_t *dlpap = (dl_phys_addr_req_t *)mp->b_rptr;
12775 12781 ill->ill_phys_addr_pend = dlpap->dl_addr_type;
12776 12782 break;
12777 12783 }
12778 12784 case DL_BIND_REQ:
12779 12785 mutex_enter(&ill->ill_lock);
12780 12786 ill->ill_state_flags &= ~ILL_DL_UNBIND_IN_PROGRESS;
12781 12787 mutex_exit(&ill->ill_lock);
12782 12788 break;
12783 12789 }
12784 12790
12785 12791 /*
12786 12792 * Except for the ACKs for the M_PCPROTO messages, all other ACKs
12787 12793 * are dropped by ip_rput() if ILL_CONDEMNED is set. Therefore
12788 12794 * we only wait for the ACK of the DL_UNBIND_REQ.
12789 12795 */
12790 12796 mutex_enter(&ill->ill_lock);
12791 12797 if (!(ill->ill_state_flags & ILL_CONDEMNED) ||
12792 12798 (prim == DL_UNBIND_REQ)) {
12793 12799 ill->ill_dlpi_pending = prim;
12794 12800 waitack = B_TRUE;
12795 12801 }
12796 12802
12797 12803 mutex_exit(&ill->ill_lock);
12798 12804 DTRACE_PROBE3(ill__dlpi, char *, "ill_dlpi_dispatch",
12799 12805 char *, dl_primstr(prim), ill_t *, ill);
12800 12806 putnext(ill->ill_wq, mp);
12801 12807
12802 12808 /*
12803 12809 * There is no ack for DL_NOTIFY_CONF messages
12804 12810 */
12805 12811 if (waitack && prim == DL_NOTIFY_CONF)
12806 12812 ill_dlpi_done(ill, prim);
12807 12813 }
12808 12814
12809 12815 /*
12810 12816 * Helper function for ill_dlpi_send().
12811 12817 */
12812 12818 /* ARGSUSED */
12813 12819 static void
12814 12820 ill_dlpi_send_writer(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *arg)
12815 12821 {
12816 12822 ill_dlpi_send(q->q_ptr, mp);
12817 12823 }
12818 12824
12819 12825 /*
12820 12826 * Send a DLPI control message to the driver but make sure there
12821 12827 * is only one outstanding message. Uses ill_dlpi_pending to tell
12822 12828 * when it must queue. ip_rput_dlpi_writer calls ill_dlpi_done()
12823 12829 * when an ACK or a NAK is received to process the next queued message.
12824 12830 */
12825 12831 void
12826 12832 ill_dlpi_send(ill_t *ill, mblk_t *mp)
12827 12833 {
12828 12834 mblk_t **mpp;
12829 12835
12830 12836 ASSERT(DB_TYPE(mp) == M_PROTO || DB_TYPE(mp) == M_PCPROTO);
12831 12837
12832 12838 /*
12833 12839 * To ensure that any DLPI requests for current exclusive operation
12834 12840 * are always completely sent before any DLPI messages for other
12835 12841 * operations, require writer access before enqueuing.
12836 12842 */
12837 12843 if (!IAM_WRITER_ILL(ill)) {
12838 12844 ill_refhold(ill);
12839 12845 /* qwriter_ip() does the ill_refrele() */
12840 12846 qwriter_ip(ill, ill->ill_wq, mp, ill_dlpi_send_writer,
12841 12847 NEW_OP, B_TRUE);
12842 12848 return;
12843 12849 }
12844 12850
12845 12851 mutex_enter(&ill->ill_lock);
12846 12852 if (ill->ill_dlpi_pending != DL_PRIM_INVAL) {
12847 12853 /* Must queue message. Tail insertion */
12848 12854 mpp = &ill->ill_dlpi_deferred;
12849 12855 while (*mpp != NULL)
12850 12856 mpp = &((*mpp)->b_next);
12851 12857
12852 12858 ip1dbg(("ill_dlpi_send: deferring request for %s "
12853 12859 "while %s pending\n", ill->ill_name,
12854 12860 dl_primstr(ill->ill_dlpi_pending)));
12855 12861
12856 12862 *mpp = mp;
12857 12863 mutex_exit(&ill->ill_lock);
12858 12864 return;
12859 12865 }
12860 12866 mutex_exit(&ill->ill_lock);
12861 12867 ill_dlpi_dispatch(ill, mp);
12862 12868 }
12863 12869
12864 12870 void
12865 12871 ill_capability_send(ill_t *ill, mblk_t *mp)
12866 12872 {
12867 12873 ill->ill_capab_pending_cnt++;
12868 12874 ill_dlpi_send(ill, mp);
12869 12875 }
12870 12876
12871 12877 void
12872 12878 ill_capability_done(ill_t *ill)
12873 12879 {
12874 12880 ASSERT(ill->ill_capab_pending_cnt != 0);
12875 12881
12876 12882 ill_dlpi_done(ill, DL_CAPABILITY_REQ);
12877 12883
12878 12884 ill->ill_capab_pending_cnt--;
12879 12885 if (ill->ill_capab_pending_cnt == 0 &&
12880 12886 ill->ill_dlpi_capab_state == IDCS_OK)
12881 12887 ill_capability_reset_alloc(ill);
12882 12888 }
12883 12889
12884 12890 /*
12885 12891 * Send all deferred DLPI messages without waiting for their ACKs.
12886 12892 */
12887 12893 void
12888 12894 ill_dlpi_send_deferred(ill_t *ill)
12889 12895 {
12890 12896 mblk_t *mp, *nextmp;
12891 12897
12892 12898 /*
12893 12899 * Clear ill_dlpi_pending so that the message is not queued in
12894 12900 * ill_dlpi_send().
12895 12901 */
12896 12902 mutex_enter(&ill->ill_lock);
12897 12903 ill->ill_dlpi_pending = DL_PRIM_INVAL;
12898 12904 mp = ill->ill_dlpi_deferred;
12899 12905 ill->ill_dlpi_deferred = NULL;
12900 12906 mutex_exit(&ill->ill_lock);
12901 12907
12902 12908 for (; mp != NULL; mp = nextmp) {
12903 12909 nextmp = mp->b_next;
12904 12910 mp->b_next = NULL;
12905 12911 ill_dlpi_send(ill, mp);
12906 12912 }
12907 12913 }
12908 12914
12909 12915 /*
12910 12916 * Clear all the deferred DLPI messages. Called on receiving an M_ERROR
12911 12917 * or M_HANGUP
12912 12918 */
12913 12919 static void
12914 12920 ill_dlpi_clear_deferred(ill_t *ill)
12915 12921 {
12916 12922 mblk_t *mp, *nextmp;
12917 12923
12918 12924 mutex_enter(&ill->ill_lock);
12919 12925 ill->ill_dlpi_pending = DL_PRIM_INVAL;
12920 12926 mp = ill->ill_dlpi_deferred;
12921 12927 ill->ill_dlpi_deferred = NULL;
12922 12928 mutex_exit(&ill->ill_lock);
12923 12929
12924 12930 for (; mp != NULL; mp = nextmp) {
12925 12931 nextmp = mp->b_next;
12926 12932 inet_freemsg(mp);
12927 12933 }
12928 12934 }
12929 12935
12930 12936 /*
12931 12937 * Check if the DLPI primitive `prim' is pending; print a warning if not.
12932 12938 */
12933 12939 boolean_t
12934 12940 ill_dlpi_pending(ill_t *ill, t_uscalar_t prim)
12935 12941 {
12936 12942 t_uscalar_t pending;
12937 12943
12938 12944 mutex_enter(&ill->ill_lock);
12939 12945 if (ill->ill_dlpi_pending == prim) {
12940 12946 mutex_exit(&ill->ill_lock);
12941 12947 return (B_TRUE);
12942 12948 }
12943 12949
12944 12950 /*
12945 12951 * During teardown, ill_dlpi_dispatch() will send DLPI requests
12946 12952 * without waiting, so don't print any warnings in that case.
12947 12953 */
12948 12954 if (ill->ill_state_flags & ILL_CONDEMNED) {
12949 12955 mutex_exit(&ill->ill_lock);
12950 12956 return (B_FALSE);
12951 12957 }
12952 12958 pending = ill->ill_dlpi_pending;
12953 12959 mutex_exit(&ill->ill_lock);
12954 12960
12955 12961 if (pending == DL_PRIM_INVAL) {
12956 12962 (void) mi_strlog(ill->ill_rq, 1, SL_CONSOLE|SL_ERROR|SL_TRACE,
12957 12963 "received unsolicited ack for %s on %s\n",
12958 12964 dl_primstr(prim), ill->ill_name);
12959 12965 } else {
12960 12966 (void) mi_strlog(ill->ill_rq, 1, SL_CONSOLE|SL_ERROR|SL_TRACE,
12961 12967 "received unexpected ack for %s on %s (expecting %s)\n",
12962 12968 dl_primstr(prim), ill->ill_name, dl_primstr(pending));
12963 12969 }
12964 12970 return (B_FALSE);
12965 12971 }
12966 12972
12967 12973 /*
12968 12974 * Complete the current DLPI operation associated with `prim' on `ill' and
12969 12975 * start the next queued DLPI operation (if any). If there are no queued DLPI
12970 12976 * operations and the ill's current exclusive IPSQ operation has finished
12971 12977 * (i.e., ipsq_current_finish() was called), then clear ipsq_current_ipif to
12972 12978 * allow the next exclusive IPSQ operation to begin upon ipsq_exit(). See
12973 12979 * the comments above ipsq_current_finish() for details.
12974 12980 */
12975 12981 void
12976 12982 ill_dlpi_done(ill_t *ill, t_uscalar_t prim)
12977 12983 {
12978 12984 mblk_t *mp;
12979 12985 ipsq_t *ipsq = ill->ill_phyint->phyint_ipsq;
12980 12986 ipxop_t *ipx = ipsq->ipsq_xop;
12981 12987
12982 12988 ASSERT(IAM_WRITER_IPSQ(ipsq));
12983 12989 mutex_enter(&ill->ill_lock);
12984 12990
12985 12991 ASSERT(prim != DL_PRIM_INVAL);
12986 12992 ASSERT(ill->ill_dlpi_pending == prim);
12987 12993
12988 12994 ip1dbg(("ill_dlpi_done: %s has completed %s (%u)\n", ill->ill_name,
12989 12995 dl_primstr(ill->ill_dlpi_pending), ill->ill_dlpi_pending));
12990 12996
12991 12997 if ((mp = ill->ill_dlpi_deferred) == NULL) {
12992 12998 ill->ill_dlpi_pending = DL_PRIM_INVAL;
12993 12999 if (ipx->ipx_current_done) {
12994 13000 mutex_enter(&ipx->ipx_lock);
12995 13001 ipx->ipx_current_ipif = NULL;
12996 13002 mutex_exit(&ipx->ipx_lock);
12997 13003 }
12998 13004 cv_signal(&ill->ill_cv);
12999 13005 mutex_exit(&ill->ill_lock);
13000 13006 return;
13001 13007 }
13002 13008
13003 13009 ill->ill_dlpi_deferred = mp->b_next;
13004 13010 mp->b_next = NULL;
13005 13011 mutex_exit(&ill->ill_lock);
13006 13012
13007 13013 ill_dlpi_dispatch(ill, mp);
13008 13014 }
13009 13015
13010 13016 /*
13011 13017 * Queue a (multicast) DLPI control message to be sent to the driver by
13012 13018 * later calling ill_dlpi_send_queued.
13013 13019 * We queue them while holding a lock (ill_mcast_lock) to ensure that they
13014 13020 * are sent in order i.e., prevent a DL_DISABMULTI_REQ and DL_ENABMULTI_REQ
13015 13021 * for the same group to race.
13016 13022 * We send DLPI control messages in order using ill_lock.
13017 13023 * For IPMP we should be called on the cast_ill.
13018 13024 */
13019 13025 void
13020 13026 ill_dlpi_queue(ill_t *ill, mblk_t *mp)
13021 13027 {
13022 13028 mblk_t **mpp;
13023 13029
13024 13030 ASSERT(DB_TYPE(mp) == M_PROTO || DB_TYPE(mp) == M_PCPROTO);
13025 13031
13026 13032 mutex_enter(&ill->ill_lock);
13027 13033 /* Must queue message. Tail insertion */
13028 13034 mpp = &ill->ill_dlpi_deferred;
13029 13035 while (*mpp != NULL)
13030 13036 mpp = &((*mpp)->b_next);
13031 13037
13032 13038 *mpp = mp;
13033 13039 mutex_exit(&ill->ill_lock);
13034 13040 }
13035 13041
13036 13042 /*
13037 13043 * Send the messages that were queued. Make sure there is only
13038 13044 * one outstanding message. ip_rput_dlpi_writer calls ill_dlpi_done()
13039 13045 * when an ACK or a NAK is received to process the next queued message.
13040 13046 * For IPMP we are called on the upper ill, but when send what is queued
13041 13047 * on the cast_ill.
13042 13048 */
13043 13049 void
13044 13050 ill_dlpi_send_queued(ill_t *ill)
13045 13051 {
13046 13052 mblk_t *mp;
13047 13053 union DL_primitives *dlp;
13048 13054 t_uscalar_t prim;
13049 13055 ill_t *release_ill = NULL;
13050 13056
13051 13057 if (IS_IPMP(ill)) {
13052 13058 /* On the upper IPMP ill. */
13053 13059 release_ill = ipmp_illgrp_hold_cast_ill(ill->ill_grp);
13054 13060 if (release_ill == NULL) {
13055 13061 /* Avoid ever sending anything down to the ipmpstub */
13056 13062 return;
13057 13063 }
13058 13064 ill = release_ill;
13059 13065 }
13060 13066 mutex_enter(&ill->ill_lock);
13061 13067 while ((mp = ill->ill_dlpi_deferred) != NULL) {
13062 13068 if (ill->ill_dlpi_pending != DL_PRIM_INVAL) {
13063 13069 /* Can't send. Somebody else will send it */
13064 13070 mutex_exit(&ill->ill_lock);
13065 13071 goto done;
13066 13072 }
13067 13073 ill->ill_dlpi_deferred = mp->b_next;
13068 13074 mp->b_next = NULL;
13069 13075 if (!ill->ill_dl_up) {
13070 13076 /*
13071 13077 * Nobody there. All multicast addresses will be
13072 13078 * re-joined when we get the DL_BIND_ACK bringing the
13073 13079 * interface up.
13074 13080 */
13075 13081 freemsg(mp);
13076 13082 continue;
13077 13083 }
13078 13084 dlp = (union DL_primitives *)mp->b_rptr;
13079 13085 prim = dlp->dl_primitive;
13080 13086
13081 13087 if (!(ill->ill_state_flags & ILL_CONDEMNED) ||
13082 13088 (prim == DL_UNBIND_REQ)) {
13083 13089 ill->ill_dlpi_pending = prim;
13084 13090 }
13085 13091 mutex_exit(&ill->ill_lock);
13086 13092
13087 13093 DTRACE_PROBE3(ill__dlpi, char *, "ill_dlpi_send_queued",
13088 13094 char *, dl_primstr(prim), ill_t *, ill);
13089 13095 putnext(ill->ill_wq, mp);
13090 13096 mutex_enter(&ill->ill_lock);
13091 13097 }
13092 13098 mutex_exit(&ill->ill_lock);
13093 13099 done:
13094 13100 if (release_ill != NULL)
13095 13101 ill_refrele(release_ill);
13096 13102 }
13097 13103
13098 13104 /*
13099 13105 * Queue an IP (IGMP/MLD) message to be sent by IP from
13100 13106 * ill_mcast_send_queued
13101 13107 * We queue them while holding a lock (ill_mcast_lock) to ensure that they
13102 13108 * are sent in order i.e., prevent a IGMP leave and IGMP join for the same
13103 13109 * group to race.
13104 13110 * We send them in order using ill_lock.
13105 13111 * For IPMP we are called on the upper ill, but we queue on the cast_ill.
13106 13112 */
13107 13113 void
13108 13114 ill_mcast_queue(ill_t *ill, mblk_t *mp)
13109 13115 {
13110 13116 mblk_t **mpp;
13111 13117 ill_t *release_ill = NULL;
13112 13118
13113 13119 ASSERT(RW_LOCK_HELD(&ill->ill_mcast_lock));
13114 13120
13115 13121 if (IS_IPMP(ill)) {
13116 13122 /* On the upper IPMP ill. */
13117 13123 release_ill = ipmp_illgrp_hold_cast_ill(ill->ill_grp);
13118 13124 if (release_ill == NULL) {
13119 13125 /* Discard instead of queuing for the ipmp interface */
13120 13126 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
13121 13127 ip_drop_output("ipIfStatsOutDiscards - no cast_ill",
13122 13128 mp, ill);
13123 13129 freemsg(mp);
13124 13130 return;
13125 13131 }
13126 13132 ill = release_ill;
13127 13133 }
13128 13134
13129 13135 mutex_enter(&ill->ill_lock);
13130 13136 /* Must queue message. Tail insertion */
13131 13137 mpp = &ill->ill_mcast_deferred;
13132 13138 while (*mpp != NULL)
13133 13139 mpp = &((*mpp)->b_next);
13134 13140
13135 13141 *mpp = mp;
13136 13142 mutex_exit(&ill->ill_lock);
13137 13143 if (release_ill != NULL)
13138 13144 ill_refrele(release_ill);
13139 13145 }
13140 13146
13141 13147 /*
13142 13148 * Send the IP packets that were queued by ill_mcast_queue.
13143 13149 * These are IGMP/MLD packets.
13144 13150 *
13145 13151 * For IPMP we are called on the upper ill, but when send what is queued
13146 13152 * on the cast_ill.
13147 13153 *
13148 13154 * Request loopback of the report if we are acting as a multicast
13149 13155 * router, so that the process-level routing demon can hear it.
13150 13156 * This will run multiple times for the same group if there are members
13151 13157 * on the same group for multiple ipif's on the same ill. The
13152 13158 * igmp_input/mld_input code will suppress this due to the loopback thus we
13153 13159 * always loopback membership report.
13154 13160 *
13155 13161 * We also need to make sure that this does not get load balanced
13156 13162 * by IPMP. We do this by passing an ill to ip_output_simple.
13157 13163 */
13158 13164 void
13159 13165 ill_mcast_send_queued(ill_t *ill)
13160 13166 {
13161 13167 mblk_t *mp;
13162 13168 ip_xmit_attr_t ixas;
13163 13169 ill_t *release_ill = NULL;
13164 13170
13165 13171 if (IS_IPMP(ill)) {
13166 13172 /* On the upper IPMP ill. */
13167 13173 release_ill = ipmp_illgrp_hold_cast_ill(ill->ill_grp);
13168 13174 if (release_ill == NULL) {
13169 13175 /*
13170 13176 * We should have no messages on the ipmp interface
13171 13177 * but no point in trying to send them.
13172 13178 */
13173 13179 return;
13174 13180 }
13175 13181 ill = release_ill;
13176 13182 }
13177 13183 bzero(&ixas, sizeof (ixas));
13178 13184 ixas.ixa_zoneid = ALL_ZONES;
13179 13185 ixas.ixa_cred = kcred;
13180 13186 ixas.ixa_cpid = NOPID;
13181 13187 ixas.ixa_tsl = NULL;
13182 13188 /*
13183 13189 * Here we set ixa_ifindex. If IPMP it will be the lower ill which
13184 13190 * makes ip_select_route pick the IRE_MULTICAST for the cast_ill.
13185 13191 * That is necessary to handle IGMP/MLD snooping switches.
13186 13192 */
13187 13193 ixas.ixa_ifindex = ill->ill_phyint->phyint_ifindex;
13188 13194 ixas.ixa_ipst = ill->ill_ipst;
13189 13195
13190 13196 mutex_enter(&ill->ill_lock);
13191 13197 while ((mp = ill->ill_mcast_deferred) != NULL) {
13192 13198 ill->ill_mcast_deferred = mp->b_next;
13193 13199 mp->b_next = NULL;
13194 13200 if (!ill->ill_dl_up) {
13195 13201 /*
13196 13202 * Nobody there. Just drop the ip packets.
13197 13203 * IGMP/MLD will resend later, if this is a replumb.
13198 13204 */
13199 13205 freemsg(mp);
13200 13206 continue;
13201 13207 }
13202 13208 mutex_enter(&ill->ill_phyint->phyint_lock);
13203 13209 if (IS_UNDER_IPMP(ill) && !ipmp_ill_is_active(ill)) {
13204 13210 /*
13205 13211 * When the ill is getting deactivated, we only want to
13206 13212 * send the DLPI messages, so drop IGMP/MLD packets.
13207 13213 * DLPI messages are handled by ill_dlpi_send_queued()
13208 13214 */
13209 13215 mutex_exit(&ill->ill_phyint->phyint_lock);
13210 13216 freemsg(mp);
13211 13217 continue;
13212 13218 }
13213 13219 mutex_exit(&ill->ill_phyint->phyint_lock);
13214 13220 mutex_exit(&ill->ill_lock);
13215 13221
13216 13222 /* Check whether we are sending IPv4 or IPv6. */
13217 13223 if (ill->ill_isv6) {
13218 13224 ip6_t *ip6h = (ip6_t *)mp->b_rptr;
13219 13225
13220 13226 ixas.ixa_multicast_ttl = ip6h->ip6_hops;
13221 13227 ixas.ixa_flags = IXAF_BASIC_SIMPLE_V6;
13222 13228 } else {
13223 13229 ipha_t *ipha = (ipha_t *)mp->b_rptr;
13224 13230
13225 13231 ixas.ixa_multicast_ttl = ipha->ipha_ttl;
13226 13232 ixas.ixa_flags = IXAF_BASIC_SIMPLE_V4;
13227 13233 ixas.ixa_flags &= ~IXAF_SET_ULP_CKSUM;
13228 13234 }
13229 13235 ixas.ixa_flags &= ~IXAF_VERIFY_SOURCE;
13230 13236 ixas.ixa_flags |= IXAF_MULTICAST_LOOP | IXAF_SET_SOURCE;
13231 13237 (void) ip_output_simple(mp, &ixas);
13232 13238 ixa_cleanup(&ixas);
13233 13239
13234 13240 mutex_enter(&ill->ill_lock);
13235 13241 }
13236 13242 mutex_exit(&ill->ill_lock);
13237 13243
13238 13244 done:
13239 13245 if (release_ill != NULL)
13240 13246 ill_refrele(release_ill);
13241 13247 }
13242 13248
13243 13249 /*
13244 13250 * Take down a specific interface, but don't lose any information about it.
13245 13251 * (Always called as writer.)
13246 13252 * This function goes through the down sequence even if the interface is
13247 13253 * already down. There are 2 reasons.
13248 13254 * a. Currently we permit interface routes that depend on down interfaces
13249 13255 * to be added. This behaviour itself is questionable. However it appears
13250 13256 * that both Solaris and 4.3 BSD have exhibited this behaviour for a long
13251 13257 * time. We go thru the cleanup in order to remove these routes.
13252 13258 * b. The bringup of the interface could fail in ill_dl_up i.e. we get
13253 13259 * DL_ERROR_ACK in response to the DL_BIND request. The interface is
13254 13260 * down, but we need to cleanup i.e. do ill_dl_down and
13255 13261 * ip_rput_dlpi_writer (DL_ERROR_ACK) -> ipif_down.
13256 13262 *
13257 13263 * IP-MT notes:
13258 13264 *
13259 13265 * Model of reference to interfaces.
13260 13266 *
13261 13267 * The following members in ipif_t track references to the ipif.
13262 13268 * int ipif_refcnt; Active reference count
13263 13269 *
13264 13270 * The following members in ill_t track references to the ill.
13265 13271 * int ill_refcnt; active refcnt
13266 13272 * uint_t ill_ire_cnt; Number of ires referencing ill
13267 13273 * uint_t ill_ncec_cnt; Number of ncecs referencing ill
13268 13274 * uint_t ill_nce_cnt; Number of nces referencing ill
13269 13275 * uint_t ill_ilm_cnt; Number of ilms referencing ill
13270 13276 *
13271 13277 * Reference to an ipif or ill can be obtained in any of the following ways.
13272 13278 *
13273 13279 * Through the lookup functions ipif_lookup_* / ill_lookup_* functions
13274 13280 * Pointers to ipif / ill from other data structures viz ire and conn.
13275 13281 * Implicit reference to the ipif / ill by holding a reference to the ire.
13276 13282 *
13277 13283 * The ipif/ill lookup functions return a reference held ipif / ill.
13278 13284 * ipif_refcnt and ill_refcnt track the reference counts respectively.
13279 13285 * This is a purely dynamic reference count associated with threads holding
13280 13286 * references to the ipif / ill. Pointers from other structures do not
13281 13287 * count towards this reference count.
13282 13288 *
13283 13289 * ill_ire_cnt is the number of ire's associated with the
13284 13290 * ill. This is incremented whenever a new ire is created referencing the
13285 13291 * ill. This is done atomically inside ire_add_v[46] where the ire is
13286 13292 * actually added to the ire hash table. The count is decremented in
13287 13293 * ire_inactive where the ire is destroyed.
13288 13294 *
13289 13295 * ill_ncec_cnt is the number of ncec's referencing the ill thru ncec_ill.
13290 13296 * This is incremented atomically in
13291 13297 * ndp_add_v4()/ndp_add_v6() where the nce is actually added to the
13292 13298 * table. Similarly it is decremented in ncec_inactive() where the ncec
13293 13299 * is destroyed.
13294 13300 *
13295 13301 * ill_nce_cnt is the number of nce's referencing the ill thru nce_ill. This is
13296 13302 * incremented atomically in nce_add() where the nce is actually added to the
13297 13303 * ill_nce. Similarly it is decremented in nce_inactive() where the nce
13298 13304 * is destroyed.
13299 13305 *
13300 13306 * ill_ilm_cnt is the ilm's reference to the ill. It is incremented in
13301 13307 * ilm_add() and decremented before the ilm is freed in ilm_delete().
13302 13308 *
13303 13309 * Flow of ioctls involving interface down/up
13304 13310 *
13305 13311 * The following is the sequence of an attempt to set some critical flags on an
13306 13312 * up interface.
13307 13313 * ip_sioctl_flags
13308 13314 * ipif_down
13309 13315 * wait for ipif to be quiescent
13310 13316 * ipif_down_tail
13311 13317 * ip_sioctl_flags_tail
13312 13318 *
13313 13319 * All set ioctls that involve down/up sequence would have a skeleton similar
13314 13320 * to the above. All the *tail functions are called after the refcounts have
13315 13321 * dropped to the appropriate values.
13316 13322 *
13317 13323 * SIOC ioctls during the IPIF_CHANGING interval.
13318 13324 *
13319 13325 * Threads handling SIOC set ioctls serialize on the squeue, but this
13320 13326 * is not done for SIOC get ioctls. Since a set ioctl can cause several
13321 13327 * steps of internal changes to the state, some of which are visible in
13322 13328 * ipif_flags (such as IFF_UP being cleared and later set), and we want
13323 13329 * the set ioctl to be atomic related to the get ioctls, the SIOC get code
13324 13330 * will wait and restart ioctls if IPIF_CHANGING is set. The mblk is then
13325 13331 * enqueued in the ipsq and the operation is restarted by ipsq_exit() when
13326 13332 * the current exclusive operation completes. The IPIF_CHANGING check
13327 13333 * and enqueue is atomic using the ill_lock and ipsq_lock. The
13328 13334 * lookup is done holding the ill_lock. Hence the ill/ipif state flags can't
13329 13335 * change while the ill_lock is held. Before dropping the ill_lock we acquire
13330 13336 * the ipsq_lock and call ipsq_enq. This ensures that ipsq_exit can't finish
13331 13337 * until we release the ipsq_lock, even though the ill/ipif state flags
13332 13338 * can change after we drop the ill_lock.
13333 13339 */
13334 13340 int
13335 13341 ipif_down(ipif_t *ipif, queue_t *q, mblk_t *mp)
13336 13342 {
13337 13343 ill_t *ill = ipif->ipif_ill;
13338 13344 conn_t *connp;
13339 13345 boolean_t success;
13340 13346 boolean_t ipif_was_up = B_FALSE;
13341 13347 ip_stack_t *ipst = ill->ill_ipst;
13342 13348
13343 13349 ASSERT(IAM_WRITER_IPIF(ipif));
13344 13350
13345 13351 ip1dbg(("ipif_down(%s:%u)\n", ill->ill_name, ipif->ipif_id));
13346 13352
13347 13353 DTRACE_PROBE3(ipif__downup, char *, "ipif_down",
13348 13354 ill_t *, ill, ipif_t *, ipif);
13349 13355
13350 13356 if (ipif->ipif_flags & IPIF_UP) {
13351 13357 mutex_enter(&ill->ill_lock);
13352 13358 ipif->ipif_flags &= ~IPIF_UP;
13353 13359 ASSERT(ill->ill_ipif_up_count > 0);
13354 13360 --ill->ill_ipif_up_count;
13355 13361 mutex_exit(&ill->ill_lock);
13356 13362 ipif_was_up = B_TRUE;
13357 13363 /* Update status in SCTP's list */
13358 13364 sctp_update_ipif(ipif, SCTP_IPIF_DOWN);
13359 13365 ill_nic_event_dispatch(ipif->ipif_ill,
13360 13366 MAP_IPIF_ID(ipif->ipif_id), NE_LIF_DOWN, NULL, 0);
13361 13367 }
13362 13368
13363 13369 /*
13364 13370 * Removal of the last ipif from an ill may result in a DL_UNBIND
13365 13371 * being sent to the driver, and we must not send any data packets to
13366 13372 * the driver after the DL_UNBIND_REQ. To ensure this, all the
13367 13373 * ire and nce entries used in the data path will be cleaned
13368 13374 * up, and we also set the ILL_DOWN_IN_PROGRESS bit to make
13369 13375 * sure on new entries will be added until the ill is bound
13370 13376 * again. The ILL_DOWN_IN_PROGRESS bit is turned off upon
13371 13377 * receipt of a DL_BIND_ACK.
13372 13378 */
13373 13379 if (ill->ill_wq != NULL && !ill->ill_logical_down &&
13374 13380 ill->ill_ipif_up_count == 0 && ill->ill_ipif_dup_count == 0 &&
13375 13381 ill->ill_dl_up) {
13376 13382 ill->ill_state_flags |= ILL_DOWN_IN_PROGRESS;
13377 13383 }
13378 13384
13379 13385 /*
13380 13386 * Blow away memberships we established in ipif_multicast_up().
13381 13387 */
13382 13388 ipif_multicast_down(ipif);
13383 13389
13384 13390 /*
13385 13391 * Remove from the mapping for __sin6_src_id. We insert only
13386 13392 * when the address is not INADDR_ANY. As IPv4 addresses are
13387 13393 * stored as mapped addresses, we need to check for mapped
13388 13394 * INADDR_ANY also.
13389 13395 */
13390 13396 if (ipif_was_up && !IN6_IS_ADDR_UNSPECIFIED(&ipif->ipif_v6lcl_addr) &&
13391 13397 !IN6_IS_ADDR_V4MAPPED_ANY(&ipif->ipif_v6lcl_addr) &&
13392 13398 !(ipif->ipif_flags & IPIF_NOLOCAL)) {
13393 13399 int err;
13394 13400
13395 13401 err = ip_srcid_remove(&ipif->ipif_v6lcl_addr,
13396 13402 ipif->ipif_zoneid, ipst);
13397 13403 if (err != 0) {
13398 13404 ip0dbg(("ipif_down: srcid_remove %d\n", err));
13399 13405 }
13400 13406 }
13401 13407
13402 13408 if (ipif_was_up) {
13403 13409 /* only delete if we'd added ire's before */
13404 13410 if (ipif->ipif_isv6)
13405 13411 ipif_delete_ires_v6(ipif);
13406 13412 else
13407 13413 ipif_delete_ires_v4(ipif);
13408 13414 }
13409 13415
13410 13416 if (ipif_was_up && ill->ill_ipif_up_count == 0) {
13411 13417 /*
13412 13418 * Since the interface is now down, it may have just become
13413 13419 * inactive. Note that this needs to be done even for a
13414 13420 * lll_logical_down(), or ARP entries will not get correctly
13415 13421 * restored when the interface comes back up.
13416 13422 */
13417 13423 if (IS_UNDER_IPMP(ill))
13418 13424 ipmp_ill_refresh_active(ill);
13419 13425 }
13420 13426
13421 13427 /*
13422 13428 * neighbor-discovery or arp entries for this interface. The ipif
13423 13429 * has to be quiesced, so we walk all the nce's and delete those
13424 13430 * that point at the ipif->ipif_ill. At the same time, we also
13425 13431 * update IPMP so that ipifs for data addresses are unbound. We dont
13426 13432 * call ipif_arp_down to DL_UNBIND the arp stream itself here, but defer
13427 13433 * that for ipif_down_tail()
13428 13434 */
13429 13435 ipif_nce_down(ipif);
13430 13436
13431 13437 /*
13432 13438 * If this is the last ipif on the ill, we also need to remove
13433 13439 * any IREs with ire_ill set. Otherwise ipif_is_quiescent() will
13434 13440 * never succeed.
13435 13441 */
13436 13442 if (ill->ill_ipif_up_count == 0 && ill->ill_ipif_dup_count == 0)
13437 13443 ire_walk_ill(0, 0, ill_downi, ill, ill);
13438 13444
13439 13445 /*
13440 13446 * Walk all CONNs that can have a reference on an ire for this
13441 13447 * ipif (we actually walk all that now have stale references).
13442 13448 */
13443 13449 ipcl_walk(conn_ixa_cleanup, (void *)B_TRUE, ipst);
13444 13450
13445 13451 /*
13446 13452 * If mp is NULL the caller will wait for the appropriate refcnt.
13447 13453 * Eg. ip_sioctl_removeif -> ipif_free -> ipif_down
13448 13454 * and ill_delete -> ipif_free -> ipif_down
13449 13455 */
13450 13456 if (mp == NULL) {
13451 13457 ASSERT(q == NULL);
13452 13458 return (0);
13453 13459 }
13454 13460
13455 13461 if (CONN_Q(q)) {
13456 13462 connp = Q_TO_CONN(q);
13457 13463 mutex_enter(&connp->conn_lock);
13458 13464 } else {
13459 13465 connp = NULL;
13460 13466 }
13461 13467 mutex_enter(&ill->ill_lock);
13462 13468 /*
13463 13469 * Are there any ire's pointing to this ipif that are still active ?
13464 13470 * If this is the last ipif going down, are there any ire's pointing
13465 13471 * to this ill that are still active ?
13466 13472 */
13467 13473 if (ipif_is_quiescent(ipif)) {
13468 13474 mutex_exit(&ill->ill_lock);
13469 13475 if (connp != NULL)
13470 13476 mutex_exit(&connp->conn_lock);
13471 13477 return (0);
13472 13478 }
13473 13479
13474 13480 ip1dbg(("ipif_down: need to wait, adding pending mp %s ill %p",
13475 13481 ill->ill_name, (void *)ill));
13476 13482 /*
13477 13483 * Enqueue the mp atomically in ipsq_pending_mp. When the refcount
13478 13484 * drops down, the operation will be restarted by ipif_ill_refrele_tail
13479 13485 * which in turn is called by the last refrele on the ipif/ill/ire.
13480 13486 */
13481 13487 success = ipsq_pending_mp_add(connp, ipif, q, mp, IPIF_DOWN);
13482 13488 if (!success) {
13483 13489 /* The conn is closing. So just return */
13484 13490 ASSERT(connp != NULL);
13485 13491 mutex_exit(&ill->ill_lock);
13486 13492 mutex_exit(&connp->conn_lock);
13487 13493 return (EINTR);
13488 13494 }
13489 13495
13490 13496 mutex_exit(&ill->ill_lock);
13491 13497 if (connp != NULL)
13492 13498 mutex_exit(&connp->conn_lock);
13493 13499 return (EINPROGRESS);
13494 13500 }
13495 13501
13496 13502 int
13497 13503 ipif_down_tail(ipif_t *ipif)
13498 13504 {
13499 13505 ill_t *ill = ipif->ipif_ill;
13500 13506 int err = 0;
13501 13507
13502 13508 DTRACE_PROBE3(ipif__downup, char *, "ipif_down_tail",
13503 13509 ill_t *, ill, ipif_t *, ipif);
13504 13510
13505 13511 /*
13506 13512 * Skip any loopback interface (null wq).
13507 13513 * If this is the last logical interface on the ill
13508 13514 * have ill_dl_down tell the driver we are gone (unbind)
13509 13515 * Note that lun 0 can ipif_down even though
13510 13516 * there are other logical units that are up.
13511 13517 * This occurs e.g. when we change a "significant" IFF_ flag.
13512 13518 */
13513 13519 if (ill->ill_wq != NULL && !ill->ill_logical_down &&
13514 13520 ill->ill_ipif_up_count == 0 && ill->ill_ipif_dup_count == 0 &&
13515 13521 ill->ill_dl_up) {
13516 13522 ill_dl_down(ill);
13517 13523 }
13518 13524 if (!ipif->ipif_isv6)
13519 13525 err = ipif_arp_down(ipif);
13520 13526
13521 13527 ill->ill_logical_down = 0;
13522 13528
13523 13529 ip_rts_ifmsg(ipif, RTSQ_DEFAULT);
13524 13530 ip_rts_newaddrmsg(RTM_DELETE, 0, ipif, RTSQ_DEFAULT);
13525 13531 return (err);
13526 13532 }
13527 13533
13528 13534 /*
13529 13535 * Bring interface logically down without bringing the physical interface
13530 13536 * down e.g. when the netmask is changed. This avoids long lasting link
13531 13537 * negotiations between an ethernet interface and a certain switches.
13532 13538 */
13533 13539 static int
13534 13540 ipif_logical_down(ipif_t *ipif, queue_t *q, mblk_t *mp)
13535 13541 {
13536 13542 DTRACE_PROBE3(ipif__downup, char *, "ipif_logical_down",
13537 13543 ill_t *, ipif->ipif_ill, ipif_t *, ipif);
13538 13544
13539 13545 /*
13540 13546 * The ill_logical_down flag is a transient flag. It is set here
13541 13547 * and is cleared once the down has completed in ipif_down_tail.
13542 13548 * This flag does not indicate whether the ill stream is in the
13543 13549 * DL_BOUND state with the driver. Instead this flag is used by
13544 13550 * ipif_down_tail to determine whether to DL_UNBIND the stream with
13545 13551 * the driver. The state of the ill stream i.e. whether it is
13546 13552 * DL_BOUND with the driver or not is indicated by the ill_dl_up flag.
13547 13553 */
13548 13554 ipif->ipif_ill->ill_logical_down = 1;
13549 13555 return (ipif_down(ipif, q, mp));
13550 13556 }
13551 13557
13552 13558 /*
13553 13559 * Initiate deallocate of an IPIF. Always called as writer. Called by
13554 13560 * ill_delete or ip_sioctl_removeif.
13555 13561 */
13556 13562 static void
13557 13563 ipif_free(ipif_t *ipif)
13558 13564 {
13559 13565 ip_stack_t *ipst = ipif->ipif_ill->ill_ipst;
13560 13566
13561 13567 ASSERT(IAM_WRITER_IPIF(ipif));
13562 13568
13563 13569 if (ipif->ipif_recovery_id != 0)
13564 13570 (void) untimeout(ipif->ipif_recovery_id);
13565 13571 ipif->ipif_recovery_id = 0;
13566 13572
13567 13573 /*
13568 13574 * Take down the interface. We can be called either from ill_delete
13569 13575 * or from ip_sioctl_removeif.
13570 13576 */
13571 13577 (void) ipif_down(ipif, NULL, NULL);
13572 13578
13573 13579 /*
13574 13580 * Now that the interface is down, there's no chance it can still
13575 13581 * become a duplicate. Cancel any timer that may have been set while
13576 13582 * tearing down.
13577 13583 */
13578 13584 if (ipif->ipif_recovery_id != 0)
13579 13585 (void) untimeout(ipif->ipif_recovery_id);
13580 13586 ipif->ipif_recovery_id = 0;
13581 13587
13582 13588 rw_enter(&ipst->ips_ill_g_lock, RW_WRITER);
13583 13589 /* Remove pointers to this ill in the multicast routing tables */
13584 13590 reset_mrt_vif_ipif(ipif);
13585 13591 /* If necessary, clear the cached source ipif rotor. */
13586 13592 if (ipif->ipif_ill->ill_src_ipif == ipif)
13587 13593 ipif->ipif_ill->ill_src_ipif = NULL;
13588 13594 rw_exit(&ipst->ips_ill_g_lock);
13589 13595 }
13590 13596
13591 13597 static void
13592 13598 ipif_free_tail(ipif_t *ipif)
13593 13599 {
13594 13600 ip_stack_t *ipst = ipif->ipif_ill->ill_ipst;
13595 13601
13596 13602 /*
13597 13603 * Need to hold both ill_g_lock and ill_lock while
13598 13604 * inserting or removing an ipif from the linked list
13599 13605 * of ipifs hanging off the ill.
13600 13606 */
13601 13607 rw_enter(&ipst->ips_ill_g_lock, RW_WRITER);
13602 13608
13603 13609 #ifdef DEBUG
13604 13610 ipif_trace_cleanup(ipif);
13605 13611 #endif
13606 13612
13607 13613 /* Ask SCTP to take it out of it list */
13608 13614 sctp_update_ipif(ipif, SCTP_IPIF_REMOVE);
13609 13615 ip_rts_newaddrmsg(RTM_FREEADDR, 0, ipif, RTSQ_DEFAULT);
13610 13616
13611 13617 /* Get it out of the ILL interface list. */
13612 13618 ipif_remove(ipif);
13613 13619 rw_exit(&ipst->ips_ill_g_lock);
13614 13620
13615 13621 ASSERT(!(ipif->ipif_flags & (IPIF_UP | IPIF_DUPLICATE)));
13616 13622 ASSERT(ipif->ipif_recovery_id == 0);
13617 13623 ASSERT(ipif->ipif_ire_local == NULL);
13618 13624 ASSERT(ipif->ipif_ire_if == NULL);
13619 13625
13620 13626 /* Free the memory. */
13621 13627 mi_free(ipif);
13622 13628 }
13623 13629
13624 13630 /*
13625 13631 * Sets `buf' to an ipif name of the form "ill_name:id", or "ill_name" if "id"
13626 13632 * is zero.
13627 13633 */
13628 13634 void
13629 13635 ipif_get_name(const ipif_t *ipif, char *buf, int len)
13630 13636 {
13631 13637 char lbuf[LIFNAMSIZ];
13632 13638 char *name;
13633 13639 size_t name_len;
13634 13640
13635 13641 buf[0] = '\0';
13636 13642 name = ipif->ipif_ill->ill_name;
13637 13643 name_len = ipif->ipif_ill->ill_name_length;
13638 13644 if (ipif->ipif_id != 0) {
13639 13645 (void) sprintf(lbuf, "%s%c%d", name, IPIF_SEPARATOR_CHAR,
13640 13646 ipif->ipif_id);
13641 13647 name = lbuf;
13642 13648 name_len = mi_strlen(name) + 1;
13643 13649 }
13644 13650 len -= 1;
13645 13651 buf[len] = '\0';
13646 13652 len = MIN(len, name_len);
13647 13653 bcopy(name, buf, len);
13648 13654 }
13649 13655
13650 13656 /*
13651 13657 * Sets `buf' to an ill name.
13652 13658 */
13653 13659 void
13654 13660 ill_get_name(const ill_t *ill, char *buf, int len)
13655 13661 {
13656 13662 char *name;
13657 13663 size_t name_len;
13658 13664
13659 13665 name = ill->ill_name;
13660 13666 name_len = ill->ill_name_length;
13661 13667 len -= 1;
13662 13668 buf[len] = '\0';
13663 13669 len = MIN(len, name_len);
13664 13670 bcopy(name, buf, len);
13665 13671 }
13666 13672
13667 13673 /*
13668 13674 * Find an IPIF based on the name passed in. Names can be of the form <phys>
13669 13675 * (e.g., le0) or <phys>:<#> (e.g., le0:1). When there is no colon, the
13670 13676 * implied unit id is zero. <phys> must correspond to the name of an ILL.
13671 13677 * (May be called as writer.)
13672 13678 */
13673 13679 static ipif_t *
13674 13680 ipif_lookup_on_name(char *name, size_t namelen, boolean_t do_alloc,
13675 13681 boolean_t *exists, boolean_t isv6, zoneid_t zoneid, ip_stack_t *ipst)
13676 13682 {
13677 13683 char *cp;
13678 13684 char *endp;
13679 13685 long id;
13680 13686 ill_t *ill;
13681 13687 ipif_t *ipif;
13682 13688 uint_t ire_type;
13683 13689 boolean_t did_alloc = B_FALSE;
13684 13690 char last;
13685 13691
13686 13692 /*
13687 13693 * If the caller wants to us to create the ipif, make sure we have a
13688 13694 * valid zoneid
13689 13695 */
13690 13696 ASSERT(!do_alloc || zoneid != ALL_ZONES);
13691 13697
13692 13698 if (namelen == 0) {
13693 13699 return (NULL);
13694 13700 }
13695 13701
13696 13702 *exists = B_FALSE;
13697 13703 /* Look for a colon in the name. */
13698 13704 endp = &name[namelen];
13699 13705 for (cp = endp; --cp > name; ) {
13700 13706 if (*cp == IPIF_SEPARATOR_CHAR)
13701 13707 break;
13702 13708 }
13703 13709
13704 13710 if (*cp == IPIF_SEPARATOR_CHAR) {
13705 13711 /*
13706 13712 * Reject any non-decimal aliases for logical
13707 13713 * interfaces. Aliases with leading zeroes
13708 13714 * are also rejected as they introduce ambiguity
13709 13715 * in the naming of the interfaces.
13710 13716 * In order to confirm with existing semantics,
13711 13717 * and to not break any programs/script relying
13712 13718 * on that behaviour, if<0>:0 is considered to be
13713 13719 * a valid interface.
13714 13720 *
13715 13721 * If alias has two or more digits and the first
13716 13722 * is zero, fail.
13717 13723 */
13718 13724 if (&cp[2] < endp && cp[1] == '0') {
13719 13725 return (NULL);
13720 13726 }
13721 13727 }
13722 13728
13723 13729 if (cp <= name) {
13724 13730 cp = endp;
13725 13731 }
13726 13732 last = *cp;
13727 13733 *cp = '\0';
13728 13734
13729 13735 /*
13730 13736 * Look up the ILL, based on the portion of the name
13731 13737 * before the slash. ill_lookup_on_name returns a held ill.
13732 13738 * Temporary to check whether ill exists already. If so
13733 13739 * ill_lookup_on_name will clear it.
13734 13740 */
13735 13741 ill = ill_lookup_on_name(name, do_alloc, isv6,
13736 13742 &did_alloc, ipst);
13737 13743 *cp = last;
13738 13744 if (ill == NULL)
13739 13745 return (NULL);
13740 13746
13741 13747 /* Establish the unit number in the name. */
13742 13748 id = 0;
13743 13749 if (cp < endp && *endp == '\0') {
13744 13750 /* If there was a colon, the unit number follows. */
13745 13751 cp++;
13746 13752 if (ddi_strtol(cp, NULL, 0, &id) != 0) {
13747 13753 ill_refrele(ill);
13748 13754 return (NULL);
13749 13755 }
13750 13756 }
13751 13757
13752 13758 mutex_enter(&ill->ill_lock);
13753 13759 /* Now see if there is an IPIF with this unit number. */
13754 13760 for (ipif = ill->ill_ipif; ipif != NULL; ipif = ipif->ipif_next) {
13755 13761 if (ipif->ipif_id == id) {
13756 13762 if (zoneid != ALL_ZONES &&
13757 13763 zoneid != ipif->ipif_zoneid &&
13758 13764 ipif->ipif_zoneid != ALL_ZONES) {
13759 13765 mutex_exit(&ill->ill_lock);
13760 13766 ill_refrele(ill);
13761 13767 return (NULL);
13762 13768 }
13763 13769 if (IPIF_CAN_LOOKUP(ipif)) {
13764 13770 ipif_refhold_locked(ipif);
13765 13771 mutex_exit(&ill->ill_lock);
13766 13772 if (!did_alloc)
13767 13773 *exists = B_TRUE;
13768 13774 /*
13769 13775 * Drop locks before calling ill_refrele
13770 13776 * since it can potentially call into
13771 13777 * ipif_ill_refrele_tail which can end up
13772 13778 * in trying to acquire any lock.
13773 13779 */
13774 13780 ill_refrele(ill);
13775 13781 return (ipif);
13776 13782 }
13777 13783 }
13778 13784 }
13779 13785
13780 13786 if (!do_alloc) {
13781 13787 mutex_exit(&ill->ill_lock);
13782 13788 ill_refrele(ill);
13783 13789 return (NULL);
13784 13790 }
13785 13791
13786 13792 /*
13787 13793 * If none found, atomically allocate and return a new one.
13788 13794 * Historically, we used IRE_LOOPBACK only for lun 0, and IRE_LOCAL
13789 13795 * to support "receive only" use of lo0:1 etc. as is still done
13790 13796 * below as an initial guess.
13791 13797 * However, this is now likely to be overriden later in ipif_up_done()
13792 13798 * when we know for sure what address has been configured on the
13793 13799 * interface, since we might have more than one loopback interface
13794 13800 * with a loopback address, e.g. in the case of zones, and all the
13795 13801 * interfaces with loopback addresses need to be marked IRE_LOOPBACK.
13796 13802 */
13797 13803 if (ill->ill_net_type == IRE_LOOPBACK && id == 0)
13798 13804 ire_type = IRE_LOOPBACK;
13799 13805 else
13800 13806 ire_type = IRE_LOCAL;
13801 13807 ipif = ipif_allocate(ill, id, ire_type, B_TRUE, B_TRUE, NULL);
13802 13808 if (ipif != NULL)
13803 13809 ipif_refhold_locked(ipif);
13804 13810 mutex_exit(&ill->ill_lock);
13805 13811 ill_refrele(ill);
13806 13812 return (ipif);
13807 13813 }
13808 13814
13809 13815 /*
13810 13816 * Variant of the above that queues the request on the ipsq when
13811 13817 * IPIF_CHANGING is set.
13812 13818 */
13813 13819 static ipif_t *
13814 13820 ipif_lookup_on_name_async(char *name, size_t namelen, boolean_t isv6,
13815 13821 zoneid_t zoneid, queue_t *q, mblk_t *mp, ipsq_func_t func, int *error,
13816 13822 ip_stack_t *ipst)
13817 13823 {
13818 13824 char *cp;
13819 13825 char *endp;
13820 13826 long id;
13821 13827 ill_t *ill;
13822 13828 ipif_t *ipif;
13823 13829 boolean_t did_alloc = B_FALSE;
13824 13830 ipsq_t *ipsq;
13825 13831
13826 13832 if (error != NULL)
13827 13833 *error = 0;
13828 13834
13829 13835 if (namelen == 0) {
13830 13836 if (error != NULL)
13831 13837 *error = ENXIO;
13832 13838 return (NULL);
13833 13839 }
13834 13840
13835 13841 /* Look for a colon in the name. */
13836 13842 endp = &name[namelen];
13837 13843 for (cp = endp; --cp > name; ) {
13838 13844 if (*cp == IPIF_SEPARATOR_CHAR)
13839 13845 break;
13840 13846 }
13841 13847
13842 13848 if (*cp == IPIF_SEPARATOR_CHAR) {
13843 13849 /*
13844 13850 * Reject any non-decimal aliases for logical
13845 13851 * interfaces. Aliases with leading zeroes
13846 13852 * are also rejected as they introduce ambiguity
13847 13853 * in the naming of the interfaces.
13848 13854 * In order to confirm with existing semantics,
13849 13855 * and to not break any programs/script relying
13850 13856 * on that behaviour, if<0>:0 is considered to be
13851 13857 * a valid interface.
13852 13858 *
13853 13859 * If alias has two or more digits and the first
13854 13860 * is zero, fail.
13855 13861 */
13856 13862 if (&cp[2] < endp && cp[1] == '0') {
13857 13863 if (error != NULL)
13858 13864 *error = EINVAL;
13859 13865 return (NULL);
13860 13866 }
13861 13867 }
13862 13868
13863 13869 if (cp <= name) {
13864 13870 cp = endp;
13865 13871 } else {
13866 13872 *cp = '\0';
13867 13873 }
13868 13874
13869 13875 /*
13870 13876 * Look up the ILL, based on the portion of the name
13871 13877 * before the slash. ill_lookup_on_name returns a held ill.
13872 13878 * Temporary to check whether ill exists already. If so
13873 13879 * ill_lookup_on_name will clear it.
13874 13880 */
13875 13881 ill = ill_lookup_on_name(name, B_FALSE, isv6, &did_alloc, ipst);
13876 13882 if (cp != endp)
13877 13883 *cp = IPIF_SEPARATOR_CHAR;
13878 13884 if (ill == NULL)
13879 13885 return (NULL);
13880 13886
13881 13887 /* Establish the unit number in the name. */
13882 13888 id = 0;
13883 13889 if (cp < endp && *endp == '\0') {
13884 13890 /* If there was a colon, the unit number follows. */
13885 13891 cp++;
13886 13892 if (ddi_strtol(cp, NULL, 0, &id) != 0) {
13887 13893 ill_refrele(ill);
13888 13894 if (error != NULL)
13889 13895 *error = ENXIO;
13890 13896 return (NULL);
13891 13897 }
13892 13898 }
13893 13899
13894 13900 GRAB_CONN_LOCK(q);
13895 13901 mutex_enter(&ill->ill_lock);
13896 13902 /* Now see if there is an IPIF with this unit number. */
13897 13903 for (ipif = ill->ill_ipif; ipif != NULL; ipif = ipif->ipif_next) {
13898 13904 if (ipif->ipif_id == id) {
13899 13905 if (zoneid != ALL_ZONES &&
13900 13906 zoneid != ipif->ipif_zoneid &&
13901 13907 ipif->ipif_zoneid != ALL_ZONES) {
13902 13908 mutex_exit(&ill->ill_lock);
13903 13909 RELEASE_CONN_LOCK(q);
13904 13910 ill_refrele(ill);
13905 13911 if (error != NULL)
13906 13912 *error = ENXIO;
13907 13913 return (NULL);
13908 13914 }
13909 13915
13910 13916 if (!(IPIF_IS_CHANGING(ipif) ||
13911 13917 IPIF_IS_CONDEMNED(ipif)) ||
13912 13918 IAM_WRITER_IPIF(ipif)) {
13913 13919 ipif_refhold_locked(ipif);
13914 13920 mutex_exit(&ill->ill_lock);
13915 13921 /*
13916 13922 * Drop locks before calling ill_refrele
13917 13923 * since it can potentially call into
13918 13924 * ipif_ill_refrele_tail which can end up
13919 13925 * in trying to acquire any lock.
13920 13926 */
13921 13927 RELEASE_CONN_LOCK(q);
13922 13928 ill_refrele(ill);
13923 13929 return (ipif);
13924 13930 } else if (q != NULL && !IPIF_IS_CONDEMNED(ipif)) {
13925 13931 ipsq = ill->ill_phyint->phyint_ipsq;
13926 13932 mutex_enter(&ipsq->ipsq_lock);
13927 13933 mutex_enter(&ipsq->ipsq_xop->ipx_lock);
13928 13934 mutex_exit(&ill->ill_lock);
13929 13935 ipsq_enq(ipsq, q, mp, func, NEW_OP, ill);
13930 13936 mutex_exit(&ipsq->ipsq_xop->ipx_lock);
13931 13937 mutex_exit(&ipsq->ipsq_lock);
13932 13938 RELEASE_CONN_LOCK(q);
13933 13939 ill_refrele(ill);
13934 13940 if (error != NULL)
13935 13941 *error = EINPROGRESS;
13936 13942 return (NULL);
13937 13943 }
13938 13944 }
13939 13945 }
13940 13946 RELEASE_CONN_LOCK(q);
13941 13947 mutex_exit(&ill->ill_lock);
13942 13948 ill_refrele(ill);
13943 13949 if (error != NULL)
13944 13950 *error = ENXIO;
13945 13951 return (NULL);
13946 13952 }
13947 13953
13948 13954 /*
13949 13955 * This routine is called whenever a new address comes up on an ipif. If
13950 13956 * we are configured to respond to address mask requests, then we are supposed
13951 13957 * to broadcast an address mask reply at this time. This routine is also
13952 13958 * called if we are already up, but a netmask change is made. This is legal
13953 13959 * but might not make the system manager very popular. (May be called
13954 13960 * as writer.)
13955 13961 */
13956 13962 void
13957 13963 ipif_mask_reply(ipif_t *ipif)
13958 13964 {
13959 13965 icmph_t *icmph;
13960 13966 ipha_t *ipha;
13961 13967 mblk_t *mp;
13962 13968 ip_stack_t *ipst = ipif->ipif_ill->ill_ipst;
13963 13969 ip_xmit_attr_t ixas;
13964 13970
13965 13971 #define REPLY_LEN (sizeof (icmp_ipha) + sizeof (icmph_t) + IP_ADDR_LEN)
13966 13972
13967 13973 if (!ipst->ips_ip_respond_to_address_mask_broadcast)
13968 13974 return;
13969 13975
13970 13976 /* ICMP mask reply is IPv4 only */
13971 13977 ASSERT(!ipif->ipif_isv6);
13972 13978 /* ICMP mask reply is not for a loopback interface */
13973 13979 ASSERT(ipif->ipif_ill->ill_wq != NULL);
13974 13980
13975 13981 if (ipif->ipif_lcl_addr == INADDR_ANY)
13976 13982 return;
13977 13983
13978 13984 mp = allocb(REPLY_LEN, BPRI_HI);
13979 13985 if (mp == NULL)
13980 13986 return;
13981 13987 mp->b_wptr = mp->b_rptr + REPLY_LEN;
13982 13988
13983 13989 ipha = (ipha_t *)mp->b_rptr;
13984 13990 bzero(ipha, REPLY_LEN);
13985 13991 *ipha = icmp_ipha;
13986 13992 ipha->ipha_ttl = ipst->ips_ip_broadcast_ttl;
13987 13993 ipha->ipha_src = ipif->ipif_lcl_addr;
13988 13994 ipha->ipha_dst = ipif->ipif_brd_addr;
13989 13995 ipha->ipha_length = htons(REPLY_LEN);
13990 13996 ipha->ipha_ident = 0;
13991 13997
13992 13998 icmph = (icmph_t *)&ipha[1];
13993 13999 icmph->icmph_type = ICMP_ADDRESS_MASK_REPLY;
13994 14000 bcopy(&ipif->ipif_net_mask, &icmph[1], IP_ADDR_LEN);
13995 14001 icmph->icmph_checksum = IP_CSUM(mp, sizeof (ipha_t), 0);
13996 14002
13997 14003 bzero(&ixas, sizeof (ixas));
13998 14004 ixas.ixa_flags = IXAF_BASIC_SIMPLE_V4;
13999 14005 ixas.ixa_zoneid = ALL_ZONES;
14000 14006 ixas.ixa_ifindex = 0;
14001 14007 ixas.ixa_ipst = ipst;
14002 14008 ixas.ixa_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
14003 14009 (void) ip_output_simple(mp, &ixas);
14004 14010 ixa_cleanup(&ixas);
14005 14011 #undef REPLY_LEN
14006 14012 }
14007 14013
14008 14014 /*
14009 14015 * Join the ipif specific multicast groups.
14010 14016 * Must be called after a mapping has been set up in the resolver. (Always
14011 14017 * called as writer.)
14012 14018 */
14013 14019 void
14014 14020 ipif_multicast_up(ipif_t *ipif)
14015 14021 {
14016 14022 int err;
14017 14023 ill_t *ill;
14018 14024 ilm_t *ilm;
14019 14025
14020 14026 ASSERT(IAM_WRITER_IPIF(ipif));
14021 14027
14022 14028 ill = ipif->ipif_ill;
14023 14029
14024 14030 ip1dbg(("ipif_multicast_up\n"));
14025 14031 if (!(ill->ill_flags & ILLF_MULTICAST) ||
14026 14032 ipif->ipif_allhosts_ilm != NULL)
14027 14033 return;
14028 14034
14029 14035 if (ipif->ipif_isv6) {
14030 14036 in6_addr_t v6allmc = ipv6_all_hosts_mcast;
14031 14037 in6_addr_t v6solmc = ipv6_solicited_node_mcast;
14032 14038
14033 14039 v6solmc.s6_addr32[3] |= ipif->ipif_v6lcl_addr.s6_addr32[3];
14034 14040
14035 14041 if (IN6_IS_ADDR_UNSPECIFIED(&ipif->ipif_v6lcl_addr))
14036 14042 return;
14037 14043
14038 14044 ip1dbg(("ipif_multicast_up - addmulti\n"));
14039 14045
14040 14046 /*
14041 14047 * Join the all hosts multicast address. We skip this for
14042 14048 * underlying IPMP interfaces since they should be invisible.
14043 14049 */
14044 14050 if (!IS_UNDER_IPMP(ill)) {
14045 14051 ilm = ip_addmulti(&v6allmc, ill, ipif->ipif_zoneid,
14046 14052 &err);
14047 14053 if (ilm == NULL) {
14048 14054 ASSERT(err != 0);
14049 14055 ip0dbg(("ipif_multicast_up: "
14050 14056 "all_hosts_mcast failed %d\n", err));
14051 14057 return;
14052 14058 }
14053 14059 ipif->ipif_allhosts_ilm = ilm;
14054 14060 }
14055 14061
14056 14062 /*
14057 14063 * Enable multicast for the solicited node multicast address.
14058 14064 * If IPMP we need to put the membership on the upper ill.
14059 14065 */
14060 14066 if (!(ipif->ipif_flags & IPIF_NOLOCAL)) {
14061 14067 ill_t *mcast_ill = NULL;
14062 14068 boolean_t need_refrele;
14063 14069
14064 14070 if (IS_UNDER_IPMP(ill) &&
14065 14071 (mcast_ill = ipmp_ill_hold_ipmp_ill(ill)) != NULL) {
14066 14072 need_refrele = B_TRUE;
14067 14073 } else {
14068 14074 mcast_ill = ill;
14069 14075 need_refrele = B_FALSE;
14070 14076 }
14071 14077
14072 14078 ilm = ip_addmulti(&v6solmc, mcast_ill,
14073 14079 ipif->ipif_zoneid, &err);
14074 14080 if (need_refrele)
14075 14081 ill_refrele(mcast_ill);
14076 14082
14077 14083 if (ilm == NULL) {
14078 14084 ASSERT(err != 0);
14079 14085 ip0dbg(("ipif_multicast_up: solicited MC"
14080 14086 " failed %d\n", err));
14081 14087 if ((ilm = ipif->ipif_allhosts_ilm) != NULL) {
14082 14088 ipif->ipif_allhosts_ilm = NULL;
14083 14089 (void) ip_delmulti(ilm);
14084 14090 }
14085 14091 return;
14086 14092 }
14087 14093 ipif->ipif_solmulti_ilm = ilm;
14088 14094 }
14089 14095 } else {
14090 14096 in6_addr_t v6group;
14091 14097
14092 14098 if (ipif->ipif_lcl_addr == INADDR_ANY || IS_UNDER_IPMP(ill))
14093 14099 return;
14094 14100
14095 14101 /* Join the all hosts multicast address */
14096 14102 ip1dbg(("ipif_multicast_up - addmulti\n"));
14097 14103 IN6_IPADDR_TO_V4MAPPED(htonl(INADDR_ALLHOSTS_GROUP), &v6group);
14098 14104
14099 14105 ilm = ip_addmulti(&v6group, ill, ipif->ipif_zoneid, &err);
14100 14106 if (ilm == NULL) {
14101 14107 ASSERT(err != 0);
14102 14108 ip0dbg(("ipif_multicast_up: failed %d\n", err));
14103 14109 return;
14104 14110 }
14105 14111 ipif->ipif_allhosts_ilm = ilm;
14106 14112 }
14107 14113 }
14108 14114
14109 14115 /*
14110 14116 * Blow away any multicast groups that we joined in ipif_multicast_up().
14111 14117 * (ilms from explicit memberships are handled in conn_update_ill.)
14112 14118 */
14113 14119 void
14114 14120 ipif_multicast_down(ipif_t *ipif)
14115 14121 {
14116 14122 ASSERT(IAM_WRITER_IPIF(ipif));
14117 14123
14118 14124 ip1dbg(("ipif_multicast_down\n"));
14119 14125
14120 14126 if (ipif->ipif_allhosts_ilm != NULL) {
14121 14127 (void) ip_delmulti(ipif->ipif_allhosts_ilm);
14122 14128 ipif->ipif_allhosts_ilm = NULL;
14123 14129 }
14124 14130 if (ipif->ipif_solmulti_ilm != NULL) {
14125 14131 (void) ip_delmulti(ipif->ipif_solmulti_ilm);
14126 14132 ipif->ipif_solmulti_ilm = NULL;
14127 14133 }
14128 14134 }
14129 14135
14130 14136 /*
14131 14137 * Used when an interface comes up to recreate any extra routes on this
14132 14138 * interface.
14133 14139 */
14134 14140 int
14135 14141 ill_recover_saved_ire(ill_t *ill)
14136 14142 {
14137 14143 mblk_t *mp;
14138 14144 ip_stack_t *ipst = ill->ill_ipst;
14139 14145
14140 14146 ip1dbg(("ill_recover_saved_ire(%s)", ill->ill_name));
14141 14147
14142 14148 mutex_enter(&ill->ill_saved_ire_lock);
14143 14149 for (mp = ill->ill_saved_ire_mp; mp != NULL; mp = mp->b_cont) {
14144 14150 ire_t *ire, *nire;
14145 14151 ifrt_t *ifrt;
14146 14152
14147 14153 ifrt = (ifrt_t *)mp->b_rptr;
14148 14154 /*
14149 14155 * Create a copy of the IRE with the saved address and netmask.
14150 14156 */
14151 14157 if (ill->ill_isv6) {
14152 14158 ire = ire_create_v6(
14153 14159 &ifrt->ifrt_v6addr,
14154 14160 &ifrt->ifrt_v6mask,
14155 14161 &ifrt->ifrt_v6gateway_addr,
14156 14162 ifrt->ifrt_type,
14157 14163 ill,
14158 14164 ifrt->ifrt_zoneid,
14159 14165 ifrt->ifrt_flags,
14160 14166 NULL,
14161 14167 ipst);
14162 14168 } else {
14163 14169 ire = ire_create(
14164 14170 (uint8_t *)&ifrt->ifrt_addr,
14165 14171 (uint8_t *)&ifrt->ifrt_mask,
14166 14172 (uint8_t *)&ifrt->ifrt_gateway_addr,
14167 14173 ifrt->ifrt_type,
14168 14174 ill,
14169 14175 ifrt->ifrt_zoneid,
14170 14176 ifrt->ifrt_flags,
14171 14177 NULL,
14172 14178 ipst);
14173 14179 }
14174 14180 if (ire == NULL) {
14175 14181 mutex_exit(&ill->ill_saved_ire_lock);
14176 14182 return (ENOMEM);
14177 14183 }
14178 14184
14179 14185 if (ifrt->ifrt_flags & RTF_SETSRC) {
14180 14186 if (ill->ill_isv6) {
14181 14187 ire->ire_setsrc_addr_v6 =
14182 14188 ifrt->ifrt_v6setsrc_addr;
14183 14189 } else {
14184 14190 ire->ire_setsrc_addr = ifrt->ifrt_setsrc_addr;
14185 14191 }
14186 14192 }
14187 14193
14188 14194 /*
14189 14195 * Some software (for example, GateD and Sun Cluster) attempts
14190 14196 * to create (what amount to) IRE_PREFIX routes with the
14191 14197 * loopback address as the gateway. This is primarily done to
14192 14198 * set up prefixes with the RTF_REJECT flag set (for example,
14193 14199 * when generating aggregate routes.)
14194 14200 *
14195 14201 * If the IRE type (as defined by ill->ill_net_type) is
14196 14202 * IRE_LOOPBACK, then we map the request into a
14197 14203 * IRE_IF_NORESOLVER.
14198 14204 */
14199 14205 if (ill->ill_net_type == IRE_LOOPBACK)
14200 14206 ire->ire_type = IRE_IF_NORESOLVER;
14201 14207
14202 14208 /*
14203 14209 * ire held by ire_add, will be refreled' towards the
14204 14210 * the end of ipif_up_done
14205 14211 */
14206 14212 nire = ire_add(ire);
14207 14213 /*
14208 14214 * Check if it was a duplicate entry. This handles
14209 14215 * the case of two racing route adds for the same route
14210 14216 */
14211 14217 if (nire == NULL) {
14212 14218 ip1dbg(("ill_recover_saved_ire: FAILED\n"));
14213 14219 } else if (nire != ire) {
14214 14220 ip1dbg(("ill_recover_saved_ire: duplicate ire %p\n",
14215 14221 (void *)nire));
14216 14222 ire_delete(nire);
14217 14223 } else {
14218 14224 ip1dbg(("ill_recover_saved_ire: added ire %p\n",
14219 14225 (void *)nire));
14220 14226 }
14221 14227 if (nire != NULL)
14222 14228 ire_refrele(nire);
14223 14229 }
14224 14230 mutex_exit(&ill->ill_saved_ire_lock);
14225 14231 return (0);
14226 14232 }
14227 14233
14228 14234 /*
14229 14235 * Used to set the netmask and broadcast address to default values when the
14230 14236 * interface is brought up. (Always called as writer.)
14231 14237 */
14232 14238 static void
14233 14239 ipif_set_default(ipif_t *ipif)
14234 14240 {
14235 14241 ASSERT(MUTEX_HELD(&ipif->ipif_ill->ill_lock));
14236 14242
14237 14243 if (!ipif->ipif_isv6) {
14238 14244 /*
14239 14245 * Interface holds an IPv4 address. Default
14240 14246 * mask is the natural netmask.
14241 14247 */
14242 14248 if (!ipif->ipif_net_mask) {
14243 14249 ipaddr_t v4mask;
14244 14250
14245 14251 v4mask = ip_net_mask(ipif->ipif_lcl_addr);
14246 14252 V4MASK_TO_V6(v4mask, ipif->ipif_v6net_mask);
14247 14253 }
14248 14254 if (ipif->ipif_flags & IPIF_POINTOPOINT) {
14249 14255 /* ipif_subnet is ipif_pp_dst_addr for pt-pt */
14250 14256 ipif->ipif_v6subnet = ipif->ipif_v6pp_dst_addr;
14251 14257 } else {
14252 14258 V6_MASK_COPY(ipif->ipif_v6lcl_addr,
14253 14259 ipif->ipif_v6net_mask, ipif->ipif_v6subnet);
14254 14260 }
14255 14261 /*
14256 14262 * NOTE: SunOS 4.X does this even if the broadcast address
14257 14263 * has been already set thus we do the same here.
14258 14264 */
14259 14265 if (ipif->ipif_flags & IPIF_BROADCAST) {
14260 14266 ipaddr_t v4addr;
14261 14267
14262 14268 v4addr = ipif->ipif_subnet | ~ipif->ipif_net_mask;
14263 14269 IN6_IPADDR_TO_V4MAPPED(v4addr, &ipif->ipif_v6brd_addr);
14264 14270 }
14265 14271 } else {
14266 14272 /*
14267 14273 * Interface holds an IPv6-only address. Default
14268 14274 * mask is all-ones.
14269 14275 */
14270 14276 if (IN6_IS_ADDR_UNSPECIFIED(&ipif->ipif_v6net_mask))
14271 14277 ipif->ipif_v6net_mask = ipv6_all_ones;
14272 14278 if (ipif->ipif_flags & IPIF_POINTOPOINT) {
14273 14279 /* ipif_subnet is ipif_pp_dst_addr for pt-pt */
14274 14280 ipif->ipif_v6subnet = ipif->ipif_v6pp_dst_addr;
14275 14281 } else {
14276 14282 V6_MASK_COPY(ipif->ipif_v6lcl_addr,
14277 14283 ipif->ipif_v6net_mask, ipif->ipif_v6subnet);
14278 14284 }
14279 14285 }
14280 14286 }
14281 14287
14282 14288 /*
14283 14289 * Return 0 if this address can be used as local address without causing
14284 14290 * duplicate address problems. Otherwise, return EADDRNOTAVAIL if the address
14285 14291 * is already up on a different ill, and EADDRINUSE if it's up on the same ill.
14286 14292 * Note that the same IPv6 link-local address is allowed as long as the ills
14287 14293 * are not on the same link.
14288 14294 */
14289 14295 int
14290 14296 ip_addr_availability_check(ipif_t *new_ipif)
14291 14297 {
14292 14298 in6_addr_t our_v6addr;
14293 14299 ill_t *ill;
14294 14300 ipif_t *ipif;
14295 14301 ill_walk_context_t ctx;
14296 14302 ip_stack_t *ipst = new_ipif->ipif_ill->ill_ipst;
14297 14303
14298 14304 ASSERT(IAM_WRITER_IPIF(new_ipif));
14299 14305 ASSERT(MUTEX_HELD(&ipst->ips_ip_addr_avail_lock));
14300 14306 ASSERT(RW_READ_HELD(&ipst->ips_ill_g_lock));
14301 14307
14302 14308 new_ipif->ipif_flags &= ~IPIF_UNNUMBERED;
14303 14309 if (IN6_IS_ADDR_UNSPECIFIED(&new_ipif->ipif_v6lcl_addr) ||
14304 14310 IN6_IS_ADDR_V4MAPPED_ANY(&new_ipif->ipif_v6lcl_addr))
14305 14311 return (0);
14306 14312
14307 14313 our_v6addr = new_ipif->ipif_v6lcl_addr;
14308 14314
14309 14315 if (new_ipif->ipif_isv6)
14310 14316 ill = ILL_START_WALK_V6(&ctx, ipst);
14311 14317 else
14312 14318 ill = ILL_START_WALK_V4(&ctx, ipst);
14313 14319
14314 14320 for (; ill != NULL; ill = ill_next(&ctx, ill)) {
14315 14321 for (ipif = ill->ill_ipif; ipif != NULL;
14316 14322 ipif = ipif->ipif_next) {
14317 14323 if ((ipif == new_ipif) ||
14318 14324 !(ipif->ipif_flags & IPIF_UP) ||
14319 14325 (ipif->ipif_flags & IPIF_UNNUMBERED) ||
14320 14326 !IN6_ARE_ADDR_EQUAL(&ipif->ipif_v6lcl_addr,
14321 14327 &our_v6addr))
14322 14328 continue;
14323 14329
14324 14330 if (new_ipif->ipif_flags & IPIF_POINTOPOINT)
14325 14331 new_ipif->ipif_flags |= IPIF_UNNUMBERED;
14326 14332 else if (ipif->ipif_flags & IPIF_POINTOPOINT)
14327 14333 ipif->ipif_flags |= IPIF_UNNUMBERED;
14328 14334 else if ((IN6_IS_ADDR_LINKLOCAL(&our_v6addr) ||
14329 14335 IN6_IS_ADDR_SITELOCAL(&our_v6addr)) &&
14330 14336 !IS_ON_SAME_LAN(ill, new_ipif->ipif_ill))
14331 14337 continue;
14332 14338 else if (new_ipif->ipif_zoneid != ipif->ipif_zoneid &&
14333 14339 ipif->ipif_zoneid != ALL_ZONES && IS_LOOPBACK(ill))
14334 14340 continue;
14335 14341 else if (new_ipif->ipif_ill == ill)
14336 14342 return (EADDRINUSE);
14337 14343 else
14338 14344 return (EADDRNOTAVAIL);
14339 14345 }
14340 14346 }
14341 14347
14342 14348 return (0);
14343 14349 }
14344 14350
14345 14351 /*
14346 14352 * Bring up an ipif: bring up arp/ndp, bring up the DLPI stream, and add
14347 14353 * IREs for the ipif.
14348 14354 * When the routine returns EINPROGRESS then mp has been consumed and
14349 14355 * the ioctl will be acked from ip_rput_dlpi.
14350 14356 */
14351 14357 int
14352 14358 ipif_up(ipif_t *ipif, queue_t *q, mblk_t *mp)
14353 14359 {
14354 14360 ill_t *ill = ipif->ipif_ill;
14355 14361 boolean_t isv6 = ipif->ipif_isv6;
14356 14362 int err = 0;
14357 14363 boolean_t success;
14358 14364 uint_t ipif_orig_id;
14359 14365 ip_stack_t *ipst = ill->ill_ipst;
14360 14366
14361 14367 ASSERT(IAM_WRITER_IPIF(ipif));
14362 14368
14363 14369 ip1dbg(("ipif_up(%s:%u)\n", ill->ill_name, ipif->ipif_id));
14364 14370 DTRACE_PROBE3(ipif__downup, char *, "ipif_up",
14365 14371 ill_t *, ill, ipif_t *, ipif);
14366 14372
14367 14373 /* Shouldn't get here if it is already up. */
14368 14374 if (ipif->ipif_flags & IPIF_UP)
14369 14375 return (EALREADY);
14370 14376
14371 14377 /*
14372 14378 * If this is a request to bring up a data address on an interface
14373 14379 * under IPMP, then move the address to its IPMP meta-interface and
14374 14380 * try to bring it up. One complication is that the zeroth ipif for
14375 14381 * an ill is special, in that every ill always has one, and that code
14376 14382 * throughout IP deferences ill->ill_ipif without holding any locks.
14377 14383 */
14378 14384 if (IS_UNDER_IPMP(ill) && ipmp_ipif_is_dataaddr(ipif) &&
14379 14385 (!ipif->ipif_isv6 || !V6_IPIF_LINKLOCAL(ipif))) {
14380 14386 ipif_t *stubipif = NULL, *moveipif = NULL;
14381 14387 ill_t *ipmp_ill = ipmp_illgrp_ipmp_ill(ill->ill_grp);
14382 14388
14383 14389 /*
14384 14390 * The ipif being brought up should be quiesced. If it's not,
14385 14391 * something has gone amiss and we need to bail out. (If it's
14386 14392 * quiesced, we know it will remain so via IPIF_CONDEMNED.)
14387 14393 */
14388 14394 mutex_enter(&ill->ill_lock);
14389 14395 if (!ipif_is_quiescent(ipif)) {
14390 14396 mutex_exit(&ill->ill_lock);
14391 14397 return (EINVAL);
14392 14398 }
14393 14399 mutex_exit(&ill->ill_lock);
14394 14400
14395 14401 /*
14396 14402 * If we're going to need to allocate ipifs, do it prior
14397 14403 * to starting the move (and grabbing locks).
14398 14404 */
14399 14405 if (ipif->ipif_id == 0) {
14400 14406 if ((moveipif = ipif_allocate(ill, 0, IRE_LOCAL, B_TRUE,
14401 14407 B_FALSE, &err)) == NULL) {
14402 14408 return (err);
14403 14409 }
14404 14410 if ((stubipif = ipif_allocate(ill, 0, IRE_LOCAL, B_TRUE,
14405 14411 B_FALSE, &err)) == NULL) {
14406 14412 mi_free(moveipif);
14407 14413 return (err);
14408 14414 }
14409 14415 }
14410 14416
14411 14417 /*
14412 14418 * Grab or transfer the ipif to move. During the move, keep
14413 14419 * ill_g_lock held to prevent any ill walker threads from
14414 14420 * seeing things in an inconsistent state.
14415 14421 */
14416 14422 rw_enter(&ipst->ips_ill_g_lock, RW_WRITER);
14417 14423 if (ipif->ipif_id != 0) {
14418 14424 ipif_remove(ipif);
14419 14425 } else {
14420 14426 ipif_transfer(ipif, moveipif, stubipif);
14421 14427 ipif = moveipif;
14422 14428 }
14423 14429
14424 14430 /*
14425 14431 * Place the ipif on the IPMP ill. If the zeroth ipif on
14426 14432 * the IPMP ill is a stub (0.0.0.0 down address) then we
14427 14433 * replace that one. Otherwise, pick the next available slot.
14428 14434 */
14429 14435 ipif->ipif_ill = ipmp_ill;
14430 14436 ipif_orig_id = ipif->ipif_id;
14431 14437
14432 14438 if (ipmp_ipif_is_stubaddr(ipmp_ill->ill_ipif)) {
14433 14439 ipif_transfer(ipif, ipmp_ill->ill_ipif, NULL);
14434 14440 ipif = ipmp_ill->ill_ipif;
14435 14441 } else {
14436 14442 ipif->ipif_id = -1;
14437 14443 if ((err = ipif_insert(ipif, B_FALSE)) != 0) {
14438 14444 /*
14439 14445 * No more available ipif_id's -- put it back
14440 14446 * on the original ill and fail the operation.
14441 14447 * Since we're writer on the ill, we can be
14442 14448 * sure our old slot is still available.
14443 14449 */
14444 14450 ipif->ipif_id = ipif_orig_id;
14445 14451 ipif->ipif_ill = ill;
14446 14452 if (ipif_orig_id == 0) {
14447 14453 ipif_transfer(ipif, ill->ill_ipif,
14448 14454 NULL);
14449 14455 } else {
14450 14456 VERIFY(ipif_insert(ipif, B_FALSE) == 0);
14451 14457 }
14452 14458 rw_exit(&ipst->ips_ill_g_lock);
14453 14459 return (err);
14454 14460 }
14455 14461 }
14456 14462 rw_exit(&ipst->ips_ill_g_lock);
14457 14463
14458 14464 /*
14459 14465 * Tell SCTP that the ipif has moved. Note that even if we
14460 14466 * had to allocate a new ipif, the original sequence id was
14461 14467 * preserved and therefore SCTP won't know.
14462 14468 */
14463 14469 sctp_move_ipif(ipif, ill, ipmp_ill);
14464 14470
14465 14471 /*
14466 14472 * If the ipif being brought up was on slot zero, then we
14467 14473 * first need to bring up the placeholder we stuck there. In
14468 14474 * ip_rput_dlpi_writer(), arp_bringup_done(), or the recursive
14469 14475 * call to ipif_up() itself, if we successfully bring up the
14470 14476 * placeholder, we'll check ill_move_ipif and bring it up too.
14471 14477 */
14472 14478 if (ipif_orig_id == 0) {
14473 14479 ASSERT(ill->ill_move_ipif == NULL);
14474 14480 ill->ill_move_ipif = ipif;
14475 14481 if ((err = ipif_up(ill->ill_ipif, q, mp)) == 0)
14476 14482 ASSERT(ill->ill_move_ipif == NULL);
14477 14483 if (err != EINPROGRESS)
14478 14484 ill->ill_move_ipif = NULL;
14479 14485 return (err);
14480 14486 }
14481 14487
14482 14488 /*
14483 14489 * Bring it up on the IPMP ill.
14484 14490 */
14485 14491 return (ipif_up(ipif, q, mp));
14486 14492 }
14487 14493
14488 14494 /* Skip arp/ndp for any loopback interface. */
14489 14495 if (ill->ill_wq != NULL) {
14490 14496 conn_t *connp = CONN_Q(q) ? Q_TO_CONN(q) : NULL;
14491 14497 ipsq_t *ipsq = ill->ill_phyint->phyint_ipsq;
14492 14498
14493 14499 if (!ill->ill_dl_up) {
14494 14500 /*
14495 14501 * ill_dl_up is not yet set. i.e. we are yet to
14496 14502 * DL_BIND with the driver and this is the first
14497 14503 * logical interface on the ill to become "up".
14498 14504 * Tell the driver to get going (via DL_BIND_REQ).
14499 14505 * Note that changing "significant" IFF_ flags
14500 14506 * address/netmask etc cause a down/up dance, but
14501 14507 * does not cause an unbind (DL_UNBIND) with the driver
14502 14508 */
14503 14509 return (ill_dl_up(ill, ipif, mp, q));
14504 14510 }
14505 14511
14506 14512 /*
14507 14513 * ipif_resolver_up may end up needeing to bind/attach
14508 14514 * the ARP stream, which in turn necessitates a
14509 14515 * DLPI message exchange with the driver. ioctls are
14510 14516 * serialized and so we cannot send more than one
14511 14517 * interface up message at a time. If ipif_resolver_up
14512 14518 * does need to wait for the DLPI handshake for the ARP stream,
14513 14519 * we get EINPROGRESS and we will complete in arp_bringup_done.
14514 14520 */
14515 14521
14516 14522 ASSERT(connp != NULL || !CONN_Q(q));
14517 14523 if (connp != NULL)
14518 14524 mutex_enter(&connp->conn_lock);
14519 14525 mutex_enter(&ill->ill_lock);
14520 14526 success = ipsq_pending_mp_add(connp, ipif, q, mp, 0);
14521 14527 mutex_exit(&ill->ill_lock);
14522 14528 if (connp != NULL)
14523 14529 mutex_exit(&connp->conn_lock);
14524 14530 if (!success)
14525 14531 return (EINTR);
14526 14532
14527 14533 /*
14528 14534 * Crank up IPv6 neighbor discovery. Unlike ARP, this should
14529 14535 * complete when ipif_ndp_up returns.
14530 14536 */
14531 14537 err = ipif_resolver_up(ipif, Res_act_initial);
14532 14538 if (err == EINPROGRESS) {
14533 14539 /* We will complete it in arp_bringup_done() */
14534 14540 return (err);
14535 14541 }
14536 14542
14537 14543 if (isv6 && err == 0)
14538 14544 err = ipif_ndp_up(ipif, B_TRUE);
14539 14545
14540 14546 ASSERT(err != EINPROGRESS);
14541 14547 mp = ipsq_pending_mp_get(ipsq, &connp);
14542 14548 ASSERT(mp != NULL);
14543 14549 if (err != 0)
14544 14550 return (err);
14545 14551 } else {
14546 14552 /*
14547 14553 * Interfaces without underlying hardware don't do duplicate
14548 14554 * address detection.
14549 14555 */
14550 14556 ASSERT(!(ipif->ipif_flags & IPIF_DUPLICATE));
14551 14557 ipif->ipif_addr_ready = 1;
14552 14558 err = ill_add_ires(ill);
14553 14559 /* allocation failure? */
14554 14560 if (err != 0)
14555 14561 return (err);
14556 14562 }
14557 14563
14558 14564 err = (isv6 ? ipif_up_done_v6(ipif) : ipif_up_done(ipif));
14559 14565 if (err == 0 && ill->ill_move_ipif != NULL) {
14560 14566 ipif = ill->ill_move_ipif;
14561 14567 ill->ill_move_ipif = NULL;
14562 14568 return (ipif_up(ipif, q, mp));
14563 14569 }
14564 14570 return (err);
14565 14571 }
14566 14572
14567 14573 /*
14568 14574 * Add any IREs tied to the ill. For now this is just an IRE_MULTICAST.
14569 14575 * The identical set of IREs need to be removed in ill_delete_ires().
14570 14576 */
14571 14577 int
14572 14578 ill_add_ires(ill_t *ill)
14573 14579 {
14574 14580 ire_t *ire;
14575 14581 in6_addr_t dummy6 = {(uint32_t)V6_MCAST, 0, 0, 1};
14576 14582 in_addr_t dummy4 = htonl(INADDR_ALLHOSTS_GROUP);
14577 14583
14578 14584 if (ill->ill_ire_multicast != NULL)
14579 14585 return (0);
14580 14586
14581 14587 /*
14582 14588 * provide some dummy ire_addr for creating the ire.
14583 14589 */
14584 14590 if (ill->ill_isv6) {
14585 14591 ire = ire_create_v6(&dummy6, 0, 0, IRE_MULTICAST, ill,
14586 14592 ALL_ZONES, RTF_UP, NULL, ill->ill_ipst);
14587 14593 } else {
14588 14594 ire = ire_create((uchar_t *)&dummy4, 0, 0, IRE_MULTICAST, ill,
14589 14595 ALL_ZONES, RTF_UP, NULL, ill->ill_ipst);
14590 14596 }
14591 14597 if (ire == NULL)
14592 14598 return (ENOMEM);
14593 14599
14594 14600 ill->ill_ire_multicast = ire;
14595 14601 return (0);
14596 14602 }
14597 14603
14598 14604 void
14599 14605 ill_delete_ires(ill_t *ill)
14600 14606 {
14601 14607 if (ill->ill_ire_multicast != NULL) {
14602 14608 /*
14603 14609 * BIND/ATTACH completed; Release the ref for ill_ire_multicast
14604 14610 * which was taken without any th_tracing enabled.
14605 14611 * We also mark it as condemned (note that it was never added)
14606 14612 * so that caching conn's can move off of it.
14607 14613 */
14608 14614 ire_make_condemned(ill->ill_ire_multicast);
14609 14615 ire_refrele_notr(ill->ill_ire_multicast);
14610 14616 ill->ill_ire_multicast = NULL;
14611 14617 }
14612 14618 }
14613 14619
14614 14620 /*
14615 14621 * Perform a bind for the physical device.
14616 14622 * When the routine returns EINPROGRESS then mp has been consumed and
14617 14623 * the ioctl will be acked from ip_rput_dlpi.
14618 14624 * Allocate an unbind message and save it until ipif_down.
14619 14625 */
14620 14626 static int
14621 14627 ill_dl_up(ill_t *ill, ipif_t *ipif, mblk_t *mp, queue_t *q)
14622 14628 {
14623 14629 mblk_t *bind_mp = NULL;
14624 14630 mblk_t *unbind_mp = NULL;
14625 14631 conn_t *connp;
14626 14632 boolean_t success;
14627 14633 int err;
14628 14634
14629 14635 DTRACE_PROBE2(ill__downup, char *, "ill_dl_up", ill_t *, ill);
14630 14636
14631 14637 ip1dbg(("ill_dl_up(%s)\n", ill->ill_name));
14632 14638 ASSERT(IAM_WRITER_ILL(ill));
14633 14639 ASSERT(mp != NULL);
14634 14640
14635 14641 /*
14636 14642 * Make sure we have an IRE_MULTICAST in case we immediately
14637 14643 * start receiving packets.
14638 14644 */
14639 14645 err = ill_add_ires(ill);
14640 14646 if (err != 0)
14641 14647 goto bad;
14642 14648
14643 14649 bind_mp = ip_dlpi_alloc(sizeof (dl_bind_req_t) + sizeof (long),
14644 14650 DL_BIND_REQ);
14645 14651 if (bind_mp == NULL)
14646 14652 goto bad;
14647 14653 ((dl_bind_req_t *)bind_mp->b_rptr)->dl_sap = ill->ill_sap;
14648 14654 ((dl_bind_req_t *)bind_mp->b_rptr)->dl_service_mode = DL_CLDLS;
14649 14655
14650 14656 /*
14651 14657 * ill_unbind_mp would be non-null if the following sequence had
14652 14658 * happened:
14653 14659 * - send DL_BIND_REQ to driver, wait for response
14654 14660 * - multiple ioctls that need to bring the ipif up are encountered,
14655 14661 * but they cannot enter the ipsq due to the outstanding DL_BIND_REQ.
14656 14662 * These ioctls will then be enqueued on the ipsq
14657 14663 * - a DL_ERROR_ACK is returned for the DL_BIND_REQ
14658 14664 * At this point, the pending ioctls in the ipsq will be drained, and
14659 14665 * since ill->ill_dl_up was not set, ill_dl_up would be invoked with
14660 14666 * a non-null ill->ill_unbind_mp
14661 14667 */
14662 14668 if (ill->ill_unbind_mp == NULL) {
14663 14669 unbind_mp = ip_dlpi_alloc(sizeof (dl_unbind_req_t),
14664 14670 DL_UNBIND_REQ);
14665 14671 if (unbind_mp == NULL)
14666 14672 goto bad;
14667 14673 }
14668 14674 /*
14669 14675 * Record state needed to complete this operation when the
14670 14676 * DL_BIND_ACK shows up. Also remember the pre-allocated mblks.
14671 14677 */
14672 14678 connp = CONN_Q(q) ? Q_TO_CONN(q) : NULL;
14673 14679 ASSERT(connp != NULL || !CONN_Q(q));
14674 14680 GRAB_CONN_LOCK(q);
14675 14681 mutex_enter(&ipif->ipif_ill->ill_lock);
14676 14682 success = ipsq_pending_mp_add(connp, ipif, q, mp, 0);
14677 14683 mutex_exit(&ipif->ipif_ill->ill_lock);
14678 14684 RELEASE_CONN_LOCK(q);
14679 14685 if (!success)
14680 14686 goto bad;
14681 14687
14682 14688 /*
14683 14689 * Save the unbind message for ill_dl_down(); it will be consumed when
14684 14690 * the interface goes down.
14685 14691 */
14686 14692 if (ill->ill_unbind_mp == NULL)
14687 14693 ill->ill_unbind_mp = unbind_mp;
14688 14694
14689 14695 ill_dlpi_send(ill, bind_mp);
14690 14696 /* Send down link-layer capabilities probe if not already done. */
14691 14697 ill_capability_probe(ill);
14692 14698
14693 14699 /*
14694 14700 * Sysid used to rely on the fact that netboots set domainname
14695 14701 * and the like. Now that miniroot boots aren't strictly netboots
14696 14702 * and miniroot network configuration is driven from userland
14697 14703 * these things still need to be set. This situation can be detected
14698 14704 * by comparing the interface being configured here to the one
14699 14705 * dhcifname was set to reference by the boot loader. Once sysid is
14700 14706 * converted to use dhcp_ipc_getinfo() this call can go away.
14701 14707 */
14702 14708 if ((ipif->ipif_flags & IPIF_DHCPRUNNING) &&
14703 14709 (strcmp(ill->ill_name, dhcifname) == 0) &&
14704 14710 (strlen(srpc_domain) == 0)) {
14705 14711 if (dhcpinit() != 0)
14706 14712 cmn_err(CE_WARN, "no cached dhcp response");
14707 14713 }
14708 14714
14709 14715 /*
14710 14716 * This operation will complete in ip_rput_dlpi with either
14711 14717 * a DL_BIND_ACK or DL_ERROR_ACK.
14712 14718 */
14713 14719 return (EINPROGRESS);
14714 14720 bad:
14715 14721 ip1dbg(("ill_dl_up(%s) FAILED\n", ill->ill_name));
14716 14722
14717 14723 freemsg(bind_mp);
14718 14724 freemsg(unbind_mp);
14719 14725 return (ENOMEM);
14720 14726 }
14721 14727
14722 14728 /* Add room for tcp+ip headers */
14723 14729 uint_t ip_loopback_mtuplus = IP_LOOPBACK_MTU + IP_SIMPLE_HDR_LENGTH + 20;
14724 14730
14725 14731 /*
14726 14732 * DLPI and ARP is up.
14727 14733 * Create all the IREs associated with an interface. Bring up multicast.
14728 14734 * Set the interface flag and finish other initialization
14729 14735 * that potentially had to be deferred to after DL_BIND_ACK.
14730 14736 */
14731 14737 int
14732 14738 ipif_up_done(ipif_t *ipif)
14733 14739 {
14734 14740 ill_t *ill = ipif->ipif_ill;
14735 14741 int err = 0;
14736 14742 boolean_t loopback = B_FALSE;
14737 14743 boolean_t update_src_selection = B_TRUE;
14738 14744 ipif_t *tmp_ipif;
14739 14745
14740 14746 ip1dbg(("ipif_up_done(%s:%u)\n",
14741 14747 ipif->ipif_ill->ill_name, ipif->ipif_id));
14742 14748 DTRACE_PROBE3(ipif__downup, char *, "ipif_up_done",
14743 14749 ill_t *, ill, ipif_t *, ipif);
14744 14750
14745 14751 /* Check if this is a loopback interface */
14746 14752 if (ipif->ipif_ill->ill_wq == NULL)
14747 14753 loopback = B_TRUE;
14748 14754
14749 14755 ASSERT(!MUTEX_HELD(&ipif->ipif_ill->ill_lock));
14750 14756
14751 14757 /*
14752 14758 * If all other interfaces for this ill are down or DEPRECATED,
14753 14759 * or otherwise unsuitable for source address selection,
14754 14760 * reset the src generation numbers to make sure source
14755 14761 * address selection gets to take this new ipif into account.
14756 14762 * No need to hold ill_lock while traversing the ipif list since
14757 14763 * we are writer
14758 14764 */
14759 14765 for (tmp_ipif = ill->ill_ipif; tmp_ipif;
14760 14766 tmp_ipif = tmp_ipif->ipif_next) {
14761 14767 if (((tmp_ipif->ipif_flags &
14762 14768 (IPIF_NOXMIT|IPIF_ANYCAST|IPIF_NOLOCAL|IPIF_DEPRECATED)) ||
14763 14769 !(tmp_ipif->ipif_flags & IPIF_UP)) ||
14764 14770 (tmp_ipif == ipif))
14765 14771 continue;
14766 14772 /* first useable pre-existing interface */
14767 14773 update_src_selection = B_FALSE;
14768 14774 break;
14769 14775 }
14770 14776 if (update_src_selection)
14771 14777 ip_update_source_selection(ill->ill_ipst);
14772 14778
14773 14779 if (IS_LOOPBACK(ill) || ill->ill_net_type == IRE_IF_NORESOLVER) {
14774 14780 nce_t *loop_nce = NULL;
14775 14781 uint16_t flags = (NCE_F_MYADDR | NCE_F_AUTHORITY | NCE_F_NONUD);
14776 14782
14777 14783 /*
14778 14784 * lo0:1 and subsequent ipifs were marked IRE_LOCAL in
14779 14785 * ipif_lookup_on_name(), but in the case of zones we can have
14780 14786 * several loopback addresses on lo0. So all the interfaces with
14781 14787 * loopback addresses need to be marked IRE_LOOPBACK.
14782 14788 */
14783 14789 if (V4_PART_OF_V6(ipif->ipif_v6lcl_addr) ==
14784 14790 htonl(INADDR_LOOPBACK))
14785 14791 ipif->ipif_ire_type = IRE_LOOPBACK;
14786 14792 else
14787 14793 ipif->ipif_ire_type = IRE_LOCAL;
14788 14794 if (ill->ill_net_type != IRE_LOOPBACK)
14789 14795 flags |= NCE_F_PUBLISH;
14790 14796
14791 14797 /* add unicast nce for the local addr */
14792 14798 err = nce_lookup_then_add_v4(ill, NULL,
14793 14799 ill->ill_phys_addr_length, &ipif->ipif_lcl_addr, flags,
14794 14800 ND_REACHABLE, &loop_nce);
14795 14801 /* A shared-IP zone sees EEXIST for lo0:N */
14796 14802 if (err == 0 || err == EEXIST) {
14797 14803 ipif->ipif_added_nce = 1;
14798 14804 loop_nce->nce_ipif_cnt++;
14799 14805 nce_refrele(loop_nce);
14800 14806 err = 0;
14801 14807 } else {
14802 14808 ASSERT(loop_nce == NULL);
14803 14809 return (err);
14804 14810 }
14805 14811 }
14806 14812
14807 14813 /* Create all the IREs associated with this interface */
14808 14814 err = ipif_add_ires_v4(ipif, loopback);
14809 14815 if (err != 0) {
14810 14816 /*
14811 14817 * see comments about return value from
14812 14818 * ip_addr_availability_check() in ipif_add_ires_v4().
14813 14819 */
14814 14820 if (err != EADDRINUSE) {
14815 14821 (void) ipif_arp_down(ipif);
14816 14822 } else {
14817 14823 /*
14818 14824 * Make IPMP aware of the deleted ipif so that
14819 14825 * the needed ipmp cleanup (e.g., of ipif_bound_ill)
14820 14826 * can be completed. Note that we do not want to
14821 14827 * destroy the nce that was created on the ipmp_ill
14822 14828 * for the active copy of the duplicate address in
14823 14829 * use.
14824 14830 */
14825 14831 if (IS_IPMP(ill))
14826 14832 ipmp_illgrp_del_ipif(ill->ill_grp, ipif);
14827 14833 err = EADDRNOTAVAIL;
14828 14834 }
14829 14835 return (err);
14830 14836 }
14831 14837
14832 14838 if (ill->ill_ipif_up_count == 1 && !loopback) {
14833 14839 /* Recover any additional IREs entries for this ill */
14834 14840 (void) ill_recover_saved_ire(ill);
14835 14841 }
14836 14842
14837 14843 if (ill->ill_need_recover_multicast) {
14838 14844 /*
14839 14845 * Need to recover all multicast memberships in the driver.
14840 14846 * This had to be deferred until we had attached. The same
14841 14847 * code exists in ipif_up_done_v6() to recover IPv6
14842 14848 * memberships.
14843 14849 *
14844 14850 * Note that it would be preferable to unconditionally do the
14845 14851 * ill_recover_multicast() in ill_dl_up(), but we cannot do
14846 14852 * that since ill_join_allmulti() depends on ill_dl_up being
14847 14853 * set, and it is not set until we receive a DL_BIND_ACK after
14848 14854 * having called ill_dl_up().
14849 14855 */
14850 14856 ill_recover_multicast(ill);
14851 14857 }
14852 14858
14853 14859 if (ill->ill_ipif_up_count == 1) {
14854 14860 /*
14855 14861 * Since the interface is now up, it may now be active.
14856 14862 */
14857 14863 if (IS_UNDER_IPMP(ill))
14858 14864 ipmp_ill_refresh_active(ill);
14859 14865
14860 14866 /*
14861 14867 * If this is an IPMP interface, we may now be able to
14862 14868 * establish ARP entries.
14863 14869 */
14864 14870 if (IS_IPMP(ill))
14865 14871 ipmp_illgrp_refresh_arpent(ill->ill_grp);
14866 14872 }
14867 14873
14868 14874 /* Join the allhosts multicast address */
14869 14875 ipif_multicast_up(ipif);
14870 14876
14871 14877 if (!loopback && !update_src_selection &&
14872 14878 !(ipif->ipif_flags & (IPIF_NOLOCAL|IPIF_ANYCAST|IPIF_DEPRECATED)))
14873 14879 ip_update_source_selection(ill->ill_ipst);
14874 14880
14875 14881 if (!loopback && ipif->ipif_addr_ready) {
14876 14882 /* Broadcast an address mask reply. */
14877 14883 ipif_mask_reply(ipif);
14878 14884 }
14879 14885 /* Perhaps ilgs should use this ill */
14880 14886 update_conn_ill(NULL, ill->ill_ipst);
14881 14887
14882 14888 /*
14883 14889 * This had to be deferred until we had bound. Tell routing sockets and
14884 14890 * others that this interface is up if it looks like the address has
14885 14891 * been validated. Otherwise, if it isn't ready yet, wait for
14886 14892 * duplicate address detection to do its thing.
14887 14893 */
14888 14894 if (ipif->ipif_addr_ready)
14889 14895 ipif_up_notify(ipif);
14890 14896 return (0);
14891 14897 }
14892 14898
14893 14899 /*
14894 14900 * Add the IREs associated with the ipif.
14895 14901 * Those MUST be explicitly removed in ipif_delete_ires_v4.
14896 14902 */
14897 14903 static int
14898 14904 ipif_add_ires_v4(ipif_t *ipif, boolean_t loopback)
14899 14905 {
14900 14906 ill_t *ill = ipif->ipif_ill;
14901 14907 ip_stack_t *ipst = ill->ill_ipst;
14902 14908 ire_t *ire_array[20];
14903 14909 ire_t **irep = ire_array;
14904 14910 ire_t **irep1;
14905 14911 ipaddr_t net_mask = 0;
14906 14912 ipaddr_t subnet_mask, route_mask;
14907 14913 int err;
14908 14914 ire_t *ire_local = NULL; /* LOCAL or LOOPBACK */
14909 14915 ire_t *ire_if = NULL;
14910 14916 uchar_t *gw;
14911 14917
14912 14918 if ((ipif->ipif_lcl_addr != INADDR_ANY) &&
14913 14919 !(ipif->ipif_flags & IPIF_NOLOCAL)) {
14914 14920 /*
14915 14921 * If we're on a labeled system then make sure that zone-
14916 14922 * private addresses have proper remote host database entries.
14917 14923 */
14918 14924 if (is_system_labeled() &&
14919 14925 ipif->ipif_ire_type != IRE_LOOPBACK &&
14920 14926 !tsol_check_interface_address(ipif))
14921 14927 return (EINVAL);
14922 14928
14923 14929 /* Register the source address for __sin6_src_id */
14924 14930 err = ip_srcid_insert(&ipif->ipif_v6lcl_addr,
14925 14931 ipif->ipif_zoneid, ipst);
14926 14932 if (err != 0) {
14927 14933 ip0dbg(("ipif_add_ires: srcid_insert %d\n", err));
14928 14934 return (err);
14929 14935 }
14930 14936
14931 14937 if (loopback)
14932 14938 gw = (uchar_t *)&ipif->ipif_lcl_addr;
14933 14939 else
14934 14940 gw = NULL;
14935 14941
14936 14942 /* If the interface address is set, create the local IRE. */
14937 14943 ire_local = ire_create(
14938 14944 (uchar_t *)&ipif->ipif_lcl_addr, /* dest address */
14939 14945 (uchar_t *)&ip_g_all_ones, /* mask */
14940 14946 gw, /* gateway */
14941 14947 ipif->ipif_ire_type, /* LOCAL or LOOPBACK */
14942 14948 ipif->ipif_ill,
14943 14949 ipif->ipif_zoneid,
14944 14950 ((ipif->ipif_flags & IPIF_PRIVATE) ?
14945 14951 RTF_PRIVATE : 0) | RTF_KERNEL,
14946 14952 NULL,
14947 14953 ipst);
14948 14954 ip1dbg(("ipif_add_ires: 0x%p creating IRE %p type 0x%x"
14949 14955 " for 0x%x\n", (void *)ipif, (void *)ire_local,
14950 14956 ipif->ipif_ire_type,
14951 14957 ntohl(ipif->ipif_lcl_addr)));
14952 14958 if (ire_local == NULL) {
14953 14959 ip1dbg(("ipif_up_done: NULL ire_local\n"));
14954 14960 err = ENOMEM;
14955 14961 goto bad;
14956 14962 }
14957 14963 } else {
14958 14964 ip1dbg((
14959 14965 "ipif_add_ires: not creating IRE %d for 0x%x: flags 0x%x\n",
14960 14966 ipif->ipif_ire_type,
14961 14967 ntohl(ipif->ipif_lcl_addr),
14962 14968 (uint_t)ipif->ipif_flags));
14963 14969 }
14964 14970 if ((ipif->ipif_lcl_addr != INADDR_ANY) &&
14965 14971 !(ipif->ipif_flags & IPIF_NOLOCAL)) {
14966 14972 net_mask = ip_net_mask(ipif->ipif_lcl_addr);
14967 14973 } else {
14968 14974 net_mask = htonl(IN_CLASSA_NET); /* fallback */
14969 14975 }
14970 14976
14971 14977 subnet_mask = ipif->ipif_net_mask;
14972 14978
14973 14979 /*
14974 14980 * If mask was not specified, use natural netmask of
14975 14981 * interface address. Also, store this mask back into the
14976 14982 * ipif struct.
14977 14983 */
14978 14984 if (subnet_mask == 0) {
14979 14985 subnet_mask = net_mask;
14980 14986 V4MASK_TO_V6(subnet_mask, ipif->ipif_v6net_mask);
14981 14987 V6_MASK_COPY(ipif->ipif_v6lcl_addr, ipif->ipif_v6net_mask,
14982 14988 ipif->ipif_v6subnet);
14983 14989 }
14984 14990
14985 14991 /* Set up the IRE_IF_RESOLVER or IRE_IF_NORESOLVER, as appropriate. */
14986 14992 if (!loopback && !(ipif->ipif_flags & IPIF_NOXMIT) &&
14987 14993 ipif->ipif_subnet != INADDR_ANY) {
14988 14994 /* ipif_subnet is ipif_pp_dst_addr for pt-pt */
14989 14995
14990 14996 if (ipif->ipif_flags & IPIF_POINTOPOINT) {
14991 14997 route_mask = IP_HOST_MASK;
14992 14998 } else {
14993 14999 route_mask = subnet_mask;
14994 15000 }
14995 15001
14996 15002 ip1dbg(("ipif_add_ires: ipif 0x%p ill 0x%p "
14997 15003 "creating if IRE ill_net_type 0x%x for 0x%x\n",
14998 15004 (void *)ipif, (void *)ill, ill->ill_net_type,
14999 15005 ntohl(ipif->ipif_subnet)));
15000 15006 ire_if = ire_create(
15001 15007 (uchar_t *)&ipif->ipif_subnet,
15002 15008 (uchar_t *)&route_mask,
15003 15009 (uchar_t *)&ipif->ipif_lcl_addr,
15004 15010 ill->ill_net_type,
15005 15011 ill,
15006 15012 ipif->ipif_zoneid,
15007 15013 ((ipif->ipif_flags & IPIF_PRIVATE) ?
15008 15014 RTF_PRIVATE: 0) | RTF_KERNEL,
15009 15015 NULL,
15010 15016 ipst);
15011 15017 if (ire_if == NULL) {
15012 15018 ip1dbg(("ipif_up_done: NULL ire_if\n"));
15013 15019 err = ENOMEM;
15014 15020 goto bad;
15015 15021 }
15016 15022 }
15017 15023
15018 15024 /*
15019 15025 * Create any necessary broadcast IREs.
15020 15026 */
15021 15027 if ((ipif->ipif_flags & IPIF_BROADCAST) &&
15022 15028 !(ipif->ipif_flags & IPIF_NOXMIT))
15023 15029 irep = ipif_create_bcast_ires(ipif, irep);
15024 15030
15025 15031 /* If an earlier ire_create failed, get out now */
15026 15032 for (irep1 = irep; irep1 > ire_array; ) {
15027 15033 irep1--;
15028 15034 if (*irep1 == NULL) {
15029 15035 ip1dbg(("ipif_up_done: NULL ire found in ire_array\n"));
15030 15036 err = ENOMEM;
15031 15037 goto bad;
15032 15038 }
15033 15039 }
15034 15040
15035 15041 /*
15036 15042 * Need to atomically check for IP address availability under
15037 15043 * ip_addr_avail_lock. ill_g_lock is held as reader to ensure no new
15038 15044 * ills or new ipifs can be added while we are checking availability.
15039 15045 */
15040 15046 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
15041 15047 mutex_enter(&ipst->ips_ip_addr_avail_lock);
15042 15048 /* Mark it up, and increment counters. */
15043 15049 ipif->ipif_flags |= IPIF_UP;
15044 15050 ill->ill_ipif_up_count++;
15045 15051 err = ip_addr_availability_check(ipif);
15046 15052 mutex_exit(&ipst->ips_ip_addr_avail_lock);
15047 15053 rw_exit(&ipst->ips_ill_g_lock);
15048 15054
15049 15055 if (err != 0) {
15050 15056 /*
15051 15057 * Our address may already be up on the same ill. In this case,
15052 15058 * the ARP entry for our ipif replaced the one for the other
15053 15059 * ipif. So we don't want to delete it (otherwise the other ipif
15054 15060 * would be unable to send packets).
15055 15061 * ip_addr_availability_check() identifies this case for us and
15056 15062 * returns EADDRINUSE; Caller should turn it into EADDRNOTAVAIL
15057 15063 * which is the expected error code.
15058 15064 */
15059 15065 ill->ill_ipif_up_count--;
15060 15066 ipif->ipif_flags &= ~IPIF_UP;
15061 15067 goto bad;
15062 15068 }
15063 15069
15064 15070 /*
15065 15071 * Add in all newly created IREs. ire_create_bcast() has
15066 15072 * already checked for duplicates of the IRE_BROADCAST type.
15067 15073 * We add the IRE_INTERFACE before the IRE_LOCAL to ensure
15068 15074 * that lookups find the IRE_LOCAL even if the IRE_INTERFACE is
15069 15075 * a /32 route.
15070 15076 */
15071 15077 if (ire_if != NULL) {
15072 15078 ire_if = ire_add(ire_if);
15073 15079 if (ire_if == NULL) {
15074 15080 err = ENOMEM;
15075 15081 goto bad2;
15076 15082 }
15077 15083 #ifdef DEBUG
15078 15084 ire_refhold_notr(ire_if);
15079 15085 ire_refrele(ire_if);
15080 15086 #endif
15081 15087 }
15082 15088 if (ire_local != NULL) {
15083 15089 ire_local = ire_add(ire_local);
15084 15090 if (ire_local == NULL) {
15085 15091 err = ENOMEM;
15086 15092 goto bad2;
15087 15093 }
15088 15094 #ifdef DEBUG
15089 15095 ire_refhold_notr(ire_local);
15090 15096 ire_refrele(ire_local);
15091 15097 #endif
15092 15098 }
15093 15099 rw_enter(&ipst->ips_ill_g_lock, RW_WRITER);
15094 15100 if (ire_local != NULL)
15095 15101 ipif->ipif_ire_local = ire_local;
15096 15102 if (ire_if != NULL)
15097 15103 ipif->ipif_ire_if = ire_if;
15098 15104 rw_exit(&ipst->ips_ill_g_lock);
15099 15105 ire_local = NULL;
15100 15106 ire_if = NULL;
15101 15107
15102 15108 /*
15103 15109 * We first add all of them, and if that succeeds we refrele the
15104 15110 * bunch. That enables us to delete all of them should any of the
15105 15111 * ire_adds fail.
15106 15112 */
15107 15113 for (irep1 = irep; irep1 > ire_array; ) {
15108 15114 irep1--;
15109 15115 ASSERT(!MUTEX_HELD(&((*irep1)->ire_ill->ill_lock)));
15110 15116 *irep1 = ire_add(*irep1);
15111 15117 if (*irep1 == NULL) {
15112 15118 err = ENOMEM;
15113 15119 goto bad2;
15114 15120 }
15115 15121 }
15116 15122
15117 15123 for (irep1 = irep; irep1 > ire_array; ) {
15118 15124 irep1--;
15119 15125 /* refheld by ire_add. */
15120 15126 if (*irep1 != NULL) {
15121 15127 ire_refrele(*irep1);
15122 15128 *irep1 = NULL;
15123 15129 }
15124 15130 }
15125 15131
15126 15132 if (!loopback) {
15127 15133 /*
15128 15134 * If the broadcast address has been set, make sure it makes
15129 15135 * sense based on the interface address.
15130 15136 * Only match on ill since we are sharing broadcast addresses.
15131 15137 */
15132 15138 if ((ipif->ipif_brd_addr != INADDR_ANY) &&
15133 15139 (ipif->ipif_flags & IPIF_BROADCAST)) {
15134 15140 ire_t *ire;
15135 15141
15136 15142 ire = ire_ftable_lookup_v4(ipif->ipif_brd_addr, 0, 0,
15137 15143 IRE_BROADCAST, ipif->ipif_ill, ALL_ZONES, NULL,
15138 15144 (MATCH_IRE_TYPE | MATCH_IRE_ILL), 0, ipst, NULL);
15139 15145
15140 15146 if (ire == NULL) {
15141 15147 /*
15142 15148 * If there isn't a matching broadcast IRE,
15143 15149 * revert to the default for this netmask.
15144 15150 */
15145 15151 ipif->ipif_v6brd_addr = ipv6_all_zeros;
15146 15152 mutex_enter(&ipif->ipif_ill->ill_lock);
15147 15153 ipif_set_default(ipif);
15148 15154 mutex_exit(&ipif->ipif_ill->ill_lock);
15149 15155 } else {
15150 15156 ire_refrele(ire);
15151 15157 }
15152 15158 }
15153 15159
15154 15160 }
15155 15161 return (0);
15156 15162
15157 15163 bad2:
15158 15164 ill->ill_ipif_up_count--;
15159 15165 ipif->ipif_flags &= ~IPIF_UP;
15160 15166
15161 15167 bad:
15162 15168 ip1dbg(("ipif_add_ires: FAILED \n"));
15163 15169 if (ire_local != NULL)
15164 15170 ire_delete(ire_local);
15165 15171 if (ire_if != NULL)
15166 15172 ire_delete(ire_if);
15167 15173
15168 15174 rw_enter(&ipst->ips_ill_g_lock, RW_WRITER);
15169 15175 ire_local = ipif->ipif_ire_local;
15170 15176 ipif->ipif_ire_local = NULL;
15171 15177 ire_if = ipif->ipif_ire_if;
15172 15178 ipif->ipif_ire_if = NULL;
15173 15179 rw_exit(&ipst->ips_ill_g_lock);
15174 15180 if (ire_local != NULL) {
15175 15181 ire_delete(ire_local);
15176 15182 ire_refrele_notr(ire_local);
15177 15183 }
15178 15184 if (ire_if != NULL) {
15179 15185 ire_delete(ire_if);
15180 15186 ire_refrele_notr(ire_if);
15181 15187 }
15182 15188
15183 15189 while (irep > ire_array) {
15184 15190 irep--;
15185 15191 if (*irep != NULL) {
15186 15192 ire_delete(*irep);
15187 15193 }
15188 15194 }
15189 15195 (void) ip_srcid_remove(&ipif->ipif_v6lcl_addr, ipif->ipif_zoneid, ipst);
15190 15196
15191 15197 return (err);
15192 15198 }
15193 15199
15194 15200 /* Remove all the IREs created by ipif_add_ires_v4 */
15195 15201 void
15196 15202 ipif_delete_ires_v4(ipif_t *ipif)
15197 15203 {
15198 15204 ill_t *ill = ipif->ipif_ill;
15199 15205 ip_stack_t *ipst = ill->ill_ipst;
15200 15206 ire_t *ire;
15201 15207
15202 15208 rw_enter(&ipst->ips_ill_g_lock, RW_WRITER);
15203 15209 ire = ipif->ipif_ire_local;
15204 15210 ipif->ipif_ire_local = NULL;
15205 15211 rw_exit(&ipst->ips_ill_g_lock);
15206 15212 if (ire != NULL) {
15207 15213 /*
15208 15214 * Move count to ipif so we don't loose the count due to
15209 15215 * a down/up dance.
15210 15216 */
15211 15217 atomic_add_32(&ipif->ipif_ib_pkt_count, ire->ire_ib_pkt_count);
15212 15218
15213 15219 ire_delete(ire);
15214 15220 ire_refrele_notr(ire);
15215 15221 }
15216 15222 rw_enter(&ipst->ips_ill_g_lock, RW_WRITER);
15217 15223 ire = ipif->ipif_ire_if;
15218 15224 ipif->ipif_ire_if = NULL;
15219 15225 rw_exit(&ipst->ips_ill_g_lock);
15220 15226 if (ire != NULL) {
15221 15227 ire_delete(ire);
15222 15228 ire_refrele_notr(ire);
15223 15229 }
15224 15230
15225 15231 /*
15226 15232 * Delete the broadcast IREs.
15227 15233 */
15228 15234 if ((ipif->ipif_flags & IPIF_BROADCAST) &&
15229 15235 !(ipif->ipif_flags & IPIF_NOXMIT))
15230 15236 ipif_delete_bcast_ires(ipif);
15231 15237 }
15232 15238
15233 15239 /*
15234 15240 * Checks for availbility of a usable source address (if there is one) when the
15235 15241 * destination ILL has the ill_usesrc_ifindex pointing to another ILL. Note
15236 15242 * this selection is done regardless of the destination.
15237 15243 */
15238 15244 boolean_t
15239 15245 ipif_zone_avail(uint_t ifindex, boolean_t isv6, zoneid_t zoneid,
15240 15246 ip_stack_t *ipst)
15241 15247 {
15242 15248 ipif_t *ipif = NULL;
15243 15249 ill_t *uill;
15244 15250
15245 15251 ASSERT(ifindex != 0);
15246 15252
15247 15253 uill = ill_lookup_on_ifindex(ifindex, isv6, ipst);
15248 15254 if (uill == NULL)
15249 15255 return (B_FALSE);
15250 15256
15251 15257 mutex_enter(&uill->ill_lock);
15252 15258 for (ipif = uill->ill_ipif; ipif != NULL; ipif = ipif->ipif_next) {
15253 15259 if (IPIF_IS_CONDEMNED(ipif))
15254 15260 continue;
15255 15261 if (ipif->ipif_flags & (IPIF_NOLOCAL|IPIF_ANYCAST))
15256 15262 continue;
15257 15263 if (!(ipif->ipif_flags & IPIF_UP))
15258 15264 continue;
15259 15265 if (ipif->ipif_zoneid != zoneid)
15260 15266 continue;
15261 15267 if (isv6 ? IN6_IS_ADDR_UNSPECIFIED(&ipif->ipif_v6lcl_addr) :
15262 15268 ipif->ipif_lcl_addr == INADDR_ANY)
15263 15269 continue;
15264 15270 mutex_exit(&uill->ill_lock);
15265 15271 ill_refrele(uill);
15266 15272 return (B_TRUE);
15267 15273 }
15268 15274 mutex_exit(&uill->ill_lock);
15269 15275 ill_refrele(uill);
15270 15276 return (B_FALSE);
15271 15277 }
15272 15278
15273 15279 /*
15274 15280 * Find an ipif with a good local address on the ill+zoneid.
15275 15281 */
15276 15282 ipif_t *
15277 15283 ipif_good_addr(ill_t *ill, zoneid_t zoneid)
15278 15284 {
15279 15285 ipif_t *ipif;
15280 15286
15281 15287 mutex_enter(&ill->ill_lock);
15282 15288 for (ipif = ill->ill_ipif; ipif != NULL; ipif = ipif->ipif_next) {
15283 15289 if (IPIF_IS_CONDEMNED(ipif))
15284 15290 continue;
15285 15291 if (ipif->ipif_flags & (IPIF_NOLOCAL|IPIF_ANYCAST))
15286 15292 continue;
15287 15293 if (!(ipif->ipif_flags & IPIF_UP))
15288 15294 continue;
15289 15295 if (ipif->ipif_zoneid != zoneid &&
15290 15296 ipif->ipif_zoneid != ALL_ZONES && zoneid != ALL_ZONES)
15291 15297 continue;
15292 15298 if (ill->ill_isv6 ?
15293 15299 IN6_IS_ADDR_UNSPECIFIED(&ipif->ipif_v6lcl_addr) :
15294 15300 ipif->ipif_lcl_addr == INADDR_ANY)
15295 15301 continue;
15296 15302 ipif_refhold_locked(ipif);
15297 15303 mutex_exit(&ill->ill_lock);
15298 15304 return (ipif);
15299 15305 }
15300 15306 mutex_exit(&ill->ill_lock);
15301 15307 return (NULL);
15302 15308 }
15303 15309
15304 15310 /*
15305 15311 * IP source address type, sorted from worst to best. For a given type,
15306 15312 * always prefer IP addresses on the same subnet. All-zones addresses are
15307 15313 * suboptimal because they pose problems with unlabeled destinations.
15308 15314 */
15309 15315 typedef enum {
15310 15316 IPIF_NONE,
15311 15317 IPIF_DIFFNET_DEPRECATED, /* deprecated and different subnet */
15312 15318 IPIF_SAMENET_DEPRECATED, /* deprecated and same subnet */
15313 15319 IPIF_DIFFNET_ALLZONES, /* allzones and different subnet */
15314 15320 IPIF_SAMENET_ALLZONES, /* allzones and same subnet */
15315 15321 IPIF_DIFFNET, /* normal and different subnet */
15316 15322 IPIF_SAMENET, /* normal and same subnet */
15317 15323 IPIF_LOCALADDR /* local loopback */
15318 15324 } ipif_type_t;
15319 15325
15320 15326 /*
15321 15327 * Pick the optimal ipif on `ill' for sending to destination `dst' from zone
15322 15328 * `zoneid'. We rate usable ipifs from low -> high as per the ipif_type_t
15323 15329 * enumeration, and return the highest-rated ipif. If there's a tie, we pick
15324 15330 * the first one, unless IPMP is used in which case we round-robin among them;
15325 15331 * see below for more.
15326 15332 *
15327 15333 * Returns NULL if there is no suitable source address for the ill.
15328 15334 * This only occurs when there is no valid source address for the ill.
15329 15335 */
15330 15336 ipif_t *
15331 15337 ipif_select_source_v4(ill_t *ill, ipaddr_t dst, zoneid_t zoneid,
15332 15338 boolean_t allow_usesrc, boolean_t *notreadyp)
15333 15339 {
15334 15340 ill_t *usill = NULL;
15335 15341 ill_t *ipmp_ill = NULL;
15336 15342 ipif_t *start_ipif, *next_ipif, *ipif, *best_ipif;
15337 15343 ipif_type_t type, best_type;
15338 15344 tsol_tpc_t *src_rhtp, *dst_rhtp;
15339 15345 ip_stack_t *ipst = ill->ill_ipst;
15340 15346 boolean_t samenet;
15341 15347
15342 15348 if (ill->ill_usesrc_ifindex != 0 && allow_usesrc) {
15343 15349 usill = ill_lookup_on_ifindex(ill->ill_usesrc_ifindex,
15344 15350 B_FALSE, ipst);
15345 15351 if (usill != NULL)
15346 15352 ill = usill; /* Select source from usesrc ILL */
15347 15353 else
15348 15354 return (NULL);
15349 15355 }
15350 15356
15351 15357 /*
15352 15358 * Test addresses should never be used for source address selection,
15353 15359 * so if we were passed one, switch to the IPMP meta-interface.
15354 15360 */
15355 15361 if (IS_UNDER_IPMP(ill)) {
15356 15362 if ((ipmp_ill = ipmp_ill_hold_ipmp_ill(ill)) != NULL)
15357 15363 ill = ipmp_ill; /* Select source from IPMP ill */
15358 15364 else
15359 15365 return (NULL);
15360 15366 }
15361 15367
15362 15368 /*
15363 15369 * If we're dealing with an unlabeled destination on a labeled system,
15364 15370 * make sure that we ignore source addresses that are incompatible with
15365 15371 * the destination's default label. That destination's default label
15366 15372 * must dominate the minimum label on the source address.
15367 15373 */
15368 15374 dst_rhtp = NULL;
15369 15375 if (is_system_labeled()) {
15370 15376 dst_rhtp = find_tpc(&dst, IPV4_VERSION, B_FALSE);
15371 15377 if (dst_rhtp == NULL)
15372 15378 return (NULL);
15373 15379 if (dst_rhtp->tpc_tp.host_type != UNLABELED) {
15374 15380 TPC_RELE(dst_rhtp);
15375 15381 dst_rhtp = NULL;
15376 15382 }
15377 15383 }
15378 15384
15379 15385 /*
15380 15386 * Hold the ill_g_lock as reader. This makes sure that no ipif/ill
15381 15387 * can be deleted. But an ipif/ill can get CONDEMNED any time.
15382 15388 * After selecting the right ipif, under ill_lock make sure ipif is
15383 15389 * not condemned, and increment refcnt. If ipif is CONDEMNED,
15384 15390 * we retry. Inside the loop we still need to check for CONDEMNED,
15385 15391 * but not under a lock.
15386 15392 */
15387 15393 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
15388 15394 retry:
15389 15395 /*
15390 15396 * For source address selection, we treat the ipif list as circular
15391 15397 * and continue until we get back to where we started. This allows
15392 15398 * IPMP to vary source address selection (which improves inbound load
15393 15399 * spreading) by caching its last ending point and starting from
15394 15400 * there. NOTE: we don't have to worry about ill_src_ipif changing
15395 15401 * ills since that can't happen on the IPMP ill.
15396 15402 */
15397 15403 start_ipif = ill->ill_ipif;
15398 15404 if (IS_IPMP(ill) && ill->ill_src_ipif != NULL)
15399 15405 start_ipif = ill->ill_src_ipif;
15400 15406
15401 15407 ipif = start_ipif;
15402 15408 best_ipif = NULL;
15403 15409 best_type = IPIF_NONE;
15404 15410 do {
15405 15411 if ((next_ipif = ipif->ipif_next) == NULL)
15406 15412 next_ipif = ill->ill_ipif;
15407 15413
15408 15414 if (IPIF_IS_CONDEMNED(ipif))
15409 15415 continue;
15410 15416 /* Always skip NOLOCAL and ANYCAST interfaces */
15411 15417 if (ipif->ipif_flags & (IPIF_NOLOCAL|IPIF_ANYCAST))
15412 15418 continue;
15413 15419 /* Always skip NOACCEPT interfaces */
15414 15420 if (ipif->ipif_ill->ill_flags & ILLF_NOACCEPT)
15415 15421 continue;
15416 15422 if (!(ipif->ipif_flags & IPIF_UP))
15417 15423 continue;
15418 15424
15419 15425 if (!ipif->ipif_addr_ready) {
15420 15426 if (notreadyp != NULL)
15421 15427 *notreadyp = B_TRUE;
15422 15428 continue;
15423 15429 }
15424 15430
15425 15431 if (zoneid != ALL_ZONES &&
15426 15432 ipif->ipif_zoneid != zoneid &&
15427 15433 ipif->ipif_zoneid != ALL_ZONES)
15428 15434 continue;
15429 15435
15430 15436 /*
15431 15437 * Interfaces with 0.0.0.0 address are allowed to be UP, but
15432 15438 * are not valid as source addresses.
15433 15439 */
15434 15440 if (ipif->ipif_lcl_addr == INADDR_ANY)
15435 15441 continue;
15436 15442
15437 15443 /*
15438 15444 * Check compatibility of local address for destination's
15439 15445 * default label if we're on a labeled system. Incompatible
15440 15446 * addresses can't be used at all.
15441 15447 */
15442 15448 if (dst_rhtp != NULL) {
15443 15449 boolean_t incompat;
15444 15450
15445 15451 src_rhtp = find_tpc(&ipif->ipif_lcl_addr,
15446 15452 IPV4_VERSION, B_FALSE);
15447 15453 if (src_rhtp == NULL)
15448 15454 continue;
15449 15455 incompat = src_rhtp->tpc_tp.host_type != SUN_CIPSO ||
15450 15456 src_rhtp->tpc_tp.tp_doi !=
15451 15457 dst_rhtp->tpc_tp.tp_doi ||
15452 15458 (!_blinrange(&dst_rhtp->tpc_tp.tp_def_label,
15453 15459 &src_rhtp->tpc_tp.tp_sl_range_cipso) &&
15454 15460 !blinlset(&dst_rhtp->tpc_tp.tp_def_label,
15455 15461 src_rhtp->tpc_tp.tp_sl_set_cipso));
15456 15462 TPC_RELE(src_rhtp);
15457 15463 if (incompat)
15458 15464 continue;
15459 15465 }
15460 15466
15461 15467 samenet = ((ipif->ipif_net_mask & dst) == ipif->ipif_subnet);
15462 15468
15463 15469 if (ipif->ipif_lcl_addr == dst) {
15464 15470 type = IPIF_LOCALADDR;
15465 15471 } else if (ipif->ipif_flags & IPIF_DEPRECATED) {
15466 15472 type = samenet ? IPIF_SAMENET_DEPRECATED :
15467 15473 IPIF_DIFFNET_DEPRECATED;
15468 15474 } else if (ipif->ipif_zoneid == ALL_ZONES) {
15469 15475 type = samenet ? IPIF_SAMENET_ALLZONES :
15470 15476 IPIF_DIFFNET_ALLZONES;
15471 15477 } else {
15472 15478 type = samenet ? IPIF_SAMENET : IPIF_DIFFNET;
15473 15479 }
15474 15480
15475 15481 if (type > best_type) {
15476 15482 best_type = type;
15477 15483 best_ipif = ipif;
15478 15484 if (best_type == IPIF_LOCALADDR)
15479 15485 break; /* can't get better */
15480 15486 }
15481 15487 } while ((ipif = next_ipif) != start_ipif);
15482 15488
15483 15489 if ((ipif = best_ipif) != NULL) {
15484 15490 mutex_enter(&ipif->ipif_ill->ill_lock);
15485 15491 if (IPIF_IS_CONDEMNED(ipif)) {
15486 15492 mutex_exit(&ipif->ipif_ill->ill_lock);
15487 15493 goto retry;
15488 15494 }
15489 15495 ipif_refhold_locked(ipif);
15490 15496
15491 15497 /*
15492 15498 * For IPMP, update the source ipif rotor to the next ipif,
15493 15499 * provided we can look it up. (We must not use it if it's
15494 15500 * IPIF_CONDEMNED since we may have grabbed ill_g_lock after
15495 15501 * ipif_free() checked ill_src_ipif.)
15496 15502 */
15497 15503 if (IS_IPMP(ill) && ipif != NULL) {
15498 15504 next_ipif = ipif->ipif_next;
15499 15505 if (next_ipif != NULL && !IPIF_IS_CONDEMNED(next_ipif))
15500 15506 ill->ill_src_ipif = next_ipif;
15501 15507 else
15502 15508 ill->ill_src_ipif = NULL;
15503 15509 }
15504 15510 mutex_exit(&ipif->ipif_ill->ill_lock);
15505 15511 }
15506 15512
15507 15513 rw_exit(&ipst->ips_ill_g_lock);
15508 15514 if (usill != NULL)
15509 15515 ill_refrele(usill);
15510 15516 if (ipmp_ill != NULL)
15511 15517 ill_refrele(ipmp_ill);
15512 15518 if (dst_rhtp != NULL)
15513 15519 TPC_RELE(dst_rhtp);
15514 15520
15515 15521 #ifdef DEBUG
15516 15522 if (ipif == NULL) {
15517 15523 char buf1[INET6_ADDRSTRLEN];
15518 15524
15519 15525 ip1dbg(("ipif_select_source_v4(%s, %s) -> NULL\n",
15520 15526 ill->ill_name,
15521 15527 inet_ntop(AF_INET, &dst, buf1, sizeof (buf1))));
15522 15528 } else {
15523 15529 char buf1[INET6_ADDRSTRLEN];
15524 15530 char buf2[INET6_ADDRSTRLEN];
15525 15531
15526 15532 ip1dbg(("ipif_select_source_v4(%s, %s) -> %s\n",
15527 15533 ipif->ipif_ill->ill_name,
15528 15534 inet_ntop(AF_INET, &dst, buf1, sizeof (buf1)),
15529 15535 inet_ntop(AF_INET, &ipif->ipif_lcl_addr,
15530 15536 buf2, sizeof (buf2))));
15531 15537 }
15532 15538 #endif /* DEBUG */
15533 15539 return (ipif);
15534 15540 }
15535 15541
15536 15542 /*
15537 15543 * Pick a source address based on the destination ill and an optional setsrc
15538 15544 * address.
15539 15545 * The result is stored in srcp. If generation is set, then put the source
15540 15546 * generation number there before we look for the source address (to avoid
15541 15547 * missing changes in the set of source addresses.
15542 15548 * If flagsp is set, then us it to pass back ipif_flags.
15543 15549 *
15544 15550 * If the caller wants to cache the returned source address and detect when
15545 15551 * that might be stale, the caller should pass in a generation argument,
15546 15552 * which the caller can later compare against ips_src_generation
15547 15553 *
15548 15554 * The precedence order for selecting an IPv4 source address is:
15549 15555 * - RTF_SETSRC on the offlink ire always wins.
15550 15556 * - If usrsrc is set, swap the ill to be the usesrc one.
15551 15557 * - If IPMP is used on the ill, select a random address from the most
15552 15558 * preferred ones below:
15553 15559 * 1. If onlink destination, same subnet and not deprecated, not ALL_ZONES
15554 15560 * 2. Not deprecated, not ALL_ZONES
15555 15561 * 3. If onlink destination, same subnet and not deprecated, ALL_ZONES
15556 15562 * 4. Not deprecated, ALL_ZONES
15557 15563 * 5. If onlink destination, same subnet and deprecated
15558 15564 * 6. Deprecated.
15559 15565 *
15560 15566 * We have lower preference for ALL_ZONES IP addresses,
15561 15567 * as they pose problems with unlabeled destinations.
15562 15568 *
15563 15569 * Note that when multiple IP addresses match e.g., #1 we pick
15564 15570 * the first one if IPMP is not in use. With IPMP we randomize.
15565 15571 */
15566 15572 int
15567 15573 ip_select_source_v4(ill_t *ill, ipaddr_t setsrc, ipaddr_t dst,
15568 15574 ipaddr_t multicast_ifaddr,
15569 15575 zoneid_t zoneid, ip_stack_t *ipst, ipaddr_t *srcp,
15570 15576 uint32_t *generation, uint64_t *flagsp)
15571 15577 {
15572 15578 ipif_t *ipif;
15573 15579 boolean_t notready = B_FALSE; /* Set if !ipif_addr_ready found */
15574 15580
15575 15581 if (flagsp != NULL)
15576 15582 *flagsp = 0;
15577 15583
15578 15584 /*
15579 15585 * Need to grab the generation number before we check to
15580 15586 * avoid a race with a change to the set of local addresses.
15581 15587 * No lock needed since the thread which updates the set of local
15582 15588 * addresses use ipif/ill locks and exit those (hence a store memory
15583 15589 * barrier) before doing the atomic increase of ips_src_generation.
15584 15590 */
15585 15591 if (generation != NULL) {
15586 15592 *generation = ipst->ips_src_generation;
15587 15593 }
15588 15594
15589 15595 if (CLASSD(dst) && multicast_ifaddr != INADDR_ANY) {
15590 15596 *srcp = multicast_ifaddr;
15591 15597 return (0);
15592 15598 }
15593 15599
15594 15600 /* Was RTF_SETSRC set on the first IRE in the recursive lookup? */
15595 15601 if (setsrc != INADDR_ANY) {
15596 15602 *srcp = setsrc;
15597 15603 return (0);
15598 15604 }
15599 15605 ipif = ipif_select_source_v4(ill, dst, zoneid, B_TRUE, ¬ready);
15600 15606 if (ipif == NULL) {
15601 15607 if (notready)
15602 15608 return (ENETDOWN);
15603 15609 else
15604 15610 return (EADDRNOTAVAIL);
15605 15611 }
15606 15612 *srcp = ipif->ipif_lcl_addr;
15607 15613 if (flagsp != NULL)
15608 15614 *flagsp = ipif->ipif_flags;
15609 15615 ipif_refrele(ipif);
15610 15616 return (0);
15611 15617 }
15612 15618
15613 15619 /* ARGSUSED */
15614 15620 int
15615 15621 if_unitsel_restart(ipif_t *ipif, sin_t *dummy_sin, queue_t *q, mblk_t *mp,
15616 15622 ip_ioctl_cmd_t *ipip, void *dummy_ifreq)
15617 15623 {
15618 15624 /*
15619 15625 * ill_phyint_reinit merged the v4 and v6 into a single
15620 15626 * ipsq. We might not have been able to complete the
15621 15627 * operation in ipif_set_values, if we could not become
15622 15628 * exclusive. If so restart it here.
15623 15629 */
15624 15630 return (ipif_set_values_tail(ipif->ipif_ill, ipif, mp, q));
15625 15631 }
15626 15632
15627 15633 /*
15628 15634 * Can operate on either a module or a driver queue.
15629 15635 * Returns an error if not a module queue.
15630 15636 */
15631 15637 /* ARGSUSED */
15632 15638 int
15633 15639 if_unitsel(ipif_t *dummy_ipif, sin_t *dummy_sin, queue_t *q, mblk_t *mp,
15634 15640 ip_ioctl_cmd_t *ipip, void *dummy_ifreq)
15635 15641 {
15636 15642 queue_t *q1 = q;
15637 15643 char *cp;
15638 15644 char interf_name[LIFNAMSIZ];
15639 15645 uint_t ppa = *(uint_t *)mp->b_cont->b_cont->b_rptr;
15640 15646
15641 15647 if (q->q_next == NULL) {
15642 15648 ip1dbg((
15643 15649 "if_unitsel: IF_UNITSEL: no q_next\n"));
15644 15650 return (EINVAL);
15645 15651 }
15646 15652
15647 15653 if (((ill_t *)(q->q_ptr))->ill_name[0] != '\0')
15648 15654 return (EALREADY);
15649 15655
15650 15656 do {
15651 15657 q1 = q1->q_next;
15652 15658 } while (q1->q_next);
15653 15659 cp = q1->q_qinfo->qi_minfo->mi_idname;
15654 15660 (void) sprintf(interf_name, "%s%d", cp, ppa);
15655 15661
15656 15662 /*
15657 15663 * Here we are not going to delay the ioack until after
15658 15664 * ACKs from DL_ATTACH_REQ/DL_BIND_REQ. So no need to save the
15659 15665 * original ioctl message before sending the requests.
15660 15666 */
15661 15667 return (ipif_set_values(q, mp, interf_name, &ppa));
15662 15668 }
15663 15669
15664 15670 /* ARGSUSED */
15665 15671 int
15666 15672 ip_sioctl_sifname(ipif_t *dummy_ipif, sin_t *dummy_sin, queue_t *q, mblk_t *mp,
15667 15673 ip_ioctl_cmd_t *ipip, void *dummy_ifreq)
15668 15674 {
15669 15675 return (ENXIO);
15670 15676 }
15671 15677
15672 15678 /*
15673 15679 * Create any IRE_BROADCAST entries for `ipif', and store those entries in
15674 15680 * `irep'. Returns a pointer to the next free `irep' entry
15675 15681 * A mirror exists in ipif_delete_bcast_ires().
15676 15682 *
15677 15683 * The management of any "extra" or seemingly duplicate IRE_BROADCASTs is
15678 15684 * done in ire_add.
15679 15685 */
15680 15686 static ire_t **
15681 15687 ipif_create_bcast_ires(ipif_t *ipif, ire_t **irep)
15682 15688 {
15683 15689 ipaddr_t addr;
15684 15690 ipaddr_t netmask = ip_net_mask(ipif->ipif_lcl_addr);
15685 15691 ipaddr_t subnetmask = ipif->ipif_net_mask;
15686 15692 ill_t *ill = ipif->ipif_ill;
15687 15693 zoneid_t zoneid = ipif->ipif_zoneid;
15688 15694
15689 15695 ip1dbg(("ipif_create_bcast_ires: creating broadcast IREs\n"));
15690 15696
15691 15697 ASSERT(ipif->ipif_flags & IPIF_BROADCAST);
15692 15698 ASSERT(!(ipif->ipif_flags & IPIF_NOXMIT));
15693 15699
15694 15700 if (ipif->ipif_lcl_addr == INADDR_ANY ||
15695 15701 (ipif->ipif_flags & IPIF_NOLOCAL))
15696 15702 netmask = htonl(IN_CLASSA_NET); /* fallback */
15697 15703
15698 15704 irep = ire_create_bcast(ill, 0, zoneid, irep);
15699 15705 irep = ire_create_bcast(ill, INADDR_BROADCAST, zoneid, irep);
15700 15706
15701 15707 /*
15702 15708 * For backward compatibility, we create net broadcast IREs based on
15703 15709 * the old "IP address class system", since some old machines only
15704 15710 * respond to these class derived net broadcast. However, we must not
15705 15711 * create these net broadcast IREs if the subnetmask is shorter than
15706 15712 * the IP address class based derived netmask. Otherwise, we may
15707 15713 * create a net broadcast address which is the same as an IP address
15708 15714 * on the subnet -- and then TCP will refuse to talk to that address.
15709 15715 */
15710 15716 if (netmask < subnetmask) {
15711 15717 addr = netmask & ipif->ipif_subnet;
15712 15718 irep = ire_create_bcast(ill, addr, zoneid, irep);
15713 15719 irep = ire_create_bcast(ill, ~netmask | addr, zoneid, irep);
15714 15720 }
15715 15721
15716 15722 /*
15717 15723 * Don't create IRE_BROADCAST IREs for the interface if the subnetmask
15718 15724 * is 0xFFFFFFFF, as an IRE_LOCAL for that interface is already
15719 15725 * created. Creating these broadcast IREs will only create confusion
15720 15726 * as `addr' will be the same as the IP address.
15721 15727 */
15722 15728 if (subnetmask != 0xFFFFFFFF) {
15723 15729 addr = ipif->ipif_subnet;
15724 15730 irep = ire_create_bcast(ill, addr, zoneid, irep);
15725 15731 irep = ire_create_bcast(ill, ~subnetmask | addr, zoneid, irep);
15726 15732 }
15727 15733
15728 15734 return (irep);
15729 15735 }
15730 15736
15731 15737 /*
15732 15738 * Mirror of ipif_create_bcast_ires()
15733 15739 */
15734 15740 static void
15735 15741 ipif_delete_bcast_ires(ipif_t *ipif)
15736 15742 {
15737 15743 ipaddr_t addr;
15738 15744 ipaddr_t netmask = ip_net_mask(ipif->ipif_lcl_addr);
15739 15745 ipaddr_t subnetmask = ipif->ipif_net_mask;
15740 15746 ill_t *ill = ipif->ipif_ill;
15741 15747 zoneid_t zoneid = ipif->ipif_zoneid;
15742 15748 ire_t *ire;
15743 15749
15744 15750 ASSERT(ipif->ipif_flags & IPIF_BROADCAST);
15745 15751 ASSERT(!(ipif->ipif_flags & IPIF_NOXMIT));
15746 15752
15747 15753 if (ipif->ipif_lcl_addr == INADDR_ANY ||
15748 15754 (ipif->ipif_flags & IPIF_NOLOCAL))
15749 15755 netmask = htonl(IN_CLASSA_NET); /* fallback */
15750 15756
15751 15757 ire = ire_lookup_bcast(ill, 0, zoneid);
15752 15758 ASSERT(ire != NULL);
15753 15759 ire_delete(ire); ire_refrele(ire);
15754 15760 ire = ire_lookup_bcast(ill, INADDR_BROADCAST, zoneid);
15755 15761 ASSERT(ire != NULL);
15756 15762 ire_delete(ire); ire_refrele(ire);
15757 15763
15758 15764 /*
15759 15765 * For backward compatibility, we create net broadcast IREs based on
15760 15766 * the old "IP address class system", since some old machines only
15761 15767 * respond to these class derived net broadcast. However, we must not
15762 15768 * create these net broadcast IREs if the subnetmask is shorter than
15763 15769 * the IP address class based derived netmask. Otherwise, we may
15764 15770 * create a net broadcast address which is the same as an IP address
15765 15771 * on the subnet -- and then TCP will refuse to talk to that address.
15766 15772 */
15767 15773 if (netmask < subnetmask) {
15768 15774 addr = netmask & ipif->ipif_subnet;
15769 15775 ire = ire_lookup_bcast(ill, addr, zoneid);
15770 15776 ASSERT(ire != NULL);
15771 15777 ire_delete(ire); ire_refrele(ire);
15772 15778 ire = ire_lookup_bcast(ill, ~netmask | addr, zoneid);
15773 15779 ASSERT(ire != NULL);
15774 15780 ire_delete(ire); ire_refrele(ire);
15775 15781 }
15776 15782
15777 15783 /*
15778 15784 * Don't create IRE_BROADCAST IREs for the interface if the subnetmask
15779 15785 * is 0xFFFFFFFF, as an IRE_LOCAL for that interface is already
15780 15786 * created. Creating these broadcast IREs will only create confusion
15781 15787 * as `addr' will be the same as the IP address.
15782 15788 */
15783 15789 if (subnetmask != 0xFFFFFFFF) {
15784 15790 addr = ipif->ipif_subnet;
15785 15791 ire = ire_lookup_bcast(ill, addr, zoneid);
15786 15792 ASSERT(ire != NULL);
15787 15793 ire_delete(ire); ire_refrele(ire);
15788 15794 ire = ire_lookup_bcast(ill, ~subnetmask | addr, zoneid);
15789 15795 ASSERT(ire != NULL);
15790 15796 ire_delete(ire); ire_refrele(ire);
15791 15797 }
15792 15798 }
15793 15799
15794 15800 /*
15795 15801 * Extract both the flags (including IFF_CANTCHANGE) such as IFF_IPV*
15796 15802 * from lifr_flags and the name from lifr_name.
15797 15803 * Set IFF_IPV* and ill_isv6 prior to doing the lookup
15798 15804 * since ipif_lookup_on_name uses the _isv6 flags when matching.
15799 15805 * Returns EINPROGRESS when mp has been consumed by queueing it on
15800 15806 * ipx_pending_mp and the ioctl will complete in ip_rput.
15801 15807 *
15802 15808 * Can operate on either a module or a driver queue.
15803 15809 * Returns an error if not a module queue.
15804 15810 */
15805 15811 /* ARGSUSED */
15806 15812 int
15807 15813 ip_sioctl_slifname(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
15808 15814 ip_ioctl_cmd_t *ipip, void *if_req)
15809 15815 {
15810 15816 ill_t *ill = q->q_ptr;
15811 15817 phyint_t *phyi;
15812 15818 ip_stack_t *ipst;
15813 15819 struct lifreq *lifr = if_req;
15814 15820 uint64_t new_flags;
15815 15821
15816 15822 ASSERT(ipif != NULL);
15817 15823 ip1dbg(("ip_sioctl_slifname %s\n", lifr->lifr_name));
15818 15824
15819 15825 if (q->q_next == NULL) {
15820 15826 ip1dbg(("if_sioctl_slifname: SIOCSLIFNAME: no q_next\n"));
15821 15827 return (EINVAL);
15822 15828 }
15823 15829
15824 15830 /*
15825 15831 * If we are not writer on 'q' then this interface exists already
15826 15832 * and previous lookups (ip_extract_lifreq()) found this ipif --
15827 15833 * so return EALREADY.
15828 15834 */
15829 15835 if (ill != ipif->ipif_ill)
15830 15836 return (EALREADY);
15831 15837
15832 15838 if (ill->ill_name[0] != '\0')
15833 15839 return (EALREADY);
15834 15840
15835 15841 /*
15836 15842 * If there's another ill already with the requested name, ensure
15837 15843 * that it's of the same type. Otherwise, ill_phyint_reinit() will
15838 15844 * fuse together two unrelated ills, which will cause chaos.
15839 15845 */
15840 15846 ipst = ill->ill_ipst;
15841 15847 phyi = avl_find(&ipst->ips_phyint_g_list->phyint_list_avl_by_name,
15842 15848 lifr->lifr_name, NULL);
15843 15849 if (phyi != NULL) {
15844 15850 ill_t *ill_mate = phyi->phyint_illv4;
15845 15851
15846 15852 if (ill_mate == NULL)
15847 15853 ill_mate = phyi->phyint_illv6;
15848 15854 ASSERT(ill_mate != NULL);
15849 15855
15850 15856 if (ill_mate->ill_media->ip_m_mac_type !=
15851 15857 ill->ill_media->ip_m_mac_type) {
15852 15858 ip1dbg(("if_sioctl_slifname: SIOCSLIFNAME: attempt to "
15853 15859 "use the same ill name on differing media\n"));
15854 15860 return (EINVAL);
15855 15861 }
15856 15862 }
15857 15863
15858 15864 /*
15859 15865 * We start off as IFF_IPV4 in ipif_allocate and become
15860 15866 * IFF_IPV4 or IFF_IPV6 here depending on lifr_flags value.
15861 15867 * The only flags that we read from user space are IFF_IPV4,
15862 15868 * IFF_IPV6, and IFF_BROADCAST.
15863 15869 *
15864 15870 * This ill has not been inserted into the global list.
15865 15871 * So we are still single threaded and don't need any lock
15866 15872 *
15867 15873 * Saniy check the flags.
15868 15874 */
15869 15875
15870 15876 if ((lifr->lifr_flags & IFF_BROADCAST) &&
15871 15877 ((lifr->lifr_flags & IFF_IPV6) ||
15872 15878 (!ill->ill_needs_attach && ill->ill_bcast_addr_length == 0))) {
15873 15879 ip1dbg(("ip_sioctl_slifname: link not broadcast capable "
15874 15880 "or IPv6 i.e., no broadcast \n"));
15875 15881 return (EINVAL);
15876 15882 }
15877 15883
15878 15884 new_flags =
15879 15885 lifr->lifr_flags & (IFF_IPV6|IFF_IPV4|IFF_BROADCAST);
15880 15886
15881 15887 if ((new_flags ^ (IFF_IPV6|IFF_IPV4)) == 0) {
15882 15888 ip1dbg(("ip_sioctl_slifname: flags must be exactly one of "
15883 15889 "IFF_IPV4 or IFF_IPV6\n"));
15884 15890 return (EINVAL);
15885 15891 }
15886 15892
15887 15893 /*
15888 15894 * We always start off as IPv4, so only need to check for IPv6.
15889 15895 */
15890 15896 if ((new_flags & IFF_IPV6) != 0) {
15891 15897 ill->ill_flags |= ILLF_IPV6;
15892 15898 ill->ill_flags &= ~ILLF_IPV4;
15893 15899
15894 15900 if (lifr->lifr_flags & IFF_NOLINKLOCAL)
15895 15901 ill->ill_flags |= ILLF_NOLINKLOCAL;
15896 15902 }
15897 15903
15898 15904 if ((new_flags & IFF_BROADCAST) != 0)
15899 15905 ipif->ipif_flags |= IPIF_BROADCAST;
15900 15906 else
15901 15907 ipif->ipif_flags &= ~IPIF_BROADCAST;
15902 15908
15903 15909 /* We started off as V4. */
15904 15910 if (ill->ill_flags & ILLF_IPV6) {
15905 15911 ill->ill_phyint->phyint_illv6 = ill;
15906 15912 ill->ill_phyint->phyint_illv4 = NULL;
15907 15913 }
15908 15914
15909 15915 return (ipif_set_values(q, mp, lifr->lifr_name, &lifr->lifr_ppa));
15910 15916 }
15911 15917
15912 15918 /* ARGSUSED */
15913 15919 int
15914 15920 ip_sioctl_slifname_restart(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
15915 15921 ip_ioctl_cmd_t *ipip, void *if_req)
15916 15922 {
15917 15923 /*
15918 15924 * ill_phyint_reinit merged the v4 and v6 into a single
15919 15925 * ipsq. We might not have been able to complete the
15920 15926 * slifname in ipif_set_values, if we could not become
15921 15927 * exclusive. If so restart it here
15922 15928 */
15923 15929 return (ipif_set_values_tail(ipif->ipif_ill, ipif, mp, q));
15924 15930 }
15925 15931
15926 15932 /*
15927 15933 * Return a pointer to the ipif which matches the index, IP version type and
15928 15934 * zoneid.
15929 15935 */
15930 15936 ipif_t *
15931 15937 ipif_lookup_on_ifindex(uint_t index, boolean_t isv6, zoneid_t zoneid,
15932 15938 ip_stack_t *ipst)
15933 15939 {
15934 15940 ill_t *ill;
15935 15941 ipif_t *ipif = NULL;
15936 15942
15937 15943 ill = ill_lookup_on_ifindex(index, isv6, ipst);
15938 15944 if (ill != NULL) {
15939 15945 mutex_enter(&ill->ill_lock);
15940 15946 for (ipif = ill->ill_ipif; ipif != NULL;
15941 15947 ipif = ipif->ipif_next) {
15942 15948 if (!IPIF_IS_CONDEMNED(ipif) && (zoneid == ALL_ZONES ||
15943 15949 zoneid == ipif->ipif_zoneid ||
15944 15950 ipif->ipif_zoneid == ALL_ZONES)) {
15945 15951 ipif_refhold_locked(ipif);
15946 15952 break;
15947 15953 }
15948 15954 }
15949 15955 mutex_exit(&ill->ill_lock);
15950 15956 ill_refrele(ill);
15951 15957 }
15952 15958 return (ipif);
15953 15959 }
15954 15960
15955 15961 /*
15956 15962 * Change an existing physical interface's index. If the new index
15957 15963 * is acceptable we update the index and the phyint_list_avl_by_index tree.
15958 15964 * Finally, we update other systems which may have a dependence on the
15959 15965 * index value.
15960 15966 */
15961 15967 /* ARGSUSED */
15962 15968 int
15963 15969 ip_sioctl_slifindex(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
15964 15970 ip_ioctl_cmd_t *ipip, void *ifreq)
15965 15971 {
15966 15972 ill_t *ill;
15967 15973 phyint_t *phyi;
15968 15974 struct ifreq *ifr = (struct ifreq *)ifreq;
15969 15975 struct lifreq *lifr = (struct lifreq *)ifreq;
15970 15976 uint_t old_index, index;
15971 15977 ip_stack_t *ipst = ipif->ipif_ill->ill_ipst;
15972 15978 avl_index_t where;
15973 15979
15974 15980 if (ipip->ipi_cmd_type == IF_CMD)
15975 15981 index = ifr->ifr_index;
15976 15982 else
15977 15983 index = lifr->lifr_index;
15978 15984
15979 15985 /*
15980 15986 * Only allow on physical interface. Also, index zero is illegal.
15981 15987 */
15982 15988 ill = ipif->ipif_ill;
15983 15989 phyi = ill->ill_phyint;
15984 15990 if (ipif->ipif_id != 0 || index == 0 || index > IF_INDEX_MAX) {
15985 15991 return (EINVAL);
15986 15992 }
15987 15993
15988 15994 /* If the index is not changing, no work to do */
15989 15995 if (phyi->phyint_ifindex == index)
15990 15996 return (0);
15991 15997
15992 15998 /*
15993 15999 * Use phyint_exists() to determine if the new interface index
15994 16000 * is already in use. If the index is unused then we need to
15995 16001 * change the phyint's position in the phyint_list_avl_by_index
15996 16002 * tree. If we do not do this, subsequent lookups (using the new
15997 16003 * index value) will not find the phyint.
15998 16004 */
15999 16005 rw_enter(&ipst->ips_ill_g_lock, RW_WRITER);
16000 16006 if (phyint_exists(index, ipst)) {
16001 16007 rw_exit(&ipst->ips_ill_g_lock);
16002 16008 return (EEXIST);
16003 16009 }
16004 16010
16005 16011 /*
16006 16012 * The new index is unused. Set it in the phyint. However we must not
16007 16013 * forget to trigger NE_IFINDEX_CHANGE event before the ifindex
16008 16014 * changes. The event must be bound to old ifindex value.
16009 16015 */
16010 16016 ill_nic_event_dispatch(ill, 0, NE_IFINDEX_CHANGE,
16011 16017 &index, sizeof (index));
16012 16018
16013 16019 old_index = phyi->phyint_ifindex;
16014 16020 phyi->phyint_ifindex = index;
16015 16021
16016 16022 avl_remove(&ipst->ips_phyint_g_list->phyint_list_avl_by_index, phyi);
16017 16023 (void) avl_find(&ipst->ips_phyint_g_list->phyint_list_avl_by_index,
16018 16024 &index, &where);
16019 16025 avl_insert(&ipst->ips_phyint_g_list->phyint_list_avl_by_index,
16020 16026 phyi, where);
16021 16027 rw_exit(&ipst->ips_ill_g_lock);
16022 16028
16023 16029 /* Update SCTP's ILL list */
16024 16030 sctp_ill_reindex(ill, old_index);
16025 16031
16026 16032 /* Send the routing sockets message */
16027 16033 ip_rts_ifmsg(ipif, RTSQ_DEFAULT);
16028 16034 if (ILL_OTHER(ill))
16029 16035 ip_rts_ifmsg(ILL_OTHER(ill)->ill_ipif, RTSQ_DEFAULT);
16030 16036
16031 16037 /* Perhaps ilgs should use this ill */
16032 16038 update_conn_ill(NULL, ill->ill_ipst);
16033 16039 return (0);
16034 16040 }
16035 16041
16036 16042 /* ARGSUSED */
16037 16043 int
16038 16044 ip_sioctl_get_lifindex(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
16039 16045 ip_ioctl_cmd_t *ipip, void *ifreq)
16040 16046 {
16041 16047 struct ifreq *ifr = (struct ifreq *)ifreq;
16042 16048 struct lifreq *lifr = (struct lifreq *)ifreq;
16043 16049
16044 16050 ip1dbg(("ip_sioctl_get_lifindex(%s:%u %p)\n",
16045 16051 ipif->ipif_ill->ill_name, ipif->ipif_id, (void *)ipif));
16046 16052 /* Get the interface index */
16047 16053 if (ipip->ipi_cmd_type == IF_CMD) {
16048 16054 ifr->ifr_index = ipif->ipif_ill->ill_phyint->phyint_ifindex;
16049 16055 } else {
16050 16056 lifr->lifr_index = ipif->ipif_ill->ill_phyint->phyint_ifindex;
16051 16057 }
16052 16058 return (0);
16053 16059 }
16054 16060
16055 16061 /* ARGSUSED */
16056 16062 int
16057 16063 ip_sioctl_get_lifzone(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
16058 16064 ip_ioctl_cmd_t *ipip, void *ifreq)
16059 16065 {
16060 16066 struct lifreq *lifr = (struct lifreq *)ifreq;
16061 16067
16062 16068 ip1dbg(("ip_sioctl_get_lifzone(%s:%u %p)\n",
16063 16069 ipif->ipif_ill->ill_name, ipif->ipif_id, (void *)ipif));
16064 16070 /* Get the interface zone */
16065 16071 ASSERT(ipip->ipi_cmd_type == LIF_CMD);
16066 16072 lifr->lifr_zoneid = ipif->ipif_zoneid;
16067 16073 return (0);
16068 16074 }
16069 16075
16070 16076 /*
16071 16077 * Set the zoneid of an interface.
16072 16078 */
16073 16079 /* ARGSUSED */
16074 16080 int
16075 16081 ip_sioctl_slifzone(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
16076 16082 ip_ioctl_cmd_t *ipip, void *ifreq)
16077 16083 {
16078 16084 struct lifreq *lifr = (struct lifreq *)ifreq;
16079 16085 int err = 0;
16080 16086 boolean_t need_up = B_FALSE;
16081 16087 zone_t *zptr;
16082 16088 zone_status_t status;
16083 16089 zoneid_t zoneid;
16084 16090
16085 16091 ASSERT(ipip->ipi_cmd_type == LIF_CMD);
16086 16092 if ((zoneid = lifr->lifr_zoneid) == ALL_ZONES) {
16087 16093 if (!is_system_labeled())
16088 16094 return (ENOTSUP);
16089 16095 zoneid = GLOBAL_ZONEID;
16090 16096 }
16091 16097
16092 16098 /* cannot assign instance zero to a non-global zone */
16093 16099 if (ipif->ipif_id == 0 && zoneid != GLOBAL_ZONEID)
16094 16100 return (ENOTSUP);
16095 16101
16096 16102 /*
16097 16103 * Cannot assign to a zone that doesn't exist or is shutting down. In
16098 16104 * the event of a race with the zone shutdown processing, since IP
16099 16105 * serializes this ioctl and SIOCGLIFCONF/SIOCLIFREMOVEIF, we know the
16100 16106 * interface will be cleaned up even if the zone is shut down
16101 16107 * immediately after the status check. If the interface can't be brought
16102 16108 * down right away, and the zone is shut down before the restart
16103 16109 * function is called, we resolve the possible races by rechecking the
16104 16110 * zone status in the restart function.
16105 16111 */
16106 16112 if ((zptr = zone_find_by_id(zoneid)) == NULL)
16107 16113 return (EINVAL);
16108 16114 status = zone_status_get(zptr);
16109 16115 zone_rele(zptr);
16110 16116
16111 16117 if (status != ZONE_IS_READY && status != ZONE_IS_RUNNING)
16112 16118 return (EINVAL);
16113 16119
16114 16120 if (ipif->ipif_flags & IPIF_UP) {
16115 16121 /*
16116 16122 * If the interface is already marked up,
16117 16123 * we call ipif_down which will take care
16118 16124 * of ditching any IREs that have been set
16119 16125 * up based on the old interface address.
16120 16126 */
16121 16127 err = ipif_logical_down(ipif, q, mp);
16122 16128 if (err == EINPROGRESS)
16123 16129 return (err);
16124 16130 (void) ipif_down_tail(ipif);
16125 16131 need_up = B_TRUE;
16126 16132 }
16127 16133
16128 16134 err = ip_sioctl_slifzone_tail(ipif, lifr->lifr_zoneid, q, mp, need_up);
16129 16135 return (err);
16130 16136 }
16131 16137
16132 16138 static int
16133 16139 ip_sioctl_slifzone_tail(ipif_t *ipif, zoneid_t zoneid,
16134 16140 queue_t *q, mblk_t *mp, boolean_t need_up)
16135 16141 {
16136 16142 int err = 0;
16137 16143 ip_stack_t *ipst;
16138 16144
16139 16145 ip1dbg(("ip_sioctl_zoneid_tail(%s:%u %p)\n",
16140 16146 ipif->ipif_ill->ill_name, ipif->ipif_id, (void *)ipif));
16141 16147
16142 16148 if (CONN_Q(q))
16143 16149 ipst = CONNQ_TO_IPST(q);
16144 16150 else
16145 16151 ipst = ILLQ_TO_IPST(q);
16146 16152
16147 16153 /*
16148 16154 * For exclusive stacks we don't allow a different zoneid than
16149 16155 * global.
16150 16156 */
16151 16157 if (ipst->ips_netstack->netstack_stackid != GLOBAL_NETSTACKID &&
16152 16158 zoneid != GLOBAL_ZONEID)
16153 16159 return (EINVAL);
16154 16160
16155 16161 /* Set the new zone id. */
16156 16162 ipif->ipif_zoneid = zoneid;
16157 16163
16158 16164 /* Update sctp list */
16159 16165 sctp_update_ipif(ipif, SCTP_IPIF_UPDATE);
16160 16166
16161 16167 /* The default multicast interface might have changed */
16162 16168 ire_increment_multicast_generation(ipst, ipif->ipif_ill->ill_isv6);
16163 16169
16164 16170 if (need_up) {
16165 16171 /*
16166 16172 * Now bring the interface back up. If this
16167 16173 * is the only IPIF for the ILL, ipif_up
16168 16174 * will have to re-bind to the device, so
16169 16175 * we may get back EINPROGRESS, in which
16170 16176 * case, this IOCTL will get completed in
16171 16177 * ip_rput_dlpi when we see the DL_BIND_ACK.
16172 16178 */
16173 16179 err = ipif_up(ipif, q, mp);
16174 16180 }
16175 16181 return (err);
16176 16182 }
16177 16183
16178 16184 /* ARGSUSED */
16179 16185 int
16180 16186 ip_sioctl_slifzone_restart(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
16181 16187 ip_ioctl_cmd_t *ipip, void *if_req)
16182 16188 {
16183 16189 struct lifreq *lifr = (struct lifreq *)if_req;
16184 16190 zoneid_t zoneid;
16185 16191 zone_t *zptr;
16186 16192 zone_status_t status;
16187 16193
16188 16194 ASSERT(ipip->ipi_cmd_type == LIF_CMD);
16189 16195 if ((zoneid = lifr->lifr_zoneid) == ALL_ZONES)
16190 16196 zoneid = GLOBAL_ZONEID;
16191 16197
16192 16198 ip1dbg(("ip_sioctl_slifzone_restart(%s:%u %p)\n",
16193 16199 ipif->ipif_ill->ill_name, ipif->ipif_id, (void *)ipif));
16194 16200
16195 16201 /*
16196 16202 * We recheck the zone status to resolve the following race condition:
16197 16203 * 1) process sends SIOCSLIFZONE to put hme0:1 in zone "myzone";
16198 16204 * 2) hme0:1 is up and can't be brought down right away;
16199 16205 * ip_sioctl_slifzone() returns EINPROGRESS and the request is queued;
16200 16206 * 3) zone "myzone" is halted; the zone status switches to
16201 16207 * 'shutting_down' and the zones framework sends SIOCGLIFCONF to list
16202 16208 * the interfaces to remove - hme0:1 is not returned because it's not
16203 16209 * yet in "myzone", so it won't be removed;
16204 16210 * 4) the restart function for SIOCSLIFZONE is called; without the
16205 16211 * status check here, we would have hme0:1 in "myzone" after it's been
16206 16212 * destroyed.
16207 16213 * Note that if the status check fails, we need to bring the interface
16208 16214 * back to its state prior to ip_sioctl_slifzone(), hence the call to
16209 16215 * ipif_up_done[_v6]().
16210 16216 */
16211 16217 status = ZONE_IS_UNINITIALIZED;
16212 16218 if ((zptr = zone_find_by_id(zoneid)) != NULL) {
16213 16219 status = zone_status_get(zptr);
16214 16220 zone_rele(zptr);
16215 16221 }
16216 16222 if (status != ZONE_IS_READY && status != ZONE_IS_RUNNING) {
16217 16223 if (ipif->ipif_isv6) {
16218 16224 (void) ipif_up_done_v6(ipif);
16219 16225 } else {
16220 16226 (void) ipif_up_done(ipif);
16221 16227 }
16222 16228 return (EINVAL);
16223 16229 }
16224 16230
16225 16231 (void) ipif_down_tail(ipif);
16226 16232
16227 16233 return (ip_sioctl_slifzone_tail(ipif, lifr->lifr_zoneid, q, mp,
16228 16234 B_TRUE));
16229 16235 }
16230 16236
16231 16237 /*
16232 16238 * Return the number of addresses on `ill' with one or more of the values
16233 16239 * in `set' set and all of the values in `clear' clear.
16234 16240 */
16235 16241 static uint_t
16236 16242 ill_flagaddr_cnt(const ill_t *ill, uint64_t set, uint64_t clear)
16237 16243 {
16238 16244 ipif_t *ipif;
16239 16245 uint_t cnt = 0;
16240 16246
16241 16247 ASSERT(IAM_WRITER_ILL(ill));
16242 16248
16243 16249 for (ipif = ill->ill_ipif; ipif != NULL; ipif = ipif->ipif_next)
16244 16250 if ((ipif->ipif_flags & set) && !(ipif->ipif_flags & clear))
16245 16251 cnt++;
16246 16252
16247 16253 return (cnt);
16248 16254 }
16249 16255
16250 16256 /*
16251 16257 * Return the number of migratable addresses on `ill' that are under
16252 16258 * application control.
16253 16259 */
16254 16260 uint_t
16255 16261 ill_appaddr_cnt(const ill_t *ill)
16256 16262 {
16257 16263 return (ill_flagaddr_cnt(ill, IPIF_DHCPRUNNING | IPIF_ADDRCONF,
16258 16264 IPIF_NOFAILOVER));
16259 16265 }
16260 16266
16261 16267 /*
16262 16268 * Return the number of point-to-point addresses on `ill'.
16263 16269 */
16264 16270 uint_t
16265 16271 ill_ptpaddr_cnt(const ill_t *ill)
16266 16272 {
16267 16273 return (ill_flagaddr_cnt(ill, IPIF_POINTOPOINT, 0));
16268 16274 }
16269 16275
16270 16276 /* ARGSUSED */
16271 16277 int
16272 16278 ip_sioctl_get_lifusesrc(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
16273 16279 ip_ioctl_cmd_t *ipip, void *ifreq)
16274 16280 {
16275 16281 struct lifreq *lifr = ifreq;
16276 16282
16277 16283 ASSERT(q->q_next == NULL);
16278 16284 ASSERT(CONN_Q(q));
16279 16285
16280 16286 ip1dbg(("ip_sioctl_get_lifusesrc(%s:%u %p)\n",
16281 16287 ipif->ipif_ill->ill_name, ipif->ipif_id, (void *)ipif));
16282 16288 lifr->lifr_index = ipif->ipif_ill->ill_usesrc_ifindex;
16283 16289 ip1dbg(("ip_sioctl_get_lifusesrc:lifr_index = %d\n", lifr->lifr_index));
16284 16290
16285 16291 return (0);
16286 16292 }
16287 16293
16288 16294 /* Find the previous ILL in this usesrc group */
16289 16295 static ill_t *
16290 16296 ill_prev_usesrc(ill_t *uill)
16291 16297 {
16292 16298 ill_t *ill;
16293 16299
16294 16300 for (ill = uill->ill_usesrc_grp_next;
16295 16301 ASSERT(ill), ill->ill_usesrc_grp_next != uill;
16296 16302 ill = ill->ill_usesrc_grp_next)
16297 16303 /* do nothing */;
16298 16304 return (ill);
16299 16305 }
16300 16306
16301 16307 /*
16302 16308 * Release all members of the usesrc group. This routine is called
16303 16309 * from ill_delete when the interface being unplumbed is the
16304 16310 * group head.
16305 16311 *
16306 16312 * This silently clears the usesrc that ifconfig setup.
16307 16313 * An alternative would be to keep that ifindex, and drop packets on the floor
16308 16314 * since no source address can be selected.
16309 16315 * Even if we keep the current semantics, don't need a lock and a linked list.
16310 16316 * Can walk all the ills checking if they have a ill_usesrc_ifindex matching
16311 16317 * the one that is being removed. Issue is how we return the usesrc users
16312 16318 * (SIOCGLIFSRCOF). We want to be able to find the ills which have an
16313 16319 * ill_usesrc_ifindex matching a target ill. We could also do that with an
16314 16320 * ill walk, but the walker would need to insert in the ioctl response.
16315 16321 */
16316 16322 static void
16317 16323 ill_disband_usesrc_group(ill_t *uill)
16318 16324 {
16319 16325 ill_t *next_ill, *tmp_ill;
16320 16326 ip_stack_t *ipst = uill->ill_ipst;
16321 16327
16322 16328 ASSERT(RW_WRITE_HELD(&ipst->ips_ill_g_usesrc_lock));
16323 16329 next_ill = uill->ill_usesrc_grp_next;
16324 16330
16325 16331 do {
16326 16332 ASSERT(next_ill != NULL);
16327 16333 tmp_ill = next_ill->ill_usesrc_grp_next;
16328 16334 ASSERT(tmp_ill != NULL);
16329 16335 next_ill->ill_usesrc_grp_next = NULL;
16330 16336 next_ill->ill_usesrc_ifindex = 0;
16331 16337 next_ill = tmp_ill;
16332 16338 } while (next_ill->ill_usesrc_ifindex != 0);
16333 16339 uill->ill_usesrc_grp_next = NULL;
16334 16340 }
16335 16341
16336 16342 /*
16337 16343 * Remove the client usesrc ILL from the list and relink to a new list
16338 16344 */
16339 16345 int
16340 16346 ill_relink_usesrc_ills(ill_t *ucill, ill_t *uill, uint_t ifindex)
16341 16347 {
16342 16348 ill_t *ill, *tmp_ill;
16343 16349 ip_stack_t *ipst = ucill->ill_ipst;
16344 16350
16345 16351 ASSERT((ucill != NULL) && (ucill->ill_usesrc_grp_next != NULL) &&
16346 16352 (uill != NULL) && RW_WRITE_HELD(&ipst->ips_ill_g_usesrc_lock));
16347 16353
16348 16354 /*
16349 16355 * Check if the usesrc client ILL passed in is not already
16350 16356 * in use as a usesrc ILL i.e one whose source address is
16351 16357 * in use OR a usesrc ILL is not already in use as a usesrc
16352 16358 * client ILL
16353 16359 */
16354 16360 if ((ucill->ill_usesrc_ifindex == 0) ||
16355 16361 (uill->ill_usesrc_ifindex != 0)) {
16356 16362 return (-1);
16357 16363 }
16358 16364
16359 16365 ill = ill_prev_usesrc(ucill);
16360 16366 ASSERT(ill->ill_usesrc_grp_next != NULL);
16361 16367
16362 16368 /* Remove from the current list */
16363 16369 if (ill->ill_usesrc_grp_next->ill_usesrc_grp_next == ill) {
16364 16370 /* Only two elements in the list */
16365 16371 ASSERT(ill->ill_usesrc_ifindex == 0);
16366 16372 ill->ill_usesrc_grp_next = NULL;
16367 16373 } else {
16368 16374 ill->ill_usesrc_grp_next = ucill->ill_usesrc_grp_next;
16369 16375 }
16370 16376
16371 16377 if (ifindex == 0) {
16372 16378 ucill->ill_usesrc_ifindex = 0;
16373 16379 ucill->ill_usesrc_grp_next = NULL;
16374 16380 return (0);
16375 16381 }
16376 16382
16377 16383 ucill->ill_usesrc_ifindex = ifindex;
16378 16384 tmp_ill = uill->ill_usesrc_grp_next;
16379 16385 uill->ill_usesrc_grp_next = ucill;
16380 16386 ucill->ill_usesrc_grp_next =
16381 16387 (tmp_ill != NULL) ? tmp_ill : uill;
16382 16388 return (0);
16383 16389 }
16384 16390
16385 16391 /*
16386 16392 * Set the ill_usesrc and ill_usesrc_head fields. See synchronization notes in
16387 16393 * ip.c for locking details.
16388 16394 */
16389 16395 /* ARGSUSED */
16390 16396 int
16391 16397 ip_sioctl_slifusesrc(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
16392 16398 ip_ioctl_cmd_t *ipip, void *ifreq)
16393 16399 {
16394 16400 struct lifreq *lifr = (struct lifreq *)ifreq;
16395 16401 boolean_t isv6 = B_FALSE, reset_flg = B_FALSE;
16396 16402 ill_t *usesrc_ill, *usesrc_cli_ill = ipif->ipif_ill;
16397 16403 int err = 0, ret;
16398 16404 uint_t ifindex;
16399 16405 ipsq_t *ipsq = NULL;
16400 16406 ip_stack_t *ipst = ipif->ipif_ill->ill_ipst;
16401 16407
16402 16408 ASSERT(IAM_WRITER_IPIF(ipif));
16403 16409 ASSERT(q->q_next == NULL);
16404 16410 ASSERT(CONN_Q(q));
16405 16411
16406 16412 isv6 = (Q_TO_CONN(q))->conn_family == AF_INET6;
16407 16413
16408 16414 ifindex = lifr->lifr_index;
16409 16415 if (ifindex == 0) {
16410 16416 if (usesrc_cli_ill->ill_usesrc_grp_next == NULL) {
16411 16417 /* non usesrc group interface, nothing to reset */
16412 16418 return (0);
16413 16419 }
16414 16420 ifindex = usesrc_cli_ill->ill_usesrc_ifindex;
16415 16421 /* valid reset request */
16416 16422 reset_flg = B_TRUE;
16417 16423 }
16418 16424
16419 16425 usesrc_ill = ill_lookup_on_ifindex(ifindex, isv6, ipst);
16420 16426 if (usesrc_ill == NULL)
16421 16427 return (ENXIO);
16422 16428 if (usesrc_ill == ipif->ipif_ill) {
16423 16429 ill_refrele(usesrc_ill);
16424 16430 return (EINVAL);
16425 16431 }
16426 16432
16427 16433 ipsq = ipsq_try_enter(NULL, usesrc_ill, q, mp, ip_process_ioctl,
16428 16434 NEW_OP, B_TRUE);
16429 16435 if (ipsq == NULL) {
16430 16436 err = EINPROGRESS;
16431 16437 /* Operation enqueued on the ipsq of the usesrc ILL */
16432 16438 goto done;
16433 16439 }
16434 16440
16435 16441 /* USESRC isn't currently supported with IPMP */
16436 16442 if (IS_IPMP(usesrc_ill) || IS_UNDER_IPMP(usesrc_ill)) {
16437 16443 err = ENOTSUP;
16438 16444 goto done;
16439 16445 }
16440 16446
16441 16447 /*
16442 16448 * USESRC isn't compatible with the STANDBY flag. (STANDBY is only
16443 16449 * used by IPMP underlying interfaces, but someone might think it's
16444 16450 * more general and try to use it independently with VNI.)
16445 16451 */
16446 16452 if (usesrc_ill->ill_phyint->phyint_flags & PHYI_STANDBY) {
16447 16453 err = ENOTSUP;
16448 16454 goto done;
16449 16455 }
16450 16456
16451 16457 /*
16452 16458 * If the client is already in use as a usesrc_ill or a usesrc_ill is
16453 16459 * already a client then return EINVAL
16454 16460 */
16455 16461 if (IS_USESRC_ILL(usesrc_cli_ill) || IS_USESRC_CLI_ILL(usesrc_ill)) {
16456 16462 err = EINVAL;
16457 16463 goto done;
16458 16464 }
16459 16465
16460 16466 /*
16461 16467 * If the ill_usesrc_ifindex field is already set to what it needs to
16462 16468 * be then this is a duplicate operation.
16463 16469 */
16464 16470 if (!reset_flg && usesrc_cli_ill->ill_usesrc_ifindex == ifindex) {
16465 16471 err = 0;
16466 16472 goto done;
16467 16473 }
16468 16474
16469 16475 ip1dbg(("ip_sioctl_slifusesrc: usesrc_cli_ill %s, usesrc_ill %s,"
16470 16476 " v6 = %d", usesrc_cli_ill->ill_name, usesrc_ill->ill_name,
16471 16477 usesrc_ill->ill_isv6));
16472 16478
16473 16479 /*
16474 16480 * ill_g_usesrc_lock global lock protects the ill_usesrc_grp_next
16475 16481 * and the ill_usesrc_ifindex fields
16476 16482 */
16477 16483 rw_enter(&ipst->ips_ill_g_usesrc_lock, RW_WRITER);
16478 16484
16479 16485 if (reset_flg) {
16480 16486 ret = ill_relink_usesrc_ills(usesrc_cli_ill, usesrc_ill, 0);
16481 16487 if (ret != 0) {
16482 16488 err = EINVAL;
16483 16489 }
16484 16490 rw_exit(&ipst->ips_ill_g_usesrc_lock);
16485 16491 goto done;
16486 16492 }
16487 16493
16488 16494 /*
16489 16495 * Four possibilities to consider:
16490 16496 * 1. Both usesrc_ill and usesrc_cli_ill are not part of any usesrc grp
16491 16497 * 2. usesrc_ill is part of a group but usesrc_cli_ill isn't
16492 16498 * 3. usesrc_cli_ill is part of a group but usesrc_ill isn't
16493 16499 * 4. Both are part of their respective usesrc groups
16494 16500 */
16495 16501 if ((usesrc_ill->ill_usesrc_grp_next == NULL) &&
16496 16502 (usesrc_cli_ill->ill_usesrc_grp_next == NULL)) {
16497 16503 ASSERT(usesrc_ill->ill_usesrc_ifindex == 0);
16498 16504 usesrc_cli_ill->ill_usesrc_ifindex = ifindex;
16499 16505 usesrc_ill->ill_usesrc_grp_next = usesrc_cli_ill;
16500 16506 usesrc_cli_ill->ill_usesrc_grp_next = usesrc_ill;
16501 16507 } else if ((usesrc_ill->ill_usesrc_grp_next != NULL) &&
16502 16508 (usesrc_cli_ill->ill_usesrc_grp_next == NULL)) {
16503 16509 usesrc_cli_ill->ill_usesrc_ifindex = ifindex;
16504 16510 /* Insert at head of list */
16505 16511 usesrc_cli_ill->ill_usesrc_grp_next =
16506 16512 usesrc_ill->ill_usesrc_grp_next;
16507 16513 usesrc_ill->ill_usesrc_grp_next = usesrc_cli_ill;
16508 16514 } else {
16509 16515 ret = ill_relink_usesrc_ills(usesrc_cli_ill, usesrc_ill,
16510 16516 ifindex);
16511 16517 if (ret != 0)
16512 16518 err = EINVAL;
16513 16519 }
16514 16520 rw_exit(&ipst->ips_ill_g_usesrc_lock);
16515 16521
16516 16522 done:
16517 16523 if (ipsq != NULL)
16518 16524 ipsq_exit(ipsq);
16519 16525 /* The refrele on the lifr_name ipif is done by ip_process_ioctl */
16520 16526 ill_refrele(usesrc_ill);
16521 16527
16522 16528 /* Let conn_ixa caching know that source address selection changed */
16523 16529 ip_update_source_selection(ipst);
16524 16530
16525 16531 return (err);
16526 16532 }
16527 16533
16528 16534 /* ARGSUSED */
16529 16535 int
16530 16536 ip_sioctl_get_dadstate(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
16531 16537 ip_ioctl_cmd_t *ipip, void *if_req)
16532 16538 {
16533 16539 struct lifreq *lifr = (struct lifreq *)if_req;
16534 16540 ill_t *ill = ipif->ipif_ill;
16535 16541
16536 16542 /*
16537 16543 * Need a lock since IFF_UP can be set even when there are
16538 16544 * references to the ipif.
16539 16545 */
16540 16546 mutex_enter(&ill->ill_lock);
16541 16547 if ((ipif->ipif_flags & IPIF_UP) && ipif->ipif_addr_ready == 0)
16542 16548 lifr->lifr_dadstate = DAD_IN_PROGRESS;
16543 16549 else
16544 16550 lifr->lifr_dadstate = DAD_DONE;
16545 16551 mutex_exit(&ill->ill_lock);
16546 16552 return (0);
16547 16553 }
16548 16554
16549 16555 /*
16550 16556 * comparison function used by avl.
16551 16557 */
16552 16558 static int
16553 16559 ill_phyint_compare_index(const void *index_ptr, const void *phyip)
16554 16560 {
16555 16561
16556 16562 uint_t index;
16557 16563
16558 16564 ASSERT(phyip != NULL && index_ptr != NULL);
16559 16565
16560 16566 index = *((uint_t *)index_ptr);
16561 16567 /*
16562 16568 * let the phyint with the lowest index be on top.
16563 16569 */
16564 16570 if (((phyint_t *)phyip)->phyint_ifindex < index)
16565 16571 return (1);
16566 16572 if (((phyint_t *)phyip)->phyint_ifindex > index)
16567 16573 return (-1);
16568 16574 return (0);
16569 16575 }
16570 16576
16571 16577 /*
16572 16578 * comparison function used by avl.
16573 16579 */
16574 16580 static int
16575 16581 ill_phyint_compare_name(const void *name_ptr, const void *phyip)
16576 16582 {
16577 16583 ill_t *ill;
16578 16584 int res = 0;
16579 16585
16580 16586 ASSERT(phyip != NULL && name_ptr != NULL);
16581 16587
16582 16588 if (((phyint_t *)phyip)->phyint_illv4)
16583 16589 ill = ((phyint_t *)phyip)->phyint_illv4;
16584 16590 else
16585 16591 ill = ((phyint_t *)phyip)->phyint_illv6;
16586 16592 ASSERT(ill != NULL);
16587 16593
16588 16594 res = strcmp(ill->ill_name, (char *)name_ptr);
16589 16595 if (res > 0)
16590 16596 return (1);
16591 16597 else if (res < 0)
16592 16598 return (-1);
16593 16599 return (0);
16594 16600 }
16595 16601
16596 16602 /*
16597 16603 * This function is called on the unplumb path via ill_glist_delete() when
16598 16604 * there are no ills left on the phyint and thus the phyint can be freed.
16599 16605 */
16600 16606 static void
16601 16607 phyint_free(phyint_t *phyi)
16602 16608 {
16603 16609 ip_stack_t *ipst = PHYINT_TO_IPST(phyi);
16604 16610
16605 16611 ASSERT(phyi->phyint_illv4 == NULL && phyi->phyint_illv6 == NULL);
16606 16612
16607 16613 /*
16608 16614 * If this phyint was an IPMP meta-interface, blow away the group.
16609 16615 * This is safe to do because all of the illgrps have already been
16610 16616 * removed by I_PUNLINK, and thus SIOCSLIFGROUPNAME cannot find us.
16611 16617 * If we're cleaning up as a result of failed initialization,
16612 16618 * phyint_grp may be NULL.
16613 16619 */
16614 16620 if ((phyi->phyint_flags & PHYI_IPMP) && (phyi->phyint_grp != NULL)) {
16615 16621 rw_enter(&ipst->ips_ipmp_lock, RW_WRITER);
16616 16622 ipmp_grp_destroy(phyi->phyint_grp);
16617 16623 phyi->phyint_grp = NULL;
16618 16624 rw_exit(&ipst->ips_ipmp_lock);
16619 16625 }
16620 16626
16621 16627 /*
16622 16628 * If this interface was under IPMP, take it out of the group.
16623 16629 */
16624 16630 if (phyi->phyint_grp != NULL)
16625 16631 ipmp_phyint_leave_grp(phyi);
16626 16632
16627 16633 /*
16628 16634 * Delete the phyint and disassociate its ipsq. The ipsq itself
16629 16635 * will be freed in ipsq_exit().
16630 16636 */
16631 16637 phyi->phyint_ipsq->ipsq_phyint = NULL;
16632 16638 phyi->phyint_name[0] = '\0';
16633 16639
16634 16640 mi_free(phyi);
16635 16641 }
16636 16642
16637 16643 /*
16638 16644 * Attach the ill to the phyint structure which can be shared by both
16639 16645 * IPv4 and IPv6 ill. ill_init allocates a phyint to just hold flags. This
16640 16646 * function is called from ipif_set_values and ill_lookup_on_name (for
16641 16647 * loopback) where we know the name of the ill. We lookup the ill and if
16642 16648 * there is one present already with the name use that phyint. Otherwise
16643 16649 * reuse the one allocated by ill_init.
16644 16650 */
16645 16651 static void
16646 16652 ill_phyint_reinit(ill_t *ill)
16647 16653 {
16648 16654 boolean_t isv6 = ill->ill_isv6;
16649 16655 phyint_t *phyi_old;
16650 16656 phyint_t *phyi;
16651 16657 avl_index_t where = 0;
16652 16658 ill_t *ill_other = NULL;
16653 16659 ip_stack_t *ipst = ill->ill_ipst;
16654 16660
16655 16661 ASSERT(RW_WRITE_HELD(&ipst->ips_ill_g_lock));
16656 16662
16657 16663 phyi_old = ill->ill_phyint;
16658 16664 ASSERT(isv6 || (phyi_old->phyint_illv4 == ill &&
16659 16665 phyi_old->phyint_illv6 == NULL));
16660 16666 ASSERT(!isv6 || (phyi_old->phyint_illv6 == ill &&
16661 16667 phyi_old->phyint_illv4 == NULL));
16662 16668 ASSERT(phyi_old->phyint_ifindex == 0);
16663 16669
16664 16670 /*
16665 16671 * Now that our ill has a name, set it in the phyint.
16666 16672 */
16667 16673 (void) strlcpy(ill->ill_phyint->phyint_name, ill->ill_name, LIFNAMSIZ);
16668 16674
16669 16675 phyi = avl_find(&ipst->ips_phyint_g_list->phyint_list_avl_by_name,
16670 16676 ill->ill_name, &where);
16671 16677
16672 16678 /*
16673 16679 * 1. We grabbed the ill_g_lock before inserting this ill into
16674 16680 * the global list of ills. So no other thread could have located
16675 16681 * this ill and hence the ipsq of this ill is guaranteed to be empty.
16676 16682 * 2. Now locate the other protocol instance of this ill.
16677 16683 * 3. Now grab both ill locks in the right order, and the phyint lock of
16678 16684 * the new ipsq. Holding ill locks + ill_g_lock ensures that the ipsq
16679 16685 * of neither ill can change.
16680 16686 * 4. Merge the phyint and thus the ipsq as well of this ill onto the
16681 16687 * other ill.
16682 16688 * 5. Release all locks.
16683 16689 */
16684 16690
16685 16691 /*
16686 16692 * Look for IPv4 if we are initializing IPv6 or look for IPv6 if
16687 16693 * we are initializing IPv4.
16688 16694 */
16689 16695 if (phyi != NULL) {
16690 16696 ill_other = (isv6) ? phyi->phyint_illv4 : phyi->phyint_illv6;
16691 16697 ASSERT(ill_other->ill_phyint != NULL);
16692 16698 ASSERT((isv6 && !ill_other->ill_isv6) ||
16693 16699 (!isv6 && ill_other->ill_isv6));
16694 16700 GRAB_ILL_LOCKS(ill, ill_other);
16695 16701 /*
16696 16702 * We are potentially throwing away phyint_flags which
16697 16703 * could be different from the one that we obtain from
16698 16704 * ill_other->ill_phyint. But it is okay as we are assuming
16699 16705 * that the state maintained within IP is correct.
16700 16706 */
16701 16707 mutex_enter(&phyi->phyint_lock);
16702 16708 if (isv6) {
16703 16709 ASSERT(phyi->phyint_illv6 == NULL);
16704 16710 phyi->phyint_illv6 = ill;
16705 16711 } else {
16706 16712 ASSERT(phyi->phyint_illv4 == NULL);
16707 16713 phyi->phyint_illv4 = ill;
16708 16714 }
16709 16715
16710 16716 /*
16711 16717 * Delete the old phyint and make its ipsq eligible
16712 16718 * to be freed in ipsq_exit().
16713 16719 */
16714 16720 phyi_old->phyint_illv4 = NULL;
16715 16721 phyi_old->phyint_illv6 = NULL;
16716 16722 phyi_old->phyint_ipsq->ipsq_phyint = NULL;
16717 16723 phyi_old->phyint_name[0] = '\0';
16718 16724 mi_free(phyi_old);
16719 16725 } else {
16720 16726 mutex_enter(&ill->ill_lock);
16721 16727 /*
16722 16728 * We don't need to acquire any lock, since
16723 16729 * the ill is not yet visible globally and we
16724 16730 * have not yet released the ill_g_lock.
16725 16731 */
16726 16732 phyi = phyi_old;
16727 16733 mutex_enter(&phyi->phyint_lock);
16728 16734 /* XXX We need a recovery strategy here. */
16729 16735 if (!phyint_assign_ifindex(phyi, ipst))
16730 16736 cmn_err(CE_PANIC, "phyint_assign_ifindex() failed");
16731 16737
16732 16738 avl_insert(&ipst->ips_phyint_g_list->phyint_list_avl_by_name,
16733 16739 (void *)phyi, where);
16734 16740
16735 16741 (void) avl_find(&ipst->ips_phyint_g_list->
16736 16742 phyint_list_avl_by_index,
16737 16743 &phyi->phyint_ifindex, &where);
16738 16744 avl_insert(&ipst->ips_phyint_g_list->phyint_list_avl_by_index,
16739 16745 (void *)phyi, where);
16740 16746 }
16741 16747
16742 16748 /*
16743 16749 * Reassigning ill_phyint automatically reassigns the ipsq also.
16744 16750 * pending mp is not affected because that is per ill basis.
16745 16751 */
16746 16752 ill->ill_phyint = phyi;
16747 16753
16748 16754 /*
16749 16755 * Now that the phyint's ifindex has been assigned, complete the
16750 16756 * remaining
16751 16757 */
16752 16758 ill->ill_ip_mib->ipIfStatsIfIndex = ill->ill_phyint->phyint_ifindex;
16753 16759 if (ill->ill_isv6) {
16754 16760 ill->ill_icmp6_mib->ipv6IfIcmpIfIndex =
16755 16761 ill->ill_phyint->phyint_ifindex;
16756 16762 ill->ill_mcast_type = ipst->ips_mld_max_version;
16757 16763 } else {
16758 16764 ill->ill_mcast_type = ipst->ips_igmp_max_version;
16759 16765 }
16760 16766
16761 16767 /*
16762 16768 * Generate an event within the hooks framework to indicate that
16763 16769 * a new interface has just been added to IP. For this event to
16764 16770 * be generated, the network interface must, at least, have an
16765 16771 * ifindex assigned to it. (We don't generate the event for
16766 16772 * loopback since ill_lookup_on_name() has its own NE_PLUMB event.)
16767 16773 *
16768 16774 * This needs to be run inside the ill_g_lock perimeter to ensure
16769 16775 * that the ordering of delivered events to listeners matches the
16770 16776 * order of them in the kernel.
16771 16777 */
16772 16778 if (!IS_LOOPBACK(ill)) {
16773 16779 ill_nic_event_dispatch(ill, 0, NE_PLUMB, ill->ill_name,
16774 16780 ill->ill_name_length);
16775 16781 }
16776 16782 RELEASE_ILL_LOCKS(ill, ill_other);
16777 16783 mutex_exit(&phyi->phyint_lock);
16778 16784 }
16779 16785
16780 16786 /*
16781 16787 * Notify any downstream modules of the name of this interface.
16782 16788 * An M_IOCTL is used even though we don't expect a successful reply.
16783 16789 * Any reply message from the driver (presumably an M_IOCNAK) will
16784 16790 * eventually get discarded somewhere upstream. The message format is
16785 16791 * simply an SIOCSLIFNAME ioctl just as might be sent from ifconfig
16786 16792 * to IP.
16787 16793 */
16788 16794 static void
16789 16795 ip_ifname_notify(ill_t *ill, queue_t *q)
16790 16796 {
16791 16797 mblk_t *mp1, *mp2;
16792 16798 struct iocblk *iocp;
16793 16799 struct lifreq *lifr;
16794 16800
16795 16801 mp1 = mkiocb(SIOCSLIFNAME);
16796 16802 if (mp1 == NULL)
16797 16803 return;
16798 16804 mp2 = allocb(sizeof (struct lifreq), BPRI_HI);
16799 16805 if (mp2 == NULL) {
16800 16806 freeb(mp1);
16801 16807 return;
16802 16808 }
16803 16809
16804 16810 mp1->b_cont = mp2;
16805 16811 iocp = (struct iocblk *)mp1->b_rptr;
16806 16812 iocp->ioc_count = sizeof (struct lifreq);
16807 16813
16808 16814 lifr = (struct lifreq *)mp2->b_rptr;
16809 16815 mp2->b_wptr += sizeof (struct lifreq);
16810 16816 bzero(lifr, sizeof (struct lifreq));
16811 16817
16812 16818 (void) strncpy(lifr->lifr_name, ill->ill_name, LIFNAMSIZ);
16813 16819 lifr->lifr_ppa = ill->ill_ppa;
16814 16820 lifr->lifr_flags = (ill->ill_flags & (ILLF_IPV4|ILLF_IPV6));
16815 16821
16816 16822 DTRACE_PROBE3(ill__dlpi, char *, "ip_ifname_notify",
16817 16823 char *, "SIOCSLIFNAME", ill_t *, ill);
16818 16824 putnext(q, mp1);
16819 16825 }
16820 16826
16821 16827 static int
16822 16828 ipif_set_values_tail(ill_t *ill, ipif_t *ipif, mblk_t *mp, queue_t *q)
16823 16829 {
16824 16830 int err;
16825 16831 ip_stack_t *ipst = ill->ill_ipst;
16826 16832 phyint_t *phyi = ill->ill_phyint;
16827 16833
16828 16834 /*
16829 16835 * Now that ill_name is set, the configuration for the IPMP
16830 16836 * meta-interface can be performed.
16831 16837 */
16832 16838 if (IS_IPMP(ill)) {
16833 16839 rw_enter(&ipst->ips_ipmp_lock, RW_WRITER);
16834 16840 /*
16835 16841 * If phyi->phyint_grp is NULL, then this is the first IPMP
16836 16842 * meta-interface and we need to create the IPMP group.
16837 16843 */
16838 16844 if (phyi->phyint_grp == NULL) {
16839 16845 /*
16840 16846 * If someone has renamed another IPMP group to have
16841 16847 * the same name as our interface, bail.
16842 16848 */
16843 16849 if (ipmp_grp_lookup(ill->ill_name, ipst) != NULL) {
16844 16850 rw_exit(&ipst->ips_ipmp_lock);
16845 16851 return (EEXIST);
16846 16852 }
16847 16853 phyi->phyint_grp = ipmp_grp_create(ill->ill_name, phyi);
16848 16854 if (phyi->phyint_grp == NULL) {
16849 16855 rw_exit(&ipst->ips_ipmp_lock);
16850 16856 return (ENOMEM);
16851 16857 }
16852 16858 }
16853 16859 rw_exit(&ipst->ips_ipmp_lock);
16854 16860 }
16855 16861
16856 16862 /* Tell downstream modules where they are. */
16857 16863 ip_ifname_notify(ill, q);
16858 16864
16859 16865 /*
16860 16866 * ill_dl_phys returns EINPROGRESS in the usual case.
16861 16867 * Error cases are ENOMEM ...
16862 16868 */
16863 16869 err = ill_dl_phys(ill, ipif, mp, q);
16864 16870
16865 16871 if (ill->ill_isv6) {
16866 16872 mutex_enter(&ipst->ips_mld_slowtimeout_lock);
16867 16873 if (ipst->ips_mld_slowtimeout_id == 0) {
16868 16874 ipst->ips_mld_slowtimeout_id = timeout(mld_slowtimo,
16869 16875 (void *)ipst,
16870 16876 MSEC_TO_TICK(MCAST_SLOWTIMO_INTERVAL));
16871 16877 }
16872 16878 mutex_exit(&ipst->ips_mld_slowtimeout_lock);
16873 16879 } else {
16874 16880 mutex_enter(&ipst->ips_igmp_slowtimeout_lock);
16875 16881 if (ipst->ips_igmp_slowtimeout_id == 0) {
16876 16882 ipst->ips_igmp_slowtimeout_id = timeout(igmp_slowtimo,
16877 16883 (void *)ipst,
16878 16884 MSEC_TO_TICK(MCAST_SLOWTIMO_INTERVAL));
16879 16885 }
16880 16886 mutex_exit(&ipst->ips_igmp_slowtimeout_lock);
16881 16887 }
16882 16888
16883 16889 return (err);
16884 16890 }
16885 16891
16886 16892 /*
16887 16893 * Common routine for ppa and ifname setting. Should be called exclusive.
16888 16894 *
16889 16895 * Returns EINPROGRESS when mp has been consumed by queueing it on
16890 16896 * ipx_pending_mp and the ioctl will complete in ip_rput.
16891 16897 *
16892 16898 * NOTE : If ppa is UNIT_MAX, we assign the next valid ppa and return
16893 16899 * the new name and new ppa in lifr_name and lifr_ppa respectively.
16894 16900 * For SLIFNAME, we pass these values back to the userland.
16895 16901 */
16896 16902 static int
16897 16903 ipif_set_values(queue_t *q, mblk_t *mp, char *interf_name, uint_t *new_ppa_ptr)
16898 16904 {
16899 16905 ill_t *ill;
16900 16906 ipif_t *ipif;
16901 16907 ipsq_t *ipsq;
16902 16908 char *ppa_ptr;
16903 16909 char *old_ptr;
16904 16910 char old_char;
16905 16911 int error;
16906 16912 ip_stack_t *ipst;
16907 16913
16908 16914 ip1dbg(("ipif_set_values: interface %s\n", interf_name));
16909 16915 ASSERT(q->q_next != NULL);
16910 16916 ASSERT(interf_name != NULL);
16911 16917
16912 16918 ill = (ill_t *)q->q_ptr;
16913 16919 ipst = ill->ill_ipst;
16914 16920
16915 16921 ASSERT(ill->ill_ipst != NULL);
16916 16922 ASSERT(ill->ill_name[0] == '\0');
16917 16923 ASSERT(IAM_WRITER_ILL(ill));
16918 16924 ASSERT((mi_strlen(interf_name) + 1) <= LIFNAMSIZ);
16919 16925 ASSERT(ill->ill_ppa == UINT_MAX);
16920 16926
16921 16927 ill->ill_defend_start = ill->ill_defend_count = 0;
16922 16928 /* The ppa is sent down by ifconfig or is chosen */
16923 16929 if ((ppa_ptr = ill_get_ppa_ptr(interf_name)) == NULL) {
16924 16930 return (EINVAL);
16925 16931 }
16926 16932
16927 16933 /*
16928 16934 * make sure ppa passed in is same as ppa in the name.
16929 16935 * This check is not made when ppa == UINT_MAX in that case ppa
16930 16936 * in the name could be anything. System will choose a ppa and
16931 16937 * update new_ppa_ptr and inter_name to contain the choosen ppa.
16932 16938 */
16933 16939 if (*new_ppa_ptr != UINT_MAX) {
16934 16940 /* stoi changes the pointer */
16935 16941 old_ptr = ppa_ptr;
16936 16942 /*
16937 16943 * ifconfig passed in 0 for the ppa for DLPI 1 style devices
16938 16944 * (they don't have an externally visible ppa). We assign one
16939 16945 * here so that we can manage the interface. Note that in
16940 16946 * the past this value was always 0 for DLPI 1 drivers.
16941 16947 */
16942 16948 if (*new_ppa_ptr == 0)
16943 16949 *new_ppa_ptr = stoi(&old_ptr);
16944 16950 else if (*new_ppa_ptr != (uint_t)stoi(&old_ptr))
16945 16951 return (EINVAL);
16946 16952 }
16947 16953 /*
16948 16954 * terminate string before ppa
16949 16955 * save char at that location.
16950 16956 */
16951 16957 old_char = ppa_ptr[0];
16952 16958 ppa_ptr[0] = '\0';
16953 16959
16954 16960 ill->ill_ppa = *new_ppa_ptr;
16955 16961 /*
16956 16962 * Finish as much work now as possible before calling ill_glist_insert
16957 16963 * which makes the ill globally visible and also merges it with the
16958 16964 * other protocol instance of this phyint. The remaining work is
16959 16965 * done after entering the ipsq which may happen sometime later.
16960 16966 */
16961 16967 ipif = ill->ill_ipif;
16962 16968
16963 16969 /* We didn't do this when we allocated ipif in ip_ll_subnet_defaults */
16964 16970 ipif_assign_seqid(ipif);
16965 16971
16966 16972 if (!(ill->ill_flags & (ILLF_IPV4|ILLF_IPV6)))
16967 16973 ill->ill_flags |= ILLF_IPV4;
16968 16974
16969 16975 ASSERT(ipif->ipif_next == NULL); /* Only one ipif on ill */
16970 16976 ASSERT((ipif->ipif_flags & IPIF_UP) == 0);
16971 16977
16972 16978 if (ill->ill_flags & ILLF_IPV6) {
16973 16979
16974 16980 ill->ill_isv6 = B_TRUE;
16975 16981 ill_set_inputfn(ill);
16976 16982 if (ill->ill_rq != NULL) {
16977 16983 ill->ill_rq->q_qinfo = &iprinitv6;
16978 16984 }
16979 16985
16980 16986 /* Keep the !IN6_IS_ADDR_V4MAPPED assertions happy */
16981 16987 ipif->ipif_v6lcl_addr = ipv6_all_zeros;
16982 16988 ipif->ipif_v6subnet = ipv6_all_zeros;
16983 16989 ipif->ipif_v6net_mask = ipv6_all_zeros;
16984 16990 ipif->ipif_v6brd_addr = ipv6_all_zeros;
16985 16991 ipif->ipif_v6pp_dst_addr = ipv6_all_zeros;
16986 16992 ill->ill_reachable_retrans_time = ND_RETRANS_TIMER;
16987 16993 /*
16988 16994 * point-to-point or Non-mulicast capable
16989 16995 * interfaces won't do NUD unless explicitly
16990 16996 * configured to do so.
16991 16997 */
16992 16998 if (ipif->ipif_flags & IPIF_POINTOPOINT ||
16993 16999 !(ill->ill_flags & ILLF_MULTICAST)) {
16994 17000 ill->ill_flags |= ILLF_NONUD;
16995 17001 }
16996 17002 /* Make sure IPv4 specific flag is not set on IPv6 if */
16997 17003 if (ill->ill_flags & ILLF_NOARP) {
16998 17004 /*
16999 17005 * Note: xresolv interfaces will eventually need
17000 17006 * NOARP set here as well, but that will require
17001 17007 * those external resolvers to have some
17002 17008 * knowledge of that flag and act appropriately.
17003 17009 * Not to be changed at present.
17004 17010 */
17005 17011 ill->ill_flags &= ~ILLF_NOARP;
17006 17012 }
17007 17013 /*
17008 17014 * Set the ILLF_ROUTER flag according to the global
17009 17015 * IPv6 forwarding policy.
17010 17016 */
17011 17017 if (ipst->ips_ipv6_forwarding != 0)
17012 17018 ill->ill_flags |= ILLF_ROUTER;
17013 17019 } else if (ill->ill_flags & ILLF_IPV4) {
17014 17020 ill->ill_isv6 = B_FALSE;
17015 17021 ill_set_inputfn(ill);
17016 17022 ill->ill_reachable_retrans_time = ARP_RETRANS_TIMER;
17017 17023 IN6_IPADDR_TO_V4MAPPED(INADDR_ANY, &ipif->ipif_v6lcl_addr);
17018 17024 IN6_IPADDR_TO_V4MAPPED(INADDR_ANY, &ipif->ipif_v6subnet);
17019 17025 IN6_IPADDR_TO_V4MAPPED(INADDR_ANY, &ipif->ipif_v6net_mask);
17020 17026 IN6_IPADDR_TO_V4MAPPED(INADDR_ANY, &ipif->ipif_v6brd_addr);
17021 17027 IN6_IPADDR_TO_V4MAPPED(INADDR_ANY, &ipif->ipif_v6pp_dst_addr);
17022 17028 /*
17023 17029 * Set the ILLF_ROUTER flag according to the global
17024 17030 * IPv4 forwarding policy.
17025 17031 */
17026 17032 if (ipst->ips_ip_forwarding != 0)
17027 17033 ill->ill_flags |= ILLF_ROUTER;
17028 17034 }
17029 17035
17030 17036 ASSERT(ill->ill_phyint != NULL);
17031 17037
17032 17038 /*
17033 17039 * The ipIfStatsIfindex and ipv6IfIcmpIfIndex assignments will
17034 17040 * be completed in ill_glist_insert -> ill_phyint_reinit
17035 17041 */
17036 17042 if (!ill_allocate_mibs(ill))
17037 17043 return (ENOMEM);
17038 17044
17039 17045 /*
17040 17046 * Pick a default sap until we get the DL_INFO_ACK back from
17041 17047 * the driver.
17042 17048 */
17043 17049 ill->ill_sap = (ill->ill_isv6) ? ill->ill_media->ip_m_ipv6sap :
17044 17050 ill->ill_media->ip_m_ipv4sap;
17045 17051
17046 17052 ill->ill_ifname_pending = 1;
17047 17053 ill->ill_ifname_pending_err = 0;
17048 17054
17049 17055 /*
17050 17056 * When the first ipif comes up in ipif_up_done(), multicast groups
17051 17057 * that were joined while this ill was not bound to the DLPI link need
17052 17058 * to be recovered by ill_recover_multicast().
17053 17059 */
17054 17060 ill->ill_need_recover_multicast = 1;
17055 17061
17056 17062 ill_refhold(ill);
17057 17063 rw_enter(&ipst->ips_ill_g_lock, RW_WRITER);
17058 17064 if ((error = ill_glist_insert(ill, interf_name,
17059 17065 (ill->ill_flags & ILLF_IPV6) == ILLF_IPV6)) > 0) {
17060 17066 ill->ill_ppa = UINT_MAX;
17061 17067 ill->ill_name[0] = '\0';
17062 17068 /*
17063 17069 * undo null termination done above.
17064 17070 */
17065 17071 ppa_ptr[0] = old_char;
17066 17072 rw_exit(&ipst->ips_ill_g_lock);
17067 17073 ill_refrele(ill);
17068 17074 return (error);
17069 17075 }
17070 17076
17071 17077 ASSERT(ill->ill_name_length <= LIFNAMSIZ);
17072 17078
17073 17079 /*
17074 17080 * When we return the buffer pointed to by interf_name should contain
17075 17081 * the same name as in ill_name.
17076 17082 * If a ppa was choosen by the system (ppa passed in was UINT_MAX)
17077 17083 * the buffer pointed to by new_ppa_ptr would not contain the right ppa
17078 17084 * so copy full name and update the ppa ptr.
17079 17085 * When ppa passed in != UINT_MAX all values are correct just undo
17080 17086 * null termination, this saves a bcopy.
17081 17087 */
17082 17088 if (*new_ppa_ptr == UINT_MAX) {
17083 17089 bcopy(ill->ill_name, interf_name, ill->ill_name_length);
17084 17090 *new_ppa_ptr = ill->ill_ppa;
17085 17091 } else {
17086 17092 /*
17087 17093 * undo null termination done above.
17088 17094 */
17089 17095 ppa_ptr[0] = old_char;
17090 17096 }
17091 17097
17092 17098 /* Let SCTP know about this ILL */
17093 17099 sctp_update_ill(ill, SCTP_ILL_INSERT);
17094 17100
17095 17101 /*
17096 17102 * ill_glist_insert has made the ill visible globally, and
17097 17103 * ill_phyint_reinit could have changed the ipsq. At this point,
17098 17104 * we need to hold the ips_ill_g_lock across the call to enter the
17099 17105 * ipsq to enforce atomicity and prevent reordering. In the event
17100 17106 * the ipsq has changed, and if the new ipsq is currently busy,
17101 17107 * we need to make sure that this half-completed ioctl is ahead of
17102 17108 * any subsequent ioctl. We achieve this by not dropping the
17103 17109 * ips_ill_g_lock which prevents any ill lookup itself thereby
17104 17110 * ensuring that new ioctls can't start.
17105 17111 */
17106 17112 ipsq = ipsq_try_enter_internal(ill, q, mp, ip_reprocess_ioctl, NEW_OP,
17107 17113 B_TRUE);
17108 17114
17109 17115 rw_exit(&ipst->ips_ill_g_lock);
17110 17116 ill_refrele(ill);
17111 17117 if (ipsq == NULL)
17112 17118 return (EINPROGRESS);
17113 17119
17114 17120 /*
17115 17121 * If ill_phyint_reinit() changed our ipsq, then start on the new ipsq.
17116 17122 */
17117 17123 if (ipsq->ipsq_xop->ipx_current_ipif == NULL)
17118 17124 ipsq_current_start(ipsq, ipif, SIOCSLIFNAME);
17119 17125 else
17120 17126 ASSERT(ipsq->ipsq_xop->ipx_current_ipif == ipif);
17121 17127
17122 17128 error = ipif_set_values_tail(ill, ipif, mp, q);
17123 17129 ipsq_exit(ipsq);
17124 17130 if (error != 0 && error != EINPROGRESS) {
17125 17131 /*
17126 17132 * restore previous values
17127 17133 */
17128 17134 ill->ill_isv6 = B_FALSE;
17129 17135 ill_set_inputfn(ill);
17130 17136 }
17131 17137 return (error);
17132 17138 }
17133 17139
17134 17140 void
17135 17141 ipif_init(ip_stack_t *ipst)
17136 17142 {
17137 17143 int i;
17138 17144
17139 17145 for (i = 0; i < MAX_G_HEADS; i++) {
17140 17146 ipst->ips_ill_g_heads[i].ill_g_list_head =
17141 17147 (ill_if_t *)&ipst->ips_ill_g_heads[i];
17142 17148 ipst->ips_ill_g_heads[i].ill_g_list_tail =
17143 17149 (ill_if_t *)&ipst->ips_ill_g_heads[i];
17144 17150 }
17145 17151
17146 17152 avl_create(&ipst->ips_phyint_g_list->phyint_list_avl_by_index,
17147 17153 ill_phyint_compare_index,
17148 17154 sizeof (phyint_t),
17149 17155 offsetof(struct phyint, phyint_avl_by_index));
17150 17156 avl_create(&ipst->ips_phyint_g_list->phyint_list_avl_by_name,
17151 17157 ill_phyint_compare_name,
17152 17158 sizeof (phyint_t),
17153 17159 offsetof(struct phyint, phyint_avl_by_name));
17154 17160 }
17155 17161
17156 17162 /*
17157 17163 * Save enough information so that we can recreate the IRE if
17158 17164 * the interface goes down and then up.
17159 17165 */
17160 17166 void
17161 17167 ill_save_ire(ill_t *ill, ire_t *ire)
17162 17168 {
17163 17169 mblk_t *save_mp;
17164 17170
17165 17171 save_mp = allocb(sizeof (ifrt_t), BPRI_MED);
17166 17172 if (save_mp != NULL) {
17167 17173 ifrt_t *ifrt;
17168 17174
17169 17175 save_mp->b_wptr += sizeof (ifrt_t);
17170 17176 ifrt = (ifrt_t *)save_mp->b_rptr;
17171 17177 bzero(ifrt, sizeof (ifrt_t));
17172 17178 ifrt->ifrt_type = ire->ire_type;
17173 17179 if (ire->ire_ipversion == IPV4_VERSION) {
17174 17180 ASSERT(!ill->ill_isv6);
17175 17181 ifrt->ifrt_addr = ire->ire_addr;
17176 17182 ifrt->ifrt_gateway_addr = ire->ire_gateway_addr;
17177 17183 ifrt->ifrt_setsrc_addr = ire->ire_setsrc_addr;
17178 17184 ifrt->ifrt_mask = ire->ire_mask;
17179 17185 } else {
17180 17186 ASSERT(ill->ill_isv6);
17181 17187 ifrt->ifrt_v6addr = ire->ire_addr_v6;
17182 17188 /* ire_gateway_addr_v6 can change due to RTM_CHANGE */
17183 17189 mutex_enter(&ire->ire_lock);
17184 17190 ifrt->ifrt_v6gateway_addr = ire->ire_gateway_addr_v6;
17185 17191 mutex_exit(&ire->ire_lock);
17186 17192 ifrt->ifrt_v6setsrc_addr = ire->ire_setsrc_addr_v6;
17187 17193 ifrt->ifrt_v6mask = ire->ire_mask_v6;
17188 17194 }
17189 17195 ifrt->ifrt_flags = ire->ire_flags;
17190 17196 ifrt->ifrt_zoneid = ire->ire_zoneid;
17191 17197 mutex_enter(&ill->ill_saved_ire_lock);
17192 17198 save_mp->b_cont = ill->ill_saved_ire_mp;
17193 17199 ill->ill_saved_ire_mp = save_mp;
17194 17200 ill->ill_saved_ire_cnt++;
17195 17201 mutex_exit(&ill->ill_saved_ire_lock);
17196 17202 }
17197 17203 }
17198 17204
17199 17205 /*
17200 17206 * Remove one entry from ill_saved_ire_mp.
17201 17207 */
17202 17208 void
17203 17209 ill_remove_saved_ire(ill_t *ill, ire_t *ire)
17204 17210 {
17205 17211 mblk_t **mpp;
17206 17212 mblk_t *mp;
17207 17213 ifrt_t *ifrt;
17208 17214
17209 17215 /* Remove from ill_saved_ire_mp list if it is there */
17210 17216 mutex_enter(&ill->ill_saved_ire_lock);
17211 17217 for (mpp = &ill->ill_saved_ire_mp; *mpp != NULL;
17212 17218 mpp = &(*mpp)->b_cont) {
17213 17219 in6_addr_t gw_addr_v6;
17214 17220
17215 17221 /*
17216 17222 * On a given ill, the tuple of address, gateway, mask,
17217 17223 * ire_type, and zoneid is unique for each saved IRE.
17218 17224 */
17219 17225 mp = *mpp;
17220 17226 ifrt = (ifrt_t *)mp->b_rptr;
17221 17227 /* ire_gateway_addr_v6 can change - need lock */
17222 17228 mutex_enter(&ire->ire_lock);
17223 17229 gw_addr_v6 = ire->ire_gateway_addr_v6;
17224 17230 mutex_exit(&ire->ire_lock);
17225 17231
17226 17232 if (ifrt->ifrt_zoneid != ire->ire_zoneid ||
17227 17233 ifrt->ifrt_type != ire->ire_type)
17228 17234 continue;
17229 17235
17230 17236 if (ill->ill_isv6 ?
17231 17237 (IN6_ARE_ADDR_EQUAL(&ifrt->ifrt_v6addr,
17232 17238 &ire->ire_addr_v6) &&
17233 17239 IN6_ARE_ADDR_EQUAL(&ifrt->ifrt_v6gateway_addr,
17234 17240 &gw_addr_v6) &&
17235 17241 IN6_ARE_ADDR_EQUAL(&ifrt->ifrt_v6mask,
17236 17242 &ire->ire_mask_v6)) :
17237 17243 (ifrt->ifrt_addr == ire->ire_addr &&
17238 17244 ifrt->ifrt_gateway_addr == ire->ire_gateway_addr &&
17239 17245 ifrt->ifrt_mask == ire->ire_mask)) {
17240 17246 *mpp = mp->b_cont;
17241 17247 ill->ill_saved_ire_cnt--;
17242 17248 freeb(mp);
17243 17249 break;
17244 17250 }
17245 17251 }
17246 17252 mutex_exit(&ill->ill_saved_ire_lock);
17247 17253 }
17248 17254
17249 17255 /*
17250 17256 * IP multirouting broadcast routes handling
17251 17257 * Append CGTP broadcast IREs to regular ones created
17252 17258 * at ifconfig time.
17253 17259 * The usage is a route add <cgtp_bc> <nic_bc> -multirt i.e., both
17254 17260 * the destination and the gateway are broadcast addresses.
17255 17261 * The caller has verified that the destination is an IRE_BROADCAST and that
17256 17262 * RTF_MULTIRT was set. Here if the gateway is a broadcast address, then
17257 17263 * we create a MULTIRT IRE_BROADCAST.
17258 17264 * Note that the IRE_HOST created by ire_rt_add doesn't get found by anything
17259 17265 * since the IRE_BROADCAST takes precedence; ire_add_v4 does head insertion.
17260 17266 */
17261 17267 static void
17262 17268 ip_cgtp_bcast_add(ire_t *ire, ip_stack_t *ipst)
17263 17269 {
17264 17270 ire_t *ire_prim;
17265 17271
17266 17272 ASSERT(ire != NULL);
17267 17273
17268 17274 ire_prim = ire_ftable_lookup_v4(ire->ire_gateway_addr, 0, 0,
17269 17275 IRE_BROADCAST, NULL, ALL_ZONES, NULL, MATCH_IRE_TYPE, 0, ipst,
17270 17276 NULL);
17271 17277 if (ire_prim != NULL) {
17272 17278 /*
17273 17279 * We are in the special case of broadcasts for
17274 17280 * CGTP. We add an IRE_BROADCAST that holds
17275 17281 * the RTF_MULTIRT flag, the destination
17276 17282 * address and the low level
17277 17283 * info of ire_prim. In other words, CGTP
17278 17284 * broadcast is added to the redundant ipif.
17279 17285 */
17280 17286 ill_t *ill_prim;
17281 17287 ire_t *bcast_ire;
17282 17288
17283 17289 ill_prim = ire_prim->ire_ill;
17284 17290
17285 17291 ip2dbg(("ip_cgtp_filter_bcast_add: ire_prim %p, ill_prim %p\n",
17286 17292 (void *)ire_prim, (void *)ill_prim));
17287 17293
17288 17294 bcast_ire = ire_create(
17289 17295 (uchar_t *)&ire->ire_addr,
17290 17296 (uchar_t *)&ip_g_all_ones,
17291 17297 (uchar_t *)&ire->ire_gateway_addr,
17292 17298 IRE_BROADCAST,
17293 17299 ill_prim,
17294 17300 GLOBAL_ZONEID, /* CGTP is only for the global zone */
17295 17301 ire->ire_flags | RTF_KERNEL,
17296 17302 NULL,
17297 17303 ipst);
17298 17304
17299 17305 /*
17300 17306 * Here we assume that ire_add does head insertion so that
17301 17307 * the added IRE_BROADCAST comes before the existing IRE_HOST.
17302 17308 */
17303 17309 if (bcast_ire != NULL) {
17304 17310 if (ire->ire_flags & RTF_SETSRC) {
17305 17311 bcast_ire->ire_setsrc_addr =
17306 17312 ire->ire_setsrc_addr;
17307 17313 }
17308 17314 bcast_ire = ire_add(bcast_ire);
17309 17315 if (bcast_ire != NULL) {
17310 17316 ip2dbg(("ip_cgtp_filter_bcast_add: "
17311 17317 "added bcast_ire %p\n",
17312 17318 (void *)bcast_ire));
17313 17319
17314 17320 ill_save_ire(ill_prim, bcast_ire);
17315 17321 ire_refrele(bcast_ire);
17316 17322 }
17317 17323 }
17318 17324 ire_refrele(ire_prim);
17319 17325 }
17320 17326 }
17321 17327
17322 17328 /*
17323 17329 * IP multirouting broadcast routes handling
17324 17330 * Remove the broadcast ire.
17325 17331 * The usage is a route delete <cgtp_bc> <nic_bc> -multirt i.e., both
17326 17332 * the destination and the gateway are broadcast addresses.
17327 17333 * The caller has only verified that RTF_MULTIRT was set. We check
17328 17334 * that the destination is broadcast and that the gateway is a broadcast
17329 17335 * address, and if so delete the IRE added by ip_cgtp_bcast_add().
17330 17336 */
17331 17337 static void
17332 17338 ip_cgtp_bcast_delete(ire_t *ire, ip_stack_t *ipst)
17333 17339 {
17334 17340 ASSERT(ire != NULL);
17335 17341
17336 17342 if (ip_type_v4(ire->ire_addr, ipst) == IRE_BROADCAST) {
17337 17343 ire_t *ire_prim;
17338 17344
17339 17345 ire_prim = ire_ftable_lookup_v4(ire->ire_gateway_addr, 0, 0,
17340 17346 IRE_BROADCAST, NULL, ALL_ZONES, NULL, MATCH_IRE_TYPE, 0,
17341 17347 ipst, NULL);
17342 17348 if (ire_prim != NULL) {
17343 17349 ill_t *ill_prim;
17344 17350 ire_t *bcast_ire;
17345 17351
17346 17352 ill_prim = ire_prim->ire_ill;
17347 17353
17348 17354 ip2dbg(("ip_cgtp_filter_bcast_delete: "
17349 17355 "ire_prim %p, ill_prim %p\n",
17350 17356 (void *)ire_prim, (void *)ill_prim));
17351 17357
17352 17358 bcast_ire = ire_ftable_lookup_v4(ire->ire_addr, 0,
17353 17359 ire->ire_gateway_addr, IRE_BROADCAST,
17354 17360 ill_prim, ALL_ZONES, NULL,
17355 17361 MATCH_IRE_TYPE | MATCH_IRE_GW | MATCH_IRE_ILL |
17356 17362 MATCH_IRE_MASK, 0, ipst, NULL);
17357 17363
17358 17364 if (bcast_ire != NULL) {
17359 17365 ip2dbg(("ip_cgtp_filter_bcast_delete: "
17360 17366 "looked up bcast_ire %p\n",
17361 17367 (void *)bcast_ire));
17362 17368 ill_remove_saved_ire(bcast_ire->ire_ill,
17363 17369 bcast_ire);
17364 17370 ire_delete(bcast_ire);
17365 17371 ire_refrele(bcast_ire);
17366 17372 }
17367 17373 ire_refrele(ire_prim);
17368 17374 }
17369 17375 }
17370 17376 }
17371 17377
17372 17378 /*
17373 17379 * Derive an interface id from the link layer address.
17374 17380 * Knows about IEEE 802 and IEEE EUI-64 mappings.
17375 17381 */
17376 17382 static void
17377 17383 ip_ether_v6intfid(ill_t *ill, in6_addr_t *v6addr)
17378 17384 {
17379 17385 char *addr;
17380 17386
17381 17387 /*
17382 17388 * Note that some IPv6 interfaces get plumbed over links that claim to
17383 17389 * be DL_ETHER, but don't actually have Ethernet MAC addresses (e.g.
17384 17390 * PPP links). The ETHERADDRL check here ensures that we only set the
17385 17391 * interface ID on IPv6 interfaces above links that actually have real
17386 17392 * Ethernet addresses.
17387 17393 */
17388 17394 if (ill->ill_phys_addr_length == ETHERADDRL) {
17389 17395 /* Form EUI-64 like address */
17390 17396 addr = (char *)&v6addr->s6_addr32[2];
17391 17397 bcopy(ill->ill_phys_addr, addr, 3);
17392 17398 addr[0] ^= 0x2; /* Toggle Universal/Local bit */
17393 17399 addr[3] = (char)0xff;
17394 17400 addr[4] = (char)0xfe;
17395 17401 bcopy(ill->ill_phys_addr + 3, addr + 5, 3);
17396 17402 }
17397 17403 }
17398 17404
17399 17405 /* ARGSUSED */
17400 17406 static void
17401 17407 ip_nodef_v6intfid(ill_t *ill, in6_addr_t *v6addr)
17402 17408 {
17403 17409 }
17404 17410
17405 17411 typedef struct ipmp_ifcookie {
17406 17412 uint32_t ic_hostid;
17407 17413 char ic_ifname[LIFNAMSIZ];
17408 17414 char ic_zonename[ZONENAME_MAX];
17409 17415 } ipmp_ifcookie_t;
17410 17416
17411 17417 /*
17412 17418 * Construct a pseudo-random interface ID for the IPMP interface that's both
17413 17419 * predictable and (almost) guaranteed to be unique.
17414 17420 */
17415 17421 static void
17416 17422 ip_ipmp_v6intfid(ill_t *ill, in6_addr_t *v6addr)
17417 17423 {
17418 17424 zone_t *zp;
17419 17425 uint8_t *addr;
17420 17426 uchar_t hash[16];
17421 17427 ulong_t hostid;
17422 17428 MD5_CTX ctx;
17423 17429 ipmp_ifcookie_t ic = { 0 };
17424 17430
17425 17431 ASSERT(IS_IPMP(ill));
17426 17432
17427 17433 (void) ddi_strtoul(hw_serial, NULL, 10, &hostid);
17428 17434 ic.ic_hostid = htonl((uint32_t)hostid);
17429 17435
17430 17436 (void) strlcpy(ic.ic_ifname, ill->ill_name, LIFNAMSIZ);
17431 17437
17432 17438 if ((zp = zone_find_by_id(ill->ill_zoneid)) != NULL) {
17433 17439 (void) strlcpy(ic.ic_zonename, zp->zone_name, ZONENAME_MAX);
17434 17440 zone_rele(zp);
17435 17441 }
17436 17442
17437 17443 MD5Init(&ctx);
17438 17444 MD5Update(&ctx, &ic, sizeof (ic));
17439 17445 MD5Final(hash, &ctx);
17440 17446
17441 17447 /*
17442 17448 * Map the hash to an interface ID per the basic approach in RFC3041.
17443 17449 */
17444 17450 addr = &v6addr->s6_addr8[8];
17445 17451 bcopy(hash + 8, addr, sizeof (uint64_t));
17446 17452 addr[0] &= ~0x2; /* set local bit */
17447 17453 }
17448 17454
17449 17455 /*
17450 17456 * Map the multicast in6_addr_t in m_ip6addr to the physaddr for ethernet.
17451 17457 */
17452 17458 static void
17453 17459 ip_ether_v6_mapping(ill_t *ill, uchar_t *m_ip6addr, uchar_t *m_physaddr)
17454 17460 {
17455 17461 phyint_t *phyi = ill->ill_phyint;
17456 17462
17457 17463 /*
17458 17464 * Check PHYI_MULTI_BCAST and length of physical
17459 17465 * address to determine if we use the mapping or the
17460 17466 * broadcast address.
17461 17467 */
17462 17468 if ((phyi->phyint_flags & PHYI_MULTI_BCAST) != 0 ||
17463 17469 ill->ill_phys_addr_length != ETHERADDRL) {
17464 17470 ip_mbcast_mapping(ill, m_ip6addr, m_physaddr);
17465 17471 return;
17466 17472 }
17467 17473 m_physaddr[0] = 0x33;
17468 17474 m_physaddr[1] = 0x33;
17469 17475 m_physaddr[2] = m_ip6addr[12];
17470 17476 m_physaddr[3] = m_ip6addr[13];
17471 17477 m_physaddr[4] = m_ip6addr[14];
17472 17478 m_physaddr[5] = m_ip6addr[15];
17473 17479 }
17474 17480
17475 17481 /*
17476 17482 * Map the multicast ipaddr_t in m_ipaddr to the physaddr for ethernet.
17477 17483 */
17478 17484 static void
17479 17485 ip_ether_v4_mapping(ill_t *ill, uchar_t *m_ipaddr, uchar_t *m_physaddr)
17480 17486 {
17481 17487 phyint_t *phyi = ill->ill_phyint;
17482 17488
17483 17489 /*
17484 17490 * Check PHYI_MULTI_BCAST and length of physical
17485 17491 * address to determine if we use the mapping or the
17486 17492 * broadcast address.
17487 17493 */
17488 17494 if ((phyi->phyint_flags & PHYI_MULTI_BCAST) != 0 ||
17489 17495 ill->ill_phys_addr_length != ETHERADDRL) {
17490 17496 ip_mbcast_mapping(ill, m_ipaddr, m_physaddr);
17491 17497 return;
17492 17498 }
17493 17499 m_physaddr[0] = 0x01;
17494 17500 m_physaddr[1] = 0x00;
17495 17501 m_physaddr[2] = 0x5e;
17496 17502 m_physaddr[3] = m_ipaddr[1] & 0x7f;
17497 17503 m_physaddr[4] = m_ipaddr[2];
17498 17504 m_physaddr[5] = m_ipaddr[3];
17499 17505 }
17500 17506
17501 17507 /* ARGSUSED */
17502 17508 static void
17503 17509 ip_mbcast_mapping(ill_t *ill, uchar_t *m_ipaddr, uchar_t *m_physaddr)
17504 17510 {
17505 17511 /*
17506 17512 * for the MULTI_BCAST case and other cases when we want to
17507 17513 * use the link-layer broadcast address for multicast.
17508 17514 */
17509 17515 uint8_t *bphys_addr;
17510 17516 dl_unitdata_req_t *dlur;
17511 17517
17512 17518 dlur = (dl_unitdata_req_t *)ill->ill_bcast_mp->b_rptr;
17513 17519 if (ill->ill_sap_length < 0) {
17514 17520 bphys_addr = (uchar_t *)dlur +
17515 17521 dlur->dl_dest_addr_offset;
17516 17522 } else {
17517 17523 bphys_addr = (uchar_t *)dlur +
17518 17524 dlur->dl_dest_addr_offset + ill->ill_sap_length;
17519 17525 }
17520 17526
17521 17527 bcopy(bphys_addr, m_physaddr, ill->ill_phys_addr_length);
17522 17528 }
17523 17529
17524 17530 /*
17525 17531 * Derive IPoIB interface id from the link layer address.
17526 17532 */
17527 17533 static void
17528 17534 ip_ib_v6intfid(ill_t *ill, in6_addr_t *v6addr)
17529 17535 {
17530 17536 char *addr;
17531 17537
17532 17538 ASSERT(ill->ill_phys_addr_length == 20);
17533 17539 addr = (char *)&v6addr->s6_addr32[2];
17534 17540 bcopy(ill->ill_phys_addr + 12, addr, 8);
17535 17541 /*
17536 17542 * In IBA 1.1 timeframe, some vendors erroneously set the u/l bit
17537 17543 * in the globally assigned EUI-64 GUID to 1, in violation of IEEE
17538 17544 * rules. In these cases, the IBA considers these GUIDs to be in
17539 17545 * "Modified EUI-64" format, and thus toggling the u/l bit is not
17540 17546 * required; vendors are required not to assign global EUI-64's
17541 17547 * that differ only in u/l bit values, thus guaranteeing uniqueness
17542 17548 * of the interface identifier. Whether the GUID is in modified
17543 17549 * or proper EUI-64 format, the ipv6 identifier must have the u/l
17544 17550 * bit set to 1.
17545 17551 */
17546 17552 addr[0] |= 2; /* Set Universal/Local bit to 1 */
17547 17553 }
17548 17554
17549 17555 /*
17550 17556 * Map the multicast ipaddr_t in m_ipaddr to the physaddr for InfiniBand.
17551 17557 * Note on mapping from multicast IP addresses to IPoIB multicast link
17552 17558 * addresses. IPoIB multicast link addresses are based on IBA link addresses.
17553 17559 * The format of an IPoIB multicast address is:
17554 17560 *
17555 17561 * 4 byte QPN Scope Sign. Pkey
17556 17562 * +--------------------------------------------+
17557 17563 * | 00FFFFFF | FF | 1X | X01B | Pkey | GroupID |
17558 17564 * +--------------------------------------------+
17559 17565 *
17560 17566 * The Scope and Pkey components are properties of the IBA port and
17561 17567 * network interface. They can be ascertained from the broadcast address.
17562 17568 * The Sign. part is the signature, and is 401B for IPv4 and 601B for IPv6.
17563 17569 */
17564 17570 static void
17565 17571 ip_ib_v4_mapping(ill_t *ill, uchar_t *m_ipaddr, uchar_t *m_physaddr)
17566 17572 {
17567 17573 static uint8_t ipv4_g_phys_ibmulti_addr[] = { 0x00, 0xff, 0xff, 0xff,
17568 17574 0xff, 0x10, 0x40, 0x1b, 0x00, 0x00, 0x00, 0x00,
17569 17575 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00 };
17570 17576 uint8_t *bphys_addr;
17571 17577 dl_unitdata_req_t *dlur;
17572 17578
17573 17579 bcopy(ipv4_g_phys_ibmulti_addr, m_physaddr, ill->ill_phys_addr_length);
17574 17580
17575 17581 /*
17576 17582 * RFC 4391: IPv4 MGID is 28-bit long.
17577 17583 */
17578 17584 m_physaddr[16] = m_ipaddr[0] & 0x0f;
17579 17585 m_physaddr[17] = m_ipaddr[1];
17580 17586 m_physaddr[18] = m_ipaddr[2];
17581 17587 m_physaddr[19] = m_ipaddr[3];
17582 17588
17583 17589
17584 17590 dlur = (dl_unitdata_req_t *)ill->ill_bcast_mp->b_rptr;
17585 17591 if (ill->ill_sap_length < 0) {
17586 17592 bphys_addr = (uchar_t *)dlur + dlur->dl_dest_addr_offset;
17587 17593 } else {
17588 17594 bphys_addr = (uchar_t *)dlur + dlur->dl_dest_addr_offset +
17589 17595 ill->ill_sap_length;
17590 17596 }
17591 17597 /*
17592 17598 * Now fill in the IBA scope/Pkey values from the broadcast address.
17593 17599 */
17594 17600 m_physaddr[5] = bphys_addr[5];
17595 17601 m_physaddr[8] = bphys_addr[8];
17596 17602 m_physaddr[9] = bphys_addr[9];
17597 17603 }
17598 17604
17599 17605 static void
17600 17606 ip_ib_v6_mapping(ill_t *ill, uchar_t *m_ipaddr, uchar_t *m_physaddr)
17601 17607 {
17602 17608 static uint8_t ipv4_g_phys_ibmulti_addr[] = { 0x00, 0xff, 0xff, 0xff,
17603 17609 0xff, 0x10, 0x60, 0x1b, 0x00, 0x00, 0x00, 0x00,
17604 17610 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00 };
17605 17611 uint8_t *bphys_addr;
17606 17612 dl_unitdata_req_t *dlur;
17607 17613
17608 17614 bcopy(ipv4_g_phys_ibmulti_addr, m_physaddr, ill->ill_phys_addr_length);
17609 17615
17610 17616 /*
17611 17617 * RFC 4391: IPv4 MGID is 80-bit long.
17612 17618 */
17613 17619 bcopy(&m_ipaddr[6], &m_physaddr[10], 10);
17614 17620
17615 17621 dlur = (dl_unitdata_req_t *)ill->ill_bcast_mp->b_rptr;
17616 17622 if (ill->ill_sap_length < 0) {
17617 17623 bphys_addr = (uchar_t *)dlur + dlur->dl_dest_addr_offset;
17618 17624 } else {
17619 17625 bphys_addr = (uchar_t *)dlur + dlur->dl_dest_addr_offset +
17620 17626 ill->ill_sap_length;
17621 17627 }
17622 17628 /*
17623 17629 * Now fill in the IBA scope/Pkey values from the broadcast address.
17624 17630 */
17625 17631 m_physaddr[5] = bphys_addr[5];
17626 17632 m_physaddr[8] = bphys_addr[8];
17627 17633 m_physaddr[9] = bphys_addr[9];
17628 17634 }
17629 17635
17630 17636 /*
17631 17637 * Derive IPv6 interface id from an IPv4 link-layer address (e.g. from an IPv4
17632 17638 * tunnel). The IPv4 address simply get placed in the lower 4 bytes of the
17633 17639 * IPv6 interface id. This is a suggested mechanism described in section 3.7
17634 17640 * of RFC4213.
17635 17641 */
17636 17642 static void
17637 17643 ip_ipv4_genv6intfid(ill_t *ill, uint8_t *physaddr, in6_addr_t *v6addr)
17638 17644 {
17639 17645 ASSERT(ill->ill_phys_addr_length == sizeof (ipaddr_t));
17640 17646 v6addr->s6_addr32[2] = 0;
17641 17647 bcopy(physaddr, &v6addr->s6_addr32[3], sizeof (ipaddr_t));
17642 17648 }
17643 17649
17644 17650 /*
17645 17651 * Derive IPv6 interface id from an IPv6 link-layer address (e.g. from an IPv6
17646 17652 * tunnel). The lower 8 bytes of the IPv6 address simply become the interface
17647 17653 * id.
17648 17654 */
17649 17655 static void
17650 17656 ip_ipv6_genv6intfid(ill_t *ill, uint8_t *physaddr, in6_addr_t *v6addr)
17651 17657 {
17652 17658 in6_addr_t *v6lladdr = (in6_addr_t *)physaddr;
17653 17659
17654 17660 ASSERT(ill->ill_phys_addr_length == sizeof (in6_addr_t));
17655 17661 bcopy(&v6lladdr->s6_addr32[2], &v6addr->s6_addr32[2], 8);
17656 17662 }
17657 17663
17658 17664 static void
17659 17665 ip_ipv6_v6intfid(ill_t *ill, in6_addr_t *v6addr)
17660 17666 {
17661 17667 ip_ipv6_genv6intfid(ill, ill->ill_phys_addr, v6addr);
17662 17668 }
17663 17669
17664 17670 static void
17665 17671 ip_ipv6_v6destintfid(ill_t *ill, in6_addr_t *v6addr)
17666 17672 {
17667 17673 ip_ipv6_genv6intfid(ill, ill->ill_dest_addr, v6addr);
17668 17674 }
17669 17675
17670 17676 static void
17671 17677 ip_ipv4_v6intfid(ill_t *ill, in6_addr_t *v6addr)
17672 17678 {
17673 17679 ip_ipv4_genv6intfid(ill, ill->ill_phys_addr, v6addr);
17674 17680 }
17675 17681
17676 17682 static void
17677 17683 ip_ipv4_v6destintfid(ill_t *ill, in6_addr_t *v6addr)
17678 17684 {
17679 17685 ip_ipv4_genv6intfid(ill, ill->ill_dest_addr, v6addr);
17680 17686 }
17681 17687
17682 17688 /*
17683 17689 * Lookup an ill and verify that the zoneid has an ipif on that ill.
17684 17690 * Returns an held ill, or NULL.
17685 17691 */
17686 17692 ill_t *
17687 17693 ill_lookup_on_ifindex_zoneid(uint_t index, zoneid_t zoneid, boolean_t isv6,
17688 17694 ip_stack_t *ipst)
17689 17695 {
17690 17696 ill_t *ill;
17691 17697 ipif_t *ipif;
17692 17698
17693 17699 ill = ill_lookup_on_ifindex(index, isv6, ipst);
17694 17700 if (ill == NULL)
17695 17701 return (NULL);
17696 17702
17697 17703 mutex_enter(&ill->ill_lock);
17698 17704 for (ipif = ill->ill_ipif; ipif != NULL; ipif = ipif->ipif_next) {
17699 17705 if (IPIF_IS_CONDEMNED(ipif))
17700 17706 continue;
17701 17707 if (zoneid != ALL_ZONES && ipif->ipif_zoneid != zoneid &&
17702 17708 ipif->ipif_zoneid != ALL_ZONES)
17703 17709 continue;
17704 17710
17705 17711 mutex_exit(&ill->ill_lock);
17706 17712 return (ill);
17707 17713 }
17708 17714 mutex_exit(&ill->ill_lock);
17709 17715 ill_refrele(ill);
17710 17716 return (NULL);
17711 17717 }
17712 17718
17713 17719 /*
17714 17720 * Return a pointer to an ipif_t given a combination of (ill_idx,ipif_id)
17715 17721 * If a pointer to an ipif_t is returned then the caller will need to do
17716 17722 * an ill_refrele().
17717 17723 */
17718 17724 ipif_t *
17719 17725 ipif_getby_indexes(uint_t ifindex, uint_t lifidx, boolean_t isv6,
17720 17726 ip_stack_t *ipst)
17721 17727 {
17722 17728 ipif_t *ipif;
17723 17729 ill_t *ill;
17724 17730
17725 17731 ill = ill_lookup_on_ifindex(ifindex, isv6, ipst);
17726 17732 if (ill == NULL)
17727 17733 return (NULL);
17728 17734
17729 17735 mutex_enter(&ill->ill_lock);
17730 17736 if (ill->ill_state_flags & ILL_CONDEMNED) {
17731 17737 mutex_exit(&ill->ill_lock);
17732 17738 ill_refrele(ill);
17733 17739 return (NULL);
17734 17740 }
17735 17741
17736 17742 for (ipif = ill->ill_ipif; ipif != NULL; ipif = ipif->ipif_next) {
17737 17743 if (!IPIF_CAN_LOOKUP(ipif))
17738 17744 continue;
17739 17745 if (lifidx == ipif->ipif_id) {
17740 17746 ipif_refhold_locked(ipif);
17741 17747 break;
17742 17748 }
17743 17749 }
17744 17750
17745 17751 mutex_exit(&ill->ill_lock);
17746 17752 ill_refrele(ill);
17747 17753 return (ipif);
17748 17754 }
17749 17755
17750 17756 /*
17751 17757 * Set ill_inputfn based on the current know state.
17752 17758 * This needs to be called when any of the factors taken into
17753 17759 * account changes.
17754 17760 */
17755 17761 void
17756 17762 ill_set_inputfn(ill_t *ill)
17757 17763 {
17758 17764 ip_stack_t *ipst = ill->ill_ipst;
17759 17765
17760 17766 if (ill->ill_isv6) {
17761 17767 if (is_system_labeled())
17762 17768 ill->ill_inputfn = ill_input_full_v6;
17763 17769 else
17764 17770 ill->ill_inputfn = ill_input_short_v6;
17765 17771 } else {
17766 17772 if (is_system_labeled())
17767 17773 ill->ill_inputfn = ill_input_full_v4;
17768 17774 else if (ill->ill_dhcpinit != 0)
17769 17775 ill->ill_inputfn = ill_input_full_v4;
17770 17776 else if (ipst->ips_ipcl_proto_fanout_v4[IPPROTO_RSVP].connf_head
17771 17777 != NULL)
17772 17778 ill->ill_inputfn = ill_input_full_v4;
17773 17779 else if (ipst->ips_ip_cgtp_filter &&
17774 17780 ipst->ips_ip_cgtp_filter_ops != NULL)
17775 17781 ill->ill_inputfn = ill_input_full_v4;
17776 17782 else
17777 17783 ill->ill_inputfn = ill_input_short_v4;
17778 17784 }
17779 17785 }
17780 17786
17781 17787 /*
17782 17788 * Re-evaluate ill_inputfn for all the IPv4 ills.
17783 17789 * Used when RSVP and CGTP comes and goes.
17784 17790 */
17785 17791 void
17786 17792 ill_set_inputfn_all(ip_stack_t *ipst)
17787 17793 {
17788 17794 ill_walk_context_t ctx;
17789 17795 ill_t *ill;
17790 17796
17791 17797 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
17792 17798 ill = ILL_START_WALK_V4(&ctx, ipst);
17793 17799 for (; ill != NULL; ill = ill_next(&ctx, ill))
17794 17800 ill_set_inputfn(ill);
17795 17801
17796 17802 rw_exit(&ipst->ips_ill_g_lock);
17797 17803 }
17798 17804
17799 17805 /*
17800 17806 * Set the physical address information for `ill' to the contents of the
17801 17807 * dl_notify_ind_t pointed to by `mp'. Must be called as writer, and will be
17802 17808 * asynchronous if `ill' cannot immediately be quiesced -- in which case
17803 17809 * EINPROGRESS will be returned.
17804 17810 */
17805 17811 int
17806 17812 ill_set_phys_addr(ill_t *ill, mblk_t *mp)
17807 17813 {
17808 17814 ipsq_t *ipsq = ill->ill_phyint->phyint_ipsq;
17809 17815 dl_notify_ind_t *dlindp = (dl_notify_ind_t *)mp->b_rptr;
17810 17816
17811 17817 ASSERT(IAM_WRITER_IPSQ(ipsq));
17812 17818
17813 17819 if (dlindp->dl_data != DL_IPV6_LINK_LAYER_ADDR &&
17814 17820 dlindp->dl_data != DL_CURR_DEST_ADDR &&
17815 17821 dlindp->dl_data != DL_CURR_PHYS_ADDR) {
17816 17822 /* Changing DL_IPV6_TOKEN is not yet supported */
17817 17823 return (0);
17818 17824 }
17819 17825
17820 17826 /*
17821 17827 * We need to store up to two copies of `mp' in `ill'. Due to the
17822 17828 * design of ipsq_pending_mp_add(), we can't pass them as separate
17823 17829 * arguments to ill_set_phys_addr_tail(). Instead, chain them
17824 17830 * together here, then pull 'em apart in ill_set_phys_addr_tail().
17825 17831 */
17826 17832 if ((mp = copyb(mp)) == NULL || (mp->b_cont = copyb(mp)) == NULL) {
17827 17833 freemsg(mp);
17828 17834 return (ENOMEM);
17829 17835 }
17830 17836
17831 17837 ipsq_current_start(ipsq, ill->ill_ipif, 0);
17832 17838
17833 17839 /*
17834 17840 * Since we'll only do a logical down, we can't rely on ipif_down
17835 17841 * to turn on ILL_DOWN_IN_PROGRESS, or for the DL_BIND_ACK to reset
17836 17842 * ILL_DOWN_IN_PROGRESS. We instead manage this separately for this
17837 17843 * case, to quiesce ire's and nce's for ill_is_quiescent.
17838 17844 */
17839 17845 mutex_enter(&ill->ill_lock);
17840 17846 ill->ill_state_flags |= ILL_DOWN_IN_PROGRESS;
17841 17847 /* no more ire/nce addition allowed */
17842 17848 mutex_exit(&ill->ill_lock);
17843 17849
17844 17850 /*
17845 17851 * If we can quiesce the ill, then set the address. If not, then
17846 17852 * ill_set_phys_addr_tail() will be called from ipif_ill_refrele_tail().
17847 17853 */
17848 17854 ill_down_ipifs(ill, B_TRUE);
17849 17855 mutex_enter(&ill->ill_lock);
17850 17856 if (!ill_is_quiescent(ill)) {
17851 17857 /* call cannot fail since `conn_t *' argument is NULL */
17852 17858 (void) ipsq_pending_mp_add(NULL, ill->ill_ipif, ill->ill_rq,
17853 17859 mp, ILL_DOWN);
17854 17860 mutex_exit(&ill->ill_lock);
17855 17861 return (EINPROGRESS);
17856 17862 }
17857 17863 mutex_exit(&ill->ill_lock);
17858 17864
17859 17865 ill_set_phys_addr_tail(ipsq, ill->ill_rq, mp, NULL);
17860 17866 return (0);
17861 17867 }
17862 17868
17863 17869 /*
17864 17870 * When the allowed-ips link property is set on the datalink, IP receives a
17865 17871 * DL_NOTE_ALLOWED_IPS notification that is processed in ill_set_allowed_ips()
17866 17872 * to initialize the ill_allowed_ips[] array in the ill_t. This array is then
17867 17873 * used to vet addresses passed to ip_sioctl_addr() and to ensure that the
17868 17874 * only IP addresses configured on the ill_t are those in the ill_allowed_ips[]
17869 17875 * array.
17870 17876 */
17871 17877 void
17872 17878 ill_set_allowed_ips(ill_t *ill, mblk_t *mp)
17873 17879 {
17874 17880 ipsq_t *ipsq = ill->ill_phyint->phyint_ipsq;
17875 17881 dl_notify_ind_t *dlip = (dl_notify_ind_t *)mp->b_rptr;
17876 17882 mac_protect_t *mrp;
17877 17883 int i;
17878 17884
17879 17885 ASSERT(IAM_WRITER_IPSQ(ipsq));
17880 17886 mrp = (mac_protect_t *)&dlip[1];
17881 17887
17882 17888 if (mrp->mp_ipaddrcnt == 0) { /* reset allowed-ips */
17883 17889 kmem_free(ill->ill_allowed_ips,
17884 17890 ill->ill_allowed_ips_cnt * sizeof (in6_addr_t));
17885 17891 ill->ill_allowed_ips_cnt = 0;
17886 17892 ill->ill_allowed_ips = NULL;
17887 17893 mutex_enter(&ill->ill_phyint->phyint_lock);
17888 17894 ill->ill_phyint->phyint_flags &= ~PHYI_L3PROTECT;
17889 17895 mutex_exit(&ill->ill_phyint->phyint_lock);
17890 17896 return;
17891 17897 }
17892 17898
17893 17899 if (ill->ill_allowed_ips != NULL) {
17894 17900 kmem_free(ill->ill_allowed_ips,
17895 17901 ill->ill_allowed_ips_cnt * sizeof (in6_addr_t));
17896 17902 }
17897 17903 ill->ill_allowed_ips_cnt = mrp->mp_ipaddrcnt;
17898 17904 ill->ill_allowed_ips = kmem_alloc(
17899 17905 ill->ill_allowed_ips_cnt * sizeof (in6_addr_t), KM_SLEEP);
17900 17906 for (i = 0; i < mrp->mp_ipaddrcnt; i++)
17901 17907 ill->ill_allowed_ips[i] = mrp->mp_ipaddrs[i].ip_addr;
17902 17908
17903 17909 mutex_enter(&ill->ill_phyint->phyint_lock);
17904 17910 ill->ill_phyint->phyint_flags |= PHYI_L3PROTECT;
17905 17911 mutex_exit(&ill->ill_phyint->phyint_lock);
17906 17912 }
17907 17913
17908 17914 /*
17909 17915 * Once the ill associated with `q' has quiesced, set its physical address
17910 17916 * information to the values in `addrmp'. Note that two copies of `addrmp'
17911 17917 * are passed (linked by b_cont), since we sometimes need to save two distinct
17912 17918 * copies in the ill_t, and our context doesn't permit sleeping or allocation
17913 17919 * failure (we'll free the other copy if it's not needed). Since the ill_t
17914 17920 * is quiesced, we know any stale nce's with the old address information have
17915 17921 * already been removed, so we don't need to call nce_flush().
17916 17922 */
17917 17923 /* ARGSUSED */
17918 17924 static void
17919 17925 ill_set_phys_addr_tail(ipsq_t *ipsq, queue_t *q, mblk_t *addrmp, void *dummy)
17920 17926 {
17921 17927 ill_t *ill = q->q_ptr;
17922 17928 mblk_t *addrmp2 = unlinkb(addrmp);
17923 17929 dl_notify_ind_t *dlindp = (dl_notify_ind_t *)addrmp->b_rptr;
17924 17930 uint_t addrlen, addroff;
17925 17931 int status;
17926 17932
17927 17933 ASSERT(IAM_WRITER_IPSQ(ipsq));
17928 17934
17929 17935 addroff = dlindp->dl_addr_offset;
17930 17936 addrlen = dlindp->dl_addr_length - ABS(ill->ill_sap_length);
17931 17937
17932 17938 switch (dlindp->dl_data) {
17933 17939 case DL_IPV6_LINK_LAYER_ADDR:
17934 17940 ill_set_ndmp(ill, addrmp, addroff, addrlen);
17935 17941 freemsg(addrmp2);
17936 17942 break;
17937 17943
17938 17944 case DL_CURR_DEST_ADDR:
17939 17945 freemsg(ill->ill_dest_addr_mp);
17940 17946 ill->ill_dest_addr = addrmp->b_rptr + addroff;
17941 17947 ill->ill_dest_addr_mp = addrmp;
17942 17948 if (ill->ill_isv6) {
17943 17949 ill_setdesttoken(ill);
17944 17950 ipif_setdestlinklocal(ill->ill_ipif);
17945 17951 }
17946 17952 freemsg(addrmp2);
17947 17953 break;
17948 17954
17949 17955 case DL_CURR_PHYS_ADDR:
17950 17956 freemsg(ill->ill_phys_addr_mp);
17951 17957 ill->ill_phys_addr = addrmp->b_rptr + addroff;
17952 17958 ill->ill_phys_addr_mp = addrmp;
17953 17959 ill->ill_phys_addr_length = addrlen;
17954 17960 if (ill->ill_isv6)
17955 17961 ill_set_ndmp(ill, addrmp2, addroff, addrlen);
17956 17962 else
17957 17963 freemsg(addrmp2);
17958 17964 if (ill->ill_isv6) {
17959 17965 ill_setdefaulttoken(ill);
17960 17966 ipif_setlinklocal(ill->ill_ipif);
17961 17967 }
17962 17968 break;
17963 17969 default:
17964 17970 ASSERT(0);
17965 17971 }
17966 17972
17967 17973 /*
17968 17974 * reset ILL_DOWN_IN_PROGRESS so that we can successfully add ires
17969 17975 * as we bring the ipifs up again.
17970 17976 */
17971 17977 mutex_enter(&ill->ill_lock);
17972 17978 ill->ill_state_flags &= ~ILL_DOWN_IN_PROGRESS;
17973 17979 mutex_exit(&ill->ill_lock);
17974 17980 /*
17975 17981 * If there are ipifs to bring up, ill_up_ipifs() will return
17976 17982 * EINPROGRESS, and ipsq_current_finish() will be called by
17977 17983 * ip_rput_dlpi_writer() or arp_bringup_done() when the last ipif is
17978 17984 * brought up.
17979 17985 */
17980 17986 status = ill_up_ipifs(ill, q, addrmp);
17981 17987 if (status != EINPROGRESS)
17982 17988 ipsq_current_finish(ipsq);
17983 17989 }
17984 17990
17985 17991 /*
17986 17992 * Helper routine for setting the ill_nd_lla fields.
17987 17993 */
17988 17994 void
17989 17995 ill_set_ndmp(ill_t *ill, mblk_t *ndmp, uint_t addroff, uint_t addrlen)
17990 17996 {
17991 17997 freemsg(ill->ill_nd_lla_mp);
17992 17998 ill->ill_nd_lla = ndmp->b_rptr + addroff;
17993 17999 ill->ill_nd_lla_mp = ndmp;
17994 18000 ill->ill_nd_lla_len = addrlen;
17995 18001 }
17996 18002
17997 18003 /*
17998 18004 * Replumb the ill.
17999 18005 */
18000 18006 int
18001 18007 ill_replumb(ill_t *ill, mblk_t *mp)
18002 18008 {
18003 18009 ipsq_t *ipsq = ill->ill_phyint->phyint_ipsq;
18004 18010
18005 18011 ASSERT(IAM_WRITER_IPSQ(ipsq));
18006 18012
18007 18013 ipsq_current_start(ipsq, ill->ill_ipif, 0);
18008 18014
18009 18015 /*
18010 18016 * If we can quiesce the ill, then continue. If not, then
18011 18017 * ill_replumb_tail() will be called from ipif_ill_refrele_tail().
18012 18018 */
18013 18019 ill_down_ipifs(ill, B_FALSE);
18014 18020
18015 18021 mutex_enter(&ill->ill_lock);
18016 18022 if (!ill_is_quiescent(ill)) {
18017 18023 /* call cannot fail since `conn_t *' argument is NULL */
18018 18024 (void) ipsq_pending_mp_add(NULL, ill->ill_ipif, ill->ill_rq,
18019 18025 mp, ILL_DOWN);
18020 18026 mutex_exit(&ill->ill_lock);
18021 18027 return (EINPROGRESS);
18022 18028 }
18023 18029 mutex_exit(&ill->ill_lock);
18024 18030
18025 18031 ill_replumb_tail(ipsq, ill->ill_rq, mp, NULL);
18026 18032 return (0);
18027 18033 }
18028 18034
18029 18035 /* ARGSUSED */
18030 18036 static void
18031 18037 ill_replumb_tail(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *dummy)
18032 18038 {
18033 18039 ill_t *ill = q->q_ptr;
18034 18040 int err;
18035 18041 conn_t *connp = NULL;
18036 18042
18037 18043 ASSERT(IAM_WRITER_IPSQ(ipsq));
18038 18044 freemsg(ill->ill_replumb_mp);
18039 18045 ill->ill_replumb_mp = copyb(mp);
18040 18046
18041 18047 if (ill->ill_replumb_mp == NULL) {
18042 18048 /* out of memory */
18043 18049 ipsq_current_finish(ipsq);
18044 18050 return;
18045 18051 }
18046 18052
18047 18053 mutex_enter(&ill->ill_lock);
18048 18054 ill->ill_up_ipifs = ipsq_pending_mp_add(NULL, ill->ill_ipif,
18049 18055 ill->ill_rq, ill->ill_replumb_mp, 0);
18050 18056 mutex_exit(&ill->ill_lock);
18051 18057
18052 18058 if (!ill->ill_up_ipifs) {
18053 18059 /* already closing */
18054 18060 ipsq_current_finish(ipsq);
18055 18061 return;
18056 18062 }
18057 18063 ill->ill_replumbing = 1;
18058 18064 err = ill_down_ipifs_tail(ill);
18059 18065
18060 18066 /*
18061 18067 * Successfully quiesced and brought down the interface, now we send
18062 18068 * the DL_NOTE_REPLUMB_DONE message down to the driver. Reuse the
18063 18069 * DL_NOTE_REPLUMB message.
18064 18070 */
18065 18071 mp = mexchange(NULL, mp, sizeof (dl_notify_conf_t), M_PROTO,
18066 18072 DL_NOTIFY_CONF);
18067 18073 ASSERT(mp != NULL);
18068 18074 ((dl_notify_conf_t *)mp->b_rptr)->dl_notification =
18069 18075 DL_NOTE_REPLUMB_DONE;
18070 18076 ill_dlpi_send(ill, mp);
18071 18077
18072 18078 /*
18073 18079 * For IPv4, we would usually get EINPROGRESS because the ETHERTYPE_ARP
18074 18080 * streams have to be unbound. When all the DLPI exchanges are done,
18075 18081 * ipsq_current_finish() will be called by arp_bringup_done(). The
18076 18082 * remainder of ipif bringup via ill_up_ipifs() will also be done in
18077 18083 * arp_bringup_done().
18078 18084 */
18079 18085 ASSERT(ill->ill_replumb_mp != NULL);
18080 18086 if (err == EINPROGRESS)
18081 18087 return;
18082 18088 else
18083 18089 ill->ill_replumb_mp = ipsq_pending_mp_get(ipsq, &connp);
18084 18090 ASSERT(connp == NULL);
18085 18091 if (err == 0 && ill->ill_replumb_mp != NULL &&
18086 18092 ill_up_ipifs(ill, q, ill->ill_replumb_mp) == EINPROGRESS) {
18087 18093 return;
18088 18094 }
18089 18095 ipsq_current_finish(ipsq);
18090 18096 }
18091 18097
18092 18098 /*
18093 18099 * Issue ioctl `cmd' on `lh'; caller provides the initial payload in `buf'
18094 18100 * which is `bufsize' bytes. On success, zero is returned and `buf' updated
18095 18101 * as per the ioctl. On failure, an errno is returned.
18096 18102 */
18097 18103 static int
18098 18104 ip_ioctl(ldi_handle_t lh, int cmd, void *buf, uint_t bufsize, cred_t *cr)
18099 18105 {
18100 18106 int rval;
18101 18107 struct strioctl iocb;
18102 18108
18103 18109 iocb.ic_cmd = cmd;
18104 18110 iocb.ic_timout = 15;
18105 18111 iocb.ic_len = bufsize;
18106 18112 iocb.ic_dp = buf;
18107 18113
18108 18114 return (ldi_ioctl(lh, I_STR, (intptr_t)&iocb, FKIOCTL, cr, &rval));
18109 18115 }
18110 18116
18111 18117 /*
18112 18118 * Issue an SIOCGLIFCONF for address family `af' and store the result into a
18113 18119 * dynamically-allocated `lifcp' that will be `bufsizep' bytes on success.
18114 18120 */
18115 18121 static int
18116 18122 ip_lifconf_ioctl(ldi_handle_t lh, int af, struct lifconf *lifcp,
18117 18123 uint_t *bufsizep, cred_t *cr)
18118 18124 {
18119 18125 int err;
18120 18126 struct lifnum lifn;
18121 18127
18122 18128 bzero(&lifn, sizeof (lifn));
18123 18129 lifn.lifn_family = af;
18124 18130 lifn.lifn_flags = LIFC_UNDER_IPMP;
18125 18131
18126 18132 if ((err = ip_ioctl(lh, SIOCGLIFNUM, &lifn, sizeof (lifn), cr)) != 0)
18127 18133 return (err);
18128 18134
18129 18135 /*
18130 18136 * Pad the interface count to account for additional interfaces that
18131 18137 * may have been configured between the SIOCGLIFNUM and SIOCGLIFCONF.
18132 18138 */
18133 18139 lifn.lifn_count += 4;
18134 18140 bzero(lifcp, sizeof (*lifcp));
18135 18141 lifcp->lifc_flags = LIFC_UNDER_IPMP;
18136 18142 lifcp->lifc_family = af;
18137 18143 lifcp->lifc_len = *bufsizep = lifn.lifn_count * sizeof (struct lifreq);
18138 18144 lifcp->lifc_buf = kmem_zalloc(*bufsizep, KM_SLEEP);
18139 18145
18140 18146 err = ip_ioctl(lh, SIOCGLIFCONF, lifcp, sizeof (*lifcp), cr);
18141 18147 if (err != 0) {
18142 18148 kmem_free(lifcp->lifc_buf, *bufsizep);
18143 18149 return (err);
18144 18150 }
18145 18151
18146 18152 return (0);
18147 18153 }
18148 18154
18149 18155 /*
18150 18156 * Helper for ip_interface_cleanup() that removes the loopback interface.
18151 18157 */
18152 18158 static void
18153 18159 ip_loopback_removeif(ldi_handle_t lh, boolean_t isv6, cred_t *cr)
18154 18160 {
18155 18161 int err;
18156 18162 struct lifreq lifr;
18157 18163
18158 18164 bzero(&lifr, sizeof (lifr));
18159 18165 (void) strcpy(lifr.lifr_name, ipif_loopback_name);
18160 18166
18161 18167 /*
18162 18168 * Attempt to remove the interface. It may legitimately not exist
18163 18169 * (e.g. the zone administrator unplumbed it), so ignore ENXIO.
18164 18170 */
18165 18171 err = ip_ioctl(lh, SIOCLIFREMOVEIF, &lifr, sizeof (lifr), cr);
18166 18172 if (err != 0 && err != ENXIO) {
18167 18173 ip0dbg(("ip_loopback_removeif: IP%s SIOCLIFREMOVEIF failed: "
18168 18174 "error %d\n", isv6 ? "v6" : "v4", err));
18169 18175 }
18170 18176 }
18171 18177
18172 18178 /*
18173 18179 * Helper for ip_interface_cleanup() that ensures no IP interfaces are in IPMP
18174 18180 * groups and that IPMP data addresses are down. These conditions must be met
18175 18181 * so that IPMP interfaces can be I_PUNLINK'd, as per ip_sioctl_plink_ipmp().
18176 18182 */
18177 18183 static void
18178 18184 ip_ipmp_cleanup(ldi_handle_t lh, boolean_t isv6, cred_t *cr)
18179 18185 {
18180 18186 int af = isv6 ? AF_INET6 : AF_INET;
18181 18187 int i, nifs;
18182 18188 int err;
18183 18189 uint_t bufsize;
18184 18190 uint_t lifrsize = sizeof (struct lifreq);
18185 18191 struct lifconf lifc;
18186 18192 struct lifreq *lifrp;
18187 18193
18188 18194 if ((err = ip_lifconf_ioctl(lh, af, &lifc, &bufsize, cr)) != 0) {
18189 18195 cmn_err(CE_WARN, "ip_ipmp_cleanup: cannot get interface list "
18190 18196 "(error %d); any IPMP interfaces cannot be shutdown", err);
18191 18197 return;
18192 18198 }
18193 18199
18194 18200 nifs = lifc.lifc_len / lifrsize;
18195 18201 for (lifrp = lifc.lifc_req, i = 0; i < nifs; i++, lifrp++) {
18196 18202 err = ip_ioctl(lh, SIOCGLIFFLAGS, lifrp, lifrsize, cr);
18197 18203 if (err != 0) {
18198 18204 cmn_err(CE_WARN, "ip_ipmp_cleanup: %s: cannot get "
18199 18205 "flags: error %d", lifrp->lifr_name, err);
18200 18206 continue;
18201 18207 }
18202 18208
18203 18209 if (lifrp->lifr_flags & IFF_IPMP) {
18204 18210 if ((lifrp->lifr_flags & (IFF_UP|IFF_DUPLICATE)) == 0)
18205 18211 continue;
18206 18212
18207 18213 lifrp->lifr_flags &= ~IFF_UP;
18208 18214 err = ip_ioctl(lh, SIOCSLIFFLAGS, lifrp, lifrsize, cr);
18209 18215 if (err != 0) {
18210 18216 cmn_err(CE_WARN, "ip_ipmp_cleanup: %s: cannot "
18211 18217 "bring down (error %d); IPMP interface may "
18212 18218 "not be shutdown", lifrp->lifr_name, err);
18213 18219 }
18214 18220
18215 18221 /*
18216 18222 * Check if IFF_DUPLICATE is still set -- and if so,
18217 18223 * reset the address to clear it.
18218 18224 */
18219 18225 err = ip_ioctl(lh, SIOCGLIFFLAGS, lifrp, lifrsize, cr);
18220 18226 if (err != 0 || !(lifrp->lifr_flags & IFF_DUPLICATE))
18221 18227 continue;
18222 18228
18223 18229 err = ip_ioctl(lh, SIOCGLIFADDR, lifrp, lifrsize, cr);
18224 18230 if (err != 0 || (err = ip_ioctl(lh, SIOCGLIFADDR,
18225 18231 lifrp, lifrsize, cr)) != 0) {
18226 18232 cmn_err(CE_WARN, "ip_ipmp_cleanup: %s: cannot "
18227 18233 "reset DAD (error %d); IPMP interface may "
18228 18234 "not be shutdown", lifrp->lifr_name, err);
18229 18235 }
18230 18236 continue;
18231 18237 }
18232 18238
18233 18239 if (strchr(lifrp->lifr_name, IPIF_SEPARATOR_CHAR) == 0) {
18234 18240 lifrp->lifr_groupname[0] = '\0';
18235 18241 if ((err = ip_ioctl(lh, SIOCSLIFGROUPNAME, lifrp,
18236 18242 lifrsize, cr)) != 0) {
18237 18243 cmn_err(CE_WARN, "ip_ipmp_cleanup: %s: cannot "
18238 18244 "leave IPMP group (error %d); associated "
18239 18245 "IPMP interface may not be shutdown",
18240 18246 lifrp->lifr_name, err);
18241 18247 continue;
18242 18248 }
18243 18249 }
18244 18250 }
18245 18251
18246 18252 kmem_free(lifc.lifc_buf, bufsize);
18247 18253 }
18248 18254
18249 18255 #define UDPDEV "/devices/pseudo/udp@0:udp"
18250 18256 #define UDP6DEV "/devices/pseudo/udp6@0:udp6"
18251 18257
18252 18258 /*
18253 18259 * Remove the loopback interfaces and prep the IPMP interfaces to be torn down.
18254 18260 * Non-loopback interfaces are either I_LINK'd or I_PLINK'd; the former go away
18255 18261 * when the user-level processes in the zone are killed and the latter are
18256 18262 * cleaned up by str_stack_shutdown().
18257 18263 */
18258 18264 void
18259 18265 ip_interface_cleanup(ip_stack_t *ipst)
18260 18266 {
18261 18267 ldi_handle_t lh;
18262 18268 ldi_ident_t li;
18263 18269 cred_t *cr;
18264 18270 int err;
18265 18271 int i;
18266 18272 char *devs[] = { UDP6DEV, UDPDEV };
18267 18273 netstackid_t stackid = ipst->ips_netstack->netstack_stackid;
18268 18274
18269 18275 if ((err = ldi_ident_from_major(ddi_name_to_major("ip"), &li)) != 0) {
18270 18276 cmn_err(CE_WARN, "ip_interface_cleanup: cannot get ldi ident:"
18271 18277 " error %d", err);
18272 18278 return;
18273 18279 }
18274 18280
18275 18281 cr = zone_get_kcred(netstackid_to_zoneid(stackid));
18276 18282 ASSERT(cr != NULL);
18277 18283
18278 18284 /*
18279 18285 * NOTE: loop executes exactly twice and is hardcoded to know that the
18280 18286 * first iteration is IPv6. (Unrolling yields repetitious code, hence
18281 18287 * the loop.)
18282 18288 */
18283 18289 for (i = 0; i < 2; i++) {
18284 18290 err = ldi_open_by_name(devs[i], FREAD|FWRITE, cr, &lh, li);
18285 18291 if (err != 0) {
18286 18292 cmn_err(CE_WARN, "ip_interface_cleanup: cannot open %s:"
18287 18293 " error %d", devs[i], err);
18288 18294 continue;
18289 18295 }
18290 18296
18291 18297 ip_loopback_removeif(lh, i == 0, cr);
18292 18298 ip_ipmp_cleanup(lh, i == 0, cr);
18293 18299
18294 18300 (void) ldi_close(lh, FREAD|FWRITE, cr);
18295 18301 }
18296 18302
18297 18303 ldi_ident_release(li);
18298 18304 crfree(cr);
18299 18305 }
18300 18306
18301 18307 /*
18302 18308 * This needs to be in-sync with nic_event_t definition
18303 18309 */
18304 18310 static const char *
18305 18311 ill_hook_event2str(nic_event_t event)
18306 18312 {
18307 18313 switch (event) {
18308 18314 case NE_PLUMB:
18309 18315 return ("PLUMB");
18310 18316 case NE_UNPLUMB:
18311 18317 return ("UNPLUMB");
18312 18318 case NE_UP:
18313 18319 return ("UP");
18314 18320 case NE_DOWN:
18315 18321 return ("DOWN");
18316 18322 case NE_ADDRESS_CHANGE:
18317 18323 return ("ADDRESS_CHANGE");
18318 18324 case NE_LIF_UP:
18319 18325 return ("LIF_UP");
18320 18326 case NE_LIF_DOWN:
18321 18327 return ("LIF_DOWN");
18322 18328 case NE_IFINDEX_CHANGE:
18323 18329 return ("IFINDEX_CHANGE");
18324 18330 default:
18325 18331 return ("UNKNOWN");
18326 18332 }
18327 18333 }
18328 18334
18329 18335 void
18330 18336 ill_nic_event_dispatch(ill_t *ill, lif_if_t lif, nic_event_t event,
18331 18337 nic_event_data_t data, size_t datalen)
18332 18338 {
18333 18339 ip_stack_t *ipst = ill->ill_ipst;
18334 18340 hook_nic_event_int_t *info;
18335 18341 const char *str = NULL;
18336 18342
18337 18343 /* create a new nic event info */
18338 18344 if ((info = kmem_alloc(sizeof (*info), KM_NOSLEEP)) == NULL)
18339 18345 goto fail;
18340 18346
18341 18347 info->hnei_event.hne_nic = ill->ill_phyint->phyint_ifindex;
18342 18348 info->hnei_event.hne_lif = lif;
18343 18349 info->hnei_event.hne_event = event;
18344 18350 info->hnei_event.hne_protocol = ill->ill_isv6 ?
18345 18351 ipst->ips_ipv6_net_data : ipst->ips_ipv4_net_data;
18346 18352 info->hnei_event.hne_data = NULL;
18347 18353 info->hnei_event.hne_datalen = 0;
18348 18354 info->hnei_stackid = ipst->ips_netstack->netstack_stackid;
18349 18355
18350 18356 if (data != NULL && datalen != 0) {
18351 18357 info->hnei_event.hne_data = kmem_alloc(datalen, KM_NOSLEEP);
18352 18358 if (info->hnei_event.hne_data == NULL)
18353 18359 goto fail;
18354 18360 bcopy(data, info->hnei_event.hne_data, datalen);
18355 18361 info->hnei_event.hne_datalen = datalen;
18356 18362 }
18357 18363
18358 18364 if (ddi_taskq_dispatch(eventq_queue_nic, ip_ne_queue_func, info,
18359 18365 DDI_NOSLEEP) == DDI_SUCCESS)
18360 18366 return;
18361 18367
18362 18368 fail:
18363 18369 if (info != NULL) {
18364 18370 if (info->hnei_event.hne_data != NULL) {
18365 18371 kmem_free(info->hnei_event.hne_data,
18366 18372 info->hnei_event.hne_datalen);
18367 18373 }
18368 18374 kmem_free(info, sizeof (hook_nic_event_t));
18369 18375 }
18370 18376 str = ill_hook_event2str(event);
18371 18377 ip2dbg(("ill_nic_event_dispatch: could not dispatch %s nic event "
18372 18378 "information for %s (ENOMEM)\n", str, ill->ill_name));
18373 18379 }
18374 18380
18375 18381 static int
18376 18382 ipif_arp_up_done_tail(ipif_t *ipif, enum ip_resolver_action res_act)
18377 18383 {
18378 18384 int err = 0;
18379 18385 const in_addr_t *addr = NULL;
18380 18386 nce_t *nce = NULL;
18381 18387 ill_t *ill = ipif->ipif_ill;
18382 18388 ill_t *bound_ill;
18383 18389 boolean_t added_ipif = B_FALSE;
18384 18390 uint16_t state;
18385 18391 uint16_t flags;
18386 18392
18387 18393 DTRACE_PROBE3(ipif__downup, char *, "ipif_arp_up_done_tail",
18388 18394 ill_t *, ill, ipif_t *, ipif);
18389 18395 if (ipif->ipif_lcl_addr != INADDR_ANY) {
18390 18396 addr = &ipif->ipif_lcl_addr;
18391 18397 }
18392 18398
18393 18399 if ((ipif->ipif_flags & IPIF_UNNUMBERED) || addr == NULL) {
18394 18400 if (res_act != Res_act_initial)
18395 18401 return (EINVAL);
18396 18402 }
18397 18403
18398 18404 if (addr != NULL) {
18399 18405 ipmp_illgrp_t *illg = ill->ill_grp;
18400 18406
18401 18407 /* add unicast nce for the local addr */
18402 18408
18403 18409 if (IS_IPMP(ill)) {
18404 18410 /*
18405 18411 * If we're here via ipif_up(), then the ipif
18406 18412 * won't be bound yet -- add it to the group,
18407 18413 * which will bind it if possible. (We would
18408 18414 * add it in ipif_up(), but deleting on failure
18409 18415 * there is gruesome.) If we're here via
18410 18416 * ipmp_ill_bind_ipif(), then the ipif has
18411 18417 * already been added to the group and we
18412 18418 * just need to use the binding.
18413 18419 */
18414 18420 if ((bound_ill = ipmp_ipif_bound_ill(ipif)) == NULL) {
18415 18421 bound_ill = ipmp_illgrp_add_ipif(illg, ipif);
18416 18422 if (bound_ill == NULL) {
18417 18423 /*
18418 18424 * We couldn't bind the ipif to an ill
18419 18425 * yet, so we have nothing to publish.
18420 18426 * Mark the address as ready and return.
18421 18427 */
18422 18428 ipif->ipif_addr_ready = 1;
18423 18429 return (0);
18424 18430 }
18425 18431 added_ipif = B_TRUE;
18426 18432 }
18427 18433 } else {
18428 18434 bound_ill = ill;
18429 18435 }
18430 18436
18431 18437 flags = (NCE_F_MYADDR | NCE_F_PUBLISH | NCE_F_AUTHORITY |
18432 18438 NCE_F_NONUD);
18433 18439 /*
18434 18440 * If this is an initial bring-up (or the ipif was never
18435 18441 * completely brought up), do DAD. Otherwise, we're here
18436 18442 * because IPMP has rebound an address to this ill: send
18437 18443 * unsolicited advertisements (ARP announcements) to
18438 18444 * inform others.
18439 18445 */
18440 18446 if (res_act == Res_act_initial || !ipif->ipif_addr_ready) {
18441 18447 state = ND_UNCHANGED; /* compute in nce_add_common() */
18442 18448 } else {
18443 18449 state = ND_REACHABLE;
18444 18450 flags |= NCE_F_UNSOL_ADV;
18445 18451 }
18446 18452
18447 18453 retry:
18448 18454 err = nce_lookup_then_add_v4(ill,
18449 18455 bound_ill->ill_phys_addr, bound_ill->ill_phys_addr_length,
18450 18456 addr, flags, state, &nce);
18451 18457
18452 18458 /*
18453 18459 * note that we may encounter EEXIST if we are moving
18454 18460 * the nce as a result of a rebind operation.
18455 18461 */
18456 18462 switch (err) {
18457 18463 case 0:
18458 18464 ipif->ipif_added_nce = 1;
18459 18465 nce->nce_ipif_cnt++;
18460 18466 break;
18461 18467 case EEXIST:
18462 18468 ip1dbg(("ipif_arp_up: NCE already exists for %s\n",
18463 18469 ill->ill_name));
18464 18470 if (!NCE_MYADDR(nce->nce_common)) {
18465 18471 /*
18466 18472 * A leftover nce from before this address
18467 18473 * existed
18468 18474 */
18469 18475 ncec_delete(nce->nce_common);
18470 18476 nce_refrele(nce);
18471 18477 nce = NULL;
18472 18478 goto retry;
18473 18479 }
18474 18480 if ((ipif->ipif_flags & IPIF_POINTOPOINT) == 0) {
18475 18481 nce_refrele(nce);
18476 18482 nce = NULL;
18477 18483 ip1dbg(("ipif_arp_up: NCE already exists "
18478 18484 "for %s:%u\n", ill->ill_name,
18479 18485 ipif->ipif_id));
18480 18486 goto arp_up_done;
18481 18487 }
18482 18488 /*
18483 18489 * Duplicate local addresses are permissible for
18484 18490 * IPIF_POINTOPOINT interfaces which will get marked
18485 18491 * IPIF_UNNUMBERED later in
18486 18492 * ip_addr_availability_check().
18487 18493 *
18488 18494 * The nce_ipif_cnt field tracks the number of
18489 18495 * ipifs that have nce_addr as their local address.
18490 18496 */
18491 18497 ipif->ipif_addr_ready = 1;
18492 18498 ipif->ipif_added_nce = 1;
18493 18499 nce->nce_ipif_cnt++;
18494 18500 err = 0;
18495 18501 break;
18496 18502 default:
18497 18503 ASSERT(nce == NULL);
18498 18504 goto arp_up_done;
18499 18505 }
18500 18506 if (arp_no_defense) {
18501 18507 if ((ipif->ipif_flags & IPIF_UP) &&
18502 18508 !ipif->ipif_addr_ready)
18503 18509 ipif_up_notify(ipif);
18504 18510 ipif->ipif_addr_ready = 1;
18505 18511 }
18506 18512 } else {
18507 18513 /* zero address. nothing to publish */
18508 18514 ipif->ipif_addr_ready = 1;
18509 18515 }
18510 18516 if (nce != NULL)
18511 18517 nce_refrele(nce);
18512 18518 arp_up_done:
18513 18519 if (added_ipif && err != 0)
18514 18520 ipmp_illgrp_del_ipif(ill->ill_grp, ipif);
18515 18521 return (err);
18516 18522 }
18517 18523
18518 18524 int
18519 18525 ipif_arp_up(ipif_t *ipif, enum ip_resolver_action res_act, boolean_t was_dup)
18520 18526 {
18521 18527 int err = 0;
18522 18528 ill_t *ill = ipif->ipif_ill;
18523 18529 boolean_t first_interface, wait_for_dlpi = B_FALSE;
18524 18530
18525 18531 DTRACE_PROBE3(ipif__downup, char *, "ipif_arp_up",
18526 18532 ill_t *, ill, ipif_t *, ipif);
18527 18533
18528 18534 /*
18529 18535 * need to bring up ARP or setup mcast mapping only
18530 18536 * when the first interface is coming UP.
18531 18537 */
18532 18538 first_interface = (ill->ill_ipif_up_count == 0 &&
18533 18539 ill->ill_ipif_dup_count == 0 && !was_dup);
18534 18540
18535 18541 if (res_act == Res_act_initial && first_interface) {
18536 18542 /*
18537 18543 * Send ATTACH + BIND
18538 18544 */
18539 18545 err = arp_ll_up(ill);
18540 18546 if (err != EINPROGRESS && err != 0)
18541 18547 return (err);
18542 18548
18543 18549 /*
18544 18550 * Add NCE for local address. Start DAD.
18545 18551 * we'll wait to hear that DAD has finished
18546 18552 * before using the interface.
18547 18553 */
18548 18554 if (err == EINPROGRESS)
18549 18555 wait_for_dlpi = B_TRUE;
18550 18556 }
18551 18557
18552 18558 if (!wait_for_dlpi)
18553 18559 (void) ipif_arp_up_done_tail(ipif, res_act);
18554 18560
18555 18561 return (!wait_for_dlpi ? 0 : EINPROGRESS);
18556 18562 }
18557 18563
18558 18564 /*
18559 18565 * Finish processing of "arp_up" after all the DLPI message
18560 18566 * exchanges have completed between arp and the driver.
18561 18567 */
18562 18568 void
18563 18569 arp_bringup_done(ill_t *ill, int err)
18564 18570 {
18565 18571 mblk_t *mp1;
18566 18572 ipif_t *ipif;
18567 18573 conn_t *connp = NULL;
18568 18574 ipsq_t *ipsq;
18569 18575 queue_t *q;
18570 18576
18571 18577 ip1dbg(("arp_bringup_done(%s)\n", ill->ill_name));
18572 18578
18573 18579 ASSERT(IAM_WRITER_ILL(ill));
18574 18580
18575 18581 ipsq = ill->ill_phyint->phyint_ipsq;
18576 18582 ipif = ipsq->ipsq_xop->ipx_pending_ipif;
18577 18583 mp1 = ipsq_pending_mp_get(ipsq, &connp);
18578 18584 ASSERT(!((mp1 != NULL) ^ (ipif != NULL)));
18579 18585 if (mp1 == NULL) /* bringup was aborted by the user */
18580 18586 return;
18581 18587
18582 18588 /*
18583 18589 * If an IOCTL is waiting on this (ipsq_current_ioctl != 0), then we
18584 18590 * must have an associated conn_t. Otherwise, we're bringing this
18585 18591 * interface back up as part of handling an asynchronous event (e.g.,
18586 18592 * physical address change).
18587 18593 */
18588 18594 if (ipsq->ipsq_xop->ipx_current_ioctl != 0) {
18589 18595 ASSERT(connp != NULL);
18590 18596 q = CONNP_TO_WQ(connp);
18591 18597 } else {
18592 18598 ASSERT(connp == NULL);
18593 18599 q = ill->ill_rq;
18594 18600 }
18595 18601 if (err == 0) {
18596 18602 if (ipif->ipif_isv6) {
18597 18603 if ((err = ipif_up_done_v6(ipif)) != 0)
18598 18604 ip0dbg(("arp_bringup_done: init failed\n"));
18599 18605 } else {
18600 18606 err = ipif_arp_up_done_tail(ipif, Res_act_initial);
18601 18607 if (err != 0 ||
18602 18608 (err = ipif_up_done(ipif)) != 0) {
18603 18609 ip0dbg(("arp_bringup_done: "
18604 18610 "init failed err %x\n", err));
18605 18611 (void) ipif_arp_down(ipif);
18606 18612 }
18607 18613
18608 18614 }
18609 18615 } else {
18610 18616 ip0dbg(("arp_bringup_done: DL_BIND_REQ failed\n"));
18611 18617 }
18612 18618
18613 18619 if ((err == 0) && (ill->ill_up_ipifs)) {
18614 18620 err = ill_up_ipifs(ill, q, mp1);
18615 18621 if (err == EINPROGRESS)
18616 18622 return;
18617 18623 }
18618 18624
18619 18625 /*
18620 18626 * If we have a moved ipif to bring up, and everything has succeeded
18621 18627 * to this point, bring it up on the IPMP ill. Otherwise, leave it
18622 18628 * down -- the admin can try to bring it up by hand if need be.
18623 18629 */
18624 18630 if (ill->ill_move_ipif != NULL) {
18625 18631 ipif = ill->ill_move_ipif;
18626 18632 ip1dbg(("bringing up ipif %p on ill %s\n", (void *)ipif,
18627 18633 ipif->ipif_ill->ill_name));
18628 18634 ill->ill_move_ipif = NULL;
18629 18635 if (err == 0) {
18630 18636 err = ipif_up(ipif, q, mp1);
18631 18637 if (err == EINPROGRESS)
18632 18638 return;
18633 18639 }
18634 18640 }
18635 18641
18636 18642 /*
18637 18643 * The operation must complete without EINPROGRESS since
18638 18644 * ipsq_pending_mp_get() has removed the mblk from ipsq_pending_mp.
18639 18645 * Otherwise, the operation will be stuck forever in the ipsq.
18640 18646 */
18641 18647 ASSERT(err != EINPROGRESS);
18642 18648 if (ipsq->ipsq_xop->ipx_current_ioctl != 0) {
18643 18649 DTRACE_PROBE4(ipif__ioctl, char *, "arp_bringup_done finish",
18644 18650 int, ipsq->ipsq_xop->ipx_current_ioctl,
18645 18651 ill_t *, ill, ipif_t *, ipif);
18646 18652 ip_ioctl_finish(q, mp1, err, NO_COPYOUT, ipsq);
18647 18653 } else {
18648 18654 ipsq_current_finish(ipsq);
18649 18655 }
18650 18656 }
18651 18657
18652 18658 /*
18653 18659 * Finish processing of arp replumb after all the DLPI message
18654 18660 * exchanges have completed between arp and the driver.
18655 18661 */
18656 18662 void
18657 18663 arp_replumb_done(ill_t *ill, int err)
18658 18664 {
18659 18665 mblk_t *mp1;
18660 18666 ipif_t *ipif;
18661 18667 conn_t *connp = NULL;
18662 18668 ipsq_t *ipsq;
18663 18669 queue_t *q;
18664 18670
18665 18671 ASSERT(IAM_WRITER_ILL(ill));
18666 18672
18667 18673 ipsq = ill->ill_phyint->phyint_ipsq;
18668 18674 ipif = ipsq->ipsq_xop->ipx_pending_ipif;
18669 18675 mp1 = ipsq_pending_mp_get(ipsq, &connp);
18670 18676 ASSERT(!((mp1 != NULL) ^ (ipif != NULL)));
18671 18677 if (mp1 == NULL) {
18672 18678 ip0dbg(("arp_replumb_done: bringup aborted ioctl %x\n",
18673 18679 ipsq->ipsq_xop->ipx_current_ioctl));
18674 18680 /* bringup was aborted by the user */
18675 18681 return;
18676 18682 }
18677 18683 /*
18678 18684 * If an IOCTL is waiting on this (ipsq_current_ioctl != 0), then we
18679 18685 * must have an associated conn_t. Otherwise, we're bringing this
18680 18686 * interface back up as part of handling an asynchronous event (e.g.,
18681 18687 * physical address change).
18682 18688 */
18683 18689 if (ipsq->ipsq_xop->ipx_current_ioctl != 0) {
18684 18690 ASSERT(connp != NULL);
18685 18691 q = CONNP_TO_WQ(connp);
18686 18692 } else {
18687 18693 ASSERT(connp == NULL);
18688 18694 q = ill->ill_rq;
18689 18695 }
18690 18696 if ((err == 0) && (ill->ill_up_ipifs)) {
18691 18697 err = ill_up_ipifs(ill, q, mp1);
18692 18698 if (err == EINPROGRESS)
18693 18699 return;
18694 18700 }
18695 18701 /*
18696 18702 * The operation must complete without EINPROGRESS since
18697 18703 * ipsq_pending_mp_get() has removed the mblk from ipsq_pending_mp.
18698 18704 * Otherwise, the operation will be stuck forever in the ipsq.
18699 18705 */
18700 18706 ASSERT(err != EINPROGRESS);
18701 18707 if (ipsq->ipsq_xop->ipx_current_ioctl != 0) {
18702 18708 DTRACE_PROBE4(ipif__ioctl, char *,
18703 18709 "arp_replumb_done finish",
18704 18710 int, ipsq->ipsq_xop->ipx_current_ioctl,
18705 18711 ill_t *, ill, ipif_t *, ipif);
18706 18712 ip_ioctl_finish(q, mp1, err, NO_COPYOUT, ipsq);
18707 18713 } else {
18708 18714 ipsq_current_finish(ipsq);
18709 18715 }
18710 18716 }
18711 18717
18712 18718 void
18713 18719 ipif_up_notify(ipif_t *ipif)
18714 18720 {
18715 18721 ip_rts_ifmsg(ipif, RTSQ_DEFAULT);
18716 18722 ip_rts_newaddrmsg(RTM_ADD, 0, ipif, RTSQ_DEFAULT);
18717 18723 sctp_update_ipif(ipif, SCTP_IPIF_UP);
18718 18724 ill_nic_event_dispatch(ipif->ipif_ill, MAP_IPIF_ID(ipif->ipif_id),
18719 18725 NE_LIF_UP, NULL, 0);
18720 18726 }
18721 18727
18722 18728 /*
18723 18729 * ILB ioctl uses cv_wait (such as deleting a rule or adding a server) and
18724 18730 * this assumes the context is cv_wait'able. Hence it shouldnt' be used on
18725 18731 * TPI end points with STREAMS modules pushed above. This is assured by not
18726 18732 * having the IPI_MODOK flag for the ioctl. And IP ensures the ILB ioctl
18727 18733 * never ends up on an ipsq, otherwise we may end up processing the ioctl
18728 18734 * while unwinding from the ispq and that could be a thread from the bottom.
18729 18735 */
18730 18736 /* ARGSUSED */
18731 18737 int
18732 18738 ip_sioctl_ilb_cmd(ipif_t *ipif, sin_t *sin, queue_t *q, mblk_t *mp,
18733 18739 ip_ioctl_cmd_t *ipip, void *arg)
18734 18740 {
18735 18741 mblk_t *cmd_mp = mp->b_cont->b_cont;
18736 18742 ilb_cmd_t command = *((ilb_cmd_t *)cmd_mp->b_rptr);
18737 18743 int ret = 0;
18738 18744 int i;
18739 18745 size_t size;
18740 18746 ip_stack_t *ipst;
18741 18747 zoneid_t zoneid;
18742 18748 ilb_stack_t *ilbs;
18743 18749
18744 18750 ipst = CONNQ_TO_IPST(q);
18745 18751 ilbs = ipst->ips_netstack->netstack_ilb;
18746 18752 zoneid = Q_TO_CONN(q)->conn_zoneid;
18747 18753
18748 18754 switch (command) {
18749 18755 case ILB_CREATE_RULE: {
18750 18756 ilb_rule_cmd_t *cmd = (ilb_rule_cmd_t *)cmd_mp->b_rptr;
18751 18757
18752 18758 if (MBLKL(cmd_mp) != sizeof (ilb_rule_cmd_t)) {
18753 18759 ret = EINVAL;
18754 18760 break;
18755 18761 }
18756 18762
18757 18763 ret = ilb_rule_add(ilbs, zoneid, cmd);
18758 18764 break;
18759 18765 }
18760 18766 case ILB_DESTROY_RULE:
18761 18767 case ILB_ENABLE_RULE:
18762 18768 case ILB_DISABLE_RULE: {
18763 18769 ilb_name_cmd_t *cmd = (ilb_name_cmd_t *)cmd_mp->b_rptr;
18764 18770
18765 18771 if (MBLKL(cmd_mp) != sizeof (ilb_name_cmd_t)) {
18766 18772 ret = EINVAL;
18767 18773 break;
18768 18774 }
18769 18775
18770 18776 if (cmd->flags & ILB_RULE_ALLRULES) {
18771 18777 if (command == ILB_DESTROY_RULE) {
18772 18778 ilb_rule_del_all(ilbs, zoneid);
18773 18779 break;
18774 18780 } else if (command == ILB_ENABLE_RULE) {
18775 18781 ilb_rule_enable_all(ilbs, zoneid);
18776 18782 break;
18777 18783 } else if (command == ILB_DISABLE_RULE) {
18778 18784 ilb_rule_disable_all(ilbs, zoneid);
18779 18785 break;
18780 18786 }
18781 18787 } else {
18782 18788 if (command == ILB_DESTROY_RULE) {
18783 18789 ret = ilb_rule_del(ilbs, zoneid, cmd->name);
18784 18790 } else if (command == ILB_ENABLE_RULE) {
18785 18791 ret = ilb_rule_enable(ilbs, zoneid, cmd->name,
18786 18792 NULL);
18787 18793 } else if (command == ILB_DISABLE_RULE) {
18788 18794 ret = ilb_rule_disable(ilbs, zoneid, cmd->name,
18789 18795 NULL);
18790 18796 }
18791 18797 }
18792 18798 break;
18793 18799 }
18794 18800 case ILB_NUM_RULES: {
18795 18801 ilb_num_rules_cmd_t *cmd;
18796 18802
18797 18803 if (MBLKL(cmd_mp) != sizeof (ilb_num_rules_cmd_t)) {
18798 18804 ret = EINVAL;
18799 18805 break;
18800 18806 }
18801 18807 cmd = (ilb_num_rules_cmd_t *)cmd_mp->b_rptr;
18802 18808 ilb_get_num_rules(ilbs, zoneid, &(cmd->num));
18803 18809 break;
18804 18810 }
18805 18811 case ILB_RULE_NAMES: {
18806 18812 ilb_rule_names_cmd_t *cmd;
18807 18813
18808 18814 cmd = (ilb_rule_names_cmd_t *)cmd_mp->b_rptr;
18809 18815 if (MBLKL(cmd_mp) < sizeof (ilb_rule_names_cmd_t) ||
18810 18816 cmd->num_names == 0) {
18811 18817 ret = EINVAL;
18812 18818 break;
18813 18819 }
18814 18820 size = cmd->num_names * ILB_RULE_NAMESZ;
18815 18821 if (cmd_mp->b_rptr + offsetof(ilb_rule_names_cmd_t, buf) +
18816 18822 size != cmd_mp->b_wptr) {
18817 18823 ret = EINVAL;
18818 18824 break;
18819 18825 }
18820 18826 ilb_get_rulenames(ilbs, zoneid, &cmd->num_names, cmd->buf);
18821 18827 break;
18822 18828 }
18823 18829 case ILB_NUM_SERVERS: {
18824 18830 ilb_num_servers_cmd_t *cmd;
18825 18831
18826 18832 if (MBLKL(cmd_mp) != sizeof (ilb_num_servers_cmd_t)) {
18827 18833 ret = EINVAL;
18828 18834 break;
18829 18835 }
18830 18836 cmd = (ilb_num_servers_cmd_t *)cmd_mp->b_rptr;
18831 18837 ret = ilb_get_num_servers(ilbs, zoneid, cmd->name,
18832 18838 &(cmd->num));
18833 18839 break;
18834 18840 }
18835 18841 case ILB_LIST_RULE: {
18836 18842 ilb_rule_cmd_t *cmd = (ilb_rule_cmd_t *)cmd_mp->b_rptr;
18837 18843
18838 18844 if (MBLKL(cmd_mp) != sizeof (ilb_rule_cmd_t)) {
18839 18845 ret = EINVAL;
18840 18846 break;
18841 18847 }
18842 18848 ret = ilb_rule_list(ilbs, zoneid, cmd);
18843 18849 break;
18844 18850 }
18845 18851 case ILB_LIST_SERVERS: {
18846 18852 ilb_servers_info_cmd_t *cmd;
18847 18853
18848 18854 cmd = (ilb_servers_info_cmd_t *)cmd_mp->b_rptr;
18849 18855 if (MBLKL(cmd_mp) < sizeof (ilb_servers_info_cmd_t) ||
18850 18856 cmd->num_servers == 0) {
18851 18857 ret = EINVAL;
18852 18858 break;
18853 18859 }
18854 18860 size = cmd->num_servers * sizeof (ilb_server_info_t);
18855 18861 if (cmd_mp->b_rptr + offsetof(ilb_servers_info_cmd_t, servers) +
18856 18862 size != cmd_mp->b_wptr) {
18857 18863 ret = EINVAL;
18858 18864 break;
18859 18865 }
18860 18866
18861 18867 ret = ilb_get_servers(ilbs, zoneid, cmd->name, cmd->servers,
18862 18868 &cmd->num_servers);
18863 18869 break;
18864 18870 }
18865 18871 case ILB_ADD_SERVERS: {
18866 18872 ilb_servers_info_cmd_t *cmd;
18867 18873 ilb_rule_t *rule;
18868 18874
18869 18875 cmd = (ilb_servers_info_cmd_t *)cmd_mp->b_rptr;
18870 18876 if (MBLKL(cmd_mp) < sizeof (ilb_servers_info_cmd_t)) {
18871 18877 ret = EINVAL;
18872 18878 break;
18873 18879 }
18874 18880 size = cmd->num_servers * sizeof (ilb_server_info_t);
18875 18881 if (cmd_mp->b_rptr + offsetof(ilb_servers_info_cmd_t, servers) +
18876 18882 size != cmd_mp->b_wptr) {
18877 18883 ret = EINVAL;
18878 18884 break;
18879 18885 }
18880 18886 rule = ilb_find_rule(ilbs, zoneid, cmd->name, &ret);
18881 18887 if (rule == NULL) {
18882 18888 ASSERT(ret != 0);
18883 18889 break;
18884 18890 }
18885 18891 for (i = 0; i < cmd->num_servers; i++) {
18886 18892 ilb_server_info_t *s;
18887 18893
18888 18894 s = &cmd->servers[i];
18889 18895 s->err = ilb_server_add(ilbs, rule, s);
18890 18896 }
18891 18897 ILB_RULE_REFRELE(rule);
18892 18898 break;
18893 18899 }
18894 18900 case ILB_DEL_SERVERS:
18895 18901 case ILB_ENABLE_SERVERS:
18896 18902 case ILB_DISABLE_SERVERS: {
18897 18903 ilb_servers_cmd_t *cmd;
18898 18904 ilb_rule_t *rule;
18899 18905 int (*f)();
18900 18906
18901 18907 cmd = (ilb_servers_cmd_t *)cmd_mp->b_rptr;
18902 18908 if (MBLKL(cmd_mp) < sizeof (ilb_servers_cmd_t)) {
18903 18909 ret = EINVAL;
18904 18910 break;
18905 18911 }
18906 18912 size = cmd->num_servers * sizeof (ilb_server_arg_t);
18907 18913 if (cmd_mp->b_rptr + offsetof(ilb_servers_cmd_t, servers) +
18908 18914 size != cmd_mp->b_wptr) {
18909 18915 ret = EINVAL;
18910 18916 break;
18911 18917 }
18912 18918
18913 18919 if (command == ILB_DEL_SERVERS)
18914 18920 f = ilb_server_del;
18915 18921 else if (command == ILB_ENABLE_SERVERS)
18916 18922 f = ilb_server_enable;
18917 18923 else if (command == ILB_DISABLE_SERVERS)
18918 18924 f = ilb_server_disable;
18919 18925
18920 18926 rule = ilb_find_rule(ilbs, zoneid, cmd->name, &ret);
18921 18927 if (rule == NULL) {
18922 18928 ASSERT(ret != 0);
18923 18929 break;
18924 18930 }
18925 18931
18926 18932 for (i = 0; i < cmd->num_servers; i++) {
18927 18933 ilb_server_arg_t *s;
18928 18934
18929 18935 s = &cmd->servers[i];
18930 18936 s->err = f(ilbs, zoneid, NULL, rule, &s->addr);
18931 18937 }
18932 18938 ILB_RULE_REFRELE(rule);
18933 18939 break;
18934 18940 }
18935 18941 case ILB_LIST_NAT_TABLE: {
18936 18942 ilb_list_nat_cmd_t *cmd;
18937 18943
18938 18944 cmd = (ilb_list_nat_cmd_t *)cmd_mp->b_rptr;
18939 18945 if (MBLKL(cmd_mp) < sizeof (ilb_list_nat_cmd_t)) {
18940 18946 ret = EINVAL;
18941 18947 break;
18942 18948 }
18943 18949 size = cmd->num_nat * sizeof (ilb_nat_entry_t);
18944 18950 if (cmd_mp->b_rptr + offsetof(ilb_list_nat_cmd_t, entries) +
18945 18951 size != cmd_mp->b_wptr) {
18946 18952 ret = EINVAL;
18947 18953 break;
18948 18954 }
18949 18955
18950 18956 ret = ilb_list_nat(ilbs, zoneid, cmd->entries, &cmd->num_nat,
18951 18957 &cmd->flags);
18952 18958 break;
18953 18959 }
18954 18960 case ILB_LIST_STICKY_TABLE: {
18955 18961 ilb_list_sticky_cmd_t *cmd;
18956 18962
18957 18963 cmd = (ilb_list_sticky_cmd_t *)cmd_mp->b_rptr;
18958 18964 if (MBLKL(cmd_mp) < sizeof (ilb_list_sticky_cmd_t)) {
18959 18965 ret = EINVAL;
18960 18966 break;
18961 18967 }
18962 18968 size = cmd->num_sticky * sizeof (ilb_sticky_entry_t);
18963 18969 if (cmd_mp->b_rptr + offsetof(ilb_list_sticky_cmd_t, entries) +
18964 18970 size != cmd_mp->b_wptr) {
18965 18971 ret = EINVAL;
18966 18972 break;
18967 18973 }
18968 18974
18969 18975 ret = ilb_list_sticky(ilbs, zoneid, cmd->entries,
18970 18976 &cmd->num_sticky, &cmd->flags);
18971 18977 break;
18972 18978 }
18973 18979 default:
18974 18980 ret = EINVAL;
18975 18981 break;
18976 18982 }
18977 18983 done:
18978 18984 return (ret);
18979 18985 }
18980 18986
18981 18987 /* Remove all cache entries for this logical interface */
18982 18988 void
18983 18989 ipif_nce_down(ipif_t *ipif)
18984 18990 {
18985 18991 ill_t *ill = ipif->ipif_ill;
18986 18992 nce_t *nce;
18987 18993
18988 18994 DTRACE_PROBE3(ipif__downup, char *, "ipif_nce_down",
18989 18995 ill_t *, ill, ipif_t *, ipif);
18990 18996 if (ipif->ipif_added_nce) {
18991 18997 if (ipif->ipif_isv6)
18992 18998 nce = nce_lookup_v6(ill, &ipif->ipif_v6lcl_addr);
18993 18999 else
18994 19000 nce = nce_lookup_v4(ill, &ipif->ipif_lcl_addr);
18995 19001 if (nce != NULL) {
18996 19002 if (--nce->nce_ipif_cnt == 0)
18997 19003 ncec_delete(nce->nce_common);
18998 19004 ipif->ipif_added_nce = 0;
18999 19005 nce_refrele(nce);
19000 19006 } else {
19001 19007 /*
19002 19008 * nce may already be NULL because it was already
19003 19009 * flushed, e.g., due to a call to nce_flush
19004 19010 */
19005 19011 ipif->ipif_added_nce = 0;
19006 19012 }
19007 19013 }
19008 19014 /*
19009 19015 * Make IPMP aware of the deleted data address.
19010 19016 */
19011 19017 if (IS_IPMP(ill))
19012 19018 ipmp_illgrp_del_ipif(ill->ill_grp, ipif);
19013 19019
19014 19020 /*
19015 19021 * Remove all other nces dependent on this ill when the last ipif
19016 19022 * is going away.
19017 19023 */
19018 19024 if (ill->ill_ipif_up_count == 0) {
19019 19025 ncec_walk(ill, (pfi_t)ncec_delete_per_ill,
19020 19026 (uchar_t *)ill, ill->ill_ipst);
19021 19027 if (IS_UNDER_IPMP(ill))
19022 19028 nce_flush(ill, B_TRUE);
19023 19029 }
19024 19030 }
19025 19031
19026 19032 /*
19027 19033 * find the first interface that uses usill for its source address.
19028 19034 */
19029 19035 ill_t *
19030 19036 ill_lookup_usesrc(ill_t *usill)
19031 19037 {
19032 19038 ip_stack_t *ipst = usill->ill_ipst;
19033 19039 ill_t *ill;
19034 19040
19035 19041 ASSERT(usill != NULL);
19036 19042
19037 19043 /* ill_g_usesrc_lock protects ill_usesrc_grp_next */
19038 19044 rw_enter(&ipst->ips_ill_g_usesrc_lock, RW_WRITER);
19039 19045 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
19040 19046 for (ill = usill->ill_usesrc_grp_next; ill != NULL && ill != usill;
19041 19047 ill = ill->ill_usesrc_grp_next) {
19042 19048 if (!IS_UNDER_IPMP(ill) && (ill->ill_flags & ILLF_MULTICAST) &&
19043 19049 !ILL_IS_CONDEMNED(ill)) {
19044 19050 ill_refhold(ill);
19045 19051 break;
19046 19052 }
19047 19053 }
19048 19054 rw_exit(&ipst->ips_ill_g_lock);
19049 19055 rw_exit(&ipst->ips_ill_g_usesrc_lock);
19050 19056 return (ill);
19051 19057 }
19052 19058
19053 19059 /*
19054 19060 * This comment applies to both ip_sioctl_get_ifhwaddr and
19055 19061 * ip_sioctl_get_lifhwaddr as the basic function of these two functions
19056 19062 * is the same.
19057 19063 *
19058 19064 * The goal here is to find an IP interface that corresponds to the name
19059 19065 * provided by the caller in the ifreq/lifreq structure held in the mblk_t
19060 19066 * chain and to fill out a sockaddr/sockaddr_storage structure with the
19061 19067 * mac address.
19062 19068 *
19063 19069 * The SIOCGIFHWADDR/SIOCGLIFHWADDR ioctl may return an error for a number
19064 19070 * of different reasons:
19065 19071 * ENXIO - the device name is not known to IP.
19066 19072 * EADDRNOTAVAIL - the device has no hardware address. This is indicated
19067 19073 * by ill_phys_addr not pointing to an actual address.
19068 19074 * EPFNOSUPPORT - this will indicate that a request is being made for a
19069 19075 * mac address that will not fit in the data structure supplier (struct
19070 19076 * sockaddr).
19071 19077 *
19072 19078 */
19073 19079 /* ARGSUSED */
19074 19080 int
19075 19081 ip_sioctl_get_ifhwaddr(ipif_t *ipif, sin_t *dummy_sin, queue_t *q, mblk_t *mp,
19076 19082 ip_ioctl_cmd_t *ipip, void *if_req)
19077 19083 {
19078 19084 struct sockaddr *sock;
19079 19085 struct ifreq *ifr;
19080 19086 mblk_t *mp1;
19081 19087 ill_t *ill;
19082 19088
19083 19089 ASSERT(ipif != NULL);
19084 19090 ill = ipif->ipif_ill;
19085 19091
19086 19092 if (ill->ill_phys_addr == NULL) {
19087 19093 return (EADDRNOTAVAIL);
19088 19094 }
19089 19095 if (ill->ill_phys_addr_length > sizeof (sock->sa_data)) {
19090 19096 return (EPFNOSUPPORT);
19091 19097 }
19092 19098
19093 19099 ip1dbg(("ip_sioctl_get_hwaddr(%s)\n", ill->ill_name));
19094 19100
19095 19101 /* Existence of mp1 has been checked in ip_wput_nondata */
19096 19102 mp1 = mp->b_cont->b_cont;
19097 19103 ifr = (struct ifreq *)mp1->b_rptr;
19098 19104
19099 19105 sock = &ifr->ifr_addr;
19100 19106 /*
19101 19107 * The "family" field in the returned structure is set to a value
19102 19108 * that represents the type of device to which the address belongs.
19103 19109 * The value returned may differ to that on Linux but it will still
19104 19110 * represent the correct symbol on Solaris.
19105 19111 */
19106 19112 sock->sa_family = arp_hw_type(ill->ill_mactype);
19107 19113 bcopy(ill->ill_phys_addr, &sock->sa_data, ill->ill_phys_addr_length);
19108 19114
19109 19115 return (0);
19110 19116 }
19111 19117
19112 19118 /*
19113 19119 * The expection of applications using SIOCGIFHWADDR is that data will
19114 19120 * be returned in the sa_data field of the sockaddr structure. With
19115 19121 * SIOCGLIFHWADDR, we're breaking new ground as there is no Linux
19116 19122 * equivalent. In light of this, struct sockaddr_dl is used as it
19117 19123 * offers more space for address storage in sll_data.
19118 19124 */
19119 19125 /* ARGSUSED */
19120 19126 int
19121 19127 ip_sioctl_get_lifhwaddr(ipif_t *ipif, sin_t *dummy_sin, queue_t *q, mblk_t *mp,
19122 19128 ip_ioctl_cmd_t *ipip, void *if_req)
19123 19129 {
19124 19130 struct sockaddr_dl *sock;
19125 19131 struct lifreq *lifr;
19126 19132 mblk_t *mp1;
19127 19133 ill_t *ill;
19128 19134
19129 19135 ASSERT(ipif != NULL);
19130 19136 ill = ipif->ipif_ill;
19131 19137
19132 19138 if (ill->ill_phys_addr == NULL) {
19133 19139 return (EADDRNOTAVAIL);
19134 19140 }
19135 19141 if (ill->ill_phys_addr_length > sizeof (sock->sdl_data)) {
19136 19142 return (EPFNOSUPPORT);
19137 19143 }
19138 19144
19139 19145 ip1dbg(("ip_sioctl_get_lifhwaddr(%s)\n", ill->ill_name));
19140 19146
19141 19147 /* Existence of mp1 has been checked in ip_wput_nondata */
19142 19148 mp1 = mp->b_cont->b_cont;
19143 19149 lifr = (struct lifreq *)mp1->b_rptr;
19144 19150
19145 19151 /*
19146 19152 * sockaddr_ll is used here because it is also the structure used in
19147 19153 * responding to the same ioctl in sockpfp. The only other choice is
19148 19154 * sockaddr_dl which contains fields that are not required here
19149 19155 * because its purpose is different.
19150 19156 */
19151 19157 lifr->lifr_type = ill->ill_type;
19152 19158 sock = (struct sockaddr_dl *)&lifr->lifr_addr;
19153 19159 sock->sdl_family = AF_LINK;
19154 19160 sock->sdl_index = ill->ill_phyint->phyint_ifindex;
19155 19161 sock->sdl_type = ill->ill_mactype;
19156 19162 sock->sdl_nlen = 0;
19157 19163 sock->sdl_slen = 0;
19158 19164 sock->sdl_alen = ill->ill_phys_addr_length;
19159 19165 bcopy(ill->ill_phys_addr, sock->sdl_data, ill->ill_phys_addr_length);
19160 19166
19161 19167 return (0);
19162 19168 }
|
↓ open down ↓ |
10248 lines elided |
↑ open up ↑ |
XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX