1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 
  22 /*
  23  * Copyright (c) 1991, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Copyright (c) 1990 Mentat Inc.
  25  * Copyright (c) 2017 OmniTI Computer Consulting, Inc. All rights reserved.
  26  * Copyright (c) 2016 by Delphix. All rights reserved.
  27  * Copyright (c) 2018 Joyent, Inc. All rights reserved.
  28  */
  29 
  30 #include <sys/types.h>
  31 #include <sys/stream.h>
  32 #include <sys/dlpi.h>
  33 #include <sys/stropts.h>
  34 #include <sys/sysmacros.h>
  35 #include <sys/strsubr.h>
  36 #include <sys/strlog.h>
  37 #include <sys/strsun.h>
  38 #include <sys/zone.h>
  39 #define _SUN_TPI_VERSION 2
  40 #include <sys/tihdr.h>
  41 #include <sys/xti_inet.h>
  42 #include <sys/ddi.h>
  43 #include <sys/suntpi.h>
  44 #include <sys/cmn_err.h>
  45 #include <sys/debug.h>
  46 #include <sys/kobj.h>
  47 #include <sys/modctl.h>
  48 #include <sys/atomic.h>
  49 #include <sys/policy.h>
  50 #include <sys/priv.h>
  51 #include <sys/taskq.h>
  52 
  53 #include <sys/systm.h>
  54 #include <sys/param.h>
  55 #include <sys/kmem.h>
  56 #include <sys/sdt.h>
  57 #include <sys/socket.h>
  58 #include <sys/vtrace.h>
  59 #include <sys/isa_defs.h>
  60 #include <sys/mac.h>
  61 #include <net/if.h>
  62 #include <net/if_arp.h>
  63 #include <net/route.h>
  64 #include <sys/sockio.h>
  65 #include <netinet/in.h>
  66 #include <net/if_dl.h>
  67 
  68 #include <inet/common.h>
  69 #include <inet/mi.h>
  70 #include <inet/mib2.h>
  71 #include <inet/nd.h>
  72 #include <inet/arp.h>
  73 #include <inet/snmpcom.h>
  74 #include <inet/optcom.h>
  75 #include <inet/kstatcom.h>
  76 
  77 #include <netinet/igmp_var.h>
  78 #include <netinet/ip6.h>
  79 #include <netinet/icmp6.h>
  80 #include <netinet/sctp.h>
  81 
  82 #include <inet/ip.h>
  83 #include <inet/ip_impl.h>
  84 #include <inet/ip6.h>
  85 #include <inet/ip6_asp.h>
  86 #include <inet/tcp.h>
  87 #include <inet/tcp_impl.h>
  88 #include <inet/ip_multi.h>
  89 #include <inet/ip_if.h>
  90 #include <inet/ip_ire.h>
  91 #include <inet/ip_ftable.h>
  92 #include <inet/ip_rts.h>
  93 #include <inet/ip_ndp.h>
  94 #include <inet/ip_listutils.h>
  95 #include <netinet/igmp.h>
  96 #include <netinet/ip_mroute.h>
  97 #include <inet/ipp_common.h>
  98 
  99 #include <net/pfkeyv2.h>
 100 #include <inet/sadb.h>
 101 #include <inet/ipsec_impl.h>
 102 #include <inet/iptun/iptun_impl.h>
 103 #include <inet/ipdrop.h>
 104 #include <inet/ip_netinfo.h>
 105 #include <inet/ilb_ip.h>
 106 
 107 #include <sys/ethernet.h>
 108 #include <net/if_types.h>
 109 #include <sys/cpuvar.h>
 110 
 111 #include <ipp/ipp.h>
 112 #include <ipp/ipp_impl.h>
 113 #include <ipp/ipgpc/ipgpc.h>
 114 
 115 #include <sys/pattr.h>
 116 #include <inet/ipclassifier.h>
 117 #include <inet/sctp_ip.h>
 118 #include <inet/sctp/sctp_impl.h>
 119 #include <inet/udp_impl.h>
 120 #include <inet/rawip_impl.h>
 121 #include <inet/rts_impl.h>
 122 
 123 #include <sys/tsol/label.h>
 124 #include <sys/tsol/tnet.h>
 125 
 126 #include <sys/squeue_impl.h>
 127 #include <inet/ip_arp.h>
 128 
 129 #include <sys/clock_impl.h>       /* For LBOLT_FASTPATH{,64} */
 130 
 131 /*
 132  * Values for squeue switch:
 133  * IP_SQUEUE_ENTER_NODRAIN: SQ_NODRAIN
 134  * IP_SQUEUE_ENTER: SQ_PROCESS
 135  * IP_SQUEUE_FILL: SQ_FILL
 136  */
 137 int ip_squeue_enter = IP_SQUEUE_ENTER;  /* Setable in /etc/system */
 138 
 139 int ip_squeue_flag;
 140 
 141 /*
 142  * Setable in /etc/system
 143  */
 144 int ip_poll_normal_ms = 100;
 145 int ip_poll_normal_ticks = 0;
 146 int ip_modclose_ackwait_ms = 3000;
 147 
 148 /*
 149  * It would be nice to have these present only in DEBUG systems, but the
 150  * current design of the global symbol checking logic requires them to be
 151  * unconditionally present.
 152  */
 153 uint_t ip_thread_data;                  /* TSD key for debug support */
 154 krwlock_t ip_thread_rwlock;
 155 list_t  ip_thread_list;
 156 
 157 /*
 158  * Structure to represent a linked list of msgblks. Used by ip_snmp_ functions.
 159  */
 160 
 161 struct listptr_s {
 162         mblk_t  *lp_head;       /* pointer to the head of the list */
 163         mblk_t  *lp_tail;       /* pointer to the tail of the list */
 164 };
 165 
 166 typedef struct listptr_s listptr_t;
 167 
 168 /*
 169  * This is used by ip_snmp_get_mib2_ip_route_media and
 170  * ip_snmp_get_mib2_ip6_route_media to carry the lists of return data.
 171  */
 172 typedef struct iproutedata_s {
 173         uint_t          ird_idx;
 174         uint_t          ird_flags;      /* see below */
 175         listptr_t       ird_route;      /* ipRouteEntryTable */
 176         listptr_t       ird_netmedia;   /* ipNetToMediaEntryTable */
 177         listptr_t       ird_attrs;      /* ipRouteAttributeTable */
 178 } iproutedata_t;
 179 
 180 /* Include ire_testhidden and IRE_IF_CLONE routes */
 181 #define IRD_REPORT_ALL  0x01
 182 
 183 /*
 184  * Cluster specific hooks. These should be NULL when booted as a non-cluster
 185  */
 186 
 187 /*
 188  * Hook functions to enable cluster networking
 189  * On non-clustered systems these vectors must always be NULL.
 190  *
 191  * Hook function to Check ip specified ip address is a shared ip address
 192  * in the cluster
 193  *
 194  */
 195 int (*cl_inet_isclusterwide)(netstackid_t stack_id, uint8_t protocol,
 196     sa_family_t addr_family, uint8_t *laddrp, void *args) = NULL;
 197 
 198 /*
 199  * Hook function to generate cluster wide ip fragment identifier
 200  */
 201 uint32_t (*cl_inet_ipident)(netstackid_t stack_id, uint8_t protocol,
 202     sa_family_t addr_family, uint8_t *laddrp, uint8_t *faddrp,
 203     void *args) = NULL;
 204 
 205 /*
 206  * Hook function to generate cluster wide SPI.
 207  */
 208 void (*cl_inet_getspi)(netstackid_t, uint8_t, uint8_t *, size_t,
 209     void *) = NULL;
 210 
 211 /*
 212  * Hook function to verify if the SPI is already utlized.
 213  */
 214 
 215 int (*cl_inet_checkspi)(netstackid_t, uint8_t, uint32_t, void *) = NULL;
 216 
 217 /*
 218  * Hook function to delete the SPI from the cluster wide repository.
 219  */
 220 
 221 void (*cl_inet_deletespi)(netstackid_t, uint8_t, uint32_t, void *) = NULL;
 222 
 223 /*
 224  * Hook function to inform the cluster when packet received on an IDLE SA
 225  */
 226 
 227 void (*cl_inet_idlesa)(netstackid_t, uint8_t, uint32_t, sa_family_t,
 228     in6_addr_t, in6_addr_t, void *) = NULL;
 229 
 230 /*
 231  * Synchronization notes:
 232  *
 233  * IP is a fully D_MP STREAMS module/driver. Thus it does not depend on any
 234  * MT level protection given by STREAMS. IP uses a combination of its own
 235  * internal serialization mechanism and standard Solaris locking techniques.
 236  * The internal serialization is per phyint.  This is used to serialize
 237  * plumbing operations, IPMP operations, most set ioctls, etc.
 238  *
 239  * Plumbing is a long sequence of operations involving message
 240  * exchanges between IP, ARP and device drivers. Many set ioctls are typically
 241  * involved in plumbing operations. A natural model is to serialize these
 242  * ioctls one per ill. For example plumbing of hme0 and qfe0 can go on in
 243  * parallel without any interference. But various set ioctls on hme0 are best
 244  * serialized, along with IPMP operations and processing of DLPI control
 245  * messages received from drivers on a per phyint basis. This serialization is
 246  * provided by the ipsq_t and primitives operating on this. Details can
 247  * be found in ip_if.c above the core primitives operating on ipsq_t.
 248  *
 249  * Lookups of an ipif or ill by a thread return a refheld ipif / ill.
 250  * Simiarly lookup of an ire by a thread also returns a refheld ire.
 251  * In addition ipif's and ill's referenced by the ire are also indirectly
 252  * refheld. Thus no ipif or ill can vanish as long as an ipif is refheld
 253  * directly or indirectly. For example an SIOCSLIFADDR ioctl that changes the
 254  * address of an ipif has to go through the ipsq_t. This ensures that only
 255  * one such exclusive operation proceeds at any time on the ipif. It then
 256  * waits for all refcnts
 257  * associated with this ipif to come down to zero. The address is changed
 258  * only after the ipif has been quiesced. Then the ipif is brought up again.
 259  * More details are described above the comment in ip_sioctl_flags.
 260  *
 261  * Packet processing is based mostly on IREs and are fully multi-threaded
 262  * using standard Solaris MT techniques.
 263  *
 264  * There are explicit locks in IP to handle:
 265  * - The ip_g_head list maintained by mi_open_link() and friends.
 266  *
 267  * - The reassembly data structures (one lock per hash bucket)
 268  *
 269  * - conn_lock is meant to protect conn_t fields. The fields actually
 270  *   protected by conn_lock are documented in the conn_t definition.
 271  *
 272  * - ire_lock to protect some of the fields of the ire, IRE tables
 273  *   (one lock per hash bucket). Refer to ip_ire.c for details.
 274  *
 275  * - ndp_g_lock and ncec_lock for protecting NCEs.
 276  *
 277  * - ill_lock protects fields of the ill and ipif. Details in ip.h
 278  *
 279  * - ill_g_lock: This is a global reader/writer lock. Protects the following
 280  *      * The AVL tree based global multi list of all ills.
 281  *      * The linked list of all ipifs of an ill
 282  *      * The <ipsq-xop> mapping
 283  *      * <ill-phyint> association
 284  *   Insertion/deletion of an ill in the system, insertion/deletion of an ipif
 285  *   into an ill, changing the <ipsq-xop> mapping of an ill, changing the
 286  *   <ill-phyint> assoc of an ill will all have to hold the ill_g_lock as
 287  *   writer for the actual duration of the insertion/deletion/change.
 288  *
 289  * - ill_lock:  This is a per ill mutex.
 290  *   It protects some members of the ill_t struct; see ip.h for details.
 291  *   It also protects the <ill-phyint> assoc.
 292  *   It also protects the list of ipifs hanging off the ill.
 293  *
 294  * - ipsq_lock: This is a per ipsq_t mutex lock.
 295  *   This protects some members of the ipsq_t struct; see ip.h for details.
 296  *   It also protects the <ipsq-ipxop> mapping
 297  *
 298  * - ipx_lock: This is a per ipxop_t mutex lock.
 299  *   This protects some members of the ipxop_t struct; see ip.h for details.
 300  *
 301  * - phyint_lock: This is a per phyint mutex lock. Protects just the
 302  *   phyint_flags
 303  *
 304  * - ip_addr_avail_lock: This is used to ensure the uniqueness of IP addresses.
 305  *   This lock is held in ipif_up_done and the ipif is marked IPIF_UP and the
 306  *   uniqueness check also done atomically.
 307  *
 308  * - ill_g_usesrc_lock: This readers/writer lock protects the usesrc
 309  *   group list linked by ill_usesrc_grp_next. It also protects the
 310  *   ill_usesrc_ifindex field. It is taken as a writer when a member of the
 311  *   group is being added or deleted.  This lock is taken as a reader when
 312  *   walking the list/group(eg: to get the number of members in a usesrc group).
 313  *   Note, it is only necessary to take this lock if the ill_usesrc_grp_next
 314  *   field is changing state i.e from NULL to non-NULL or vice-versa. For
 315  *   example, it is not necessary to take this lock in the initial portion
 316  *   of ip_sioctl_slifusesrc or at all in ip_sioctl_flags since these
 317  *   operations are executed exclusively and that ensures that the "usesrc
 318  *   group state" cannot change. The "usesrc group state" change can happen
 319  *   only in the latter part of ip_sioctl_slifusesrc and in ill_delete.
 320  *
 321  * Changing <ill-phyint>, <ipsq-xop> assocications:
 322  *
 323  * To change the <ill-phyint> association, the ill_g_lock must be held
 324  * as writer, and the ill_locks of both the v4 and v6 instance of the ill
 325  * must be held.
 326  *
 327  * To change the <ipsq-xop> association, the ill_g_lock must be held as
 328  * writer, the ipsq_lock must be held, and one must be writer on the ipsq.
 329  * This is only done when ills are added or removed from IPMP groups.
 330  *
 331  * To add or delete an ipif from the list of ipifs hanging off the ill,
 332  * ill_g_lock (writer) and ill_lock must be held and the thread must be
 333  * a writer on the associated ipsq.
 334  *
 335  * To add or delete an ill to the system, the ill_g_lock must be held as
 336  * writer and the thread must be a writer on the associated ipsq.
 337  *
 338  * To add or delete an ilm to an ill, the ill_lock must be held and the thread
 339  * must be a writer on the associated ipsq.
 340  *
 341  * Lock hierarchy
 342  *
 343  * Some lock hierarchy scenarios are listed below.
 344  *
 345  * ill_g_lock -> conn_lock -> ill_lock -> ipsq_lock -> ipx_lock
 346  * ill_g_lock -> ill_lock(s) -> phyint_lock
 347  * ill_g_lock -> ndp_g_lock -> ill_lock -> ncec_lock
 348  * ill_g_lock -> ip_addr_avail_lock
 349  * conn_lock -> irb_lock -> ill_lock -> ire_lock
 350  * ill_g_lock -> ip_g_nd_lock
 351  * ill_g_lock -> ips_ipmp_lock -> ill_lock -> nce_lock
 352  * ill_g_lock -> ndp_g_lock -> ill_lock -> ncec_lock -> nce_lock
 353  * arl_lock -> ill_lock
 354  * ips_ire_dep_lock -> irb_lock
 355  *
 356  * When more than 1 ill lock is needed to be held, all ill lock addresses
 357  * are sorted on address and locked starting from highest addressed lock
 358  * downward.
 359  *
 360  * Multicast scenarios
 361  * ips_ill_g_lock -> ill_mcast_lock
 362  * conn_ilg_lock -> ips_ill_g_lock -> ill_lock
 363  * ill_mcast_serializer -> ill_mcast_lock -> ips_ipmp_lock -> ill_lock
 364  * ill_mcast_serializer -> ill_mcast_lock -> connf_lock -> conn_lock
 365  * ill_mcast_serializer -> ill_mcast_lock -> conn_ilg_lock
 366  * ill_mcast_serializer -> ill_mcast_lock -> ips_igmp_timer_lock
 367  *
 368  * IPsec scenarios
 369  *
 370  * ipsa_lock -> ill_g_lock -> ill_lock
 371  * ill_g_usesrc_lock -> ill_g_lock -> ill_lock
 372  *
 373  * Trusted Solaris scenarios
 374  *
 375  * igsa_lock -> gcgrp_rwlock -> gcgrp_lock
 376  * igsa_lock -> gcdb_lock
 377  * gcgrp_rwlock -> ire_lock
 378  * gcgrp_rwlock -> gcdb_lock
 379  *
 380  * squeue(sq_lock), flow related (ft_lock, fe_lock) locking
 381  *
 382  * cpu_lock --> ill_lock --> sqset_lock --> sq_lock
 383  * sq_lock -> conn_lock -> QLOCK(q)
 384  * ill_lock -> ft_lock -> fe_lock
 385  *
 386  * Routing/forwarding table locking notes:
 387  *
 388  * Lock acquisition order: Radix tree lock, irb_lock.
 389  * Requirements:
 390  * i.  Walker must not hold any locks during the walker callback.
 391  * ii  Walker must not see a truncated tree during the walk because of any node
 392  *     deletion.
 393  * iii Existing code assumes ire_bucket is valid if it is non-null and is used
 394  *     in many places in the code to walk the irb list. Thus even if all the
 395  *     ires in a bucket have been deleted, we still can't free the radix node
 396  *     until the ires have actually been inactive'd (freed).
 397  *
 398  * Tree traversal - Need to hold the global tree lock in read mode.
 399  * Before dropping the global tree lock, need to either increment the ire_refcnt
 400  * to ensure that the radix node can't be deleted.
 401  *
 402  * Tree add - Need to hold the global tree lock in write mode to add a
 403  * radix node. To prevent the node from being deleted, increment the
 404  * irb_refcnt, after the node is added to the tree. The ire itself is
 405  * added later while holding the irb_lock, but not the tree lock.
 406  *
 407  * Tree delete - Need to hold the global tree lock and irb_lock in write mode.
 408  * All associated ires must be inactive (i.e. freed), and irb_refcnt
 409  * must be zero.
 410  *
 411  * Walker - Increment irb_refcnt before calling the walker callback. Hold the
 412  * global tree lock (read mode) for traversal.
 413  *
 414  * IRE dependencies - In some cases we hold ips_ire_dep_lock across ire_refrele
 415  * hence we will acquire irb_lock while holding ips_ire_dep_lock.
 416  *
 417  * IPsec notes :
 418  *
 419  * IP interacts with the IPsec code (AH/ESP) by storing IPsec attributes
 420  * in the ip_xmit_attr_t ip_recv_attr_t. For outbound datagrams, the
 421  * ip_xmit_attr_t has the
 422  * information used by the IPsec code for applying the right level of
 423  * protection. The information initialized by IP in the ip_xmit_attr_t
 424  * is determined by the per-socket policy or global policy in the system.
 425  * For inbound datagrams, the ip_recv_attr_t
 426  * starts out with nothing in it. It gets filled
 427  * with the right information if it goes through the AH/ESP code, which
 428  * happens if the incoming packet is secure. The information initialized
 429  * by AH/ESP, is later used by IP (during fanouts to ULP) to see whether
 430  * the policy requirements needed by per-socket policy or global policy
 431  * is met or not.
 432  *
 433  * For fully connected sockets i.e dst, src [addr, port] is known,
 434  * conn_policy_cached is set indicating that policy has been cached.
 435  * conn_in_enforce_policy may or may not be set depending on whether
 436  * there is a global policy match or per-socket policy match.
 437  * Policy inheriting happpens in ip_policy_set once the destination is known.
 438  * Once the right policy is set on the conn_t, policy cannot change for
 439  * this socket. This makes life simpler for TCP (UDP ?) where
 440  * re-transmissions go out with the same policy. For symmetry, policy
 441  * is cached for fully connected UDP sockets also. Thus if policy is cached,
 442  * it also implies that policy is latched i.e policy cannot change
 443  * on these sockets. As we have the right policy on the conn, we don't
 444  * have to lookup global policy for every outbound and inbound datagram
 445  * and thus serving as an optimization. Note that a global policy change
 446  * does not affect fully connected sockets if they have policy. If fully
 447  * connected sockets did not have any policy associated with it, global
 448  * policy change may affect them.
 449  *
 450  * IP Flow control notes:
 451  * ---------------------
 452  * Non-TCP streams are flow controlled by IP. The way this is accomplished
 453  * differs when ILL_CAPAB_DLD_DIRECT is enabled for that IP instance. When
 454  * ILL_DIRECT_CAPABLE(ill) is TRUE, IP can do direct function calls into
 455  * GLDv3. Otherwise packets are sent down to lower layers using STREAMS
 456  * functions.
 457  *
 458  * Per Tx ring udp flow control:
 459  * This is applicable only when ILL_CAPAB_DLD_DIRECT capability is set in
 460  * the ill (i.e. ILL_DIRECT_CAPABLE(ill) is true).
 461  *
 462  * The underlying link can expose multiple Tx rings to the GLDv3 mac layer.
 463  * To achieve best performance, outgoing traffic need to be fanned out among
 464  * these Tx ring. mac_tx() is called (via str_mdata_fastpath_put()) to send
 465  * traffic out of the NIC and it takes a fanout hint. UDP connections pass
 466  * the address of connp as fanout hint to mac_tx(). Under flow controlled
 467  * condition, mac_tx() returns a non-NULL cookie (ip_mac_tx_cookie_t). This
 468  * cookie points to a specific Tx ring that is blocked. The cookie is used to
 469  * hash into an idl_tx_list[] entry in idl_tx_list[] array. Each idl_tx_list_t
 470  * point to drain_lists (idl_t's). These drain list will store the blocked UDP
 471  * connp's. The drain list is not a single list but a configurable number of
 472  * lists.
 473  *
 474  * The diagram below shows idl_tx_list_t's and their drain_lists. ip_stack_t
 475  * has an array of idl_tx_list_t. The size of the array is TX_FANOUT_SIZE
 476  * which is equal to 128. This array in turn contains a pointer to idl_t[],
 477  * the ip drain list. The idl_t[] array size is MIN(max_ncpus, 8). The drain
 478  * list will point to the list of connp's that are flow controlled.
 479  *
 480  *                      ---------------   -------   -------   -------
 481  *                   |->|drain_list[0]|-->|connp|-->|connp|-->|connp|-->
 482  *                   |  ---------------   -------   -------   -------
 483  *                   |  ---------------   -------   -------   -------
 484  *                   |->|drain_list[1]|-->|connp|-->|connp|-->|connp|-->
 485  * ----------------  |  ---------------   -------   -------   -------
 486  * |idl_tx_list[0]|->|  ---------------   -------   -------   -------
 487  * ----------------  |->|drain_list[2]|-->|connp|-->|connp|-->|connp|-->
 488  *                   |  ---------------   -------   -------   -------
 489  *                   .        .              .         .         .
 490  *                   |  ---------------   -------   -------   -------
 491  *                   |->|drain_list[n]|-->|connp|-->|connp|-->|connp|-->
 492  *                      ---------------   -------   -------   -------
 493  *                      ---------------   -------   -------   -------
 494  *                   |->|drain_list[0]|-->|connp|-->|connp|-->|connp|-->
 495  *                   |  ---------------   -------   -------   -------
 496  *                   |  ---------------   -------   -------   -------
 497  * ----------------  |->|drain_list[1]|-->|connp|-->|connp|-->|connp|-->
 498  * |idl_tx_list[1]|->|  ---------------   -------   -------   -------
 499  * ----------------  |        .              .         .         .
 500  *                   |  ---------------   -------   -------   -------
 501  *                   |->|drain_list[n]|-->|connp|-->|connp|-->|connp|-->
 502  *                      ---------------   -------   -------   -------
 503  *     .....
 504  * ----------------
 505  * |idl_tx_list[n]|-> ...
 506  * ----------------
 507  *
 508  * When mac_tx() returns a cookie, the cookie is hashed into an index into
 509  * ips_idl_tx_list[], and conn_drain_insert() is called with the idl_tx_list
 510  * to insert the conn onto.  conn_drain_insert() asserts flow control for the
 511  * sockets via su_txq_full() (non-STREAMS) or QFULL on conn_wq (STREAMS).
 512  * Further, conn_blocked is set to indicate that the conn is blocked.
 513  *
 514  * GLDv3 calls ill_flow_enable() when flow control is relieved.  The cookie
 515  * passed in the call to ill_flow_enable() identifies the blocked Tx ring and
 516  * is again hashed to locate the appropriate idl_tx_list, which is then
 517  * drained via conn_walk_drain().  conn_walk_drain() goes through each conn in
 518  * the drain list and calls conn_drain_remove() to clear flow control (via
 519  * calling su_txq_full() or clearing QFULL), and remove the conn from the
 520  * drain list.
 521  *
 522  * Note that the drain list is not a single list but a (configurable) array of
 523  * lists (8 elements by default).  Synchronization between drain insertion and
 524  * flow control wakeup is handled by using idl_txl->txl_lock, and only
 525  * conn_drain_insert() and conn_drain_remove() manipulate the drain list.
 526  *
 527  * Flow control via STREAMS is used when ILL_DIRECT_CAPABLE() returns FALSE.
 528  * On the send side, if the packet cannot be sent down to the driver by IP
 529  * (canput() fails), ip_xmit() drops the packet and returns EWOULDBLOCK to the
 530  * caller, who may then invoke ixa_check_drain_insert() to insert the conn on
 531  * the 0'th drain list.  When ip_wsrv() runs on the ill_wq because flow
 532  * control has been relieved, the blocked conns in the 0'th drain list are
 533  * drained as in the non-STREAMS case.
 534  *
 535  * In both the STREAMS and non-STREAMS cases, the sockfs upcall to set QFULL
 536  * is done when the conn is inserted into the drain list (conn_drain_insert())
 537  * and cleared when the conn is removed from the it (conn_drain_remove()).
 538  *
 539  * IPQOS notes:
 540  *
 541  * IPQoS Policies are applied to packets using IPPF (IP Policy framework)
 542  * and IPQoS modules. IPPF includes hooks in IP at different control points
 543  * (callout positions) which direct packets to IPQoS modules for policy
 544  * processing. Policies, if present, are global.
 545  *
 546  * The callout positions are located in the following paths:
 547  *              o local_in (packets destined for this host)
 548  *              o local_out (packets orginating from this host )
 549  *              o fwd_in  (packets forwarded by this m/c - inbound)
 550  *              o fwd_out (packets forwarded by this m/c - outbound)
 551  * Hooks at these callout points can be enabled/disabled using the ndd variable
 552  * ip_policy_mask (a bit mask with the 4 LSB indicating the callout positions).
 553  * By default all the callout positions are enabled.
 554  *
 555  * Outbound (local_out)
 556  * Hooks are placed in ire_send_wire_v4 and ire_send_wire_v6.
 557  *
 558  * Inbound (local_in)
 559  * Hooks are placed in ip_fanout_v4 and ip_fanout_v6.
 560  *
 561  * Forwarding (in and out)
 562  * Hooks are placed in ire_recv_forward_v4/v6.
 563  *
 564  * IP Policy Framework processing (IPPF processing)
 565  * Policy processing for a packet is initiated by ip_process, which ascertains
 566  * that the classifier (ipgpc) is loaded and configured, failing which the
 567  * packet resumes normal processing in IP. If the clasifier is present, the
 568  * packet is acted upon by one or more IPQoS modules (action instances), per
 569  * filters configured in ipgpc and resumes normal IP processing thereafter.
 570  * An action instance can drop a packet in course of its processing.
 571  *
 572  * Zones notes:
 573  *
 574  * The partitioning rules for networking are as follows:
 575  * 1) Packets coming from a zone must have a source address belonging to that
 576  * zone.
 577  * 2) Packets coming from a zone can only be sent on a physical interface on
 578  * which the zone has an IP address.
 579  * 3) Between two zones on the same machine, packet delivery is only allowed if
 580  * there's a matching route for the destination and zone in the forwarding
 581  * table.
 582  * 4) The TCP and UDP port spaces are per-zone; that is, two processes in
 583  * different zones can bind to the same port with the wildcard address
 584  * (INADDR_ANY).
 585  *
 586  * The granularity of interface partitioning is at the logical interface level.
 587  * Therefore, every zone has its own IP addresses, and incoming packets can be
 588  * attributed to a zone unambiguously. A logical interface is placed into a zone
 589  * using the SIOCSLIFZONE ioctl; this sets the ipif_zoneid field in the ipif_t
 590  * structure. Rule (1) is implemented by modifying the source address selection
 591  * algorithm so that the list of eligible addresses is filtered based on the
 592  * sending process zone.
 593  *
 594  * The Internet Routing Entries (IREs) are either exclusive to a zone or shared
 595  * across all zones, depending on their type. Here is the break-up:
 596  *
 597  * IRE type                             Shared/exclusive
 598  * --------                             ----------------
 599  * IRE_BROADCAST                        Exclusive
 600  * IRE_DEFAULT (default routes)         Shared (*)
 601  * IRE_LOCAL                            Exclusive (x)
 602  * IRE_LOOPBACK                         Exclusive
 603  * IRE_PREFIX (net routes)              Shared (*)
 604  * IRE_IF_NORESOLVER (interface routes) Exclusive
 605  * IRE_IF_RESOLVER (interface routes)   Exclusive
 606  * IRE_IF_CLONE (interface routes)      Exclusive
 607  * IRE_HOST (host routes)               Shared (*)
 608  *
 609  * (*) A zone can only use a default or off-subnet route if the gateway is
 610  * directly reachable from the zone, that is, if the gateway's address matches
 611  * one of the zone's logical interfaces.
 612  *
 613  * (x) IRE_LOCAL are handled a bit differently.
 614  * When ip_restrict_interzone_loopback is set (the default),
 615  * ire_route_recursive restricts loopback using an IRE_LOCAL
 616  * between zone to the case when L2 would have conceptually looped the packet
 617  * back, i.e. the loopback which is required since neither Ethernet drivers
 618  * nor Ethernet hardware loops them back. This is the case when the normal
 619  * routes (ignoring IREs with different zoneids) would send out the packet on
 620  * the same ill as the ill with which is IRE_LOCAL is associated.
 621  *
 622  * Multiple zones can share a common broadcast address; typically all zones
 623  * share the 255.255.255.255 address. Incoming as well as locally originated
 624  * broadcast packets must be dispatched to all the zones on the broadcast
 625  * network. For directed broadcasts (e.g. 10.16.72.255) this is not trivial
 626  * since some zones may not be on the 10.16.72/24 network. To handle this, each
 627  * zone has its own set of IRE_BROADCAST entries; then, broadcast packets are
 628  * sent to every zone that has an IRE_BROADCAST entry for the destination
 629  * address on the input ill, see ip_input_broadcast().
 630  *
 631  * Applications in different zones can join the same multicast group address.
 632  * The same logic applies for multicast as for broadcast. ip_input_multicast
 633  * dispatches packets to all zones that have members on the physical interface.
 634  */
 635 
 636 /*
 637  * Squeue Fanout flags:
 638  *      0: No fanout.
 639  *      1: Fanout across all squeues
 640  */
 641 boolean_t       ip_squeue_fanout = 0;
 642 
 643 /*
 644  * Maximum dups allowed per packet.
 645  */
 646 uint_t ip_max_frag_dups = 10;
 647 
 648 static int      ip_open(queue_t *q, dev_t *devp, int flag, int sflag,
 649                     cred_t *credp, boolean_t isv6);
 650 static mblk_t   *ip_xmit_attach_llhdr(mblk_t *, nce_t *);
 651 
 652 static boolean_t icmp_inbound_verify_v4(mblk_t *, icmph_t *, ip_recv_attr_t *);
 653 static void     icmp_inbound_too_big_v4(icmph_t *, ip_recv_attr_t *);
 654 static void     icmp_inbound_error_fanout_v4(mblk_t *, icmph_t *,
 655     ip_recv_attr_t *);
 656 static void     icmp_options_update(ipha_t *);
 657 static void     icmp_param_problem(mblk_t *, uint8_t,  ip_recv_attr_t *);
 658 static void     icmp_pkt(mblk_t *, void *, size_t, ip_recv_attr_t *);
 659 static mblk_t   *icmp_pkt_err_ok(mblk_t *, ip_recv_attr_t *);
 660 static void     icmp_redirect_v4(mblk_t *mp, ipha_t *, icmph_t *,
 661     ip_recv_attr_t *);
 662 static void     icmp_send_redirect(mblk_t *, ipaddr_t, ip_recv_attr_t *);
 663 static void     icmp_send_reply_v4(mblk_t *, ipha_t *, icmph_t *,
 664     ip_recv_attr_t *);
 665 
 666 mblk_t          *ip_dlpi_alloc(size_t, t_uscalar_t);
 667 char            *ip_dot_addr(ipaddr_t, char *);
 668 mblk_t          *ip_carve_mp(mblk_t **, ssize_t);
 669 static char     *ip_dot_saddr(uchar_t *, char *);
 670 static int      ip_lrput(queue_t *, mblk_t *);
 671 ipaddr_t        ip_net_mask(ipaddr_t);
 672 char            *ip_nv_lookup(nv_t *, int);
 673 int             ip_rput(queue_t *, mblk_t *);
 674 static void     ip_rput_dlpi_writer(ipsq_t *dummy_sq, queue_t *q, mblk_t *mp,
 675                     void *dummy_arg);
 676 int             ip_snmp_get(queue_t *, mblk_t *, int, boolean_t);
 677 static mblk_t   *ip_snmp_get_mib2_ip(queue_t *, mblk_t *,
 678                     mib2_ipIfStatsEntry_t *, ip_stack_t *, boolean_t);
 679 static mblk_t   *ip_snmp_get_mib2_ip_traffic_stats(queue_t *, mblk_t *,
 680                     ip_stack_t *, boolean_t);
 681 static mblk_t   *ip_snmp_get_mib2_ip6(queue_t *, mblk_t *, ip_stack_t *,
 682                     boolean_t);
 683 static mblk_t   *ip_snmp_get_mib2_icmp(queue_t *, mblk_t *, ip_stack_t *ipst);
 684 static mblk_t   *ip_snmp_get_mib2_icmp6(queue_t *, mblk_t *, ip_stack_t *ipst);
 685 static mblk_t   *ip_snmp_get_mib2_igmp(queue_t *, mblk_t *, ip_stack_t *ipst);
 686 static mblk_t   *ip_snmp_get_mib2_multi(queue_t *, mblk_t *, ip_stack_t *ipst);
 687 static mblk_t   *ip_snmp_get_mib2_ip_addr(queue_t *, mblk_t *,
 688                     ip_stack_t *ipst, boolean_t);
 689 static mblk_t   *ip_snmp_get_mib2_ip6_addr(queue_t *, mblk_t *,
 690                     ip_stack_t *ipst, boolean_t);
 691 static mblk_t   *ip_snmp_get_mib2_ip_group_src(queue_t *, mblk_t *,
 692                     ip_stack_t *ipst);
 693 static mblk_t   *ip_snmp_get_mib2_ip6_group_src(queue_t *, mblk_t *,
 694                     ip_stack_t *ipst);
 695 static mblk_t   *ip_snmp_get_mib2_ip_group_mem(queue_t *, mblk_t *,
 696                     ip_stack_t *ipst);
 697 static mblk_t   *ip_snmp_get_mib2_ip6_group_mem(queue_t *, mblk_t *,
 698                     ip_stack_t *ipst);
 699 static mblk_t   *ip_snmp_get_mib2_virt_multi(queue_t *, mblk_t *,
 700                     ip_stack_t *ipst);
 701 static mblk_t   *ip_snmp_get_mib2_multi_rtable(queue_t *, mblk_t *,
 702                     ip_stack_t *ipst);
 703 static mblk_t   *ip_snmp_get_mib2_ip_route_media(queue_t *, mblk_t *, int,
 704                     ip_stack_t *ipst);
 705 static mblk_t   *ip_snmp_get_mib2_ip6_route_media(queue_t *, mblk_t *, int,
 706                     ip_stack_t *ipst);
 707 static void     ip_snmp_get2_v4(ire_t *, iproutedata_t *);
 708 static void     ip_snmp_get2_v6_route(ire_t *, iproutedata_t *);
 709 static void     ip_snmp_get2_v4_media(ncec_t *, void *);
 710 static void     ip_snmp_get2_v6_media(ncec_t *, void *);
 711 int             ip_snmp_set(queue_t *, int, int, uchar_t *, int);
 712 
 713 static mblk_t   *ip_fragment_copyhdr(uchar_t *, int, int, ip_stack_t *,
 714                     mblk_t *);
 715 
 716 static void     conn_drain_init(ip_stack_t *);
 717 static void     conn_drain_fini(ip_stack_t *);
 718 static void     conn_drain(conn_t *connp, boolean_t closing);
 719 
 720 static void     conn_walk_drain(ip_stack_t *, idl_tx_list_t *);
 721 static void     conn_walk_sctp(pfv_t, void *, zoneid_t, netstack_t *);
 722 
 723 static void     *ip_stack_init(netstackid_t stackid, netstack_t *ns);
 724 static void     ip_stack_shutdown(netstackid_t stackid, void *arg);
 725 static void     ip_stack_fini(netstackid_t stackid, void *arg);
 726 
 727 static int      ip_multirt_apply_membership(int (*fn)(conn_t *, boolean_t,
 728     const in6_addr_t *, ipaddr_t, uint_t, mcast_record_t, const in6_addr_t *),
 729     ire_t *, conn_t *, boolean_t, const in6_addr_t *,  mcast_record_t,
 730     const in6_addr_t *);
 731 
 732 static int      ip_squeue_switch(int);
 733 
 734 static void     *ip_kstat_init(netstackid_t, ip_stack_t *);
 735 static void     ip_kstat_fini(netstackid_t, kstat_t *);
 736 static int      ip_kstat_update(kstat_t *kp, int rw);
 737 static void     *icmp_kstat_init(netstackid_t);
 738 static void     icmp_kstat_fini(netstackid_t, kstat_t *);
 739 static int      icmp_kstat_update(kstat_t *kp, int rw);
 740 static void     *ip_kstat2_init(netstackid_t, ip_stat_t *);
 741 static void     ip_kstat2_fini(netstackid_t, kstat_t *);
 742 
 743 static void     ipobs_init(ip_stack_t *);
 744 static void     ipobs_fini(ip_stack_t *);
 745 
 746 static int      ip_tp_cpu_update(cpu_setup_t, int, void *);
 747 
 748 ipaddr_t        ip_g_all_ones = IP_HOST_MASK;
 749 
 750 static long ip_rput_pullups;
 751 int     dohwcksum = 1;  /* use h/w cksum if supported by the hardware */
 752 
 753 vmem_t *ip_minor_arena_sa; /* for minor nos. from INET_MIN_DEV+2 thru 2^^18-1 */
 754 vmem_t *ip_minor_arena_la; /* for minor nos. from 2^^18 thru 2^^32-1 */
 755 
 756 int     ip_debug;
 757 
 758 /*
 759  * Multirouting/CGTP stuff
 760  */
 761 int     ip_cgtp_filter_rev = CGTP_FILTER_REV;   /* CGTP hooks version */
 762 
 763 /*
 764  * IP tunables related declarations. Definitions are in ip_tunables.c
 765  */
 766 extern mod_prop_info_t ip_propinfo_tbl[];
 767 extern int ip_propinfo_count;
 768 
 769 /*
 770  * Table of IP ioctls encoding the various properties of the ioctl and
 771  * indexed based on the last byte of the ioctl command. Occasionally there
 772  * is a clash, and there is more than 1 ioctl with the same last byte.
 773  * In such a case 1 ioctl is encoded in the ndx table and the remaining
 774  * ioctls are encoded in the misc table. An entry in the ndx table is
 775  * retrieved by indexing on the last byte of the ioctl command and comparing
 776  * the ioctl command with the value in the ndx table. In the event of a
 777  * mismatch the misc table is then searched sequentially for the desired
 778  * ioctl command.
 779  *
 780  * Entry: <command> <copyin_size> <flags> <cmd_type> <function> <restart_func>
 781  */
 782 ip_ioctl_cmd_t ip_ndx_ioctl_table[] = {
 783         /* 000 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 784         /* 001 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 785         /* 002 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 786         /* 003 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 787         /* 004 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 788         /* 005 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 789         /* 006 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 790         /* 007 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 791         /* 008 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 792         /* 009 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 793 
 794         /* 010 */ { SIOCADDRT,  sizeof (struct rtentry), IPI_PRIV,
 795                         MISC_CMD, ip_siocaddrt, NULL },
 796         /* 011 */ { SIOCDELRT,  sizeof (struct rtentry), IPI_PRIV,
 797                         MISC_CMD, ip_siocdelrt, NULL },
 798 
 799         /* 012 */ { SIOCSIFADDR, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
 800                         IF_CMD, ip_sioctl_addr, ip_sioctl_addr_restart },
 801         /* 013 */ { SIOCGIFADDR, sizeof (struct ifreq), IPI_GET_CMD,
 802                         IF_CMD, ip_sioctl_get_addr, NULL },
 803 
 804         /* 014 */ { SIOCSIFDSTADDR, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
 805                         IF_CMD, ip_sioctl_dstaddr, ip_sioctl_dstaddr_restart },
 806         /* 015 */ { SIOCGIFDSTADDR, sizeof (struct ifreq),
 807                         IPI_GET_CMD, IF_CMD, ip_sioctl_get_dstaddr, NULL },
 808 
 809         /* 016 */ { SIOCSIFFLAGS, sizeof (struct ifreq),
 810                         IPI_PRIV | IPI_WR,
 811                         IF_CMD, ip_sioctl_flags, ip_sioctl_flags_restart },
 812         /* 017 */ { SIOCGIFFLAGS, sizeof (struct ifreq),
 813                         IPI_MODOK | IPI_GET_CMD,
 814                         IF_CMD, ip_sioctl_get_flags, NULL },
 815 
 816         /* 018 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 817         /* 019 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 818 
 819         /* copyin size cannot be coded for SIOCGIFCONF */
 820         /* 020 */ { O_SIOCGIFCONF, 0, IPI_GET_CMD,
 821                         MISC_CMD, ip_sioctl_get_ifconf, NULL },
 822 
 823         /* 021 */ { SIOCSIFMTU, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
 824                         IF_CMD, ip_sioctl_mtu, NULL },
 825         /* 022 */ { SIOCGIFMTU, sizeof (struct ifreq), IPI_GET_CMD,
 826                         IF_CMD, ip_sioctl_get_mtu, NULL },
 827         /* 023 */ { SIOCGIFBRDADDR, sizeof (struct ifreq),
 828                         IPI_GET_CMD, IF_CMD, ip_sioctl_get_brdaddr, NULL },
 829         /* 024 */ { SIOCSIFBRDADDR, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
 830                         IF_CMD, ip_sioctl_brdaddr, NULL },
 831         /* 025 */ { SIOCGIFNETMASK, sizeof (struct ifreq),
 832                         IPI_GET_CMD, IF_CMD, ip_sioctl_get_netmask, NULL },
 833         /* 026 */ { SIOCSIFNETMASK, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
 834                         IF_CMD, ip_sioctl_netmask, ip_sioctl_netmask_restart },
 835         /* 027 */ { SIOCGIFMETRIC, sizeof (struct ifreq),
 836                         IPI_GET_CMD, IF_CMD, ip_sioctl_get_metric, NULL },
 837         /* 028 */ { SIOCSIFMETRIC, sizeof (struct ifreq), IPI_PRIV,
 838                         IF_CMD, ip_sioctl_metric, NULL },
 839         /* 029 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 840 
 841         /* See 166-168 below for extended SIOC*XARP ioctls */
 842         /* 030 */ { SIOCSARP, sizeof (struct arpreq), IPI_PRIV | IPI_WR,
 843                         ARP_CMD, ip_sioctl_arp, NULL },
 844         /* 031 */ { SIOCGARP, sizeof (struct arpreq), IPI_GET_CMD,
 845                         ARP_CMD, ip_sioctl_arp, NULL },
 846         /* 032 */ { SIOCDARP, sizeof (struct arpreq), IPI_PRIV | IPI_WR,
 847                         ARP_CMD, ip_sioctl_arp, NULL },
 848 
 849         /* 033 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 850         /* 034 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 851         /* 035 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 852         /* 036 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 853         /* 037 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 854         /* 038 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 855         /* 039 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 856         /* 040 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 857         /* 041 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 858         /* 042 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 859         /* 043 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 860         /* 044 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 861         /* 045 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 862         /* 046 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 863         /* 047 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 864         /* 048 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 865         /* 049 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 866         /* 050 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 867         /* 051 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 868         /* 052 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 869         /* 053 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 870 
 871         /* 054 */ { IF_UNITSEL, sizeof (int), IPI_PRIV | IPI_WR | IPI_MODOK,
 872                         MISC_CMD, if_unitsel, if_unitsel_restart },
 873 
 874         /* 055 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 875         /* 056 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 876         /* 057 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 877         /* 058 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 878         /* 059 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 879         /* 060 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 880         /* 061 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 881         /* 062 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 882         /* 063 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 883         /* 064 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 884         /* 065 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 885         /* 066 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 886         /* 067 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 887         /* 068 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 888         /* 069 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 889         /* 070 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 890         /* 071 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 891         /* 072 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 892 
 893         /* 073 */ { SIOCSIFNAME, sizeof (struct ifreq),
 894                         IPI_PRIV | IPI_WR | IPI_MODOK,
 895                         IF_CMD, ip_sioctl_sifname, NULL },
 896 
 897         /* 074 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 898         /* 075 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 899         /* 076 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 900         /* 077 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 901         /* 078 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 902         /* 079 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 903         /* 080 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 904         /* 081 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 905         /* 082 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 906         /* 083 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 907         /* 084 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 908         /* 085 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 909         /* 086 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 910 
 911         /* 087 */ { SIOCGIFNUM, sizeof (int), IPI_GET_CMD,
 912                         MISC_CMD, ip_sioctl_get_ifnum, NULL },
 913         /* 088 */ { SIOCGIFMUXID, sizeof (struct ifreq), IPI_GET_CMD,
 914                         IF_CMD, ip_sioctl_get_muxid, NULL },
 915         /* 089 */ { SIOCSIFMUXID, sizeof (struct ifreq),
 916                         IPI_PRIV | IPI_WR, IF_CMD, ip_sioctl_muxid, NULL },
 917 
 918         /* Both if and lif variants share same func */
 919         /* 090 */ { SIOCGIFINDEX, sizeof (struct ifreq), IPI_GET_CMD,
 920                         IF_CMD, ip_sioctl_get_lifindex, NULL },
 921         /* Both if and lif variants share same func */
 922         /* 091 */ { SIOCSIFINDEX, sizeof (struct ifreq),
 923                         IPI_PRIV | IPI_WR, IF_CMD, ip_sioctl_slifindex, NULL },
 924 
 925         /* copyin size cannot be coded for SIOCGIFCONF */
 926         /* 092 */ { SIOCGIFCONF, 0, IPI_GET_CMD,
 927                         MISC_CMD, ip_sioctl_get_ifconf, NULL },
 928         /* 093 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 929         /* 094 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 930         /* 095 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 931         /* 096 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 932         /* 097 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 933         /* 098 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 934         /* 099 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 935         /* 100 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 936         /* 101 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 937         /* 102 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 938         /* 103 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 939         /* 104 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 940         /* 105 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 941         /* 106 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 942         /* 107 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 943         /* 108 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 944         /* 109 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 945 
 946         /* 110 */ { SIOCLIFREMOVEIF, sizeof (struct lifreq),
 947                         IPI_PRIV | IPI_WR, LIF_CMD, ip_sioctl_removeif,
 948                         ip_sioctl_removeif_restart },
 949         /* 111 */ { SIOCLIFADDIF, sizeof (struct lifreq),
 950                         IPI_GET_CMD | IPI_PRIV | IPI_WR,
 951                         LIF_CMD, ip_sioctl_addif, NULL },
 952 #define SIOCLIFADDR_NDX 112
 953         /* 112 */ { SIOCSLIFADDR, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
 954                         LIF_CMD, ip_sioctl_addr, ip_sioctl_addr_restart },
 955         /* 113 */ { SIOCGLIFADDR, sizeof (struct lifreq),
 956                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_addr, NULL },
 957         /* 114 */ { SIOCSLIFDSTADDR, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
 958                         LIF_CMD, ip_sioctl_dstaddr, ip_sioctl_dstaddr_restart },
 959         /* 115 */ { SIOCGLIFDSTADDR, sizeof (struct lifreq),
 960                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_dstaddr, NULL },
 961         /* 116 */ { SIOCSLIFFLAGS, sizeof (struct lifreq),
 962                         IPI_PRIV | IPI_WR,
 963                         LIF_CMD, ip_sioctl_flags, ip_sioctl_flags_restart },
 964         /* 117 */ { SIOCGLIFFLAGS, sizeof (struct lifreq),
 965                         IPI_GET_CMD | IPI_MODOK,
 966                         LIF_CMD, ip_sioctl_get_flags, NULL },
 967 
 968         /* 118 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 969         /* 119 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 970 
 971         /* 120 */ { O_SIOCGLIFCONF, 0, IPI_GET_CMD, MISC_CMD,
 972                         ip_sioctl_get_lifconf, NULL },
 973         /* 121 */ { SIOCSLIFMTU, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
 974                         LIF_CMD, ip_sioctl_mtu, NULL },
 975         /* 122 */ { SIOCGLIFMTU, sizeof (struct lifreq), IPI_GET_CMD,
 976                         LIF_CMD, ip_sioctl_get_mtu, NULL },
 977         /* 123 */ { SIOCGLIFBRDADDR, sizeof (struct lifreq),
 978                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_brdaddr, NULL },
 979         /* 124 */ { SIOCSLIFBRDADDR, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
 980                         LIF_CMD, ip_sioctl_brdaddr, NULL },
 981         /* 125 */ { SIOCGLIFNETMASK, sizeof (struct lifreq),
 982                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_netmask, NULL },
 983         /* 126 */ { SIOCSLIFNETMASK, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
 984                         LIF_CMD, ip_sioctl_netmask, ip_sioctl_netmask_restart },
 985         /* 127 */ { SIOCGLIFMETRIC, sizeof (struct lifreq),
 986                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_metric, NULL },
 987         /* 128 */ { SIOCSLIFMETRIC, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
 988                         LIF_CMD, ip_sioctl_metric, NULL },
 989         /* 129 */ { SIOCSLIFNAME, sizeof (struct lifreq),
 990                         IPI_PRIV | IPI_WR | IPI_MODOK,
 991                         LIF_CMD, ip_sioctl_slifname,
 992                         ip_sioctl_slifname_restart },
 993 
 994         /* 130 */ { SIOCGLIFNUM, sizeof (struct lifnum), IPI_GET_CMD,
 995                         MISC_CMD, ip_sioctl_get_lifnum, NULL },
 996         /* 131 */ { SIOCGLIFMUXID, sizeof (struct lifreq),
 997                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_muxid, NULL },
 998         /* 132 */ { SIOCSLIFMUXID, sizeof (struct lifreq),
 999                         IPI_PRIV | IPI_WR, LIF_CMD, ip_sioctl_muxid, NULL },
1000         /* 133 */ { SIOCGLIFINDEX, sizeof (struct lifreq),
1001                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_lifindex, 0 },
1002         /* 134 */ { SIOCSLIFINDEX, sizeof (struct lifreq),
1003                         IPI_PRIV | IPI_WR, LIF_CMD, ip_sioctl_slifindex, 0 },
1004         /* 135 */ { SIOCSLIFTOKEN, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1005                         LIF_CMD, ip_sioctl_token, NULL },
1006         /* 136 */ { SIOCGLIFTOKEN, sizeof (struct lifreq),
1007                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_token, NULL },
1008         /* 137 */ { SIOCSLIFSUBNET, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1009                         LIF_CMD, ip_sioctl_subnet, ip_sioctl_subnet_restart },
1010         /* 138 */ { SIOCGLIFSUBNET, sizeof (struct lifreq),
1011                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_subnet, NULL },
1012         /* 139 */ { SIOCSLIFLNKINFO, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1013                         LIF_CMD, ip_sioctl_lnkinfo, NULL },
1014 
1015         /* 140 */ { SIOCGLIFLNKINFO, sizeof (struct lifreq),
1016                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_lnkinfo, NULL },
1017         /* 141 */ { SIOCLIFDELND, sizeof (struct lifreq), IPI_PRIV,
1018                         LIF_CMD, ip_siocdelndp_v6, NULL },
1019         /* 142 */ { SIOCLIFGETND, sizeof (struct lifreq), IPI_GET_CMD,
1020                         LIF_CMD, ip_siocqueryndp_v6, NULL },
1021         /* 143 */ { SIOCLIFSETND, sizeof (struct lifreq), IPI_PRIV,
1022                         LIF_CMD, ip_siocsetndp_v6, NULL },
1023         /* 144 */ { SIOCTMYADDR, sizeof (struct sioc_addrreq), IPI_GET_CMD,
1024                         MISC_CMD, ip_sioctl_tmyaddr, NULL },
1025         /* 145 */ { SIOCTONLINK, sizeof (struct sioc_addrreq), IPI_GET_CMD,
1026                         MISC_CMD, ip_sioctl_tonlink, NULL },
1027         /* 146 */ { SIOCTMYSITE, sizeof (struct sioc_addrreq), 0,
1028                         MISC_CMD, ip_sioctl_tmysite, NULL },
1029         /* 147 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1030         /* 148 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1031 
1032         /* Old *IPSECONFIG ioctls are now deprecated, now see spdsock.c */
1033         /* 149 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1034         /* 150 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1035         /* 151 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1036         /* 152 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1037 
1038         /* 153 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1039 
1040         /* 154 */ { SIOCGLIFBINDING, sizeof (struct lifreq), IPI_GET_CMD,
1041                         LIF_CMD, ip_sioctl_get_binding, NULL },
1042         /* 155 */ { SIOCSLIFGROUPNAME, sizeof (struct lifreq),
1043                         IPI_PRIV | IPI_WR,
1044                         LIF_CMD, ip_sioctl_groupname, ip_sioctl_groupname },
1045         /* 156 */ { SIOCGLIFGROUPNAME, sizeof (struct lifreq),
1046                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_groupname, NULL },
1047         /* 157 */ { SIOCGLIFGROUPINFO, sizeof (lifgroupinfo_t),
1048                         IPI_GET_CMD, MISC_CMD, ip_sioctl_groupinfo, NULL },
1049 
1050         /* Leave 158-160 unused; used to be SIOC*IFARP ioctls */
1051         /* 158 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1052         /* 159 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1053         /* 160 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1054 
1055         /* 161 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1056 
1057         /* These are handled in ip_sioctl_copyin_setup itself */
1058         /* 162 */ { SIOCGIP6ADDRPOLICY, 0, IPI_NULL_BCONT,
1059                         MISC_CMD, NULL, NULL },
1060         /* 163 */ { SIOCSIP6ADDRPOLICY, 0, IPI_PRIV | IPI_NULL_BCONT,
1061                         MISC_CMD, NULL, NULL },
1062         /* 164 */ { SIOCGDSTINFO, 0, IPI_GET_CMD, MISC_CMD, NULL, NULL },
1063 
1064         /* 165 */ { SIOCGLIFCONF, 0, IPI_GET_CMD, MISC_CMD,
1065                         ip_sioctl_get_lifconf, NULL },
1066 
1067         /* 166 */ { SIOCSXARP, sizeof (struct xarpreq), IPI_PRIV | IPI_WR,
1068                         XARP_CMD, ip_sioctl_arp, NULL },
1069         /* 167 */ { SIOCGXARP, sizeof (struct xarpreq), IPI_GET_CMD,
1070                         XARP_CMD, ip_sioctl_arp, NULL },
1071         /* 168 */ { SIOCDXARP, sizeof (struct xarpreq), IPI_PRIV | IPI_WR,
1072                         XARP_CMD, ip_sioctl_arp, NULL },
1073 
1074         /* SIOCPOPSOCKFS is not handled by IP */
1075         /* 169 */ { IPI_DONTCARE /* SIOCPOPSOCKFS */, 0, 0, 0, NULL, NULL },
1076 
1077         /* 170 */ { SIOCGLIFZONE, sizeof (struct lifreq),
1078                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_lifzone, NULL },
1079         /* 171 */ { SIOCSLIFZONE, sizeof (struct lifreq),
1080                         IPI_PRIV | IPI_WR, LIF_CMD, ip_sioctl_slifzone,
1081                         ip_sioctl_slifzone_restart },
1082         /* 172-174 are SCTP ioctls and not handled by IP */
1083         /* 172 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1084         /* 173 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1085         /* 174 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1086         /* 175 */ { SIOCGLIFUSESRC, sizeof (struct lifreq),
1087                         IPI_GET_CMD, LIF_CMD,
1088                         ip_sioctl_get_lifusesrc, 0 },
1089         /* 176 */ { SIOCSLIFUSESRC, sizeof (struct lifreq),
1090                         IPI_PRIV | IPI_WR,
1091                         LIF_CMD, ip_sioctl_slifusesrc,
1092                         NULL },
1093         /* 177 */ { SIOCGLIFSRCOF, 0, IPI_GET_CMD, MISC_CMD,
1094                         ip_sioctl_get_lifsrcof, NULL },
1095         /* 178 */ { SIOCGMSFILTER, sizeof (struct group_filter), IPI_GET_CMD,
1096                         MSFILT_CMD, ip_sioctl_msfilter, NULL },
1097         /* 179 */ { SIOCSMSFILTER, sizeof (struct group_filter), 0,
1098                         MSFILT_CMD, ip_sioctl_msfilter, NULL },
1099         /* 180 */ { SIOCGIPMSFILTER, sizeof (struct ip_msfilter), IPI_GET_CMD,
1100                         MSFILT_CMD, ip_sioctl_msfilter, NULL },
1101         /* 181 */ { SIOCSIPMSFILTER, sizeof (struct ip_msfilter), 0,
1102                         MSFILT_CMD, ip_sioctl_msfilter, NULL },
1103         /* 182 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1104         /* SIOCSENABLESDP is handled by SDP */
1105         /* 183 */ { IPI_DONTCARE /* SIOCSENABLESDP */, 0, 0, 0, NULL, NULL },
1106         /* 184 */ { IPI_DONTCARE /* SIOCSQPTR */, 0, 0, 0, NULL, NULL },
1107         /* 185 */ { SIOCGIFHWADDR, sizeof (struct ifreq), IPI_GET_CMD,
1108                         IF_CMD, ip_sioctl_get_ifhwaddr, NULL },
1109         /* 186 */ { IPI_DONTCARE /* SIOCGSTAMP */, 0, 0, 0, NULL, NULL },
1110         /* 187 */ { SIOCILB, 0, IPI_PRIV | IPI_GET_CMD, MISC_CMD,
1111                         ip_sioctl_ilb_cmd, NULL },
1112         /* 188 */ { SIOCGETPROP, 0, IPI_GET_CMD, 0, NULL, NULL },
1113         /* 189 */ { SIOCSETPROP, 0, IPI_PRIV | IPI_WR, 0, NULL, NULL},
1114         /* 190 */ { SIOCGLIFDADSTATE, sizeof (struct lifreq),
1115                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_dadstate, NULL },
1116         /* 191 */ { SIOCSLIFPREFIX, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1117                         LIF_CMD, ip_sioctl_prefix, ip_sioctl_prefix_restart },
1118         /* 192 */ { SIOCGLIFHWADDR, sizeof (struct lifreq), IPI_GET_CMD,
1119                         LIF_CMD, ip_sioctl_get_lifhwaddr, NULL }
1120 };
1121 
1122 int ip_ndx_ioctl_count = sizeof (ip_ndx_ioctl_table) / sizeof (ip_ioctl_cmd_t);
1123 
1124 ip_ioctl_cmd_t ip_misc_ioctl_table[] = {
1125         { I_LINK,       0, IPI_PRIV | IPI_WR, 0, NULL, NULL },
1126         { I_UNLINK,     0, IPI_PRIV | IPI_WR, 0, NULL, NULL },
1127         { I_PLINK,      0, IPI_PRIV | IPI_WR, 0, NULL, NULL },
1128         { I_PUNLINK,    0, IPI_PRIV | IPI_WR, 0, NULL, NULL },
1129         { ND_GET,       0, 0, 0, NULL, NULL },
1130         { ND_SET,       0, IPI_PRIV | IPI_WR, 0, NULL, NULL },
1131         { IP_IOCTL,     0, 0, 0, NULL, NULL },
1132         { SIOCGETVIFCNT, sizeof (struct sioc_vif_req), IPI_GET_CMD,
1133                 MISC_CMD, mrt_ioctl},
1134         { SIOCGETSGCNT, sizeof (struct sioc_sg_req), IPI_GET_CMD,
1135                 MISC_CMD, mrt_ioctl},
1136         { SIOCGETLSGCNT, sizeof (struct sioc_lsg_req), IPI_GET_CMD,
1137                 MISC_CMD, mrt_ioctl}
1138 };
1139 
1140 int ip_misc_ioctl_count =
1141     sizeof (ip_misc_ioctl_table) / sizeof (ip_ioctl_cmd_t);
1142 
1143 int     conn_drain_nthreads;            /* Number of drainers reqd. */
1144                                         /* Settable in /etc/system */
1145 /* Defined in ip_ire.c */
1146 extern uint32_t ip_ire_max_bucket_cnt, ip6_ire_max_bucket_cnt;
1147 extern uint32_t ip_ire_min_bucket_cnt, ip6_ire_min_bucket_cnt;
1148 extern uint32_t ip_ire_mem_ratio, ip_ire_cpu_ratio;
1149 
1150 static nv_t     ire_nv_arr[] = {
1151         { IRE_BROADCAST, "BROADCAST" },
1152         { IRE_LOCAL, "LOCAL" },
1153         { IRE_LOOPBACK, "LOOPBACK" },
1154         { IRE_DEFAULT, "DEFAULT" },
1155         { IRE_PREFIX, "PREFIX" },
1156         { IRE_IF_NORESOLVER, "IF_NORESOL" },
1157         { IRE_IF_RESOLVER, "IF_RESOLV" },
1158         { IRE_IF_CLONE, "IF_CLONE" },
1159         { IRE_HOST, "HOST" },
1160         { IRE_MULTICAST, "MULTICAST" },
1161         { IRE_NOROUTE, "NOROUTE" },
1162         { 0 }
1163 };
1164 
1165 nv_t    *ire_nv_tbl = ire_nv_arr;
1166 
1167 /* Simple ICMP IP Header Template */
1168 static ipha_t icmp_ipha = {
1169         IP_SIMPLE_HDR_VERSION, 0, 0, 0, 0, 0, IPPROTO_ICMP
1170 };
1171 
1172 struct module_info ip_mod_info = {
1173         IP_MOD_ID, IP_MOD_NAME, IP_MOD_MINPSZ, IP_MOD_MAXPSZ, IP_MOD_HIWAT,
1174         IP_MOD_LOWAT
1175 };
1176 
1177 /*
1178  * Duplicate static symbols within a module confuses mdb; so we avoid the
1179  * problem by making the symbols here distinct from those in udp.c.
1180  */
1181 
1182 /*
1183  * Entry points for IP as a device and as a module.
1184  * We have separate open functions for the /dev/ip and /dev/ip6 devices.
1185  */
1186 static struct qinit iprinitv4 = {
1187         ip_rput, NULL, ip_openv4, ip_close, NULL, &ip_mod_info
1188 };
1189 
1190 struct qinit iprinitv6 = {
1191         ip_rput_v6, NULL, ip_openv6, ip_close, NULL, &ip_mod_info
1192 };
1193 
1194 static struct qinit ipwinit = {
1195         ip_wput_nondata, ip_wsrv, NULL, NULL, NULL, &ip_mod_info
1196 };
1197 
1198 static struct qinit iplrinit = {
1199         ip_lrput, NULL, ip_openv4, ip_close, NULL, &ip_mod_info
1200 };
1201 
1202 static struct qinit iplwinit = {
1203         ip_lwput, NULL, NULL, NULL, NULL, &ip_mod_info
1204 };
1205 
1206 /* For AF_INET aka /dev/ip */
1207 struct streamtab ipinfov4 = {
1208         &iprinitv4, &ipwinit, &iplrinit, &iplwinit
1209 };
1210 
1211 /* For AF_INET6 aka /dev/ip6 */
1212 struct streamtab ipinfov6 = {
1213         &iprinitv6, &ipwinit, &iplrinit, &iplwinit
1214 };
1215 
1216 #ifdef  DEBUG
1217 boolean_t skip_sctp_cksum = B_FALSE;
1218 #endif
1219 
1220 /*
1221  * Generate an ICMP fragmentation needed message.
1222  * When called from ip_output side a minimal ip_recv_attr_t needs to be
1223  * constructed by the caller.
1224  */
1225 void
1226 icmp_frag_needed(mblk_t *mp, int mtu, ip_recv_attr_t *ira)
1227 {
1228         icmph_t icmph;
1229         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
1230 
1231         mp = icmp_pkt_err_ok(mp, ira);
1232         if (mp == NULL)
1233                 return;
1234 
1235         bzero(&icmph, sizeof (icmph_t));
1236         icmph.icmph_type = ICMP_DEST_UNREACHABLE;
1237         icmph.icmph_code = ICMP_FRAGMENTATION_NEEDED;
1238         icmph.icmph_du_mtu = htons((uint16_t)mtu);
1239         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutFragNeeded);
1240         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDestUnreachs);
1241 
1242         icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
1243 }
1244 
1245 /*
1246  * icmp_inbound_v4 deals with ICMP messages that are handled by IP.
1247  * If the ICMP message is consumed by IP, i.e., it should not be delivered
1248  * to any IPPROTO_ICMP raw sockets, then it returns NULL.
1249  * Likewise, if the ICMP error is misformed (too short, etc), then it
1250  * returns NULL. The caller uses this to determine whether or not to send
1251  * to raw sockets.
1252  *
1253  * All error messages are passed to the matching transport stream.
1254  *
1255  * The following cases are handled by icmp_inbound:
1256  * 1) It needs to send a reply back and possibly delivering it
1257  *    to the "interested" upper clients.
1258  * 2) Return the mblk so that the caller can pass it to the RAW socket clients.
1259  * 3) It needs to change some values in IP only.
1260  * 4) It needs to change some values in IP and upper layers e.g TCP
1261  *    by delivering an error to the upper layers.
1262  *
1263  * We handle the above three cases in the context of IPsec in the
1264  * following way :
1265  *
1266  * 1) Send the reply back in the same way as the request came in.
1267  *    If it came in encrypted, it goes out encrypted. If it came in
1268  *    clear, it goes out in clear. Thus, this will prevent chosen
1269  *    plain text attack.
1270  * 2) The client may or may not expect things to come in secure.
1271  *    If it comes in secure, the policy constraints are checked
1272  *    before delivering it to the upper layers. If it comes in
1273  *    clear, ipsec_inbound_accept_clear will decide whether to
1274  *    accept this in clear or not. In both the cases, if the returned
1275  *    message (IP header + 8 bytes) that caused the icmp message has
1276  *    AH/ESP headers, it is sent up to AH/ESP for validation before
1277  *    sending up. If there are only 8 bytes of returned message, then
1278  *    upper client will not be notified.
1279  * 3) Check with global policy to see whether it matches the constaints.
1280  *    But this will be done only if icmp_accept_messages_in_clear is
1281  *    zero.
1282  * 4) If we need to change both in IP and ULP, then the decision taken
1283  *    while affecting the values in IP and while delivering up to TCP
1284  *    should be the same.
1285  *
1286  *      There are two cases.
1287  *
1288  *      a) If we reject data at the IP layer (ipsec_check_global_policy()
1289  *         failed), we will not deliver it to the ULP, even though they
1290  *         are *willing* to accept in *clear*. This is fine as our global
1291  *         disposition to icmp messages asks us reject the datagram.
1292  *
1293  *      b) If we accept data at the IP layer (ipsec_check_global_policy()
1294  *         succeeded or icmp_accept_messages_in_clear is 1), and not able
1295  *         to deliver it to ULP (policy failed), it can lead to
1296  *         consistency problems. The cases known at this time are
1297  *         ICMP_DESTINATION_UNREACHABLE  messages with following code
1298  *         values :
1299  *
1300  *         - ICMP_FRAGMENTATION_NEEDED : IP adapts to the new value
1301  *           and Upper layer rejects. Then the communication will
1302  *           come to a stop. This is solved by making similar decisions
1303  *           at both levels. Currently, when we are unable to deliver
1304  *           to the Upper Layer (due to policy failures) while IP has
1305  *           adjusted dce_pmtu, the next outbound datagram would
1306  *           generate a local ICMP_FRAGMENTATION_NEEDED message - which
1307  *           will be with the right level of protection. Thus the right
1308  *           value will be communicated even if we are not able to
1309  *           communicate when we get from the wire initially. But this
1310  *           assumes there would be at least one outbound datagram after
1311  *           IP has adjusted its dce_pmtu value. To make things
1312  *           simpler, we accept in clear after the validation of
1313  *           AH/ESP headers.
1314  *
1315  *         - Other ICMP ERRORS : We may not be able to deliver it to the
1316  *           upper layer depending on the level of protection the upper
1317  *           layer expects and the disposition in ipsec_inbound_accept_clear().
1318  *           ipsec_inbound_accept_clear() decides whether a given ICMP error
1319  *           should be accepted in clear when the Upper layer expects secure.
1320  *           Thus the communication may get aborted by some bad ICMP
1321  *           packets.
1322  */
1323 mblk_t *
1324 icmp_inbound_v4(mblk_t *mp, ip_recv_attr_t *ira)
1325 {
1326         icmph_t         *icmph;
1327         ipha_t          *ipha;          /* Outer header */
1328         int             ip_hdr_length;  /* Outer header length */
1329         boolean_t       interested;
1330         ipif_t          *ipif;
1331         uint32_t        ts;
1332         uint32_t        *tsp;
1333         timestruc_t     now;
1334         ill_t           *ill = ira->ira_ill;
1335         ip_stack_t      *ipst = ill->ill_ipst;
1336         zoneid_t        zoneid = ira->ira_zoneid;
1337         int             len_needed;
1338         mblk_t          *mp_ret = NULL;
1339 
1340         ipha = (ipha_t *)mp->b_rptr;
1341 
1342         BUMP_MIB(&ipst->ips_icmp_mib, icmpInMsgs);
1343 
1344         ip_hdr_length = ira->ira_ip_hdr_length;
1345         if ((mp->b_wptr - mp->b_rptr) < (ip_hdr_length + ICMPH_SIZE)) {
1346                 if (ira->ira_pktlen < (ip_hdr_length + ICMPH_SIZE)) {
1347                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
1348                         ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
1349                         freemsg(mp);
1350                         return (NULL);
1351                 }
1352                 /* Last chance to get real. */
1353                 ipha = ip_pullup(mp, ip_hdr_length + ICMPH_SIZE, ira);
1354                 if (ipha == NULL) {
1355                         BUMP_MIB(&ipst->ips_icmp_mib, icmpInErrors);
1356                         freemsg(mp);
1357                         return (NULL);
1358                 }
1359         }
1360 
1361         /* The IP header will always be a multiple of four bytes */
1362         icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1363         ip2dbg(("icmp_inbound_v4: type %d code %d\n", icmph->icmph_type,
1364             icmph->icmph_code));
1365 
1366         /*
1367          * We will set "interested" to "true" if we should pass a copy to
1368          * the transport or if we handle the packet locally.
1369          */
1370         interested = B_FALSE;
1371         switch (icmph->icmph_type) {
1372         case ICMP_ECHO_REPLY:
1373                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInEchoReps);
1374                 break;
1375         case ICMP_DEST_UNREACHABLE:
1376                 if (icmph->icmph_code == ICMP_FRAGMENTATION_NEEDED)
1377                         BUMP_MIB(&ipst->ips_icmp_mib, icmpInFragNeeded);
1378                 interested = B_TRUE;    /* Pass up to transport */
1379                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInDestUnreachs);
1380                 break;
1381         case ICMP_SOURCE_QUENCH:
1382                 interested = B_TRUE;    /* Pass up to transport */
1383                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInSrcQuenchs);
1384                 break;
1385         case ICMP_REDIRECT:
1386                 if (!ipst->ips_ip_ignore_redirect)
1387                         interested = B_TRUE;
1388                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInRedirects);
1389                 break;
1390         case ICMP_ECHO_REQUEST:
1391                 /*
1392                  * Whether to respond to echo requests that come in as IP
1393                  * broadcasts or as IP multicast is subject to debate
1394                  * (what isn't?).  We aim to please, you pick it.
1395                  * Default is do it.
1396                  */
1397                 if (ira->ira_flags & IRAF_MULTICAST) {
1398                         /* multicast: respond based on tunable */
1399                         interested = ipst->ips_ip_g_resp_to_echo_mcast;
1400                 } else if (ira->ira_flags & IRAF_BROADCAST) {
1401                         /* broadcast: respond based on tunable */
1402                         interested = ipst->ips_ip_g_resp_to_echo_bcast;
1403                 } else {
1404                         /* unicast: always respond */
1405                         interested = B_TRUE;
1406                 }
1407                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInEchos);
1408                 if (!interested) {
1409                         /* We never pass these to RAW sockets */
1410                         freemsg(mp);
1411                         return (NULL);
1412                 }
1413 
1414                 /* Check db_ref to make sure we can modify the packet. */
1415                 if (mp->b_datap->db_ref > 1) {
1416                         mblk_t  *mp1;
1417 
1418                         mp1 = copymsg(mp);
1419                         freemsg(mp);
1420                         if (!mp1) {
1421                                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
1422                                 return (NULL);
1423                         }
1424                         mp = mp1;
1425                         ipha = (ipha_t *)mp->b_rptr;
1426                         icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1427                 }
1428                 icmph->icmph_type = ICMP_ECHO_REPLY;
1429                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutEchoReps);
1430                 icmp_send_reply_v4(mp, ipha, icmph, ira);
1431                 return (NULL);
1432 
1433         case ICMP_ROUTER_ADVERTISEMENT:
1434         case ICMP_ROUTER_SOLICITATION:
1435                 break;
1436         case ICMP_TIME_EXCEEDED:
1437                 interested = B_TRUE;    /* Pass up to transport */
1438                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInTimeExcds);
1439                 break;
1440         case ICMP_PARAM_PROBLEM:
1441                 interested = B_TRUE;    /* Pass up to transport */
1442                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInParmProbs);
1443                 break;
1444         case ICMP_TIME_STAMP_REQUEST:
1445                 /* Response to Time Stamp Requests is local policy. */
1446                 if (ipst->ips_ip_g_resp_to_timestamp) {
1447                         if (ira->ira_flags & IRAF_MULTIBROADCAST)
1448                                 interested =
1449                                     ipst->ips_ip_g_resp_to_timestamp_bcast;
1450                         else
1451                                 interested = B_TRUE;
1452                 }
1453                 if (!interested) {
1454                         /* We never pass these to RAW sockets */
1455                         freemsg(mp);
1456                         return (NULL);
1457                 }
1458 
1459                 /* Make sure we have enough of the packet */
1460                 len_needed = ip_hdr_length + ICMPH_SIZE +
1461                     3 * sizeof (uint32_t);
1462 
1463                 if (mp->b_wptr - mp->b_rptr < len_needed) {
1464                         ipha = ip_pullup(mp, len_needed, ira);
1465                         if (ipha == NULL) {
1466                                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1467                                 ip_drop_input("ipIfStatsInDiscards - ip_pullup",
1468                                     mp, ill);
1469                                 freemsg(mp);
1470                                 return (NULL);
1471                         }
1472                         /* Refresh following the pullup. */
1473                         icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1474                 }
1475                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInTimestamps);
1476                 /* Check db_ref to make sure we can modify the packet. */
1477                 if (mp->b_datap->db_ref > 1) {
1478                         mblk_t  *mp1;
1479 
1480                         mp1 = copymsg(mp);
1481                         freemsg(mp);
1482                         if (!mp1) {
1483                                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
1484                                 return (NULL);
1485                         }
1486                         mp = mp1;
1487                         ipha = (ipha_t *)mp->b_rptr;
1488                         icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1489                 }
1490                 icmph->icmph_type = ICMP_TIME_STAMP_REPLY;
1491                 tsp = (uint32_t *)&icmph[1];
1492                 tsp++;          /* Skip past 'originate time' */
1493                 /* Compute # of milliseconds since midnight */
1494                 gethrestime(&now);
1495                 ts = (now.tv_sec % (24 * 60 * 60)) * 1000 +
1496                     NSEC2MSEC(now.tv_nsec);
1497                 *tsp++ = htonl(ts);     /* Lay in 'receive time' */
1498                 *tsp++ = htonl(ts);     /* Lay in 'send time' */
1499                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutTimestampReps);
1500                 icmp_send_reply_v4(mp, ipha, icmph, ira);
1501                 return (NULL);
1502 
1503         case ICMP_TIME_STAMP_REPLY:
1504                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInTimestampReps);
1505                 break;
1506         case ICMP_INFO_REQUEST:
1507                 /* Per RFC 1122 3.2.2.7, ignore this. */
1508         case ICMP_INFO_REPLY:
1509                 break;
1510         case ICMP_ADDRESS_MASK_REQUEST:
1511                 if (ira->ira_flags & IRAF_MULTIBROADCAST) {
1512                         interested =
1513                             ipst->ips_ip_respond_to_address_mask_broadcast;
1514                 } else {
1515                         interested = B_TRUE;
1516                 }
1517                 if (!interested) {
1518                         /* We never pass these to RAW sockets */
1519                         freemsg(mp);
1520                         return (NULL);
1521                 }
1522                 len_needed = ip_hdr_length + ICMPH_SIZE + IP_ADDR_LEN;
1523                 if (mp->b_wptr - mp->b_rptr < len_needed) {
1524                         ipha = ip_pullup(mp, len_needed, ira);
1525                         if (ipha == NULL) {
1526                                 BUMP_MIB(ill->ill_ip_mib,
1527                                     ipIfStatsInTruncatedPkts);
1528                                 ip_drop_input("ipIfStatsInTruncatedPkts", mp,
1529                                     ill);
1530                                 freemsg(mp);
1531                                 return (NULL);
1532                         }
1533                         /* Refresh following the pullup. */
1534                         icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1535                 }
1536                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInAddrMasks);
1537                 /* Check db_ref to make sure we can modify the packet. */
1538                 if (mp->b_datap->db_ref > 1) {
1539                         mblk_t  *mp1;
1540 
1541                         mp1 = copymsg(mp);
1542                         freemsg(mp);
1543                         if (!mp1) {
1544                                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
1545                                 return (NULL);
1546                         }
1547                         mp = mp1;
1548                         ipha = (ipha_t *)mp->b_rptr;
1549                         icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1550                 }
1551                 /*
1552                  * Need the ipif with the mask be the same as the source
1553                  * address of the mask reply. For unicast we have a specific
1554                  * ipif. For multicast/broadcast we only handle onlink
1555                  * senders, and use the source address to pick an ipif.
1556                  */
1557                 ipif = ipif_lookup_addr(ipha->ipha_dst, ill, zoneid, ipst);
1558                 if (ipif == NULL) {
1559                         /* Broadcast or multicast */
1560                         ipif = ipif_lookup_remote(ill, ipha->ipha_src, zoneid);
1561                         if (ipif == NULL) {
1562                                 freemsg(mp);
1563                                 return (NULL);
1564                         }
1565                 }
1566                 icmph->icmph_type = ICMP_ADDRESS_MASK_REPLY;
1567                 bcopy(&ipif->ipif_net_mask, &icmph[1], IP_ADDR_LEN);
1568                 ipif_refrele(ipif);
1569                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutAddrMaskReps);
1570                 icmp_send_reply_v4(mp, ipha, icmph, ira);
1571                 return (NULL);
1572 
1573         case ICMP_ADDRESS_MASK_REPLY:
1574                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInAddrMaskReps);
1575                 break;
1576         default:
1577                 interested = B_TRUE;    /* Pass up to transport */
1578                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInUnknowns);
1579                 break;
1580         }
1581         /*
1582          * See if there is an ICMP client to avoid an extra copymsg/freemsg
1583          * if there isn't one.
1584          */
1585         if (ipst->ips_ipcl_proto_fanout_v4[IPPROTO_ICMP].connf_head != NULL) {
1586                 /* If there is an ICMP client and we want one too, copy it. */
1587 
1588                 if (!interested) {
1589                         /* Caller will deliver to RAW sockets */
1590                         return (mp);
1591                 }
1592                 mp_ret = copymsg(mp);
1593                 if (mp_ret == NULL) {
1594                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1595                         ip_drop_input("ipIfStatsInDiscards - copymsg", mp, ill);
1596                 }
1597         } else if (!interested) {
1598                 /* Neither we nor raw sockets are interested. Drop packet now */
1599                 freemsg(mp);
1600                 return (NULL);
1601         }
1602 
1603         /*
1604          * ICMP error or redirect packet. Make sure we have enough of
1605          * the header and that db_ref == 1 since we might end up modifying
1606          * the packet.
1607          */
1608         if (mp->b_cont != NULL) {
1609                 if (ip_pullup(mp, -1, ira) == NULL) {
1610                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1611                         ip_drop_input("ipIfStatsInDiscards - ip_pullup",
1612                             mp, ill);
1613                         freemsg(mp);
1614                         return (mp_ret);
1615                 }
1616         }
1617 
1618         if (mp->b_datap->db_ref > 1) {
1619                 mblk_t  *mp1;
1620 
1621                 mp1 = copymsg(mp);
1622                 if (mp1 == NULL) {
1623                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1624                         ip_drop_input("ipIfStatsInDiscards - copymsg", mp, ill);
1625                         freemsg(mp);
1626                         return (mp_ret);
1627                 }
1628                 freemsg(mp);
1629                 mp = mp1;
1630         }
1631 
1632         /*
1633          * In case mp has changed, verify the message before any further
1634          * processes.
1635          */
1636         ipha = (ipha_t *)mp->b_rptr;
1637         icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1638         if (!icmp_inbound_verify_v4(mp, icmph, ira)) {
1639                 freemsg(mp);
1640                 return (mp_ret);
1641         }
1642 
1643         switch (icmph->icmph_type) {
1644         case ICMP_REDIRECT:
1645                 icmp_redirect_v4(mp, ipha, icmph, ira);
1646                 break;
1647         case ICMP_DEST_UNREACHABLE:
1648                 if (icmph->icmph_code == ICMP_FRAGMENTATION_NEEDED) {
1649                         /* Update DCE and adjust MTU is icmp header if needed */
1650                         icmp_inbound_too_big_v4(icmph, ira);
1651                 }
1652                 /* FALLTHROUGH */
1653         default:
1654                 icmp_inbound_error_fanout_v4(mp, icmph, ira);
1655                 break;
1656         }
1657         return (mp_ret);
1658 }
1659 
1660 /*
1661  * Send an ICMP echo, timestamp or address mask reply.
1662  * The caller has already updated the payload part of the packet.
1663  * We handle the ICMP checksum, IP source address selection and feed
1664  * the packet into ip_output_simple.
1665  */
1666 static void
1667 icmp_send_reply_v4(mblk_t *mp, ipha_t *ipha, icmph_t *icmph,
1668     ip_recv_attr_t *ira)
1669 {
1670         uint_t          ip_hdr_length = ira->ira_ip_hdr_length;
1671         ill_t           *ill = ira->ira_ill;
1672         ip_stack_t      *ipst = ill->ill_ipst;
1673         ip_xmit_attr_t  ixas;
1674 
1675         /* Send out an ICMP packet */
1676         icmph->icmph_checksum = 0;
1677         icmph->icmph_checksum = IP_CSUM(mp, ip_hdr_length, 0);
1678         /* Reset time to live. */
1679         ipha->ipha_ttl = ipst->ips_ip_def_ttl;
1680         {
1681                 /* Swap source and destination addresses */
1682                 ipaddr_t tmp;
1683 
1684                 tmp = ipha->ipha_src;
1685                 ipha->ipha_src = ipha->ipha_dst;
1686                 ipha->ipha_dst = tmp;
1687         }
1688         ipha->ipha_ident = 0;
1689         if (!IS_SIMPLE_IPH(ipha))
1690                 icmp_options_update(ipha);
1691 
1692         bzero(&ixas, sizeof (ixas));
1693         ixas.ixa_flags = IXAF_BASIC_SIMPLE_V4;
1694         ixas.ixa_zoneid = ira->ira_zoneid;
1695         ixas.ixa_cred = kcred;
1696         ixas.ixa_cpid = NOPID;
1697         ixas.ixa_tsl = ira->ira_tsl; /* Behave as a multi-level responder */
1698         ixas.ixa_ifindex = 0;
1699         ixas.ixa_ipst = ipst;
1700         ixas.ixa_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1701 
1702         if (!(ira->ira_flags & IRAF_IPSEC_SECURE)) {
1703                 /*
1704                  * This packet should go out the same way as it
1705                  * came in i.e in clear, independent of the IPsec policy
1706                  * for transmitting packets.
1707                  */
1708                 ixas.ixa_flags |= IXAF_NO_IPSEC;
1709         } else {
1710                 if (!ipsec_in_to_out(ira, &ixas, mp, ipha, NULL)) {
1711                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1712                         /* Note: mp already consumed and ip_drop_packet done */
1713                         return;
1714                 }
1715         }
1716         if (ira->ira_flags & IRAF_MULTIBROADCAST) {
1717                 /*
1718                  * Not one or our addresses (IRE_LOCALs), thus we let
1719                  * ip_output_simple pick the source.
1720                  */
1721                 ipha->ipha_src = INADDR_ANY;
1722                 ixas.ixa_flags |= IXAF_SET_SOURCE;
1723         }
1724         /* Should we send with DF and use dce_pmtu? */
1725         if (ipst->ips_ipv4_icmp_return_pmtu) {
1726                 ixas.ixa_flags |= IXAF_PMTU_DISCOVERY;
1727                 ipha->ipha_fragment_offset_and_flags |= IPH_DF_HTONS;
1728         }
1729 
1730         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutMsgs);
1731 
1732         (void) ip_output_simple(mp, &ixas);
1733         ixa_cleanup(&ixas);
1734 }
1735 
1736 /*
1737  * Verify the ICMP messages for either for ICMP error or redirect packet.
1738  * The caller should have fully pulled up the message. If it's a redirect
1739  * packet, only basic checks on IP header will be done; otherwise, verify
1740  * the packet by looking at the included ULP header.
1741  *
1742  * Called before icmp_inbound_error_fanout_v4 is called.
1743  */
1744 static boolean_t
1745 icmp_inbound_verify_v4(mblk_t *mp, icmph_t *icmph, ip_recv_attr_t *ira)
1746 {
1747         ill_t           *ill = ira->ira_ill;
1748         int             hdr_length;
1749         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
1750         conn_t          *connp;
1751         ipha_t          *ipha;  /* Inner IP header */
1752 
1753         ipha = (ipha_t *)&icmph[1];
1754         if ((uchar_t *)ipha + IP_SIMPLE_HDR_LENGTH > mp->b_wptr)
1755                 goto truncated;
1756 
1757         hdr_length = IPH_HDR_LENGTH(ipha);
1758 
1759         if ((IPH_HDR_VERSION(ipha) != IPV4_VERSION))
1760                 goto discard_pkt;
1761 
1762         if (hdr_length < sizeof (ipha_t))
1763                 goto truncated;
1764 
1765         if ((uchar_t *)ipha + hdr_length > mp->b_wptr)
1766                 goto truncated;
1767 
1768         /*
1769          * Stop here for ICMP_REDIRECT.
1770          */
1771         if (icmph->icmph_type == ICMP_REDIRECT)
1772                 return (B_TRUE);
1773 
1774         /*
1775          * ICMP errors only.
1776          */
1777         switch (ipha->ipha_protocol) {
1778         case IPPROTO_UDP:
1779                 /*
1780                  * Verify we have at least ICMP_MIN_TP_HDR_LEN bytes of
1781                  * transport header.
1782                  */
1783                 if ((uchar_t *)ipha + hdr_length + ICMP_MIN_TP_HDR_LEN >
1784                     mp->b_wptr)
1785                         goto truncated;
1786                 break;
1787         case IPPROTO_TCP: {
1788                 tcpha_t         *tcpha;
1789 
1790                 /*
1791                  * Verify we have at least ICMP_MIN_TP_HDR_LEN bytes of
1792                  * transport header.
1793                  */
1794                 if ((uchar_t *)ipha + hdr_length + ICMP_MIN_TP_HDR_LEN >
1795                     mp->b_wptr)
1796                         goto truncated;
1797 
1798                 tcpha = (tcpha_t *)((uchar_t *)ipha + hdr_length);
1799                 connp = ipcl_tcp_lookup_reversed_ipv4(ipha, tcpha, TCPS_LISTEN,
1800                     ipst);
1801                 if (connp == NULL)
1802                         goto discard_pkt;
1803 
1804                 if ((connp->conn_verifyicmp != NULL) &&
1805                     !connp->conn_verifyicmp(connp, tcpha, icmph, NULL, ira)) {
1806                         CONN_DEC_REF(connp);
1807                         goto discard_pkt;
1808                 }
1809                 CONN_DEC_REF(connp);
1810                 break;
1811         }
1812         case IPPROTO_SCTP:
1813                 /*
1814                  * Verify we have at least ICMP_MIN_TP_HDR_LEN bytes of
1815                  * transport header.
1816                  */
1817                 if ((uchar_t *)ipha + hdr_length + ICMP_MIN_TP_HDR_LEN >
1818                     mp->b_wptr)
1819                         goto truncated;
1820                 break;
1821         case IPPROTO_ESP:
1822         case IPPROTO_AH:
1823                 break;
1824         case IPPROTO_ENCAP:
1825                 if ((uchar_t *)ipha + hdr_length + sizeof (ipha_t) >
1826                     mp->b_wptr)
1827                         goto truncated;
1828                 break;
1829         default:
1830                 break;
1831         }
1832 
1833         return (B_TRUE);
1834 
1835 discard_pkt:
1836         /* Bogus ICMP error. */
1837         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1838         return (B_FALSE);
1839 
1840 truncated:
1841         /* We pulled up everthing already. Must be truncated */
1842         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
1843         ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
1844         return (B_FALSE);
1845 }
1846 
1847 /* Table from RFC 1191 */
1848 static int icmp_frag_size_table[] =
1849 { 32000, 17914, 8166, 4352, 2002, 1496, 1006, 508, 296, 68 };
1850 
1851 /*
1852  * Process received ICMP Packet too big.
1853  * Just handles the DCE create/update, including using the above table of
1854  * PMTU guesses. The caller is responsible for validating the packet before
1855  * passing it in and also to fanout the ICMP error to any matching transport
1856  * conns. Assumes the message has been fully pulled up and verified.
1857  *
1858  * Before getting here, the caller has called icmp_inbound_verify_v4()
1859  * that should have verified with ULP to prevent undoing the changes we're
1860  * going to make to DCE. For example, TCP might have verified that the packet
1861  * which generated error is in the send window.
1862  *
1863  * In some cases modified this MTU in the ICMP header packet; the caller
1864  * should pass to the matching ULP after this returns.
1865  */
1866 static void
1867 icmp_inbound_too_big_v4(icmph_t *icmph, ip_recv_attr_t *ira)
1868 {
1869         dce_t           *dce;
1870         int             old_mtu;
1871         int             mtu, orig_mtu;
1872         ipaddr_t        dst;
1873         boolean_t       disable_pmtud;
1874         ill_t           *ill = ira->ira_ill;
1875         ip_stack_t      *ipst = ill->ill_ipst;
1876         uint_t          hdr_length;
1877         ipha_t          *ipha;
1878 
1879         /* Caller already pulled up everything. */
1880         ipha = (ipha_t *)&icmph[1];
1881         ASSERT(icmph->icmph_type == ICMP_DEST_UNREACHABLE &&
1882             icmph->icmph_code == ICMP_FRAGMENTATION_NEEDED);
1883         ASSERT(ill != NULL);
1884 
1885         hdr_length = IPH_HDR_LENGTH(ipha);
1886 
1887         /*
1888          * We handle path MTU for source routed packets since the DCE
1889          * is looked up using the final destination.
1890          */
1891         dst = ip_get_dst(ipha);
1892 
1893         dce = dce_lookup_and_add_v4(dst, ipst);
1894         if (dce == NULL) {
1895                 /* Couldn't add a unique one - ENOMEM */
1896                 ip1dbg(("icmp_inbound_too_big_v4: no dce for 0x%x\n",
1897                     ntohl(dst)));
1898                 return;
1899         }
1900 
1901         /* Check for MTU discovery advice as described in RFC 1191 */
1902         mtu = ntohs(icmph->icmph_du_mtu);
1903         orig_mtu = mtu;
1904         disable_pmtud = B_FALSE;
1905 
1906         mutex_enter(&dce->dce_lock);
1907         if (dce->dce_flags & DCEF_PMTU)
1908                 old_mtu = dce->dce_pmtu;
1909         else
1910                 old_mtu = ill->ill_mtu;
1911 
1912         if (icmph->icmph_du_zero != 0 || mtu < ipst->ips_ip_pmtu_min) {
1913                 uint32_t length;
1914                 int     i;
1915 
1916                 /*
1917                  * Use the table from RFC 1191 to figure out
1918                  * the next "plateau" based on the length in
1919                  * the original IP packet.
1920                  */
1921                 length = ntohs(ipha->ipha_length);
1922                 DTRACE_PROBE2(ip4__pmtu__guess, dce_t *, dce,
1923                     uint32_t, length);
1924                 if (old_mtu <= length &&
1925                     old_mtu >= length - hdr_length) {
1926                         /*
1927                          * Handle broken BSD 4.2 systems that
1928                          * return the wrong ipha_length in ICMP
1929                          * errors.
1930                          */
1931                         ip1dbg(("Wrong mtu: sent %d, dce %d\n",
1932                             length, old_mtu));
1933                         length -= hdr_length;
1934                 }
1935                 for (i = 0; i < A_CNT(icmp_frag_size_table); i++) {
1936                         if (length > icmp_frag_size_table[i])
1937                                 break;
1938                 }
1939                 if (i == A_CNT(icmp_frag_size_table)) {
1940                         /* Smaller than IP_MIN_MTU! */
1941                         ip1dbg(("Too big for packet size %d\n",
1942                             length));
1943                         disable_pmtud = B_TRUE;
1944                         mtu = ipst->ips_ip_pmtu_min;
1945                 } else {
1946                         mtu = icmp_frag_size_table[i];
1947                         ip1dbg(("Calculated mtu %d, packet size %d, "
1948                             "before %d\n", mtu, length, old_mtu));
1949                         if (mtu < ipst->ips_ip_pmtu_min) {
1950                                 mtu = ipst->ips_ip_pmtu_min;
1951                                 disable_pmtud = B_TRUE;
1952                         }
1953                 }
1954         }
1955         if (disable_pmtud)
1956                 dce->dce_flags |= DCEF_TOO_SMALL_PMTU;
1957         else
1958                 dce->dce_flags &= ~DCEF_TOO_SMALL_PMTU;
1959 
1960         dce->dce_pmtu = MIN(old_mtu, mtu);
1961         /* Prepare to send the new max frag size for the ULP. */
1962         icmph->icmph_du_zero = 0;
1963         icmph->icmph_du_mtu =  htons((uint16_t)dce->dce_pmtu);
1964         DTRACE_PROBE4(ip4__pmtu__change, icmph_t *, icmph, dce_t *,
1965             dce, int, orig_mtu, int, mtu);
1966 
1967         /* We now have a PMTU for sure */
1968         dce->dce_flags |= DCEF_PMTU;
1969         dce->dce_last_change_time = TICK_TO_SEC(ddi_get_lbolt64());
1970         mutex_exit(&dce->dce_lock);
1971         /*
1972          * After dropping the lock the new value is visible to everyone.
1973          * Then we bump the generation number so any cached values reinspect
1974          * the dce_t.
1975          */
1976         dce_increment_generation(dce);
1977         dce_refrele(dce);
1978 }
1979 
1980 /*
1981  * If the packet in error is Self-Encapsulated, icmp_inbound_error_fanout_v4
1982  * calls this function.
1983  */
1984 static mblk_t *
1985 icmp_inbound_self_encap_error_v4(mblk_t *mp, ipha_t *ipha, ipha_t *in_ipha)
1986 {
1987         int length;
1988 
1989         ASSERT(mp->b_datap->db_type == M_DATA);
1990 
1991         /* icmp_inbound_v4 has already pulled up the whole error packet */
1992         ASSERT(mp->b_cont == NULL);
1993 
1994         /*
1995          * The length that we want to overlay is the inner header
1996          * and what follows it.
1997          */
1998         length = msgdsize(mp) - ((uchar_t *)in_ipha - mp->b_rptr);
1999 
2000         /*
2001          * Overlay the inner header and whatever follows it over the
2002          * outer header.
2003          */
2004         bcopy((uchar_t *)in_ipha, (uchar_t *)ipha, length);
2005 
2006         /* Adjust for what we removed */
2007         mp->b_wptr -= (uchar_t *)in_ipha - (uchar_t *)ipha;
2008         return (mp);
2009 }
2010 
2011 /*
2012  * Try to pass the ICMP message upstream in case the ULP cares.
2013  *
2014  * If the packet that caused the ICMP error is secure, we send
2015  * it to AH/ESP to make sure that the attached packet has a
2016  * valid association. ipha in the code below points to the
2017  * IP header of the packet that caused the error.
2018  *
2019  * For IPsec cases, we let the next-layer-up (which has access to
2020  * cached policy on the conn_t, or can query the SPD directly)
2021  * subtract out any IPsec overhead if they must.  We therefore make no
2022  * adjustments here for IPsec overhead.
2023  *
2024  * IFN could have been generated locally or by some router.
2025  *
2026  * LOCAL : ire_send_wire (before calling ipsec_out_process) can call
2027  * icmp_frag_needed/icmp_pkt2big_v6 to generated a local IFN.
2028  *          This happens because IP adjusted its value of MTU on an
2029  *          earlier IFN message and could not tell the upper layer,
2030  *          the new adjusted value of MTU e.g. Packet was encrypted
2031  *          or there was not enough information to fanout to upper
2032  *          layers. Thus on the next outbound datagram, ire_send_wire
2033  *          generates the IFN, where IPsec processing has *not* been
2034  *          done.
2035  *
2036  *          Note that we retain ixa_fragsize across IPsec thus once
2037  *          we have picking ixa_fragsize and entered ipsec_out_process we do
2038  *          no change the fragsize even if the path MTU changes before
2039  *          we reach ip_output_post_ipsec.
2040  *
2041  *          In the local case, IRAF_LOOPBACK will be set indicating
2042  *          that IFN was generated locally.
2043  *
2044  * ROUTER : IFN could be secure or non-secure.
2045  *
2046  *          * SECURE : We use the IPSEC_IN to fanout to AH/ESP if the
2047  *            packet in error has AH/ESP headers to validate the AH/ESP
2048  *            headers. AH/ESP will verify whether there is a valid SA or
2049  *            not and send it back. We will fanout again if we have more
2050  *            data in the packet.
2051  *
2052  *            If the packet in error does not have AH/ESP, we handle it
2053  *            like any other case.
2054  *
2055  *          * NON_SECURE : If the packet in error has AH/ESP headers, we send it
2056  *            up to AH/ESP for validation. AH/ESP will verify whether there is a
2057  *            valid SA or not and send it back. We will fanout again if
2058  *            we have more data in the packet.
2059  *
2060  *            If the packet in error does not have AH/ESP, we handle it
2061  *            like any other case.
2062  *
2063  * The caller must have called icmp_inbound_verify_v4.
2064  */
2065 static void
2066 icmp_inbound_error_fanout_v4(mblk_t *mp, icmph_t *icmph, ip_recv_attr_t *ira)
2067 {
2068         uint16_t        *up;    /* Pointer to ports in ULP header */
2069         uint32_t        ports;  /* reversed ports for fanout */
2070         ipha_t          ripha;  /* With reversed addresses */
2071         ipha_t          *ipha;  /* Inner IP header */
2072         uint_t          hdr_length; /* Inner IP header length */
2073         tcpha_t         *tcpha;
2074         conn_t          *connp;
2075         ill_t           *ill = ira->ira_ill;
2076         ip_stack_t      *ipst = ill->ill_ipst;
2077         ipsec_stack_t   *ipss = ipst->ips_netstack->netstack_ipsec;
2078         ill_t           *rill = ira->ira_rill;
2079 
2080         /* Caller already pulled up everything. */
2081         ipha = (ipha_t *)&icmph[1];
2082         ASSERT((uchar_t *)&ipha[1] <= mp->b_wptr);
2083         ASSERT(mp->b_cont == NULL);
2084 
2085         hdr_length = IPH_HDR_LENGTH(ipha);
2086         ira->ira_protocol = ipha->ipha_protocol;
2087 
2088         /*
2089          * We need a separate IP header with the source and destination
2090          * addresses reversed to do fanout/classification because the ipha in
2091          * the ICMP error is in the form we sent it out.
2092          */
2093         ripha.ipha_src = ipha->ipha_dst;
2094         ripha.ipha_dst = ipha->ipha_src;
2095         ripha.ipha_protocol = ipha->ipha_protocol;
2096         ripha.ipha_version_and_hdr_length = ipha->ipha_version_and_hdr_length;
2097 
2098         ip2dbg(("icmp_inbound_error_v4: proto %d %x to %x: %d/%d\n",
2099             ripha.ipha_protocol, ntohl(ipha->ipha_src),
2100             ntohl(ipha->ipha_dst),
2101             icmph->icmph_type, icmph->icmph_code));
2102 
2103         switch (ipha->ipha_protocol) {
2104         case IPPROTO_UDP:
2105                 up = (uint16_t *)((uchar_t *)ipha + hdr_length);
2106 
2107                 /* Attempt to find a client stream based on port. */
2108                 ip2dbg(("icmp_inbound_error_v4: UDP ports %d to %d\n",
2109                     ntohs(up[0]), ntohs(up[1])));
2110 
2111                 /* Note that we send error to all matches. */
2112                 ira->ira_flags |= IRAF_ICMP_ERROR;
2113                 ip_fanout_udp_multi_v4(mp, &ripha, up[0], up[1], ira);
2114                 ira->ira_flags &= ~IRAF_ICMP_ERROR;
2115                 return;
2116 
2117         case IPPROTO_TCP:
2118                 /*
2119                  * Find a TCP client stream for this packet.
2120                  * Note that we do a reverse lookup since the header is
2121                  * in the form we sent it out.
2122                  */
2123                 tcpha = (tcpha_t *)((uchar_t *)ipha + hdr_length);
2124                 connp = ipcl_tcp_lookup_reversed_ipv4(ipha, tcpha, TCPS_LISTEN,
2125                     ipst);
2126                 if (connp == NULL)
2127                         goto discard_pkt;
2128 
2129                 if (CONN_INBOUND_POLICY_PRESENT(connp, ipss) ||
2130                     (ira->ira_flags & IRAF_IPSEC_SECURE)) {
2131                         mp = ipsec_check_inbound_policy(mp, connp,
2132                             ipha, NULL, ira);
2133                         if (mp == NULL) {
2134                                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
2135                                 /* Note that mp is NULL */
2136                                 ip_drop_input("ipIfStatsInDiscards", mp, ill);
2137                                 CONN_DEC_REF(connp);
2138                                 return;
2139                         }
2140                 }
2141 
2142                 ira->ira_flags |= IRAF_ICMP_ERROR;
2143                 ira->ira_ill = ira->ira_rill = NULL;
2144                 if (IPCL_IS_TCP(connp)) {
2145                         SQUEUE_ENTER_ONE(connp->conn_sqp, mp,
2146                             connp->conn_recvicmp, connp, ira, SQ_FILL,
2147                             SQTAG_TCP_INPUT_ICMP_ERR);
2148                 } else {
2149                         /* Not TCP; must be SOCK_RAW, IPPROTO_TCP */
2150                         (connp->conn_recv)(connp, mp, NULL, ira);
2151                         CONN_DEC_REF(connp);
2152                 }
2153                 ira->ira_ill = ill;
2154                 ira->ira_rill = rill;
2155                 ira->ira_flags &= ~IRAF_ICMP_ERROR;
2156                 return;
2157 
2158         case IPPROTO_SCTP:
2159                 up = (uint16_t *)((uchar_t *)ipha + hdr_length);
2160                 /* Find a SCTP client stream for this packet. */
2161                 ((uint16_t *)&ports)[0] = up[1];
2162                 ((uint16_t *)&ports)[1] = up[0];
2163 
2164                 ira->ira_flags |= IRAF_ICMP_ERROR;
2165                 ip_fanout_sctp(mp, &ripha, NULL, ports, ira);
2166                 ira->ira_flags &= ~IRAF_ICMP_ERROR;
2167                 return;
2168 
2169         case IPPROTO_ESP:
2170         case IPPROTO_AH:
2171                 if (!ipsec_loaded(ipss)) {
2172                         ip_proto_not_sup(mp, ira);
2173                         return;
2174                 }
2175 
2176                 if (ipha->ipha_protocol == IPPROTO_ESP)
2177                         mp = ipsecesp_icmp_error(mp, ira);
2178                 else
2179                         mp = ipsecah_icmp_error(mp, ira);
2180                 if (mp == NULL)
2181                         return;
2182 
2183                 /* Just in case ipsec didn't preserve the NULL b_cont */
2184                 if (mp->b_cont != NULL) {
2185                         if (!pullupmsg(mp, -1))
2186                                 goto discard_pkt;
2187                 }
2188 
2189                 /*
2190                  * Note that ira_pktlen and ira_ip_hdr_length are no longer
2191                  * correct, but we don't use them any more here.
2192                  *
2193                  * If succesful, the mp has been modified to not include
2194                  * the ESP/AH header so we can fanout to the ULP's icmp
2195                  * error handler.
2196                  */
2197                 if (mp->b_wptr - mp->b_rptr < IP_SIMPLE_HDR_LENGTH)
2198                         goto truncated;
2199 
2200                 /* Verify the modified message before any further processes. */
2201                 ipha = (ipha_t *)mp->b_rptr;
2202                 hdr_length = IPH_HDR_LENGTH(ipha);
2203                 icmph = (icmph_t *)&mp->b_rptr[hdr_length];
2204                 if (!icmp_inbound_verify_v4(mp, icmph, ira)) {
2205                         freemsg(mp);
2206                         return;
2207                 }
2208 
2209                 icmp_inbound_error_fanout_v4(mp, icmph, ira);
2210                 return;
2211 
2212         case IPPROTO_ENCAP: {
2213                 /* Look for self-encapsulated packets that caused an error */
2214                 ipha_t *in_ipha;
2215 
2216                 /*
2217                  * Caller has verified that length has to be
2218                  * at least the size of IP header.
2219                  */
2220                 ASSERT(hdr_length >= sizeof (ipha_t));
2221                 /*
2222                  * Check the sanity of the inner IP header like
2223                  * we did for the outer header.
2224                  */
2225                 in_ipha = (ipha_t *)((uchar_t *)ipha + hdr_length);
2226                 if ((IPH_HDR_VERSION(in_ipha) != IPV4_VERSION)) {
2227                         goto discard_pkt;
2228                 }
2229                 if (IPH_HDR_LENGTH(in_ipha) < sizeof (ipha_t)) {
2230                         goto discard_pkt;
2231                 }
2232                 /* Check for Self-encapsulated tunnels */
2233                 if (in_ipha->ipha_src == ipha->ipha_src &&
2234                     in_ipha->ipha_dst == ipha->ipha_dst) {
2235 
2236                         mp = icmp_inbound_self_encap_error_v4(mp, ipha,
2237                             in_ipha);
2238                         if (mp == NULL)
2239                                 goto discard_pkt;
2240 
2241                         /*
2242                          * Just in case self_encap didn't preserve the NULL
2243                          * b_cont
2244                          */
2245                         if (mp->b_cont != NULL) {
2246                                 if (!pullupmsg(mp, -1))
2247                                         goto discard_pkt;
2248                         }
2249                         /*
2250                          * Note that ira_pktlen and ira_ip_hdr_length are no
2251                          * longer correct, but we don't use them any more here.
2252                          */
2253                         if (mp->b_wptr - mp->b_rptr < IP_SIMPLE_HDR_LENGTH)
2254                                 goto truncated;
2255 
2256                         /*
2257                          * Verify the modified message before any further
2258                          * processes.
2259                          */
2260                         ipha = (ipha_t *)mp->b_rptr;
2261                         hdr_length = IPH_HDR_LENGTH(ipha);
2262                         icmph = (icmph_t *)&mp->b_rptr[hdr_length];
2263                         if (!icmp_inbound_verify_v4(mp, icmph, ira)) {
2264                                 freemsg(mp);
2265                                 return;
2266                         }
2267 
2268                         /*
2269                          * The packet in error is self-encapsualted.
2270                          * And we are finding it further encapsulated
2271                          * which we could not have possibly generated.
2272                          */
2273                         if (ipha->ipha_protocol == IPPROTO_ENCAP) {
2274                                 goto discard_pkt;
2275                         }
2276                         icmp_inbound_error_fanout_v4(mp, icmph, ira);
2277                         return;
2278                 }
2279                 /* No self-encapsulated */
2280         }
2281         /* FALLTHROUGH */
2282         case IPPROTO_IPV6:
2283                 if ((connp = ipcl_iptun_classify_v4(&ripha.ipha_src,
2284                     &ripha.ipha_dst, ipst)) != NULL) {
2285                         ira->ira_flags |= IRAF_ICMP_ERROR;
2286                         connp->conn_recvicmp(connp, mp, NULL, ira);
2287                         CONN_DEC_REF(connp);
2288                         ira->ira_flags &= ~IRAF_ICMP_ERROR;
2289                         return;
2290                 }
2291                 /*
2292                  * No IP tunnel is interested, fallthrough and see
2293                  * if a raw socket will want it.
2294                  */
2295                 /* FALLTHROUGH */
2296         default:
2297                 ira->ira_flags |= IRAF_ICMP_ERROR;
2298                 ip_fanout_proto_v4(mp, &ripha, ira);
2299                 ira->ira_flags &= ~IRAF_ICMP_ERROR;
2300                 return;
2301         }
2302         /* NOTREACHED */
2303 discard_pkt:
2304         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
2305         ip1dbg(("icmp_inbound_error_fanout_v4: drop pkt\n"));
2306         ip_drop_input("ipIfStatsInDiscards", mp, ill);
2307         freemsg(mp);
2308         return;
2309 
2310 truncated:
2311         /* We pulled up everthing already. Must be truncated */
2312         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
2313         ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
2314         freemsg(mp);
2315 }
2316 
2317 /*
2318  * Common IP options parser.
2319  *
2320  * Setup routine: fill in *optp with options-parsing state, then
2321  * tail-call ipoptp_next to return the first option.
2322  */
2323 uint8_t
2324 ipoptp_first(ipoptp_t *optp, ipha_t *ipha)
2325 {
2326         uint32_t totallen; /* total length of all options */
2327 
2328         totallen = ipha->ipha_version_and_hdr_length -
2329             (uint8_t)((IP_VERSION << 4) + IP_SIMPLE_HDR_LENGTH_IN_WORDS);
2330         totallen <<= 2;
2331         optp->ipoptp_next = (uint8_t *)(&ipha[1]);
2332         optp->ipoptp_end = optp->ipoptp_next + totallen;
2333         optp->ipoptp_flags = 0;
2334         return (ipoptp_next(optp));
2335 }
2336 
2337 /* Like above but without an ipha_t */
2338 uint8_t
2339 ipoptp_first2(ipoptp_t *optp, uint32_t totallen, uint8_t *opt)
2340 {
2341         optp->ipoptp_next = opt;
2342         optp->ipoptp_end = optp->ipoptp_next + totallen;
2343         optp->ipoptp_flags = 0;
2344         return (ipoptp_next(optp));
2345 }
2346 
2347 /*
2348  * Common IP options parser: extract next option.
2349  */
2350 uint8_t
2351 ipoptp_next(ipoptp_t *optp)
2352 {
2353         uint8_t *end = optp->ipoptp_end;
2354         uint8_t *cur = optp->ipoptp_next;
2355         uint8_t opt, len, pointer;
2356 
2357         /*
2358          * If cur > end already, then the ipoptp_end or ipoptp_next pointer
2359          * has been corrupted.
2360          */
2361         ASSERT(cur <= end);
2362 
2363         if (cur == end)
2364                 return (IPOPT_EOL);
2365 
2366         opt = cur[IPOPT_OPTVAL];
2367 
2368         /*
2369          * Skip any NOP options.
2370          */
2371         while (opt == IPOPT_NOP) {
2372                 cur++;
2373                 if (cur == end)
2374                         return (IPOPT_EOL);
2375                 opt = cur[IPOPT_OPTVAL];
2376         }
2377 
2378         if (opt == IPOPT_EOL)
2379                 return (IPOPT_EOL);
2380 
2381         /*
2382          * Option requiring a length.
2383          */
2384         if ((cur + 1) >= end) {
2385                 optp->ipoptp_flags |= IPOPTP_ERROR;
2386                 return (IPOPT_EOL);
2387         }
2388         len = cur[IPOPT_OLEN];
2389         if (len < 2) {
2390                 optp->ipoptp_flags |= IPOPTP_ERROR;
2391                 return (IPOPT_EOL);
2392         }
2393         optp->ipoptp_cur = cur;
2394         optp->ipoptp_len = len;
2395         optp->ipoptp_next = cur + len;
2396         if (cur + len > end) {
2397                 optp->ipoptp_flags |= IPOPTP_ERROR;
2398                 return (IPOPT_EOL);
2399         }
2400 
2401         /*
2402          * For the options which require a pointer field, make sure
2403          * its there, and make sure it points to either something
2404          * inside this option, or the end of the option.
2405          */
2406         switch (opt) {
2407         case IPOPT_RR:
2408         case IPOPT_TS:
2409         case IPOPT_LSRR:
2410         case IPOPT_SSRR:
2411                 if (len <= IPOPT_OFFSET) {
2412                         optp->ipoptp_flags |= IPOPTP_ERROR;
2413                         return (opt);
2414                 }
2415                 pointer = cur[IPOPT_OFFSET];
2416                 if (pointer - 1 > len) {
2417                         optp->ipoptp_flags |= IPOPTP_ERROR;
2418                         return (opt);
2419                 }
2420                 break;
2421         }
2422 
2423         /*
2424          * Sanity check the pointer field based on the type of the
2425          * option.
2426          */
2427         switch (opt) {
2428         case IPOPT_RR:
2429         case IPOPT_SSRR:
2430         case IPOPT_LSRR:
2431                 if (pointer < IPOPT_MINOFF_SR)
2432                         optp->ipoptp_flags |= IPOPTP_ERROR;
2433                 break;
2434         case IPOPT_TS:
2435                 if (pointer < IPOPT_MINOFF_IT)
2436                         optp->ipoptp_flags |= IPOPTP_ERROR;
2437                 /*
2438                  * Note that the Internet Timestamp option also
2439                  * contains two four bit fields (the Overflow field,
2440                  * and the Flag field), which follow the pointer
2441                  * field.  We don't need to check that these fields
2442                  * fall within the length of the option because this
2443                  * was implicitely done above.  We've checked that the
2444                  * pointer value is at least IPOPT_MINOFF_IT, and that
2445                  * it falls within the option.  Since IPOPT_MINOFF_IT >
2446                  * IPOPT_POS_OV_FLG, we don't need the explicit check.
2447                  */
2448                 ASSERT(len > IPOPT_POS_OV_FLG);
2449                 break;
2450         }
2451 
2452         return (opt);
2453 }
2454 
2455 /*
2456  * Use the outgoing IP header to create an IP_OPTIONS option the way
2457  * it was passed down from the application.
2458  *
2459  * This is compatible with BSD in that it returns
2460  * the reverse source route with the final destination
2461  * as the last entry. The first 4 bytes of the option
2462  * will contain the final destination.
2463  */
2464 int
2465 ip_opt_get_user(conn_t *connp, uchar_t *buf)
2466 {
2467         ipoptp_t        opts;
2468         uchar_t         *opt;
2469         uint8_t         optval;
2470         uint8_t         optlen;
2471         uint32_t        len = 0;
2472         uchar_t         *buf1 = buf;
2473         uint32_t        totallen;
2474         ipaddr_t        dst;
2475         ip_pkt_t        *ipp = &connp->conn_xmit_ipp;
2476 
2477         if (!(ipp->ipp_fields & IPPF_IPV4_OPTIONS))
2478                 return (0);
2479 
2480         totallen = ipp->ipp_ipv4_options_len;
2481         if (totallen & 0x3)
2482                 return (0);
2483 
2484         buf += IP_ADDR_LEN;     /* Leave room for final destination */
2485         len += IP_ADDR_LEN;
2486         bzero(buf1, IP_ADDR_LEN);
2487 
2488         dst = connp->conn_faddr_v4;
2489 
2490         for (optval = ipoptp_first2(&opts, totallen, ipp->ipp_ipv4_options);
2491             optval != IPOPT_EOL;
2492             optval = ipoptp_next(&opts)) {
2493                 int     off;
2494 
2495                 opt = opts.ipoptp_cur;
2496                 if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
2497                         break;
2498                 }
2499                 optlen = opts.ipoptp_len;
2500 
2501                 switch (optval) {
2502                 case IPOPT_SSRR:
2503                 case IPOPT_LSRR:
2504 
2505                         /*
2506                          * Insert destination as the first entry in the source
2507                          * route and move down the entries on step.
2508                          * The last entry gets placed at buf1.
2509                          */
2510                         buf[IPOPT_OPTVAL] = optval;
2511                         buf[IPOPT_OLEN] = optlen;
2512                         buf[IPOPT_OFFSET] = optlen;
2513 
2514                         off = optlen - IP_ADDR_LEN;
2515                         if (off < 0) {
2516                                 /* No entries in source route */
2517                                 break;
2518                         }
2519                         /* Last entry in source route if not already set */
2520                         if (dst == INADDR_ANY)
2521                                 bcopy(opt + off, buf1, IP_ADDR_LEN);
2522                         off -= IP_ADDR_LEN;
2523 
2524                         while (off > 0) {
2525                                 bcopy(opt + off,
2526                                     buf + off + IP_ADDR_LEN,
2527                                     IP_ADDR_LEN);
2528                                 off -= IP_ADDR_LEN;
2529                         }
2530                         /* ipha_dst into first slot */
2531                         bcopy(&dst, buf + off + IP_ADDR_LEN,
2532                             IP_ADDR_LEN);
2533                         buf += optlen;
2534                         len += optlen;
2535                         break;
2536 
2537                 default:
2538                         bcopy(opt, buf, optlen);
2539                         buf += optlen;
2540                         len += optlen;
2541                         break;
2542                 }
2543         }
2544 done:
2545         /* Pad the resulting options */
2546         while (len & 0x3) {
2547                 *buf++ = IPOPT_EOL;
2548                 len++;
2549         }
2550         return (len);
2551 }
2552 
2553 /*
2554  * Update any record route or timestamp options to include this host.
2555  * Reverse any source route option.
2556  * This routine assumes that the options are well formed i.e. that they
2557  * have already been checked.
2558  */
2559 static void
2560 icmp_options_update(ipha_t *ipha)
2561 {
2562         ipoptp_t        opts;
2563         uchar_t         *opt;
2564         uint8_t         optval;
2565         ipaddr_t        src;            /* Our local address */
2566         ipaddr_t        dst;
2567 
2568         ip2dbg(("icmp_options_update\n"));
2569         src = ipha->ipha_src;
2570         dst = ipha->ipha_dst;
2571 
2572         for (optval = ipoptp_first(&opts, ipha);
2573             optval != IPOPT_EOL;
2574             optval = ipoptp_next(&opts)) {
2575                 ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
2576                 opt = opts.ipoptp_cur;
2577                 ip2dbg(("icmp_options_update: opt %d, len %d\n",
2578                     optval, opts.ipoptp_len));
2579                 switch (optval) {
2580                         int off1, off2;
2581                 case IPOPT_SSRR:
2582                 case IPOPT_LSRR:
2583                         /*
2584                          * Reverse the source route.  The first entry
2585                          * should be the next to last one in the current
2586                          * source route (the last entry is our address).
2587                          * The last entry should be the final destination.
2588                          */
2589                         off1 = IPOPT_MINOFF_SR - 1;
2590                         off2 = opt[IPOPT_OFFSET] - IP_ADDR_LEN - 1;
2591                         if (off2 < 0) {
2592                                 /* No entries in source route */
2593                                 ip1dbg((
2594                                     "icmp_options_update: bad src route\n"));
2595                                 break;
2596                         }
2597                         bcopy((char *)opt + off2, &dst, IP_ADDR_LEN);
2598                         bcopy(&ipha->ipha_dst, (char *)opt + off2, IP_ADDR_LEN);
2599                         bcopy(&dst, &ipha->ipha_dst, IP_ADDR_LEN);
2600                         off2 -= IP_ADDR_LEN;
2601 
2602                         while (off1 < off2) {
2603                                 bcopy((char *)opt + off1, &src, IP_ADDR_LEN);
2604                                 bcopy((char *)opt + off2, (char *)opt + off1,
2605                                     IP_ADDR_LEN);
2606                                 bcopy(&src, (char *)opt + off2, IP_ADDR_LEN);
2607                                 off1 += IP_ADDR_LEN;
2608                                 off2 -= IP_ADDR_LEN;
2609                         }
2610                         opt[IPOPT_OFFSET] = IPOPT_MINOFF_SR;
2611                         break;
2612                 }
2613         }
2614 }
2615 
2616 /*
2617  * Process received ICMP Redirect messages.
2618  * Assumes the caller has verified that the headers are in the pulled up mblk.
2619  * Consumes mp.
2620  */
2621 static void
2622 icmp_redirect_v4(mblk_t *mp, ipha_t *ipha, icmph_t *icmph, ip_recv_attr_t *ira)
2623 {
2624         ire_t           *ire, *nire;
2625         ire_t           *prev_ire;
2626         ipaddr_t        src, dst, gateway;
2627         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
2628         ipha_t          *inner_ipha;    /* Inner IP header */
2629 
2630         /* Caller already pulled up everything. */
2631         inner_ipha = (ipha_t *)&icmph[1];
2632         src = ipha->ipha_src;
2633         dst = inner_ipha->ipha_dst;
2634         gateway = icmph->icmph_rd_gateway;
2635         /* Make sure the new gateway is reachable somehow. */
2636         ire = ire_ftable_lookup_v4(gateway, 0, 0, IRE_ONLINK, NULL,
2637             ALL_ZONES, NULL, MATCH_IRE_TYPE, 0, ipst, NULL);
2638         /*
2639          * Make sure we had a route for the dest in question and that
2640          * that route was pointing to the old gateway (the source of the
2641          * redirect packet.)
2642          * We do longest match and then compare ire_gateway_addr below.
2643          */
2644         prev_ire = ire_ftable_lookup_v4(dst, 0, 0, 0, NULL, ALL_ZONES,
2645             NULL, MATCH_IRE_DSTONLY, 0, ipst, NULL);
2646         /*
2647          * Check that
2648          *      the redirect was not from ourselves
2649          *      the new gateway and the old gateway are directly reachable
2650          */
2651         if (prev_ire == NULL || ire == NULL ||
2652             (prev_ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK)) ||
2653             (prev_ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) ||
2654             !(ire->ire_type & IRE_IF_ALL) ||
2655             prev_ire->ire_gateway_addr != src) {
2656                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInBadRedirects);
2657                 ip_drop_input("icmpInBadRedirects - ire", mp, ira->ira_ill);
2658                 freemsg(mp);
2659                 if (ire != NULL)
2660                         ire_refrele(ire);
2661                 if (prev_ire != NULL)
2662                         ire_refrele(prev_ire);
2663                 return;
2664         }
2665 
2666         ire_refrele(prev_ire);
2667         ire_refrele(ire);
2668 
2669         /*
2670          * TODO: more precise handling for cases 0, 2, 3, the latter two
2671          * require TOS routing
2672          */
2673         switch (icmph->icmph_code) {
2674         case 0:
2675         case 1:
2676                 /* TODO: TOS specificity for cases 2 and 3 */
2677         case 2:
2678         case 3:
2679                 break;
2680         default:
2681                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInBadRedirects);
2682                 ip_drop_input("icmpInBadRedirects - code", mp, ira->ira_ill);
2683                 freemsg(mp);
2684                 return;
2685         }
2686         /*
2687          * Create a Route Association.  This will allow us to remember that
2688          * someone we believe told us to use the particular gateway.
2689          */
2690         ire = ire_create(
2691             (uchar_t *)&dst,                        /* dest addr */
2692             (uchar_t *)&ip_g_all_ones,              /* mask */
2693             (uchar_t *)&gateway,            /* gateway addr */
2694             IRE_HOST,
2695             NULL,                               /* ill */
2696             ALL_ZONES,
2697             (RTF_DYNAMIC | RTF_GATEWAY | RTF_HOST),
2698             NULL,                               /* tsol_gc_t */
2699             ipst);
2700 
2701         if (ire == NULL) {
2702                 freemsg(mp);
2703                 return;
2704         }
2705         nire = ire_add(ire);
2706         /* Check if it was a duplicate entry */
2707         if (nire != NULL && nire != ire) {
2708                 ASSERT(nire->ire_identical_ref > 1);
2709                 ire_delete(nire);
2710                 ire_refrele(nire);
2711                 nire = NULL;
2712         }
2713         ire = nire;
2714         if (ire != NULL) {
2715                 ire_refrele(ire);               /* Held in ire_add */
2716 
2717                 /* tell routing sockets that we received a redirect */
2718                 ip_rts_change(RTM_REDIRECT, dst, gateway, IP_HOST_MASK, 0, src,
2719                     (RTF_DYNAMIC | RTF_GATEWAY | RTF_HOST), 0,
2720                     (RTA_DST | RTA_GATEWAY | RTA_NETMASK | RTA_AUTHOR), ipst);
2721         }
2722 
2723         /*
2724          * Delete any existing IRE_HOST type redirect ires for this destination.
2725          * This together with the added IRE has the effect of
2726          * modifying an existing redirect.
2727          */
2728         prev_ire = ire_ftable_lookup_v4(dst, 0, src, IRE_HOST, NULL,
2729             ALL_ZONES, NULL, (MATCH_IRE_GW | MATCH_IRE_TYPE), 0, ipst, NULL);
2730         if (prev_ire != NULL) {
2731                 if (prev_ire ->ire_flags & RTF_DYNAMIC)
2732                         ire_delete(prev_ire);
2733                 ire_refrele(prev_ire);
2734         }
2735 
2736         freemsg(mp);
2737 }
2738 
2739 /*
2740  * Generate an ICMP parameter problem message.
2741  * When called from ip_output side a minimal ip_recv_attr_t needs to be
2742  * constructed by the caller.
2743  */
2744 static void
2745 icmp_param_problem(mblk_t *mp, uint8_t ptr, ip_recv_attr_t *ira)
2746 {
2747         icmph_t icmph;
2748         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
2749 
2750         mp = icmp_pkt_err_ok(mp, ira);
2751         if (mp == NULL)
2752                 return;
2753 
2754         bzero(&icmph, sizeof (icmph_t));
2755         icmph.icmph_type = ICMP_PARAM_PROBLEM;
2756         icmph.icmph_pp_ptr = ptr;
2757         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutParmProbs);
2758         icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
2759 }
2760 
2761 /*
2762  * Build and ship an IPv4 ICMP message using the packet data in mp, and
2763  * the ICMP header pointed to by "stuff".  (May be called as writer.)
2764  * Note: assumes that icmp_pkt_err_ok has been called to verify that
2765  * an icmp error packet can be sent.
2766  * Assigns an appropriate source address to the packet. If ipha_dst is
2767  * one of our addresses use it for source. Otherwise let ip_output_simple
2768  * pick the source address.
2769  */
2770 static void
2771 icmp_pkt(mblk_t *mp, void *stuff, size_t len, ip_recv_attr_t *ira)
2772 {
2773         ipaddr_t dst;
2774         icmph_t *icmph;
2775         ipha_t  *ipha;
2776         uint_t  len_needed;
2777         size_t  msg_len;
2778         mblk_t  *mp1;
2779         ipaddr_t src;
2780         ire_t   *ire;
2781         ip_xmit_attr_t ixas;
2782         ip_stack_t *ipst = ira->ira_ill->ill_ipst;
2783 
2784         ipha = (ipha_t *)mp->b_rptr;
2785 
2786         bzero(&ixas, sizeof (ixas));
2787         ixas.ixa_flags = IXAF_BASIC_SIMPLE_V4;
2788         ixas.ixa_zoneid = ira->ira_zoneid;
2789         ixas.ixa_ifindex = 0;
2790         ixas.ixa_ipst = ipst;
2791         ixas.ixa_cred = kcred;
2792         ixas.ixa_cpid = NOPID;
2793         ixas.ixa_tsl = ira->ira_tsl; /* Behave as a multi-level responder */
2794         ixas.ixa_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
2795 
2796         if (ira->ira_flags & IRAF_IPSEC_SECURE) {
2797                 /*
2798                  * Apply IPsec based on how IPsec was applied to
2799                  * the packet that had the error.
2800                  *
2801                  * If it was an outbound packet that caused the ICMP
2802                  * error, then the caller will have setup the IRA
2803                  * appropriately.
2804                  */
2805                 if (!ipsec_in_to_out(ira, &ixas, mp, ipha, NULL)) {
2806                         BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsOutDiscards);
2807                         /* Note: mp already consumed and ip_drop_packet done */
2808                         return;
2809                 }
2810         } else {
2811                 /*
2812                  * This is in clear. The icmp message we are building
2813                  * here should go out in clear, independent of our policy.
2814                  */
2815                 ixas.ixa_flags |= IXAF_NO_IPSEC;
2816         }
2817 
2818         /* Remember our eventual destination */
2819         dst = ipha->ipha_src;
2820 
2821         /*
2822          * If the packet was for one of our unicast addresses, make
2823          * sure we respond with that as the source. Otherwise
2824          * have ip_output_simple pick the source address.
2825          */
2826         ire = ire_ftable_lookup_v4(ipha->ipha_dst, 0, 0,
2827             (IRE_LOCAL|IRE_LOOPBACK), NULL, ira->ira_zoneid, NULL,
2828             MATCH_IRE_TYPE|MATCH_IRE_ZONEONLY, 0, ipst, NULL);
2829         if (ire != NULL) {
2830                 ire_refrele(ire);
2831                 src = ipha->ipha_dst;
2832         } else {
2833                 src = INADDR_ANY;
2834                 ixas.ixa_flags |= IXAF_SET_SOURCE;
2835         }
2836 
2837         /*
2838          * Check if we can send back more then 8 bytes in addition to
2839          * the IP header.  We try to send 64 bytes of data and the internal
2840          * header in the special cases of ipv4 encapsulated ipv4 or ipv6.
2841          */
2842         len_needed = IPH_HDR_LENGTH(ipha);
2843         if (ipha->ipha_protocol == IPPROTO_ENCAP ||
2844             ipha->ipha_protocol == IPPROTO_IPV6) {
2845                 if (!pullupmsg(mp, -1)) {
2846                         BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsOutDiscards);
2847                         ip_drop_output("ipIfStatsOutDiscards", mp, NULL);
2848                         freemsg(mp);
2849                         return;
2850                 }
2851                 ipha = (ipha_t *)mp->b_rptr;
2852 
2853                 if (ipha->ipha_protocol == IPPROTO_ENCAP) {
2854                         len_needed += IPH_HDR_LENGTH(((uchar_t *)ipha +
2855                             len_needed));
2856                 } else {
2857                         ip6_t *ip6h = (ip6_t *)((uchar_t *)ipha + len_needed);
2858 
2859                         ASSERT(ipha->ipha_protocol == IPPROTO_IPV6);
2860                         len_needed += ip_hdr_length_v6(mp, ip6h);
2861                 }
2862         }
2863         len_needed += ipst->ips_ip_icmp_return;
2864         msg_len = msgdsize(mp);
2865         if (msg_len > len_needed) {
2866                 (void) adjmsg(mp, len_needed - msg_len);
2867                 msg_len = len_needed;
2868         }
2869         mp1 = allocb(sizeof (icmp_ipha) + len, BPRI_MED);
2870         if (mp1 == NULL) {
2871                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutErrors);
2872                 freemsg(mp);
2873                 return;
2874         }
2875         mp1->b_cont = mp;
2876         mp = mp1;
2877 
2878         /*
2879          * Set IXAF_TRUSTED_ICMP so we can let the ICMP messages this
2880          * node generates be accepted in peace by all on-host destinations.
2881          * If we do NOT assume that all on-host destinations trust
2882          * self-generated ICMP messages, then rework here, ip6.c, and spd.c.
2883          * (Look for IXAF_TRUSTED_ICMP).
2884          */
2885         ixas.ixa_flags |= IXAF_TRUSTED_ICMP;
2886 
2887         ipha = (ipha_t *)mp->b_rptr;
2888         mp1->b_wptr = (uchar_t *)ipha + (sizeof (icmp_ipha) + len);
2889         *ipha = icmp_ipha;
2890         ipha->ipha_src = src;
2891         ipha->ipha_dst = dst;
2892         ipha->ipha_ttl = ipst->ips_ip_def_ttl;
2893         msg_len += sizeof (icmp_ipha) + len;
2894         if (msg_len > IP_MAXPACKET) {
2895                 (void) adjmsg(mp, IP_MAXPACKET - msg_len);
2896                 msg_len = IP_MAXPACKET;
2897         }
2898         ipha->ipha_length = htons((uint16_t)msg_len);
2899         icmph = (icmph_t *)&ipha[1];
2900         bcopy(stuff, icmph, len);
2901         icmph->icmph_checksum = 0;
2902         icmph->icmph_checksum = IP_CSUM(mp, (int32_t)sizeof (ipha_t), 0);
2903         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutMsgs);
2904 
2905         (void) ip_output_simple(mp, &ixas);
2906         ixa_cleanup(&ixas);
2907 }
2908 
2909 /*
2910  * Determine if an ICMP error packet can be sent given the rate limit.
2911  * The limit consists of an average frequency (icmp_pkt_err_interval measured
2912  * in milliseconds) and a burst size. Burst size number of packets can
2913  * be sent arbitrarely closely spaced.
2914  * The state is tracked using two variables to implement an approximate
2915  * token bucket filter:
2916  *      icmp_pkt_err_last - lbolt value when the last burst started
2917  *      icmp_pkt_err_sent - number of packets sent in current burst
2918  */
2919 boolean_t
2920 icmp_err_rate_limit(ip_stack_t *ipst)
2921 {
2922         clock_t now = TICK_TO_MSEC(ddi_get_lbolt());
2923         uint_t refilled; /* Number of packets refilled in tbf since last */
2924         /* Guard against changes by loading into local variable */
2925         uint_t err_interval = ipst->ips_ip_icmp_err_interval;
2926 
2927         if (err_interval == 0)
2928                 return (B_FALSE);
2929 
2930         if (ipst->ips_icmp_pkt_err_last > now) {
2931                 /* 100HZ lbolt in ms for 32bit arch wraps every 49.7 days */
2932                 ipst->ips_icmp_pkt_err_last = 0;
2933                 ipst->ips_icmp_pkt_err_sent = 0;
2934         }
2935         /*
2936          * If we are in a burst update the token bucket filter.
2937          * Update the "last" time to be close to "now" but make sure
2938          * we don't loose precision.
2939          */
2940         if (ipst->ips_icmp_pkt_err_sent != 0) {
2941                 refilled = (now - ipst->ips_icmp_pkt_err_last)/err_interval;
2942                 if (refilled > ipst->ips_icmp_pkt_err_sent) {
2943                         ipst->ips_icmp_pkt_err_sent = 0;
2944                 } else {
2945                         ipst->ips_icmp_pkt_err_sent -= refilled;
2946                         ipst->ips_icmp_pkt_err_last += refilled * err_interval;
2947                 }
2948         }
2949         if (ipst->ips_icmp_pkt_err_sent == 0) {
2950                 /* Start of new burst */
2951                 ipst->ips_icmp_pkt_err_last = now;
2952         }
2953         if (ipst->ips_icmp_pkt_err_sent < ipst->ips_ip_icmp_err_burst) {
2954                 ipst->ips_icmp_pkt_err_sent++;
2955                 ip1dbg(("icmp_err_rate_limit: %d sent in burst\n",
2956                     ipst->ips_icmp_pkt_err_sent));
2957                 return (B_FALSE);
2958         }
2959         ip1dbg(("icmp_err_rate_limit: dropped\n"));
2960         return (B_TRUE);
2961 }
2962 
2963 /*
2964  * Check if it is ok to send an IPv4 ICMP error packet in
2965  * response to the IPv4 packet in mp.
2966  * Free the message and return null if no
2967  * ICMP error packet should be sent.
2968  */
2969 static mblk_t *
2970 icmp_pkt_err_ok(mblk_t *mp, ip_recv_attr_t *ira)
2971 {
2972         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
2973         icmph_t *icmph;
2974         ipha_t  *ipha;
2975         uint_t  len_needed;
2976 
2977         if (!mp)
2978                 return (NULL);
2979         ipha = (ipha_t *)mp->b_rptr;
2980         if (ip_csum_hdr(ipha)) {
2981                 BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsInCksumErrs);
2982                 ip_drop_input("ipIfStatsInCksumErrs", mp, NULL);
2983                 freemsg(mp);
2984                 return (NULL);
2985         }
2986         if (ip_type_v4(ipha->ipha_dst, ipst) == IRE_BROADCAST ||
2987             ip_type_v4(ipha->ipha_src, ipst) == IRE_BROADCAST ||
2988             CLASSD(ipha->ipha_dst) ||
2989             CLASSD(ipha->ipha_src) ||
2990             (ntohs(ipha->ipha_fragment_offset_and_flags) & IPH_OFFSET)) {
2991                 /* Note: only errors to the fragment with offset 0 */
2992                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
2993                 freemsg(mp);
2994                 return (NULL);
2995         }
2996         if (ipha->ipha_protocol == IPPROTO_ICMP) {
2997                 /*
2998                  * Check the ICMP type.  RFC 1122 sez:  don't send ICMP
2999                  * errors in response to any ICMP errors.
3000                  */
3001                 len_needed = IPH_HDR_LENGTH(ipha) + ICMPH_SIZE;
3002                 if (mp->b_wptr - mp->b_rptr < len_needed) {
3003                         if (!pullupmsg(mp, len_needed)) {
3004                                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInErrors);
3005                                 freemsg(mp);
3006                                 return (NULL);
3007                         }
3008                         ipha = (ipha_t *)mp->b_rptr;
3009                 }
3010                 icmph = (icmph_t *)
3011                     (&((char *)ipha)[IPH_HDR_LENGTH(ipha)]);
3012                 switch (icmph->icmph_type) {
3013                 case ICMP_DEST_UNREACHABLE:
3014                 case ICMP_SOURCE_QUENCH:
3015                 case ICMP_TIME_EXCEEDED:
3016                 case ICMP_PARAM_PROBLEM:
3017                 case ICMP_REDIRECT:
3018                         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
3019                         freemsg(mp);
3020                         return (NULL);
3021                 default:
3022                         break;
3023                 }
3024         }
3025         /*
3026          * If this is a labeled system, then check to see if we're allowed to
3027          * send a response to this particular sender.  If not, then just drop.
3028          */
3029         if (is_system_labeled() && !tsol_can_reply_error(mp, ira)) {
3030                 ip2dbg(("icmp_pkt_err_ok: can't respond to packet\n"));
3031                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
3032                 freemsg(mp);
3033                 return (NULL);
3034         }
3035         if (icmp_err_rate_limit(ipst)) {
3036                 /*
3037                  * Only send ICMP error packets every so often.
3038                  * This should be done on a per port/source basis,
3039                  * but for now this will suffice.
3040                  */
3041                 freemsg(mp);
3042                 return (NULL);
3043         }
3044         return (mp);
3045 }
3046 
3047 /*
3048  * Called when a packet was sent out the same link that it arrived on.
3049  * Check if it is ok to send a redirect and then send it.
3050  */
3051 void
3052 ip_send_potential_redirect_v4(mblk_t *mp, ipha_t *ipha, ire_t *ire,
3053     ip_recv_attr_t *ira)
3054 {
3055         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
3056         ipaddr_t        src, nhop;
3057         mblk_t          *mp1;
3058         ire_t           *nhop_ire;
3059 
3060         /*
3061          * Check the source address to see if it originated
3062          * on the same logical subnet it is going back out on.
3063          * If so, we should be able to send it a redirect.
3064          * Avoid sending a redirect if the destination
3065          * is directly connected (i.e., we matched an IRE_ONLINK),
3066          * or if the packet was source routed out this interface.
3067          *
3068          * We avoid sending a redirect if the
3069          * destination is directly connected
3070          * because it is possible that multiple
3071          * IP subnets may have been configured on
3072          * the link, and the source may not
3073          * be on the same subnet as ip destination,
3074          * even though they are on the same
3075          * physical link.
3076          */
3077         if ((ire->ire_type & IRE_ONLINK) ||
3078             ip_source_routed(ipha, ipst))
3079                 return;
3080 
3081         nhop_ire = ire_nexthop(ire);
3082         if (nhop_ire == NULL)
3083                 return;
3084 
3085         nhop = nhop_ire->ire_addr;
3086 
3087         if (nhop_ire->ire_type & IRE_IF_CLONE) {
3088                 ire_t   *ire2;
3089 
3090                 /* Follow ire_dep_parent to find non-clone IRE_INTERFACE */
3091                 mutex_enter(&nhop_ire->ire_lock);
3092                 ire2 = nhop_ire->ire_dep_parent;
3093                 if (ire2 != NULL)
3094                         ire_refhold(ire2);
3095                 mutex_exit(&nhop_ire->ire_lock);
3096                 ire_refrele(nhop_ire);
3097                 nhop_ire = ire2;
3098         }
3099         if (nhop_ire == NULL)
3100                 return;
3101 
3102         ASSERT(!(nhop_ire->ire_type & IRE_IF_CLONE));
3103 
3104         src = ipha->ipha_src;
3105 
3106         /*
3107          * We look at the interface ire for the nexthop,
3108          * to see if ipha_src is in the same subnet
3109          * as the nexthop.
3110          */
3111         if ((src & nhop_ire->ire_mask) == (nhop & nhop_ire->ire_mask)) {
3112                 /*
3113                  * The source is directly connected.
3114                  */
3115                 mp1 = copymsg(mp);
3116                 if (mp1 != NULL) {
3117                         icmp_send_redirect(mp1, nhop, ira);
3118                 }
3119         }
3120         ire_refrele(nhop_ire);
3121 }
3122 
3123 /*
3124  * Generate an ICMP redirect message.
3125  */
3126 static void
3127 icmp_send_redirect(mblk_t *mp, ipaddr_t gateway, ip_recv_attr_t *ira)
3128 {
3129         icmph_t icmph;
3130         ip_stack_t *ipst = ira->ira_ill->ill_ipst;
3131 
3132         mp = icmp_pkt_err_ok(mp, ira);
3133         if (mp == NULL)
3134                 return;
3135 
3136         bzero(&icmph, sizeof (icmph_t));
3137         icmph.icmph_type = ICMP_REDIRECT;
3138         icmph.icmph_code = 1;
3139         icmph.icmph_rd_gateway = gateway;
3140         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutRedirects);
3141         icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
3142 }
3143 
3144 /*
3145  * Generate an ICMP time exceeded message.
3146  */
3147 void
3148 icmp_time_exceeded(mblk_t *mp, uint8_t code, ip_recv_attr_t *ira)
3149 {
3150         icmph_t icmph;
3151         ip_stack_t *ipst = ira->ira_ill->ill_ipst;
3152 
3153         mp = icmp_pkt_err_ok(mp, ira);
3154         if (mp == NULL)
3155                 return;
3156 
3157         bzero(&icmph, sizeof (icmph_t));
3158         icmph.icmph_type = ICMP_TIME_EXCEEDED;
3159         icmph.icmph_code = code;
3160         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutTimeExcds);
3161         icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
3162 }
3163 
3164 /*
3165  * Generate an ICMP unreachable message.
3166  * When called from ip_output side a minimal ip_recv_attr_t needs to be
3167  * constructed by the caller.
3168  */
3169 void
3170 icmp_unreachable(mblk_t *mp, uint8_t code, ip_recv_attr_t *ira)
3171 {
3172         icmph_t icmph;
3173         ip_stack_t *ipst = ira->ira_ill->ill_ipst;
3174 
3175         mp = icmp_pkt_err_ok(mp, ira);
3176         if (mp == NULL)
3177                 return;
3178 
3179         bzero(&icmph, sizeof (icmph_t));
3180         icmph.icmph_type = ICMP_DEST_UNREACHABLE;
3181         icmph.icmph_code = code;
3182         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDestUnreachs);
3183         icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
3184 }
3185 
3186 /*
3187  * Latch in the IPsec state for a stream based the policy in the listener
3188  * and the actions in the ip_recv_attr_t.
3189  * Called directly from TCP and SCTP.
3190  */
3191 boolean_t
3192 ip_ipsec_policy_inherit(conn_t *connp, conn_t *lconnp, ip_recv_attr_t *ira)
3193 {
3194         ASSERT(lconnp->conn_policy != NULL);
3195         ASSERT(connp->conn_policy == NULL);
3196 
3197         IPPH_REFHOLD(lconnp->conn_policy);
3198         connp->conn_policy = lconnp->conn_policy;
3199 
3200         if (ira->ira_ipsec_action != NULL) {
3201                 if (connp->conn_latch == NULL) {
3202                         connp->conn_latch = iplatch_create();
3203                         if (connp->conn_latch == NULL)
3204                                 return (B_FALSE);
3205                 }
3206                 ipsec_latch_inbound(connp, ira);
3207         }
3208         return (B_TRUE);
3209 }
3210 
3211 /*
3212  * Verify whether or not the IP address is a valid local address.
3213  * Could be a unicast, including one for a down interface.
3214  * If allow_mcbc then a multicast or broadcast address is also
3215  * acceptable.
3216  *
3217  * In the case of a broadcast/multicast address, however, the
3218  * upper protocol is expected to reset the src address
3219  * to zero when we return IPVL_MCAST/IPVL_BCAST so that
3220  * no packets are emitted with broadcast/multicast address as
3221  * source address (that violates hosts requirements RFC 1122)
3222  * The addresses valid for bind are:
3223  *      (1) - INADDR_ANY (0)
3224  *      (2) - IP address of an UP interface
3225  *      (3) - IP address of a DOWN interface
3226  *      (4) - valid local IP broadcast addresses. In this case
3227  *      the conn will only receive packets destined to
3228  *      the specified broadcast address.
3229  *      (5) - a multicast address. In this case
3230  *      the conn will only receive packets destined to
3231  *      the specified multicast address. Note: the
3232  *      application still has to issue an
3233  *      IP_ADD_MEMBERSHIP socket option.
3234  *
3235  * In all the above cases, the bound address must be valid in the current zone.
3236  * When the address is loopback, multicast or broadcast, there might be many
3237  * matching IREs so bind has to look up based on the zone.
3238  */
3239 ip_laddr_t
3240 ip_laddr_verify_v4(ipaddr_t src_addr, zoneid_t zoneid,
3241     ip_stack_t *ipst, boolean_t allow_mcbc)
3242 {
3243         ire_t *src_ire;
3244 
3245         ASSERT(src_addr != INADDR_ANY);
3246 
3247         src_ire = ire_ftable_lookup_v4(src_addr, 0, 0, 0,
3248             NULL, zoneid, NULL, MATCH_IRE_ZONEONLY, 0, ipst, NULL);
3249 
3250         /*
3251          * If an address other than in6addr_any is requested,
3252          * we verify that it is a valid address for bind
3253          * Note: Following code is in if-else-if form for
3254          * readability compared to a condition check.
3255          */
3256         if (src_ire != NULL && (src_ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK))) {
3257                 /*
3258                  * (2) Bind to address of local UP interface
3259                  */
3260                 ire_refrele(src_ire);
3261                 return (IPVL_UNICAST_UP);
3262         } else if (src_ire != NULL && src_ire->ire_type & IRE_BROADCAST) {
3263                 /*
3264                  * (4) Bind to broadcast address
3265                  */
3266                 ire_refrele(src_ire);
3267                 if (allow_mcbc)
3268                         return (IPVL_BCAST);
3269                 else
3270                         return (IPVL_BAD);
3271         } else if (CLASSD(src_addr)) {
3272                 /* (5) bind to multicast address. */
3273                 if (src_ire != NULL)
3274                         ire_refrele(src_ire);
3275 
3276                 if (allow_mcbc)
3277                         return (IPVL_MCAST);
3278                 else
3279                         return (IPVL_BAD);
3280         } else {
3281                 ipif_t *ipif;
3282 
3283                 /*
3284                  * (3) Bind to address of local DOWN interface?
3285                  * (ipif_lookup_addr() looks up all interfaces
3286                  * but we do not get here for UP interfaces
3287                  * - case (2) above)
3288                  */
3289                 if (src_ire != NULL)
3290                         ire_refrele(src_ire);
3291 
3292                 ipif = ipif_lookup_addr(src_addr, NULL, zoneid, ipst);
3293                 if (ipif == NULL)
3294                         return (IPVL_BAD);
3295 
3296                 /* Not a useful source? */
3297                 if (ipif->ipif_flags & (IPIF_NOLOCAL | IPIF_ANYCAST)) {
3298                         ipif_refrele(ipif);
3299                         return (IPVL_BAD);
3300                 }
3301                 ipif_refrele(ipif);
3302                 return (IPVL_UNICAST_DOWN);
3303         }
3304 }
3305 
3306 /*
3307  * Insert in the bind fanout for IPv4 and IPv6.
3308  * The caller should already have used ip_laddr_verify_v*() before calling
3309  * this.
3310  */
3311 int
3312 ip_laddr_fanout_insert(conn_t *connp)
3313 {
3314         int             error;
3315 
3316         /*
3317          * Allow setting new policies. For example, disconnects result
3318          * in us being called. As we would have set conn_policy_cached
3319          * to B_TRUE before, we should set it to B_FALSE, so that policy
3320          * can change after the disconnect.
3321          */
3322         connp->conn_policy_cached = B_FALSE;
3323 
3324         error = ipcl_bind_insert(connp);
3325         if (error != 0) {
3326                 if (connp->conn_anon_port) {
3327                         (void) tsol_mlp_anon(crgetzone(connp->conn_cred),
3328                             connp->conn_mlp_type, connp->conn_proto,
3329                             ntohs(connp->conn_lport), B_FALSE);
3330                 }
3331                 connp->conn_mlp_type = mlptSingle;
3332         }
3333         return (error);
3334 }
3335 
3336 /*
3337  * Verify that both the source and destination addresses are valid. If
3338  * IPDF_VERIFY_DST is not set, then the destination address may be unreachable,
3339  * i.e. have no route to it.  Protocols like TCP want to verify destination
3340  * reachability, while tunnels do not.
3341  *
3342  * Determine the route, the interface, and (optionally) the source address
3343  * to use to reach a given destination.
3344  * Note that we allow connect to broadcast and multicast addresses when
3345  * IPDF_ALLOW_MCBC is set.
3346  * first_hop and dst_addr are normally the same, but if source routing
3347  * they will differ; in that case the first_hop is what we'll use for the
3348  * routing lookup but the dce and label checks will be done on dst_addr,
3349  *
3350  * If uinfo is set, then we fill in the best available information
3351  * we have for the destination. This is based on (in priority order) any
3352  * metrics and path MTU stored in a dce_t, route metrics, and finally the
3353  * ill_mtu/ill_mc_mtu.
3354  *
3355  * Tsol note: If we have a source route then dst_addr != firsthop. But we
3356  * always do the label check on dst_addr.
3357  */
3358 int
3359 ip_set_destination_v4(ipaddr_t *src_addrp, ipaddr_t dst_addr, ipaddr_t firsthop,
3360     ip_xmit_attr_t *ixa, iulp_t *uinfo, uint32_t flags, uint_t mac_mode)
3361 {
3362         ire_t           *ire = NULL;
3363         int             error = 0;
3364         ipaddr_t        setsrc;                         /* RTF_SETSRC */
3365         zoneid_t        zoneid = ixa->ixa_zoneid;    /* Honors SO_ALLZONES */
3366         ip_stack_t      *ipst = ixa->ixa_ipst;
3367         dce_t           *dce;
3368         uint_t          pmtu;
3369         uint_t          generation;
3370         nce_t           *nce;
3371         ill_t           *ill = NULL;
3372         boolean_t       multirt = B_FALSE;
3373 
3374         ASSERT(ixa->ixa_flags & IXAF_IS_IPV4);
3375 
3376         /*
3377          * We never send to zero; the ULPs map it to the loopback address.
3378          * We can't allow it since we use zero to mean unitialized in some
3379          * places.
3380          */
3381         ASSERT(dst_addr != INADDR_ANY);
3382 
3383         if (is_system_labeled()) {
3384                 ts_label_t *tsl = NULL;
3385 
3386                 error = tsol_check_dest(ixa->ixa_tsl, &dst_addr, IPV4_VERSION,
3387                     mac_mode, (flags & IPDF_ZONE_IS_GLOBAL) != 0, &tsl);
3388                 if (error != 0)
3389                         return (error);
3390                 if (tsl != NULL) {
3391                         /* Update the label */
3392                         ip_xmit_attr_replace_tsl(ixa, tsl);
3393                 }
3394         }
3395 
3396         setsrc = INADDR_ANY;
3397         /*
3398          * Select a route; For IPMP interfaces, we would only select
3399          * a "hidden" route (i.e., going through a specific under_ill)
3400          * if ixa_ifindex has been specified.
3401          */
3402         ire = ip_select_route_v4(firsthop, *src_addrp, ixa,
3403             &generation, &setsrc, &error, &multirt);
3404         ASSERT(ire != NULL);    /* IRE_NOROUTE if none found */
3405         if (error != 0)
3406                 goto bad_addr;
3407 
3408         /*
3409          * ire can't be a broadcast or multicast unless IPDF_ALLOW_MCBC is set.
3410          * If IPDF_VERIFY_DST is set, the destination must be reachable;
3411          * Otherwise the destination needn't be reachable.
3412          *
3413          * If we match on a reject or black hole, then we've got a
3414          * local failure.  May as well fail out the connect() attempt,
3415          * since it's never going to succeed.
3416          */
3417         if (ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
3418                 /*
3419                  * If we're verifying destination reachability, we always want
3420                  * to complain here.
3421                  *
3422                  * If we're not verifying destination reachability but the
3423                  * destination has a route, we still want to fail on the
3424                  * temporary address and broadcast address tests.
3425                  *
3426                  * In both cases do we let the code continue so some reasonable
3427                  * information is returned to the caller. That enables the
3428                  * caller to use (and even cache) the IRE. conn_ip_ouput will
3429                  * use the generation mismatch path to check for the unreachable
3430                  * case thereby avoiding any specific check in the main path.
3431                  */
3432                 ASSERT(generation == IRE_GENERATION_VERIFY);
3433                 if (flags & IPDF_VERIFY_DST) {
3434                         /*
3435                          * Set errno but continue to set up ixa_ire to be
3436                          * the RTF_REJECT|RTF_BLACKHOLE IRE.
3437                          * That allows callers to use ip_output to get an
3438                          * ICMP error back.
3439                          */
3440                         if (!(ire->ire_type & IRE_HOST))
3441                                 error = ENETUNREACH;
3442                         else
3443                                 error = EHOSTUNREACH;
3444                 }
3445         }
3446 
3447         if ((ire->ire_type & (IRE_BROADCAST|IRE_MULTICAST)) &&
3448             !(flags & IPDF_ALLOW_MCBC)) {
3449                 ire_refrele(ire);
3450                 ire = ire_reject(ipst, B_FALSE);
3451                 generation = IRE_GENERATION_VERIFY;
3452                 error = ENETUNREACH;
3453         }
3454 
3455         /* Cache things */
3456         if (ixa->ixa_ire != NULL)
3457                 ire_refrele_notr(ixa->ixa_ire);
3458 #ifdef DEBUG
3459         ire_refhold_notr(ire);
3460         ire_refrele(ire);
3461 #endif
3462         ixa->ixa_ire = ire;
3463         ixa->ixa_ire_generation = generation;
3464 
3465         /*
3466          * Ensure that ixa_dce is always set any time that ixa_ire is set,
3467          * since some callers will send a packet to conn_ip_output() even if
3468          * there's an error.
3469          */
3470         if (flags & IPDF_UNIQUE_DCE) {
3471                 /* Fallback to the default dce if allocation fails */
3472                 dce = dce_lookup_and_add_v4(dst_addr, ipst);
3473                 if (dce != NULL)
3474                         generation = dce->dce_generation;
3475                 else
3476                         dce = dce_lookup_v4(dst_addr, ipst, &generation);
3477         } else {
3478                 dce = dce_lookup_v4(dst_addr, ipst, &generation);
3479         }
3480         ASSERT(dce != NULL);
3481         if (ixa->ixa_dce != NULL)
3482                 dce_refrele_notr(ixa->ixa_dce);
3483 #ifdef DEBUG
3484         dce_refhold_notr(dce);
3485         dce_refrele(dce);
3486 #endif
3487         ixa->ixa_dce = dce;
3488         ixa->ixa_dce_generation = generation;
3489 
3490         /*
3491          * For multicast with multirt we have a flag passed back from
3492          * ire_lookup_multi_ill_v4 since we don't have an IRE for each
3493          * possible multicast address.
3494          * We also need a flag for multicast since we can't check
3495          * whether RTF_MULTIRT is set in ixa_ire for multicast.
3496          */
3497         if (multirt) {
3498                 ixa->ixa_postfragfn = ip_postfrag_multirt_v4;
3499                 ixa->ixa_flags |= IXAF_MULTIRT_MULTICAST;
3500         } else {
3501                 ixa->ixa_postfragfn = ire->ire_postfragfn;
3502                 ixa->ixa_flags &= ~IXAF_MULTIRT_MULTICAST;
3503         }
3504         if (!(ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE))) {
3505                 /* Get an nce to cache. */
3506                 nce = ire_to_nce(ire, firsthop, NULL);
3507                 if (nce == NULL) {
3508                         /* Allocation failure? */
3509                         ixa->ixa_ire_generation = IRE_GENERATION_VERIFY;
3510                 } else {
3511                         if (ixa->ixa_nce != NULL)
3512                                 nce_refrele(ixa->ixa_nce);
3513                         ixa->ixa_nce = nce;
3514                 }
3515         }
3516 
3517         /*
3518          * If the source address is a loopback address, the
3519          * destination had best be local or multicast.
3520          * If we are sending to an IRE_LOCAL using a loopback source then
3521          * it had better be the same zoneid.
3522          */
3523         if (*src_addrp == htonl(INADDR_LOOPBACK)) {
3524                 if ((ire->ire_type & IRE_LOCAL) && ire->ire_zoneid != zoneid) {
3525                         ire = NULL;     /* Stored in ixa_ire */
3526                         error = EADDRNOTAVAIL;
3527                         goto bad_addr;
3528                 }
3529                 if (!(ire->ire_type & (IRE_LOOPBACK|IRE_LOCAL|IRE_MULTICAST))) {
3530                         ire = NULL;     /* Stored in ixa_ire */
3531                         error = EADDRNOTAVAIL;
3532                         goto bad_addr;
3533                 }
3534         }
3535         if (ire->ire_type & IRE_BROADCAST) {
3536                 /*
3537                  * If the ULP didn't have a specified source, then we
3538                  * make sure we reselect the source when sending
3539                  * broadcasts out different interfaces.
3540                  */
3541                 if (flags & IPDF_SELECT_SRC)
3542                         ixa->ixa_flags |= IXAF_SET_SOURCE;
3543                 else
3544                         ixa->ixa_flags &= ~IXAF_SET_SOURCE;
3545         }
3546 
3547         /*
3548          * Does the caller want us to pick a source address?
3549          */
3550         if (flags & IPDF_SELECT_SRC) {
3551                 ipaddr_t        src_addr;
3552 
3553                 /*
3554                  * We use use ire_nexthop_ill to avoid the under ipmp
3555                  * interface for source address selection. Note that for ipmp
3556                  * probe packets, ixa_ifindex would have been specified, and
3557                  * the ip_select_route() invocation would have picked an ire
3558                  * will ire_ill pointing at an under interface.
3559                  */
3560                 ill = ire_nexthop_ill(ire);
3561 
3562                 /* If unreachable we have no ill but need some source */
3563                 if (ill == NULL) {
3564                         src_addr = htonl(INADDR_LOOPBACK);
3565                         /* Make sure we look for a better source address */
3566                         generation = SRC_GENERATION_VERIFY;
3567                 } else {
3568                         error = ip_select_source_v4(ill, setsrc, dst_addr,
3569                             ixa->ixa_multicast_ifaddr, zoneid,
3570                             ipst, &src_addr, &generation, NULL);
3571                         if (error != 0) {
3572                                 ire = NULL;     /* Stored in ixa_ire */
3573                                 goto bad_addr;
3574                         }
3575                 }
3576 
3577                 /*
3578                  * We allow the source address to to down.
3579                  * However, we check that we don't use the loopback address
3580                  * as a source when sending out on the wire.
3581                  */
3582                 if ((src_addr == htonl(INADDR_LOOPBACK)) &&
3583                     !(ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK|IRE_MULTICAST)) &&
3584                     !(ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE))) {
3585                         ire = NULL;     /* Stored in ixa_ire */
3586                         error = EADDRNOTAVAIL;
3587                         goto bad_addr;
3588                 }
3589 
3590                 *src_addrp = src_addr;
3591                 ixa->ixa_src_generation = generation;
3592         }
3593 
3594         /*
3595          * Make sure we don't leave an unreachable ixa_nce in place
3596          * since ip_select_route is used when we unplumb i.e., remove
3597          * references on ixa_ire, ixa_nce, and ixa_dce.
3598          */
3599         nce = ixa->ixa_nce;
3600         if (nce != NULL && nce->nce_is_condemned) {
3601                 nce_refrele(nce);
3602                 ixa->ixa_nce = NULL;
3603                 ixa->ixa_ire_generation = IRE_GENERATION_VERIFY;
3604         }
3605 
3606         /*
3607          * The caller has set IXAF_PMTU_DISCOVERY if path MTU is desired.
3608          * However, we can't do it for IPv4 multicast or broadcast.
3609          */
3610         if (ire->ire_type & (IRE_BROADCAST|IRE_MULTICAST))
3611                 ixa->ixa_flags &= ~IXAF_PMTU_DISCOVERY;
3612 
3613         /*
3614          * Set initial value for fragmentation limit. Either conn_ip_output
3615          * or ULP might updates it when there are routing changes.
3616          * Handles a NULL ixa_ire->ire_ill or a NULL ixa_nce for RTF_REJECT.
3617          */
3618         pmtu = ip_get_pmtu(ixa);
3619         ixa->ixa_fragsize = pmtu;
3620         /* Make sure ixa_fragsize and ixa_pmtu remain identical */
3621         if (ixa->ixa_flags & IXAF_VERIFY_PMTU)
3622                 ixa->ixa_pmtu = pmtu;
3623 
3624         /*
3625          * Extract information useful for some transports.
3626          * First we look for DCE metrics. Then we take what we have in
3627          * the metrics in the route, where the offlink is used if we have
3628          * one.
3629          */
3630         if (uinfo != NULL) {
3631                 bzero(uinfo, sizeof (*uinfo));
3632 
3633                 if (dce->dce_flags & DCEF_UINFO)
3634                         *uinfo = dce->dce_uinfo;
3635 
3636                 rts_merge_metrics(uinfo, &ire->ire_metrics);
3637 
3638                 /* Allow ire_metrics to decrease the path MTU from above */
3639                 if (uinfo->iulp_mtu == 0 || uinfo->iulp_mtu > pmtu)
3640                         uinfo->iulp_mtu = pmtu;
3641 
3642                 uinfo->iulp_localnet = (ire->ire_type & IRE_ONLINK) != 0;
3643                 uinfo->iulp_loopback = (ire->ire_type & IRE_LOOPBACK) != 0;
3644                 uinfo->iulp_local = (ire->ire_type & IRE_LOCAL) != 0;
3645         }
3646 
3647         if (ill != NULL)
3648                 ill_refrele(ill);
3649 
3650         return (error);
3651 
3652 bad_addr:
3653         if (ire != NULL)
3654                 ire_refrele(ire);
3655 
3656         if (ill != NULL)
3657                 ill_refrele(ill);
3658 
3659         /*
3660          * Make sure we don't leave an unreachable ixa_nce in place
3661          * since ip_select_route is used when we unplumb i.e., remove
3662          * references on ixa_ire, ixa_nce, and ixa_dce.
3663          */
3664         nce = ixa->ixa_nce;
3665         if (nce != NULL && nce->nce_is_condemned) {
3666                 nce_refrele(nce);
3667                 ixa->ixa_nce = NULL;
3668                 ixa->ixa_ire_generation = IRE_GENERATION_VERIFY;
3669         }
3670 
3671         return (error);
3672 }
3673 
3674 
3675 /*
3676  * Get the base MTU for the case when path MTU discovery is not used.
3677  * Takes the MTU of the IRE into account.
3678  */
3679 uint_t
3680 ip_get_base_mtu(ill_t *ill, ire_t *ire)
3681 {
3682         uint_t mtu;
3683         uint_t iremtu = ire->ire_metrics.iulp_mtu;
3684 
3685         if (ire->ire_type & (IRE_MULTICAST|IRE_BROADCAST))
3686                 mtu = ill->ill_mc_mtu;
3687         else
3688                 mtu = ill->ill_mtu;
3689 
3690         if (iremtu != 0 && iremtu < mtu)
3691                 mtu = iremtu;
3692 
3693         return (mtu);
3694 }
3695 
3696 /*
3697  * Get the PMTU for the attributes. Handles both IPv4 and IPv6.
3698  * Assumes that ixa_ire, dce, and nce have already been set up.
3699  *
3700  * The caller has set IXAF_PMTU_DISCOVERY if path MTU discovery is desired.
3701  * We avoid path MTU discovery if it is disabled with ndd.
3702  * Furtermore, if the path MTU is too small, then we don't set DF for IPv4.
3703  *
3704  * NOTE: We also used to turn it off for source routed packets. That
3705  * is no longer required since the dce is per final destination.
3706  */
3707 uint_t
3708 ip_get_pmtu(ip_xmit_attr_t *ixa)
3709 {
3710         ip_stack_t      *ipst = ixa->ixa_ipst;
3711         dce_t           *dce;
3712         nce_t           *nce;
3713         ire_t           *ire;
3714         uint_t          pmtu;
3715 
3716         ire = ixa->ixa_ire;
3717         dce = ixa->ixa_dce;
3718         nce = ixa->ixa_nce;
3719 
3720         /*
3721          * If path MTU discovery has been turned off by ndd, then we ignore
3722          * any dce_pmtu and for IPv4 we will not set DF.
3723          */
3724         if (!ipst->ips_ip_path_mtu_discovery)
3725                 ixa->ixa_flags &= ~IXAF_PMTU_DISCOVERY;
3726 
3727         pmtu = IP_MAXPACKET;
3728         /*
3729          * Decide whether whether IPv4 sets DF
3730          * For IPv6 "no DF" means to use the 1280 mtu
3731          */
3732         if (ixa->ixa_flags & IXAF_PMTU_DISCOVERY) {
3733                 ixa->ixa_flags |= IXAF_PMTU_IPV4_DF;
3734         } else {
3735                 ixa->ixa_flags &= ~IXAF_PMTU_IPV4_DF;
3736                 if (!(ixa->ixa_flags & IXAF_IS_IPV4))
3737                         pmtu = IPV6_MIN_MTU;
3738         }
3739 
3740         /* Check if the PMTU is to old before we use it */
3741         if ((dce->dce_flags & DCEF_PMTU) &&
3742             TICK_TO_SEC(ddi_get_lbolt64()) - dce->dce_last_change_time >
3743             ipst->ips_ip_pathmtu_interval) {
3744                 /*
3745                  * Older than 20 minutes. Drop the path MTU information.
3746                  */
3747                 mutex_enter(&dce->dce_lock);
3748                 dce->dce_flags &= ~(DCEF_PMTU|DCEF_TOO_SMALL_PMTU);
3749                 dce->dce_last_change_time = TICK_TO_SEC(ddi_get_lbolt64());
3750                 mutex_exit(&dce->dce_lock);
3751                 dce_increment_generation(dce);
3752         }
3753 
3754         /* The metrics on the route can lower the path MTU */
3755         if (ire->ire_metrics.iulp_mtu != 0 &&
3756             ire->ire_metrics.iulp_mtu < pmtu)
3757                 pmtu = ire->ire_metrics.iulp_mtu;
3758 
3759         /*
3760          * If the path MTU is smaller than some minimum, we still use dce_pmtu
3761          * above (would be 576 for IPv4 and 1280 for IPv6), but we clear
3762          * IXAF_PMTU_IPV4_DF so that we avoid setting DF for IPv4.
3763          */
3764         if (ixa->ixa_flags & IXAF_PMTU_DISCOVERY) {
3765                 if (dce->dce_flags & DCEF_PMTU) {
3766                         if (dce->dce_pmtu < pmtu)
3767                                 pmtu = dce->dce_pmtu;
3768 
3769                         if (dce->dce_flags & DCEF_TOO_SMALL_PMTU) {
3770                                 ixa->ixa_flags |= IXAF_PMTU_TOO_SMALL;
3771                                 ixa->ixa_flags &= ~IXAF_PMTU_IPV4_DF;
3772                         } else {
3773                                 ixa->ixa_flags &= ~IXAF_PMTU_TOO_SMALL;
3774                                 ixa->ixa_flags |= IXAF_PMTU_IPV4_DF;
3775                         }
3776                 } else {
3777                         ixa->ixa_flags &= ~IXAF_PMTU_TOO_SMALL;
3778                         ixa->ixa_flags |= IXAF_PMTU_IPV4_DF;
3779                 }
3780         }
3781 
3782         /*
3783          * If we have an IRE_LOCAL we use the loopback mtu instead of
3784          * the ill for going out the wire i.e., IRE_LOCAL gets the same
3785          * mtu as IRE_LOOPBACK.
3786          */
3787         if (ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK)) {
3788                 uint_t loopback_mtu;
3789 
3790                 loopback_mtu = (ire->ire_ipversion == IPV6_VERSION) ?
3791                     ip_loopback_mtu_v6plus : ip_loopback_mtuplus;
3792 
3793                 if (loopback_mtu < pmtu)
3794                         pmtu = loopback_mtu;
3795         } else if (nce != NULL) {
3796                 /*
3797                  * Make sure we don't exceed the interface MTU.
3798                  * In the case of RTF_REJECT or RTF_BLACKHOLE we might not have
3799                  * an ill. We'd use the above IP_MAXPACKET in that case just
3800                  * to tell the transport something larger than zero.
3801                  */
3802                 if (ire->ire_type & (IRE_MULTICAST|IRE_BROADCAST)) {
3803                         if (nce->nce_common->ncec_ill->ill_mc_mtu < pmtu)
3804                                 pmtu = nce->nce_common->ncec_ill->ill_mc_mtu;
3805                         if (nce->nce_common->ncec_ill != nce->nce_ill &&
3806                             nce->nce_ill->ill_mc_mtu < pmtu) {
3807                                 /*
3808                                  * for interfaces in an IPMP group, the mtu of
3809                                  * the nce_ill (under_ill) could be different
3810                                  * from the mtu of the ncec_ill, so we take the
3811                                  * min of the two.
3812                                  */
3813                                 pmtu = nce->nce_ill->ill_mc_mtu;
3814                         }
3815                 } else {
3816                         if (nce->nce_common->ncec_ill->ill_mtu < pmtu)
3817                                 pmtu = nce->nce_common->ncec_ill->ill_mtu;
3818                         if (nce->nce_common->ncec_ill != nce->nce_ill &&
3819                             nce->nce_ill->ill_mtu < pmtu) {
3820                                 /*
3821                                  * for interfaces in an IPMP group, the mtu of
3822                                  * the nce_ill (under_ill) could be different
3823                                  * from the mtu of the ncec_ill, so we take the
3824                                  * min of the two.
3825                                  */
3826                                 pmtu = nce->nce_ill->ill_mtu;
3827                         }
3828                 }
3829         }
3830 
3831         /*
3832          * Handle the IPV6_USE_MIN_MTU socket option or ancillary data.
3833          * Only applies to IPv6.
3834          */
3835         if (!(ixa->ixa_flags & IXAF_IS_IPV4)) {
3836                 if (ixa->ixa_flags & IXAF_USE_MIN_MTU) {
3837                         switch (ixa->ixa_use_min_mtu) {
3838                         case IPV6_USE_MIN_MTU_MULTICAST:
3839                                 if (ire->ire_type & IRE_MULTICAST)
3840                                         pmtu = IPV6_MIN_MTU;
3841                                 break;
3842                         case IPV6_USE_MIN_MTU_ALWAYS:
3843                                 pmtu = IPV6_MIN_MTU;
3844                                 break;
3845                         case IPV6_USE_MIN_MTU_NEVER:
3846                                 break;
3847                         }
3848                 } else {
3849                         /* Default is IPV6_USE_MIN_MTU_MULTICAST */
3850                         if (ire->ire_type & IRE_MULTICAST)
3851                                 pmtu = IPV6_MIN_MTU;
3852                 }
3853         }
3854 
3855         /*
3856          * For multirouted IPv6 packets, the IP layer will insert a 8-byte
3857          * fragment header in every packet. We compensate for those cases by
3858          * returning a smaller path MTU to the ULP.
3859          *
3860          * In the case of CGTP then ip_output will add a fragment header.
3861          * Make sure there is room for it by telling a smaller number
3862          * to the transport.
3863          *
3864          * When IXAF_IPV6_ADDR_FRAGHDR we subtract the frag hdr here
3865          * so the ULPs consistently see a iulp_pmtu and ip_get_pmtu()
3866          * which is the size of the packets it can send.
3867          */
3868         if (!(ixa->ixa_flags & IXAF_IS_IPV4)) {
3869                 if ((ire->ire_flags & RTF_MULTIRT) ||
3870                     (ixa->ixa_flags & IXAF_MULTIRT_MULTICAST)) {
3871                         pmtu -= sizeof (ip6_frag_t);
3872                         ixa->ixa_flags |= IXAF_IPV6_ADD_FRAGHDR;
3873                 }
3874         }
3875 
3876         return (pmtu);
3877 }
3878 
3879 /*
3880  * Carve "len" bytes out of an mblk chain, consuming any we empty, and duping
3881  * the final piece where we don't.  Return a pointer to the first mblk in the
3882  * result, and update the pointer to the next mblk to chew on.  If anything
3883  * goes wrong (i.e., dupb fails), we waste everything in sight and return a
3884  * NULL pointer.
3885  */
3886 mblk_t *
3887 ip_carve_mp(mblk_t **mpp, ssize_t len)
3888 {
3889         mblk_t  *mp0;
3890         mblk_t  *mp1;
3891         mblk_t  *mp2;
3892 
3893         if (!len || !mpp || !(mp0 = *mpp))
3894                 return (NULL);
3895         /* If we aren't going to consume the first mblk, we need a dup. */
3896         if (mp0->b_wptr - mp0->b_rptr > len) {
3897                 mp1 = dupb(mp0);
3898                 if (mp1) {
3899                         /* Partition the data between the two mblks. */
3900                         mp1->b_wptr = mp1->b_rptr + len;
3901                         mp0->b_rptr = mp1->b_wptr;
3902                         /*
3903                          * after adjustments if mblk not consumed is now
3904                          * unaligned, try to align it. If this fails free
3905                          * all messages and let upper layer recover.
3906                          */
3907                         if (!OK_32PTR(mp0->b_rptr)) {
3908                                 if (!pullupmsg(mp0, -1)) {
3909                                         freemsg(mp0);
3910                                         freemsg(mp1);
3911                                         *mpp = NULL;
3912                                         return (NULL);
3913                                 }
3914                         }
3915                 }
3916                 return (mp1);
3917         }
3918         /* Eat through as many mblks as we need to get len bytes. */
3919         len -= mp0->b_wptr - mp0->b_rptr;
3920         for (mp2 = mp1 = mp0; (mp2 = mp2->b_cont) != 0 && len; mp1 = mp2) {
3921                 if (mp2->b_wptr - mp2->b_rptr > len) {
3922                         /*
3923                          * We won't consume the entire last mblk.  Like
3924                          * above, dup and partition it.
3925                          */
3926                         mp1->b_cont = dupb(mp2);
3927                         mp1 = mp1->b_cont;
3928                         if (!mp1) {
3929                                 /*
3930                                  * Trouble.  Rather than go to a lot of
3931                                  * trouble to clean up, we free the messages.
3932                                  * This won't be any worse than losing it on
3933                                  * the wire.
3934                                  */
3935                                 freemsg(mp0);
3936                                 freemsg(mp2);
3937                                 *mpp = NULL;
3938                                 return (NULL);
3939                         }
3940                         mp1->b_wptr = mp1->b_rptr + len;
3941                         mp2->b_rptr = mp1->b_wptr;
3942                         /*
3943                          * after adjustments if mblk not consumed is now
3944                          * unaligned, try to align it. If this fails free
3945                          * all messages and let upper layer recover.
3946                          */
3947                         if (!OK_32PTR(mp2->b_rptr)) {
3948                                 if (!pullupmsg(mp2, -1)) {
3949                                         freemsg(mp0);
3950                                         freemsg(mp2);
3951                                         *mpp = NULL;
3952                                         return (NULL);
3953                                 }
3954                         }
3955                         *mpp = mp2;
3956                         return (mp0);
3957                 }
3958                 /* Decrement len by the amount we just got. */
3959                 len -= mp2->b_wptr - mp2->b_rptr;
3960         }
3961         /*
3962          * len should be reduced to zero now.  If not our caller has
3963          * screwed up.
3964          */
3965         if (len) {
3966                 /* Shouldn't happen! */
3967                 freemsg(mp0);
3968                 *mpp = NULL;
3969                 return (NULL);
3970         }
3971         /*
3972          * We consumed up to exactly the end of an mblk.  Detach the part
3973          * we are returning from the rest of the chain.
3974          */
3975         mp1->b_cont = NULL;
3976         *mpp = mp2;
3977         return (mp0);
3978 }
3979 
3980 /* The ill stream is being unplumbed. Called from ip_close */
3981 int
3982 ip_modclose(ill_t *ill)
3983 {
3984         boolean_t success;
3985         ipsq_t  *ipsq;
3986         ipif_t  *ipif;
3987         queue_t *q = ill->ill_rq;
3988         ip_stack_t      *ipst = ill->ill_ipst;
3989         int     i;
3990         arl_ill_common_t *ai = ill->ill_common;
3991 
3992         /*
3993          * The punlink prior to this may have initiated a capability
3994          * negotiation. But ipsq_enter will block until that finishes or
3995          * times out.
3996          */
3997         success = ipsq_enter(ill, B_FALSE, NEW_OP);
3998 
3999         /*
4000          * Open/close/push/pop is guaranteed to be single threaded
4001          * per stream by STREAMS. FS guarantees that all references
4002          * from top are gone before close is called. So there can't
4003          * be another close thread that has set CONDEMNED on this ill.
4004          * and cause ipsq_enter to return failure.
4005          */
4006         ASSERT(success);
4007         ipsq = ill->ill_phyint->phyint_ipsq;
4008 
4009         /*
4010          * Mark it condemned. No new reference will be made to this ill.
4011          * Lookup functions will return an error. Threads that try to
4012          * increment the refcnt must check for ILL_CAN_LOOKUP. This ensures
4013          * that the refcnt will drop down to zero.
4014          */
4015         mutex_enter(&ill->ill_lock);
4016         ill->ill_state_flags |= ILL_CONDEMNED;
4017         for (ipif = ill->ill_ipif; ipif != NULL;
4018             ipif = ipif->ipif_next) {
4019                 ipif->ipif_state_flags |= IPIF_CONDEMNED;
4020         }
4021         /*
4022          * Wake up anybody waiting to enter the ipsq. ipsq_enter
4023          * returns  error if ILL_CONDEMNED is set
4024          */
4025         cv_broadcast(&ill->ill_cv);
4026         mutex_exit(&ill->ill_lock);
4027 
4028         /*
4029          * Send all the deferred DLPI messages downstream which came in
4030          * during the small window right before ipsq_enter(). We do this
4031          * without waiting for the ACKs because all the ACKs for M_PROTO
4032          * messages are ignored in ip_rput() when ILL_CONDEMNED is set.
4033          */
4034         ill_dlpi_send_deferred(ill);
4035 
4036         /*
4037          * Shut down fragmentation reassembly.
4038          * ill_frag_timer won't start a timer again.
4039          * Now cancel any existing timer
4040          */
4041         (void) untimeout(ill->ill_frag_timer_id);
4042         (void) ill_frag_timeout(ill, 0);
4043 
4044         /*
4045          * Call ill_delete to bring down the ipifs, ilms and ill on
4046          * this ill. Then wait for the refcnts to drop to zero.
4047          * ill_is_freeable checks whether the ill is really quiescent.
4048          * Then make sure that threads that are waiting to enter the
4049          * ipsq have seen the error returned by ipsq_enter and have
4050          * gone away. Then we call ill_delete_tail which does the
4051          * DL_UNBIND_REQ with the driver and then qprocsoff.
4052          */
4053         ill_delete(ill);
4054         mutex_enter(&ill->ill_lock);
4055         while (!ill_is_freeable(ill))
4056                 cv_wait(&ill->ill_cv, &ill->ill_lock);
4057 
4058         while (ill->ill_waiters)
4059                 cv_wait(&ill->ill_cv, &ill->ill_lock);
4060 
4061         mutex_exit(&ill->ill_lock);
4062 
4063         /*
4064          * ill_delete_tail drops reference on ill_ipst, but we need to keep
4065          * it held until the end of the function since the cleanup
4066          * below needs to be able to use the ip_stack_t.
4067          */
4068         netstack_hold(ipst->ips_netstack);
4069 
4070         /* qprocsoff is done via ill_delete_tail */
4071         ill_delete_tail(ill);
4072         /*
4073          * synchronously wait for arp stream to unbind. After this, we
4074          * cannot get any data packets up from the driver.
4075          */
4076         arp_unbind_complete(ill);
4077         ASSERT(ill->ill_ipst == NULL);
4078 
4079         /*
4080          * Walk through all conns and qenable those that have queued data.
4081          * Close synchronization needs this to
4082          * be done to ensure that all upper layers blocked
4083          * due to flow control to the closing device
4084          * get unblocked.
4085          */
4086         ip1dbg(("ip_wsrv: walking\n"));
4087         for (i = 0; i < TX_FANOUT_SIZE; i++) {
4088                 conn_walk_drain(ipst, &ipst->ips_idl_tx_list[i]);
4089         }
4090 
4091         /*
4092          * ai can be null if this is an IPv6 ill, or if the IPv4
4093          * stream is being torn down before ARP was plumbed (e.g.,
4094          * /sbin/ifconfig plumbing a stream twice, and encountering
4095          * an error
4096          */
4097         if (ai != NULL) {
4098                 ASSERT(!ill->ill_isv6);
4099                 mutex_enter(&ai->ai_lock);
4100                 ai->ai_ill = NULL;
4101                 if (ai->ai_arl == NULL) {
4102                         mutex_destroy(&ai->ai_lock);
4103                         kmem_free(ai, sizeof (*ai));
4104                 } else {
4105                         cv_signal(&ai->ai_ill_unplumb_done);
4106                         mutex_exit(&ai->ai_lock);
4107                 }
4108         }
4109 
4110         mutex_enter(&ipst->ips_ip_mi_lock);
4111         mi_close_unlink(&ipst->ips_ip_g_head, (IDP)ill);
4112         mutex_exit(&ipst->ips_ip_mi_lock);
4113 
4114         /*
4115          * credp could be null if the open didn't succeed and ip_modopen
4116          * itself calls ip_close.
4117          */
4118         if (ill->ill_credp != NULL)
4119                 crfree(ill->ill_credp);
4120 
4121         mutex_destroy(&ill->ill_saved_ire_lock);
4122         mutex_destroy(&ill->ill_lock);
4123         rw_destroy(&ill->ill_mcast_lock);
4124         mutex_destroy(&ill->ill_mcast_serializer);
4125         list_destroy(&ill->ill_nce);
4126 
4127         /*
4128          * Now we are done with the module close pieces that
4129          * need the netstack_t.
4130          */
4131         netstack_rele(ipst->ips_netstack);
4132 
4133         mi_close_free((IDP)ill);
4134         q->q_ptr = WR(q)->q_ptr = NULL;
4135 
4136         ipsq_exit(ipsq);
4137 
4138         return (0);
4139 }
4140 
4141 /*
4142  * This is called as part of close() for IP, UDP, ICMP, and RTS
4143  * in order to quiesce the conn.
4144  */
4145 void
4146 ip_quiesce_conn(conn_t *connp)
4147 {
4148         boolean_t       drain_cleanup_reqd = B_FALSE;
4149         boolean_t       conn_ioctl_cleanup_reqd = B_FALSE;
4150         boolean_t       ilg_cleanup_reqd = B_FALSE;
4151         ip_stack_t      *ipst;
4152 
4153         ASSERT(!IPCL_IS_TCP(connp));
4154         ipst = connp->conn_netstack->netstack_ip;
4155 
4156         /*
4157          * Mark the conn as closing, and this conn must not be
4158          * inserted in future into any list. Eg. conn_drain_insert(),
4159          * won't insert this conn into the conn_drain_list.
4160          *
4161          * conn_idl, and conn_ilg cannot get set henceforth.
4162          */
4163         mutex_enter(&connp->conn_lock);
4164         ASSERT(!(connp->conn_state_flags & CONN_QUIESCED));
4165         connp->conn_state_flags |= CONN_CLOSING;
4166         if (connp->conn_idl != NULL)
4167                 drain_cleanup_reqd = B_TRUE;
4168         if (connp->conn_oper_pending_ill != NULL)
4169                 conn_ioctl_cleanup_reqd = B_TRUE;
4170         if (connp->conn_dhcpinit_ill != NULL) {
4171                 ASSERT(connp->conn_dhcpinit_ill->ill_dhcpinit != 0);
4172                 atomic_dec_32(&connp->conn_dhcpinit_ill->ill_dhcpinit);
4173                 ill_set_inputfn(connp->conn_dhcpinit_ill);
4174                 connp->conn_dhcpinit_ill = NULL;
4175         }
4176         if (connp->conn_ilg != NULL)
4177                 ilg_cleanup_reqd = B_TRUE;
4178         mutex_exit(&connp->conn_lock);
4179 
4180         if (conn_ioctl_cleanup_reqd)
4181                 conn_ioctl_cleanup(connp);
4182 
4183         if (is_system_labeled() && connp->conn_anon_port) {
4184                 (void) tsol_mlp_anon(crgetzone(connp->conn_cred),
4185                     connp->conn_mlp_type, connp->conn_proto,
4186                     ntohs(connp->conn_lport), B_FALSE);
4187                 connp->conn_anon_port = 0;
4188         }
4189         connp->conn_mlp_type = mlptSingle;
4190 
4191         /*
4192          * Remove this conn from any fanout list it is on.
4193          * and then wait for any threads currently operating
4194          * on this endpoint to finish
4195          */
4196         ipcl_hash_remove(connp);
4197 
4198         /*
4199          * Remove this conn from the drain list, and do any other cleanup that
4200          * may be required.  (TCP conns are never flow controlled, and
4201          * conn_idl will be NULL.)
4202          */
4203         if (drain_cleanup_reqd && connp->conn_idl != NULL) {
4204                 idl_t *idl = connp->conn_idl;
4205 
4206                 mutex_enter(&idl->idl_lock);
4207                 conn_drain(connp, B_TRUE);
4208                 mutex_exit(&idl->idl_lock);
4209         }
4210 
4211         if (connp == ipst->ips_ip_g_mrouter)
4212                 (void) ip_mrouter_done(ipst);
4213 
4214         if (ilg_cleanup_reqd)
4215                 ilg_delete_all(connp);
4216 
4217         /*
4218          * Now conn refcnt can increase only thru CONN_INC_REF_LOCKED.
4219          * callers from write side can't be there now because close
4220          * is in progress. The only other caller is ipcl_walk
4221          * which checks for the condemned flag.
4222          */
4223         mutex_enter(&connp->conn_lock);
4224         connp->conn_state_flags |= CONN_CONDEMNED;
4225         while (connp->conn_ref != 1)
4226                 cv_wait(&connp->conn_cv, &connp->conn_lock);
4227         connp->conn_state_flags |= CONN_QUIESCED;
4228         mutex_exit(&connp->conn_lock);
4229 }
4230 
4231 /* ARGSUSED */
4232 int
4233 ip_close(queue_t *q, int flags, cred_t *credp __unused)
4234 {
4235         conn_t          *connp;
4236 
4237         /*
4238          * Call the appropriate delete routine depending on whether this is
4239          * a module or device.
4240          */
4241         if (WR(q)->q_next != NULL) {
4242                 /* This is a module close */
4243                 return (ip_modclose((ill_t *)q->q_ptr));
4244         }
4245 
4246         connp = q->q_ptr;
4247         ip_quiesce_conn(connp);
4248 
4249         qprocsoff(q);
4250 
4251         /*
4252          * Now we are truly single threaded on this stream, and can
4253          * delete the things hanging off the connp, and finally the connp.
4254          * We removed this connp from the fanout list, it cannot be
4255          * accessed thru the fanouts, and we already waited for the
4256          * conn_ref to drop to 0. We are already in close, so
4257          * there cannot be any other thread from the top. qprocsoff
4258          * has completed, and service has completed or won't run in
4259          * future.
4260          */
4261         ASSERT(connp->conn_ref == 1);
4262 
4263         inet_minor_free(connp->conn_minor_arena, connp->conn_dev);
4264 
4265         connp->conn_ref--;
4266         ipcl_conn_destroy(connp);
4267 
4268         q->q_ptr = WR(q)->q_ptr = NULL;
4269         return (0);
4270 }
4271 
4272 /*
4273  * Wapper around putnext() so that ip_rts_request can merely use
4274  * conn_recv.
4275  */
4276 /*ARGSUSED2*/
4277 static void
4278 ip_conn_input(void *arg1, mblk_t *mp, void *arg2, ip_recv_attr_t *ira)
4279 {
4280         conn_t *connp = (conn_t *)arg1;
4281 
4282         putnext(connp->conn_rq, mp);
4283 }
4284 
4285 /* Dummy in case ICMP error delivery is attempted to a /dev/ip instance */
4286 /* ARGSUSED */
4287 static void
4288 ip_conn_input_icmp(void *arg1, mblk_t *mp, void *arg2, ip_recv_attr_t *ira)
4289 {
4290         freemsg(mp);
4291 }
4292 
4293 /*
4294  * Called when the module is about to be unloaded
4295  */
4296 void
4297 ip_ddi_destroy(void)
4298 {
4299         /* This needs to be called before destroying any transports. */
4300         mutex_enter(&cpu_lock);
4301         unregister_cpu_setup_func(ip_tp_cpu_update, NULL);
4302         mutex_exit(&cpu_lock);
4303 
4304         tnet_fini();
4305 
4306         icmp_ddi_g_destroy();
4307         rts_ddi_g_destroy();
4308         udp_ddi_g_destroy();
4309         sctp_ddi_g_destroy();
4310         tcp_ddi_g_destroy();
4311         ilb_ddi_g_destroy();
4312         dce_g_destroy();
4313         ipsec_policy_g_destroy();
4314         ipcl_g_destroy();
4315         ip_net_g_destroy();
4316         ip_ire_g_fini();
4317         inet_minor_destroy(ip_minor_arena_sa);
4318 #if defined(_LP64)
4319         inet_minor_destroy(ip_minor_arena_la);
4320 #endif
4321 
4322 #ifdef DEBUG
4323         list_destroy(&ip_thread_list);
4324         rw_destroy(&ip_thread_rwlock);
4325         tsd_destroy(&ip_thread_data);
4326 #endif
4327 
4328         netstack_unregister(NS_IP);
4329 }
4330 
4331 /*
4332  * First step in cleanup.
4333  */
4334 /* ARGSUSED */
4335 static void
4336 ip_stack_shutdown(netstackid_t stackid, void *arg)
4337 {
4338         ip_stack_t *ipst = (ip_stack_t *)arg;
4339         kt_did_t ktid;
4340 
4341 #ifdef NS_DEBUG
4342         printf("ip_stack_shutdown(%p, stack %d)\n", (void *)ipst, stackid);
4343 #endif
4344 
4345         /*
4346          * Perform cleanup for special interfaces (loopback and IPMP).
4347          */
4348         ip_interface_cleanup(ipst);
4349 
4350         /*
4351          * The *_hook_shutdown()s start the process of notifying any
4352          * consumers that things are going away.... nothing is destroyed.
4353          */
4354         ipv4_hook_shutdown(ipst);
4355         ipv6_hook_shutdown(ipst);
4356         arp_hook_shutdown(ipst);
4357 
4358         mutex_enter(&ipst->ips_capab_taskq_lock);
4359         ktid = ipst->ips_capab_taskq_thread->t_did;
4360         ipst->ips_capab_taskq_quit = B_TRUE;
4361         cv_signal(&ipst->ips_capab_taskq_cv);
4362         mutex_exit(&ipst->ips_capab_taskq_lock);
4363 
4364         /*
4365          * In rare occurrences, particularly on virtual hardware where CPUs can
4366          * be de-scheduled, the thread that we just signaled will not run until
4367          * after we have gotten through parts of ip_stack_fini. If that happens
4368          * then we'll try to grab the ips_capab_taskq_lock as part of returning
4369          * from cv_wait which no longer exists.
4370          */
4371         thread_join(ktid);
4372 }
4373 
4374 /*
4375  * Free the IP stack instance.
4376  */
4377 static void
4378 ip_stack_fini(netstackid_t stackid, void *arg)
4379 {
4380         ip_stack_t *ipst = (ip_stack_t *)arg;
4381         int ret;
4382 
4383 #ifdef NS_DEBUG
4384         printf("ip_stack_fini(%p, stack %d)\n", (void *)ipst, stackid);
4385 #endif
4386         /*
4387          * At this point, all of the notifications that the events and
4388          * protocols are going away have been run, meaning that we can
4389          * now set about starting to clean things up.
4390          */
4391         ipobs_fini(ipst);
4392         ipv4_hook_destroy(ipst);
4393         ipv6_hook_destroy(ipst);
4394         arp_hook_destroy(ipst);
4395         ip_net_destroy(ipst);
4396 
4397         ipmp_destroy(ipst);
4398 
4399         ip_kstat_fini(stackid, ipst->ips_ip_mibkp);
4400         ipst->ips_ip_mibkp = NULL;
4401         icmp_kstat_fini(stackid, ipst->ips_icmp_mibkp);
4402         ipst->ips_icmp_mibkp = NULL;
4403         ip_kstat2_fini(stackid, ipst->ips_ip_kstat);
4404         ipst->ips_ip_kstat = NULL;
4405         bzero(&ipst->ips_ip_statistics, sizeof (ipst->ips_ip_statistics));
4406         ip6_kstat_fini(stackid, ipst->ips_ip6_kstat);
4407         ipst->ips_ip6_kstat = NULL;
4408         bzero(&ipst->ips_ip6_statistics, sizeof (ipst->ips_ip6_statistics));
4409 
4410         kmem_free(ipst->ips_propinfo_tbl,
4411             ip_propinfo_count * sizeof (mod_prop_info_t));
4412         ipst->ips_propinfo_tbl = NULL;
4413 
4414         dce_stack_destroy(ipst);
4415         ip_mrouter_stack_destroy(ipst);
4416 
4417         /*
4418          * Quiesce all of our timers. Note we set the quiesce flags before we
4419          * call untimeout. The slowtimers may actually kick off another instance
4420          * of the non-slow timers.
4421          */
4422         mutex_enter(&ipst->ips_igmp_timer_lock);
4423         ipst->ips_igmp_timer_quiesce = B_TRUE;
4424         mutex_exit(&ipst->ips_igmp_timer_lock);
4425 
4426         mutex_enter(&ipst->ips_mld_timer_lock);
4427         ipst->ips_mld_timer_quiesce = B_TRUE;
4428         mutex_exit(&ipst->ips_mld_timer_lock);
4429 
4430         mutex_enter(&ipst->ips_igmp_slowtimeout_lock);
4431         ipst->ips_igmp_slowtimeout_quiesce = B_TRUE;
4432         mutex_exit(&ipst->ips_igmp_slowtimeout_lock);
4433 
4434         mutex_enter(&ipst->ips_mld_slowtimeout_lock);
4435         ipst->ips_mld_slowtimeout_quiesce = B_TRUE;
4436         mutex_exit(&ipst->ips_mld_slowtimeout_lock);
4437 
4438         ret = untimeout(ipst->ips_igmp_timeout_id);
4439         if (ret == -1) {
4440                 ASSERT(ipst->ips_igmp_timeout_id == 0);
4441         } else {
4442                 ASSERT(ipst->ips_igmp_timeout_id != 0);
4443                 ipst->ips_igmp_timeout_id = 0;
4444         }
4445         ret = untimeout(ipst->ips_igmp_slowtimeout_id);
4446         if (ret == -1) {
4447                 ASSERT(ipst->ips_igmp_slowtimeout_id == 0);
4448         } else {
4449                 ASSERT(ipst->ips_igmp_slowtimeout_id != 0);
4450                 ipst->ips_igmp_slowtimeout_id = 0;
4451         }
4452         ret = untimeout(ipst->ips_mld_timeout_id);
4453         if (ret == -1) {
4454                 ASSERT(ipst->ips_mld_timeout_id == 0);
4455         } else {
4456                 ASSERT(ipst->ips_mld_timeout_id != 0);
4457                 ipst->ips_mld_timeout_id = 0;
4458         }
4459         ret = untimeout(ipst->ips_mld_slowtimeout_id);
4460         if (ret == -1) {
4461                 ASSERT(ipst->ips_mld_slowtimeout_id == 0);
4462         } else {
4463                 ASSERT(ipst->ips_mld_slowtimeout_id != 0);
4464                 ipst->ips_mld_slowtimeout_id = 0;
4465         }
4466 
4467         ip_ire_fini(ipst);
4468         ip6_asp_free(ipst);
4469         conn_drain_fini(ipst);
4470         ipcl_destroy(ipst);
4471 
4472         mutex_destroy(&ipst->ips_ndp4->ndp_g_lock);
4473         mutex_destroy(&ipst->ips_ndp6->ndp_g_lock);
4474         kmem_free(ipst->ips_ndp4, sizeof (ndp_g_t));
4475         ipst->ips_ndp4 = NULL;
4476         kmem_free(ipst->ips_ndp6, sizeof (ndp_g_t));
4477         ipst->ips_ndp6 = NULL;
4478 
4479         if (ipst->ips_loopback_ksp != NULL) {
4480                 kstat_delete_netstack(ipst->ips_loopback_ksp, stackid);
4481                 ipst->ips_loopback_ksp = NULL;
4482         }
4483 
4484         mutex_destroy(&ipst->ips_capab_taskq_lock);
4485         cv_destroy(&ipst->ips_capab_taskq_cv);
4486 
4487         rw_destroy(&ipst->ips_srcid_lock);
4488 
4489         mutex_destroy(&ipst->ips_ip_mi_lock);
4490         rw_destroy(&ipst->ips_ill_g_usesrc_lock);
4491 
4492         mutex_destroy(&ipst->ips_igmp_timer_lock);
4493         mutex_destroy(&ipst->ips_mld_timer_lock);
4494         mutex_destroy(&ipst->ips_igmp_slowtimeout_lock);
4495         mutex_destroy(&ipst->ips_mld_slowtimeout_lock);
4496         mutex_destroy(&ipst->ips_ip_addr_avail_lock);
4497         rw_destroy(&ipst->ips_ill_g_lock);
4498 
4499         kmem_free(ipst->ips_phyint_g_list, sizeof (phyint_list_t));
4500         ipst->ips_phyint_g_list = NULL;
4501         kmem_free(ipst->ips_ill_g_heads, sizeof (ill_g_head_t) * MAX_G_HEADS);
4502         ipst->ips_ill_g_heads = NULL;
4503 
4504         ldi_ident_release(ipst->ips_ldi_ident);
4505         kmem_free(ipst, sizeof (*ipst));
4506 }
4507 
4508 /*
4509  * This function is called from the TSD destructor, and is used to debug
4510  * reference count issues in IP. See block comment in <inet/ip_if.h> for
4511  * details.
4512  */
4513 static void
4514 ip_thread_exit(void *phash)
4515 {
4516         th_hash_t *thh = phash;
4517 
4518         rw_enter(&ip_thread_rwlock, RW_WRITER);
4519         list_remove(&ip_thread_list, thh);
4520         rw_exit(&ip_thread_rwlock);
4521         mod_hash_destroy_hash(thh->thh_hash);
4522         kmem_free(thh, sizeof (*thh));
4523 }
4524 
4525 /*
4526  * Called when the IP kernel module is loaded into the kernel
4527  */
4528 void
4529 ip_ddi_init(void)
4530 {
4531         ip_squeue_flag = ip_squeue_switch(ip_squeue_enter);
4532 
4533         /*
4534          * For IP and TCP the minor numbers should start from 2 since we have 4
4535          * initial devices: ip, ip6, tcp, tcp6.
4536          */
4537         /*
4538          * If this is a 64-bit kernel, then create two separate arenas -
4539          * one for TLIs in the range of INET_MIN_DEV+2 through 2^^18-1, and the
4540          * other for socket apps in the range 2^^18 through 2^^32-1.
4541          */
4542         ip_minor_arena_la = NULL;
4543         ip_minor_arena_sa = NULL;
4544 #if defined(_LP64)
4545         if ((ip_minor_arena_sa = inet_minor_create("ip_minor_arena_sa",
4546             INET_MIN_DEV + 2, MAXMIN32, KM_SLEEP)) == NULL) {
4547                 cmn_err(CE_PANIC,
4548                     "ip_ddi_init: ip_minor_arena_sa creation failed\n");
4549         }
4550         if ((ip_minor_arena_la = inet_minor_create("ip_minor_arena_la",
4551             MAXMIN32 + 1, MAXMIN64, KM_SLEEP)) == NULL) {
4552                 cmn_err(CE_PANIC,
4553                     "ip_ddi_init: ip_minor_arena_la creation failed\n");
4554         }
4555 #else
4556         if ((ip_minor_arena_sa = inet_minor_create("ip_minor_arena_sa",
4557             INET_MIN_DEV + 2, MAXMIN, KM_SLEEP)) == NULL) {
4558                 cmn_err(CE_PANIC,
4559                     "ip_ddi_init: ip_minor_arena_sa creation failed\n");
4560         }
4561 #endif
4562         ip_poll_normal_ticks = MSEC_TO_TICK_ROUNDUP(ip_poll_normal_ms);
4563 
4564         ipcl_g_init();
4565         ip_ire_g_init();
4566         ip_net_g_init();
4567 
4568 #ifdef DEBUG
4569         tsd_create(&ip_thread_data, ip_thread_exit);
4570         rw_init(&ip_thread_rwlock, NULL, RW_DEFAULT, NULL);
4571         list_create(&ip_thread_list, sizeof (th_hash_t),
4572             offsetof(th_hash_t, thh_link));
4573 #endif
4574         ipsec_policy_g_init();
4575         tcp_ddi_g_init();
4576         sctp_ddi_g_init();
4577         dce_g_init();
4578 
4579         /*
4580          * We want to be informed each time a stack is created or
4581          * destroyed in the kernel, so we can maintain the
4582          * set of udp_stack_t's.
4583          */
4584         netstack_register(NS_IP, ip_stack_init, ip_stack_shutdown,
4585             ip_stack_fini);
4586 
4587         tnet_init();
4588 
4589         udp_ddi_g_init();
4590         rts_ddi_g_init();
4591         icmp_ddi_g_init();
4592         ilb_ddi_g_init();
4593 
4594         /* This needs to be called after all transports are initialized. */
4595         mutex_enter(&cpu_lock);
4596         register_cpu_setup_func(ip_tp_cpu_update, NULL);
4597         mutex_exit(&cpu_lock);
4598 }
4599 
4600 /*
4601  * Initialize the IP stack instance.
4602  */
4603 static void *
4604 ip_stack_init(netstackid_t stackid, netstack_t *ns)
4605 {
4606         ip_stack_t      *ipst;
4607         size_t          arrsz;
4608         major_t         major;
4609 
4610 #ifdef NS_DEBUG
4611         printf("ip_stack_init(stack %d)\n", stackid);
4612 #endif
4613 
4614         ipst = (ip_stack_t *)kmem_zalloc(sizeof (*ipst), KM_SLEEP);
4615         ipst->ips_netstack = ns;
4616 
4617         ipst->ips_ill_g_heads = kmem_zalloc(sizeof (ill_g_head_t) * MAX_G_HEADS,
4618             KM_SLEEP);
4619         ipst->ips_phyint_g_list = kmem_zalloc(sizeof (phyint_list_t),
4620             KM_SLEEP);
4621         ipst->ips_ndp4 = kmem_zalloc(sizeof (ndp_g_t), KM_SLEEP);
4622         ipst->ips_ndp6 = kmem_zalloc(sizeof (ndp_g_t), KM_SLEEP);
4623         mutex_init(&ipst->ips_ndp4->ndp_g_lock, NULL, MUTEX_DEFAULT, NULL);
4624         mutex_init(&ipst->ips_ndp6->ndp_g_lock, NULL, MUTEX_DEFAULT, NULL);
4625 
4626         mutex_init(&ipst->ips_igmp_timer_lock, NULL, MUTEX_DEFAULT, NULL);
4627         ipst->ips_igmp_deferred_next = INFINITY;
4628         mutex_init(&ipst->ips_mld_timer_lock, NULL, MUTEX_DEFAULT, NULL);
4629         ipst->ips_mld_deferred_next = INFINITY;
4630         mutex_init(&ipst->ips_igmp_slowtimeout_lock, NULL, MUTEX_DEFAULT, NULL);
4631         mutex_init(&ipst->ips_mld_slowtimeout_lock, NULL, MUTEX_DEFAULT, NULL);
4632         mutex_init(&ipst->ips_ip_mi_lock, NULL, MUTEX_DEFAULT, NULL);
4633         mutex_init(&ipst->ips_ip_addr_avail_lock, NULL, MUTEX_DEFAULT, NULL);
4634         rw_init(&ipst->ips_ill_g_lock, NULL, RW_DEFAULT, NULL);
4635         rw_init(&ipst->ips_ill_g_usesrc_lock, NULL, RW_DEFAULT, NULL);
4636 
4637         ipcl_init(ipst);
4638         ip_ire_init(ipst);
4639         ip6_asp_init(ipst);
4640         ipif_init(ipst);
4641         conn_drain_init(ipst);
4642         ip_mrouter_stack_init(ipst);
4643         dce_stack_init(ipst);
4644 
4645         ipst->ips_ip_multirt_log_interval = 1000;
4646 
4647         ipst->ips_ill_index = 1;
4648 
4649         ipst->ips_saved_ip_forwarding = -1;
4650         ipst->ips_reg_vif_num = ALL_VIFS;    /* Index to Register vif */
4651 
4652         arrsz = ip_propinfo_count * sizeof (mod_prop_info_t);
4653         ipst->ips_propinfo_tbl = (mod_prop_info_t *)kmem_alloc(arrsz, KM_SLEEP);
4654         bcopy(ip_propinfo_tbl, ipst->ips_propinfo_tbl, arrsz);
4655 
4656         ipst->ips_ip_mibkp = ip_kstat_init(stackid, ipst);
4657         ipst->ips_icmp_mibkp = icmp_kstat_init(stackid);
4658         ipst->ips_ip_kstat = ip_kstat2_init(stackid, &ipst->ips_ip_statistics);
4659         ipst->ips_ip6_kstat =
4660             ip6_kstat_init(stackid, &ipst->ips_ip6_statistics);
4661 
4662         ipst->ips_ip_src_id = 1;
4663         rw_init(&ipst->ips_srcid_lock, NULL, RW_DEFAULT, NULL);
4664 
4665         ipst->ips_src_generation = SRC_GENERATION_INITIAL;
4666 
4667         ip_net_init(ipst, ns);
4668         ipv4_hook_init(ipst);
4669         ipv6_hook_init(ipst);
4670         arp_hook_init(ipst);
4671         ipmp_init(ipst);
4672         ipobs_init(ipst);
4673 
4674         /*
4675          * Create the taskq dispatcher thread and initialize related stuff.
4676          */
4677         mutex_init(&ipst->ips_capab_taskq_lock, NULL, MUTEX_DEFAULT, NULL);
4678         cv_init(&ipst->ips_capab_taskq_cv, NULL, CV_DEFAULT, NULL);
4679         ipst->ips_capab_taskq_thread = thread_create(NULL, 0,
4680             ill_taskq_dispatch, ipst, 0, &p0, TS_RUN, minclsyspri);
4681 
4682         major = mod_name_to_major(INET_NAME);
4683         (void) ldi_ident_from_major(major, &ipst->ips_ldi_ident);
4684         return (ipst);
4685 }
4686 
4687 /*
4688  * Allocate and initialize a DLPI template of the specified length.  (May be
4689  * called as writer.)
4690  */
4691 mblk_t *
4692 ip_dlpi_alloc(size_t len, t_uscalar_t prim)
4693 {
4694         mblk_t  *mp;
4695 
4696         mp = allocb(len, BPRI_MED);
4697         if (!mp)
4698                 return (NULL);
4699 
4700         /*
4701          * DLPIv2 says that DL_INFO_REQ and DL_TOKEN_REQ (the latter
4702          * of which we don't seem to use) are sent with M_PCPROTO, and
4703          * that other DLPI are M_PROTO.
4704          */
4705         if (prim == DL_INFO_REQ) {
4706                 mp->b_datap->db_type = M_PCPROTO;
4707         } else {
4708                 mp->b_datap->db_type = M_PROTO;
4709         }
4710 
4711         mp->b_wptr = mp->b_rptr + len;
4712         bzero(mp->b_rptr, len);
4713         ((dl_unitdata_req_t *)mp->b_rptr)->dl_primitive = prim;
4714         return (mp);
4715 }
4716 
4717 /*
4718  * Allocate and initialize a DLPI notification.  (May be called as writer.)
4719  */
4720 mblk_t *
4721 ip_dlnotify_alloc(uint_t notification, uint_t data)
4722 {
4723         dl_notify_ind_t *notifyp;
4724         mblk_t          *mp;
4725 
4726         if ((mp = ip_dlpi_alloc(DL_NOTIFY_IND_SIZE, DL_NOTIFY_IND)) == NULL)
4727                 return (NULL);
4728 
4729         notifyp = (dl_notify_ind_t *)mp->b_rptr;
4730         notifyp->dl_notification = notification;
4731         notifyp->dl_data = data;
4732         return (mp);
4733 }
4734 
4735 mblk_t *
4736 ip_dlnotify_alloc2(uint_t notification, uint_t data1, uint_t data2)
4737 {
4738         dl_notify_ind_t *notifyp;
4739         mblk_t          *mp;
4740 
4741         if ((mp = ip_dlpi_alloc(DL_NOTIFY_IND_SIZE, DL_NOTIFY_IND)) == NULL)
4742                 return (NULL);
4743 
4744         notifyp = (dl_notify_ind_t *)mp->b_rptr;
4745         notifyp->dl_notification = notification;
4746         notifyp->dl_data1 = data1;
4747         notifyp->dl_data2 = data2;
4748         return (mp);
4749 }
4750 
4751 /*
4752  * Debug formatting routine.  Returns a character string representation of the
4753  * addr in buf, of the form xxx.xxx.xxx.xxx.  This routine takes the address
4754  * in the form of a ipaddr_t and calls ip_dot_saddr with a pointer.
4755  *
4756  * Once the ndd table-printing interfaces are removed, this can be changed to
4757  * standard dotted-decimal form.
4758  */
4759 char *
4760 ip_dot_addr(ipaddr_t addr, char *buf)
4761 {
4762         uint8_t *ap = (uint8_t *)&addr;
4763 
4764         (void) mi_sprintf(buf, "%03d.%03d.%03d.%03d",
4765             ap[0] & 0xFF, ap[1] & 0xFF, ap[2] & 0xFF, ap[3] & 0xFF);
4766         return (buf);
4767 }
4768 
4769 /*
4770  * Write the given MAC address as a printable string in the usual colon-
4771  * separated format.
4772  */
4773 const char *
4774 mac_colon_addr(const uint8_t *addr, size_t alen, char *buf, size_t buflen)
4775 {
4776         char *bp;
4777 
4778         if (alen == 0 || buflen < 4)
4779                 return ("?");
4780         bp = buf;
4781         for (;;) {
4782                 /*
4783                  * If there are more MAC address bytes available, but we won't
4784                  * have any room to print them, then add "..." to the string
4785                  * instead.  See below for the 'magic number' explanation.
4786                  */
4787                 if ((alen == 2 && buflen < 6) || (alen > 2 && buflen < 7)) {
4788                         (void) strcpy(bp, "...");
4789                         break;
4790                 }
4791                 (void) sprintf(bp, "%02x", *addr++);
4792                 bp += 2;
4793                 if (--alen == 0)
4794                         break;
4795                 *bp++ = ':';
4796                 buflen -= 3;
4797                 /*
4798                  * At this point, based on the first 'if' statement above,
4799                  * either alen == 1 and buflen >= 3, or alen > 1 and
4800                  * buflen >= 4.  The first case leaves room for the final "xx"
4801                  * number and trailing NUL byte.  The second leaves room for at
4802                  * least "...".  Thus the apparently 'magic' numbers chosen for
4803                  * that statement.
4804                  */
4805         }
4806         return (buf);
4807 }
4808 
4809 /*
4810  * Called when it is conceptually a ULP that would sent the packet
4811  * e.g., port unreachable and protocol unreachable. Check that the packet
4812  * would have passed the IPsec global policy before sending the error.
4813  *
4814  * Send an ICMP error after patching up the packet appropriately.
4815  * Uses ip_drop_input and bumps the appropriate MIB.
4816  */
4817 void
4818 ip_fanout_send_icmp_v4(mblk_t *mp, uint_t icmp_type, uint_t icmp_code,
4819     ip_recv_attr_t *ira)
4820 {
4821         ipha_t          *ipha;
4822         boolean_t       secure;
4823         ill_t           *ill = ira->ira_ill;
4824         ip_stack_t      *ipst = ill->ill_ipst;
4825         netstack_t      *ns = ipst->ips_netstack;
4826         ipsec_stack_t   *ipss = ns->netstack_ipsec;
4827 
4828         secure = ira->ira_flags & IRAF_IPSEC_SECURE;
4829 
4830         /*
4831          * We are generating an icmp error for some inbound packet.
4832          * Called from all ip_fanout_(udp, tcp, proto) functions.
4833          * Before we generate an error, check with global policy
4834          * to see whether this is allowed to enter the system. As
4835          * there is no "conn", we are checking with global policy.
4836          */
4837         ipha = (ipha_t *)mp->b_rptr;
4838         if (secure || ipss->ipsec_inbound_v4_policy_present) {
4839                 mp = ipsec_check_global_policy(mp, NULL, ipha, NULL, ira, ns);
4840                 if (mp == NULL)
4841                         return;
4842         }
4843 
4844         /* We never send errors for protocols that we do implement */
4845         if (ira->ira_protocol == IPPROTO_ICMP ||
4846             ira->ira_protocol == IPPROTO_IGMP) {
4847                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
4848                 ip_drop_input("ip_fanout_send_icmp_v4", mp, ill);
4849                 freemsg(mp);
4850                 return;
4851         }
4852         /*
4853          * Have to correct checksum since
4854          * the packet might have been
4855          * fragmented and the reassembly code in ip_rput
4856          * does not restore the IP checksum.
4857          */
4858         ipha->ipha_hdr_checksum = 0;
4859         ipha->ipha_hdr_checksum = ip_csum_hdr(ipha);
4860 
4861         switch (icmp_type) {
4862         case ICMP_DEST_UNREACHABLE:
4863                 switch (icmp_code) {
4864                 case ICMP_PROTOCOL_UNREACHABLE:
4865                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInUnknownProtos);
4866                         ip_drop_input("ipIfStatsInUnknownProtos", mp, ill);
4867                         break;
4868                 case ICMP_PORT_UNREACHABLE:
4869                         BUMP_MIB(ill->ill_ip_mib, udpIfStatsNoPorts);
4870                         ip_drop_input("ipIfStatsNoPorts", mp, ill);
4871                         break;
4872                 }
4873 
4874                 icmp_unreachable(mp, icmp_code, ira);
4875                 break;
4876         default:
4877 #ifdef DEBUG
4878                 panic("ip_fanout_send_icmp_v4: wrong type");
4879                 /*NOTREACHED*/
4880 #else
4881                 freemsg(mp);
4882                 break;
4883 #endif
4884         }
4885 }
4886 
4887 /*
4888  * Used to send an ICMP error message when a packet is received for
4889  * a protocol that is not supported. The mblk passed as argument
4890  * is consumed by this function.
4891  */
4892 void
4893 ip_proto_not_sup(mblk_t *mp, ip_recv_attr_t *ira)
4894 {
4895         ipha_t          *ipha;
4896 
4897         ipha = (ipha_t *)mp->b_rptr;
4898         if (ira->ira_flags & IRAF_IS_IPV4) {
4899                 ASSERT(IPH_HDR_VERSION(ipha) == IP_VERSION);
4900                 ip_fanout_send_icmp_v4(mp, ICMP_DEST_UNREACHABLE,
4901                     ICMP_PROTOCOL_UNREACHABLE, ira);
4902         } else {
4903                 ASSERT(IPH_HDR_VERSION(ipha) == IPV6_VERSION);
4904                 ip_fanout_send_icmp_v6(mp, ICMP6_PARAM_PROB,
4905                     ICMP6_PARAMPROB_NEXTHEADER, ira);
4906         }
4907 }
4908 
4909 /*
4910  * Deliver a rawip packet to the given conn, possibly applying ipsec policy.
4911  * Handles IPv4 and IPv6.
4912  * We are responsible for disposing of mp, such as by freemsg() or putnext()
4913  * Caller is responsible for dropping references to the conn.
4914  */
4915 void
4916 ip_fanout_proto_conn(conn_t *connp, mblk_t *mp, ipha_t *ipha, ip6_t *ip6h,
4917     ip_recv_attr_t *ira)
4918 {
4919         ill_t           *ill = ira->ira_ill;
4920         ip_stack_t      *ipst = ill->ill_ipst;
4921         ipsec_stack_t   *ipss = ipst->ips_netstack->netstack_ipsec;
4922         boolean_t       secure;
4923         uint_t          protocol = ira->ira_protocol;
4924         iaflags_t       iraflags = ira->ira_flags;
4925         queue_t         *rq;
4926 
4927         secure = iraflags & IRAF_IPSEC_SECURE;
4928 
4929         rq = connp->conn_rq;
4930         if (IPCL_IS_NONSTR(connp) ? connp->conn_flow_cntrld : !canputnext(rq)) {
4931                 switch (protocol) {
4932                 case IPPROTO_ICMPV6:
4933                         BUMP_MIB(ill->ill_icmp6_mib, ipv6IfIcmpInOverflows);
4934                         break;
4935                 case IPPROTO_ICMP:
4936                         BUMP_MIB(&ipst->ips_icmp_mib, icmpInOverflows);
4937                         break;
4938                 default:
4939                         BUMP_MIB(ill->ill_ip_mib, rawipIfStatsInOverflows);
4940                         break;
4941                 }
4942                 freemsg(mp);
4943                 return;
4944         }
4945 
4946         ASSERT(!(IPCL_IS_IPTUN(connp)));
4947 
4948         if (((iraflags & IRAF_IS_IPV4) ?
4949             CONN_INBOUND_POLICY_PRESENT(connp, ipss) :
4950             CONN_INBOUND_POLICY_PRESENT_V6(connp, ipss)) ||
4951             secure) {
4952                 mp = ipsec_check_inbound_policy(mp, connp, ipha,
4953                     ip6h, ira);
4954                 if (mp == NULL) {
4955                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
4956                         /* Note that mp is NULL */
4957                         ip_drop_input("ipIfStatsInDiscards", mp, ill);
4958                         return;
4959                 }
4960         }
4961 
4962         if (iraflags & IRAF_ICMP_ERROR) {
4963                 (connp->conn_recvicmp)(connp, mp, NULL, ira);
4964         } else {
4965                 ill_t *rill = ira->ira_rill;
4966 
4967                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCInDelivers);
4968                 ira->ira_ill = ira->ira_rill = NULL;
4969                 /* Send it upstream */
4970                 (connp->conn_recv)(connp, mp, NULL, ira);
4971                 ira->ira_ill = ill;
4972                 ira->ira_rill = rill;
4973         }
4974 }
4975 
4976 /*
4977  * Handle protocols with which IP is less intimate.  There
4978  * can be more than one stream bound to a particular
4979  * protocol.  When this is the case, normally each one gets a copy
4980  * of any incoming packets.
4981  *
4982  * IPsec NOTE :
4983  *
4984  * Don't allow a secure packet going up a non-secure connection.
4985  * We don't allow this because
4986  *
4987  * 1) Reply might go out in clear which will be dropped at
4988  *    the sending side.
4989  * 2) If the reply goes out in clear it will give the
4990  *    adversary enough information for getting the key in
4991  *    most of the cases.
4992  *
4993  * Moreover getting a secure packet when we expect clear
4994  * implies that SA's were added without checking for
4995  * policy on both ends. This should not happen once ISAKMP
4996  * is used to negotiate SAs as SAs will be added only after
4997  * verifying the policy.
4998  *
4999  * Zones notes:
5000  * Earlier in ip_input on a system with multiple shared-IP zones we
5001  * duplicate the multicast and broadcast packets and send them up
5002  * with each explicit zoneid that exists on that ill.
5003  * This means that here we can match the zoneid with SO_ALLZONES being special.
5004  */
5005 void
5006 ip_fanout_proto_v4(mblk_t *mp, ipha_t *ipha, ip_recv_attr_t *ira)
5007 {
5008         mblk_t          *mp1;
5009         ipaddr_t        laddr;
5010         conn_t          *connp, *first_connp, *next_connp;
5011         connf_t         *connfp;
5012         ill_t           *ill = ira->ira_ill;
5013         ip_stack_t      *ipst = ill->ill_ipst;
5014 
5015         laddr = ipha->ipha_dst;
5016 
5017         connfp = &ipst->ips_ipcl_proto_fanout_v4[ira->ira_protocol];
5018         mutex_enter(&connfp->connf_lock);
5019         connp = connfp->connf_head;
5020         for (connp = connfp->connf_head; connp != NULL;
5021             connp = connp->conn_next) {
5022                 /* Note: IPCL_PROTO_MATCH includes conn_wantpacket */
5023                 if (IPCL_PROTO_MATCH(connp, ira, ipha) &&
5024                     (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5025                     tsol_receive_local(mp, &laddr, IPV4_VERSION, ira, connp))) {
5026                         break;
5027                 }
5028         }
5029 
5030         if (connp == NULL) {
5031                 /*
5032                  * No one bound to these addresses.  Is
5033                  * there a client that wants all
5034                  * unclaimed datagrams?
5035                  */
5036                 mutex_exit(&connfp->connf_lock);
5037                 ip_fanout_send_icmp_v4(mp, ICMP_DEST_UNREACHABLE,
5038                     ICMP_PROTOCOL_UNREACHABLE, ira);
5039                 return;
5040         }
5041 
5042         ASSERT(IPCL_IS_NONSTR(connp) || connp->conn_rq != NULL);
5043 
5044         CONN_INC_REF(connp);
5045         first_connp = connp;
5046         connp = connp->conn_next;
5047 
5048         for (;;) {
5049                 while (connp != NULL) {
5050                         /* Note: IPCL_PROTO_MATCH includes conn_wantpacket */
5051                         if (IPCL_PROTO_MATCH(connp, ira, ipha) &&
5052                             (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5053                             tsol_receive_local(mp, &laddr, IPV4_VERSION,
5054                             ira, connp)))
5055                                 break;
5056                         connp = connp->conn_next;
5057                 }
5058 
5059                 if (connp == NULL) {
5060                         /* No more interested clients */
5061                         connp = first_connp;
5062                         break;
5063                 }
5064                 if (((mp1 = dupmsg(mp)) == NULL) &&
5065                     ((mp1 = copymsg(mp)) == NULL)) {
5066                         /* Memory allocation failed */
5067                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
5068                         ip_drop_input("ipIfStatsInDiscards", mp, ill);
5069                         connp = first_connp;
5070                         break;
5071                 }
5072 
5073                 CONN_INC_REF(connp);
5074                 mutex_exit(&connfp->connf_lock);
5075 
5076                 ip_fanout_proto_conn(connp, mp1, (ipha_t *)mp1->b_rptr, NULL,
5077                     ira);
5078 
5079                 mutex_enter(&connfp->connf_lock);
5080                 /* Follow the next pointer before releasing the conn. */
5081                 next_connp = connp->conn_next;
5082                 CONN_DEC_REF(connp);
5083                 connp = next_connp;
5084         }
5085 
5086         /* Last one.  Send it upstream. */
5087         mutex_exit(&connfp->connf_lock);
5088 
5089         ip_fanout_proto_conn(connp, mp, ipha, NULL, ira);
5090 
5091         CONN_DEC_REF(connp);
5092 }
5093 
5094 /*
5095  * If we have a IPsec NAT-Traversal packet, strip the zero-SPI or
5096  * pass it along to ESP if the SPI is non-zero.  Returns the mblk if the mblk
5097  * is not consumed.
5098  *
5099  * One of three things can happen, all of which affect the passed-in mblk:
5100  *
5101  * 1.) The packet is stock UDP and gets its zero-SPI stripped.  Return mblk..
5102  *
5103  * 2.) The packet is ESP-in-UDP, gets transformed into an equivalent
5104  *     ESP packet, and is passed along to ESP for consumption.  Return NULL.
5105  *
5106  * 3.) The packet is an ESP-in-UDP Keepalive.  Drop it and return NULL.
5107  */
5108 mblk_t *
5109 zero_spi_check(mblk_t *mp, ip_recv_attr_t *ira)
5110 {
5111         int shift, plen, iph_len;
5112         ipha_t *ipha;
5113         udpha_t *udpha;
5114         uint32_t *spi;
5115         uint32_t esp_ports;
5116         uint8_t *orptr;
5117         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
5118         ipsec_stack_t   *ipss = ipst->ips_netstack->netstack_ipsec;
5119 
5120         ipha = (ipha_t *)mp->b_rptr;
5121         iph_len = ira->ira_ip_hdr_length;
5122         plen = ira->ira_pktlen;
5123 
5124         if (plen - iph_len - sizeof (udpha_t) < sizeof (uint32_t)) {
5125                 /*
5126                  * Most likely a keepalive for the benefit of an intervening
5127                  * NAT.  These aren't for us, per se, so drop it.
5128                  *
5129                  * RFC 3947/8 doesn't say for sure what to do for 2-3
5130                  * byte packets (keepalives are 1-byte), but we'll drop them
5131                  * also.
5132                  */
5133                 ip_drop_packet(mp, B_TRUE, ira->ira_ill,
5134                     DROPPER(ipss, ipds_esp_nat_t_ka), &ipss->ipsec_dropper);
5135                 return (NULL);
5136         }
5137 
5138         if (MBLKL(mp) < iph_len + sizeof (udpha_t) + sizeof (*spi)) {
5139                 /* might as well pull it all up - it might be ESP. */
5140                 if (!pullupmsg(mp, -1)) {
5141                         ip_drop_packet(mp, B_TRUE, ira->ira_ill,
5142                             DROPPER(ipss, ipds_esp_nomem),
5143                             &ipss->ipsec_dropper);
5144                         return (NULL);
5145                 }
5146 
5147                 ipha = (ipha_t *)mp->b_rptr;
5148         }
5149         spi = (uint32_t *)(mp->b_rptr + iph_len + sizeof (udpha_t));
5150         if (*spi == 0) {
5151                 /* UDP packet - remove 0-spi. */
5152                 shift = sizeof (uint32_t);
5153         } else {
5154                 /* ESP-in-UDP packet - reduce to ESP. */
5155                 ipha->ipha_protocol = IPPROTO_ESP;
5156                 shift = sizeof (udpha_t);
5157         }
5158 
5159         /* Fix IP header */
5160         ira->ira_pktlen = (plen - shift);
5161         ipha->ipha_length = htons(ira->ira_pktlen);
5162         ipha->ipha_hdr_checksum = 0;
5163 
5164         orptr = mp->b_rptr;
5165         mp->b_rptr += shift;
5166 
5167         udpha = (udpha_t *)(orptr + iph_len);
5168         if (*spi == 0) {
5169                 ASSERT((uint8_t *)ipha == orptr);
5170                 udpha->uha_length = htons(plen - shift - iph_len);
5171                 iph_len += sizeof (udpha_t);    /* For the call to ovbcopy(). */
5172                 esp_ports = 0;
5173         } else {
5174                 esp_ports = *((uint32_t *)udpha);
5175                 ASSERT(esp_ports != 0);
5176         }
5177         ovbcopy(orptr, orptr + shift, iph_len);
5178         if (esp_ports != 0) /* Punt up for ESP processing. */ {
5179                 ipha = (ipha_t *)(orptr + shift);
5180 
5181                 ira->ira_flags |= IRAF_ESP_UDP_PORTS;
5182                 ira->ira_esp_udp_ports = esp_ports;
5183                 ip_fanout_v4(mp, ipha, ira);
5184                 return (NULL);
5185         }
5186         return (mp);
5187 }
5188 
5189 /*
5190  * Deliver a udp packet to the given conn, possibly applying ipsec policy.
5191  * Handles IPv4 and IPv6.
5192  * We are responsible for disposing of mp, such as by freemsg() or putnext()
5193  * Caller is responsible for dropping references to the conn.
5194  */
5195 void
5196 ip_fanout_udp_conn(conn_t *connp, mblk_t *mp, ipha_t *ipha, ip6_t *ip6h,
5197     ip_recv_attr_t *ira)
5198 {
5199         ill_t           *ill = ira->ira_ill;
5200         ip_stack_t      *ipst = ill->ill_ipst;
5201         ipsec_stack_t   *ipss = ipst->ips_netstack->netstack_ipsec;
5202         boolean_t       secure;
5203         iaflags_t       iraflags = ira->ira_flags;
5204 
5205         secure = iraflags & IRAF_IPSEC_SECURE;
5206 
5207         if (IPCL_IS_NONSTR(connp) ? connp->conn_flow_cntrld :
5208             !canputnext(connp->conn_rq)) {
5209                 BUMP_MIB(ill->ill_ip_mib, udpIfStatsInOverflows);
5210                 freemsg(mp);
5211                 return;
5212         }
5213 
5214         if (((iraflags & IRAF_IS_IPV4) ?
5215             CONN_INBOUND_POLICY_PRESENT(connp, ipss) :
5216             CONN_INBOUND_POLICY_PRESENT_V6(connp, ipss)) ||
5217             secure) {
5218                 mp = ipsec_check_inbound_policy(mp, connp, ipha,
5219                     ip6h, ira);
5220                 if (mp == NULL) {
5221                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
5222                         /* Note that mp is NULL */
5223                         ip_drop_input("ipIfStatsInDiscards", mp, ill);
5224                         return;
5225                 }
5226         }
5227 
5228         /*
5229          * Since this code is not used for UDP unicast we don't need a NAT_T
5230          * check. Only ip_fanout_v4 has that check.
5231          */
5232         if (ira->ira_flags & IRAF_ICMP_ERROR) {
5233                 (connp->conn_recvicmp)(connp, mp, NULL, ira);
5234         } else {
5235                 ill_t *rill = ira->ira_rill;
5236 
5237                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCInDelivers);
5238                 ira->ira_ill = ira->ira_rill = NULL;
5239                 /* Send it upstream */
5240                 (connp->conn_recv)(connp, mp, NULL, ira);
5241                 ira->ira_ill = ill;
5242                 ira->ira_rill = rill;
5243         }
5244 }
5245 
5246 /*
5247  * Fanout for UDP packets that are multicast or broadcast, and ICMP errors.
5248  * (Unicast fanout is handled in ip_input_v4.)
5249  *
5250  * If SO_REUSEADDR is set all multicast and broadcast packets
5251  * will be delivered to all conns bound to the same port.
5252  *
5253  * If there is at least one matching AF_INET receiver, then we will
5254  * ignore any AF_INET6 receivers.
5255  * In the special case where an AF_INET socket binds to 0.0.0.0/<port> and an
5256  * AF_INET6 socket binds to ::/<port>, only the AF_INET socket receives the IPv4
5257  * packets.
5258  *
5259  * Zones notes:
5260  * Earlier in ip_input on a system with multiple shared-IP zones we
5261  * duplicate the multicast and broadcast packets and send them up
5262  * with each explicit zoneid that exists on that ill.
5263  * This means that here we can match the zoneid with SO_ALLZONES being special.
5264  */
5265 void
5266 ip_fanout_udp_multi_v4(mblk_t *mp, ipha_t *ipha, uint16_t lport, uint16_t fport,
5267     ip_recv_attr_t *ira)
5268 {
5269         ipaddr_t        laddr;
5270         in6_addr_t      v6faddr;
5271         conn_t          *connp;
5272         connf_t         *connfp;
5273         ipaddr_t        faddr;
5274         ill_t           *ill = ira->ira_ill;
5275         ip_stack_t      *ipst = ill->ill_ipst;
5276 
5277         ASSERT(ira->ira_flags & (IRAF_MULTIBROADCAST|IRAF_ICMP_ERROR));
5278 
5279         laddr = ipha->ipha_dst;
5280         faddr = ipha->ipha_src;
5281 
5282         connfp = &ipst->ips_ipcl_udp_fanout[IPCL_UDP_HASH(lport, ipst)];
5283         mutex_enter(&connfp->connf_lock);
5284         connp = connfp->connf_head;
5285 
5286         /*
5287          * If SO_REUSEADDR has been set on the first we send the
5288          * packet to all clients that have joined the group and
5289          * match the port.
5290          */
5291         while (connp != NULL) {
5292                 if ((IPCL_UDP_MATCH(connp, lport, laddr, fport, faddr)) &&
5293                     conn_wantpacket(connp, ira, ipha) &&
5294                     (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5295                     tsol_receive_local(mp, &laddr, IPV4_VERSION, ira, connp)))
5296                         break;
5297                 connp = connp->conn_next;
5298         }
5299 
5300         if (connp == NULL)
5301                 goto notfound;
5302 
5303         CONN_INC_REF(connp);
5304 
5305         if (connp->conn_reuseaddr) {
5306                 conn_t          *first_connp = connp;
5307                 conn_t          *next_connp;
5308                 mblk_t          *mp1;
5309 
5310                 connp = connp->conn_next;
5311                 for (;;) {
5312                         while (connp != NULL) {
5313                                 if (IPCL_UDP_MATCH(connp, lport, laddr,
5314                                     fport, faddr) &&
5315                                     conn_wantpacket(connp, ira, ipha) &&
5316                                     (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5317                                     tsol_receive_local(mp, &laddr, IPV4_VERSION,
5318                                     ira, connp)))
5319                                         break;
5320                                 connp = connp->conn_next;
5321                         }
5322                         if (connp == NULL) {
5323                                 /* No more interested clients */
5324                                 connp = first_connp;
5325                                 break;
5326                         }
5327                         if (((mp1 = dupmsg(mp)) == NULL) &&
5328                             ((mp1 = copymsg(mp)) == NULL)) {
5329                                 /* Memory allocation failed */
5330                                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
5331                                 ip_drop_input("ipIfStatsInDiscards", mp, ill);
5332                                 connp = first_connp;
5333                                 break;
5334                         }
5335                         CONN_INC_REF(connp);
5336                         mutex_exit(&connfp->connf_lock);
5337 
5338                         IP_STAT(ipst, ip_udp_fanmb);
5339                         ip_fanout_udp_conn(connp, mp1, (ipha_t *)mp1->b_rptr,
5340                             NULL, ira);
5341                         mutex_enter(&connfp->connf_lock);
5342                         /* Follow the next pointer before releasing the conn */
5343                         next_connp = connp->conn_next;
5344                         CONN_DEC_REF(connp);
5345                         connp = next_connp;
5346                 }
5347         }
5348 
5349         /* Last one.  Send it upstream. */
5350         mutex_exit(&connfp->connf_lock);
5351         IP_STAT(ipst, ip_udp_fanmb);
5352         ip_fanout_udp_conn(connp, mp, ipha, NULL, ira);
5353         CONN_DEC_REF(connp);
5354         return;
5355 
5356 notfound:
5357         mutex_exit(&connfp->connf_lock);
5358         /*
5359          * IPv6 endpoints bound to multicast IPv4-mapped addresses
5360          * have already been matched above, since they live in the IPv4
5361          * fanout tables. This implies we only need to
5362          * check for IPv6 in6addr_any endpoints here.
5363          * Thus we compare using ipv6_all_zeros instead of the destination
5364          * address, except for the multicast group membership lookup which
5365          * uses the IPv4 destination.
5366          */
5367         IN6_IPADDR_TO_V4MAPPED(ipha->ipha_src, &v6faddr);
5368         connfp = &ipst->ips_ipcl_udp_fanout[IPCL_UDP_HASH(lport, ipst)];
5369         mutex_enter(&connfp->connf_lock);
5370         connp = connfp->connf_head;
5371         /*
5372          * IPv4 multicast packet being delivered to an AF_INET6
5373          * in6addr_any endpoint.
5374          * Need to check conn_wantpacket(). Note that we use conn_wantpacket()
5375          * and not conn_wantpacket_v6() since any multicast membership is
5376          * for an IPv4-mapped multicast address.
5377          */
5378         while (connp != NULL) {
5379                 if (IPCL_UDP_MATCH_V6(connp, lport, ipv6_all_zeros,
5380                     fport, v6faddr) &&
5381                     conn_wantpacket(connp, ira, ipha) &&
5382                     (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5383                     tsol_receive_local(mp, &laddr, IPV4_VERSION, ira, connp)))
5384                         break;
5385                 connp = connp->conn_next;
5386         }
5387 
5388         if (connp == NULL) {
5389                 /*
5390                  * No one bound to this port.  Is
5391                  * there a client that wants all
5392                  * unclaimed datagrams?
5393                  */
5394                 mutex_exit(&connfp->connf_lock);
5395 
5396                 if (ipst->ips_ipcl_proto_fanout_v4[IPPROTO_UDP].connf_head !=
5397                     NULL) {
5398                         ASSERT(ira->ira_protocol == IPPROTO_UDP);
5399                         ip_fanout_proto_v4(mp, ipha, ira);
5400                 } else {
5401                         /*
5402                          * We used to attempt to send an icmp error here, but
5403                          * since this is known to be a multicast packet
5404                          * and we don't send icmp errors in response to
5405                          * multicast, just drop the packet and give up sooner.
5406                          */
5407                         BUMP_MIB(ill->ill_ip_mib, udpIfStatsNoPorts);
5408                         freemsg(mp);
5409                 }
5410                 return;
5411         }
5412         CONN_INC_REF(connp);
5413         ASSERT(IPCL_IS_NONSTR(connp) || connp->conn_rq != NULL);
5414 
5415         /*
5416          * If SO_REUSEADDR has been set on the first we send the
5417          * packet to all clients that have joined the group and
5418          * match the port.
5419          */
5420         if (connp->conn_reuseaddr) {
5421                 conn_t          *first_connp = connp;
5422                 conn_t          *next_connp;
5423                 mblk_t          *mp1;
5424 
5425                 connp = connp->conn_next;
5426                 for (;;) {
5427                         while (connp != NULL) {
5428                                 if (IPCL_UDP_MATCH_V6(connp, lport,
5429                                     ipv6_all_zeros, fport, v6faddr) &&
5430                                     conn_wantpacket(connp, ira, ipha) &&
5431                                     (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5432                                     tsol_receive_local(mp, &laddr, IPV4_VERSION,
5433                                     ira, connp)))
5434                                         break;
5435                                 connp = connp->conn_next;
5436                         }
5437                         if (connp == NULL) {
5438                                 /* No more interested clients */
5439                                 connp = first_connp;
5440                                 break;
5441                         }
5442                         if (((mp1 = dupmsg(mp)) == NULL) &&
5443                             ((mp1 = copymsg(mp)) == NULL)) {
5444                                 /* Memory allocation failed */
5445                                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
5446                                 ip_drop_input("ipIfStatsInDiscards", mp, ill);
5447                                 connp = first_connp;
5448                                 break;
5449                         }
5450                         CONN_INC_REF(connp);
5451                         mutex_exit(&connfp->connf_lock);
5452 
5453                         IP_STAT(ipst, ip_udp_fanmb);
5454                         ip_fanout_udp_conn(connp, mp1, (ipha_t *)mp1->b_rptr,
5455                             NULL, ira);
5456                         mutex_enter(&connfp->connf_lock);
5457                         /* Follow the next pointer before releasing the conn */
5458                         next_connp = connp->conn_next;
5459                         CONN_DEC_REF(connp);
5460                         connp = next_connp;
5461                 }
5462         }
5463 
5464         /* Last one.  Send it upstream. */
5465         mutex_exit(&connfp->connf_lock);
5466         IP_STAT(ipst, ip_udp_fanmb);
5467         ip_fanout_udp_conn(connp, mp, ipha, NULL, ira);
5468         CONN_DEC_REF(connp);
5469 }
5470 
5471 /*
5472  * Split an incoming packet's IPv4 options into the label and the other options.
5473  * If 'allocate' is set it does memory allocation for the ip_pkt_t, including
5474  * clearing out any leftover label or options.
5475  * Otherwise it just makes ipp point into the packet.
5476  *
5477  * Returns zero if ok; ENOMEM if the buffer couldn't be allocated.
5478  */
5479 int
5480 ip_find_hdr_v4(ipha_t *ipha, ip_pkt_t *ipp, boolean_t allocate)
5481 {
5482         uchar_t         *opt;
5483         uint32_t        totallen;
5484         uint32_t        optval;
5485         uint32_t        optlen;
5486 
5487         ipp->ipp_fields |= IPPF_HOPLIMIT | IPPF_TCLASS | IPPF_ADDR;
5488         ipp->ipp_hoplimit = ipha->ipha_ttl;
5489         ipp->ipp_type_of_service = ipha->ipha_type_of_service;
5490         IN6_IPADDR_TO_V4MAPPED(ipha->ipha_dst, &ipp->ipp_addr);
5491 
5492         /*
5493          * Get length (in 4 byte octets) of IP header options.
5494          */
5495         totallen = ipha->ipha_version_and_hdr_length -
5496             (uint8_t)((IP_VERSION << 4) + IP_SIMPLE_HDR_LENGTH_IN_WORDS);
5497 
5498         if (totallen == 0) {
5499                 if (!allocate)
5500                         return (0);
5501 
5502                 /* Clear out anything from a previous packet */
5503                 if (ipp->ipp_fields & IPPF_IPV4_OPTIONS) {
5504                         kmem_free(ipp->ipp_ipv4_options,
5505                             ipp->ipp_ipv4_options_len);
5506                         ipp->ipp_ipv4_options = NULL;
5507                         ipp->ipp_ipv4_options_len = 0;
5508                         ipp->ipp_fields &= ~IPPF_IPV4_OPTIONS;
5509                 }
5510                 if (ipp->ipp_fields & IPPF_LABEL_V4) {
5511                         kmem_free(ipp->ipp_label_v4, ipp->ipp_label_len_v4);
5512                         ipp->ipp_label_v4 = NULL;
5513                         ipp->ipp_label_len_v4 = 0;
5514                         ipp->ipp_fields &= ~IPPF_LABEL_V4;
5515                 }
5516                 return (0);
5517         }
5518 
5519         totallen <<= 2;
5520         opt = (uchar_t *)&ipha[1];
5521         if (!is_system_labeled()) {
5522 
5523         copyall:
5524                 if (!allocate) {
5525                         if (totallen != 0) {
5526                                 ipp->ipp_ipv4_options = opt;
5527                                 ipp->ipp_ipv4_options_len = totallen;
5528                                 ipp->ipp_fields |= IPPF_IPV4_OPTIONS;
5529                         }
5530                         return (0);
5531                 }
5532                 /* Just copy all of options */
5533                 if (ipp->ipp_fields & IPPF_IPV4_OPTIONS) {
5534                         if (totallen == ipp->ipp_ipv4_options_len) {
5535                                 bcopy(opt, ipp->ipp_ipv4_options, totallen);
5536                                 return (0);
5537                         }
5538                         kmem_free(ipp->ipp_ipv4_options,
5539                             ipp->ipp_ipv4_options_len);
5540                         ipp->ipp_ipv4_options = NULL;
5541                         ipp->ipp_ipv4_options_len = 0;
5542                         ipp->ipp_fields &= ~IPPF_IPV4_OPTIONS;
5543                 }
5544                 if (totallen == 0)
5545                         return (0);
5546 
5547                 ipp->ipp_ipv4_options = kmem_alloc(totallen, KM_NOSLEEP);
5548                 if (ipp->ipp_ipv4_options == NULL)
5549                         return (ENOMEM);
5550                 ipp->ipp_ipv4_options_len = totallen;
5551                 ipp->ipp_fields |= IPPF_IPV4_OPTIONS;
5552                 bcopy(opt, ipp->ipp_ipv4_options, totallen);
5553                 return (0);
5554         }
5555 
5556         if (allocate && (ipp->ipp_fields & IPPF_LABEL_V4)) {
5557                 kmem_free(ipp->ipp_label_v4, ipp->ipp_label_len_v4);
5558                 ipp->ipp_label_v4 = NULL;
5559                 ipp->ipp_label_len_v4 = 0;
5560                 ipp->ipp_fields &= ~IPPF_LABEL_V4;
5561         }
5562 
5563         /*
5564          * Search for CIPSO option.
5565          * We assume CIPSO is first in options if it is present.
5566          * If it isn't, then ipp_opt_ipv4_options will not include the options
5567          * prior to the CIPSO option.
5568          */
5569         while (totallen != 0) {
5570                 switch (optval = opt[IPOPT_OPTVAL]) {
5571                 case IPOPT_EOL:
5572                         return (0);
5573                 case IPOPT_NOP:
5574                         optlen = 1;
5575                         break;
5576                 default:
5577                         if (totallen <= IPOPT_OLEN)
5578                                 return (EINVAL);
5579                         optlen = opt[IPOPT_OLEN];
5580                         if (optlen < 2)
5581                                 return (EINVAL);
5582                 }
5583                 if (optlen > totallen)
5584                         return (EINVAL);
5585 
5586                 switch (optval) {
5587                 case IPOPT_COMSEC:
5588                         if (!allocate) {
5589                                 ipp->ipp_label_v4 = opt;
5590                                 ipp->ipp_label_len_v4 = optlen;
5591                                 ipp->ipp_fields |= IPPF_LABEL_V4;
5592                         } else {
5593                                 ipp->ipp_label_v4 = kmem_alloc(optlen,
5594                                     KM_NOSLEEP);
5595                                 if (ipp->ipp_label_v4 == NULL)
5596                                         return (ENOMEM);
5597                                 ipp->ipp_label_len_v4 = optlen;
5598                                 ipp->ipp_fields |= IPPF_LABEL_V4;
5599                                 bcopy(opt, ipp->ipp_label_v4, optlen);
5600                         }
5601                         totallen -= optlen;
5602                         opt += optlen;
5603 
5604                         /* Skip padding bytes until we get to a multiple of 4 */
5605                         while ((totallen & 3) != 0 && opt[0] == IPOPT_NOP) {
5606                                 totallen--;
5607                                 opt++;
5608                         }
5609                         /* Remaining as ipp_ipv4_options */
5610                         goto copyall;
5611                 }
5612                 totallen -= optlen;
5613                 opt += optlen;
5614         }
5615         /* No CIPSO found; return everything as ipp_ipv4_options */
5616         totallen = ipha->ipha_version_and_hdr_length -
5617             (uint8_t)((IP_VERSION << 4) + IP_SIMPLE_HDR_LENGTH_IN_WORDS);
5618         totallen <<= 2;
5619         opt = (uchar_t *)&ipha[1];
5620         goto copyall;
5621 }
5622 
5623 /*
5624  * Efficient versions of lookup for an IRE when we only
5625  * match the address.
5626  * For RTF_REJECT or BLACKHOLE we return IRE_NOROUTE.
5627  * Does not handle multicast addresses.
5628  */
5629 uint_t
5630 ip_type_v4(ipaddr_t addr, ip_stack_t *ipst)
5631 {
5632         ire_t *ire;
5633         uint_t result;
5634 
5635         ire = ire_ftable_lookup_simple_v4(addr, 0, ipst, NULL);
5636         ASSERT(ire != NULL);
5637         if (ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE))
5638                 result = IRE_NOROUTE;
5639         else
5640                 result = ire->ire_type;
5641         ire_refrele(ire);
5642         return (result);
5643 }
5644 
5645 /*
5646  * Efficient versions of lookup for an IRE when we only
5647  * match the address.
5648  * For RTF_REJECT or BLACKHOLE we return IRE_NOROUTE.
5649  * Does not handle multicast addresses.
5650  */
5651 uint_t
5652 ip_type_v6(const in6_addr_t *addr, ip_stack_t *ipst)
5653 {
5654         ire_t *ire;
5655         uint_t result;
5656 
5657         ire = ire_ftable_lookup_simple_v6(addr, 0, ipst, NULL);
5658         ASSERT(ire != NULL);
5659         if (ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE))
5660                 result = IRE_NOROUTE;
5661         else
5662                 result = ire->ire_type;
5663         ire_refrele(ire);
5664         return (result);
5665 }
5666 
5667 /*
5668  * Nobody should be sending
5669  * packets up this stream
5670  */
5671 static int
5672 ip_lrput(queue_t *q, mblk_t *mp)
5673 {
5674         switch (mp->b_datap->db_type) {
5675         case M_FLUSH:
5676                 /* Turn around */
5677                 if (*mp->b_rptr & FLUSHW) {
5678                         *mp->b_rptr &= ~FLUSHR;
5679                         qreply(q, mp);
5680                         return (0);
5681                 }
5682                 break;
5683         }
5684         freemsg(mp);
5685         return (0);
5686 }
5687 
5688 /* Nobody should be sending packets down this stream */
5689 /* ARGSUSED */
5690 int
5691 ip_lwput(queue_t *q, mblk_t *mp)
5692 {
5693         freemsg(mp);
5694         return (0);
5695 }
5696 
5697 /*
5698  * Move the first hop in any source route to ipha_dst and remove that part of
5699  * the source route.  Called by other protocols.  Errors in option formatting
5700  * are ignored - will be handled by ip_output_options. Return the final
5701  * destination (either ipha_dst or the last entry in a source route.)
5702  */
5703 ipaddr_t
5704 ip_massage_options(ipha_t *ipha, netstack_t *ns)
5705 {
5706         ipoptp_t        opts;
5707         uchar_t         *opt;
5708         uint8_t         optval;
5709         uint8_t         optlen;
5710         ipaddr_t        dst;
5711         int             i;
5712         ip_stack_t      *ipst = ns->netstack_ip;
5713 
5714         ip2dbg(("ip_massage_options\n"));
5715         dst = ipha->ipha_dst;
5716         for (optval = ipoptp_first(&opts, ipha);
5717             optval != IPOPT_EOL;
5718             optval = ipoptp_next(&opts)) {
5719                 opt = opts.ipoptp_cur;
5720                 switch (optval) {
5721                         uint8_t off;
5722                 case IPOPT_SSRR:
5723                 case IPOPT_LSRR:
5724                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
5725                                 ip1dbg(("ip_massage_options: bad src route\n"));
5726                                 break;
5727                         }
5728                         optlen = opts.ipoptp_len;
5729                         off = opt[IPOPT_OFFSET];
5730                         off--;
5731                 redo_srr:
5732                         if (optlen < IP_ADDR_LEN ||
5733                             off > optlen - IP_ADDR_LEN) {
5734                                 /* End of source route */
5735                                 ip1dbg(("ip_massage_options: end of SR\n"));
5736                                 break;
5737                         }
5738                         bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
5739                         ip1dbg(("ip_massage_options: next hop 0x%x\n",
5740                             ntohl(dst)));
5741                         /*
5742                          * Check if our address is present more than
5743                          * once as consecutive hops in source route.
5744                          * XXX verify per-interface ip_forwarding
5745                          * for source route?
5746                          */
5747                         if (ip_type_v4(dst, ipst) == IRE_LOCAL) {
5748                                 off += IP_ADDR_LEN;
5749                                 goto redo_srr;
5750                         }
5751                         if (dst == htonl(INADDR_LOOPBACK)) {
5752                                 ip1dbg(("ip_massage_options: loopback addr in "
5753                                     "source route!\n"));
5754                                 break;
5755                         }
5756                         /*
5757                          * Update ipha_dst to be the first hop and remove the
5758                          * first hop from the source route (by overwriting
5759                          * part of the option with NOP options).
5760                          */
5761                         ipha->ipha_dst = dst;
5762                         /* Put the last entry in dst */
5763                         off = ((optlen - IP_ADDR_LEN - 3) & ~(IP_ADDR_LEN-1)) +
5764                             3;
5765                         bcopy(&opt[off], &dst, IP_ADDR_LEN);
5766 
5767                         ip1dbg(("ip_massage_options: last hop 0x%x\n",
5768                             ntohl(dst)));
5769                         /* Move down and overwrite */
5770                         opt[IP_ADDR_LEN] = opt[0];
5771                         opt[IP_ADDR_LEN+1] = opt[IPOPT_OLEN] - IP_ADDR_LEN;
5772                         opt[IP_ADDR_LEN+2] = opt[IPOPT_OFFSET];
5773                         for (i = 0; i < IP_ADDR_LEN; i++)
5774                                 opt[i] = IPOPT_NOP;
5775                         break;
5776                 }
5777         }
5778         return (dst);
5779 }
5780 
5781 /*
5782  * Return the network mask
5783  * associated with the specified address.
5784  */
5785 ipaddr_t
5786 ip_net_mask(ipaddr_t addr)
5787 {
5788         uchar_t *up = (uchar_t *)&addr;
5789         ipaddr_t mask = 0;
5790         uchar_t *maskp = (uchar_t *)&mask;
5791 
5792 #if defined(__i386) || defined(__amd64)
5793 #define TOTALLY_BRAIN_DAMAGED_C_COMPILER
5794 #endif
5795 #ifdef  TOTALLY_BRAIN_DAMAGED_C_COMPILER
5796         maskp[0] = maskp[1] = maskp[2] = maskp[3] = 0;
5797 #endif
5798         if (CLASSD(addr)) {
5799                 maskp[0] = 0xF0;
5800                 return (mask);
5801         }
5802 
5803         /* We assume Class E default netmask to be 32 */
5804         if (CLASSE(addr))
5805                 return (0xffffffffU);
5806 
5807         if (addr == 0)
5808                 return (0);
5809         maskp[0] = 0xFF;
5810         if ((up[0] & 0x80) == 0)
5811                 return (mask);
5812 
5813         maskp[1] = 0xFF;
5814         if ((up[0] & 0xC0) == 0x80)
5815                 return (mask);
5816 
5817         maskp[2] = 0xFF;
5818         if ((up[0] & 0xE0) == 0xC0)
5819                 return (mask);
5820 
5821         /* Otherwise return no mask */
5822         return ((ipaddr_t)0);
5823 }
5824 
5825 /* Name/Value Table Lookup Routine */
5826 char *
5827 ip_nv_lookup(nv_t *nv, int value)
5828 {
5829         if (!nv)
5830                 return (NULL);
5831         for (; nv->nv_name; nv++) {
5832                 if (nv->nv_value == value)
5833                         return (nv->nv_name);
5834         }
5835         return ("unknown");
5836 }
5837 
5838 static int
5839 ip_wait_for_info_ack(ill_t *ill)
5840 {
5841         int err;
5842 
5843         mutex_enter(&ill->ill_lock);
5844         while (ill->ill_state_flags & ILL_LL_SUBNET_PENDING) {
5845                 /*
5846                  * Return value of 0 indicates a pending signal.
5847                  */
5848                 err = cv_wait_sig(&ill->ill_cv, &ill->ill_lock);
5849                 if (err == 0) {
5850                         mutex_exit(&ill->ill_lock);
5851                         return (EINTR);
5852                 }
5853         }
5854         mutex_exit(&ill->ill_lock);
5855         /*
5856          * ip_rput_other could have set an error  in ill_error on
5857          * receipt of M_ERROR.
5858          */
5859         return (ill->ill_error);
5860 }
5861 
5862 /*
5863  * This is a module open, i.e. this is a control stream for access
5864  * to a DLPI device.  We allocate an ill_t as the instance data in
5865  * this case.
5866  */
5867 static int
5868 ip_modopen(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
5869 {
5870         ill_t   *ill;
5871         int     err;
5872         zoneid_t zoneid;
5873         netstack_t *ns;
5874         ip_stack_t *ipst;
5875 
5876         /*
5877          * Prevent unprivileged processes from pushing IP so that
5878          * they can't send raw IP.
5879          */
5880         if (secpolicy_net_rawaccess(credp) != 0)
5881                 return (EPERM);
5882 
5883         ns = netstack_find_by_cred(credp);
5884         ASSERT(ns != NULL);
5885         ipst = ns->netstack_ip;
5886         ASSERT(ipst != NULL);
5887 
5888         /*
5889          * For exclusive stacks we set the zoneid to zero
5890          * to make IP operate as if in the global zone.
5891          */
5892         if (ipst->ips_netstack->netstack_stackid != GLOBAL_NETSTACKID)
5893                 zoneid = GLOBAL_ZONEID;
5894         else
5895                 zoneid = crgetzoneid(credp);
5896 
5897         ill = (ill_t *)mi_open_alloc_sleep(sizeof (ill_t));
5898         q->q_ptr = WR(q)->q_ptr = ill;
5899         ill->ill_ipst = ipst;
5900         ill->ill_zoneid = zoneid;
5901 
5902         /*
5903          * ill_init initializes the ill fields and then sends down
5904          * down a DL_INFO_REQ after calling qprocson.
5905          */
5906         err = ill_init(q, ill);
5907 
5908         if (err != 0) {
5909                 mi_free(ill);
5910                 netstack_rele(ipst->ips_netstack);
5911                 q->q_ptr = NULL;
5912                 WR(q)->q_ptr = NULL;
5913                 return (err);
5914         }
5915 
5916         /*
5917          * Wait for the DL_INFO_ACK if a DL_INFO_REQ was sent.
5918          *
5919          * ill_init initializes the ipsq marking this thread as
5920          * writer
5921          */
5922         ipsq_exit(ill->ill_phyint->phyint_ipsq);
5923         err = ip_wait_for_info_ack(ill);
5924         if (err == 0)
5925                 ill->ill_credp = credp;
5926         else
5927                 goto fail;
5928 
5929         crhold(credp);
5930 
5931         mutex_enter(&ipst->ips_ip_mi_lock);
5932         err = mi_open_link(&ipst->ips_ip_g_head, (IDP)q->q_ptr, devp, flag,
5933             sflag, credp);
5934         mutex_exit(&ipst->ips_ip_mi_lock);
5935 fail:
5936         if (err) {
5937                 (void) ip_close(q, 0, credp);
5938                 return (err);
5939         }
5940         return (0);
5941 }
5942 
5943 /* For /dev/ip aka AF_INET open */
5944 int
5945 ip_openv4(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
5946 {
5947         return (ip_open(q, devp, flag, sflag, credp, B_FALSE));
5948 }
5949 
5950 /* For /dev/ip6 aka AF_INET6 open */
5951 int
5952 ip_openv6(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
5953 {
5954         return (ip_open(q, devp, flag, sflag, credp, B_TRUE));
5955 }
5956 
5957 /* IP open routine. */
5958 int
5959 ip_open(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp,
5960     boolean_t isv6)
5961 {
5962         conn_t          *connp;
5963         major_t         maj;
5964         zoneid_t        zoneid;
5965         netstack_t      *ns;
5966         ip_stack_t      *ipst;
5967 
5968         /* Allow reopen. */
5969         if (q->q_ptr != NULL)
5970                 return (0);
5971 
5972         if (sflag & MODOPEN) {
5973                 /* This is a module open */
5974                 return (ip_modopen(q, devp, flag, sflag, credp));
5975         }
5976 
5977         if ((flag & ~(FKLYR)) == IP_HELPER_STR) {
5978                 /*
5979                  * Non streams based socket looking for a stream
5980                  * to access IP
5981                  */
5982                 return (ip_helper_stream_setup(q, devp, flag, sflag,
5983                     credp, isv6));
5984         }
5985 
5986         ns = netstack_find_by_cred(credp);
5987         ASSERT(ns != NULL);
5988         ipst = ns->netstack_ip;
5989         ASSERT(ipst != NULL);
5990 
5991         /*
5992          * For exclusive stacks we set the zoneid to zero
5993          * to make IP operate as if in the global zone.
5994          */
5995         if (ipst->ips_netstack->netstack_stackid != GLOBAL_NETSTACKID)
5996                 zoneid = GLOBAL_ZONEID;
5997         else
5998                 zoneid = crgetzoneid(credp);
5999 
6000         /*
6001          * We are opening as a device. This is an IP client stream, and we
6002          * allocate an conn_t as the instance data.
6003          */
6004         connp = ipcl_conn_create(IPCL_IPCCONN, KM_SLEEP, ipst->ips_netstack);
6005 
6006         /*
6007          * ipcl_conn_create did a netstack_hold. Undo the hold that was
6008          * done by netstack_find_by_cred()
6009          */
6010         netstack_rele(ipst->ips_netstack);
6011 
6012         connp->conn_ixa->ixa_flags |= IXAF_MULTICAST_LOOP | IXAF_SET_ULP_CKSUM;
6013         /* conn_allzones can not be set this early, hence no IPCL_ZONEID */
6014         connp->conn_ixa->ixa_zoneid = zoneid;
6015         connp->conn_zoneid = zoneid;
6016 
6017         connp->conn_rq = q;
6018         q->q_ptr = WR(q)->q_ptr = connp;
6019 
6020         /* Minor tells us which /dev entry was opened */
6021         if (isv6) {
6022                 connp->conn_family = AF_INET6;
6023                 connp->conn_ipversion = IPV6_VERSION;
6024                 connp->conn_ixa->ixa_flags &= ~IXAF_IS_IPV4;
6025                 connp->conn_ixa->ixa_src_preferences = IPV6_PREFER_SRC_DEFAULT;
6026         } else {
6027                 connp->conn_family = AF_INET;
6028                 connp->conn_ipversion = IPV4_VERSION;
6029                 connp->conn_ixa->ixa_flags |= IXAF_IS_IPV4;
6030         }
6031 
6032         if ((ip_minor_arena_la != NULL) && (flag & SO_SOCKSTR) &&
6033             ((connp->conn_dev = inet_minor_alloc(ip_minor_arena_la)) != 0)) {
6034                 connp->conn_minor_arena = ip_minor_arena_la;
6035         } else {
6036                 /*
6037                  * Either minor numbers in the large arena were exhausted
6038                  * or a non socket application is doing the open.
6039                  * Try to allocate from the small arena.
6040                  */
6041                 if ((connp->conn_dev =
6042                     inet_minor_alloc(ip_minor_arena_sa)) == 0) {
6043                         /* CONN_DEC_REF takes care of netstack_rele() */
6044                         q->q_ptr = WR(q)->q_ptr = NULL;
6045                         CONN_DEC_REF(connp);
6046                         return (EBUSY);
6047                 }
6048                 connp->conn_minor_arena = ip_minor_arena_sa;
6049         }
6050 
6051         maj = getemajor(*devp);
6052         *devp = makedevice(maj, (minor_t)connp->conn_dev);
6053 
6054         /*
6055          * connp->conn_cred is crfree()ed in ipcl_conn_destroy()
6056          */
6057         connp->conn_cred = credp;
6058         connp->conn_cpid = curproc->p_pid;
6059         /* Cache things in ixa without an extra refhold */
6060         ASSERT(!(connp->conn_ixa->ixa_free_flags & IXA_FREE_CRED));
6061         connp->conn_ixa->ixa_cred = connp->conn_cred;
6062         connp->conn_ixa->ixa_cpid = connp->conn_cpid;
6063         if (is_system_labeled())
6064                 connp->conn_ixa->ixa_tsl = crgetlabel(connp->conn_cred);
6065 
6066         /*
6067          * Handle IP_IOC_RTS_REQUEST and other ioctls which use conn_recv
6068          */
6069         connp->conn_recv = ip_conn_input;
6070         connp->conn_recvicmp = ip_conn_input_icmp;
6071 
6072         crhold(connp->conn_cred);
6073 
6074         /*
6075          * If the caller has the process-wide flag set, then default to MAC
6076          * exempt mode.  This allows read-down to unlabeled hosts.
6077          */
6078         if (getpflags(NET_MAC_AWARE, credp) != 0)
6079                 connp->conn_mac_mode = CONN_MAC_AWARE;
6080 
6081         connp->conn_zone_is_global = (crgetzoneid(credp) == GLOBAL_ZONEID);
6082 
6083         connp->conn_rq = q;
6084         connp->conn_wq = WR(q);
6085 
6086         /* Non-zero default values */
6087         connp->conn_ixa->ixa_flags |= IXAF_MULTICAST_LOOP;
6088 
6089         /*
6090          * Make the conn globally visible to walkers
6091          */
6092         ASSERT(connp->conn_ref == 1);
6093         mutex_enter(&connp->conn_lock);
6094         connp->conn_state_flags &= ~CONN_INCIPIENT;
6095         mutex_exit(&connp->conn_lock);
6096 
6097         qprocson(q);
6098 
6099         return (0);
6100 }
6101 
6102 /*
6103  * Set IPsec policy from an ipsec_req_t. If the req is not "zero" and valid,
6104  * all of them are copied to the conn_t. If the req is "zero", the policy is
6105  * zeroed out. A "zero" policy has zero ipsr_{ah,req,self_encap}_req
6106  * fields.
6107  * We keep only the latest setting of the policy and thus policy setting
6108  * is not incremental/cumulative.
6109  *
6110  * Requests to set policies with multiple alternative actions will
6111  * go through a different API.
6112  */
6113 int
6114 ipsec_set_req(cred_t *cr, conn_t *connp, ipsec_req_t *req)
6115 {
6116         uint_t ah_req = 0;
6117         uint_t esp_req = 0;
6118         uint_t se_req = 0;
6119         ipsec_act_t *actp = NULL;
6120         uint_t nact;
6121         ipsec_policy_head_t *ph;
6122         boolean_t is_pol_reset, is_pol_inserted = B_FALSE;
6123         int error = 0;
6124         netstack_t      *ns = connp->conn_netstack;
6125         ip_stack_t      *ipst = ns->netstack_ip;
6126         ipsec_stack_t   *ipss = ns->netstack_ipsec;
6127 
6128 #define REQ_MASK (IPSEC_PREF_REQUIRED|IPSEC_PREF_NEVER)
6129 
6130         /*
6131          * The IP_SEC_OPT option does not allow variable length parameters,
6132          * hence a request cannot be NULL.
6133          */
6134         if (req == NULL)
6135                 return (EINVAL);
6136 
6137         ah_req = req->ipsr_ah_req;
6138         esp_req = req->ipsr_esp_req;
6139         se_req = req->ipsr_self_encap_req;
6140 
6141         /* Don't allow setting self-encap without one or more of AH/ESP. */
6142         if (se_req != 0 && esp_req == 0 && ah_req == 0)
6143                 return (EINVAL);
6144 
6145         /*
6146          * Are we dealing with a request to reset the policy (i.e.
6147          * zero requests).
6148          */
6149         is_pol_reset = ((ah_req & REQ_MASK) == 0 &&
6150             (esp_req & REQ_MASK) == 0 &&
6151             (se_req & REQ_MASK) == 0);
6152 
6153         if (!is_pol_reset) {
6154                 /*
6155                  * If we couldn't load IPsec, fail with "protocol
6156                  * not supported".
6157                  * IPsec may not have been loaded for a request with zero
6158                  * policies, so we don't fail in this case.
6159                  */
6160                 mutex_enter(&ipss->ipsec_loader_lock);
6161                 if (ipss->ipsec_loader_state != IPSEC_LOADER_SUCCEEDED) {
6162                         mutex_exit(&ipss->ipsec_loader_lock);
6163                         return (EPROTONOSUPPORT);
6164                 }
6165                 mutex_exit(&ipss->ipsec_loader_lock);
6166 
6167                 /*
6168                  * Test for valid requests. Invalid algorithms
6169                  * need to be tested by IPsec code because new
6170                  * algorithms can be added dynamically.
6171                  */
6172                 if ((ah_req & ~(REQ_MASK|IPSEC_PREF_UNIQUE)) != 0 ||
6173                     (esp_req & ~(REQ_MASK|IPSEC_PREF_UNIQUE)) != 0 ||
6174                     (se_req & ~(REQ_MASK|IPSEC_PREF_UNIQUE)) != 0) {
6175                         return (EINVAL);
6176                 }
6177 
6178                 /*
6179                  * Only privileged users can issue these
6180                  * requests.
6181                  */
6182                 if (((ah_req & IPSEC_PREF_NEVER) ||
6183                     (esp_req & IPSEC_PREF_NEVER) ||
6184                     (se_req & IPSEC_PREF_NEVER)) &&
6185                     secpolicy_ip_config(cr, B_FALSE) != 0) {
6186                         return (EPERM);
6187                 }
6188 
6189                 /*
6190                  * The IPSEC_PREF_REQUIRED and IPSEC_PREF_NEVER
6191                  * are mutually exclusive.
6192                  */
6193                 if (((ah_req & REQ_MASK) == REQ_MASK) ||
6194                     ((esp_req & REQ_MASK) == REQ_MASK) ||
6195                     ((se_req & REQ_MASK) == REQ_MASK)) {
6196                         /* Both of them are set */
6197                         return (EINVAL);
6198                 }
6199         }
6200 
6201         ASSERT(MUTEX_HELD(&connp->conn_lock));
6202 
6203         /*
6204          * If we have already cached policies in conn_connect(), don't
6205          * let them change now. We cache policies for connections
6206          * whose src,dst [addr, port] is known.
6207          */
6208         if (connp->conn_policy_cached) {
6209                 return (EINVAL);
6210         }
6211 
6212         /*
6213          * We have a zero policies, reset the connection policy if already
6214          * set. This will cause the connection to inherit the
6215          * global policy, if any.
6216          */
6217         if (is_pol_reset) {
6218                 if (connp->conn_policy != NULL) {
6219                         IPPH_REFRELE(connp->conn_policy, ipst->ips_netstack);
6220                         connp->conn_policy = NULL;
6221                 }
6222                 connp->conn_in_enforce_policy = B_FALSE;
6223                 connp->conn_out_enforce_policy = B_FALSE;
6224                 return (0);
6225         }
6226 
6227         ph = connp->conn_policy = ipsec_polhead_split(connp->conn_policy,
6228             ipst->ips_netstack);
6229         if (ph == NULL)
6230                 goto enomem;
6231 
6232         ipsec_actvec_from_req(req, &actp, &nact, ipst->ips_netstack);
6233         if (actp == NULL)
6234                 goto enomem;
6235 
6236         /*
6237          * Always insert IPv4 policy entries, since they can also apply to
6238          * ipv6 sockets being used in ipv4-compat mode.
6239          */
6240         if (!ipsec_polhead_insert(ph, actp, nact, IPSEC_AF_V4,
6241             IPSEC_TYPE_INBOUND, ns))
6242                 goto enomem;
6243         is_pol_inserted = B_TRUE;
6244         if (!ipsec_polhead_insert(ph, actp, nact, IPSEC_AF_V4,
6245             IPSEC_TYPE_OUTBOUND, ns))
6246                 goto enomem;
6247 
6248         /*
6249          * We're looking at a v6 socket, also insert the v6-specific
6250          * entries.
6251          */
6252         if (connp->conn_family == AF_INET6) {
6253                 if (!ipsec_polhead_insert(ph, actp, nact, IPSEC_AF_V6,
6254                     IPSEC_TYPE_INBOUND, ns))
6255                         goto enomem;
6256                 if (!ipsec_polhead_insert(ph, actp, nact, IPSEC_AF_V6,
6257                     IPSEC_TYPE_OUTBOUND, ns))
6258                         goto enomem;
6259         }
6260 
6261         ipsec_actvec_free(actp, nact);
6262 
6263         /*
6264          * If the requests need security, set enforce_policy.
6265          * If the requests are IPSEC_PREF_NEVER, one should
6266          * still set conn_out_enforce_policy so that ip_set_destination
6267          * marks the ip_xmit_attr_t appropriatly. This is needed so that
6268          * for connections that we don't cache policy in at connect time,
6269          * if global policy matches in ip_output_attach_policy, we
6270          * don't wrongly inherit global policy. Similarly, we need
6271          * to set conn_in_enforce_policy also so that we don't verify
6272          * policy wrongly.
6273          */
6274         if ((ah_req & REQ_MASK) != 0 ||
6275             (esp_req & REQ_MASK) != 0 ||
6276             (se_req & REQ_MASK) != 0) {
6277                 connp->conn_in_enforce_policy = B_TRUE;
6278                 connp->conn_out_enforce_policy = B_TRUE;
6279         }
6280 
6281         return (error);
6282 #undef REQ_MASK
6283 
6284         /*
6285          * Common memory-allocation-failure exit path.
6286          */
6287 enomem:
6288         if (actp != NULL)
6289                 ipsec_actvec_free(actp, nact);
6290         if (is_pol_inserted)
6291                 ipsec_polhead_flush(ph, ns);
6292         return (ENOMEM);
6293 }
6294 
6295 /*
6296  * Set socket options for joining and leaving multicast groups.
6297  * Common to IPv4 and IPv6; inet6 indicates the type of socket.
6298  * The caller has already check that the option name is consistent with
6299  * the address family of the socket.
6300  */
6301 int
6302 ip_opt_set_multicast_group(conn_t *connp, t_scalar_t name,
6303     uchar_t *invalp, boolean_t inet6, boolean_t checkonly)
6304 {
6305         int             *i1 = (int *)invalp;
6306         int             error = 0;
6307         ip_stack_t      *ipst = connp->conn_netstack->netstack_ip;
6308         struct ip_mreq  *v4_mreqp;
6309         struct ipv6_mreq *v6_mreqp;
6310         struct group_req *greqp;
6311         ire_t *ire;
6312         boolean_t done = B_FALSE;
6313         ipaddr_t ifaddr;
6314         in6_addr_t v6group;
6315         uint_t ifindex;
6316         boolean_t mcast_opt = B_TRUE;
6317         mcast_record_t fmode;
6318         int (*optfn)(conn_t *, boolean_t, const in6_addr_t *,
6319             ipaddr_t, uint_t, mcast_record_t, const in6_addr_t *);
6320 
6321         switch (name) {
6322         case IP_ADD_MEMBERSHIP:
6323         case IPV6_JOIN_GROUP:
6324                 mcast_opt = B_FALSE;
6325                 /* FALLTHROUGH */
6326         case MCAST_JOIN_GROUP:
6327                 fmode = MODE_IS_EXCLUDE;
6328                 optfn = ip_opt_add_group;
6329                 break;
6330 
6331         case IP_DROP_MEMBERSHIP:
6332         case IPV6_LEAVE_GROUP:
6333                 mcast_opt = B_FALSE;
6334                 /* FALLTHROUGH */
6335         case MCAST_LEAVE_GROUP:
6336                 fmode = MODE_IS_INCLUDE;
6337                 optfn = ip_opt_delete_group;
6338                 break;
6339         default:
6340                 ASSERT(0);
6341         }
6342 
6343         if (mcast_opt) {
6344                 struct sockaddr_in *sin;
6345                 struct sockaddr_in6 *sin6;
6346 
6347                 greqp = (struct group_req *)i1;
6348                 if (greqp->gr_group.ss_family == AF_INET) {
6349                         sin = (struct sockaddr_in *)&(greqp->gr_group);
6350                         IN6_INADDR_TO_V4MAPPED(&sin->sin_addr, &v6group);
6351                 } else {
6352                         if (!inet6)
6353                                 return (EINVAL);        /* Not on INET socket */
6354 
6355                         sin6 = (struct sockaddr_in6 *)&(greqp->gr_group);
6356                         v6group = sin6->sin6_addr;
6357                 }
6358                 ifaddr = INADDR_ANY;
6359                 ifindex = greqp->gr_interface;
6360         } else if (inet6) {
6361                 v6_mreqp = (struct ipv6_mreq *)i1;
6362                 v6group = v6_mreqp->ipv6mr_multiaddr;
6363                 ifaddr = INADDR_ANY;
6364                 ifindex = v6_mreqp->ipv6mr_interface;
6365         } else {
6366                 v4_mreqp = (struct ip_mreq *)i1;
6367                 IN6_INADDR_TO_V4MAPPED(&v4_mreqp->imr_multiaddr, &v6group);
6368                 ifaddr = (ipaddr_t)v4_mreqp->imr_interface.s_addr;
6369                 ifindex = 0;
6370         }
6371 
6372         /*
6373          * In the multirouting case, we need to replicate
6374          * the request on all interfaces that will take part
6375          * in replication.  We do so because multirouting is
6376          * reflective, thus we will probably receive multi-
6377          * casts on those interfaces.
6378          * The ip_multirt_apply_membership() succeeds if
6379          * the operation succeeds on at least one interface.
6380          */
6381         if (IN6_IS_ADDR_V4MAPPED(&v6group)) {
6382                 ipaddr_t group;
6383 
6384                 IN6_V4MAPPED_TO_IPADDR(&v6group, group);
6385 
6386                 ire = ire_ftable_lookup_v4(group, IP_HOST_MASK, 0,
6387                     IRE_HOST | IRE_INTERFACE, NULL, ALL_ZONES, NULL,
6388                     MATCH_IRE_MASK | MATCH_IRE_TYPE, 0, ipst, NULL);
6389         } else {
6390                 ire = ire_ftable_lookup_v6(&v6group, &ipv6_all_ones, 0,
6391                     IRE_HOST | IRE_INTERFACE, NULL, ALL_ZONES, NULL,
6392                     MATCH_IRE_MASK | MATCH_IRE_TYPE, 0, ipst, NULL);
6393         }
6394         if (ire != NULL) {
6395                 if (ire->ire_flags & RTF_MULTIRT) {
6396                         error = ip_multirt_apply_membership(optfn, ire, connp,
6397                             checkonly, &v6group, fmode, &ipv6_all_zeros);
6398                         done = B_TRUE;
6399                 }
6400                 ire_refrele(ire);
6401         }
6402 
6403         if (!done) {
6404                 error = optfn(connp, checkonly, &v6group, ifaddr, ifindex,
6405                     fmode, &ipv6_all_zeros);
6406         }
6407         return (error);
6408 }
6409 
6410 /*
6411  * Set socket options for joining and leaving multicast groups
6412  * for specific sources.
6413  * Common to IPv4 and IPv6; inet6 indicates the type of socket.
6414  * The caller has already check that the option name is consistent with
6415  * the address family of the socket.
6416  */
6417 int
6418 ip_opt_set_multicast_sources(conn_t *connp, t_scalar_t name,
6419     uchar_t *invalp, boolean_t inet6, boolean_t checkonly)
6420 {
6421         int             *i1 = (int *)invalp;
6422         int             error = 0;
6423         ip_stack_t      *ipst = connp->conn_netstack->netstack_ip;
6424         struct ip_mreq_source *imreqp;
6425         struct group_source_req *gsreqp;
6426         in6_addr_t v6group, v6src;
6427         uint32_t ifindex;
6428         ipaddr_t ifaddr;
6429         boolean_t mcast_opt = B_TRUE;
6430         mcast_record_t fmode;
6431         ire_t *ire;
6432         boolean_t done = B_FALSE;
6433         int (*optfn)(conn_t *, boolean_t, const in6_addr_t *,
6434             ipaddr_t, uint_t, mcast_record_t, const in6_addr_t *);
6435 
6436         switch (name) {
6437         case IP_BLOCK_SOURCE:
6438                 mcast_opt = B_FALSE;
6439                 /* FALLTHROUGH */
6440         case MCAST_BLOCK_SOURCE:
6441                 fmode = MODE_IS_EXCLUDE;
6442                 optfn = ip_opt_add_group;
6443                 break;
6444 
6445         case IP_UNBLOCK_SOURCE:
6446                 mcast_opt = B_FALSE;
6447                 /* FALLTHROUGH */
6448         case MCAST_UNBLOCK_SOURCE:
6449                 fmode = MODE_IS_EXCLUDE;
6450                 optfn = ip_opt_delete_group;
6451                 break;
6452 
6453         case IP_ADD_SOURCE_MEMBERSHIP:
6454                 mcast_opt = B_FALSE;
6455                 /* FALLTHROUGH */
6456         case MCAST_JOIN_SOURCE_GROUP:
6457                 fmode = MODE_IS_INCLUDE;
6458                 optfn = ip_opt_add_group;
6459                 break;
6460 
6461         case IP_DROP_SOURCE_MEMBERSHIP:
6462                 mcast_opt = B_FALSE;
6463                 /* FALLTHROUGH */
6464         case MCAST_LEAVE_SOURCE_GROUP:
6465                 fmode = MODE_IS_INCLUDE;
6466                 optfn = ip_opt_delete_group;
6467                 break;
6468         default:
6469                 ASSERT(0);
6470         }
6471 
6472         if (mcast_opt) {
6473                 gsreqp = (struct group_source_req *)i1;
6474                 ifindex = gsreqp->gsr_interface;
6475                 if (gsreqp->gsr_group.ss_family == AF_INET) {
6476                         struct sockaddr_in *s;
6477                         s = (struct sockaddr_in *)&gsreqp->gsr_group;
6478                         IN6_INADDR_TO_V4MAPPED(&s->sin_addr, &v6group);
6479                         s = (struct sockaddr_in *)&gsreqp->gsr_source;
6480                         IN6_INADDR_TO_V4MAPPED(&s->sin_addr, &v6src);
6481                 } else {
6482                         struct sockaddr_in6 *s6;
6483 
6484                         if (!inet6)
6485                                 return (EINVAL);        /* Not on INET socket */
6486 
6487                         s6 = (struct sockaddr_in6 *)&gsreqp->gsr_group;
6488                         v6group = s6->sin6_addr;
6489                         s6 = (struct sockaddr_in6 *)&gsreqp->gsr_source;
6490                         v6src = s6->sin6_addr;
6491                 }
6492                 ifaddr = INADDR_ANY;
6493         } else {
6494                 imreqp = (struct ip_mreq_source *)i1;
6495                 IN6_INADDR_TO_V4MAPPED(&imreqp->imr_multiaddr, &v6group);
6496                 IN6_INADDR_TO_V4MAPPED(&imreqp->imr_sourceaddr, &v6src);
6497                 ifaddr = (ipaddr_t)imreqp->imr_interface.s_addr;
6498                 ifindex = 0;
6499         }
6500 
6501         /*
6502          * Handle src being mapped INADDR_ANY by changing it to unspecified.
6503          */
6504         if (IN6_IS_ADDR_V4MAPPED_ANY(&v6src))
6505                 v6src = ipv6_all_zeros;
6506 
6507         /*
6508          * In the multirouting case, we need to replicate
6509          * the request as noted in the mcast cases above.
6510          */
6511         if (IN6_IS_ADDR_V4MAPPED(&v6group)) {
6512                 ipaddr_t group;
6513 
6514                 IN6_V4MAPPED_TO_IPADDR(&v6group, group);
6515 
6516                 ire = ire_ftable_lookup_v4(group, IP_HOST_MASK, 0,
6517                     IRE_HOST | IRE_INTERFACE, NULL, ALL_ZONES, NULL,
6518                     MATCH_IRE_MASK | MATCH_IRE_TYPE, 0, ipst, NULL);
6519         } else {
6520                 ire = ire_ftable_lookup_v6(&v6group, &ipv6_all_ones, 0,
6521                     IRE_HOST | IRE_INTERFACE, NULL, ALL_ZONES, NULL,
6522                     MATCH_IRE_MASK | MATCH_IRE_TYPE, 0, ipst, NULL);
6523         }
6524         if (ire != NULL) {
6525                 if (ire->ire_flags & RTF_MULTIRT) {
6526                         error = ip_multirt_apply_membership(optfn, ire, connp,
6527                             checkonly, &v6group, fmode, &v6src);
6528                         done = B_TRUE;
6529                 }
6530                 ire_refrele(ire);
6531         }
6532         if (!done) {
6533                 error = optfn(connp, checkonly, &v6group, ifaddr, ifindex,
6534                     fmode, &v6src);
6535         }
6536         return (error);
6537 }
6538 
6539 /*
6540  * Given a destination address and a pointer to where to put the information
6541  * this routine fills in the mtuinfo.
6542  * The socket must be connected.
6543  * For sctp conn_faddr is the primary address.
6544  */
6545 int
6546 ip_fill_mtuinfo(conn_t *connp, ip_xmit_attr_t *ixa, struct ip6_mtuinfo *mtuinfo)
6547 {
6548         uint32_t        pmtu = IP_MAXPACKET;
6549         uint_t          scopeid;
6550 
6551         if (IN6_IS_ADDR_UNSPECIFIED(&connp->conn_faddr_v6))
6552                 return (-1);
6553 
6554         /* In case we never sent or called ip_set_destination_v4/v6 */
6555         if (ixa->ixa_ire != NULL)
6556                 pmtu = ip_get_pmtu(ixa);
6557 
6558         if (ixa->ixa_flags & IXAF_SCOPEID_SET)
6559                 scopeid = ixa->ixa_scopeid;
6560         else
6561                 scopeid = 0;
6562 
6563         bzero(mtuinfo, sizeof (*mtuinfo));
6564         mtuinfo->ip6m_addr.sin6_family = AF_INET6;
6565         mtuinfo->ip6m_addr.sin6_port = connp->conn_fport;
6566         mtuinfo->ip6m_addr.sin6_addr = connp->conn_faddr_v6;
6567         mtuinfo->ip6m_addr.sin6_scope_id = scopeid;
6568         mtuinfo->ip6m_mtu = pmtu;
6569 
6570         return (sizeof (struct ip6_mtuinfo));
6571 }
6572 
6573 /*
6574  * When the src multihoming is changed from weak to [strong, preferred]
6575  * ip_ire_rebind_walker is called to walk the list of all ire_t entries
6576  * and identify routes that were created by user-applications in the
6577  * unbound state (i.e., without RTA_IFP), and for which an ire_ill is not
6578  * currently defined. These routes are then 'rebound', i.e., their ire_ill
6579  * is selected by finding an interface route for the gateway.
6580  */
6581 /* ARGSUSED */
6582 void
6583 ip_ire_rebind_walker(ire_t *ire, void *notused)
6584 {
6585         if (!ire->ire_unbound || ire->ire_ill != NULL)
6586                 return;
6587         ire_rebind(ire);
6588         ire_delete(ire);
6589 }
6590 
6591 /*
6592  * When the src multihoming is changed from  [strong, preferred] to weak,
6593  * ip_ire_unbind_walker is called to walk the list of all ire_t entries, and
6594  * set any entries that were created by user-applications in the unbound state
6595  * (i.e., without RTA_IFP) back to having a NULL ire_ill.
6596  */
6597 /* ARGSUSED */
6598 void
6599 ip_ire_unbind_walker(ire_t *ire, void *notused)
6600 {
6601         ire_t *new_ire;
6602 
6603         if (!ire->ire_unbound || ire->ire_ill == NULL)
6604                 return;
6605         if (ire->ire_ipversion == IPV6_VERSION) {
6606                 new_ire = ire_create_v6(&ire->ire_addr_v6, &ire->ire_mask_v6,
6607                     &ire->ire_gateway_addr_v6, ire->ire_type, NULL,
6608                     ire->ire_zoneid, ire->ire_flags, NULL, ire->ire_ipst);
6609         } else {
6610                 new_ire = ire_create((uchar_t *)&ire->ire_addr,
6611                     (uchar_t *)&ire->ire_mask,
6612                     (uchar_t *)&ire->ire_gateway_addr, ire->ire_type, NULL,
6613                     ire->ire_zoneid, ire->ire_flags, NULL, ire->ire_ipst);
6614         }
6615         if (new_ire == NULL)
6616                 return;
6617         new_ire->ire_unbound = B_TRUE;
6618         /*
6619          * The bound ire must first be deleted so that we don't return
6620          * the existing one on the attempt to add the unbound new_ire.
6621          */
6622         ire_delete(ire);
6623         new_ire = ire_add(new_ire);
6624         if (new_ire != NULL)
6625                 ire_refrele(new_ire);
6626 }
6627 
6628 /*
6629  * When the settings of ip*_strict_src_multihoming tunables are changed,
6630  * all cached routes need to be recomputed. This recomputation needs to be
6631  * done when going from weaker to stronger modes so that the cached ire
6632  * for the connection does not violate the current ip*_strict_src_multihoming
6633  * setting. It also needs to be done when going from stronger to weaker modes,
6634  * so that we fall back to matching on the longest-matching-route (as opposed
6635  * to a shorter match that may have been selected in the strong mode
6636  * to satisfy src_multihoming settings).
6637  *
6638  * The cached ixa_ire entires for all conn_t entries are marked as
6639  * "verify" so that they will be recomputed for the next packet.
6640  */
6641 void
6642 conn_ire_revalidate(conn_t *connp, void *arg)
6643 {
6644         boolean_t isv6 = (boolean_t)arg;
6645 
6646         if ((isv6 && connp->conn_ipversion != IPV6_VERSION) ||
6647             (!isv6 && connp->conn_ipversion != IPV4_VERSION))
6648                 return;
6649         connp->conn_ixa->ixa_ire_generation = IRE_GENERATION_VERIFY;
6650 }
6651 
6652 /*
6653  * Handles both IPv4 and IPv6 reassembly - doing the out-of-order cases,
6654  * When an ipf is passed here for the first time, if
6655  * we already have in-order fragments on the queue, we convert from the fast-
6656  * path reassembly scheme to the hard-case scheme.  From then on, additional
6657  * fragments are reassembled here.  We keep track of the start and end offsets
6658  * of each piece, and the number of holes in the chain.  When the hole count
6659  * goes to zero, we are done!
6660  *
6661  * The ipf_count will be updated to account for any mblk(s) added (pointed to
6662  * by mp) or subtracted (freeb()ed dups), upon return the caller must update
6663  * ipfb_count and ill_frag_count by the difference of ipf_count before and
6664  * after the call to ip_reassemble().
6665  */
6666 int
6667 ip_reassemble(mblk_t *mp, ipf_t *ipf, uint_t start, boolean_t more, ill_t *ill,
6668     size_t msg_len)
6669 {
6670         uint_t  end;
6671         mblk_t  *next_mp;
6672         mblk_t  *mp1;
6673         uint_t  offset;
6674         boolean_t incr_dups = B_TRUE;
6675         boolean_t offset_zero_seen = B_FALSE;
6676         boolean_t pkt_boundary_checked = B_FALSE;
6677 
6678         /* If start == 0 then ipf_nf_hdr_len has to be set. */
6679         ASSERT(start != 0 || ipf->ipf_nf_hdr_len != 0);
6680 
6681         /* Add in byte count */
6682         ipf->ipf_count += msg_len;
6683         if (ipf->ipf_end) {
6684                 /*
6685                  * We were part way through in-order reassembly, but now there
6686                  * is a hole.  We walk through messages already queued, and
6687                  * mark them for hard case reassembly.  We know that up till
6688                  * now they were in order starting from offset zero.
6689                  */
6690                 offset = 0;
6691                 for (mp1 = ipf->ipf_mp->b_cont; mp1; mp1 = mp1->b_cont) {
6692                         IP_REASS_SET_START(mp1, offset);
6693                         if (offset == 0) {
6694                                 ASSERT(ipf->ipf_nf_hdr_len != 0);
6695                                 offset = -ipf->ipf_nf_hdr_len;
6696                         }
6697                         offset += mp1->b_wptr - mp1->b_rptr;
6698                         IP_REASS_SET_END(mp1, offset);
6699                 }
6700                 /* One hole at the end. */
6701                 ipf->ipf_hole_cnt = 1;
6702                 /* Brand it as a hard case, forever. */
6703                 ipf->ipf_end = 0;
6704         }
6705         /* Walk through all the new pieces. */
6706         do {
6707                 end = start + (mp->b_wptr - mp->b_rptr);
6708                 /*
6709                  * If start is 0, decrease 'end' only for the first mblk of
6710                  * the fragment. Otherwise 'end' can get wrong value in the
6711                  * second pass of the loop if first mblk is exactly the
6712                  * size of ipf_nf_hdr_len.
6713                  */
6714                 if (start == 0 && !offset_zero_seen) {
6715                         /* First segment */
6716                         ASSERT(ipf->ipf_nf_hdr_len != 0);
6717                         end -= ipf->ipf_nf_hdr_len;
6718                         offset_zero_seen = B_TRUE;
6719                 }
6720                 next_mp = mp->b_cont;
6721                 /*
6722                  * We are checking to see if there is any interesing data
6723                  * to process.  If there isn't and the mblk isn't the
6724                  * one which carries the unfragmentable header then we
6725                  * drop it.  It's possible to have just the unfragmentable
6726                  * header come through without any data.  That needs to be
6727                  * saved.
6728                  *
6729                  * If the assert at the top of this function holds then the
6730                  * term "ipf->ipf_nf_hdr_len != 0" isn't needed.  This code
6731                  * is infrequently traveled enough that the test is left in
6732                  * to protect against future code changes which break that
6733                  * invariant.
6734                  */
6735                 if (start == end && start != 0 && ipf->ipf_nf_hdr_len != 0) {
6736                         /* Empty.  Blast it. */
6737                         IP_REASS_SET_START(mp, 0);
6738                         IP_REASS_SET_END(mp, 0);
6739                         /*
6740                          * If the ipf points to the mblk we are about to free,
6741                          * update ipf to point to the next mblk (or NULL
6742                          * if none).
6743                          */
6744                         if (ipf->ipf_mp->b_cont == mp)
6745                                 ipf->ipf_mp->b_cont = next_mp;
6746                         freeb(mp);
6747                         continue;
6748                 }
6749                 mp->b_cont = NULL;
6750                 IP_REASS_SET_START(mp, start);
6751                 IP_REASS_SET_END(mp, end);
6752                 if (!ipf->ipf_tail_mp) {
6753                         ipf->ipf_tail_mp = mp;
6754                         ipf->ipf_mp->b_cont = mp;
6755                         if (start == 0 || !more) {
6756                                 ipf->ipf_hole_cnt = 1;
6757                                 /*
6758                                  * if the first fragment comes in more than one
6759                                  * mblk, this loop will be executed for each
6760                                  * mblk. Need to adjust hole count so exiting
6761                                  * this routine will leave hole count at 1.
6762                                  */
6763                                 if (next_mp)
6764                                         ipf->ipf_hole_cnt++;
6765                         } else
6766                                 ipf->ipf_hole_cnt = 2;
6767                         continue;
6768                 } else if (ipf->ipf_last_frag_seen && !more &&
6769                     !pkt_boundary_checked) {
6770                         /*
6771                          * We check datagram boundary only if this fragment
6772                          * claims to be the last fragment and we have seen a
6773                          * last fragment in the past too. We do this only
6774                          * once for a given fragment.
6775                          *
6776                          * start cannot be 0 here as fragments with start=0
6777                          * and MF=0 gets handled as a complete packet. These
6778                          * fragments should not reach here.
6779                          */
6780 
6781                         if (start + msgdsize(mp) !=
6782                             IP_REASS_END(ipf->ipf_tail_mp)) {
6783                                 /*
6784                                  * We have two fragments both of which claim
6785                                  * to be the last fragment but gives conflicting
6786                                  * information about the whole datagram size.
6787                                  * Something fishy is going on. Drop the
6788                                  * fragment and free up the reassembly list.
6789                                  */
6790                                 return (IP_REASS_FAILED);
6791                         }
6792 
6793                         /*
6794                          * We shouldn't come to this code block again for this
6795                          * particular fragment.
6796                          */
6797                         pkt_boundary_checked = B_TRUE;
6798                 }
6799 
6800                 /* New stuff at or beyond tail? */
6801                 offset = IP_REASS_END(ipf->ipf_tail_mp);
6802                 if (start >= offset) {
6803                         if (ipf->ipf_last_frag_seen) {
6804                                 /* current fragment is beyond last fragment */
6805                                 return (IP_REASS_FAILED);
6806                         }
6807                         /* Link it on end. */
6808                         ipf->ipf_tail_mp->b_cont = mp;
6809                         ipf->ipf_tail_mp = mp;
6810                         if (more) {
6811                                 if (start != offset)
6812                                         ipf->ipf_hole_cnt++;
6813                         } else if (start == offset && next_mp == NULL)
6814                                         ipf->ipf_hole_cnt--;
6815                         continue;
6816                 }
6817                 mp1 = ipf->ipf_mp->b_cont;
6818                 offset = IP_REASS_START(mp1);
6819                 /* New stuff at the front? */
6820                 if (start < offset) {
6821                         if (start == 0) {
6822                                 if (end >= offset) {
6823                                         /* Nailed the hole at the begining. */
6824                                         ipf->ipf_hole_cnt--;
6825                                 }
6826                         } else if (end < offset) {
6827                                 /*
6828                                  * A hole, stuff, and a hole where there used
6829                                  * to be just a hole.
6830                                  */
6831                                 ipf->ipf_hole_cnt++;
6832                         }
6833                         mp->b_cont = mp1;
6834                         /* Check for overlap. */
6835                         while (end > offset) {
6836                                 if (end < IP_REASS_END(mp1)) {
6837                                         mp->b_wptr -= end - offset;
6838                                         IP_REASS_SET_END(mp, offset);
6839                                         BUMP_MIB(ill->ill_ip_mib,
6840                                             ipIfStatsReasmPartDups);
6841                                         break;
6842                                 }
6843                                 /* Did we cover another hole? */
6844                                 if ((mp1->b_cont &&
6845                                     IP_REASS_END(mp1) !=
6846                                     IP_REASS_START(mp1->b_cont) &&
6847                                     end >= IP_REASS_START(mp1->b_cont)) ||
6848                                     (!ipf->ipf_last_frag_seen && !more)) {
6849                                         ipf->ipf_hole_cnt--;
6850                                 }
6851                                 /* Clip out mp1. */
6852                                 if ((mp->b_cont = mp1->b_cont) == NULL) {
6853                                         /*
6854                                          * After clipping out mp1, this guy
6855                                          * is now hanging off the end.
6856                                          */
6857                                         ipf->ipf_tail_mp = mp;
6858                                 }
6859                                 IP_REASS_SET_START(mp1, 0);
6860                                 IP_REASS_SET_END(mp1, 0);
6861                                 /* Subtract byte count */
6862                                 ipf->ipf_count -= mp1->b_datap->db_lim -
6863                                     mp1->b_datap->db_base;
6864                                 freeb(mp1);
6865                                 BUMP_MIB(ill->ill_ip_mib,
6866                                     ipIfStatsReasmPartDups);
6867                                 mp1 = mp->b_cont;
6868                                 if (!mp1)
6869                                         break;
6870                                 offset = IP_REASS_START(mp1);
6871                         }
6872                         ipf->ipf_mp->b_cont = mp;
6873                         continue;
6874                 }
6875                 /*
6876                  * The new piece starts somewhere between the start of the head
6877                  * and before the end of the tail.
6878                  */
6879                 for (; mp1; mp1 = mp1->b_cont) {
6880                         offset = IP_REASS_END(mp1);
6881                         if (start < offset) {
6882                                 if (end <= offset) {
6883                                         /* Nothing new. */
6884                                         IP_REASS_SET_START(mp, 0);
6885                                         IP_REASS_SET_END(mp, 0);
6886                                         /* Subtract byte count */
6887                                         ipf->ipf_count -= mp->b_datap->db_lim -
6888                                             mp->b_datap->db_base;
6889                                         if (incr_dups) {
6890                                                 ipf->ipf_num_dups++;
6891                                                 incr_dups = B_FALSE;
6892                                         }
6893                                         freeb(mp);
6894                                         BUMP_MIB(ill->ill_ip_mib,
6895                                             ipIfStatsReasmDuplicates);
6896                                         break;
6897                                 }
6898                                 /*
6899                                  * Trim redundant stuff off beginning of new
6900                                  * piece.
6901                                  */
6902                                 IP_REASS_SET_START(mp, offset);
6903                                 mp->b_rptr += offset - start;
6904                                 BUMP_MIB(ill->ill_ip_mib,
6905                                     ipIfStatsReasmPartDups);
6906                                 start = offset;
6907                                 if (!mp1->b_cont) {
6908                                         /*
6909                                          * After trimming, this guy is now
6910                                          * hanging off the end.
6911                                          */
6912                                         mp1->b_cont = mp;
6913                                         ipf->ipf_tail_mp = mp;
6914                                         if (!more) {
6915                                                 ipf->ipf_hole_cnt--;
6916                                         }
6917                                         break;
6918                                 }
6919                         }
6920                         if (start >= IP_REASS_START(mp1->b_cont))
6921                                 continue;
6922                         /* Fill a hole */
6923                         if (start > offset)
6924                                 ipf->ipf_hole_cnt++;
6925                         mp->b_cont = mp1->b_cont;
6926                         mp1->b_cont = mp;
6927                         mp1 = mp->b_cont;
6928                         offset = IP_REASS_START(mp1);
6929                         if (end >= offset) {
6930                                 ipf->ipf_hole_cnt--;
6931                                 /* Check for overlap. */
6932                                 while (end > offset) {
6933                                         if (end < IP_REASS_END(mp1)) {
6934                                                 mp->b_wptr -= end - offset;
6935                                                 IP_REASS_SET_END(mp, offset);
6936                                                 /*
6937                                                  * TODO we might bump
6938                                                  * this up twice if there is
6939                                                  * overlap at both ends.
6940                                                  */
6941                                                 BUMP_MIB(ill->ill_ip_mib,
6942                                                     ipIfStatsReasmPartDups);
6943                                                 break;
6944                                         }
6945                                         /* Did we cover another hole? */
6946                                         if ((mp1->b_cont &&
6947                                             IP_REASS_END(mp1)
6948                                             != IP_REASS_START(mp1->b_cont) &&
6949                                             end >=
6950                                             IP_REASS_START(mp1->b_cont)) ||
6951                                             (!ipf->ipf_last_frag_seen &&
6952                                             !more)) {
6953                                                 ipf->ipf_hole_cnt--;
6954                                         }
6955                                         /* Clip out mp1. */
6956                                         if ((mp->b_cont = mp1->b_cont) ==
6957                                             NULL) {
6958                                                 /*
6959                                                  * After clipping out mp1,
6960                                                  * this guy is now hanging
6961                                                  * off the end.
6962                                                  */
6963                                                 ipf->ipf_tail_mp = mp;
6964                                         }
6965                                         IP_REASS_SET_START(mp1, 0);
6966                                         IP_REASS_SET_END(mp1, 0);
6967                                         /* Subtract byte count */
6968                                         ipf->ipf_count -=
6969                                             mp1->b_datap->db_lim -
6970                                             mp1->b_datap->db_base;
6971                                         freeb(mp1);
6972                                         BUMP_MIB(ill->ill_ip_mib,
6973                                             ipIfStatsReasmPartDups);
6974                                         mp1 = mp->b_cont;
6975                                         if (!mp1)
6976                                                 break;
6977                                         offset = IP_REASS_START(mp1);
6978                                 }
6979                         }
6980                         break;
6981                 }
6982         } while (start = end, mp = next_mp);
6983 
6984         /* Fragment just processed could be the last one. Remember this fact */
6985         if (!more)
6986                 ipf->ipf_last_frag_seen = B_TRUE;
6987 
6988         /* Still got holes? */
6989         if (ipf->ipf_hole_cnt)
6990                 return (IP_REASS_PARTIAL);
6991         /* Clean up overloaded fields to avoid upstream disasters. */
6992         for (mp1 = ipf->ipf_mp->b_cont; mp1; mp1 = mp1->b_cont) {
6993                 IP_REASS_SET_START(mp1, 0);
6994                 IP_REASS_SET_END(mp1, 0);
6995         }
6996         return (IP_REASS_COMPLETE);
6997 }
6998 
6999 /*
7000  * Fragmentation reassembly.  Each ILL has a hash table for
7001  * queuing packets undergoing reassembly for all IPIFs
7002  * associated with the ILL.  The hash is based on the packet
7003  * IP ident field.  The ILL frag hash table was allocated
7004  * as a timer block at the time the ILL was created.  Whenever
7005  * there is anything on the reassembly queue, the timer will
7006  * be running.  Returns the reassembled packet if reassembly completes.
7007  */
7008 mblk_t *
7009 ip_input_fragment(mblk_t *mp, ipha_t *ipha, ip_recv_attr_t *ira)
7010 {
7011         uint32_t        frag_offset_flags;
7012         mblk_t          *t_mp;
7013         ipaddr_t        dst;
7014         uint8_t         proto = ipha->ipha_protocol;
7015         uint32_t        sum_val;
7016         uint16_t        sum_flags;
7017         ipf_t           *ipf;
7018         ipf_t           **ipfp;
7019         ipfb_t          *ipfb;
7020         uint16_t        ident;
7021         uint32_t        offset;
7022         ipaddr_t        src;
7023         uint_t          hdr_length;
7024         uint32_t        end;
7025         mblk_t          *mp1;
7026         mblk_t          *tail_mp;
7027         size_t          count;
7028         size_t          msg_len;
7029         uint8_t         ecn_info = 0;
7030         uint32_t        packet_size;
7031         boolean_t       pruned = B_FALSE;
7032         ill_t           *ill = ira->ira_ill;
7033         ip_stack_t      *ipst = ill->ill_ipst;
7034 
7035         /*
7036          * Drop the fragmented as early as possible, if
7037          * we don't have resource(s) to re-assemble.
7038          */
7039         if (ipst->ips_ip_reass_queue_bytes == 0) {
7040                 freemsg(mp);
7041                 return (NULL);
7042         }
7043 
7044         /* Check for fragmentation offset; return if there's none */
7045         if ((frag_offset_flags = ntohs(ipha->ipha_fragment_offset_and_flags) &
7046             (IPH_MF | IPH_OFFSET)) == 0)
7047                 return (mp);
7048 
7049         /*
7050          * We utilize hardware computed checksum info only for UDP since
7051          * IP fragmentation is a normal occurrence for the protocol.  In
7052          * addition, checksum offload support for IP fragments carrying
7053          * UDP payload is commonly implemented across network adapters.
7054          */
7055         ASSERT(ira->ira_rill != NULL);
7056         if (proto == IPPROTO_UDP && dohwcksum &&
7057             ILL_HCKSUM_CAPABLE(ira->ira_rill) &&
7058             (DB_CKSUMFLAGS(mp) & (HCK_FULLCKSUM | HCK_PARTIALCKSUM))) {
7059                 mblk_t *mp1 = mp->b_cont;
7060                 int32_t len;
7061 
7062                 /* Record checksum information from the packet */
7063                 sum_val = (uint32_t)DB_CKSUM16(mp);
7064                 sum_flags = DB_CKSUMFLAGS(mp);
7065 
7066                 /* IP payload offset from beginning of mblk */
7067                 offset = ((uchar_t *)ipha + IPH_HDR_LENGTH(ipha)) - mp->b_rptr;
7068 
7069                 if ((sum_flags & HCK_PARTIALCKSUM) &&
7070                     (mp1 == NULL || mp1->b_cont == NULL) &&
7071                     offset >= DB_CKSUMSTART(mp) &&
7072                     ((len = offset - DB_CKSUMSTART(mp)) & 1) == 0) {
7073                         uint32_t adj;
7074                         /*
7075                          * Partial checksum has been calculated by hardware
7076                          * and attached to the packet; in addition, any
7077                          * prepended extraneous data is even byte aligned.
7078                          * If any such data exists, we adjust the checksum;
7079                          * this would also handle any postpended data.
7080                          */
7081                         IP_ADJCKSUM_PARTIAL(mp->b_rptr + DB_CKSUMSTART(mp),
7082                             mp, mp1, len, adj);
7083 
7084                         /* One's complement subtract extraneous checksum */
7085                         if (adj >= sum_val)
7086                                 sum_val = ~(adj - sum_val) & 0xFFFF;
7087                         else
7088                                 sum_val -= adj;
7089                 }
7090         } else {
7091                 sum_val = 0;
7092                 sum_flags = 0;
7093         }
7094 
7095         /* Clear hardware checksumming flag */
7096         DB_CKSUMFLAGS(mp) = 0;
7097 
7098         ident = ipha->ipha_ident;
7099         offset = (frag_offset_flags << 3) & 0xFFFF;
7100         src = ipha->ipha_src;
7101         dst = ipha->ipha_dst;
7102         hdr_length = IPH_HDR_LENGTH(ipha);
7103         end = ntohs(ipha->ipha_length) - hdr_length;
7104 
7105         /* If end == 0 then we have a packet with no data, so just free it */
7106         if (end == 0) {
7107                 freemsg(mp);
7108                 return (NULL);
7109         }
7110 
7111         /* Record the ECN field info. */
7112         ecn_info = (ipha->ipha_type_of_service & 0x3);
7113         if (offset != 0) {
7114                 /*
7115                  * If this isn't the first piece, strip the header, and
7116                  * add the offset to the end value.
7117                  */
7118                 mp->b_rptr += hdr_length;
7119                 end += offset;
7120         }
7121 
7122         /* Handle vnic loopback of fragments */
7123         if (mp->b_datap->db_ref > 2)
7124                 msg_len = 0;
7125         else
7126                 msg_len = MBLKSIZE(mp);
7127 
7128         tail_mp = mp;
7129         while (tail_mp->b_cont != NULL) {
7130                 tail_mp = tail_mp->b_cont;
7131                 if (tail_mp->b_datap->db_ref <= 2)
7132                         msg_len += MBLKSIZE(tail_mp);
7133         }
7134 
7135         /* If the reassembly list for this ILL will get too big, prune it */
7136         if ((msg_len + sizeof (*ipf) + ill->ill_frag_count) >=
7137             ipst->ips_ip_reass_queue_bytes) {
7138                 DTRACE_PROBE3(ip_reass_queue_bytes, uint_t, msg_len,
7139                     uint_t, ill->ill_frag_count,
7140                     uint_t, ipst->ips_ip_reass_queue_bytes);
7141                 ill_frag_prune(ill,
7142                     (ipst->ips_ip_reass_queue_bytes < msg_len) ? 0 :
7143                     (ipst->ips_ip_reass_queue_bytes - msg_len));
7144                 pruned = B_TRUE;
7145         }
7146 
7147         ipfb = &ill->ill_frag_hash_tbl[ILL_FRAG_HASH(src, ident)];
7148         mutex_enter(&ipfb->ipfb_lock);
7149 
7150         ipfp = &ipfb->ipfb_ipf;
7151         /* Try to find an existing fragment queue for this packet. */
7152         for (;;) {
7153                 ipf = ipfp[0];
7154                 if (ipf != NULL) {
7155                         /*
7156                          * It has to match on ident and src/dst address.
7157                          */
7158                         if (ipf->ipf_ident == ident &&
7159                             ipf->ipf_src == src &&
7160                             ipf->ipf_dst == dst &&
7161                             ipf->ipf_protocol == proto) {
7162                                 /*
7163                                  * If we have received too many
7164                                  * duplicate fragments for this packet
7165                                  * free it.
7166                                  */
7167                                 if (ipf->ipf_num_dups > ip_max_frag_dups) {
7168                                         ill_frag_free_pkts(ill, ipfb, ipf, 1);
7169                                         freemsg(mp);
7170                                         mutex_exit(&ipfb->ipfb_lock);
7171                                         return (NULL);
7172                                 }
7173                                 /* Found it. */
7174                                 break;
7175                         }
7176                         ipfp = &ipf->ipf_hash_next;
7177                         continue;
7178                 }
7179 
7180                 /*
7181                  * If we pruned the list, do we want to store this new
7182                  * fragment?. We apply an optimization here based on the
7183                  * fact that most fragments will be received in order.
7184                  * So if the offset of this incoming fragment is zero,
7185                  * it is the first fragment of a new packet. We will
7186                  * keep it.  Otherwise drop the fragment, as we have
7187                  * probably pruned the packet already (since the
7188                  * packet cannot be found).
7189                  */
7190                 if (pruned && offset != 0) {
7191                         mutex_exit(&ipfb->ipfb_lock);
7192                         freemsg(mp);
7193                         return (NULL);
7194                 }
7195 
7196                 if (ipfb->ipfb_frag_pkts >= MAX_FRAG_PKTS(ipst))  {
7197                         /*
7198                          * Too many fragmented packets in this hash
7199                          * bucket. Free the oldest.
7200                          */
7201                         ill_frag_free_pkts(ill, ipfb, ipfb->ipfb_ipf, 1);
7202                 }
7203 
7204                 /* New guy.  Allocate a frag message. */
7205                 mp1 = allocb(sizeof (*ipf), BPRI_MED);
7206                 if (mp1 == NULL) {
7207                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
7208                         ip_drop_input("ipIfStatsInDiscards", mp, ill);
7209                         freemsg(mp);
7210 reass_done:
7211                         mutex_exit(&ipfb->ipfb_lock);
7212                         return (NULL);
7213                 }
7214 
7215                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsReasmReqds);
7216                 mp1->b_cont = mp;
7217 
7218                 /* Initialize the fragment header. */
7219                 ipf = (ipf_t *)mp1->b_rptr;
7220                 ipf->ipf_mp = mp1;
7221                 ipf->ipf_ptphn = ipfp;
7222                 ipfp[0] = ipf;
7223                 ipf->ipf_hash_next = NULL;
7224                 ipf->ipf_ident = ident;
7225                 ipf->ipf_protocol = proto;
7226                 ipf->ipf_src = src;
7227                 ipf->ipf_dst = dst;
7228                 ipf->ipf_nf_hdr_len = 0;
7229                 /* Record reassembly start time. */
7230                 ipf->ipf_timestamp = gethrestime_sec();
7231                 /* Record ipf generation and account for frag header */
7232                 ipf->ipf_gen = ill->ill_ipf_gen++;
7233                 ipf->ipf_count = MBLKSIZE(mp1);
7234                 ipf->ipf_last_frag_seen = B_FALSE;
7235                 ipf->ipf_ecn = ecn_info;
7236                 ipf->ipf_num_dups = 0;
7237                 ipfb->ipfb_frag_pkts++;
7238                 ipf->ipf_checksum = 0;
7239                 ipf->ipf_checksum_flags = 0;
7240 
7241                 /* Store checksum value in fragment header */
7242                 if (sum_flags != 0) {
7243                         sum_val = (sum_val & 0xFFFF) + (sum_val >> 16);
7244                         sum_val = (sum_val & 0xFFFF) + (sum_val >> 16);
7245                         ipf->ipf_checksum = sum_val;
7246                         ipf->ipf_checksum_flags = sum_flags;
7247                 }
7248 
7249                 /*
7250                  * We handle reassembly two ways.  In the easy case,
7251                  * where all the fragments show up in order, we do
7252                  * minimal bookkeeping, and just clip new pieces on
7253                  * the end.  If we ever see a hole, then we go off
7254                  * to ip_reassemble which has to mark the pieces and
7255                  * keep track of the number of holes, etc.  Obviously,
7256                  * the point of having both mechanisms is so we can
7257                  * handle the easy case as efficiently as possible.
7258                  */
7259                 if (offset == 0) {
7260                         /* Easy case, in-order reassembly so far. */
7261                         ipf->ipf_count += msg_len;
7262                         ipf->ipf_tail_mp = tail_mp;
7263                         /*
7264                          * Keep track of next expected offset in
7265                          * ipf_end.
7266                          */
7267                         ipf->ipf_end = end;
7268                         ipf->ipf_nf_hdr_len = hdr_length;
7269                 } else {
7270                         /* Hard case, hole at the beginning. */
7271                         ipf->ipf_tail_mp = NULL;
7272                         /*
7273                          * ipf_end == 0 means that we have given up
7274                          * on easy reassembly.
7275                          */
7276                         ipf->ipf_end = 0;
7277 
7278                         /* Forget checksum offload from now on */
7279                         ipf->ipf_checksum_flags = 0;
7280 
7281                         /*
7282                          * ipf_hole_cnt is set by ip_reassemble.
7283                          * ipf_count is updated by ip_reassemble.
7284                          * No need to check for return value here
7285                          * as we don't expect reassembly to complete
7286                          * or fail for the first fragment itself.
7287                          */
7288                         (void) ip_reassemble(mp, ipf,
7289                             (frag_offset_flags & IPH_OFFSET) << 3,
7290                             (frag_offset_flags & IPH_MF), ill, msg_len);
7291                 }
7292                 /* Update per ipfb and ill byte counts */
7293                 ipfb->ipfb_count += ipf->ipf_count;
7294                 ASSERT(ipfb->ipfb_count > 0);     /* Wraparound */
7295                 atomic_add_32(&ill->ill_frag_count, ipf->ipf_count);
7296                 /* If the frag timer wasn't already going, start it. */
7297                 mutex_enter(&ill->ill_lock);
7298                 ill_frag_timer_start(ill);
7299                 mutex_exit(&ill->ill_lock);
7300                 goto reass_done;
7301         }
7302 
7303         /*
7304          * If the packet's flag has changed (it could be coming up
7305          * from an interface different than the previous, therefore
7306          * possibly different checksum capability), then forget about
7307          * any stored checksum states.  Otherwise add the value to
7308          * the existing one stored in the fragment header.
7309          */
7310         if (sum_flags != 0 && sum_flags == ipf->ipf_checksum_flags) {
7311                 sum_val += ipf->ipf_checksum;
7312                 sum_val = (sum_val & 0xFFFF) + (sum_val >> 16);
7313                 sum_val = (sum_val & 0xFFFF) + (sum_val >> 16);
7314                 ipf->ipf_checksum = sum_val;
7315         } else if (ipf->ipf_checksum_flags != 0) {
7316                 /* Forget checksum offload from now on */
7317                 ipf->ipf_checksum_flags = 0;
7318         }
7319 
7320         /*
7321          * We have a new piece of a datagram which is already being
7322          * reassembled.  Update the ECN info if all IP fragments
7323          * are ECN capable.  If there is one which is not, clear
7324          * all the info.  If there is at least one which has CE
7325          * code point, IP needs to report that up to transport.
7326          */
7327         if (ecn_info != IPH_ECN_NECT && ipf->ipf_ecn != IPH_ECN_NECT) {
7328                 if (ecn_info == IPH_ECN_CE)
7329                         ipf->ipf_ecn = IPH_ECN_CE;
7330         } else {
7331                 ipf->ipf_ecn = IPH_ECN_NECT;
7332         }
7333         if (offset && ipf->ipf_end == offset) {
7334                 /* The new fragment fits at the end */
7335                 ipf->ipf_tail_mp->b_cont = mp;
7336                 /* Update the byte count */
7337                 ipf->ipf_count += msg_len;
7338                 /* Update per ipfb and ill byte counts */
7339                 ipfb->ipfb_count += msg_len;
7340                 ASSERT(ipfb->ipfb_count > 0);     /* Wraparound */
7341                 atomic_add_32(&ill->ill_frag_count, msg_len);
7342                 if (frag_offset_flags & IPH_MF) {
7343                         /* More to come. */
7344                         ipf->ipf_end = end;
7345                         ipf->ipf_tail_mp = tail_mp;
7346                         goto reass_done;
7347                 }
7348         } else {
7349                 /* Go do the hard cases. */
7350                 int ret;
7351 
7352                 if (offset == 0)
7353                         ipf->ipf_nf_hdr_len = hdr_length;
7354 
7355                 /* Save current byte count */
7356                 count = ipf->ipf_count;
7357                 ret = ip_reassemble(mp, ipf,
7358                     (frag_offset_flags & IPH_OFFSET) << 3,
7359                     (frag_offset_flags & IPH_MF), ill, msg_len);
7360                 /* Count of bytes added and subtracted (freeb()ed) */
7361                 count = ipf->ipf_count - count;
7362                 if (count) {
7363                         /* Update per ipfb and ill byte counts */
7364                         ipfb->ipfb_count += count;
7365                         ASSERT(ipfb->ipfb_count > 0); /* Wraparound */
7366                         atomic_add_32(&ill->ill_frag_count, count);
7367                 }
7368                 if (ret == IP_REASS_PARTIAL) {
7369                         goto reass_done;
7370                 } else if (ret == IP_REASS_FAILED) {
7371                         /* Reassembly failed. Free up all resources */
7372                         ill_frag_free_pkts(ill, ipfb, ipf, 1);
7373                         for (t_mp = mp; t_mp != NULL; t_mp = t_mp->b_cont) {
7374                                 IP_REASS_SET_START(t_mp, 0);
7375                                 IP_REASS_SET_END(t_mp, 0);
7376                         }
7377                         freemsg(mp);
7378                         goto reass_done;
7379                 }
7380                 /* We will reach here iff 'ret' is IP_REASS_COMPLETE */
7381         }
7382         /*
7383          * We have completed reassembly.  Unhook the frag header from
7384          * the reassembly list.
7385          *
7386          * Before we free the frag header, record the ECN info
7387          * to report back to the transport.
7388          */
7389         ecn_info = ipf->ipf_ecn;
7390         BUMP_MIB(ill->ill_ip_mib, ipIfStatsReasmOKs);
7391         ipfp = ipf->ipf_ptphn;
7392 
7393         /* We need to supply these to caller */
7394         if ((sum_flags = ipf->ipf_checksum_flags) != 0)
7395                 sum_val = ipf->ipf_checksum;
7396         else
7397                 sum_val = 0;
7398 
7399         mp1 = ipf->ipf_mp;
7400         count = ipf->ipf_count;
7401         ipf = ipf->ipf_hash_next;
7402         if (ipf != NULL)
7403                 ipf->ipf_ptphn = ipfp;
7404         ipfp[0] = ipf;
7405         atomic_add_32(&ill->ill_frag_count, -count);
7406         ASSERT(ipfb->ipfb_count >= count);
7407         ipfb->ipfb_count -= count;
7408         ipfb->ipfb_frag_pkts--;
7409         mutex_exit(&ipfb->ipfb_lock);
7410         /* Ditch the frag header. */
7411         mp = mp1->b_cont;
7412 
7413         freeb(mp1);
7414 
7415         /* Restore original IP length in header. */
7416         packet_size = (uint32_t)msgdsize(mp);
7417         if (packet_size > IP_MAXPACKET) {
7418                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7419                 ip_drop_input("Reassembled packet too large", mp, ill);
7420                 freemsg(mp);
7421                 return (NULL);
7422         }
7423 
7424         if (DB_REF(mp) > 1) {
7425                 mblk_t *mp2 = copymsg(mp);
7426 
7427                 if (mp2 == NULL) {
7428                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
7429                         ip_drop_input("ipIfStatsInDiscards", mp, ill);
7430                         freemsg(mp);
7431                         return (NULL);
7432                 }
7433                 freemsg(mp);
7434                 mp = mp2;
7435         }
7436         ipha = (ipha_t *)mp->b_rptr;
7437 
7438         ipha->ipha_length = htons((uint16_t)packet_size);
7439         /* We're now complete, zip the frag state */
7440         ipha->ipha_fragment_offset_and_flags = 0;
7441         /* Record the ECN info. */
7442         ipha->ipha_type_of_service &= 0xFC;
7443         ipha->ipha_type_of_service |= ecn_info;
7444 
7445         /* Update the receive attributes */
7446         ira->ira_pktlen = packet_size;
7447         ira->ira_ip_hdr_length = IPH_HDR_LENGTH(ipha);
7448 
7449         /* Reassembly is successful; set checksum information in packet */
7450         DB_CKSUM16(mp) = (uint16_t)sum_val;
7451         DB_CKSUMFLAGS(mp) = sum_flags;
7452         DB_CKSUMSTART(mp) = ira->ira_ip_hdr_length;
7453 
7454         return (mp);
7455 }
7456 
7457 /*
7458  * Pullup function that should be used for IP input in order to
7459  * ensure we do not loose the L2 source address; we need the l2 source
7460  * address for IP_RECVSLLA and for ndp_input.
7461  *
7462  * We return either NULL or b_rptr.
7463  */
7464 void *
7465 ip_pullup(mblk_t *mp, ssize_t len, ip_recv_attr_t *ira)
7466 {
7467         ill_t           *ill = ira->ira_ill;
7468 
7469         if (ip_rput_pullups++ == 0) {
7470                 (void) mi_strlog(ill->ill_rq, 1, SL_ERROR|SL_TRACE,
7471                     "ip_pullup: %s forced us to "
7472                     " pullup pkt, hdr len %ld, hdr addr %p",
7473                     ill->ill_name, len, (void *)mp->b_rptr);
7474         }
7475         if (!(ira->ira_flags & IRAF_L2SRC_SET))
7476                 ip_setl2src(mp, ira, ira->ira_rill);
7477         ASSERT(ira->ira_flags & IRAF_L2SRC_SET);
7478         if (!pullupmsg(mp, len))
7479                 return (NULL);
7480         else
7481                 return (mp->b_rptr);
7482 }
7483 
7484 /*
7485  * Make sure ira_l2src has an address. If we don't have one fill with zeros.
7486  * When called from the ULP ira_rill will be NULL hence the caller has to
7487  * pass in the ill.
7488  */
7489 /* ARGSUSED */
7490 void
7491 ip_setl2src(mblk_t *mp, ip_recv_attr_t *ira, ill_t *ill)
7492 {
7493         const uchar_t *addr;
7494         int alen;
7495 
7496         if (ira->ira_flags & IRAF_L2SRC_SET)
7497                 return;
7498 
7499         ASSERT(ill != NULL);
7500         alen = ill->ill_phys_addr_length;
7501         ASSERT(alen <= sizeof (ira->ira_l2src));
7502         if (ira->ira_mhip != NULL &&
7503             (addr = ira->ira_mhip->mhi_saddr) != NULL) {
7504                 bcopy(addr, ira->ira_l2src, alen);
7505         } else if ((ira->ira_flags & IRAF_L2SRC_LOOPBACK) &&
7506             (addr = ill->ill_phys_addr) != NULL) {
7507                 bcopy(addr, ira->ira_l2src, alen);
7508         } else {
7509                 bzero(ira->ira_l2src, alen);
7510         }
7511         ira->ira_flags |= IRAF_L2SRC_SET;
7512 }
7513 
7514 /*
7515  * check ip header length and align it.
7516  */
7517 mblk_t *
7518 ip_check_and_align_header(mblk_t *mp, uint_t min_size, ip_recv_attr_t *ira)
7519 {
7520         ill_t   *ill = ira->ira_ill;
7521         ssize_t len;
7522 
7523         len = MBLKL(mp);
7524 
7525         if (!OK_32PTR(mp->b_rptr))
7526                 IP_STAT(ill->ill_ipst, ip_notaligned);
7527         else
7528                 IP_STAT(ill->ill_ipst, ip_recv_pullup);
7529 
7530         /* Guard against bogus device drivers */
7531         if (len < 0) {
7532                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7533                 ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7534                 freemsg(mp);
7535                 return (NULL);
7536         }
7537 
7538         if (len == 0) {
7539                 /* GLD sometimes sends up mblk with b_rptr == b_wptr! */
7540                 mblk_t *mp1 = mp->b_cont;
7541 
7542                 if (!(ira->ira_flags & IRAF_L2SRC_SET))
7543                         ip_setl2src(mp, ira, ira->ira_rill);
7544                 ASSERT(ira->ira_flags & IRAF_L2SRC_SET);
7545 
7546                 freeb(mp);
7547                 mp = mp1;
7548                 if (mp == NULL)
7549                         return (NULL);
7550 
7551                 if (OK_32PTR(mp->b_rptr) && MBLKL(mp) >= min_size)
7552                         return (mp);
7553         }
7554         if (ip_pullup(mp, min_size, ira) == NULL) {
7555                 if (msgdsize(mp) < min_size) {
7556                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7557                         ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7558                 } else {
7559                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
7560                         ip_drop_input("ipIfStatsInDiscards", mp, ill);
7561                 }
7562                 freemsg(mp);
7563                 return (NULL);
7564         }
7565         return (mp);
7566 }
7567 
7568 /*
7569  * Common code for IPv4 and IPv6 to check and pullup multi-mblks
7570  */
7571 mblk_t *
7572 ip_check_length(mblk_t *mp, uchar_t *rptr, ssize_t len, uint_t pkt_len,
7573     uint_t min_size, ip_recv_attr_t *ira)
7574 {
7575         ill_t   *ill = ira->ira_ill;
7576 
7577         /*
7578          * Make sure we have data length consistent
7579          * with the IP header.
7580          */
7581         if (mp->b_cont == NULL) {
7582                 /* pkt_len is based on ipha_len, not the mblk length */
7583                 if (pkt_len < min_size) {
7584                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7585                         ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7586                         freemsg(mp);
7587                         return (NULL);
7588                 }
7589                 if (len < 0) {
7590                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
7591                         ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
7592                         freemsg(mp);
7593                         return (NULL);
7594                 }
7595                 /* Drop any pad */
7596                 mp->b_wptr = rptr + pkt_len;
7597         } else if ((len += msgdsize(mp->b_cont)) != 0) {
7598                 ASSERT(pkt_len >= min_size);
7599                 if (pkt_len < min_size) {
7600                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7601                         ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7602                         freemsg(mp);
7603                         return (NULL);
7604                 }
7605                 if (len < 0) {
7606                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
7607                         ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
7608                         freemsg(mp);
7609                         return (NULL);
7610                 }
7611                 /* Drop any pad */
7612                 (void) adjmsg(mp, -len);
7613                 /*
7614                  * adjmsg may have freed an mblk from the chain, hence
7615                  * invalidate any hw checksum here. This will force IP to
7616                  * calculate the checksum in sw, but only for this packet.
7617                  */
7618                 DB_CKSUMFLAGS(mp) = 0;
7619                 IP_STAT(ill->ill_ipst, ip_multimblk);
7620         }
7621         return (mp);
7622 }
7623 
7624 /*
7625  * Check that the IPv4 opt_len is consistent with the packet and pullup
7626  * the options.
7627  */
7628 mblk_t *
7629 ip_check_optlen(mblk_t *mp, ipha_t *ipha, uint_t opt_len, uint_t pkt_len,
7630     ip_recv_attr_t *ira)
7631 {
7632         ill_t   *ill = ira->ira_ill;
7633         ssize_t len;
7634 
7635         /* Assume no IPv6 packets arrive over the IPv4 queue */
7636         if (IPH_HDR_VERSION(ipha) != IPV4_VERSION) {
7637                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7638                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInWrongIPVersion);
7639                 ip_drop_input("IPvN packet on IPv4 ill", mp, ill);
7640                 freemsg(mp);
7641                 return (NULL);
7642         }
7643 
7644         if (opt_len > (15 - IP_SIMPLE_HDR_LENGTH_IN_WORDS)) {
7645                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7646                 ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7647                 freemsg(mp);
7648                 return (NULL);
7649         }
7650         /*
7651          * Recompute complete header length and make sure we
7652          * have access to all of it.
7653          */
7654         len = ((size_t)opt_len + IP_SIMPLE_HDR_LENGTH_IN_WORDS) << 2;
7655         if (len > (mp->b_wptr - mp->b_rptr)) {
7656                 if (len > pkt_len) {
7657                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7658                         ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7659                         freemsg(mp);
7660                         return (NULL);
7661                 }
7662                 if (ip_pullup(mp, len, ira) == NULL) {
7663                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
7664                         ip_drop_input("ipIfStatsInDiscards", mp, ill);
7665                         freemsg(mp);
7666                         return (NULL);
7667                 }
7668         }
7669         return (mp);
7670 }
7671 
7672 /*
7673  * Returns a new ire, or the same ire, or NULL.
7674  * If a different IRE is returned, then it is held; the caller
7675  * needs to release it.
7676  * In no case is there any hold/release on the ire argument.
7677  */
7678 ire_t *
7679 ip_check_multihome(void *addr, ire_t *ire, ill_t *ill)
7680 {
7681         ire_t           *new_ire;
7682         ill_t           *ire_ill;
7683         uint_t          ifindex;
7684         ip_stack_t      *ipst = ill->ill_ipst;
7685         boolean_t       strict_check = B_FALSE;
7686 
7687         /*
7688          * IPMP common case: if IRE and ILL are in the same group, there's no
7689          * issue (e.g. packet received on an underlying interface matched an
7690          * IRE_LOCAL on its associated group interface).
7691          */
7692         ASSERT(ire->ire_ill != NULL);
7693         if (IS_IN_SAME_ILLGRP(ill, ire->ire_ill))
7694                 return (ire);
7695 
7696         /*
7697          * Do another ire lookup here, using the ingress ill, to see if the
7698          * interface is in a usesrc group.
7699          * As long as the ills belong to the same group, we don't consider
7700          * them to be arriving on the wrong interface. Thus, if the switch
7701          * is doing inbound load spreading, we won't drop packets when the
7702          * ip*_strict_dst_multihoming switch is on.
7703          * We also need to check for IPIF_UNNUMBERED point2point interfaces
7704          * where the local address may not be unique. In this case we were
7705          * at the mercy of the initial ire lookup and the IRE_LOCAL it
7706          * actually returned. The new lookup, which is more specific, should
7707          * only find the IRE_LOCAL associated with the ingress ill if one
7708          * exists.
7709          */
7710         if (ire->ire_ipversion == IPV4_VERSION) {
7711                 if (ipst->ips_ip_strict_dst_multihoming)
7712                         strict_check = B_TRUE;
7713                 new_ire = ire_ftable_lookup_v4(*((ipaddr_t *)addr), 0, 0,
7714                     IRE_LOCAL, ill, ALL_ZONES, NULL,
7715                     (MATCH_IRE_TYPE|MATCH_IRE_ILL), 0, ipst, NULL);
7716         } else {
7717                 ASSERT(!IN6_IS_ADDR_MULTICAST((in6_addr_t *)addr));
7718                 if (ipst->ips_ipv6_strict_dst_multihoming)
7719                         strict_check = B_TRUE;
7720                 new_ire = ire_ftable_lookup_v6((in6_addr_t *)addr, NULL, NULL,
7721                     IRE_LOCAL, ill, ALL_ZONES, NULL,
7722                     (MATCH_IRE_TYPE|MATCH_IRE_ILL), 0, ipst, NULL);
7723         }
7724         /*
7725          * If the same ire that was returned in ip_input() is found then this
7726          * is an indication that usesrc groups are in use. The packet
7727          * arrived on a different ill in the group than the one associated with
7728          * the destination address.  If a different ire was found then the same
7729          * IP address must be hosted on multiple ills. This is possible with
7730          * unnumbered point2point interfaces. We switch to use this new ire in
7731          * order to have accurate interface statistics.
7732          */
7733         if (new_ire != NULL) {
7734                 /* Note: held in one case but not the other? Caller handles */
7735                 if (new_ire != ire)
7736                         return (new_ire);
7737                 /* Unchanged */
7738                 ire_refrele(new_ire);
7739                 return (ire);
7740         }
7741 
7742         /*
7743          * Chase pointers once and store locally.
7744          */
7745         ASSERT(ire->ire_ill != NULL);
7746         ire_ill = ire->ire_ill;
7747         ifindex = ill->ill_usesrc_ifindex;
7748 
7749         /*
7750          * Check if it's a legal address on the 'usesrc' interface.
7751          * For IPMP data addresses the IRE_LOCAL is the upper, hence we
7752          * can just check phyint_ifindex.
7753          */
7754         if (ifindex != 0 && ifindex == ire_ill->ill_phyint->phyint_ifindex) {
7755                 return (ire);
7756         }
7757 
7758         /*
7759          * If the ip*_strict_dst_multihoming switch is on then we can
7760          * only accept this packet if the interface is marked as routing.
7761          */
7762         if (!(strict_check))
7763                 return (ire);
7764 
7765         if ((ill->ill_flags & ire->ire_ill->ill_flags & ILLF_ROUTER) != 0) {
7766                 return (ire);
7767         }
7768         return (NULL);
7769 }
7770 
7771 /*
7772  * This function is used to construct a mac_header_info_s from a
7773  * DL_UNITDATA_IND message.
7774  * The address fields in the mhi structure points into the message,
7775  * thus the caller can't use those fields after freeing the message.
7776  *
7777  * We determine whether the packet received is a non-unicast packet
7778  * and in doing so, determine whether or not it is broadcast vs multicast.
7779  * For it to be a broadcast packet, we must have the appropriate mblk_t
7780  * hanging off the ill_t.  If this is either not present or doesn't match
7781  * the destination mac address in the DL_UNITDATA_IND, the packet is deemed
7782  * to be multicast.  Thus NICs that have no broadcast address (or no
7783  * capability for one, such as point to point links) cannot return as
7784  * the packet being broadcast.
7785  */
7786 void
7787 ip_dlur_to_mhi(ill_t *ill, mblk_t *mb, struct mac_header_info_s *mhip)
7788 {
7789         dl_unitdata_ind_t *ind = (dl_unitdata_ind_t *)mb->b_rptr;
7790         mblk_t *bmp;
7791         uint_t extra_offset;
7792 
7793         bzero(mhip, sizeof (struct mac_header_info_s));
7794 
7795         mhip->mhi_dsttype = MAC_ADDRTYPE_UNICAST;
7796 
7797         if (ill->ill_sap_length < 0)
7798                 extra_offset = 0;
7799         else
7800                 extra_offset = ill->ill_sap_length;
7801 
7802         mhip->mhi_daddr = (uchar_t *)ind + ind->dl_dest_addr_offset +
7803             extra_offset;
7804         mhip->mhi_saddr = (uchar_t *)ind + ind->dl_src_addr_offset +
7805             extra_offset;
7806 
7807         if (!ind->dl_group_address)
7808                 return;
7809 
7810         /* Multicast or broadcast */
7811         mhip->mhi_dsttype = MAC_ADDRTYPE_MULTICAST;
7812 
7813         if (ind->dl_dest_addr_offset > sizeof (*ind) &&
7814             ind->dl_dest_addr_offset + ind->dl_dest_addr_length < MBLKL(mb) &&
7815             (bmp = ill->ill_bcast_mp) != NULL) {
7816                 dl_unitdata_req_t *dlur;
7817                 uint8_t *bphys_addr;
7818 
7819                 dlur = (dl_unitdata_req_t *)bmp->b_rptr;
7820                 bphys_addr = (uchar_t *)dlur + dlur->dl_dest_addr_offset +
7821                     extra_offset;
7822 
7823                 if (bcmp(mhip->mhi_daddr, bphys_addr,
7824                     ind->dl_dest_addr_length) == 0)
7825                         mhip->mhi_dsttype = MAC_ADDRTYPE_BROADCAST;
7826         }
7827 }
7828 
7829 /*
7830  * This function is used to construct a mac_header_info_s from a
7831  * M_DATA fastpath message from a DLPI driver.
7832  * The address fields in the mhi structure points into the message,
7833  * thus the caller can't use those fields after freeing the message.
7834  *
7835  * We determine whether the packet received is a non-unicast packet
7836  * and in doing so, determine whether or not it is broadcast vs multicast.
7837  * For it to be a broadcast packet, we must have the appropriate mblk_t
7838  * hanging off the ill_t.  If this is either not present or doesn't match
7839  * the destination mac address in the DL_UNITDATA_IND, the packet is deemed
7840  * to be multicast.  Thus NICs that have no broadcast address (or no
7841  * capability for one, such as point to point links) cannot return as
7842  * the packet being broadcast.
7843  */
7844 void
7845 ip_mdata_to_mhi(ill_t *ill, mblk_t *mp, struct mac_header_info_s *mhip)
7846 {
7847         mblk_t *bmp;
7848         struct ether_header *pether;
7849 
7850         bzero(mhip, sizeof (struct mac_header_info_s));
7851 
7852         mhip->mhi_dsttype = MAC_ADDRTYPE_UNICAST;
7853 
7854         pether = (struct ether_header *)((char *)mp->b_rptr
7855             - sizeof (struct ether_header));
7856 
7857         /*
7858          * Make sure the interface is an ethernet type, since we don't
7859          * know the header format for anything but Ethernet. Also make
7860          * sure we are pointing correctly above db_base.
7861          */
7862         if (ill->ill_type != IFT_ETHER)
7863                 return;
7864 
7865 retry:
7866         if ((uchar_t *)pether < mp->b_datap->db_base)
7867                 return;
7868 
7869         /* Is there a VLAN tag? */
7870         if (ill->ill_isv6) {
7871                 if (pether->ether_type != htons(ETHERTYPE_IPV6)) {
7872                         pether = (struct ether_header *)((char *)pether - 4);
7873                         goto retry;
7874                 }
7875         } else {
7876                 if (pether->ether_type != htons(ETHERTYPE_IP)) {
7877                         pether = (struct ether_header *)((char *)pether - 4);
7878                         goto retry;
7879                 }
7880         }
7881         mhip->mhi_daddr = (uchar_t *)&pether->ether_dhost;
7882         mhip->mhi_saddr = (uchar_t *)&pether->ether_shost;
7883 
7884         if (!(mhip->mhi_daddr[0] & 0x01))
7885                 return;
7886 
7887         /* Multicast or broadcast */
7888         mhip->mhi_dsttype = MAC_ADDRTYPE_MULTICAST;
7889 
7890         if ((bmp = ill->ill_bcast_mp) != NULL) {
7891                 dl_unitdata_req_t *dlur;
7892                 uint8_t *bphys_addr;
7893                 uint_t  addrlen;
7894 
7895                 dlur = (dl_unitdata_req_t *)bmp->b_rptr;
7896                 addrlen = dlur->dl_dest_addr_length;
7897                 if (ill->ill_sap_length < 0) {
7898                         bphys_addr = (uchar_t *)dlur +
7899                             dlur->dl_dest_addr_offset;
7900                         addrlen += ill->ill_sap_length;
7901                 } else {
7902                         bphys_addr = (uchar_t *)dlur +
7903                             dlur->dl_dest_addr_offset +
7904                             ill->ill_sap_length;
7905                         addrlen -= ill->ill_sap_length;
7906                 }
7907                 if (bcmp(mhip->mhi_daddr, bphys_addr, addrlen) == 0)
7908                         mhip->mhi_dsttype = MAC_ADDRTYPE_BROADCAST;
7909         }
7910 }
7911 
7912 /*
7913  * Handle anything but M_DATA messages
7914  * We see the DL_UNITDATA_IND which are part
7915  * of the data path, and also the other messages from the driver.
7916  */
7917 void
7918 ip_rput_notdata(ill_t *ill, mblk_t *mp)
7919 {
7920         mblk_t          *first_mp;
7921         struct iocblk   *iocp;
7922         struct mac_header_info_s mhi;
7923 
7924         switch (DB_TYPE(mp)) {
7925         case M_PROTO:
7926         case M_PCPROTO: {
7927                 if (((dl_unitdata_ind_t *)mp->b_rptr)->dl_primitive !=
7928                     DL_UNITDATA_IND) {
7929                         /* Go handle anything other than data elsewhere. */
7930                         ip_rput_dlpi(ill, mp);
7931                         return;
7932                 }
7933 
7934                 first_mp = mp;
7935                 mp = first_mp->b_cont;
7936                 first_mp->b_cont = NULL;
7937 
7938                 if (mp == NULL) {
7939                         freeb(first_mp);
7940                         return;
7941                 }
7942                 ip_dlur_to_mhi(ill, first_mp, &mhi);
7943                 if (ill->ill_isv6)
7944                         ip_input_v6(ill, NULL, mp, &mhi);
7945                 else
7946                         ip_input(ill, NULL, mp, &mhi);
7947 
7948                 /* Ditch the DLPI header. */
7949                 freeb(first_mp);
7950                 return;
7951         }
7952         case M_IOCACK:
7953                 iocp = (struct iocblk *)mp->b_rptr;
7954                 switch (iocp->ioc_cmd) {
7955                 case DL_IOC_HDR_INFO:
7956                         ill_fastpath_ack(ill, mp);
7957                         return;
7958                 default:
7959                         putnext(ill->ill_rq, mp);
7960                         return;
7961                 }
7962                 /* FALLTHROUGH */
7963         case M_ERROR:
7964         case M_HANGUP:
7965                 mutex_enter(&ill->ill_lock);
7966                 if (ill->ill_state_flags & ILL_CONDEMNED) {
7967                         mutex_exit(&ill->ill_lock);
7968                         freemsg(mp);
7969                         return;
7970                 }
7971                 ill_refhold_locked(ill);
7972                 mutex_exit(&ill->ill_lock);
7973                 qwriter_ip(ill, ill->ill_rq, mp, ip_rput_other, CUR_OP,
7974                     B_FALSE);
7975                 return;
7976         case M_CTL:
7977                 putnext(ill->ill_rq, mp);
7978                 return;
7979         case M_IOCNAK:
7980                 ip1dbg(("got iocnak "));
7981                 iocp = (struct iocblk *)mp->b_rptr;
7982                 switch (iocp->ioc_cmd) {
7983                 case DL_IOC_HDR_INFO:
7984                         ip_rput_other(NULL, ill->ill_rq, mp, NULL);
7985                         return;
7986                 default:
7987                         break;
7988                 }
7989                 /* FALLTHROUGH */
7990         default:
7991                 putnext(ill->ill_rq, mp);
7992                 return;
7993         }
7994 }
7995 
7996 /* Read side put procedure.  Packets coming from the wire arrive here. */
7997 int
7998 ip_rput(queue_t *q, mblk_t *mp)
7999 {
8000         ill_t   *ill;
8001         union DL_primitives *dl;
8002 
8003         ill = (ill_t *)q->q_ptr;
8004 
8005         if (ill->ill_state_flags & (ILL_CONDEMNED | ILL_LL_SUBNET_PENDING)) {
8006                 /*
8007                  * If things are opening or closing, only accept high-priority
8008                  * DLPI messages.  (On open ill->ill_ipif has not yet been
8009                  * created; on close, things hanging off the ill may have been
8010                  * freed already.)
8011                  */
8012                 dl = (union DL_primitives *)mp->b_rptr;
8013                 if (DB_TYPE(mp) != M_PCPROTO ||
8014                     dl->dl_primitive == DL_UNITDATA_IND) {
8015                         inet_freemsg(mp);
8016                         return (0);
8017                 }
8018         }
8019         if (DB_TYPE(mp) == M_DATA) {
8020                 struct mac_header_info_s mhi;
8021 
8022                 ip_mdata_to_mhi(ill, mp, &mhi);
8023                 ip_input(ill, NULL, mp, &mhi);
8024         } else {
8025                 ip_rput_notdata(ill, mp);
8026         }
8027         return (0);
8028 }
8029 
8030 /*
8031  * Move the information to a copy.
8032  */
8033 mblk_t *
8034 ip_fix_dbref(mblk_t *mp, ip_recv_attr_t *ira)
8035 {
8036         mblk_t          *mp1;
8037         ill_t           *ill = ira->ira_ill;
8038         ip_stack_t      *ipst = ill->ill_ipst;
8039 
8040         IP_STAT(ipst, ip_db_ref);
8041 
8042         /* Make sure we have ira_l2src before we loose the original mblk */
8043         if (!(ira->ira_flags & IRAF_L2SRC_SET))
8044                 ip_setl2src(mp, ira, ira->ira_rill);
8045 
8046         mp1 = copymsg(mp);
8047         if (mp1 == NULL) {
8048                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
8049                 ip_drop_input("ipIfStatsInDiscards", mp, ill);
8050                 freemsg(mp);
8051                 return (NULL);
8052         }
8053         /* preserve the hardware checksum flags and data, if present */
8054         if (DB_CKSUMFLAGS(mp) != 0) {
8055                 DB_CKSUMFLAGS(mp1) = DB_CKSUMFLAGS(mp);
8056                 DB_CKSUMSTART(mp1) = DB_CKSUMSTART(mp);
8057                 DB_CKSUMSTUFF(mp1) = DB_CKSUMSTUFF(mp);
8058                 DB_CKSUMEND(mp1) = DB_CKSUMEND(mp);
8059                 DB_CKSUM16(mp1) = DB_CKSUM16(mp);
8060         }
8061         freemsg(mp);
8062         return (mp1);
8063 }
8064 
8065 static void
8066 ip_dlpi_error(ill_t *ill, t_uscalar_t prim, t_uscalar_t dl_err,
8067     t_uscalar_t err)
8068 {
8069         if (dl_err == DL_SYSERR) {
8070                 (void) mi_strlog(ill->ill_rq, 1, SL_CONSOLE|SL_ERROR|SL_TRACE,
8071                     "%s: %s failed: DL_SYSERR (errno %u)\n",
8072                     ill->ill_name, dl_primstr(prim), err);
8073                 return;
8074         }
8075 
8076         (void) mi_strlog(ill->ill_rq, 1, SL_CONSOLE|SL_ERROR|SL_TRACE,
8077             "%s: %s failed: %s\n", ill->ill_name, dl_primstr(prim),
8078             dl_errstr(dl_err));
8079 }
8080 
8081 /*
8082  * ip_rput_dlpi is called by ip_rput to handle all DLPI messages other
8083  * than DL_UNITDATA_IND messages. If we need to process this message
8084  * exclusively, we call qwriter_ip, in which case we also need to call
8085  * ill_refhold before that, since qwriter_ip does an ill_refrele.
8086  */
8087 void
8088 ip_rput_dlpi(ill_t *ill, mblk_t *mp)
8089 {
8090         dl_ok_ack_t     *dloa = (dl_ok_ack_t *)mp->b_rptr;
8091         dl_error_ack_t  *dlea = (dl_error_ack_t *)dloa;
8092         queue_t         *q = ill->ill_rq;
8093         t_uscalar_t     prim = dloa->dl_primitive;
8094         t_uscalar_t     reqprim = DL_PRIM_INVAL;
8095 
8096         DTRACE_PROBE3(ill__dlpi, char *, "ip_rput_dlpi",
8097             char *, dl_primstr(prim), ill_t *, ill);
8098         ip1dbg(("ip_rput_dlpi"));
8099 
8100         /*
8101          * If we received an ACK but didn't send a request for it, then it
8102          * can't be part of any pending operation; discard up-front.
8103          */
8104         switch (prim) {
8105         case DL_ERROR_ACK:
8106                 reqprim = dlea->dl_error_primitive;
8107                 ip2dbg(("ip_rput_dlpi(%s): DL_ERROR_ACK for %s (0x%x): %s "
8108                     "(0x%x), unix %u\n", ill->ill_name, dl_primstr(reqprim),
8109                     reqprim, dl_errstr(dlea->dl_errno), dlea->dl_errno,
8110                     dlea->dl_unix_errno));
8111                 break;
8112         case DL_OK_ACK:
8113                 reqprim = dloa->dl_correct_primitive;
8114                 break;
8115         case DL_INFO_ACK:
8116                 reqprim = DL_INFO_REQ;
8117                 break;
8118         case DL_BIND_ACK:
8119                 reqprim = DL_BIND_REQ;
8120                 break;
8121         case DL_PHYS_ADDR_ACK:
8122                 reqprim = DL_PHYS_ADDR_REQ;
8123                 break;
8124         case DL_NOTIFY_ACK:
8125                 reqprim = DL_NOTIFY_REQ;
8126                 break;
8127         case DL_CAPABILITY_ACK:
8128                 reqprim = DL_CAPABILITY_REQ;
8129                 break;
8130         }
8131 
8132         if (prim != DL_NOTIFY_IND) {
8133                 if (reqprim == DL_PRIM_INVAL ||
8134                     !ill_dlpi_pending(ill, reqprim)) {
8135                         /* Not a DLPI message we support or expected */
8136                         freemsg(mp);
8137                         return;
8138                 }
8139                 ip1dbg(("ip_rput: received %s for %s\n", dl_primstr(prim),
8140                     dl_primstr(reqprim)));
8141         }
8142 
8143         switch (reqprim) {
8144         case DL_UNBIND_REQ:
8145                 /*
8146                  * NOTE: we mark the unbind as complete even if we got a
8147                  * DL_ERROR_ACK, since there's not much else we can do.
8148                  */
8149                 mutex_enter(&ill->ill_lock);
8150                 ill->ill_state_flags &= ~ILL_DL_UNBIND_IN_PROGRESS;
8151                 cv_signal(&ill->ill_cv);
8152                 mutex_exit(&ill->ill_lock);
8153                 break;
8154 
8155         case DL_ENABMULTI_REQ:
8156                 if (prim == DL_OK_ACK) {
8157                         if (ill->ill_dlpi_multicast_state == IDS_INPROGRESS)
8158                                 ill->ill_dlpi_multicast_state = IDS_OK;
8159                 }
8160                 break;
8161         }
8162 
8163         /*
8164          * The message is one we're waiting for (or DL_NOTIFY_IND), but we
8165          * need to become writer to continue to process it.  Because an
8166          * exclusive operation doesn't complete until replies to all queued
8167          * DLPI messages have been received, we know we're in the middle of an
8168          * exclusive operation and pass CUR_OP (except for DL_NOTIFY_IND).
8169          *
8170          * As required by qwriter_ip(), we refhold the ill; it will refrele.
8171          * Since this is on the ill stream we unconditionally bump up the
8172          * refcount without doing ILL_CAN_LOOKUP().
8173          */
8174         ill_refhold(ill);
8175         if (prim == DL_NOTIFY_IND)
8176                 qwriter_ip(ill, q, mp, ip_rput_dlpi_writer, NEW_OP, B_FALSE);
8177         else
8178                 qwriter_ip(ill, q, mp, ip_rput_dlpi_writer, CUR_OP, B_FALSE);
8179 }
8180 
8181 /*
8182  * Handling of DLPI messages that require exclusive access to the ipsq.
8183  *
8184  * Need to do ipsq_pending_mp_get on ioctl completion, which could
8185  * happen here. (along with mi_copy_done)
8186  */
8187 /* ARGSUSED */
8188 static void
8189 ip_rput_dlpi_writer(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *dummy_arg)
8190 {
8191         dl_ok_ack_t     *dloa = (dl_ok_ack_t *)mp->b_rptr;
8192         dl_error_ack_t  *dlea = (dl_error_ack_t *)dloa;
8193         int             err = 0;
8194         ill_t           *ill = (ill_t *)q->q_ptr;
8195         ipif_t          *ipif = NULL;
8196         mblk_t          *mp1 = NULL;
8197         conn_t          *connp = NULL;
8198         t_uscalar_t     paddrreq;
8199         mblk_t          *mp_hw;
8200         boolean_t       success;
8201         boolean_t       ioctl_aborted = B_FALSE;
8202         boolean_t       log = B_TRUE;
8203 
8204         DTRACE_PROBE3(ill__dlpi, char *, "ip_rput_dlpi_writer",
8205             char *, dl_primstr(dloa->dl_primitive), ill_t *, ill);
8206 
8207         ip1dbg(("ip_rput_dlpi_writer .."));
8208         ASSERT(ipsq->ipsq_xop == ill->ill_phyint->phyint_ipsq->ipsq_xop);
8209         ASSERT(IAM_WRITER_ILL(ill));
8210 
8211         ipif = ipsq->ipsq_xop->ipx_pending_ipif;
8212         /*
8213          * The current ioctl could have been aborted by the user and a new
8214          * ioctl to bring up another ill could have started. We could still
8215          * get a response from the driver later.
8216          */
8217         if (ipif != NULL && ipif->ipif_ill != ill)
8218                 ioctl_aborted = B_TRUE;
8219 
8220         switch (dloa->dl_primitive) {
8221         case DL_ERROR_ACK:
8222                 ip1dbg(("ip_rput_dlpi_writer: got DL_ERROR_ACK for %s\n",
8223                     dl_primstr(dlea->dl_error_primitive)));
8224 
8225                 DTRACE_PROBE3(ill__dlpi, char *, "ip_rput_dlpi_writer error",
8226                     char *, dl_primstr(dlea->dl_error_primitive),
8227                     ill_t *, ill);
8228 
8229                 switch (dlea->dl_error_primitive) {
8230                 case DL_DISABMULTI_REQ:
8231                         ill_dlpi_done(ill, dlea->dl_error_primitive);
8232                         break;
8233                 case DL_PROMISCON_REQ:
8234                 case DL_PROMISCOFF_REQ:
8235                 case DL_UNBIND_REQ:
8236                 case DL_ATTACH_REQ:
8237                 case DL_INFO_REQ:
8238                         ill_dlpi_done(ill, dlea->dl_error_primitive);
8239                         break;
8240                 case DL_NOTIFY_REQ:
8241                         ill_dlpi_done(ill, DL_NOTIFY_REQ);
8242                         log = B_FALSE;
8243                         break;
8244                 case DL_PHYS_ADDR_REQ:
8245                         /*
8246                          * For IPv6 only, there are two additional
8247                          * phys_addr_req's sent to the driver to get the
8248                          * IPv6 token and lla. This allows IP to acquire
8249                          * the hardware address format for a given interface
8250                          * without having built in knowledge of the hardware
8251                          * address. ill_phys_addr_pend keeps track of the last
8252                          * DL_PAR sent so we know which response we are
8253                          * dealing with. ill_dlpi_done will update
8254                          * ill_phys_addr_pend when it sends the next req.
8255                          * We don't complete the IOCTL until all three DL_PARs
8256                          * have been attempted, so set *_len to 0 and break.
8257                          */
8258                         paddrreq = ill->ill_phys_addr_pend;
8259                         ill_dlpi_done(ill, DL_PHYS_ADDR_REQ);
8260                         if (paddrreq == DL_IPV6_TOKEN) {
8261                                 ill->ill_token_length = 0;
8262                                 log = B_FALSE;
8263                                 break;
8264                         } else if (paddrreq == DL_IPV6_LINK_LAYER_ADDR) {
8265                                 ill->ill_nd_lla_len = 0;
8266                                 log = B_FALSE;
8267                                 break;
8268                         }
8269                         /*
8270                          * Something went wrong with the DL_PHYS_ADDR_REQ.
8271                          * We presumably have an IOCTL hanging out waiting
8272                          * for completion. Find it and complete the IOCTL
8273                          * with the error noted.
8274                          * However, ill_dl_phys was called on an ill queue
8275                          * (from SIOCSLIFNAME), thus conn_pending_ill is not
8276                          * set. But the ioctl is known to be pending on ill_wq.
8277                          */
8278                         if (!ill->ill_ifname_pending)
8279                                 break;
8280                         ill->ill_ifname_pending = 0;
8281                         if (!ioctl_aborted)
8282                                 mp1 = ipsq_pending_mp_get(ipsq, &connp);
8283                         if (mp1 != NULL) {
8284                                 /*
8285                                  * This operation (SIOCSLIFNAME) must have
8286                                  * happened on the ill. Assert there is no conn
8287                                  */
8288                                 ASSERT(connp == NULL);
8289                                 q = ill->ill_wq;
8290                         }
8291                         break;
8292                 case DL_BIND_REQ:
8293                         ill_dlpi_done(ill, DL_BIND_REQ);
8294                         if (ill->ill_ifname_pending)
8295                                 break;
8296                         mutex_enter(&ill->ill_lock);
8297                         ill->ill_state_flags &= ~ILL_DOWN_IN_PROGRESS;
8298                         mutex_exit(&ill->ill_lock);
8299                         /*
8300                          * Something went wrong with the bind.  We presumably
8301                          * have an IOCTL hanging out waiting for completion.
8302                          * Find it, take down the interface that was coming
8303                          * up, and complete the IOCTL with the error noted.
8304                          */
8305                         if (!ioctl_aborted)
8306                                 mp1 = ipsq_pending_mp_get(ipsq, &connp);
8307                         if (mp1 != NULL) {
8308                                 /*
8309                                  * This might be a result of a DL_NOTE_REPLUMB
8310                                  * notification. In that case, connp is NULL.
8311                                  */
8312                                 if (connp != NULL)
8313                                         q = CONNP_TO_WQ(connp);
8314 
8315                                 (void) ipif_down(ipif, NULL, NULL);
8316                                 /* error is set below the switch */
8317                         }
8318                         break;
8319                 case DL_ENABMULTI_REQ:
8320                         ill_dlpi_done(ill, DL_ENABMULTI_REQ);
8321 
8322                         if (ill->ill_dlpi_multicast_state == IDS_INPROGRESS)
8323                                 ill->ill_dlpi_multicast_state = IDS_FAILED;
8324                         if (ill->ill_dlpi_multicast_state == IDS_FAILED) {
8325 
8326                                 printf("ip: joining multicasts failed (%d)"
8327                                     " on %s - will use link layer "
8328                                     "broadcasts for multicast\n",
8329                                     dlea->dl_errno, ill->ill_name);
8330 
8331                                 /*
8332                                  * Set up for multi_bcast; We are the
8333                                  * writer, so ok to access ill->ill_ipif
8334                                  * without any lock.
8335                                  */
8336                                 mutex_enter(&ill->ill_phyint->phyint_lock);
8337                                 ill->ill_phyint->phyint_flags |=
8338                                     PHYI_MULTI_BCAST;
8339                                 mutex_exit(&ill->ill_phyint->phyint_lock);
8340 
8341                         }
8342                         freemsg(mp);    /* Don't want to pass this up */
8343                         return;
8344                 case DL_CAPABILITY_REQ:
8345                         ip1dbg(("ip_rput_dlpi_writer: got DL_ERROR_ACK for "
8346                             "DL_CAPABILITY REQ\n"));
8347                         if (ill->ill_dlpi_capab_state == IDCS_PROBE_SENT)
8348                                 ill->ill_dlpi_capab_state = IDCS_FAILED;
8349                         ill_capability_done(ill);
8350                         freemsg(mp);
8351                         return;
8352                 }
8353                 /*
8354                  * Note the error for IOCTL completion (mp1 is set when
8355                  * ready to complete ioctl). If ill_ifname_pending_err is
8356                  * set, an error occured during plumbing (ill_ifname_pending),
8357                  * so we want to report that error.
8358                  *
8359                  * NOTE: there are two addtional DL_PHYS_ADDR_REQ's
8360                  * (DL_IPV6_TOKEN and DL_IPV6_LINK_LAYER_ADDR) that are
8361                  * expected to get errack'd if the driver doesn't support
8362                  * these flags (e.g. ethernet). log will be set to B_FALSE
8363                  * if these error conditions are encountered.
8364                  */
8365                 if (mp1 != NULL) {
8366                         if (ill->ill_ifname_pending_err != 0)  {
8367                                 err = ill->ill_ifname_pending_err;
8368                                 ill->ill_ifname_pending_err = 0;
8369                         } else {
8370                                 err = dlea->dl_unix_errno ?
8371                                     dlea->dl_unix_errno : ENXIO;
8372                         }
8373                 /*
8374                  * If we're plumbing an interface and an error hasn't already
8375                  * been saved, set ill_ifname_pending_err to the error passed
8376                  * up. Ignore the error if log is B_FALSE (see comment above).
8377                  */
8378                 } else if (log && ill->ill_ifname_pending &&
8379                     ill->ill_ifname_pending_err == 0) {
8380                         ill->ill_ifname_pending_err = dlea->dl_unix_errno ?
8381                             dlea->dl_unix_errno : ENXIO;
8382                 }
8383 
8384                 if (log)
8385                         ip_dlpi_error(ill, dlea->dl_error_primitive,
8386                             dlea->dl_errno, dlea->dl_unix_errno);
8387                 break;
8388         case DL_CAPABILITY_ACK:
8389                 ill_capability_ack(ill, mp);
8390                 /*
8391                  * The message has been handed off to ill_capability_ack
8392                  * and must not be freed below
8393                  */
8394                 mp = NULL;
8395                 break;
8396 
8397         case DL_INFO_ACK:
8398                 /* Call a routine to handle this one. */
8399                 ill_dlpi_done(ill, DL_INFO_REQ);
8400                 ip_ll_subnet_defaults(ill, mp);
8401                 ASSERT(!MUTEX_HELD(&ill->ill_phyint->phyint_ipsq->ipsq_lock));
8402                 return;
8403         case DL_BIND_ACK:
8404                 /*
8405                  * We should have an IOCTL waiting on this unless
8406                  * sent by ill_dl_phys, in which case just return
8407                  */
8408                 ill_dlpi_done(ill, DL_BIND_REQ);
8409 
8410                 if (ill->ill_ifname_pending) {
8411                         DTRACE_PROBE2(ip__rput__dlpi__ifname__pending,
8412                             ill_t *, ill, mblk_t *, mp);
8413                         break;
8414                 }
8415                 mutex_enter(&ill->ill_lock);
8416                 ill->ill_dl_up = 1;
8417                 ill->ill_state_flags &= ~ILL_DOWN_IN_PROGRESS;
8418                 mutex_exit(&ill->ill_lock);
8419 
8420                 if (!ioctl_aborted)
8421                         mp1 = ipsq_pending_mp_get(ipsq, &connp);
8422                 if (mp1 == NULL) {
8423                         DTRACE_PROBE1(ip__rput__dlpi__no__mblk, ill_t *, ill);
8424                         break;
8425                 }
8426                 /*
8427                  * mp1 was added by ill_dl_up(). if that is a result of
8428                  * a DL_NOTE_REPLUMB notification, connp could be NULL.
8429                  */
8430                 if (connp != NULL)
8431                         q = CONNP_TO_WQ(connp);
8432                 /*
8433                  * We are exclusive. So nothing can change even after
8434                  * we get the pending mp.
8435                  */
8436                 ip1dbg(("ip_rput_dlpi: bind_ack %s\n", ill->ill_name));
8437                 DTRACE_PROBE1(ip__rput__dlpi__bind__ack, ill_t *, ill);
8438                 ill_nic_event_dispatch(ill, 0, NE_UP, NULL, 0);
8439 
8440                 /*
8441                  * Now bring up the resolver; when that is complete, we'll
8442                  * create IREs.  Note that we intentionally mirror what
8443                  * ipif_up() would have done, because we got here by way of
8444                  * ill_dl_up(), which stopped ipif_up()'s processing.
8445                  */
8446                 if (ill->ill_isv6) {
8447                         /*
8448                          * v6 interfaces.
8449                          * Unlike ARP which has to do another bind
8450                          * and attach, once we get here we are
8451                          * done with NDP
8452                          */
8453                         (void) ipif_resolver_up(ipif, Res_act_initial);
8454                         if ((err = ipif_ndp_up(ipif, B_TRUE)) == 0)
8455                                 err = ipif_up_done_v6(ipif);
8456                 } else if (ill->ill_net_type == IRE_IF_RESOLVER) {
8457                         /*
8458                          * ARP and other v4 external resolvers.
8459                          * Leave the pending mblk intact so that
8460                          * the ioctl completes in ip_rput().
8461                          */
8462                         if (connp != NULL)
8463                                 mutex_enter(&connp->conn_lock);
8464                         mutex_enter(&ill->ill_lock);
8465                         success = ipsq_pending_mp_add(connp, ipif, q, mp1, 0);
8466                         mutex_exit(&ill->ill_lock);
8467                         if (connp != NULL)
8468                                 mutex_exit(&connp->conn_lock);
8469                         if (success) {
8470                                 err = ipif_resolver_up(ipif, Res_act_initial);
8471                                 if (err == EINPROGRESS) {
8472                                         freemsg(mp);
8473                                         return;
8474                                 }
8475                                 mp1 = ipsq_pending_mp_get(ipsq, &connp);
8476                         } else {
8477                                 /* The conn has started closing */
8478                                 err = EINTR;
8479                         }
8480                 } else {
8481                         /*
8482                          * This one is complete. Reply to pending ioctl.
8483                          */
8484                         (void) ipif_resolver_up(ipif, Res_act_initial);
8485                         err = ipif_up_done(ipif);
8486                 }
8487 
8488                 if ((err == 0) && (ill->ill_up_ipifs)) {
8489                         err = ill_up_ipifs(ill, q, mp1);
8490                         if (err == EINPROGRESS) {
8491                                 freemsg(mp);
8492                                 return;
8493                         }
8494                 }
8495 
8496                 /*
8497                  * If we have a moved ipif to bring up, and everything has
8498                  * succeeded to this point, bring it up on the IPMP ill.
8499                  * Otherwise, leave it down -- the admin can try to bring it
8500                  * up by hand if need be.
8501                  */
8502                 if (ill->ill_move_ipif != NULL) {
8503                         if (err != 0) {
8504                                 ill->ill_move_ipif = NULL;
8505                         } else {
8506                                 ipif = ill->ill_move_ipif;
8507                                 ill->ill_move_ipif = NULL;
8508                                 err = ipif_up(ipif, q, mp1);
8509                                 if (err == EINPROGRESS) {
8510                                         freemsg(mp);
8511                                         return;
8512                                 }
8513                         }
8514                 }
8515                 break;
8516 
8517         case DL_NOTIFY_IND: {
8518                 dl_notify_ind_t *notify = (dl_notify_ind_t *)mp->b_rptr;
8519                 uint_t orig_mtu, orig_mc_mtu;
8520 
8521                 switch (notify->dl_notification) {
8522                 case DL_NOTE_PHYS_ADDR:
8523                         err = ill_set_phys_addr(ill, mp);
8524                         break;
8525 
8526                 case DL_NOTE_REPLUMB:
8527                         /*
8528                          * Directly return after calling ill_replumb().
8529                          * Note that we should not free mp as it is reused
8530                          * in the ill_replumb() function.
8531                          */
8532                         err = ill_replumb(ill, mp);
8533                         return;
8534 
8535                 case DL_NOTE_FASTPATH_FLUSH:
8536                         nce_flush(ill, B_FALSE);
8537                         break;
8538 
8539                 case DL_NOTE_SDU_SIZE:
8540                 case DL_NOTE_SDU_SIZE2:
8541                         /*
8542                          * The dce and fragmentation code can cope with
8543                          * this changing while packets are being sent.
8544                          * When packets are sent ip_output will discover
8545                          * a change.
8546                          *
8547                          * Change the MTU size of the interface.
8548                          */
8549                         mutex_enter(&ill->ill_lock);
8550                         orig_mtu = ill->ill_mtu;
8551                         orig_mc_mtu = ill->ill_mc_mtu;
8552                         switch (notify->dl_notification) {
8553                         case DL_NOTE_SDU_SIZE:
8554                                 ill->ill_current_frag =
8555                                     (uint_t)notify->dl_data;
8556                                 ill->ill_mc_mtu = (uint_t)notify->dl_data;
8557                                 break;
8558                         case DL_NOTE_SDU_SIZE2:
8559                                 ill->ill_current_frag =
8560                                     (uint_t)notify->dl_data1;
8561                                 ill->ill_mc_mtu = (uint_t)notify->dl_data2;
8562                                 break;
8563                         }
8564                         if (ill->ill_current_frag > ill->ill_max_frag)
8565                                 ill->ill_max_frag = ill->ill_current_frag;
8566 
8567                         if (!(ill->ill_flags & ILLF_FIXEDMTU)) {
8568                                 ill->ill_mtu = ill->ill_current_frag;
8569 
8570                                 /*
8571                                  * If ill_user_mtu was set (via
8572                                  * SIOCSLIFLNKINFO), clamp ill_mtu at it.
8573                                  */
8574                                 if (ill->ill_user_mtu != 0 &&
8575                                     ill->ill_user_mtu < ill->ill_mtu)
8576                                         ill->ill_mtu = ill->ill_user_mtu;
8577 
8578                                 if (ill->ill_user_mtu != 0 &&
8579                                     ill->ill_user_mtu < ill->ill_mc_mtu)
8580                                         ill->ill_mc_mtu = ill->ill_user_mtu;
8581 
8582                                 if (ill->ill_isv6) {
8583                                         if (ill->ill_mtu < IPV6_MIN_MTU)
8584                                                 ill->ill_mtu = IPV6_MIN_MTU;
8585                                         if (ill->ill_mc_mtu < IPV6_MIN_MTU)
8586                                                 ill->ill_mc_mtu = IPV6_MIN_MTU;
8587                                 } else {
8588                                         if (ill->ill_mtu < IP_MIN_MTU)
8589                                                 ill->ill_mtu = IP_MIN_MTU;
8590                                         if (ill->ill_mc_mtu < IP_MIN_MTU)
8591                                                 ill->ill_mc_mtu = IP_MIN_MTU;
8592                                 }
8593                         } else if (ill->ill_mc_mtu > ill->ill_mtu) {
8594                                 ill->ill_mc_mtu = ill->ill_mtu;
8595                         }
8596 
8597                         mutex_exit(&ill->ill_lock);
8598                         /*
8599                          * Make sure all dce_generation checks find out
8600                          * that ill_mtu/ill_mc_mtu has changed.
8601                          */
8602                         if (orig_mtu != ill->ill_mtu ||
8603                             orig_mc_mtu != ill->ill_mc_mtu) {
8604                                 dce_increment_all_generations(ill->ill_isv6,
8605                                     ill->ill_ipst);
8606                         }
8607 
8608                         /*
8609                          * Refresh IPMP meta-interface MTU if necessary.
8610                          */
8611                         if (IS_UNDER_IPMP(ill))
8612                                 ipmp_illgrp_refresh_mtu(ill->ill_grp);
8613                         break;
8614 
8615                 case DL_NOTE_LINK_UP:
8616                 case DL_NOTE_LINK_DOWN: {
8617                         /*
8618                          * We are writer. ill / phyint / ipsq assocs stable.
8619                          * The RUNNING flag reflects the state of the link.
8620                          */
8621                         phyint_t *phyint = ill->ill_phyint;
8622                         uint64_t new_phyint_flags;
8623                         boolean_t changed = B_FALSE;
8624                         boolean_t went_up;
8625 
8626                         went_up = notify->dl_notification == DL_NOTE_LINK_UP;
8627                         mutex_enter(&phyint->phyint_lock);
8628 
8629                         new_phyint_flags = went_up ?
8630                             phyint->phyint_flags | PHYI_RUNNING :
8631                             phyint->phyint_flags & ~PHYI_RUNNING;
8632 
8633                         if (IS_IPMP(ill)) {
8634                                 new_phyint_flags = went_up ?
8635                                     new_phyint_flags & ~PHYI_FAILED :
8636                                     new_phyint_flags | PHYI_FAILED;
8637                         }
8638 
8639                         if (new_phyint_flags != phyint->phyint_flags) {
8640                                 phyint->phyint_flags = new_phyint_flags;
8641                                 changed = B_TRUE;
8642                         }
8643                         mutex_exit(&phyint->phyint_lock);
8644                         /*
8645                          * ill_restart_dad handles the DAD restart and routing
8646                          * socket notification logic.
8647                          */
8648                         if (changed) {
8649                                 ill_restart_dad(phyint->phyint_illv4, went_up);
8650                                 ill_restart_dad(phyint->phyint_illv6, went_up);
8651                         }
8652                         break;
8653                 }
8654                 case DL_NOTE_PROMISC_ON_PHYS: {
8655                         phyint_t *phyint = ill->ill_phyint;
8656 
8657                         mutex_enter(&phyint->phyint_lock);
8658                         phyint->phyint_flags |= PHYI_PROMISC;
8659                         mutex_exit(&phyint->phyint_lock);
8660                         break;
8661                 }
8662                 case DL_NOTE_PROMISC_OFF_PHYS: {
8663                         phyint_t *phyint = ill->ill_phyint;
8664 
8665                         mutex_enter(&phyint->phyint_lock);
8666                         phyint->phyint_flags &= ~PHYI_PROMISC;
8667                         mutex_exit(&phyint->phyint_lock);
8668                         break;
8669                 }
8670                 case DL_NOTE_CAPAB_RENEG:
8671                         /*
8672                          * Something changed on the driver side.
8673                          * It wants us to renegotiate the capabilities
8674                          * on this ill. One possible cause is the aggregation
8675                          * interface under us where a port got added or
8676                          * went away.
8677                          *
8678                          * If the capability negotiation is already done
8679                          * or is in progress, reset the capabilities and
8680                          * mark the ill's ill_capab_reneg to be B_TRUE,
8681                          * so that when the ack comes back, we can start
8682                          * the renegotiation process.
8683                          *
8684                          * Note that if ill_capab_reneg is already B_TRUE
8685                          * (ill_dlpi_capab_state is IDS_UNKNOWN in this case),
8686                          * the capability resetting request has been sent
8687                          * and the renegotiation has not been started yet;
8688                          * nothing needs to be done in this case.
8689                          */
8690                         ipsq_current_start(ipsq, ill->ill_ipif, 0);
8691                         ill_capability_reset(ill, B_TRUE);
8692                         ipsq_current_finish(ipsq);
8693                         break;
8694 
8695                 case DL_NOTE_ALLOWED_IPS:
8696                         ill_set_allowed_ips(ill, mp);
8697                         break;
8698                 default:
8699                         ip0dbg(("ip_rput_dlpi_writer: unknown notification "
8700                             "type 0x%x for DL_NOTIFY_IND\n",
8701                             notify->dl_notification));
8702                         break;
8703                 }
8704 
8705                 /*
8706                  * As this is an asynchronous operation, we
8707                  * should not call ill_dlpi_done
8708                  */
8709                 break;
8710         }
8711         case DL_NOTIFY_ACK: {
8712                 dl_notify_ack_t *noteack = (dl_notify_ack_t *)mp->b_rptr;
8713 
8714                 if (noteack->dl_notifications & DL_NOTE_LINK_UP)
8715                         ill->ill_note_link = 1;
8716                 ill_dlpi_done(ill, DL_NOTIFY_REQ);
8717                 break;
8718         }
8719         case DL_PHYS_ADDR_ACK: {
8720                 /*
8721                  * As part of plumbing the interface via SIOCSLIFNAME,
8722                  * ill_dl_phys() will queue a series of DL_PHYS_ADDR_REQs,
8723                  * whose answers we receive here.  As each answer is received,
8724                  * we call ill_dlpi_done() to dispatch the next request as
8725                  * we're processing the current one.  Once all answers have
8726                  * been received, we use ipsq_pending_mp_get() to dequeue the
8727                  * outstanding IOCTL and reply to it.  (Because ill_dl_phys()
8728                  * is invoked from an ill queue, conn_oper_pending_ill is not
8729                  * available, but we know the ioctl is pending on ill_wq.)
8730                  */
8731                 uint_t  paddrlen, paddroff;
8732                 uint8_t *addr;
8733 
8734                 paddrreq = ill->ill_phys_addr_pend;
8735                 paddrlen = ((dl_phys_addr_ack_t *)mp->b_rptr)->dl_addr_length;
8736                 paddroff = ((dl_phys_addr_ack_t *)mp->b_rptr)->dl_addr_offset;
8737                 addr = mp->b_rptr + paddroff;
8738 
8739                 ill_dlpi_done(ill, DL_PHYS_ADDR_REQ);
8740                 if (paddrreq == DL_IPV6_TOKEN) {
8741                         /*
8742                          * bcopy to low-order bits of ill_token
8743                          *
8744                          * XXX Temporary hack - currently, all known tokens
8745                          * are 64 bits, so I'll cheat for the moment.
8746                          */
8747                         bcopy(addr, &ill->ill_token.s6_addr32[2], paddrlen);
8748                         ill->ill_token_length = paddrlen;
8749                         break;
8750                 } else if (paddrreq == DL_IPV6_LINK_LAYER_ADDR) {
8751                         ASSERT(ill->ill_nd_lla_mp == NULL);
8752                         ill_set_ndmp(ill, mp, paddroff, paddrlen);
8753                         mp = NULL;
8754                         break;
8755                 } else if (paddrreq == DL_CURR_DEST_ADDR) {
8756                         ASSERT(ill->ill_dest_addr_mp == NULL);
8757                         ill->ill_dest_addr_mp = mp;
8758                         ill->ill_dest_addr = addr;
8759                         mp = NULL;
8760                         if (ill->ill_isv6) {
8761                                 ill_setdesttoken(ill);
8762                                 ipif_setdestlinklocal(ill->ill_ipif);
8763                         }
8764                         break;
8765                 }
8766 
8767                 ASSERT(paddrreq == DL_CURR_PHYS_ADDR);
8768                 ASSERT(ill->ill_phys_addr_mp == NULL);
8769                 if (!ill->ill_ifname_pending)
8770                         break;
8771                 ill->ill_ifname_pending = 0;
8772                 if (!ioctl_aborted)
8773                         mp1 = ipsq_pending_mp_get(ipsq, &connp);
8774                 if (mp1 != NULL) {
8775                         ASSERT(connp == NULL);
8776                         q = ill->ill_wq;
8777                 }
8778                 /*
8779                  * If any error acks received during the plumbing sequence,
8780                  * ill_ifname_pending_err will be set. Break out and send up
8781                  * the error to the pending ioctl.
8782                  */
8783                 if (ill->ill_ifname_pending_err != 0) {
8784                         err = ill->ill_ifname_pending_err;
8785                         ill->ill_ifname_pending_err = 0;
8786                         break;
8787                 }
8788 
8789                 ill->ill_phys_addr_mp = mp;
8790                 ill->ill_phys_addr = (paddrlen == 0 ? NULL : addr);
8791                 mp = NULL;
8792 
8793                 /*
8794                  * If paddrlen or ill_phys_addr_length is zero, the DLPI
8795                  * provider doesn't support physical addresses.  We check both
8796                  * paddrlen and ill_phys_addr_length because sppp (PPP) does
8797                  * not have physical addresses, but historically adversises a
8798                  * physical address length of 0 in its DL_INFO_ACK, but 6 in
8799                  * its DL_PHYS_ADDR_ACK.
8800                  */
8801                 if (paddrlen == 0 || ill->ill_phys_addr_length == 0) {
8802                         ill->ill_phys_addr = NULL;
8803                 } else if (paddrlen != ill->ill_phys_addr_length) {
8804                         ip0dbg(("DL_PHYS_ADDR_ACK: got addrlen %d, expected %d",
8805                             paddrlen, ill->ill_phys_addr_length));
8806                         err = EINVAL;
8807                         break;
8808                 }
8809 
8810                 if (ill->ill_nd_lla_mp == NULL) {
8811                         if ((mp_hw = copyb(ill->ill_phys_addr_mp)) == NULL) {
8812                                 err = ENOMEM;
8813                                 break;
8814                         }
8815                         ill_set_ndmp(ill, mp_hw, paddroff, paddrlen);
8816                 }
8817 
8818                 if (ill->ill_isv6) {
8819                         ill_setdefaulttoken(ill);
8820                         ipif_setlinklocal(ill->ill_ipif);
8821                 }
8822                 break;
8823         }
8824         case DL_OK_ACK:
8825                 ip2dbg(("DL_OK_ACK %s (0x%x)\n",
8826                     dl_primstr((int)dloa->dl_correct_primitive),
8827                     dloa->dl_correct_primitive));
8828                 DTRACE_PROBE3(ill__dlpi, char *, "ip_rput_dlpi_writer ok",
8829                     char *, dl_primstr(dloa->dl_correct_primitive),
8830                     ill_t *, ill);
8831 
8832                 switch (dloa->dl_correct_primitive) {
8833                 case DL_ENABMULTI_REQ:
8834                 case DL_DISABMULTI_REQ:
8835                         ill_dlpi_done(ill, dloa->dl_correct_primitive);
8836                         break;
8837                 case DL_PROMISCON_REQ:
8838                 case DL_PROMISCOFF_REQ:
8839                 case DL_UNBIND_REQ:
8840                 case DL_ATTACH_REQ:
8841                         ill_dlpi_done(ill, dloa->dl_correct_primitive);
8842                         break;
8843                 }
8844                 break;
8845         default:
8846                 break;
8847         }
8848 
8849         freemsg(mp);
8850         if (mp1 == NULL)
8851                 return;
8852 
8853         /*
8854          * The operation must complete without EINPROGRESS since
8855          * ipsq_pending_mp_get() has removed the mblk (mp1).  Otherwise,
8856          * the operation will be stuck forever inside the IPSQ.
8857          */
8858         ASSERT(err != EINPROGRESS);
8859 
8860         DTRACE_PROBE4(ipif__ioctl, char *, "ip_rput_dlpi_writer finish",
8861             int, ipsq->ipsq_xop->ipx_current_ioctl, ill_t *, ill,
8862             ipif_t *, NULL);
8863 
8864         switch (ipsq->ipsq_xop->ipx_current_ioctl) {
8865         case 0:
8866                 ipsq_current_finish(ipsq);
8867                 break;
8868 
8869         case SIOCSLIFNAME:
8870         case IF_UNITSEL: {
8871                 ill_t *ill_other = ILL_OTHER(ill);
8872 
8873                 /*
8874                  * If SIOCSLIFNAME or IF_UNITSEL is about to succeed, and the
8875                  * ill has a peer which is in an IPMP group, then place ill
8876                  * into the same group.  One catch: although ifconfig plumbs
8877                  * the appropriate IPMP meta-interface prior to plumbing this
8878                  * ill, it is possible for multiple ifconfig applications to
8879                  * race (or for another application to adjust plumbing), in
8880                  * which case the IPMP meta-interface we need will be missing.
8881                  * If so, kick the phyint out of the group.
8882                  */
8883                 if (err == 0 && ill_other != NULL && IS_UNDER_IPMP(ill_other)) {
8884                         ipmp_grp_t      *grp = ill->ill_phyint->phyint_grp;
8885                         ipmp_illgrp_t   *illg;
8886 
8887                         illg = ill->ill_isv6 ? grp->gr_v6 : grp->gr_v4;
8888                         if (illg == NULL)
8889                                 ipmp_phyint_leave_grp(ill->ill_phyint);
8890                         else
8891                                 ipmp_ill_join_illgrp(ill, illg);
8892                 }
8893 
8894                 if (ipsq->ipsq_xop->ipx_current_ioctl == IF_UNITSEL)
8895                         ip_ioctl_finish(q, mp1, err, NO_COPYOUT, ipsq);
8896                 else
8897                         ip_ioctl_finish(q, mp1, err, COPYOUT, ipsq);
8898                 break;
8899         }
8900         case SIOCLIFADDIF:
8901                 ip_ioctl_finish(q, mp1, err, COPYOUT, ipsq);
8902                 break;
8903 
8904         default:
8905                 ip_ioctl_finish(q, mp1, err, NO_COPYOUT, ipsq);
8906                 break;
8907         }
8908 }
8909 
8910 /*
8911  * ip_rput_other is called by ip_rput to handle messages modifying the global
8912  * state in IP.  If 'ipsq' is non-NULL, caller is writer on it.
8913  */
8914 /* ARGSUSED */
8915 void
8916 ip_rput_other(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *dummy_arg)
8917 {
8918         ill_t           *ill = q->q_ptr;
8919         struct iocblk   *iocp;
8920 
8921         ip1dbg(("ip_rput_other "));
8922         if (ipsq != NULL) {
8923                 ASSERT(IAM_WRITER_IPSQ(ipsq));
8924                 ASSERT(ipsq->ipsq_xop ==
8925                     ill->ill_phyint->phyint_ipsq->ipsq_xop);
8926         }
8927 
8928         switch (mp->b_datap->db_type) {
8929         case M_ERROR:
8930         case M_HANGUP:
8931                 /*
8932                  * The device has a problem.  We force the ILL down.  It can
8933                  * be brought up again manually using SIOCSIFFLAGS (via
8934                  * ifconfig or equivalent).
8935                  */
8936                 ASSERT(ipsq != NULL);
8937                 if (mp->b_rptr < mp->b_wptr)
8938                         ill->ill_error = (int)(*mp->b_rptr & 0xFF);
8939                 if (ill->ill_error == 0)
8940                         ill->ill_error = ENXIO;
8941                 if (!ill_down_start(q, mp))
8942                         return;
8943                 ipif_all_down_tail(ipsq, q, mp, NULL);
8944                 break;
8945         case M_IOCNAK: {
8946                 iocp = (struct iocblk *)mp->b_rptr;
8947 
8948                 ASSERT(iocp->ioc_cmd == DL_IOC_HDR_INFO);
8949                 /*
8950                  * If this was the first attempt, turn off the fastpath
8951                  * probing.
8952                  */
8953                 mutex_enter(&ill->ill_lock);
8954                 if (ill->ill_dlpi_fastpath_state == IDS_INPROGRESS) {
8955                         ill->ill_dlpi_fastpath_state = IDS_FAILED;
8956                         mutex_exit(&ill->ill_lock);
8957                         /*
8958                          * don't flush the nce_t entries: we use them
8959                          * as an index to the ncec itself.
8960                          */
8961                         ip1dbg(("ip_rput: DLPI fastpath off on interface %s\n",
8962                             ill->ill_name));
8963                 } else {
8964                         mutex_exit(&ill->ill_lock);
8965                 }
8966                 freemsg(mp);
8967                 break;
8968         }
8969         default:
8970                 ASSERT(0);
8971                 break;
8972         }
8973 }
8974 
8975 /*
8976  * Update any source route, record route or timestamp options
8977  * When it fails it has consumed the message and BUMPed the MIB.
8978  */
8979 boolean_t
8980 ip_forward_options(mblk_t *mp, ipha_t *ipha, ill_t *dst_ill,
8981     ip_recv_attr_t *ira)
8982 {
8983         ipoptp_t        opts;
8984         uchar_t         *opt;
8985         uint8_t         optval;
8986         uint8_t         optlen;
8987         ipaddr_t        dst;
8988         ipaddr_t        ifaddr;
8989         uint32_t        ts;
8990         timestruc_t     now;
8991         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
8992 
8993         ip2dbg(("ip_forward_options\n"));
8994         dst = ipha->ipha_dst;
8995         for (optval = ipoptp_first(&opts, ipha);
8996             optval != IPOPT_EOL;
8997             optval = ipoptp_next(&opts)) {
8998                 ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
8999                 opt = opts.ipoptp_cur;
9000                 optlen = opts.ipoptp_len;
9001                 ip2dbg(("ip_forward_options: opt %d, len %d\n",
9002                     optval, opts.ipoptp_len));
9003                 switch (optval) {
9004                         uint32_t off;
9005                 case IPOPT_SSRR:
9006                 case IPOPT_LSRR:
9007                         /* Check if adminstratively disabled */
9008                         if (!ipst->ips_ip_forward_src_routed) {
9009                                 BUMP_MIB(dst_ill->ill_ip_mib,
9010                                     ipIfStatsForwProhibits);
9011                                 ip_drop_input("ICMP_SOURCE_ROUTE_FAILED",
9012                                     mp, dst_ill);
9013                                 icmp_unreachable(mp, ICMP_SOURCE_ROUTE_FAILED,
9014                                     ira);
9015                                 return (B_FALSE);
9016                         }
9017                         if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
9018                                 /*
9019                                  * Must be partial since ip_input_options
9020                                  * checked for strict.
9021                                  */
9022                                 break;
9023                         }
9024                         off = opt[IPOPT_OFFSET];
9025                         off--;
9026                 redo_srr:
9027                         if (optlen < IP_ADDR_LEN ||
9028                             off > optlen - IP_ADDR_LEN) {
9029                                 /* End of source route */
9030                                 ip1dbg((
9031                                     "ip_forward_options: end of SR\n"));
9032                                 break;
9033                         }
9034                         /* Pick a reasonable address on the outbound if */
9035                         ASSERT(dst_ill != NULL);
9036                         if (ip_select_source_v4(dst_ill, INADDR_ANY, dst,
9037                             INADDR_ANY, ALL_ZONES, ipst, &ifaddr, NULL,
9038                             NULL) != 0) {
9039                                 /* No source! Shouldn't happen */
9040                                 ifaddr = INADDR_ANY;
9041                         }
9042                         bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
9043                         bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9044                         ip1dbg(("ip_forward_options: next hop 0x%x\n",
9045                             ntohl(dst)));
9046 
9047                         /*
9048                          * Check if our address is present more than
9049                          * once as consecutive hops in source route.
9050                          */
9051                         if (ip_type_v4(dst, ipst) == IRE_LOCAL) {
9052                                 off += IP_ADDR_LEN;
9053                                 opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9054                                 goto redo_srr;
9055                         }
9056                         ipha->ipha_dst = dst;
9057                         opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9058                         break;
9059                 case IPOPT_RR:
9060                         off = opt[IPOPT_OFFSET];
9061                         off--;
9062                         if (optlen < IP_ADDR_LEN ||
9063                             off > optlen - IP_ADDR_LEN) {
9064                                 /* No more room - ignore */
9065                                 ip1dbg((
9066                                     "ip_forward_options: end of RR\n"));
9067                                 break;
9068                         }
9069                         /* Pick a reasonable address on the outbound if */
9070                         ASSERT(dst_ill != NULL);
9071                         if (ip_select_source_v4(dst_ill, INADDR_ANY, dst,
9072                             INADDR_ANY, ALL_ZONES, ipst, &ifaddr, NULL,
9073                             NULL) != 0) {
9074                                 /* No source! Shouldn't happen */
9075                                 ifaddr = INADDR_ANY;
9076                         }
9077                         bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9078                         opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9079                         break;
9080                 case IPOPT_TS:
9081                         /* Insert timestamp if there is room */
9082                         switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9083                         case IPOPT_TS_TSONLY:
9084                                 off = IPOPT_TS_TIMELEN;
9085                                 break;
9086                         case IPOPT_TS_PRESPEC:
9087                         case IPOPT_TS_PRESPEC_RFC791:
9088                                 /* Verify that the address matched */
9089                                 off = opt[IPOPT_OFFSET] - 1;
9090                                 bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
9091                                 if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
9092                                         /* Not for us */
9093                                         break;
9094                                 }
9095                                 /* FALLTHROUGH */
9096                         case IPOPT_TS_TSANDADDR:
9097                                 off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
9098                                 break;
9099                         default:
9100                                 /*
9101                                  * ip_*put_options should have already
9102                                  * dropped this packet.
9103                                  */
9104                                 cmn_err(CE_PANIC, "ip_forward_options: "
9105                                     "unknown IT - bug in ip_input_options?\n");
9106                                 return (B_TRUE);        /* Keep "lint" happy */
9107                         }
9108                         if (opt[IPOPT_OFFSET] - 1 + off > optlen) {
9109                                 /* Increase overflow counter */
9110                                 off = (opt[IPOPT_POS_OV_FLG] >> 4) + 1;
9111                                 opt[IPOPT_POS_OV_FLG] =
9112                                     (uint8_t)((opt[IPOPT_POS_OV_FLG] & 0x0F) |
9113                                     (off << 4));
9114                                 break;
9115                         }
9116                         off = opt[IPOPT_OFFSET] - 1;
9117                         switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9118                         case IPOPT_TS_PRESPEC:
9119                         case IPOPT_TS_PRESPEC_RFC791:
9120                         case IPOPT_TS_TSANDADDR:
9121                                 /* Pick a reasonable addr on the outbound if */
9122                                 ASSERT(dst_ill != NULL);
9123                                 if (ip_select_source_v4(dst_ill, INADDR_ANY,
9124                                     dst, INADDR_ANY, ALL_ZONES, ipst, &ifaddr,
9125                                     NULL, NULL) != 0) {
9126                                         /* No source! Shouldn't happen */
9127                                         ifaddr = INADDR_ANY;
9128                                 }
9129                                 bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9130                                 opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9131                                 /* FALLTHROUGH */
9132                         case IPOPT_TS_TSONLY:
9133                                 off = opt[IPOPT_OFFSET] - 1;
9134                                 /* Compute # of milliseconds since midnight */
9135                                 gethrestime(&now);
9136                                 ts = (now.tv_sec % (24 * 60 * 60)) * 1000 +
9137                                     NSEC2MSEC(now.tv_nsec);
9138                                 bcopy(&ts, (char *)opt + off, IPOPT_TS_TIMELEN);
9139                                 opt[IPOPT_OFFSET] += IPOPT_TS_TIMELEN;
9140                                 break;
9141                         }
9142                         break;
9143                 }
9144         }
9145         return (B_TRUE);
9146 }
9147 
9148 /*
9149  * Call ill_frag_timeout to do garbage collection. ill_frag_timeout
9150  * returns 'true' if there are still fragments left on the queue, in
9151  * which case we restart the timer.
9152  */
9153 void
9154 ill_frag_timer(void *arg)
9155 {
9156         ill_t   *ill = (ill_t *)arg;
9157         boolean_t frag_pending;
9158         ip_stack_t *ipst = ill->ill_ipst;
9159         time_t  timeout;
9160 
9161         mutex_enter(&ill->ill_lock);
9162         ASSERT(!ill->ill_fragtimer_executing);
9163         if (ill->ill_state_flags & ILL_CONDEMNED) {
9164                 ill->ill_frag_timer_id = 0;
9165                 mutex_exit(&ill->ill_lock);
9166                 return;
9167         }
9168         ill->ill_fragtimer_executing = 1;
9169         mutex_exit(&ill->ill_lock);
9170 
9171         timeout = (ill->ill_isv6 ? ipst->ips_ipv6_reassembly_timeout :
9172             ipst->ips_ip_reassembly_timeout);
9173 
9174         frag_pending = ill_frag_timeout(ill, timeout);
9175 
9176         /*
9177          * Restart the timer, if we have fragments pending or if someone
9178          * wanted us to be scheduled again.
9179          */
9180         mutex_enter(&ill->ill_lock);
9181         ill->ill_fragtimer_executing = 0;
9182         ill->ill_frag_timer_id = 0;
9183         if (frag_pending || ill->ill_fragtimer_needrestart)
9184                 ill_frag_timer_start(ill);
9185         mutex_exit(&ill->ill_lock);
9186 }
9187 
9188 void
9189 ill_frag_timer_start(ill_t *ill)
9190 {
9191         ip_stack_t *ipst = ill->ill_ipst;
9192         clock_t timeo_ms;
9193 
9194         ASSERT(MUTEX_HELD(&ill->ill_lock));
9195 
9196         /* If the ill is closing or opening don't proceed */
9197         if (ill->ill_state_flags & ILL_CONDEMNED)
9198                 return;
9199 
9200         if (ill->ill_fragtimer_executing) {
9201                 /*
9202                  * ill_frag_timer is currently executing. Just record the
9203                  * the fact that we want the timer to be restarted.
9204                  * ill_frag_timer will post a timeout before it returns,
9205                  * ensuring it will be called again.
9206                  */
9207                 ill->ill_fragtimer_needrestart = 1;
9208                 return;
9209         }
9210 
9211         if (ill->ill_frag_timer_id == 0) {
9212                 timeo_ms = (ill->ill_isv6 ? ipst->ips_ipv6_reassembly_timeout :
9213                     ipst->ips_ip_reassembly_timeout) * SECONDS;
9214 
9215                 /*
9216                  * The timer is neither running nor is the timeout handler
9217                  * executing. Post a timeout so that ill_frag_timer will be
9218                  * called
9219                  */
9220                 ill->ill_frag_timer_id = timeout(ill_frag_timer, ill,
9221                     MSEC_TO_TICK(timeo_ms >> 1));
9222                 ill->ill_fragtimer_needrestart = 0;
9223         }
9224 }
9225 
9226 /*
9227  * Update any source route, record route or timestamp options.
9228  * Check that we are at end of strict source route.
9229  * The options have already been checked for sanity in ip_input_options().
9230  */
9231 boolean_t
9232 ip_input_local_options(mblk_t *mp, ipha_t *ipha, ip_recv_attr_t *ira)
9233 {
9234         ipoptp_t        opts;
9235         uchar_t         *opt;
9236         uint8_t         optval;
9237         uint8_t         optlen;
9238         ipaddr_t        dst;
9239         ipaddr_t        ifaddr;
9240         uint32_t        ts;
9241         timestruc_t     now;
9242         ill_t           *ill = ira->ira_ill;
9243         ip_stack_t      *ipst = ill->ill_ipst;
9244 
9245         ip2dbg(("ip_input_local_options\n"));
9246 
9247         for (optval = ipoptp_first(&opts, ipha);
9248             optval != IPOPT_EOL;
9249             optval = ipoptp_next(&opts)) {
9250                 ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
9251                 opt = opts.ipoptp_cur;
9252                 optlen = opts.ipoptp_len;
9253                 ip2dbg(("ip_input_local_options: opt %d, len %d\n",
9254                     optval, optlen));
9255                 switch (optval) {
9256                         uint32_t off;
9257                 case IPOPT_SSRR:
9258                 case IPOPT_LSRR:
9259                         off = opt[IPOPT_OFFSET];
9260                         off--;
9261                         if (optlen < IP_ADDR_LEN ||
9262                             off > optlen - IP_ADDR_LEN) {
9263                                 /* End of source route */
9264                                 ip1dbg(("ip_input_local_options: end of SR\n"));
9265                                 break;
9266                         }
9267                         /*
9268                          * This will only happen if two consecutive entries
9269                          * in the source route contains our address or if
9270                          * it is a packet with a loose source route which
9271                          * reaches us before consuming the whole source route
9272                          */
9273                         ip1dbg(("ip_input_local_options: not end of SR\n"));
9274                         if (optval == IPOPT_SSRR) {
9275                                 goto bad_src_route;
9276                         }
9277                         /*
9278                          * Hack: instead of dropping the packet truncate the
9279                          * source route to what has been used by filling the
9280                          * rest with IPOPT_NOP.
9281                          */
9282                         opt[IPOPT_OLEN] = (uint8_t)off;
9283                         while (off < optlen) {
9284                                 opt[off++] = IPOPT_NOP;
9285                         }
9286                         break;
9287                 case IPOPT_RR:
9288                         off = opt[IPOPT_OFFSET];
9289                         off--;
9290                         if (optlen < IP_ADDR_LEN ||
9291                             off > optlen - IP_ADDR_LEN) {
9292                                 /* No more room - ignore */
9293                                 ip1dbg((
9294                                     "ip_input_local_options: end of RR\n"));
9295                                 break;
9296                         }
9297                         /* Pick a reasonable address on the outbound if */
9298                         if (ip_select_source_v4(ill, INADDR_ANY, ipha->ipha_dst,
9299                             INADDR_ANY, ALL_ZONES, ipst, &ifaddr, NULL,
9300                             NULL) != 0) {
9301                                 /* No source! Shouldn't happen */
9302                                 ifaddr = INADDR_ANY;
9303                         }
9304                         bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9305                         opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9306                         break;
9307                 case IPOPT_TS:
9308                         /* Insert timestamp if there is romm */
9309                         switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9310                         case IPOPT_TS_TSONLY:
9311                                 off = IPOPT_TS_TIMELEN;
9312                                 break;
9313                         case IPOPT_TS_PRESPEC:
9314                         case IPOPT_TS_PRESPEC_RFC791:
9315                                 /* Verify that the address matched */
9316                                 off = opt[IPOPT_OFFSET] - 1;
9317                                 bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
9318                                 if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
9319                                         /* Not for us */
9320                                         break;
9321                                 }
9322                                 /* FALLTHROUGH */
9323                         case IPOPT_TS_TSANDADDR:
9324                                 off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
9325                                 break;
9326                         default:
9327                                 /*
9328                                  * ip_*put_options should have already
9329                                  * dropped this packet.
9330                                  */
9331                                 cmn_err(CE_PANIC, "ip_input_local_options: "
9332                                     "unknown IT - bug in ip_input_options?\n");
9333                                 return (B_TRUE);        /* Keep "lint" happy */
9334                         }
9335                         if (opt[IPOPT_OFFSET] - 1 + off > optlen) {
9336                                 /* Increase overflow counter */
9337                                 off = (opt[IPOPT_POS_OV_FLG] >> 4) + 1;
9338                                 opt[IPOPT_POS_OV_FLG] =
9339                                     (uint8_t)((opt[IPOPT_POS_OV_FLG] & 0x0F) |
9340                                     (off << 4));
9341                                 break;
9342                         }
9343                         off = opt[IPOPT_OFFSET] - 1;
9344                         switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9345                         case IPOPT_TS_PRESPEC:
9346                         case IPOPT_TS_PRESPEC_RFC791:
9347                         case IPOPT_TS_TSANDADDR:
9348                                 /* Pick a reasonable addr on the outbound if */
9349                                 if (ip_select_source_v4(ill, INADDR_ANY,
9350                                     ipha->ipha_dst, INADDR_ANY, ALL_ZONES, ipst,
9351                                     &ifaddr, NULL, NULL) != 0) {
9352                                         /* No source! Shouldn't happen */
9353                                         ifaddr = INADDR_ANY;
9354                                 }
9355                                 bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9356                                 opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9357                                 /* FALLTHROUGH */
9358                         case IPOPT_TS_TSONLY:
9359                                 off = opt[IPOPT_OFFSET] - 1;
9360                                 /* Compute # of milliseconds since midnight */
9361                                 gethrestime(&now);
9362                                 ts = (now.tv_sec % (24 * 60 * 60)) * 1000 +
9363                                     NSEC2MSEC(now.tv_nsec);
9364                                 bcopy(&ts, (char *)opt + off, IPOPT_TS_TIMELEN);
9365                                 opt[IPOPT_OFFSET] += IPOPT_TS_TIMELEN;
9366                                 break;
9367                         }
9368                         break;
9369                 }
9370         }
9371         return (B_TRUE);
9372 
9373 bad_src_route:
9374         /* make sure we clear any indication of a hardware checksum */
9375         DB_CKSUMFLAGS(mp) = 0;
9376         ip_drop_input("ICMP_SOURCE_ROUTE_FAILED", mp, ill);
9377         icmp_unreachable(mp, ICMP_SOURCE_ROUTE_FAILED, ira);
9378         return (B_FALSE);
9379 
9380 }
9381 
9382 /*
9383  * Process IP options in an inbound packet.  Always returns the nexthop.
9384  * Normally this is the passed in nexthop, but if there is an option
9385  * that effects the nexthop (such as a source route) that will be returned.
9386  * Sets *errorp if there is an error, in which case an ICMP error has been sent
9387  * and mp freed.
9388  */
9389 ipaddr_t
9390 ip_input_options(ipha_t *ipha, ipaddr_t dst, mblk_t *mp,
9391     ip_recv_attr_t *ira, int *errorp)
9392 {
9393         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
9394         ipoptp_t        opts;
9395         uchar_t         *opt;
9396         uint8_t         optval;
9397         uint8_t         optlen;
9398         intptr_t        code = 0;
9399         ire_t           *ire;
9400 
9401         ip2dbg(("ip_input_options\n"));
9402         *errorp = 0;
9403         for (optval = ipoptp_first(&opts, ipha);
9404             optval != IPOPT_EOL;
9405             optval = ipoptp_next(&opts)) {
9406                 opt = opts.ipoptp_cur;
9407                 optlen = opts.ipoptp_len;
9408                 ip2dbg(("ip_input_options: opt %d, len %d\n",
9409                     optval, optlen));
9410                 /*
9411                  * Note: we need to verify the checksum before we
9412                  * modify anything thus this routine only extracts the next
9413                  * hop dst from any source route.
9414                  */
9415                 switch (optval) {
9416                         uint32_t off;
9417                 case IPOPT_SSRR:
9418                 case IPOPT_LSRR:
9419                         if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
9420                                 if (optval == IPOPT_SSRR) {
9421                                         ip1dbg(("ip_input_options: not next"
9422                                             " strict source route 0x%x\n",
9423                                             ntohl(dst)));
9424                                         code = (char *)&ipha->ipha_dst -
9425                                             (char *)ipha;
9426                                         goto param_prob; /* RouterReq's */
9427                                 }
9428                                 ip2dbg(("ip_input_options: "
9429                                     "not next source route 0x%x\n",
9430                                     ntohl(dst)));
9431                                 break;
9432                         }
9433 
9434                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
9435                                 ip1dbg((
9436                                     "ip_input_options: bad option offset\n"));
9437                                 code = (char *)&opt[IPOPT_OLEN] -
9438                                     (char *)ipha;
9439                                 goto param_prob;
9440                         }
9441                         off = opt[IPOPT_OFFSET];
9442                         off--;
9443                 redo_srr:
9444                         if (optlen < IP_ADDR_LEN ||
9445                             off > optlen - IP_ADDR_LEN) {
9446                                 /* End of source route */
9447                                 ip1dbg(("ip_input_options: end of SR\n"));
9448                                 break;
9449                         }
9450                         bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
9451                         ip1dbg(("ip_input_options: next hop 0x%x\n",
9452                             ntohl(dst)));
9453 
9454                         /*
9455                          * Check if our address is present more than
9456                          * once as consecutive hops in source route.
9457                          * XXX verify per-interface ip_forwarding
9458                          * for source route?
9459                          */
9460                         if (ip_type_v4(dst, ipst) == IRE_LOCAL) {
9461                                 off += IP_ADDR_LEN;
9462                                 goto redo_srr;
9463                         }
9464 
9465                         if (dst == htonl(INADDR_LOOPBACK)) {
9466                                 ip1dbg(("ip_input_options: loopback addr in "
9467                                     "source route!\n"));
9468                                 goto bad_src_route;
9469                         }
9470                         /*
9471                          * For strict: verify that dst is directly
9472                          * reachable.
9473                          */
9474                         if (optval == IPOPT_SSRR) {
9475                                 ire = ire_ftable_lookup_v4(dst, 0, 0,
9476                                     IRE_INTERFACE, NULL, ALL_ZONES,
9477                                     ira->ira_tsl,
9478                                     MATCH_IRE_TYPE | MATCH_IRE_SECATTR, 0, ipst,
9479                                     NULL);
9480                                 if (ire == NULL) {
9481                                         ip1dbg(("ip_input_options: SSRR not "
9482                                             "directly reachable: 0x%x\n",
9483                                             ntohl(dst)));
9484                                         goto bad_src_route;
9485                                 }
9486                                 ire_refrele(ire);
9487                         }
9488                         /*
9489                          * Defer update of the offset and the record route
9490                          * until the packet is forwarded.
9491                          */
9492                         break;
9493                 case IPOPT_RR:
9494                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
9495                                 ip1dbg((
9496                                     "ip_input_options: bad option offset\n"));
9497                                 code = (char *)&opt[IPOPT_OLEN] -
9498                                     (char *)ipha;
9499                                 goto param_prob;
9500                         }
9501                         break;
9502                 case IPOPT_TS:
9503                         /*
9504                          * Verify that length >= 5 and that there is either
9505                          * room for another timestamp or that the overflow
9506                          * counter is not maxed out.
9507                          */
9508                         code = (char *)&opt[IPOPT_OLEN] - (char *)ipha;
9509                         if (optlen < IPOPT_MINLEN_IT) {
9510                                 goto param_prob;
9511                         }
9512                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
9513                                 ip1dbg((
9514                                     "ip_input_options: bad option offset\n"));
9515                                 code = (char *)&opt[IPOPT_OFFSET] -
9516                                     (char *)ipha;
9517                                 goto param_prob;
9518                         }
9519                         switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9520                         case IPOPT_TS_TSONLY:
9521                                 off = IPOPT_TS_TIMELEN;
9522                                 break;
9523                         case IPOPT_TS_TSANDADDR:
9524                         case IPOPT_TS_PRESPEC:
9525                         case IPOPT_TS_PRESPEC_RFC791:
9526                                 off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
9527                                 break;
9528                         default:
9529                                 code = (char *)&opt[IPOPT_POS_OV_FLG] -
9530                                     (char *)ipha;
9531                                 goto param_prob;
9532                         }
9533                         if (opt[IPOPT_OFFSET] - 1 + off > optlen &&
9534                             (opt[IPOPT_POS_OV_FLG] & 0xF0) == 0xF0) {
9535                                 /*
9536                                  * No room and the overflow counter is 15
9537                                  * already.
9538                                  */
9539                                 goto param_prob;
9540                         }
9541                         break;
9542                 }
9543         }
9544 
9545         if ((opts.ipoptp_flags & IPOPTP_ERROR) == 0) {
9546                 return (dst);
9547         }
9548 
9549         ip1dbg(("ip_input_options: error processing IP options."));
9550         code = (char *)&opt[IPOPT_OFFSET] - (char *)ipha;
9551 
9552 param_prob:
9553         /* make sure we clear any indication of a hardware checksum */
9554         DB_CKSUMFLAGS(mp) = 0;
9555         ip_drop_input("ICMP_PARAM_PROBLEM", mp, ira->ira_ill);
9556         icmp_param_problem(mp, (uint8_t)code, ira);
9557         *errorp = -1;
9558         return (dst);
9559 
9560 bad_src_route:
9561         /* make sure we clear any indication of a hardware checksum */
9562         DB_CKSUMFLAGS(mp) = 0;
9563         ip_drop_input("ICMP_SOURCE_ROUTE_FAILED", mp, ira->ira_ill);
9564         icmp_unreachable(mp, ICMP_SOURCE_ROUTE_FAILED, ira);
9565         *errorp = -1;
9566         return (dst);
9567 }
9568 
9569 /*
9570  * IP & ICMP info in >=14 msg's ...
9571  *  - ip fixed part (mib2_ip_t)
9572  *  - icmp fixed part (mib2_icmp_t)
9573  *  - ipAddrEntryTable (ip 20)          all IPv4 ipifs
9574  *  - ipRouteEntryTable (ip 21)         all IPv4 IREs
9575  *  - ipNetToMediaEntryTable (ip 22)    all IPv4 Neighbor Cache entries
9576  *  - ipRouteAttributeTable (ip 102)    labeled routes
9577  *  - ip multicast membership (ip_member_t)
9578  *  - ip multicast source filtering (ip_grpsrc_t)
9579  *  - igmp fixed part (struct igmpstat)
9580  *  - multicast routing stats (struct mrtstat)
9581  *  - multicast routing vifs (array of struct vifctl)
9582  *  - multicast routing routes (array of struct mfcctl)
9583  *  - ip6 fixed part (mib2_ipv6IfStatsEntry_t)
9584  *                                      One per ill plus one generic
9585  *  - icmp6 fixed part (mib2_ipv6IfIcmpEntry_t)
9586  *                                      One per ill plus one generic
9587  *  - ipv6RouteEntry                    all IPv6 IREs
9588  *  - ipv6RouteAttributeTable (ip6 102) labeled routes
9589  *  - ipv6NetToMediaEntry               all IPv6 Neighbor Cache entries
9590  *  - ipv6AddrEntry                     all IPv6 ipifs
9591  *  - ipv6 multicast membership (ipv6_member_t)
9592  *  - ipv6 multicast source filtering (ipv6_grpsrc_t)
9593  *
9594  * NOTE: original mpctl is copied for msg's 2..N, since its ctl part is
9595  * already filled in by the caller.
9596  * If legacy_req is true then MIB structures needs to be truncated to their
9597  * legacy sizes before being returned.
9598  * Return value of 0 indicates that no messages were sent and caller
9599  * should free mpctl.
9600  */
9601 int
9602 ip_snmp_get(queue_t *q, mblk_t *mpctl, int level, boolean_t legacy_req)
9603 {
9604         ip_stack_t *ipst;
9605         sctp_stack_t *sctps;
9606 
9607         if (q->q_next != NULL) {
9608                 ipst = ILLQ_TO_IPST(q);
9609         } else {
9610                 ipst = CONNQ_TO_IPST(q);
9611         }
9612         ASSERT(ipst != NULL);
9613         sctps = ipst->ips_netstack->netstack_sctp;
9614 
9615         if (mpctl == NULL || mpctl->b_cont == NULL) {
9616                 return (0);
9617         }
9618 
9619         /*
9620          * For the purposes of the (broken) packet shell use
9621          * of the level we make sure MIB2_TCP/MIB2_UDP can be used
9622          * to make TCP and UDP appear first in the list of mib items.
9623          * TBD: We could expand this and use it in netstat so that
9624          * the kernel doesn't have to produce large tables (connections,
9625          * routes, etc) when netstat only wants the statistics or a particular
9626          * table.
9627          */
9628         if (!(level == MIB2_TCP || level == MIB2_UDP)) {
9629                 if ((mpctl = icmp_snmp_get(q, mpctl)) == NULL) {
9630                         return (1);
9631                 }
9632         }
9633 
9634         if (level != MIB2_TCP) {
9635                 if ((mpctl = udp_snmp_get(q, mpctl, legacy_req)) == NULL) {
9636                         return (1);
9637                 }
9638         }
9639 
9640         if (level != MIB2_UDP) {
9641                 if ((mpctl = tcp_snmp_get(q, mpctl, legacy_req)) == NULL) {
9642                         return (1);
9643                 }
9644         }
9645 
9646         if ((mpctl = ip_snmp_get_mib2_ip_traffic_stats(q, mpctl,
9647             ipst, legacy_req)) == NULL) {
9648                 return (1);
9649         }
9650 
9651         if ((mpctl = ip_snmp_get_mib2_ip6(q, mpctl, ipst,
9652             legacy_req)) == NULL) {
9653                 return (1);
9654         }
9655 
9656         if ((mpctl = ip_snmp_get_mib2_icmp(q, mpctl, ipst)) == NULL) {
9657                 return (1);
9658         }
9659 
9660         if ((mpctl = ip_snmp_get_mib2_icmp6(q, mpctl, ipst)) == NULL) {
9661                 return (1);
9662         }
9663 
9664         if ((mpctl = ip_snmp_get_mib2_igmp(q, mpctl, ipst)) == NULL) {
9665                 return (1);
9666         }
9667 
9668         if ((mpctl = ip_snmp_get_mib2_multi(q, mpctl, ipst)) == NULL) {
9669                 return (1);
9670         }
9671 
9672         if ((mpctl = ip_snmp_get_mib2_ip_addr(q, mpctl, ipst,
9673             legacy_req)) == NULL) {
9674                 return (1);
9675         }
9676 
9677         if ((mpctl = ip_snmp_get_mib2_ip6_addr(q, mpctl, ipst,
9678             legacy_req)) == NULL) {
9679                 return (1);
9680         }
9681 
9682         if ((mpctl = ip_snmp_get_mib2_ip_group_mem(q, mpctl, ipst)) == NULL) {
9683                 return (1);
9684         }
9685 
9686         if ((mpctl = ip_snmp_get_mib2_ip6_group_mem(q, mpctl, ipst)) == NULL) {
9687                 return (1);
9688         }
9689 
9690         if ((mpctl = ip_snmp_get_mib2_ip_group_src(q, mpctl, ipst)) == NULL) {
9691                 return (1);
9692         }
9693 
9694         if ((mpctl = ip_snmp_get_mib2_ip6_group_src(q, mpctl, ipst)) == NULL) {
9695                 return (1);
9696         }
9697 
9698         if ((mpctl = ip_snmp_get_mib2_virt_multi(q, mpctl, ipst)) == NULL) {
9699                 return (1);
9700         }
9701 
9702         if ((mpctl = ip_snmp_get_mib2_multi_rtable(q, mpctl, ipst)) == NULL) {
9703                 return (1);
9704         }
9705 
9706         mpctl = ip_snmp_get_mib2_ip_route_media(q, mpctl, level, ipst);
9707         if (mpctl == NULL)
9708                 return (1);
9709 
9710         mpctl = ip_snmp_get_mib2_ip6_route_media(q, mpctl, level, ipst);
9711         if (mpctl == NULL)
9712                 return (1);
9713 
9714         if ((mpctl = sctp_snmp_get_mib2(q, mpctl, sctps)) == NULL) {
9715                 return (1);
9716         }
9717         if ((mpctl = ip_snmp_get_mib2_ip_dce(q, mpctl, ipst)) == NULL) {
9718                 return (1);
9719         }
9720         freemsg(mpctl);
9721         return (1);
9722 }
9723 
9724 /* Get global (legacy) IPv4 statistics */
9725 static mblk_t *
9726 ip_snmp_get_mib2_ip(queue_t *q, mblk_t *mpctl, mib2_ipIfStatsEntry_t *ipmib,
9727     ip_stack_t *ipst, boolean_t legacy_req)
9728 {
9729         mib2_ip_t               old_ip_mib;
9730         struct opthdr           *optp;
9731         mblk_t                  *mp2ctl;
9732         mib2_ipAddrEntry_t      mae;
9733 
9734         /*
9735          * make a copy of the original message
9736          */
9737         mp2ctl = copymsg(mpctl);
9738 
9739         /* fixed length IP structure... */
9740         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9741         optp->level = MIB2_IP;
9742         optp->name = 0;
9743         SET_MIB(old_ip_mib.ipForwarding,
9744             (WE_ARE_FORWARDING(ipst) ? 1 : 2));
9745         SET_MIB(old_ip_mib.ipDefaultTTL,
9746             (uint32_t)ipst->ips_ip_def_ttl);
9747         SET_MIB(old_ip_mib.ipReasmTimeout,
9748             ipst->ips_ip_reassembly_timeout);
9749         SET_MIB(old_ip_mib.ipAddrEntrySize,
9750             (legacy_req) ? LEGACY_MIB_SIZE(&mae, mib2_ipAddrEntry_t) :
9751             sizeof (mib2_ipAddrEntry_t));
9752         SET_MIB(old_ip_mib.ipRouteEntrySize,
9753             sizeof (mib2_ipRouteEntry_t));
9754         SET_MIB(old_ip_mib.ipNetToMediaEntrySize,
9755             sizeof (mib2_ipNetToMediaEntry_t));
9756         SET_MIB(old_ip_mib.ipMemberEntrySize, sizeof (ip_member_t));
9757         SET_MIB(old_ip_mib.ipGroupSourceEntrySize, sizeof (ip_grpsrc_t));
9758         SET_MIB(old_ip_mib.ipRouteAttributeSize,
9759             sizeof (mib2_ipAttributeEntry_t));
9760         SET_MIB(old_ip_mib.transportMLPSize, sizeof (mib2_transportMLPEntry_t));
9761         SET_MIB(old_ip_mib.ipDestEntrySize, sizeof (dest_cache_entry_t));
9762 
9763         /*
9764          * Grab the statistics from the new IP MIB
9765          */
9766         SET_MIB(old_ip_mib.ipInReceives,
9767             (uint32_t)ipmib->ipIfStatsHCInReceives);
9768         SET_MIB(old_ip_mib.ipInHdrErrors, ipmib->ipIfStatsInHdrErrors);
9769         SET_MIB(old_ip_mib.ipInAddrErrors, ipmib->ipIfStatsInAddrErrors);
9770         SET_MIB(old_ip_mib.ipForwDatagrams,
9771             (uint32_t)ipmib->ipIfStatsHCOutForwDatagrams);
9772         SET_MIB(old_ip_mib.ipInUnknownProtos,
9773             ipmib->ipIfStatsInUnknownProtos);
9774         SET_MIB(old_ip_mib.ipInDiscards, ipmib->ipIfStatsInDiscards);
9775         SET_MIB(old_ip_mib.ipInDelivers,
9776             (uint32_t)ipmib->ipIfStatsHCInDelivers);
9777         SET_MIB(old_ip_mib.ipOutRequests,
9778             (uint32_t)ipmib->ipIfStatsHCOutRequests);
9779         SET_MIB(old_ip_mib.ipOutDiscards, ipmib->ipIfStatsOutDiscards);
9780         SET_MIB(old_ip_mib.ipOutNoRoutes, ipmib->ipIfStatsOutNoRoutes);
9781         SET_MIB(old_ip_mib.ipReasmReqds, ipmib->ipIfStatsReasmReqds);
9782         SET_MIB(old_ip_mib.ipReasmOKs, ipmib->ipIfStatsReasmOKs);
9783         SET_MIB(old_ip_mib.ipReasmFails, ipmib->ipIfStatsReasmFails);
9784         SET_MIB(old_ip_mib.ipFragOKs, ipmib->ipIfStatsOutFragOKs);
9785         SET_MIB(old_ip_mib.ipFragFails, ipmib->ipIfStatsOutFragFails);
9786         SET_MIB(old_ip_mib.ipFragCreates, ipmib->ipIfStatsOutFragCreates);
9787 
9788         /* ipRoutingDiscards is not being used */
9789         SET_MIB(old_ip_mib.ipRoutingDiscards, 0);
9790         SET_MIB(old_ip_mib.tcpInErrs, ipmib->tcpIfStatsInErrs);
9791         SET_MIB(old_ip_mib.udpNoPorts, ipmib->udpIfStatsNoPorts);
9792         SET_MIB(old_ip_mib.ipInCksumErrs, ipmib->ipIfStatsInCksumErrs);
9793         SET_MIB(old_ip_mib.ipReasmDuplicates,
9794             ipmib->ipIfStatsReasmDuplicates);
9795         SET_MIB(old_ip_mib.ipReasmPartDups, ipmib->ipIfStatsReasmPartDups);
9796         SET_MIB(old_ip_mib.ipForwProhibits, ipmib->ipIfStatsForwProhibits);
9797         SET_MIB(old_ip_mib.udpInCksumErrs, ipmib->udpIfStatsInCksumErrs);
9798         SET_MIB(old_ip_mib.udpInOverflows, ipmib->udpIfStatsInOverflows);
9799         SET_MIB(old_ip_mib.rawipInOverflows,
9800             ipmib->rawipIfStatsInOverflows);
9801 
9802         SET_MIB(old_ip_mib.ipsecInSucceeded, ipmib->ipsecIfStatsInSucceeded);
9803         SET_MIB(old_ip_mib.ipsecInFailed, ipmib->ipsecIfStatsInFailed);
9804         SET_MIB(old_ip_mib.ipInIPv6, ipmib->ipIfStatsInWrongIPVersion);
9805         SET_MIB(old_ip_mib.ipOutIPv6, ipmib->ipIfStatsOutWrongIPVersion);
9806         SET_MIB(old_ip_mib.ipOutSwitchIPv6,
9807             ipmib->ipIfStatsOutSwitchIPVersion);
9808 
9809         if (!snmp_append_data(mpctl->b_cont, (char *)&old_ip_mib,
9810             (int)sizeof (old_ip_mib))) {
9811                 ip1dbg(("ip_snmp_get_mib2_ip: failed to allocate %u bytes\n",
9812                     (uint_t)sizeof (old_ip_mib)));
9813         }
9814 
9815         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
9816         ip3dbg(("ip_snmp_get_mib2_ip: level %d, name %d, len %d\n",
9817             (int)optp->level, (int)optp->name, (int)optp->len));
9818         qreply(q, mpctl);
9819         return (mp2ctl);
9820 }
9821 
9822 /* Per interface IPv4 statistics */
9823 static mblk_t *
9824 ip_snmp_get_mib2_ip_traffic_stats(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst,
9825     boolean_t legacy_req)
9826 {
9827         struct opthdr           *optp;
9828         mblk_t                  *mp2ctl;
9829         ill_t                   *ill;
9830         ill_walk_context_t      ctx;
9831         mblk_t                  *mp_tail = NULL;
9832         mib2_ipIfStatsEntry_t   global_ip_mib;
9833         mib2_ipAddrEntry_t      mae;
9834 
9835         /*
9836          * Make a copy of the original message
9837          */
9838         mp2ctl = copymsg(mpctl);
9839 
9840         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9841         optp->level = MIB2_IP;
9842         optp->name = MIB2_IP_TRAFFIC_STATS;
9843         /* Include "unknown interface" ip_mib */
9844         ipst->ips_ip_mib.ipIfStatsIPVersion = MIB2_INETADDRESSTYPE_ipv4;
9845         ipst->ips_ip_mib.ipIfStatsIfIndex =
9846             MIB2_UNKNOWN_INTERFACE; /* Flag to netstat */
9847         SET_MIB(ipst->ips_ip_mib.ipIfStatsForwarding,
9848             (ipst->ips_ip_forwarding ? 1 : 2));
9849         SET_MIB(ipst->ips_ip_mib.ipIfStatsDefaultTTL,
9850             (uint32_t)ipst->ips_ip_def_ttl);
9851         SET_MIB(ipst->ips_ip_mib.ipIfStatsEntrySize,
9852             sizeof (mib2_ipIfStatsEntry_t));
9853         SET_MIB(ipst->ips_ip_mib.ipIfStatsAddrEntrySize,
9854             sizeof (mib2_ipAddrEntry_t));
9855         SET_MIB(ipst->ips_ip_mib.ipIfStatsRouteEntrySize,
9856             sizeof (mib2_ipRouteEntry_t));
9857         SET_MIB(ipst->ips_ip_mib.ipIfStatsNetToMediaEntrySize,
9858             sizeof (mib2_ipNetToMediaEntry_t));
9859         SET_MIB(ipst->ips_ip_mib.ipIfStatsMemberEntrySize,
9860             sizeof (ip_member_t));
9861         SET_MIB(ipst->ips_ip_mib.ipIfStatsGroupSourceEntrySize,
9862             sizeof (ip_grpsrc_t));
9863 
9864         bcopy(&ipst->ips_ip_mib, &global_ip_mib, sizeof (global_ip_mib));
9865 
9866         if (legacy_req) {
9867                 SET_MIB(global_ip_mib.ipIfStatsAddrEntrySize,
9868                     LEGACY_MIB_SIZE(&mae, mib2_ipAddrEntry_t));
9869         }
9870 
9871         if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
9872             (char *)&global_ip_mib, (int)sizeof (global_ip_mib))) {
9873                 ip1dbg(("ip_snmp_get_mib2_ip_traffic_stats: "
9874                     "failed to allocate %u bytes\n",
9875                     (uint_t)sizeof (global_ip_mib)));
9876         }
9877 
9878         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
9879         ill = ILL_START_WALK_V4(&ctx, ipst);
9880         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
9881                 ill->ill_ip_mib->ipIfStatsIfIndex =
9882                     ill->ill_phyint->phyint_ifindex;
9883                 SET_MIB(ill->ill_ip_mib->ipIfStatsForwarding,
9884                     (ipst->ips_ip_forwarding ? 1 : 2));
9885                 SET_MIB(ill->ill_ip_mib->ipIfStatsDefaultTTL,
9886                     (uint32_t)ipst->ips_ip_def_ttl);
9887 
9888                 ip_mib2_add_ip_stats(&global_ip_mib, ill->ill_ip_mib);
9889                 if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
9890                     (char *)ill->ill_ip_mib,
9891                     (int)sizeof (*ill->ill_ip_mib))) {
9892                         ip1dbg(("ip_snmp_get_mib2_ip_traffic_stats: "
9893                             "failed to allocate %u bytes\n",
9894                             (uint_t)sizeof (*ill->ill_ip_mib)));
9895                 }
9896         }
9897         rw_exit(&ipst->ips_ill_g_lock);
9898 
9899         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
9900         ip3dbg(("ip_snmp_get_mib2_ip_traffic_stats: "
9901             "level %d, name %d, len %d\n",
9902             (int)optp->level, (int)optp->name, (int)optp->len));
9903         qreply(q, mpctl);
9904 
9905         if (mp2ctl == NULL)
9906                 return (NULL);
9907 
9908         return (ip_snmp_get_mib2_ip(q, mp2ctl, &global_ip_mib, ipst,
9909             legacy_req));
9910 }
9911 
9912 /* Global IPv4 ICMP statistics */
9913 static mblk_t *
9914 ip_snmp_get_mib2_icmp(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
9915 {
9916         struct opthdr           *optp;
9917         mblk_t                  *mp2ctl;
9918 
9919         /*
9920          * Make a copy of the original message
9921          */
9922         mp2ctl = copymsg(mpctl);
9923 
9924         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9925         optp->level = MIB2_ICMP;
9926         optp->name = 0;
9927         if (!snmp_append_data(mpctl->b_cont, (char *)&ipst->ips_icmp_mib,
9928             (int)sizeof (ipst->ips_icmp_mib))) {
9929                 ip1dbg(("ip_snmp_get_mib2_icmp: failed to allocate %u bytes\n",
9930                     (uint_t)sizeof (ipst->ips_icmp_mib)));
9931         }
9932         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
9933         ip3dbg(("ip_snmp_get_mib2_icmp: level %d, name %d, len %d\n",
9934             (int)optp->level, (int)optp->name, (int)optp->len));
9935         qreply(q, mpctl);
9936         return (mp2ctl);
9937 }
9938 
9939 /* Global IPv4 IGMP statistics */
9940 static mblk_t *
9941 ip_snmp_get_mib2_igmp(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
9942 {
9943         struct opthdr           *optp;
9944         mblk_t                  *mp2ctl;
9945 
9946         /*
9947          * make a copy of the original message
9948          */
9949         mp2ctl = copymsg(mpctl);
9950 
9951         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9952         optp->level = EXPER_IGMP;
9953         optp->name = 0;
9954         if (!snmp_append_data(mpctl->b_cont, (char *)&ipst->ips_igmpstat,
9955             (int)sizeof (ipst->ips_igmpstat))) {
9956                 ip1dbg(("ip_snmp_get_mib2_igmp: failed to allocate %u bytes\n",
9957                     (uint_t)sizeof (ipst->ips_igmpstat)));
9958         }
9959         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
9960         ip3dbg(("ip_snmp_get_mib2_igmp: level %d, name %d, len %d\n",
9961             (int)optp->level, (int)optp->name, (int)optp->len));
9962         qreply(q, mpctl);
9963         return (mp2ctl);
9964 }
9965 
9966 /* Global IPv4 Multicast Routing statistics */
9967 static mblk_t *
9968 ip_snmp_get_mib2_multi(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
9969 {
9970         struct opthdr           *optp;
9971         mblk_t                  *mp2ctl;
9972 
9973         /*
9974          * make a copy of the original message
9975          */
9976         mp2ctl = copymsg(mpctl);
9977 
9978         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9979         optp->level = EXPER_DVMRP;
9980         optp->name = 0;
9981         if (!ip_mroute_stats(mpctl->b_cont, ipst)) {
9982                 ip0dbg(("ip_mroute_stats: failed\n"));
9983         }
9984         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
9985         ip3dbg(("ip_snmp_get_mib2_multi: level %d, name %d, len %d\n",
9986             (int)optp->level, (int)optp->name, (int)optp->len));
9987         qreply(q, mpctl);
9988         return (mp2ctl);
9989 }
9990 
9991 /* IPv4 address information */
9992 static mblk_t *
9993 ip_snmp_get_mib2_ip_addr(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst,
9994     boolean_t legacy_req)
9995 {
9996         struct opthdr           *optp;
9997         mblk_t                  *mp2ctl;
9998         mblk_t                  *mp_tail = NULL;
9999         ill_t                   *ill;
10000         ipif_t                  *ipif;
10001         uint_t                  bitval;
10002         mib2_ipAddrEntry_t      mae;
10003         size_t                  mae_size;
10004         zoneid_t                zoneid;
10005         ill_walk_context_t      ctx;
10006 
10007         /*
10008          * make a copy of the original message
10009          */
10010         mp2ctl = copymsg(mpctl);
10011 
10012         mae_size = (legacy_req) ? LEGACY_MIB_SIZE(&mae, mib2_ipAddrEntry_t) :
10013             sizeof (mib2_ipAddrEntry_t);
10014 
10015         /* ipAddrEntryTable */
10016 
10017         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10018         optp->level = MIB2_IP;
10019         optp->name = MIB2_IP_ADDR;
10020         zoneid = Q_TO_CONN(q)->conn_zoneid;
10021 
10022         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10023         ill = ILL_START_WALK_V4(&ctx, ipst);
10024         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10025                 for (ipif = ill->ill_ipif; ipif != NULL;
10026                     ipif = ipif->ipif_next) {
10027                         if (ipif->ipif_zoneid != zoneid &&
10028                             ipif->ipif_zoneid != ALL_ZONES)
10029                                 continue;
10030                         /* Sum of count from dead IRE_LO* and our current */
10031                         mae.ipAdEntInfo.ae_ibcnt = ipif->ipif_ib_pkt_count;
10032                         if (ipif->ipif_ire_local != NULL) {
10033                                 mae.ipAdEntInfo.ae_ibcnt +=
10034                                     ipif->ipif_ire_local->ire_ib_pkt_count;
10035                         }
10036                         mae.ipAdEntInfo.ae_obcnt = 0;
10037                         mae.ipAdEntInfo.ae_focnt = 0;
10038 
10039                         ipif_get_name(ipif, mae.ipAdEntIfIndex.o_bytes,
10040                             OCTET_LENGTH);
10041                         mae.ipAdEntIfIndex.o_length =
10042                             mi_strlen(mae.ipAdEntIfIndex.o_bytes);
10043                         mae.ipAdEntAddr = ipif->ipif_lcl_addr;
10044                         mae.ipAdEntNetMask = ipif->ipif_net_mask;
10045                         mae.ipAdEntInfo.ae_subnet = ipif->ipif_subnet;
10046                         mae.ipAdEntInfo.ae_subnet_len =
10047                             ip_mask_to_plen(ipif->ipif_net_mask);
10048                         mae.ipAdEntInfo.ae_src_addr = ipif->ipif_lcl_addr;
10049                         for (bitval = 1;
10050                             bitval &&
10051                             !(bitval & ipif->ipif_brd_addr);
10052                             bitval <<= 1)
10053                                 noop;
10054                         mae.ipAdEntBcastAddr = bitval;
10055                         mae.ipAdEntReasmMaxSize = IP_MAXPACKET;
10056                         mae.ipAdEntInfo.ae_mtu = ipif->ipif_ill->ill_mtu;
10057                         mae.ipAdEntInfo.ae_metric  = ipif->ipif_ill->ill_metric;
10058                         mae.ipAdEntInfo.ae_broadcast_addr =
10059                             ipif->ipif_brd_addr;
10060                         mae.ipAdEntInfo.ae_pp_dst_addr =
10061                             ipif->ipif_pp_dst_addr;
10062                         mae.ipAdEntInfo.ae_flags = ipif->ipif_flags |
10063                             ill->ill_flags | ill->ill_phyint->phyint_flags;
10064                         mae.ipAdEntRetransmitTime =
10065                             ill->ill_reachable_retrans_time;
10066 
10067                         if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10068                             (char *)&mae, (int)mae_size)) {
10069                                 ip1dbg(("ip_snmp_get_mib2_ip_addr: failed to "
10070                                     "allocate %u bytes\n", (uint_t)mae_size));
10071                         }
10072                 }
10073         }
10074         rw_exit(&ipst->ips_ill_g_lock);
10075 
10076         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10077         ip3dbg(("ip_snmp_get_mib2_ip_addr: level %d, name %d, len %d\n",
10078             (int)optp->level, (int)optp->name, (int)optp->len));
10079         qreply(q, mpctl);
10080         return (mp2ctl);
10081 }
10082 
10083 /* IPv6 address information */
10084 static mblk_t *
10085 ip_snmp_get_mib2_ip6_addr(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst,
10086     boolean_t legacy_req)
10087 {
10088         struct opthdr           *optp;
10089         mblk_t                  *mp2ctl;
10090         mblk_t                  *mp_tail = NULL;
10091         ill_t                   *ill;
10092         ipif_t                  *ipif;
10093         mib2_ipv6AddrEntry_t    mae6;
10094         size_t                  mae6_size;
10095         zoneid_t                zoneid;
10096         ill_walk_context_t      ctx;
10097 
10098         /*
10099          * make a copy of the original message
10100          */
10101         mp2ctl = copymsg(mpctl);
10102 
10103         mae6_size = (legacy_req) ?
10104             LEGACY_MIB_SIZE(&mae6, mib2_ipv6AddrEntry_t) :
10105             sizeof (mib2_ipv6AddrEntry_t);
10106 
10107         /* ipv6AddrEntryTable */
10108 
10109         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10110         optp->level = MIB2_IP6;
10111         optp->name = MIB2_IP6_ADDR;
10112         zoneid = Q_TO_CONN(q)->conn_zoneid;
10113 
10114         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10115         ill = ILL_START_WALK_V6(&ctx, ipst);
10116         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10117                 for (ipif = ill->ill_ipif; ipif != NULL;
10118                     ipif = ipif->ipif_next) {
10119                         if (ipif->ipif_zoneid != zoneid &&
10120                             ipif->ipif_zoneid != ALL_ZONES)
10121                                 continue;
10122                         /* Sum of count from dead IRE_LO* and our current */
10123                         mae6.ipv6AddrInfo.ae_ibcnt = ipif->ipif_ib_pkt_count;
10124                         if (ipif->ipif_ire_local != NULL) {
10125                                 mae6.ipv6AddrInfo.ae_ibcnt +=
10126                                     ipif->ipif_ire_local->ire_ib_pkt_count;
10127                         }
10128                         mae6.ipv6AddrInfo.ae_obcnt = 0;
10129                         mae6.ipv6AddrInfo.ae_focnt = 0;
10130 
10131                         ipif_get_name(ipif, mae6.ipv6AddrIfIndex.o_bytes,
10132                             OCTET_LENGTH);
10133                         mae6.ipv6AddrIfIndex.o_length =
10134                             mi_strlen(mae6.ipv6AddrIfIndex.o_bytes);
10135                         mae6.ipv6AddrAddress = ipif->ipif_v6lcl_addr;
10136                         mae6.ipv6AddrPfxLength =
10137                             ip_mask_to_plen_v6(&ipif->ipif_v6net_mask);
10138                         mae6.ipv6AddrInfo.ae_subnet = ipif->ipif_v6subnet;
10139                         mae6.ipv6AddrInfo.ae_subnet_len =
10140                             mae6.ipv6AddrPfxLength;
10141                         mae6.ipv6AddrInfo.ae_src_addr = ipif->ipif_v6lcl_addr;
10142 
10143                         /* Type: stateless(1), stateful(2), unknown(3) */
10144                         if (ipif->ipif_flags & IPIF_ADDRCONF)
10145                                 mae6.ipv6AddrType = 1;
10146                         else
10147                                 mae6.ipv6AddrType = 2;
10148                         /* Anycast: true(1), false(2) */
10149                         if (ipif->ipif_flags & IPIF_ANYCAST)
10150                                 mae6.ipv6AddrAnycastFlag = 1;
10151                         else
10152                                 mae6.ipv6AddrAnycastFlag = 2;
10153 
10154                         /*
10155                          * Address status: preferred(1), deprecated(2),
10156                          * invalid(3), inaccessible(4), unknown(5)
10157                          */
10158                         if (ipif->ipif_flags & IPIF_NOLOCAL)
10159                                 mae6.ipv6AddrStatus = 3;
10160                         else if (ipif->ipif_flags & IPIF_DEPRECATED)
10161                                 mae6.ipv6AddrStatus = 2;
10162                         else
10163                                 mae6.ipv6AddrStatus = 1;
10164                         mae6.ipv6AddrInfo.ae_mtu = ipif->ipif_ill->ill_mtu;
10165                         mae6.ipv6AddrInfo.ae_metric  =
10166                             ipif->ipif_ill->ill_metric;
10167                         mae6.ipv6AddrInfo.ae_pp_dst_addr =
10168                             ipif->ipif_v6pp_dst_addr;
10169                         mae6.ipv6AddrInfo.ae_flags = ipif->ipif_flags |
10170                             ill->ill_flags | ill->ill_phyint->phyint_flags;
10171                         mae6.ipv6AddrReasmMaxSize = IP_MAXPACKET;
10172                         mae6.ipv6AddrIdentifier = ill->ill_token;
10173                         mae6.ipv6AddrIdentifierLen = ill->ill_token_length;
10174                         mae6.ipv6AddrReachableTime = ill->ill_reachable_time;
10175                         mae6.ipv6AddrRetransmitTime =
10176                             ill->ill_reachable_retrans_time;
10177                         if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10178                             (char *)&mae6, (int)mae6_size)) {
10179                                 ip1dbg(("ip_snmp_get_mib2_ip6_addr: failed to "
10180                                     "allocate %u bytes\n",
10181                                     (uint_t)mae6_size));
10182                         }
10183                 }
10184         }
10185         rw_exit(&ipst->ips_ill_g_lock);
10186 
10187         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10188         ip3dbg(("ip_snmp_get_mib2_ip6_addr: level %d, name %d, len %d\n",
10189             (int)optp->level, (int)optp->name, (int)optp->len));
10190         qreply(q, mpctl);
10191         return (mp2ctl);
10192 }
10193 
10194 /* IPv4 multicast group membership. */
10195 static mblk_t *
10196 ip_snmp_get_mib2_ip_group_mem(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10197 {
10198         struct opthdr           *optp;
10199         mblk_t                  *mp2ctl;
10200         ill_t                   *ill;
10201         ipif_t                  *ipif;
10202         ilm_t                   *ilm;
10203         ip_member_t             ipm;
10204         mblk_t                  *mp_tail = NULL;
10205         ill_walk_context_t      ctx;
10206         zoneid_t                zoneid;
10207 
10208         /*
10209          * make a copy of the original message
10210          */
10211         mp2ctl = copymsg(mpctl);
10212         zoneid = Q_TO_CONN(q)->conn_zoneid;
10213 
10214         /* ipGroupMember table */
10215         optp = (struct opthdr *)&mpctl->b_rptr[
10216             sizeof (struct T_optmgmt_ack)];
10217         optp->level = MIB2_IP;
10218         optp->name = EXPER_IP_GROUP_MEMBERSHIP;
10219 
10220         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10221         ill = ILL_START_WALK_V4(&ctx, ipst);
10222         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10223                 /* Make sure the ill isn't going away. */
10224                 if (!ill_check_and_refhold(ill))
10225                         continue;
10226                 rw_exit(&ipst->ips_ill_g_lock);
10227                 rw_enter(&ill->ill_mcast_lock, RW_READER);
10228                 for (ilm = ill->ill_ilm; ilm; ilm = ilm->ilm_next) {
10229                         if (ilm->ilm_zoneid != zoneid &&
10230                             ilm->ilm_zoneid != ALL_ZONES)
10231                                 continue;
10232 
10233                         /* Is there an ipif for ilm_ifaddr? */
10234                         for (ipif = ill->ill_ipif; ipif != NULL;
10235                             ipif = ipif->ipif_next) {
10236                                 if (!IPIF_IS_CONDEMNED(ipif) &&
10237                                     ipif->ipif_lcl_addr == ilm->ilm_ifaddr &&
10238                                     ilm->ilm_ifaddr != INADDR_ANY)
10239                                         break;
10240                         }
10241                         if (ipif != NULL) {
10242                                 ipif_get_name(ipif,
10243                                     ipm.ipGroupMemberIfIndex.o_bytes,
10244                                     OCTET_LENGTH);
10245                         } else {
10246                                 ill_get_name(ill,
10247                                     ipm.ipGroupMemberIfIndex.o_bytes,
10248                                     OCTET_LENGTH);
10249                         }
10250                         ipm.ipGroupMemberIfIndex.o_length =
10251                             mi_strlen(ipm.ipGroupMemberIfIndex.o_bytes);
10252 
10253                         ipm.ipGroupMemberAddress = ilm->ilm_addr;
10254                         ipm.ipGroupMemberRefCnt = ilm->ilm_refcnt;
10255                         ipm.ipGroupMemberFilterMode = ilm->ilm_fmode;
10256                         if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10257                             (char *)&ipm, (int)sizeof (ipm))) {
10258                                 ip1dbg(("ip_snmp_get_mib2_ip_group: "
10259                                     "failed to allocate %u bytes\n",
10260                                     (uint_t)sizeof (ipm)));
10261                         }
10262                 }
10263                 rw_exit(&ill->ill_mcast_lock);
10264                 ill_refrele(ill);
10265                 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10266         }
10267         rw_exit(&ipst->ips_ill_g_lock);
10268         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10269         ip3dbg(("ip_snmp_get: level %d, name %d, len %d\n",
10270             (int)optp->level, (int)optp->name, (int)optp->len));
10271         qreply(q, mpctl);
10272         return (mp2ctl);
10273 }
10274 
10275 /* IPv6 multicast group membership. */
10276 static mblk_t *
10277 ip_snmp_get_mib2_ip6_group_mem(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10278 {
10279         struct opthdr           *optp;
10280         mblk_t                  *mp2ctl;
10281         ill_t                   *ill;
10282         ilm_t                   *ilm;
10283         ipv6_member_t           ipm6;
10284         mblk_t                  *mp_tail = NULL;
10285         ill_walk_context_t      ctx;
10286         zoneid_t                zoneid;
10287 
10288         /*
10289          * make a copy of the original message
10290          */
10291         mp2ctl = copymsg(mpctl);
10292         zoneid = Q_TO_CONN(q)->conn_zoneid;
10293 
10294         /* ip6GroupMember table */
10295         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10296         optp->level = MIB2_IP6;
10297         optp->name = EXPER_IP6_GROUP_MEMBERSHIP;
10298 
10299         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10300         ill = ILL_START_WALK_V6(&ctx, ipst);
10301         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10302                 /* Make sure the ill isn't going away. */
10303                 if (!ill_check_and_refhold(ill))
10304                         continue;
10305                 rw_exit(&ipst->ips_ill_g_lock);
10306                 /*
10307                  * Normally we don't have any members on under IPMP interfaces.
10308                  * We report them as a debugging aid.
10309                  */
10310                 rw_enter(&ill->ill_mcast_lock, RW_READER);
10311                 ipm6.ipv6GroupMemberIfIndex = ill->ill_phyint->phyint_ifindex;
10312                 for (ilm = ill->ill_ilm; ilm; ilm = ilm->ilm_next) {
10313                         if (ilm->ilm_zoneid != zoneid &&
10314                             ilm->ilm_zoneid != ALL_ZONES)
10315                                 continue;       /* not this zone */
10316                         ipm6.ipv6GroupMemberAddress = ilm->ilm_v6addr;
10317                         ipm6.ipv6GroupMemberRefCnt = ilm->ilm_refcnt;
10318                         ipm6.ipv6GroupMemberFilterMode = ilm->ilm_fmode;
10319                         if (!snmp_append_data2(mpctl->b_cont,
10320                             &mp_tail,
10321                             (char *)&ipm6, (int)sizeof (ipm6))) {
10322                                 ip1dbg(("ip_snmp_get_mib2_ip6_group: "
10323                                     "failed to allocate %u bytes\n",
10324                                     (uint_t)sizeof (ipm6)));
10325                         }
10326                 }
10327                 rw_exit(&ill->ill_mcast_lock);
10328                 ill_refrele(ill);
10329                 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10330         }
10331         rw_exit(&ipst->ips_ill_g_lock);
10332 
10333         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10334         ip3dbg(("ip_snmp_get: level %d, name %d, len %d\n",
10335             (int)optp->level, (int)optp->name, (int)optp->len));
10336         qreply(q, mpctl);
10337         return (mp2ctl);
10338 }
10339 
10340 /* IP multicast filtered sources */
10341 static mblk_t *
10342 ip_snmp_get_mib2_ip_group_src(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10343 {
10344         struct opthdr           *optp;
10345         mblk_t                  *mp2ctl;
10346         ill_t                   *ill;
10347         ipif_t                  *ipif;
10348         ilm_t                   *ilm;
10349         ip_grpsrc_t             ips;
10350         mblk_t                  *mp_tail = NULL;
10351         ill_walk_context_t      ctx;
10352         zoneid_t                zoneid;
10353         int                     i;
10354         slist_t                 *sl;
10355 
10356         /*
10357          * make a copy of the original message
10358          */
10359         mp2ctl = copymsg(mpctl);
10360         zoneid = Q_TO_CONN(q)->conn_zoneid;
10361 
10362         /* ipGroupSource table */
10363         optp = (struct opthdr *)&mpctl->b_rptr[
10364             sizeof (struct T_optmgmt_ack)];
10365         optp->level = MIB2_IP;
10366         optp->name = EXPER_IP_GROUP_SOURCES;
10367 
10368         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10369         ill = ILL_START_WALK_V4(&ctx, ipst);
10370         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10371                 /* Make sure the ill isn't going away. */
10372                 if (!ill_check_and_refhold(ill))
10373                         continue;
10374                 rw_exit(&ipst->ips_ill_g_lock);
10375                 rw_enter(&ill->ill_mcast_lock, RW_READER);
10376                 for (ilm = ill->ill_ilm; ilm; ilm = ilm->ilm_next) {
10377                         sl = ilm->ilm_filter;
10378                         if (ilm->ilm_zoneid != zoneid &&
10379                             ilm->ilm_zoneid != ALL_ZONES)
10380                                 continue;
10381                         if (SLIST_IS_EMPTY(sl))
10382                                 continue;
10383 
10384                         /* Is there an ipif for ilm_ifaddr? */
10385                         for (ipif = ill->ill_ipif; ipif != NULL;
10386                             ipif = ipif->ipif_next) {
10387                                 if (!IPIF_IS_CONDEMNED(ipif) &&
10388                                     ipif->ipif_lcl_addr == ilm->ilm_ifaddr &&
10389                                     ilm->ilm_ifaddr != INADDR_ANY)
10390                                         break;
10391                         }
10392                         if (ipif != NULL) {
10393                                 ipif_get_name(ipif,
10394                                     ips.ipGroupSourceIfIndex.o_bytes,
10395                                     OCTET_LENGTH);
10396                         } else {
10397                                 ill_get_name(ill,
10398                                     ips.ipGroupSourceIfIndex.o_bytes,
10399                                     OCTET_LENGTH);
10400                         }
10401                         ips.ipGroupSourceIfIndex.o_length =
10402                             mi_strlen(ips.ipGroupSourceIfIndex.o_bytes);
10403 
10404                         ips.ipGroupSourceGroup = ilm->ilm_addr;
10405                         for (i = 0; i < sl->sl_numsrc; i++) {
10406                                 if (!IN6_IS_ADDR_V4MAPPED(&sl->sl_addr[i]))
10407                                         continue;
10408                                 IN6_V4MAPPED_TO_IPADDR(&sl->sl_addr[i],
10409                                     ips.ipGroupSourceAddress);
10410                                 if (snmp_append_data2(mpctl->b_cont, &mp_tail,
10411                                     (char *)&ips, (int)sizeof (ips)) == 0) {
10412                                         ip1dbg(("ip_snmp_get_mib2_ip_group_src:"
10413                                             " failed to allocate %u bytes\n",
10414                                             (uint_t)sizeof (ips)));
10415                                 }
10416                         }
10417                 }
10418                 rw_exit(&ill->ill_mcast_lock);
10419                 ill_refrele(ill);
10420                 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10421         }
10422         rw_exit(&ipst->ips_ill_g_lock);
10423         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10424         ip3dbg(("ip_snmp_get: level %d, name %d, len %d\n",
10425             (int)optp->level, (int)optp->name, (int)optp->len));
10426         qreply(q, mpctl);
10427         return (mp2ctl);
10428 }
10429 
10430 /* IPv6 multicast filtered sources. */
10431 static mblk_t *
10432 ip_snmp_get_mib2_ip6_group_src(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10433 {
10434         struct opthdr           *optp;
10435         mblk_t                  *mp2ctl;
10436         ill_t                   *ill;
10437         ilm_t                   *ilm;
10438         ipv6_grpsrc_t           ips6;
10439         mblk_t                  *mp_tail = NULL;
10440         ill_walk_context_t      ctx;
10441         zoneid_t                zoneid;
10442         int                     i;
10443         slist_t                 *sl;
10444 
10445         /*
10446          * make a copy of the original message
10447          */
10448         mp2ctl = copymsg(mpctl);
10449         zoneid = Q_TO_CONN(q)->conn_zoneid;
10450 
10451         /* ip6GroupMember table */
10452         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10453         optp->level = MIB2_IP6;
10454         optp->name = EXPER_IP6_GROUP_SOURCES;
10455 
10456         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10457         ill = ILL_START_WALK_V6(&ctx, ipst);
10458         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10459                 /* Make sure the ill isn't going away. */
10460                 if (!ill_check_and_refhold(ill))
10461                         continue;
10462                 rw_exit(&ipst->ips_ill_g_lock);
10463                 /*
10464                  * Normally we don't have any members on under IPMP interfaces.
10465                  * We report them as a debugging aid.
10466                  */
10467                 rw_enter(&ill->ill_mcast_lock, RW_READER);
10468                 ips6.ipv6GroupSourceIfIndex = ill->ill_phyint->phyint_ifindex;
10469                 for (ilm = ill->ill_ilm; ilm; ilm = ilm->ilm_next) {
10470                         sl = ilm->ilm_filter;
10471                         if (ilm->ilm_zoneid != zoneid &&
10472                             ilm->ilm_zoneid != ALL_ZONES)
10473                                 continue;
10474                         if (SLIST_IS_EMPTY(sl))
10475                                 continue;
10476                         ips6.ipv6GroupSourceGroup = ilm->ilm_v6addr;
10477                         for (i = 0; i < sl->sl_numsrc; i++) {
10478                                 ips6.ipv6GroupSourceAddress = sl->sl_addr[i];
10479                                 if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10480                                     (char *)&ips6, (int)sizeof (ips6))) {
10481                                         ip1dbg(("ip_snmp_get_mib2_ip6_"
10482                                             "group_src: failed to allocate "
10483                                             "%u bytes\n",
10484                                             (uint_t)sizeof (ips6)));
10485                                 }
10486                         }
10487                 }
10488                 rw_exit(&ill->ill_mcast_lock);
10489                 ill_refrele(ill);
10490                 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10491         }
10492         rw_exit(&ipst->ips_ill_g_lock);
10493 
10494         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10495         ip3dbg(("ip_snmp_get: level %d, name %d, len %d\n",
10496             (int)optp->level, (int)optp->name, (int)optp->len));
10497         qreply(q, mpctl);
10498         return (mp2ctl);
10499 }
10500 
10501 /* Multicast routing virtual interface table. */
10502 static mblk_t *
10503 ip_snmp_get_mib2_virt_multi(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10504 {
10505         struct opthdr           *optp;
10506         mblk_t                  *mp2ctl;
10507 
10508         /*
10509          * make a copy of the original message
10510          */
10511         mp2ctl = copymsg(mpctl);
10512 
10513         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10514         optp->level = EXPER_DVMRP;
10515         optp->name = EXPER_DVMRP_VIF;
10516         if (!ip_mroute_vif(mpctl->b_cont, ipst)) {
10517                 ip0dbg(("ip_mroute_vif: failed\n"));
10518         }
10519         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10520         ip3dbg(("ip_snmp_get_mib2_virt_multi: level %d, name %d, len %d\n",
10521             (int)optp->level, (int)optp->name, (int)optp->len));
10522         qreply(q, mpctl);
10523         return (mp2ctl);
10524 }
10525 
10526 /* Multicast routing table. */
10527 static mblk_t *
10528 ip_snmp_get_mib2_multi_rtable(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10529 {
10530         struct opthdr           *optp;
10531         mblk_t                  *mp2ctl;
10532 
10533         /*
10534          * make a copy of the original message
10535          */
10536         mp2ctl = copymsg(mpctl);
10537 
10538         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10539         optp->level = EXPER_DVMRP;
10540         optp->name = EXPER_DVMRP_MRT;
10541         if (!ip_mroute_mrt(mpctl->b_cont, ipst)) {
10542                 ip0dbg(("ip_mroute_mrt: failed\n"));
10543         }
10544         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10545         ip3dbg(("ip_snmp_get_mib2_multi_rtable: level %d, name %d, len %d\n",
10546             (int)optp->level, (int)optp->name, (int)optp->len));
10547         qreply(q, mpctl);
10548         return (mp2ctl);
10549 }
10550 
10551 /*
10552  * Return ipRouteEntryTable, ipNetToMediaEntryTable, and ipRouteAttributeTable
10553  * in one IRE walk.
10554  */
10555 static mblk_t *
10556 ip_snmp_get_mib2_ip_route_media(queue_t *q, mblk_t *mpctl, int level,
10557     ip_stack_t *ipst)
10558 {
10559         struct opthdr   *optp;
10560         mblk_t          *mp2ctl;        /* Returned */
10561         mblk_t          *mp3ctl;        /* nettomedia */
10562         mblk_t          *mp4ctl;        /* routeattrs */
10563         iproutedata_t   ird;
10564         zoneid_t        zoneid;
10565 
10566         /*
10567          * make copies of the original message
10568          *      - mp2ctl is returned unchanged to the caller for its use
10569          *      - mpctl is sent upstream as ipRouteEntryTable
10570          *      - mp3ctl is sent upstream as ipNetToMediaEntryTable
10571          *      - mp4ctl is sent upstream as ipRouteAttributeTable
10572          */
10573         mp2ctl = copymsg(mpctl);
10574         mp3ctl = copymsg(mpctl);
10575         mp4ctl = copymsg(mpctl);
10576         if (mp3ctl == NULL || mp4ctl == NULL) {
10577                 freemsg(mp4ctl);
10578                 freemsg(mp3ctl);
10579                 freemsg(mp2ctl);
10580                 freemsg(mpctl);
10581                 return (NULL);
10582         }
10583 
10584         bzero(&ird, sizeof (ird));
10585 
10586         ird.ird_route.lp_head = mpctl->b_cont;
10587         ird.ird_netmedia.lp_head = mp3ctl->b_cont;
10588         ird.ird_attrs.lp_head = mp4ctl->b_cont;
10589         /*
10590          * If the level has been set the special EXPER_IP_AND_ALL_IRES value,
10591          * then also include ire_testhidden IREs and IRE_IF_CLONE.  This is
10592          * intended a temporary solution until a proper MIB API is provided
10593          * that provides complete filtering/caller-opt-in.
10594          */
10595         if (level == EXPER_IP_AND_ALL_IRES)
10596                 ird.ird_flags |= IRD_REPORT_ALL;
10597 
10598         zoneid = Q_TO_CONN(q)->conn_zoneid;
10599         ire_walk_v4(ip_snmp_get2_v4, &ird, zoneid, ipst);
10600 
10601         /* ipRouteEntryTable in mpctl */
10602         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10603         optp->level = MIB2_IP;
10604         optp->name = MIB2_IP_ROUTE;
10605         optp->len = msgdsize(ird.ird_route.lp_head);
10606         ip3dbg(("ip_snmp_get_mib2_ip_route_media: level %d, name %d, len %d\n",
10607             (int)optp->level, (int)optp->name, (int)optp->len));
10608         qreply(q, mpctl);
10609 
10610         /* ipNetToMediaEntryTable in mp3ctl */
10611         ncec_walk(NULL, ip_snmp_get2_v4_media, &ird, ipst);
10612 
10613         optp = (struct opthdr *)&mp3ctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10614         optp->level = MIB2_IP;
10615         optp->name = MIB2_IP_MEDIA;
10616         optp->len = msgdsize(ird.ird_netmedia.lp_head);
10617         ip3dbg(("ip_snmp_get_mib2_ip_route_media: level %d, name %d, len %d\n",
10618             (int)optp->level, (int)optp->name, (int)optp->len));
10619         qreply(q, mp3ctl);
10620 
10621         /* ipRouteAttributeTable in mp4ctl */
10622         optp = (struct opthdr *)&mp4ctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10623         optp->level = MIB2_IP;
10624         optp->name = EXPER_IP_RTATTR;
10625         optp->len = msgdsize(ird.ird_attrs.lp_head);
10626         ip3dbg(("ip_snmp_get_mib2_ip_route_media: level %d, name %d, len %d\n",
10627             (int)optp->level, (int)optp->name, (int)optp->len));
10628         if (optp->len == 0)
10629                 freemsg(mp4ctl);
10630         else
10631                 qreply(q, mp4ctl);
10632 
10633         return (mp2ctl);
10634 }
10635 
10636 /*
10637  * Return ipv6RouteEntryTable and ipv6RouteAttributeTable in one IRE walk, and
10638  * ipv6NetToMediaEntryTable in an NDP walk.
10639  */
10640 static mblk_t *
10641 ip_snmp_get_mib2_ip6_route_media(queue_t *q, mblk_t *mpctl, int level,
10642     ip_stack_t *ipst)
10643 {
10644         struct opthdr   *optp;
10645         mblk_t          *mp2ctl;        /* Returned */
10646         mblk_t          *mp3ctl;        /* nettomedia */
10647         mblk_t          *mp4ctl;        /* routeattrs */
10648         iproutedata_t   ird;
10649         zoneid_t        zoneid;
10650 
10651         /*
10652          * make copies of the original message
10653          *      - mp2ctl is returned unchanged to the caller for its use
10654          *      - mpctl is sent upstream as ipv6RouteEntryTable
10655          *      - mp3ctl is sent upstream as ipv6NetToMediaEntryTable
10656          *      - mp4ctl is sent upstream as ipv6RouteAttributeTable
10657          */
10658         mp2ctl = copymsg(mpctl);
10659         mp3ctl = copymsg(mpctl);
10660         mp4ctl = copymsg(mpctl);
10661         if (mp3ctl == NULL || mp4ctl == NULL) {
10662                 freemsg(mp4ctl);
10663                 freemsg(mp3ctl);
10664                 freemsg(mp2ctl);
10665                 freemsg(mpctl);
10666                 return (NULL);
10667         }
10668 
10669         bzero(&ird, sizeof (ird));
10670 
10671         ird.ird_route.lp_head = mpctl->b_cont;
10672         ird.ird_netmedia.lp_head = mp3ctl->b_cont;
10673         ird.ird_attrs.lp_head = mp4ctl->b_cont;
10674         /*
10675          * If the level has been set the special EXPER_IP_AND_ALL_IRES value,
10676          * then also include ire_testhidden IREs and IRE_IF_CLONE.  This is
10677          * intended a temporary solution until a proper MIB API is provided
10678          * that provides complete filtering/caller-opt-in.
10679          */
10680         if (level == EXPER_IP_AND_ALL_IRES)
10681                 ird.ird_flags |= IRD_REPORT_ALL;
10682 
10683         zoneid = Q_TO_CONN(q)->conn_zoneid;
10684         ire_walk_v6(ip_snmp_get2_v6_route, &ird, zoneid, ipst);
10685 
10686         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10687         optp->level = MIB2_IP6;
10688         optp->name = MIB2_IP6_ROUTE;
10689         optp->len = msgdsize(ird.ird_route.lp_head);
10690         ip3dbg(("ip_snmp_get_mib2_ip6_route_media: level %d, name %d, len %d\n",
10691             (int)optp->level, (int)optp->name, (int)optp->len));
10692         qreply(q, mpctl);
10693 
10694         /* ipv6NetToMediaEntryTable in mp3ctl */
10695         ncec_walk(NULL, ip_snmp_get2_v6_media, &ird, ipst);
10696 
10697         optp = (struct opthdr *)&mp3ctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10698         optp->level = MIB2_IP6;
10699         optp->name = MIB2_IP6_MEDIA;
10700         optp->len = msgdsize(ird.ird_netmedia.lp_head);
10701         ip3dbg(("ip_snmp_get_mib2_ip6_route_media: level %d, name %d, len %d\n",
10702             (int)optp->level, (int)optp->name, (int)optp->len));
10703         qreply(q, mp3ctl);
10704 
10705         /* ipv6RouteAttributeTable in mp4ctl */
10706         optp = (struct opthdr *)&mp4ctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10707         optp->level = MIB2_IP6;
10708         optp->name = EXPER_IP_RTATTR;
10709         optp->len = msgdsize(ird.ird_attrs.lp_head);
10710         ip3dbg(("ip_snmp_get_mib2_ip6_route_media: level %d, name %d, len %d\n",
10711             (int)optp->level, (int)optp->name, (int)optp->len));
10712         if (optp->len == 0)
10713                 freemsg(mp4ctl);
10714         else
10715                 qreply(q, mp4ctl);
10716 
10717         return (mp2ctl);
10718 }
10719 
10720 /*
10721  * IPv6 mib: One per ill
10722  */
10723 static mblk_t *
10724 ip_snmp_get_mib2_ip6(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst,
10725     boolean_t legacy_req)
10726 {
10727         struct opthdr           *optp;
10728         mblk_t                  *mp2ctl;
10729         ill_t                   *ill;
10730         ill_walk_context_t      ctx;
10731         mblk_t                  *mp_tail = NULL;
10732         mib2_ipv6AddrEntry_t    mae6;
10733         mib2_ipIfStatsEntry_t   *ise;
10734         size_t                  ise_size, iae_size;
10735 
10736         /*
10737          * Make a copy of the original message
10738          */
10739         mp2ctl = copymsg(mpctl);
10740 
10741         /* fixed length IPv6 structure ... */
10742 
10743         if (legacy_req) {
10744                 ise_size = LEGACY_MIB_SIZE(&ipst->ips_ip6_mib,
10745                     mib2_ipIfStatsEntry_t);
10746                 iae_size = LEGACY_MIB_SIZE(&mae6, mib2_ipv6AddrEntry_t);
10747         } else {
10748                 ise_size = sizeof (mib2_ipIfStatsEntry_t);
10749                 iae_size = sizeof (mib2_ipv6AddrEntry_t);
10750         }
10751 
10752         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10753         optp->level = MIB2_IP6;
10754         optp->name = 0;
10755         /* Include "unknown interface" ip6_mib */
10756         ipst->ips_ip6_mib.ipIfStatsIPVersion = MIB2_INETADDRESSTYPE_ipv6;
10757         ipst->ips_ip6_mib.ipIfStatsIfIndex =
10758             MIB2_UNKNOWN_INTERFACE; /* Flag to netstat */
10759         SET_MIB(ipst->ips_ip6_mib.ipIfStatsForwarding,
10760             ipst->ips_ipv6_forwarding ? 1 : 2);
10761         SET_MIB(ipst->ips_ip6_mib.ipIfStatsDefaultHopLimit,
10762             ipst->ips_ipv6_def_hops);
10763         SET_MIB(ipst->ips_ip6_mib.ipIfStatsEntrySize,
10764             sizeof (mib2_ipIfStatsEntry_t));
10765         SET_MIB(ipst->ips_ip6_mib.ipIfStatsAddrEntrySize,
10766             sizeof (mib2_ipv6AddrEntry_t));
10767         SET_MIB(ipst->ips_ip6_mib.ipIfStatsRouteEntrySize,
10768             sizeof (mib2_ipv6RouteEntry_t));
10769         SET_MIB(ipst->ips_ip6_mib.ipIfStatsNetToMediaEntrySize,
10770             sizeof (mib2_ipv6NetToMediaEntry_t));
10771         SET_MIB(ipst->ips_ip6_mib.ipIfStatsMemberEntrySize,
10772             sizeof (ipv6_member_t));
10773         SET_MIB(ipst->ips_ip6_mib.ipIfStatsGroupSourceEntrySize,
10774             sizeof (ipv6_grpsrc_t));
10775 
10776         /*
10777          * Synchronize 64- and 32-bit counters
10778          */
10779         SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsInReceives,
10780             ipIfStatsHCInReceives);
10781         SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsInDelivers,
10782             ipIfStatsHCInDelivers);
10783         SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsOutRequests,
10784             ipIfStatsHCOutRequests);
10785         SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsOutForwDatagrams,
10786             ipIfStatsHCOutForwDatagrams);
10787         SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsOutMcastPkts,
10788             ipIfStatsHCOutMcastPkts);
10789         SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsInMcastPkts,
10790             ipIfStatsHCInMcastPkts);
10791 
10792         if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10793             (char *)&ipst->ips_ip6_mib, (int)ise_size)) {
10794                 ip1dbg(("ip_snmp_get_mib2_ip6: failed to allocate %u bytes\n",
10795                     (uint_t)ise_size));
10796         } else if (legacy_req) {
10797                 /* Adjust the EntrySize fields for legacy requests. */
10798                 ise =
10799                     (mib2_ipIfStatsEntry_t *)(mp_tail->b_wptr - (int)ise_size);
10800                 SET_MIB(ise->ipIfStatsEntrySize, ise_size);
10801                 SET_MIB(ise->ipIfStatsAddrEntrySize, iae_size);
10802         }
10803 
10804         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10805         ill = ILL_START_WALK_V6(&ctx, ipst);
10806         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10807                 ill->ill_ip_mib->ipIfStatsIfIndex =
10808                     ill->ill_phyint->phyint_ifindex;
10809                 SET_MIB(ill->ill_ip_mib->ipIfStatsForwarding,
10810                     ipst->ips_ipv6_forwarding ? 1 : 2);
10811                 SET_MIB(ill->ill_ip_mib->ipIfStatsDefaultHopLimit,
10812                     ill->ill_max_hops);
10813 
10814                 /*
10815                  * Synchronize 64- and 32-bit counters
10816                  */
10817                 SYNC32_MIB(ill->ill_ip_mib, ipIfStatsInReceives,
10818                     ipIfStatsHCInReceives);
10819                 SYNC32_MIB(ill->ill_ip_mib, ipIfStatsInDelivers,
10820                     ipIfStatsHCInDelivers);
10821                 SYNC32_MIB(ill->ill_ip_mib, ipIfStatsOutRequests,
10822                     ipIfStatsHCOutRequests);
10823                 SYNC32_MIB(ill->ill_ip_mib, ipIfStatsOutForwDatagrams,
10824                     ipIfStatsHCOutForwDatagrams);
10825                 SYNC32_MIB(ill->ill_ip_mib, ipIfStatsOutMcastPkts,
10826                     ipIfStatsHCOutMcastPkts);
10827                 SYNC32_MIB(ill->ill_ip_mib, ipIfStatsInMcastPkts,
10828                     ipIfStatsHCInMcastPkts);
10829 
10830                 if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10831                     (char *)ill->ill_ip_mib, (int)ise_size)) {
10832                         ip1dbg(("ip_snmp_get_mib2_ip6: failed to allocate "
10833                         "%u bytes\n", (uint_t)ise_size));
10834                 } else if (legacy_req) {
10835                         /* Adjust the EntrySize fields for legacy requests. */
10836                         ise = (mib2_ipIfStatsEntry_t *)(mp_tail->b_wptr -
10837                             (int)ise_size);
10838                         SET_MIB(ise->ipIfStatsEntrySize, ise_size);
10839                         SET_MIB(ise->ipIfStatsAddrEntrySize, iae_size);
10840                 }
10841         }
10842         rw_exit(&ipst->ips_ill_g_lock);
10843 
10844         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10845         ip3dbg(("ip_snmp_get_mib2_ip6: level %d, name %d, len %d\n",
10846             (int)optp->level, (int)optp->name, (int)optp->len));
10847         qreply(q, mpctl);
10848         return (mp2ctl);
10849 }
10850 
10851 /*
10852  * ICMPv6 mib: One per ill
10853  */
10854 static mblk_t *
10855 ip_snmp_get_mib2_icmp6(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10856 {
10857         struct opthdr           *optp;
10858         mblk_t                  *mp2ctl;
10859         ill_t                   *ill;
10860         ill_walk_context_t      ctx;
10861         mblk_t                  *mp_tail = NULL;
10862         /*
10863          * Make a copy of the original message
10864          */
10865         mp2ctl = copymsg(mpctl);
10866 
10867         /* fixed length ICMPv6 structure ... */
10868 
10869         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10870         optp->level = MIB2_ICMP6;
10871         optp->name = 0;
10872         /* Include "unknown interface" icmp6_mib */
10873         ipst->ips_icmp6_mib.ipv6IfIcmpIfIndex =
10874             MIB2_UNKNOWN_INTERFACE; /* netstat flag */
10875         ipst->ips_icmp6_mib.ipv6IfIcmpEntrySize =
10876             sizeof (mib2_ipv6IfIcmpEntry_t);
10877         if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10878             (char *)&ipst->ips_icmp6_mib,
10879             (int)sizeof (ipst->ips_icmp6_mib))) {
10880                 ip1dbg(("ip_snmp_get_mib2_icmp6: failed to allocate %u bytes\n",
10881                     (uint_t)sizeof (ipst->ips_icmp6_mib)));
10882         }
10883 
10884         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10885         ill = ILL_START_WALK_V6(&ctx, ipst);
10886         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10887                 ill->ill_icmp6_mib->ipv6IfIcmpIfIndex =
10888                     ill->ill_phyint->phyint_ifindex;
10889                 if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10890                     (char *)ill->ill_icmp6_mib,
10891                     (int)sizeof (*ill->ill_icmp6_mib))) {
10892                         ip1dbg(("ip_snmp_get_mib2_icmp6: failed to allocate "
10893                             "%u bytes\n",
10894                             (uint_t)sizeof (*ill->ill_icmp6_mib)));
10895                 }
10896         }
10897         rw_exit(&ipst->ips_ill_g_lock);
10898 
10899         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10900         ip3dbg(("ip_snmp_get_mib2_icmp6: level %d, name %d, len %d\n",
10901             (int)optp->level, (int)optp->name, (int)optp->len));
10902         qreply(q, mpctl);
10903         return (mp2ctl);
10904 }
10905 
10906 /*
10907  * ire_walk routine to create both ipRouteEntryTable and
10908  * ipRouteAttributeTable in one IRE walk
10909  */
10910 static void
10911 ip_snmp_get2_v4(ire_t *ire, iproutedata_t *ird)
10912 {
10913         ill_t                           *ill;
10914         mib2_ipRouteEntry_t             *re;
10915         mib2_ipAttributeEntry_t         iaes;
10916         tsol_ire_gw_secattr_t           *attrp;
10917         tsol_gc_t                       *gc = NULL;
10918         tsol_gcgrp_t                    *gcgrp = NULL;
10919         ip_stack_t                      *ipst = ire->ire_ipst;
10920 
10921         ASSERT(ire->ire_ipversion == IPV4_VERSION);
10922 
10923         if (!(ird->ird_flags & IRD_REPORT_ALL)) {
10924                 if (ire->ire_testhidden)
10925                         return;
10926                 if (ire->ire_type & IRE_IF_CLONE)
10927                         return;
10928         }
10929 
10930         if ((re = kmem_zalloc(sizeof (*re), KM_NOSLEEP)) == NULL)
10931                 return;
10932 
10933         if ((attrp = ire->ire_gw_secattr) != NULL) {
10934                 mutex_enter(&attrp->igsa_lock);
10935                 if ((gc = attrp->igsa_gc) != NULL) {
10936                         gcgrp = gc->gc_grp;
10937                         ASSERT(gcgrp != NULL);
10938                         rw_enter(&gcgrp->gcgrp_rwlock, RW_READER);
10939                 }
10940                 mutex_exit(&attrp->igsa_lock);
10941         }
10942         /*
10943          * Return all IRE types for route table... let caller pick and choose
10944          */
10945         re->ipRouteDest = ire->ire_addr;
10946         ill = ire->ire_ill;
10947         re->ipRouteIfIndex.o_length = 0;
10948         if (ill != NULL) {
10949                 ill_get_name(ill, re->ipRouteIfIndex.o_bytes, OCTET_LENGTH);
10950                 re->ipRouteIfIndex.o_length =
10951                     mi_strlen(re->ipRouteIfIndex.o_bytes);
10952         }
10953         re->ipRouteMetric1 = -1;
10954         re->ipRouteMetric2 = -1;
10955         re->ipRouteMetric3 = -1;
10956         re->ipRouteMetric4 = -1;
10957 
10958         re->ipRouteNextHop = ire->ire_gateway_addr;
10959         /* indirect(4), direct(3), or invalid(2) */
10960         if (ire->ire_flags & (RTF_REJECT | RTF_BLACKHOLE))
10961                 re->ipRouteType = 2;
10962         else if (ire->ire_type & IRE_ONLINK)
10963                 re->ipRouteType = 3;
10964         else
10965                 re->ipRouteType = 4;
10966 
10967         re->ipRouteProto = -1;
10968         re->ipRouteAge = gethrestime_sec() - ire->ire_create_time;
10969         re->ipRouteMask = ire->ire_mask;
10970         re->ipRouteMetric5 = -1;
10971         re->ipRouteInfo.re_max_frag = ire->ire_metrics.iulp_mtu;
10972         if (ire->ire_ill != NULL && re->ipRouteInfo.re_max_frag == 0)
10973                 re->ipRouteInfo.re_max_frag = ire->ire_ill->ill_mtu;
10974 
10975         re->ipRouteInfo.re_frag_flag = 0;
10976         re->ipRouteInfo.re_rtt               = 0;
10977         re->ipRouteInfo.re_src_addr  = 0;
10978         re->ipRouteInfo.re_ref               = ire->ire_refcnt;
10979         re->ipRouteInfo.re_obpkt     = ire->ire_ob_pkt_count;
10980         re->ipRouteInfo.re_ibpkt     = ire->ire_ib_pkt_count;
10981         re->ipRouteInfo.re_flags     = ire->ire_flags;
10982 
10983         /* Add the IRE_IF_CLONE's counters to their parent IRE_INTERFACE */
10984         if (ire->ire_type & IRE_INTERFACE) {
10985                 ire_t *child;
10986 
10987                 rw_enter(&ipst->ips_ire_dep_lock, RW_READER);
10988                 child = ire->ire_dep_children;
10989                 while (child != NULL) {
10990                         re->ipRouteInfo.re_obpkt += child->ire_ob_pkt_count;
10991                         re->ipRouteInfo.re_ibpkt += child->ire_ib_pkt_count;
10992                         child = child->ire_dep_sib_next;
10993                 }
10994                 rw_exit(&ipst->ips_ire_dep_lock);
10995         }
10996 
10997         if (ire->ire_flags & RTF_DYNAMIC) {
10998                 re->ipRouteInfo.re_ire_type  = IRE_HOST_REDIRECT;
10999         } else {
11000                 re->ipRouteInfo.re_ire_type  = ire->ire_type;
11001         }
11002 
11003         if (!snmp_append_data2(ird->ird_route.lp_head, &ird->ird_route.lp_tail,
11004             (char *)re, (int)sizeof (*re))) {
11005                 ip1dbg(("ip_snmp_get2_v4: failed to allocate %u bytes\n",
11006                     (uint_t)sizeof (*re)));
11007         }
11008 
11009         if (gc != NULL) {
11010                 iaes.iae_routeidx = ird->ird_idx;
11011                 iaes.iae_doi = gc->gc_db->gcdb_doi;
11012                 iaes.iae_slrange = gc->gc_db->gcdb_slrange;
11013 
11014                 if (!snmp_append_data2(ird->ird_attrs.lp_head,
11015                     &ird->ird_attrs.lp_tail, (char *)&iaes, sizeof (iaes))) {
11016                         ip1dbg(("ip_snmp_get2_v4: failed to allocate %u "
11017                             "bytes\n", (uint_t)sizeof (iaes)));
11018                 }
11019         }
11020 
11021         /* bump route index for next pass */
11022         ird->ird_idx++;
11023 
11024         kmem_free(re, sizeof (*re));
11025         if (gcgrp != NULL)
11026                 rw_exit(&gcgrp->gcgrp_rwlock);
11027 }
11028 
11029 /*
11030  * ire_walk routine to create ipv6RouteEntryTable and ipRouteEntryTable.
11031  */
11032 static void
11033 ip_snmp_get2_v6_route(ire_t *ire, iproutedata_t *ird)
11034 {
11035         ill_t                           *ill;
11036         mib2_ipv6RouteEntry_t           *re;
11037         mib2_ipAttributeEntry_t         iaes;
11038         tsol_ire_gw_secattr_t           *attrp;
11039         tsol_gc_t                       *gc = NULL;
11040         tsol_gcgrp_t                    *gcgrp = NULL;
11041         ip_stack_t                      *ipst = ire->ire_ipst;
11042 
11043         ASSERT(ire->ire_ipversion == IPV6_VERSION);
11044 
11045         if (!(ird->ird_flags & IRD_REPORT_ALL)) {
11046                 if (ire->ire_testhidden)
11047                         return;
11048                 if (ire->ire_type & IRE_IF_CLONE)
11049                         return;
11050         }
11051 
11052         if ((re = kmem_zalloc(sizeof (*re), KM_NOSLEEP)) == NULL)
11053                 return;
11054 
11055         if ((attrp = ire->ire_gw_secattr) != NULL) {
11056                 mutex_enter(&attrp->igsa_lock);
11057                 if ((gc = attrp->igsa_gc) != NULL) {
11058                         gcgrp = gc->gc_grp;
11059                         ASSERT(gcgrp != NULL);
11060                         rw_enter(&gcgrp->gcgrp_rwlock, RW_READER);
11061                 }
11062                 mutex_exit(&attrp->igsa_lock);
11063         }
11064         /*
11065          * Return all IRE types for route table... let caller pick and choose
11066          */
11067         re->ipv6RouteDest = ire->ire_addr_v6;
11068         re->ipv6RoutePfxLength = ip_mask_to_plen_v6(&ire->ire_mask_v6);
11069         re->ipv6RouteIndex = 0;      /* Unique when multiple with same dest/plen */
11070         re->ipv6RouteIfIndex.o_length = 0;
11071         ill = ire->ire_ill;
11072         if (ill != NULL) {
11073                 ill_get_name(ill, re->ipv6RouteIfIndex.o_bytes, OCTET_LENGTH);
11074                 re->ipv6RouteIfIndex.o_length =
11075                     mi_strlen(re->ipv6RouteIfIndex.o_bytes);
11076         }
11077 
11078         ASSERT(!(ire->ire_type & IRE_BROADCAST));
11079 
11080         mutex_enter(&ire->ire_lock);
11081         re->ipv6RouteNextHop = ire->ire_gateway_addr_v6;
11082         mutex_exit(&ire->ire_lock);
11083 
11084         /* remote(4), local(3), or discard(2) */
11085         if (ire->ire_flags & (RTF_REJECT | RTF_BLACKHOLE))
11086                 re->ipv6RouteType = 2;
11087         else if (ire->ire_type & IRE_ONLINK)
11088                 re->ipv6RouteType = 3;
11089         else
11090                 re->ipv6RouteType = 4;
11091 
11092         re->ipv6RouteProtocol        = -1;
11093         re->ipv6RoutePolicy  = 0;
11094         re->ipv6RouteAge     = gethrestime_sec() - ire->ire_create_time;
11095         re->ipv6RouteNextHopRDI      = 0;
11096         re->ipv6RouteWeight  = 0;
11097         re->ipv6RouteMetric  = 0;
11098         re->ipv6RouteInfo.re_max_frag = ire->ire_metrics.iulp_mtu;
11099         if (ire->ire_ill != NULL && re->ipv6RouteInfo.re_max_frag == 0)
11100                 re->ipv6RouteInfo.re_max_frag = ire->ire_ill->ill_mtu;
11101 
11102         re->ipv6RouteInfo.re_frag_flag       = 0;
11103         re->ipv6RouteInfo.re_rtt     = 0;
11104         re->ipv6RouteInfo.re_src_addr        = ipv6_all_zeros;
11105         re->ipv6RouteInfo.re_obpkt   = ire->ire_ob_pkt_count;
11106         re->ipv6RouteInfo.re_ibpkt   = ire->ire_ib_pkt_count;
11107         re->ipv6RouteInfo.re_ref     = ire->ire_refcnt;
11108         re->ipv6RouteInfo.re_flags   = ire->ire_flags;
11109 
11110         /* Add the IRE_IF_CLONE's counters to their parent IRE_INTERFACE */
11111         if (ire->ire_type & IRE_INTERFACE) {
11112                 ire_t *child;
11113 
11114                 rw_enter(&ipst->ips_ire_dep_lock, RW_READER);
11115                 child = ire->ire_dep_children;
11116                 while (child != NULL) {
11117                         re->ipv6RouteInfo.re_obpkt += child->ire_ob_pkt_count;
11118                         re->ipv6RouteInfo.re_ibpkt += child->ire_ib_pkt_count;
11119                         child = child->ire_dep_sib_next;
11120                 }
11121                 rw_exit(&ipst->ips_ire_dep_lock);
11122         }
11123         if (ire->ire_flags & RTF_DYNAMIC) {
11124                 re->ipv6RouteInfo.re_ire_type        = IRE_HOST_REDIRECT;
11125         } else {
11126                 re->ipv6RouteInfo.re_ire_type        = ire->ire_type;
11127         }
11128 
11129         if (!snmp_append_data2(ird->ird_route.lp_head, &ird->ird_route.lp_tail,
11130             (char *)re, (int)sizeof (*re))) {
11131                 ip1dbg(("ip_snmp_get2_v6: failed to allocate %u bytes\n",
11132                     (uint_t)sizeof (*re)));
11133         }
11134 
11135         if (gc != NULL) {
11136                 iaes.iae_routeidx = ird->ird_idx;
11137                 iaes.iae_doi = gc->gc_db->gcdb_doi;
11138                 iaes.iae_slrange = gc->gc_db->gcdb_slrange;
11139 
11140                 if (!snmp_append_data2(ird->ird_attrs.lp_head,
11141                     &ird->ird_attrs.lp_tail, (char *)&iaes, sizeof (iaes))) {
11142                         ip1dbg(("ip_snmp_get2_v6: failed to allocate %u "
11143                             "bytes\n", (uint_t)sizeof (iaes)));
11144                 }
11145         }
11146 
11147         /* bump route index for next pass */
11148         ird->ird_idx++;
11149 
11150         kmem_free(re, sizeof (*re));
11151         if (gcgrp != NULL)
11152                 rw_exit(&gcgrp->gcgrp_rwlock);
11153 }
11154 
11155 /*
11156  * ncec_walk routine to create ipv6NetToMediaEntryTable
11157  */
11158 static void
11159 ip_snmp_get2_v6_media(ncec_t *ncec, void *ptr)
11160 {
11161         iproutedata_t *ird              = ptr;
11162         ill_t                           *ill;
11163         mib2_ipv6NetToMediaEntry_t      ntme;
11164 
11165         ill = ncec->ncec_ill;
11166         /* skip arpce entries, and loopback ncec entries */
11167         if (ill->ill_isv6 == B_FALSE || ill->ill_net_type == IRE_LOOPBACK)
11168                 return;
11169         /*
11170          * Neighbor cache entry attached to IRE with on-link
11171          * destination.
11172          * We report all IPMP groups on ncec_ill which is normally the upper.
11173          */
11174         ntme.ipv6NetToMediaIfIndex = ill->ill_phyint->phyint_ifindex;
11175         ntme.ipv6NetToMediaNetAddress = ncec->ncec_addr;
11176         ntme.ipv6NetToMediaPhysAddress.o_length = ill->ill_phys_addr_length;
11177         if (ncec->ncec_lladdr != NULL) {
11178                 bcopy(ncec->ncec_lladdr, ntme.ipv6NetToMediaPhysAddress.o_bytes,
11179                     ntme.ipv6NetToMediaPhysAddress.o_length);
11180         }
11181         /*
11182          * Note: Returns ND_* states. Should be:
11183          * reachable(1), stale(2), delay(3), probe(4),
11184          * invalid(5), unknown(6)
11185          */
11186         ntme.ipv6NetToMediaState = ncec->ncec_state;
11187         ntme.ipv6NetToMediaLastUpdated = 0;
11188 
11189         /* other(1), dynamic(2), static(3), local(4) */
11190         if (NCE_MYADDR(ncec)) {
11191                 ntme.ipv6NetToMediaType = 4;
11192         } else if (ncec->ncec_flags & NCE_F_PUBLISH) {
11193                 ntme.ipv6NetToMediaType = 1; /* proxy */
11194         } else if (ncec->ncec_flags & NCE_F_STATIC) {
11195                 ntme.ipv6NetToMediaType = 3;
11196         } else if (ncec->ncec_flags & (NCE_F_MCAST|NCE_F_BCAST)) {
11197                 ntme.ipv6NetToMediaType = 1;
11198         } else {
11199                 ntme.ipv6NetToMediaType = 2;
11200         }
11201 
11202         if (!snmp_append_data2(ird->ird_netmedia.lp_head,
11203             &ird->ird_netmedia.lp_tail, (char *)&ntme, sizeof (ntme))) {
11204                 ip1dbg(("ip_snmp_get2_v6_media: failed to allocate %u bytes\n",
11205                     (uint_t)sizeof (ntme)));
11206         }
11207 }
11208 
11209 int
11210 nce2ace(ncec_t *ncec)
11211 {
11212         int flags = 0;
11213 
11214         if (NCE_ISREACHABLE(ncec))
11215                 flags |= ACE_F_RESOLVED;
11216         if (ncec->ncec_flags & NCE_F_AUTHORITY)
11217                 flags |= ACE_F_AUTHORITY;
11218         if (ncec->ncec_flags & NCE_F_PUBLISH)
11219                 flags |= ACE_F_PUBLISH;
11220         if ((ncec->ncec_flags & NCE_F_NONUD) != 0)
11221                 flags |= ACE_F_PERMANENT;
11222         if (NCE_MYADDR(ncec))
11223                 flags |= (ACE_F_MYADDR | ACE_F_AUTHORITY);
11224         if (ncec->ncec_flags & NCE_F_UNVERIFIED)
11225                 flags |= ACE_F_UNVERIFIED;
11226         if (ncec->ncec_flags & NCE_F_AUTHORITY)
11227                 flags |= ACE_F_AUTHORITY;
11228         if (ncec->ncec_flags & NCE_F_DELAYED)
11229                 flags |= ACE_F_DELAYED;
11230         return (flags);
11231 }
11232 
11233 /*
11234  * ncec_walk routine to create ipNetToMediaEntryTable
11235  */
11236 static void
11237 ip_snmp_get2_v4_media(ncec_t *ncec, void *ptr)
11238 {
11239         iproutedata_t *ird              = ptr;
11240         ill_t                           *ill;
11241         mib2_ipNetToMediaEntry_t        ntme;
11242         const char                      *name = "unknown";
11243         ipaddr_t                        ncec_addr;
11244 
11245         ill = ncec->ncec_ill;
11246         if (ill->ill_isv6 || (ncec->ncec_flags & NCE_F_BCAST) ||
11247             ill->ill_net_type == IRE_LOOPBACK)
11248                 return;
11249 
11250         /* We report all IPMP groups on ncec_ill which is normally the upper. */
11251         name = ill->ill_name;
11252         /* Based on RFC 4293: other(1), inval(2), dyn(3), stat(4) */
11253         if (NCE_MYADDR(ncec)) {
11254                 ntme.ipNetToMediaType = 4;
11255         } else if (ncec->ncec_flags & (NCE_F_MCAST|NCE_F_BCAST|NCE_F_PUBLISH)) {
11256                 ntme.ipNetToMediaType = 1;
11257         } else {
11258                 ntme.ipNetToMediaType = 3;
11259         }
11260         ntme.ipNetToMediaIfIndex.o_length = MIN(OCTET_LENGTH, strlen(name));
11261         bcopy(name, ntme.ipNetToMediaIfIndex.o_bytes,
11262             ntme.ipNetToMediaIfIndex.o_length);
11263 
11264         IN6_V4MAPPED_TO_IPADDR(&ncec->ncec_addr, ncec_addr);
11265         bcopy(&ncec_addr, &ntme.ipNetToMediaNetAddress, sizeof (ncec_addr));
11266 
11267         ntme.ipNetToMediaInfo.ntm_mask.o_length = sizeof (ipaddr_t);
11268         ncec_addr = INADDR_BROADCAST;
11269         bcopy(&ncec_addr, ntme.ipNetToMediaInfo.ntm_mask.o_bytes,
11270             sizeof (ncec_addr));
11271         /*
11272          * map all the flags to the ACE counterpart.
11273          */
11274         ntme.ipNetToMediaInfo.ntm_flags = nce2ace(ncec);
11275 
11276         ntme.ipNetToMediaPhysAddress.o_length =
11277             MIN(OCTET_LENGTH, ill->ill_phys_addr_length);
11278 
11279         if (!NCE_ISREACHABLE(ncec))
11280                 ntme.ipNetToMediaPhysAddress.o_length = 0;
11281         else {
11282                 if (ncec->ncec_lladdr != NULL) {
11283                         bcopy(ncec->ncec_lladdr,
11284                             ntme.ipNetToMediaPhysAddress.o_bytes,
11285                             ntme.ipNetToMediaPhysAddress.o_length);
11286                 }
11287         }
11288 
11289         if (!snmp_append_data2(ird->ird_netmedia.lp_head,
11290             &ird->ird_netmedia.lp_tail, (char *)&ntme, sizeof (ntme))) {
11291                 ip1dbg(("ip_snmp_get2_v4_media: failed to allocate %u bytes\n",
11292                     (uint_t)sizeof (ntme)));
11293         }
11294 }
11295 
11296 /*
11297  * return (0) if invalid set request, 1 otherwise, including non-tcp requests
11298  */
11299 /* ARGSUSED */
11300 int
11301 ip_snmp_set(queue_t *q, int level, int name, uchar_t *ptr, int len)
11302 {
11303         switch (level) {
11304         case MIB2_IP:
11305         case MIB2_ICMP:
11306                 switch (name) {
11307                 default:
11308                         break;
11309                 }
11310                 return (1);
11311         default:
11312                 return (1);
11313         }
11314 }
11315 
11316 /*
11317  * When there exists both a 64- and 32-bit counter of a particular type
11318  * (i.e., InReceives), only the 64-bit counters are added.
11319  */
11320 void
11321 ip_mib2_add_ip_stats(mib2_ipIfStatsEntry_t *o1, mib2_ipIfStatsEntry_t *o2)
11322 {
11323         UPDATE_MIB(o1, ipIfStatsInHdrErrors, o2->ipIfStatsInHdrErrors);
11324         UPDATE_MIB(o1, ipIfStatsInTooBigErrors, o2->ipIfStatsInTooBigErrors);
11325         UPDATE_MIB(o1, ipIfStatsInNoRoutes, o2->ipIfStatsInNoRoutes);
11326         UPDATE_MIB(o1, ipIfStatsInAddrErrors, o2->ipIfStatsInAddrErrors);
11327         UPDATE_MIB(o1, ipIfStatsInUnknownProtos, o2->ipIfStatsInUnknownProtos);
11328         UPDATE_MIB(o1, ipIfStatsInTruncatedPkts, o2->ipIfStatsInTruncatedPkts);
11329         UPDATE_MIB(o1, ipIfStatsInDiscards, o2->ipIfStatsInDiscards);
11330         UPDATE_MIB(o1, ipIfStatsOutDiscards, o2->ipIfStatsOutDiscards);
11331         UPDATE_MIB(o1, ipIfStatsOutFragOKs, o2->ipIfStatsOutFragOKs);
11332         UPDATE_MIB(o1, ipIfStatsOutFragFails, o2->ipIfStatsOutFragFails);
11333         UPDATE_MIB(o1, ipIfStatsOutFragCreates, o2->ipIfStatsOutFragCreates);
11334         UPDATE_MIB(o1, ipIfStatsReasmReqds, o2->ipIfStatsReasmReqds);
11335         UPDATE_MIB(o1, ipIfStatsReasmOKs, o2->ipIfStatsReasmOKs);
11336         UPDATE_MIB(o1, ipIfStatsReasmFails, o2->ipIfStatsReasmFails);
11337         UPDATE_MIB(o1, ipIfStatsOutNoRoutes, o2->ipIfStatsOutNoRoutes);
11338         UPDATE_MIB(o1, ipIfStatsReasmDuplicates, o2->ipIfStatsReasmDuplicates);
11339         UPDATE_MIB(o1, ipIfStatsReasmPartDups, o2->ipIfStatsReasmPartDups);
11340         UPDATE_MIB(o1, ipIfStatsForwProhibits, o2->ipIfStatsForwProhibits);
11341         UPDATE_MIB(o1, udpInCksumErrs, o2->udpInCksumErrs);
11342         UPDATE_MIB(o1, udpInOverflows, o2->udpInOverflows);
11343         UPDATE_MIB(o1, rawipInOverflows, o2->rawipInOverflows);
11344         UPDATE_MIB(o1, ipIfStatsInWrongIPVersion,
11345             o2->ipIfStatsInWrongIPVersion);
11346         UPDATE_MIB(o1, ipIfStatsOutWrongIPVersion,
11347             o2->ipIfStatsInWrongIPVersion);
11348         UPDATE_MIB(o1, ipIfStatsOutSwitchIPVersion,
11349             o2->ipIfStatsOutSwitchIPVersion);
11350         UPDATE_MIB(o1, ipIfStatsHCInReceives, o2->ipIfStatsHCInReceives);
11351         UPDATE_MIB(o1, ipIfStatsHCInOctets, o2->ipIfStatsHCInOctets);
11352         UPDATE_MIB(o1, ipIfStatsHCInForwDatagrams,
11353             o2->ipIfStatsHCInForwDatagrams);
11354         UPDATE_MIB(o1, ipIfStatsHCInDelivers, o2->ipIfStatsHCInDelivers);
11355         UPDATE_MIB(o1, ipIfStatsHCOutRequests, o2->ipIfStatsHCOutRequests);
11356         UPDATE_MIB(o1, ipIfStatsHCOutForwDatagrams,
11357             o2->ipIfStatsHCOutForwDatagrams);
11358         UPDATE_MIB(o1, ipIfStatsOutFragReqds, o2->ipIfStatsOutFragReqds);
11359         UPDATE_MIB(o1, ipIfStatsHCOutTransmits, o2->ipIfStatsHCOutTransmits);
11360         UPDATE_MIB(o1, ipIfStatsHCOutOctets, o2->ipIfStatsHCOutOctets);
11361         UPDATE_MIB(o1, ipIfStatsHCInMcastPkts, o2->ipIfStatsHCInMcastPkts);
11362         UPDATE_MIB(o1, ipIfStatsHCInMcastOctets, o2->ipIfStatsHCInMcastOctets);
11363         UPDATE_MIB(o1, ipIfStatsHCOutMcastPkts, o2->ipIfStatsHCOutMcastPkts);
11364         UPDATE_MIB(o1, ipIfStatsHCOutMcastOctets,
11365             o2->ipIfStatsHCOutMcastOctets);
11366         UPDATE_MIB(o1, ipIfStatsHCInBcastPkts, o2->ipIfStatsHCInBcastPkts);
11367         UPDATE_MIB(o1, ipIfStatsHCOutBcastPkts, o2->ipIfStatsHCOutBcastPkts);
11368         UPDATE_MIB(o1, ipsecInSucceeded, o2->ipsecInSucceeded);
11369         UPDATE_MIB(o1, ipsecInFailed, o2->ipsecInFailed);
11370         UPDATE_MIB(o1, ipInCksumErrs, o2->ipInCksumErrs);
11371         UPDATE_MIB(o1, tcpInErrs, o2->tcpInErrs);
11372         UPDATE_MIB(o1, udpNoPorts, o2->udpNoPorts);
11373 }
11374 
11375 void
11376 ip_mib2_add_icmp6_stats(mib2_ipv6IfIcmpEntry_t *o1, mib2_ipv6IfIcmpEntry_t *o2)
11377 {
11378         UPDATE_MIB(o1, ipv6IfIcmpInMsgs, o2->ipv6IfIcmpInMsgs);
11379         UPDATE_MIB(o1, ipv6IfIcmpInErrors, o2->ipv6IfIcmpInErrors);
11380         UPDATE_MIB(o1, ipv6IfIcmpInDestUnreachs, o2->ipv6IfIcmpInDestUnreachs);
11381         UPDATE_MIB(o1, ipv6IfIcmpInAdminProhibs, o2->ipv6IfIcmpInAdminProhibs);
11382         UPDATE_MIB(o1, ipv6IfIcmpInTimeExcds, o2->ipv6IfIcmpInTimeExcds);
11383         UPDATE_MIB(o1, ipv6IfIcmpInParmProblems, o2->ipv6IfIcmpInParmProblems);
11384         UPDATE_MIB(o1, ipv6IfIcmpInPktTooBigs, o2->ipv6IfIcmpInPktTooBigs);
11385         UPDATE_MIB(o1, ipv6IfIcmpInEchos, o2->ipv6IfIcmpInEchos);
11386         UPDATE_MIB(o1, ipv6IfIcmpInEchoReplies, o2->ipv6IfIcmpInEchoReplies);
11387         UPDATE_MIB(o1, ipv6IfIcmpInRouterSolicits,
11388             o2->ipv6IfIcmpInRouterSolicits);
11389         UPDATE_MIB(o1, ipv6IfIcmpInRouterAdvertisements,
11390             o2->ipv6IfIcmpInRouterAdvertisements);
11391         UPDATE_MIB(o1, ipv6IfIcmpInNeighborSolicits,
11392             o2->ipv6IfIcmpInNeighborSolicits);
11393         UPDATE_MIB(o1, ipv6IfIcmpInNeighborAdvertisements,
11394             o2->ipv6IfIcmpInNeighborAdvertisements);
11395         UPDATE_MIB(o1, ipv6IfIcmpInRedirects, o2->ipv6IfIcmpInRedirects);
11396         UPDATE_MIB(o1, ipv6IfIcmpInGroupMembQueries,
11397             o2->ipv6IfIcmpInGroupMembQueries);
11398         UPDATE_MIB(o1, ipv6IfIcmpInGroupMembResponses,
11399             o2->ipv6IfIcmpInGroupMembResponses);
11400         UPDATE_MIB(o1, ipv6IfIcmpInGroupMembReductions,
11401             o2->ipv6IfIcmpInGroupMembReductions);
11402         UPDATE_MIB(o1, ipv6IfIcmpOutMsgs, o2->ipv6IfIcmpOutMsgs);
11403         UPDATE_MIB(o1, ipv6IfIcmpOutErrors, o2->ipv6IfIcmpOutErrors);
11404         UPDATE_MIB(o1, ipv6IfIcmpOutDestUnreachs,
11405             o2->ipv6IfIcmpOutDestUnreachs);
11406         UPDATE_MIB(o1, ipv6IfIcmpOutAdminProhibs,
11407             o2->ipv6IfIcmpOutAdminProhibs);
11408         UPDATE_MIB(o1, ipv6IfIcmpOutTimeExcds, o2->ipv6IfIcmpOutTimeExcds);
11409         UPDATE_MIB(o1, ipv6IfIcmpOutParmProblems,
11410             o2->ipv6IfIcmpOutParmProblems);
11411         UPDATE_MIB(o1, ipv6IfIcmpOutPktTooBigs, o2->ipv6IfIcmpOutPktTooBigs);
11412         UPDATE_MIB(o1, ipv6IfIcmpOutEchos, o2->ipv6IfIcmpOutEchos);
11413         UPDATE_MIB(o1, ipv6IfIcmpOutEchoReplies, o2->ipv6IfIcmpOutEchoReplies);
11414         UPDATE_MIB(o1, ipv6IfIcmpOutRouterSolicits,
11415             o2->ipv6IfIcmpOutRouterSolicits);
11416         UPDATE_MIB(o1, ipv6IfIcmpOutRouterAdvertisements,
11417             o2->ipv6IfIcmpOutRouterAdvertisements);
11418         UPDATE_MIB(o1, ipv6IfIcmpOutNeighborSolicits,
11419             o2->ipv6IfIcmpOutNeighborSolicits);
11420         UPDATE_MIB(o1, ipv6IfIcmpOutNeighborAdvertisements,
11421             o2->ipv6IfIcmpOutNeighborAdvertisements);
11422         UPDATE_MIB(o1, ipv6IfIcmpOutRedirects, o2->ipv6IfIcmpOutRedirects);
11423         UPDATE_MIB(o1, ipv6IfIcmpOutGroupMembQueries,
11424             o2->ipv6IfIcmpOutGroupMembQueries);
11425         UPDATE_MIB(o1, ipv6IfIcmpOutGroupMembResponses,
11426             o2->ipv6IfIcmpOutGroupMembResponses);
11427         UPDATE_MIB(o1, ipv6IfIcmpOutGroupMembReductions,
11428             o2->ipv6IfIcmpOutGroupMembReductions);
11429         UPDATE_MIB(o1, ipv6IfIcmpInOverflows, o2->ipv6IfIcmpInOverflows);
11430         UPDATE_MIB(o1, ipv6IfIcmpBadHoplimit, o2->ipv6IfIcmpBadHoplimit);
11431         UPDATE_MIB(o1, ipv6IfIcmpInBadNeighborAdvertisements,
11432             o2->ipv6IfIcmpInBadNeighborAdvertisements);
11433         UPDATE_MIB(o1, ipv6IfIcmpInBadNeighborSolicitations,
11434             o2->ipv6IfIcmpInBadNeighborSolicitations);
11435         UPDATE_MIB(o1, ipv6IfIcmpInBadRedirects, o2->ipv6IfIcmpInBadRedirects);
11436         UPDATE_MIB(o1, ipv6IfIcmpInGroupMembTotal,
11437             o2->ipv6IfIcmpInGroupMembTotal);
11438         UPDATE_MIB(o1, ipv6IfIcmpInGroupMembBadQueries,
11439             o2->ipv6IfIcmpInGroupMembBadQueries);
11440         UPDATE_MIB(o1, ipv6IfIcmpInGroupMembBadReports,
11441             o2->ipv6IfIcmpInGroupMembBadReports);
11442         UPDATE_MIB(o1, ipv6IfIcmpInGroupMembOurReports,
11443             o2->ipv6IfIcmpInGroupMembOurReports);
11444 }
11445 
11446 /*
11447  * Called before the options are updated to check if this packet will
11448  * be source routed from here.
11449  * This routine assumes that the options are well formed i.e. that they
11450  * have already been checked.
11451  */
11452 boolean_t
11453 ip_source_routed(ipha_t *ipha, ip_stack_t *ipst)
11454 {
11455         ipoptp_t        opts;
11456         uchar_t         *opt;
11457         uint8_t         optval;
11458         uint8_t         optlen;
11459         ipaddr_t        dst;
11460 
11461         if (IS_SIMPLE_IPH(ipha)) {
11462                 ip2dbg(("not source routed\n"));
11463                 return (B_FALSE);
11464         }
11465         dst = ipha->ipha_dst;
11466         for (optval = ipoptp_first(&opts, ipha);
11467             optval != IPOPT_EOL;
11468             optval = ipoptp_next(&opts)) {
11469                 ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
11470                 opt = opts.ipoptp_cur;
11471                 optlen = opts.ipoptp_len;
11472                 ip2dbg(("ip_source_routed: opt %d, len %d\n",
11473                     optval, optlen));
11474                 switch (optval) {
11475                         uint32_t off;
11476                 case IPOPT_SSRR:
11477                 case IPOPT_LSRR:
11478                         /*
11479                          * If dst is one of our addresses and there are some
11480                          * entries left in the source route return (true).
11481                          */
11482                         if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
11483                                 ip2dbg(("ip_source_routed: not next"
11484                                     " source route 0x%x\n",
11485                                     ntohl(dst)));
11486                                 return (B_FALSE);
11487                         }
11488                         off = opt[IPOPT_OFFSET];
11489                         off--;
11490                         if (optlen < IP_ADDR_LEN ||
11491                             off > optlen - IP_ADDR_LEN) {
11492                                 /* End of source route */
11493                                 ip1dbg(("ip_source_routed: end of SR\n"));
11494                                 return (B_FALSE);
11495                         }
11496                         return (B_TRUE);
11497                 }
11498         }
11499         ip2dbg(("not source routed\n"));
11500         return (B_FALSE);
11501 }
11502 
11503 /*
11504  * ip_unbind is called by the transports to remove a conn from
11505  * the fanout table.
11506  */
11507 void
11508 ip_unbind(conn_t *connp)
11509 {
11510 
11511         ASSERT(!MUTEX_HELD(&connp->conn_lock));
11512 
11513         if (is_system_labeled() && connp->conn_anon_port) {
11514                 (void) tsol_mlp_anon(crgetzone(connp->conn_cred),
11515                     connp->conn_mlp_type, connp->conn_proto,
11516                     ntohs(connp->conn_lport), B_FALSE);
11517                 connp->conn_anon_port = 0;
11518         }
11519         connp->conn_mlp_type = mlptSingle;
11520 
11521         ipcl_hash_remove(connp);
11522 }
11523 
11524 /*
11525  * Used for deciding the MSS size for the upper layer. Thus
11526  * we need to check the outbound policy values in the conn.
11527  */
11528 int
11529 conn_ipsec_length(conn_t *connp)
11530 {
11531         ipsec_latch_t *ipl;
11532 
11533         ipl = connp->conn_latch;
11534         if (ipl == NULL)
11535                 return (0);
11536 
11537         if (connp->conn_ixa->ixa_ipsec_policy == NULL)
11538                 return (0);
11539 
11540         return (connp->conn_ixa->ixa_ipsec_policy->ipsp_act->ipa_ovhd);
11541 }
11542 
11543 /*
11544  * Returns an estimate of the IPsec headers size. This is used if
11545  * we don't want to call into IPsec to get the exact size.
11546  */
11547 int
11548 ipsec_out_extra_length(ip_xmit_attr_t *ixa)
11549 {
11550         ipsec_action_t *a;
11551 
11552         if (!(ixa->ixa_flags & IXAF_IPSEC_SECURE))
11553                 return (0);
11554 
11555         a = ixa->ixa_ipsec_action;
11556         if (a == NULL) {
11557                 ASSERT(ixa->ixa_ipsec_policy != NULL);
11558                 a = ixa->ixa_ipsec_policy->ipsp_act;
11559         }
11560         ASSERT(a != NULL);
11561 
11562         return (a->ipa_ovhd);
11563 }
11564 
11565 /*
11566  * If there are any source route options, return the true final
11567  * destination. Otherwise, return the destination.
11568  */
11569 ipaddr_t
11570 ip_get_dst(ipha_t *ipha)
11571 {
11572         ipoptp_t        opts;
11573         uchar_t         *opt;
11574         uint8_t         optval;
11575         uint8_t         optlen;
11576         ipaddr_t        dst;
11577         uint32_t off;
11578 
11579         dst = ipha->ipha_dst;
11580 
11581         if (IS_SIMPLE_IPH(ipha))
11582                 return (dst);
11583 
11584         for (optval = ipoptp_first(&opts, ipha);
11585             optval != IPOPT_EOL;
11586             optval = ipoptp_next(&opts)) {
11587                 opt = opts.ipoptp_cur;
11588                 optlen = opts.ipoptp_len;
11589                 ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
11590                 switch (optval) {
11591                 case IPOPT_SSRR:
11592                 case IPOPT_LSRR:
11593                         off = opt[IPOPT_OFFSET];
11594                         /*
11595                          * If one of the conditions is true, it means
11596                          * end of options and dst already has the right
11597                          * value.
11598                          */
11599                         if (!(optlen < IP_ADDR_LEN || off > optlen - 3)) {
11600                                 off = optlen - IP_ADDR_LEN;
11601                                 bcopy(&opt[off], &dst, IP_ADDR_LEN);
11602                         }
11603                         return (dst);
11604                 default:
11605                         break;
11606                 }
11607         }
11608 
11609         return (dst);
11610 }
11611 
11612 /*
11613  * Outbound IP fragmentation routine.
11614  * Assumes the caller has checked whether or not fragmentation should
11615  * be allowed. Here we copy the DF bit from the header to all the generated
11616  * fragments.
11617  */
11618 int
11619 ip_fragment_v4(mblk_t *mp_orig, nce_t *nce, iaflags_t ixaflags,
11620     uint_t pkt_len, uint32_t max_frag, uint32_t xmit_hint, zoneid_t szone,
11621     zoneid_t nolzid, pfirepostfrag_t postfragfn, uintptr_t *ixa_cookie)
11622 {
11623         int             i1;
11624         int             hdr_len;
11625         mblk_t          *hdr_mp;
11626         ipha_t          *ipha;
11627         int             ip_data_end;
11628         int             len;
11629         mblk_t          *mp = mp_orig;
11630         int             offset;
11631         ill_t           *ill = nce->nce_ill;
11632         ip_stack_t      *ipst = ill->ill_ipst;
11633         mblk_t          *carve_mp;
11634         uint32_t        frag_flag;
11635         uint_t          priority = mp->b_band;
11636         int             error = 0;
11637 
11638         BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragReqds);
11639 
11640         if (pkt_len != msgdsize(mp)) {
11641                 ip0dbg(("Packet length mismatch: %d, %ld\n",
11642                     pkt_len, msgdsize(mp)));
11643                 freemsg(mp);
11644                 return (EINVAL);
11645         }
11646 
11647         if (max_frag == 0) {
11648                 ip1dbg(("ip_fragment_v4: max_frag is zero. Dropping packet\n"));
11649                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11650                 ip_drop_output("FragFails: zero max_frag", mp, ill);
11651                 freemsg(mp);
11652                 return (EINVAL);
11653         }
11654 
11655         ASSERT(MBLKL(mp) >= sizeof (ipha_t));
11656         ipha = (ipha_t *)mp->b_rptr;
11657         ASSERT(ntohs(ipha->ipha_length) == pkt_len);
11658         frag_flag = ntohs(ipha->ipha_fragment_offset_and_flags) & IPH_DF;
11659 
11660         /*
11661          * Establish the starting offset.  May not be zero if we are fragging
11662          * a fragment that is being forwarded.
11663          */
11664         offset = ntohs(ipha->ipha_fragment_offset_and_flags) & IPH_OFFSET;
11665 
11666         /* TODO why is this test needed? */
11667         if (((max_frag - ntohs(ipha->ipha_length)) & ~7) < 8) {
11668                 /* TODO: notify ulp somehow */
11669                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11670                 ip_drop_output("FragFails: bad starting offset", mp, ill);
11671                 freemsg(mp);
11672                 return (EINVAL);
11673         }
11674 
11675         hdr_len = IPH_HDR_LENGTH(ipha);
11676         ipha->ipha_hdr_checksum = 0;
11677 
11678         /*
11679          * Establish the number of bytes maximum per frag, after putting
11680          * in the header.
11681          */
11682         len = (max_frag - hdr_len) & ~7;
11683 
11684         /* Get a copy of the header for the trailing frags */
11685         hdr_mp = ip_fragment_copyhdr((uchar_t *)ipha, hdr_len, offset, ipst,
11686             mp);
11687         if (hdr_mp == NULL) {
11688                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11689                 ip_drop_output("FragFails: no hdr_mp", mp, ill);
11690                 freemsg(mp);
11691                 return (ENOBUFS);
11692         }
11693 
11694         /* Store the starting offset, with the MoreFrags flag. */
11695         i1 = offset | IPH_MF | frag_flag;
11696         ipha->ipha_fragment_offset_and_flags = htons((uint16_t)i1);
11697 
11698         /* Establish the ending byte offset, based on the starting offset. */
11699         offset <<= 3;
11700         ip_data_end = offset + ntohs(ipha->ipha_length) - hdr_len;
11701 
11702         /* Store the length of the first fragment in the IP header. */
11703         i1 = len + hdr_len;
11704         ASSERT(i1 <= IP_MAXPACKET);
11705         ipha->ipha_length = htons((uint16_t)i1);
11706 
11707         /*
11708          * Compute the IP header checksum for the first frag.  We have to
11709          * watch out that we stop at the end of the header.
11710          */
11711         ipha->ipha_hdr_checksum = ip_csum_hdr(ipha);
11712 
11713         /*
11714          * Now carve off the first frag.  Note that this will include the
11715          * original IP header.
11716          */
11717         if (!(mp = ip_carve_mp(&mp_orig, i1))) {
11718                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11719                 ip_drop_output("FragFails: could not carve mp", mp_orig, ill);
11720                 freeb(hdr_mp);
11721                 freemsg(mp_orig);
11722                 return (ENOBUFS);
11723         }
11724 
11725         BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragCreates);
11726 
11727         error = postfragfn(mp, nce, ixaflags, i1, xmit_hint, szone, nolzid,
11728             ixa_cookie);
11729         if (error != 0 && error != EWOULDBLOCK) {
11730                 /* No point in sending the other fragments */
11731                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11732                 ip_drop_output("FragFails: postfragfn failed", mp_orig, ill);
11733                 freeb(hdr_mp);
11734                 freemsg(mp_orig);
11735                 return (error);
11736         }
11737 
11738         /* No need to redo state machine in loop */
11739         ixaflags &= ~IXAF_REACH_CONF;
11740 
11741         /* Advance the offset to the second frag starting point. */
11742         offset += len;
11743         /*
11744          * Update hdr_len from the copied header - there might be less options
11745          * in the later fragments.
11746          */
11747         hdr_len = IPH_HDR_LENGTH(hdr_mp->b_rptr);
11748         /* Loop until done. */
11749         for (;;) {
11750                 uint16_t        offset_and_flags;
11751                 uint16_t        ip_len;
11752 
11753                 if (ip_data_end - offset > len) {
11754                         /*
11755                          * Carve off the appropriate amount from the original
11756                          * datagram.
11757                          */
11758                         if (!(carve_mp = ip_carve_mp(&mp_orig, len))) {
11759                                 mp = NULL;
11760                                 break;
11761                         }
11762                         /*
11763                          * More frags after this one.  Get another copy
11764                          * of the header.
11765                          */
11766                         if (carve_mp->b_datap->db_ref == 1 &&
11767                             hdr_mp->b_wptr - hdr_mp->b_rptr <
11768                             carve_mp->b_rptr - carve_mp->b_datap->db_base) {
11769                                 /* Inline IP header */
11770                                 carve_mp->b_rptr -= hdr_mp->b_wptr -
11771                                     hdr_mp->b_rptr;
11772                                 bcopy(hdr_mp->b_rptr, carve_mp->b_rptr,
11773                                     hdr_mp->b_wptr - hdr_mp->b_rptr);
11774                                 mp = carve_mp;
11775                         } else {
11776                                 if (!(mp = copyb(hdr_mp))) {
11777                                         freemsg(carve_mp);
11778                                         break;
11779                                 }
11780                                 /* Get priority marking, if any. */
11781                                 mp->b_band = priority;
11782                                 mp->b_cont = carve_mp;
11783                         }
11784                         ipha = (ipha_t *)mp->b_rptr;
11785                         offset_and_flags = IPH_MF;
11786                 } else {
11787                         /*
11788                          * Last frag.  Consume the header. Set len to
11789                          * the length of this last piece.
11790                          */
11791                         len = ip_data_end - offset;
11792 
11793                         /*
11794                          * Carve off the appropriate amount from the original
11795                          * datagram.
11796                          */
11797                         if (!(carve_mp = ip_carve_mp(&mp_orig, len))) {
11798                                 mp = NULL;
11799                                 break;
11800                         }
11801                         if (carve_mp->b_datap->db_ref == 1 &&
11802                             hdr_mp->b_wptr - hdr_mp->b_rptr <
11803                             carve_mp->b_rptr - carve_mp->b_datap->db_base) {
11804                                 /* Inline IP header */
11805                                 carve_mp->b_rptr -= hdr_mp->b_wptr -
11806                                     hdr_mp->b_rptr;
11807                                 bcopy(hdr_mp->b_rptr, carve_mp->b_rptr,
11808                                     hdr_mp->b_wptr - hdr_mp->b_rptr);
11809                                 mp = carve_mp;
11810                                 freeb(hdr_mp);
11811                                 hdr_mp = mp;
11812                         } else {
11813                                 mp = hdr_mp;
11814                                 /* Get priority marking, if any. */
11815                                 mp->b_band = priority;
11816                                 mp->b_cont = carve_mp;
11817                         }
11818                         ipha = (ipha_t *)mp->b_rptr;
11819                         /* A frag of a frag might have IPH_MF non-zero */
11820                         offset_and_flags =
11821                             ntohs(ipha->ipha_fragment_offset_and_flags) &
11822                             IPH_MF;
11823                 }
11824                 offset_and_flags |= (uint16_t)(offset >> 3);
11825                 offset_and_flags |= (uint16_t)frag_flag;
11826                 /* Store the offset and flags in the IP header. */
11827                 ipha->ipha_fragment_offset_and_flags = htons(offset_and_flags);
11828 
11829                 /* Store the length in the IP header. */
11830                 ip_len = (uint16_t)(len + hdr_len);
11831                 ipha->ipha_length = htons(ip_len);
11832 
11833                 /*
11834                  * Set the IP header checksum.  Note that mp is just
11835                  * the header, so this is easy to pass to ip_csum.
11836                  */
11837                 ipha->ipha_hdr_checksum = ip_csum_hdr(ipha);
11838 
11839                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragCreates);
11840 
11841                 error = postfragfn(mp, nce, ixaflags, ip_len, xmit_hint, szone,
11842                     nolzid, ixa_cookie);
11843                 /* All done if we just consumed the hdr_mp. */
11844                 if (mp == hdr_mp) {
11845                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragOKs);
11846                         return (error);
11847                 }
11848                 if (error != 0 && error != EWOULDBLOCK) {
11849                         DTRACE_PROBE2(ip__xmit__frag__fail, ill_t *, ill,
11850                             mblk_t *, hdr_mp);
11851                         /* No point in sending the other fragments */
11852                         break;
11853                 }
11854 
11855                 /* Otherwise, advance and loop. */
11856                 offset += len;
11857         }
11858         /* Clean up following allocation failure. */
11859         BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11860         ip_drop_output("FragFails: loop ended", NULL, ill);
11861         if (mp != hdr_mp)
11862                 freeb(hdr_mp);
11863         if (mp != mp_orig)
11864                 freemsg(mp_orig);
11865         return (error);
11866 }
11867 
11868 /*
11869  * Copy the header plus those options which have the copy bit set
11870  */
11871 static mblk_t *
11872 ip_fragment_copyhdr(uchar_t *rptr, int hdr_len, int offset, ip_stack_t *ipst,
11873     mblk_t *src)
11874 {
11875         mblk_t  *mp;
11876         uchar_t *up;
11877 
11878         /*
11879          * Quick check if we need to look for options without the copy bit
11880          * set
11881          */
11882         mp = allocb_tmpl(ipst->ips_ip_wroff_extra + hdr_len, src);
11883         if (!mp)
11884                 return (mp);
11885         mp->b_rptr += ipst->ips_ip_wroff_extra;
11886         if (hdr_len == IP_SIMPLE_HDR_LENGTH || offset != 0) {
11887                 bcopy(rptr, mp->b_rptr, hdr_len);
11888                 mp->b_wptr += hdr_len + ipst->ips_ip_wroff_extra;
11889                 return (mp);
11890         }
11891         up  = mp->b_rptr;
11892         bcopy(rptr, up, IP_SIMPLE_HDR_LENGTH);
11893         up += IP_SIMPLE_HDR_LENGTH;
11894         rptr += IP_SIMPLE_HDR_LENGTH;
11895         hdr_len -= IP_SIMPLE_HDR_LENGTH;
11896         while (hdr_len > 0) {
11897                 uint32_t optval;
11898                 uint32_t optlen;
11899 
11900                 optval = *rptr;
11901                 if (optval == IPOPT_EOL)
11902                         break;
11903                 if (optval == IPOPT_NOP)
11904                         optlen = 1;
11905                 else
11906                         optlen = rptr[1];
11907                 if (optval & IPOPT_COPY) {
11908                         bcopy(rptr, up, optlen);
11909                         up += optlen;
11910                 }
11911                 rptr += optlen;
11912                 hdr_len -= optlen;
11913         }
11914         /*
11915          * Make sure that we drop an even number of words by filling
11916          * with EOL to the next word boundary.
11917          */
11918         for (hdr_len = up - (mp->b_rptr + IP_SIMPLE_HDR_LENGTH);
11919             hdr_len & 0x3; hdr_len++)
11920                 *up++ = IPOPT_EOL;
11921         mp->b_wptr = up;
11922         /* Update header length */
11923         mp->b_rptr[0] = (uint8_t)((IP_VERSION << 4) | ((up - mp->b_rptr) >> 2));
11924         return (mp);
11925 }
11926 
11927 /*
11928  * Update any source route, record route, or timestamp options when
11929  * sending a packet back to ourselves.
11930  * Check that we are at end of strict source route.
11931  * The options have been sanity checked by ip_output_options().
11932  */
11933 void
11934 ip_output_local_options(ipha_t *ipha, ip_stack_t *ipst)
11935 {
11936         ipoptp_t        opts;
11937         uchar_t         *opt;
11938         uint8_t         optval;
11939         uint8_t         optlen;
11940         ipaddr_t        dst;
11941         uint32_t        ts;
11942         timestruc_t     now;
11943 
11944         for (optval = ipoptp_first(&opts, ipha);
11945             optval != IPOPT_EOL;
11946             optval = ipoptp_next(&opts)) {
11947                 opt = opts.ipoptp_cur;
11948                 optlen = opts.ipoptp_len;
11949                 ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
11950                 switch (optval) {
11951                         uint32_t off;
11952                 case IPOPT_SSRR:
11953                 case IPOPT_LSRR:
11954                         off = opt[IPOPT_OFFSET];
11955                         off--;
11956                         if (optlen < IP_ADDR_LEN ||
11957                             off > optlen - IP_ADDR_LEN) {
11958                                 /* End of source route */
11959                                 break;
11960                         }
11961                         /*
11962                          * This will only happen if two consecutive entries
11963                          * in the source route contains our address or if
11964                          * it is a packet with a loose source route which
11965                          * reaches us before consuming the whole source route
11966                          */
11967 
11968                         if (optval == IPOPT_SSRR) {
11969                                 return;
11970                         }
11971                         /*
11972                          * Hack: instead of dropping the packet truncate the
11973                          * source route to what has been used by filling the
11974                          * rest with IPOPT_NOP.
11975                          */
11976                         opt[IPOPT_OLEN] = (uint8_t)off;
11977                         while (off < optlen) {
11978                                 opt[off++] = IPOPT_NOP;
11979                         }
11980                         break;
11981                 case IPOPT_RR:
11982                         off = opt[IPOPT_OFFSET];
11983                         off--;
11984                         if (optlen < IP_ADDR_LEN ||
11985                             off > optlen - IP_ADDR_LEN) {
11986                                 /* No more room - ignore */
11987                                 ip1dbg((
11988                                     "ip_output_local_options: end of RR\n"));
11989                                 break;
11990                         }
11991                         dst = htonl(INADDR_LOOPBACK);
11992                         bcopy(&dst, (char *)opt + off, IP_ADDR_LEN);
11993                         opt[IPOPT_OFFSET] += IP_ADDR_LEN;
11994                         break;
11995                 case IPOPT_TS:
11996                         /* Insert timestamp if there is romm */
11997                         switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
11998                         case IPOPT_TS_TSONLY:
11999                                 off = IPOPT_TS_TIMELEN;
12000                                 break;
12001                         case IPOPT_TS_PRESPEC:
12002                         case IPOPT_TS_PRESPEC_RFC791:
12003                                 /* Verify that the address matched */
12004                                 off = opt[IPOPT_OFFSET] - 1;
12005                                 bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
12006                                 if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
12007                                         /* Not for us */
12008                                         break;
12009                                 }
12010                                 /* FALLTHROUGH */
12011                         case IPOPT_TS_TSANDADDR:
12012                                 off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
12013                                 break;
12014                         default:
12015                                 /*
12016                                  * ip_*put_options should have already
12017                                  * dropped this packet.
12018                                  */
12019                                 cmn_err(CE_PANIC, "ip_output_local_options: "
12020                                     "unknown IT - bug in ip_output_options?\n");
12021                                 return; /* Keep "lint" happy */
12022                         }
12023                         if (opt[IPOPT_OFFSET] - 1 + off > optlen) {
12024                                 /* Increase overflow counter */
12025                                 off = (opt[IPOPT_POS_OV_FLG] >> 4) + 1;
12026                                 opt[IPOPT_POS_OV_FLG] = (uint8_t)
12027                                     (opt[IPOPT_POS_OV_FLG] & 0x0F) |
12028                                     (off << 4);
12029                                 break;
12030                         }
12031                         off = opt[IPOPT_OFFSET] - 1;
12032                         switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
12033                         case IPOPT_TS_PRESPEC:
12034                         case IPOPT_TS_PRESPEC_RFC791:
12035                         case IPOPT_TS_TSANDADDR:
12036                                 dst = htonl(INADDR_LOOPBACK);
12037                                 bcopy(&dst, (char *)opt + off, IP_ADDR_LEN);
12038                                 opt[IPOPT_OFFSET] += IP_ADDR_LEN;
12039                                 /* FALLTHROUGH */
12040                         case IPOPT_TS_TSONLY:
12041                                 off = opt[IPOPT_OFFSET] - 1;
12042                                 /* Compute # of milliseconds since midnight */
12043                                 gethrestime(&now);
12044                                 ts = (now.tv_sec % (24 * 60 * 60)) * 1000 +
12045                                     NSEC2MSEC(now.tv_nsec);
12046                                 bcopy(&ts, (char *)opt + off, IPOPT_TS_TIMELEN);
12047                                 opt[IPOPT_OFFSET] += IPOPT_TS_TIMELEN;
12048                                 break;
12049                         }
12050                         break;
12051                 }
12052         }
12053 }
12054 
12055 /*
12056  * Prepend an M_DATA fastpath header, and if none present prepend a
12057  * DL_UNITDATA_REQ. Frees the mblk on failure.
12058  *
12059  * nce_dlur_mp and nce_fp_mp can not disappear once they have been set.
12060  * If there is a change to them, the nce will be deleted (condemned) and
12061  * a new nce_t will be created when packets are sent. Thus we need no locks
12062  * to access those fields.
12063  *
12064  * We preserve b_band to support IPQoS. If a DL_UNITDATA_REQ is prepended
12065  * we place b_band in dl_priority.dl_max.
12066  */
12067 static mblk_t *
12068 ip_xmit_attach_llhdr(mblk_t *mp, nce_t *nce)
12069 {
12070         uint_t  hlen;
12071         mblk_t *mp1;
12072         uint_t  priority;
12073         uchar_t *rptr;
12074 
12075         rptr = mp->b_rptr;
12076 
12077         ASSERT(DB_TYPE(mp) == M_DATA);
12078         priority = mp->b_band;
12079 
12080         ASSERT(nce != NULL);
12081         if ((mp1 = nce->nce_fp_mp) != NULL) {
12082                 hlen = MBLKL(mp1);
12083                 /*
12084                  * Check if we have enough room to prepend fastpath
12085                  * header
12086                  */
12087                 if (hlen != 0 && (rptr - mp->b_datap->db_base) >= hlen) {
12088                         rptr -= hlen;
12089                         bcopy(mp1->b_rptr, rptr, hlen);
12090                         /*
12091                          * Set the b_rptr to the start of the link layer
12092                          * header
12093                          */
12094                         mp->b_rptr = rptr;
12095                         return (mp);
12096                 }
12097                 mp1 = copyb(mp1);
12098                 if (mp1 == NULL) {
12099                         ill_t *ill = nce->nce_ill;
12100 
12101                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
12102                         ip_drop_output("ipIfStatsOutDiscards", mp, ill);
12103                         freemsg(mp);
12104                         return (NULL);
12105                 }
12106                 mp1->b_band = priority;
12107                 mp1->b_cont = mp;
12108                 DB_CKSUMSTART(mp1) = DB_CKSUMSTART(mp);
12109                 DB_CKSUMSTUFF(mp1) = DB_CKSUMSTUFF(mp);
12110                 DB_CKSUMEND(mp1) = DB_CKSUMEND(mp);
12111                 DB_CKSUMFLAGS(mp1) = DB_CKSUMFLAGS(mp);
12112                 DB_LSOMSS(mp1) = DB_LSOMSS(mp);
12113                 DTRACE_PROBE1(ip__xmit__copyb, (mblk_t *), mp1);
12114                 /*
12115                  * XXX disable ICK_VALID and compute checksum
12116                  * here; can happen if nce_fp_mp changes and
12117                  * it can't be copied now due to insufficient
12118                  * space. (unlikely, fp mp can change, but it
12119                  * does not increase in length)
12120                  */
12121                 return (mp1);
12122         }
12123         mp1 = copyb(nce->nce_dlur_mp);
12124 
12125         if (mp1 == NULL) {
12126                 ill_t *ill = nce->nce_ill;
12127 
12128                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
12129                 ip_drop_output("ipIfStatsOutDiscards", mp, ill);
12130                 freemsg(mp);
12131                 return (NULL);
12132         }
12133         mp1->b_cont = mp;
12134         if (priority != 0) {
12135                 mp1->b_band = priority;
12136                 ((dl_unitdata_req_t *)(mp1->b_rptr))->dl_priority.dl_max =
12137                     priority;
12138         }
12139         return (mp1);
12140 }
12141 
12142 /*
12143  * Finish the outbound IPsec processing. This function is called from
12144  * ipsec_out_process() if the IPsec packet was processed
12145  * synchronously, or from {ah,esp}_kcf_callback_outbound() if it was processed
12146  * asynchronously.
12147  *
12148  * This is common to IPv4 and IPv6.
12149  */
12150 int
12151 ip_output_post_ipsec(mblk_t *mp, ip_xmit_attr_t *ixa)
12152 {
12153         iaflags_t       ixaflags = ixa->ixa_flags;
12154         uint_t          pktlen;
12155 
12156 
12157         /* AH/ESP don't update ixa_pktlen when they modify the packet */
12158         if (ixaflags & IXAF_IS_IPV4) {
12159                 ipha_t          *ipha = (ipha_t *)mp->b_rptr;
12160 
12161                 ASSERT(IPH_HDR_VERSION(ipha) == IPV4_VERSION);
12162                 pktlen = ntohs(ipha->ipha_length);
12163         } else {
12164                 ip6_t           *ip6h = (ip6_t *)mp->b_rptr;
12165 
12166                 ASSERT(IPH_HDR_VERSION(mp->b_rptr) == IPV6_VERSION);
12167                 pktlen = ntohs(ip6h->ip6_plen) + IPV6_HDR_LEN;
12168         }
12169 
12170         /*
12171          * We release any hard reference on the SAs here to make
12172          * sure the SAs can be garbage collected. ipsr_sa has a soft reference
12173          * on the SAs.
12174          * If in the future we want the hard latching of the SAs in the
12175          * ip_xmit_attr_t then we should remove this.
12176          */
12177         if (ixa->ixa_ipsec_esp_sa != NULL) {
12178                 IPSA_REFRELE(ixa->ixa_ipsec_esp_sa);
12179                 ixa->ixa_ipsec_esp_sa = NULL;
12180         }
12181         if (ixa->ixa_ipsec_ah_sa != NULL) {
12182                 IPSA_REFRELE(ixa->ixa_ipsec_ah_sa);
12183                 ixa->ixa_ipsec_ah_sa = NULL;
12184         }
12185 
12186         /* Do we need to fragment? */
12187         if ((ixa->ixa_flags & IXAF_IPV6_ADD_FRAGHDR) ||
12188             pktlen > ixa->ixa_fragsize) {
12189                 if (ixaflags & IXAF_IS_IPV4) {
12190                         ASSERT(!(ixa->ixa_flags & IXAF_IPV6_ADD_FRAGHDR));
12191                         /*
12192                          * We check for the DF case in ipsec_out_process
12193                          * hence this only handles the non-DF case.
12194                          */
12195                         return (ip_fragment_v4(mp, ixa->ixa_nce, ixa->ixa_flags,
12196                             pktlen, ixa->ixa_fragsize,
12197                             ixa->ixa_xmit_hint, ixa->ixa_zoneid,
12198                             ixa->ixa_no_loop_zoneid, ixa->ixa_postfragfn,
12199                             &ixa->ixa_cookie));
12200                 } else {
12201                         mp = ip_fraghdr_add_v6(mp, ixa->ixa_ident, ixa);
12202                         if (mp == NULL) {
12203                                 /* MIB and ip_drop_output already done */
12204                                 return (ENOMEM);
12205                         }
12206                         pktlen += sizeof (ip6_frag_t);
12207                         if (pktlen > ixa->ixa_fragsize) {
12208                                 return (ip_fragment_v6(mp, ixa->ixa_nce,
12209                                     ixa->ixa_flags, pktlen,
12210                                     ixa->ixa_fragsize, ixa->ixa_xmit_hint,
12211                                     ixa->ixa_zoneid, ixa->ixa_no_loop_zoneid,
12212                                     ixa->ixa_postfragfn, &ixa->ixa_cookie));
12213                         }
12214                 }
12215         }
12216         return ((ixa->ixa_postfragfn)(mp, ixa->ixa_nce, ixa->ixa_flags,
12217             pktlen, ixa->ixa_xmit_hint, ixa->ixa_zoneid,
12218             ixa->ixa_no_loop_zoneid, NULL));
12219 }
12220 
12221 /*
12222  * Finish the inbound IPsec processing. This function is called from
12223  * ipsec_out_process() if the IPsec packet was processed
12224  * synchronously, or from {ah,esp}_kcf_callback_outbound() if it was processed
12225  * asynchronously.
12226  *
12227  * This is common to IPv4 and IPv6.
12228  */
12229 void
12230 ip_input_post_ipsec(mblk_t *mp, ip_recv_attr_t *ira)
12231 {
12232         iaflags_t       iraflags = ira->ira_flags;
12233 
12234         /* Length might have changed */
12235         if (iraflags & IRAF_IS_IPV4) {
12236                 ipha_t          *ipha = (ipha_t *)mp->b_rptr;
12237 
12238                 ASSERT(IPH_HDR_VERSION(ipha) == IPV4_VERSION);
12239                 ira->ira_pktlen = ntohs(ipha->ipha_length);
12240                 ira->ira_ip_hdr_length = IPH_HDR_LENGTH(ipha);
12241                 ira->ira_protocol = ipha->ipha_protocol;
12242 
12243                 ip_fanout_v4(mp, ipha, ira);
12244         } else {
12245                 ip6_t           *ip6h = (ip6_t *)mp->b_rptr;
12246                 uint8_t         *nexthdrp;
12247 
12248                 ASSERT(IPH_HDR_VERSION(mp->b_rptr) == IPV6_VERSION);
12249                 ira->ira_pktlen = ntohs(ip6h->ip6_plen) + IPV6_HDR_LEN;
12250                 if (!ip_hdr_length_nexthdr_v6(mp, ip6h, &ira->ira_ip_hdr_length,
12251                     &nexthdrp)) {
12252                         /* Malformed packet */
12253                         BUMP_MIB(ira->ira_ill->ill_ip_mib, ipIfStatsInDiscards);
12254                         ip_drop_input("ipIfStatsInDiscards", mp, ira->ira_ill);
12255                         freemsg(mp);
12256                         return;
12257                 }
12258                 ira->ira_protocol = *nexthdrp;
12259                 ip_fanout_v6(mp, ip6h, ira);
12260         }
12261 }
12262 
12263 /*
12264  * Select which AH & ESP SA's to use (if any) for the outbound packet.
12265  *
12266  * If this function returns B_TRUE, the requested SA's have been filled
12267  * into the ixa_ipsec_*_sa pointers.
12268  *
12269  * If the function returns B_FALSE, the packet has been "consumed", most
12270  * likely by an ACQUIRE sent up via PF_KEY to a key management daemon.
12271  *
12272  * The SA references created by the protocol-specific "select"
12273  * function will be released in ip_output_post_ipsec.
12274  */
12275 static boolean_t
12276 ipsec_out_select_sa(mblk_t *mp, ip_xmit_attr_t *ixa)
12277 {
12278         boolean_t need_ah_acquire = B_FALSE, need_esp_acquire = B_FALSE;
12279         ipsec_policy_t *pp;
12280         ipsec_action_t *ap;
12281 
12282         ASSERT(ixa->ixa_flags & IXAF_IPSEC_SECURE);
12283         ASSERT((ixa->ixa_ipsec_policy != NULL) ||
12284             (ixa->ixa_ipsec_action != NULL));
12285 
12286         ap = ixa->ixa_ipsec_action;
12287         if (ap == NULL) {
12288                 pp = ixa->ixa_ipsec_policy;
12289                 ASSERT(pp != NULL);
12290                 ap = pp->ipsp_act;
12291                 ASSERT(ap != NULL);
12292         }
12293 
12294         /*
12295          * We have an action.  now, let's select SA's.
12296          * A side effect of setting ixa_ipsec_*_sa is that it will
12297          * be cached in the conn_t.
12298          */
12299         if (ap->ipa_want_esp) {
12300                 if (ixa->ixa_ipsec_esp_sa == NULL) {
12301                         need_esp_acquire = !ipsec_outbound_sa(mp, ixa,
12302                             IPPROTO_ESP);
12303                 }
12304                 ASSERT(need_esp_acquire || ixa->ixa_ipsec_esp_sa != NULL);
12305         }
12306 
12307         if (ap->ipa_want_ah) {
12308                 if (ixa->ixa_ipsec_ah_sa == NULL) {
12309                         need_ah_acquire = !ipsec_outbound_sa(mp, ixa,
12310                             IPPROTO_AH);
12311                 }
12312                 ASSERT(need_ah_acquire || ixa->ixa_ipsec_ah_sa != NULL);
12313                 /*
12314                  * The ESP and AH processing order needs to be preserved
12315                  * when both protocols are required (ESP should be applied
12316                  * before AH for an outbound packet). Force an ESP ACQUIRE
12317                  * when both ESP and AH are required, and an AH ACQUIRE
12318                  * is needed.
12319                  */
12320                 if (ap->ipa_want_esp && need_ah_acquire)
12321                         need_esp_acquire = B_TRUE;
12322         }
12323 
12324         /*
12325          * Send an ACQUIRE (extended, regular, or both) if we need one.
12326          * Release SAs that got referenced, but will not be used until we
12327          * acquire _all_ of the SAs we need.
12328          */
12329         if (need_ah_acquire || need_esp_acquire) {
12330                 if (ixa->ixa_ipsec_ah_sa != NULL) {
12331                         IPSA_REFRELE(ixa->ixa_ipsec_ah_sa);
12332                         ixa->ixa_ipsec_ah_sa = NULL;
12333                 }
12334                 if (ixa->ixa_ipsec_esp_sa != NULL) {
12335                         IPSA_REFRELE(ixa->ixa_ipsec_esp_sa);
12336                         ixa->ixa_ipsec_esp_sa = NULL;
12337                 }
12338 
12339                 sadb_acquire(mp, ixa, need_ah_acquire, need_esp_acquire);
12340                 return (B_FALSE);
12341         }
12342 
12343         return (B_TRUE);
12344 }
12345 
12346 /*
12347  * Handle IPsec output processing.
12348  * This function is only entered once for a given packet.
12349  * We try to do things synchronously, but if we need to have user-level
12350  * set up SAs, or ESP or AH uses asynchronous kEF, then the operation
12351  * will be completed
12352  *  - when the SAs are added in esp_add_sa_finish/ah_add_sa_finish
12353  *  - when asynchronous ESP is done it will do AH
12354  *
12355  * In all cases we come back in ip_output_post_ipsec() to fragment and
12356  * send out the packet.
12357  */
12358 int
12359 ipsec_out_process(mblk_t *mp, ip_xmit_attr_t *ixa)
12360 {
12361         ill_t           *ill = ixa->ixa_nce->nce_ill;
12362         ip_stack_t      *ipst = ixa->ixa_ipst;
12363         ipsec_stack_t   *ipss;
12364         ipsec_policy_t  *pp;
12365         ipsec_action_t  *ap;
12366 
12367         ASSERT(ixa->ixa_flags & IXAF_IPSEC_SECURE);
12368 
12369         ASSERT((ixa->ixa_ipsec_policy != NULL) ||
12370             (ixa->ixa_ipsec_action != NULL));
12371 
12372         ipss = ipst->ips_netstack->netstack_ipsec;
12373         if (!ipsec_loaded(ipss)) {
12374                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
12375                 ip_drop_packet(mp, B_TRUE, ill,
12376                     DROPPER(ipss, ipds_ip_ipsec_not_loaded),
12377                     &ipss->ipsec_dropper);
12378                 return (ENOTSUP);
12379         }
12380 
12381         ap = ixa->ixa_ipsec_action;
12382         if (ap == NULL) {
12383                 pp = ixa->ixa_ipsec_policy;
12384                 ASSERT(pp != NULL);
12385                 ap = pp->ipsp_act;
12386                 ASSERT(ap != NULL);
12387         }
12388 
12389         /* Handle explicit drop action and bypass. */
12390         switch (ap->ipa_act.ipa_type) {
12391         case IPSEC_ACT_DISCARD:
12392         case IPSEC_ACT_REJECT:
12393                 ip_drop_packet(mp, B_FALSE, ill,
12394                     DROPPER(ipss, ipds_spd_explicit), &ipss->ipsec_spd_dropper);
12395                 return (EHOSTUNREACH);  /* IPsec policy failure */
12396         case IPSEC_ACT_BYPASS:
12397                 return (ip_output_post_ipsec(mp, ixa));
12398         }
12399 
12400         /*
12401          * The order of processing is first insert a IP header if needed.
12402          * Then insert the ESP header and then the AH header.
12403          */
12404         if ((ixa->ixa_flags & IXAF_IS_IPV4) && ap->ipa_want_se) {
12405                 /*
12406                  * First get the outer IP header before sending
12407                  * it to ESP.
12408                  */
12409                 ipha_t *oipha, *iipha;
12410                 mblk_t *outer_mp, *inner_mp;
12411 
12412                 if ((outer_mp = allocb(sizeof (ipha_t), BPRI_HI)) == NULL) {
12413                         (void) mi_strlog(ill->ill_rq, 0,
12414                             SL_ERROR|SL_TRACE|SL_CONSOLE,
12415                             "ipsec_out_process: "
12416                             "Self-Encapsulation failed: Out of memory\n");
12417                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
12418                         ip_drop_output("ipIfStatsOutDiscards", mp, ill);
12419                         freemsg(mp);
12420                         return (ENOBUFS);
12421                 }
12422                 inner_mp = mp;
12423                 ASSERT(inner_mp->b_datap->db_type == M_DATA);
12424                 oipha = (ipha_t *)outer_mp->b_rptr;
12425                 iipha = (ipha_t *)inner_mp->b_rptr;
12426                 *oipha = *iipha;
12427                 outer_mp->b_wptr += sizeof (ipha_t);
12428                 oipha->ipha_length = htons(ntohs(iipha->ipha_length) +
12429                     sizeof (ipha_t));
12430                 oipha->ipha_protocol = IPPROTO_ENCAP;
12431                 oipha->ipha_version_and_hdr_length =
12432                     IP_SIMPLE_HDR_VERSION;
12433                 oipha->ipha_hdr_checksum = 0;
12434                 oipha->ipha_hdr_checksum = ip_csum_hdr(oipha);
12435                 outer_mp->b_cont = inner_mp;
12436                 mp = outer_mp;
12437 
12438                 ixa->ixa_flags |= IXAF_IPSEC_TUNNEL;
12439         }
12440 
12441         /* If we need to wait for a SA then we can't return any errno */
12442         if (((ap->ipa_want_ah && (ixa->ixa_ipsec_ah_sa == NULL)) ||
12443             (ap->ipa_want_esp && (ixa->ixa_ipsec_esp_sa == NULL))) &&
12444             !ipsec_out_select_sa(mp, ixa))
12445                 return (0);
12446 
12447         /*
12448          * By now, we know what SA's to use.  Toss over to ESP & AH
12449          * to do the heavy lifting.
12450          */
12451         if (ap->ipa_want_esp) {
12452                 ASSERT(ixa->ixa_ipsec_esp_sa != NULL);
12453 
12454                 mp = ixa->ixa_ipsec_esp_sa->ipsa_output_func(mp, ixa);
12455                 if (mp == NULL) {
12456                         /*
12457                          * Either it failed or is pending. In the former case
12458                          * ipIfStatsInDiscards was increased.
12459                          */
12460                         return (0);
12461                 }
12462         }
12463 
12464         if (ap->ipa_want_ah) {
12465                 ASSERT(ixa->ixa_ipsec_ah_sa != NULL);
12466 
12467                 mp = ixa->ixa_ipsec_ah_sa->ipsa_output_func(mp, ixa);
12468                 if (mp == NULL) {
12469                         /*
12470                          * Either it failed or is pending. In the former case
12471                          * ipIfStatsInDiscards was increased.
12472                          */
12473                         return (0);
12474                 }
12475         }
12476         /*
12477          * We are done with IPsec processing. Send it over
12478          * the wire.
12479          */
12480         return (ip_output_post_ipsec(mp, ixa));
12481 }
12482 
12483 /*
12484  * ioctls that go through a down/up sequence may need to wait for the down
12485  * to complete. This involves waiting for the ire and ipif refcnts to go down
12486  * to zero. Subsequently the ioctl is restarted from ipif_ill_refrele_tail.
12487  */
12488 /* ARGSUSED */
12489 void
12490 ip_reprocess_ioctl(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *dummy_arg)
12491 {
12492         struct iocblk *iocp;
12493         mblk_t *mp1;
12494         ip_ioctl_cmd_t *ipip;
12495         int err;
12496         sin_t   *sin;
12497         struct lifreq *lifr;
12498         struct ifreq *ifr;
12499 
12500         iocp = (struct iocblk *)mp->b_rptr;
12501         ASSERT(ipsq != NULL);
12502         /* Existence of mp1 verified in ip_wput_nondata */
12503         mp1 = mp->b_cont->b_cont;
12504         ipip = ip_sioctl_lookup(iocp->ioc_cmd);
12505         if (ipip->ipi_cmd == SIOCSLIFNAME || ipip->ipi_cmd == IF_UNITSEL) {
12506                 /*
12507                  * Special case where ipx_current_ipif is not set:
12508                  * ill_phyint_reinit merged the v4 and v6 into a single ipsq.
12509                  * We are here as were not able to complete the operation in
12510                  * ipif_set_values because we could not become exclusive on
12511                  * the new ipsq.
12512                  */
12513                 ill_t *ill = q->q_ptr;
12514                 ipsq_current_start(ipsq, ill->ill_ipif, ipip->ipi_cmd);
12515         }
12516         ASSERT(ipsq->ipsq_xop->ipx_current_ipif != NULL);
12517 
12518         if (ipip->ipi_cmd_type == IF_CMD) {
12519                 /* This a old style SIOC[GS]IF* command */
12520                 ifr = (struct ifreq *)mp1->b_rptr;
12521                 sin = (sin_t *)&ifr->ifr_addr;
12522         } else if (ipip->ipi_cmd_type == LIF_CMD) {
12523                 /* This a new style SIOC[GS]LIF* command */
12524                 lifr = (struct lifreq *)mp1->b_rptr;
12525                 sin = (sin_t *)&lifr->lifr_addr;
12526         } else {
12527                 sin = NULL;
12528         }
12529 
12530         err = (*ipip->ipi_func_restart)(ipsq->ipsq_xop->ipx_current_ipif, sin,
12531             q, mp, ipip, mp1->b_rptr);
12532 
12533         DTRACE_PROBE4(ipif__ioctl, char *, "ip_reprocess_ioctl finish",
12534             int, ipip->ipi_cmd,
12535             ill_t *, ipsq->ipsq_xop->ipx_current_ipif->ipif_ill,
12536             ipif_t *, ipsq->ipsq_xop->ipx_current_ipif);
12537 
12538         ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), ipsq);
12539 }
12540 
12541 /*
12542  * ioctl processing
12543  *
12544  * ioctl processing starts with ip_sioctl_copyin_setup(), which looks up
12545  * the ioctl command in the ioctl tables, determines the copyin data size
12546  * from the ipi_copyin_size field, and does an mi_copyin() of that size.
12547  *
12548  * ioctl processing then continues when the M_IOCDATA makes its way down to
12549  * ip_wput_nondata().  The ioctl is looked up again in the ioctl table, its
12550  * associated 'conn' is refheld till the end of the ioctl and the general
12551  * ioctl processing function ip_process_ioctl() is called to extract the
12552  * arguments and process the ioctl.  To simplify extraction, ioctl commands
12553  * are "typed" based on the arguments they take (e.g., LIF_CMD which takes a
12554  * `struct lifreq'), and a common extract function (e.g., ip_extract_lifreq())
12555  * is used to extract the ioctl's arguments.
12556  *
12557  * ip_process_ioctl determines if the ioctl needs to be serialized, and if
12558  * so goes thru the serialization primitive ipsq_try_enter. Then the
12559  * appropriate function to handle the ioctl is called based on the entry in
12560  * the ioctl table. ioctl completion is encapsulated in ip_ioctl_finish
12561  * which also refreleases the 'conn' that was refheld at the start of the
12562  * ioctl. Finally ipsq_exit is called if needed to exit the ipsq.
12563  *
12564  * Many exclusive ioctls go thru an internal down up sequence as part of
12565  * the operation. For example an attempt to change the IP address of an
12566  * ipif entails ipif_down, set address, ipif_up. Bringing down the interface
12567  * does all the cleanup such as deleting all ires that use this address.
12568  * Then we need to wait till all references to the interface go away.
12569  */
12570 void
12571 ip_process_ioctl(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *arg)
12572 {
12573         struct iocblk *iocp = (struct iocblk *)mp->b_rptr;
12574         ip_ioctl_cmd_t *ipip = arg;
12575         ip_extract_func_t *extract_funcp;
12576         cmd_info_t ci;
12577         int err;
12578         boolean_t entered_ipsq = B_FALSE;
12579 
12580         ip3dbg(("ip_process_ioctl: ioctl %X\n", iocp->ioc_cmd));
12581 
12582         if (ipip == NULL)
12583                 ipip = ip_sioctl_lookup(iocp->ioc_cmd);
12584 
12585         /*
12586          * SIOCLIFADDIF needs to go thru a special path since the
12587          * ill may not exist yet. This happens in the case of lo0
12588          * which is created using this ioctl.
12589          */
12590         if (ipip->ipi_cmd == SIOCLIFADDIF) {
12591                 err = ip_sioctl_addif(NULL, NULL, q, mp, NULL, NULL);
12592                 DTRACE_PROBE4(ipif__ioctl, char *, "ip_process_ioctl finish",
12593                     int, ipip->ipi_cmd, ill_t *, NULL, ipif_t *, NULL);
12594                 ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), NULL);
12595                 return;
12596         }
12597 
12598         ci.ci_ipif = NULL;
12599         switch (ipip->ipi_cmd_type) {
12600         case MISC_CMD:
12601         case MSFILT_CMD:
12602                 /*
12603                  * All MISC_CMD ioctls come in here -- e.g. SIOCGLIFCONF.
12604                  */
12605                 if (ipip->ipi_cmd == IF_UNITSEL) {
12606                         /* ioctl comes down the ill */
12607                         ci.ci_ipif = ((ill_t *)q->q_ptr)->ill_ipif;
12608                         ipif_refhold(ci.ci_ipif);
12609                 }
12610                 err = 0;
12611                 ci.ci_sin = NULL;
12612                 ci.ci_sin6 = NULL;
12613                 ci.ci_lifr = NULL;
12614                 extract_funcp = NULL;
12615                 break;
12616 
12617         case IF_CMD:
12618         case LIF_CMD:
12619                 extract_funcp = ip_extract_lifreq;
12620                 break;
12621 
12622         case ARP_CMD:
12623         case XARP_CMD:
12624                 extract_funcp = ip_extract_arpreq;
12625                 break;
12626 
12627         default:
12628                 ASSERT(0);
12629         }
12630 
12631         if (extract_funcp != NULL) {
12632                 err = (*extract_funcp)(q, mp, ipip, &ci);
12633                 if (err != 0) {
12634                         DTRACE_PROBE4(ipif__ioctl,
12635                             char *, "ip_process_ioctl finish err",
12636                             int, ipip->ipi_cmd, ill_t *, NULL, ipif_t *, NULL);
12637                         ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), NULL);
12638                         return;
12639                 }
12640 
12641                 /*
12642                  * All of the extraction functions return a refheld ipif.
12643                  */
12644                 ASSERT(ci.ci_ipif != NULL);
12645         }
12646 
12647         if (!(ipip->ipi_flags & IPI_WR)) {
12648                 /*
12649                  * A return value of EINPROGRESS means the ioctl is
12650                  * either queued and waiting for some reason or has
12651                  * already completed.
12652                  */
12653                 err = (*ipip->ipi_func)(ci.ci_ipif, ci.ci_sin, q, mp, ipip,
12654                     ci.ci_lifr);
12655                 if (ci.ci_ipif != NULL) {
12656                         DTRACE_PROBE4(ipif__ioctl,
12657                             char *, "ip_process_ioctl finish RD",
12658                             int, ipip->ipi_cmd, ill_t *, ci.ci_ipif->ipif_ill,
12659                             ipif_t *, ci.ci_ipif);
12660                         ipif_refrele(ci.ci_ipif);
12661                 } else {
12662                         DTRACE_PROBE4(ipif__ioctl,
12663                             char *, "ip_process_ioctl finish RD",
12664                             int, ipip->ipi_cmd, ill_t *, NULL, ipif_t *, NULL);
12665                 }
12666                 ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), NULL);
12667                 return;
12668         }
12669 
12670         ASSERT(ci.ci_ipif != NULL);
12671 
12672         /*
12673          * If ipsq is non-NULL, we are already being called exclusively
12674          */
12675         ASSERT(ipsq == NULL || IAM_WRITER_IPSQ(ipsq));
12676         if (ipsq == NULL) {
12677                 ipsq = ipsq_try_enter(ci.ci_ipif, NULL, q, mp, ip_process_ioctl,
12678                     NEW_OP, B_TRUE);
12679                 if (ipsq == NULL) {
12680                         ipif_refrele(ci.ci_ipif);
12681                         return;
12682                 }
12683                 entered_ipsq = B_TRUE;
12684         }
12685         /*
12686          * Release the ipif so that ipif_down and friends that wait for
12687          * references to go away are not misled about the current ipif_refcnt
12688          * values. We are writer so we can access the ipif even after releasing
12689          * the ipif.
12690          */
12691         ipif_refrele(ci.ci_ipif);
12692 
12693         ipsq_current_start(ipsq, ci.ci_ipif, ipip->ipi_cmd);
12694 
12695         /*
12696          * A return value of EINPROGRESS means the ioctl is
12697          * either queued and waiting for some reason or has
12698          * already completed.
12699          */
12700         err = (*ipip->ipi_func)(ci.ci_ipif, ci.ci_sin, q, mp, ipip, ci.ci_lifr);
12701 
12702         DTRACE_PROBE4(ipif__ioctl, char *, "ip_process_ioctl finish WR",
12703             int, ipip->ipi_cmd,
12704             ill_t *, ci.ci_ipif == NULL ? NULL : ci.ci_ipif->ipif_ill,
12705             ipif_t *, ci.ci_ipif);
12706         ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), ipsq);
12707 
12708         if (entered_ipsq)
12709                 ipsq_exit(ipsq);
12710 }
12711 
12712 /*
12713  * Complete the ioctl. Typically ioctls use the mi package and need to
12714  * do mi_copyout/mi_copy_done.
12715  */
12716 void
12717 ip_ioctl_finish(queue_t *q, mblk_t *mp, int err, int mode, ipsq_t *ipsq)
12718 {
12719         conn_t  *connp = NULL;
12720 
12721         if (err == EINPROGRESS)
12722                 return;
12723 
12724         if (CONN_Q(q)) {
12725                 connp = Q_TO_CONN(q);
12726                 ASSERT(connp->conn_ref >= 2);
12727         }
12728 
12729         switch (mode) {
12730         case COPYOUT:
12731                 if (err == 0)
12732                         mi_copyout(q, mp);
12733                 else
12734                         mi_copy_done(q, mp, err);
12735                 break;
12736 
12737         case NO_COPYOUT:
12738                 mi_copy_done(q, mp, err);
12739                 break;
12740 
12741         default:
12742                 ASSERT(mode == CONN_CLOSE);     /* aborted through CONN_CLOSE */
12743                 break;
12744         }
12745 
12746         /*
12747          * The conn refhold and ioctlref placed on the conn at the start of the
12748          * ioctl are released here.
12749          */
12750         if (connp != NULL) {
12751                 CONN_DEC_IOCTLREF(connp);
12752                 CONN_OPER_PENDING_DONE(connp);
12753         }
12754 
12755         if (ipsq != NULL)
12756                 ipsq_current_finish(ipsq);
12757 }
12758 
12759 /* Handles all non data messages */
12760 int
12761 ip_wput_nondata(queue_t *q, mblk_t *mp)
12762 {
12763         mblk_t          *mp1;
12764         struct iocblk   *iocp;
12765         ip_ioctl_cmd_t  *ipip;
12766         conn_t          *connp;
12767         cred_t          *cr;
12768         char            *proto_str;
12769 
12770         if (CONN_Q(q))
12771                 connp = Q_TO_CONN(q);
12772         else
12773                 connp = NULL;
12774 
12775         switch (DB_TYPE(mp)) {
12776         case M_IOCTL:
12777                 /*
12778                  * IOCTL processing begins in ip_sioctl_copyin_setup which
12779                  * will arrange to copy in associated control structures.
12780                  */
12781                 ip_sioctl_copyin_setup(q, mp);
12782                 return (0);
12783         case M_IOCDATA:
12784                 /*
12785                  * Ensure that this is associated with one of our trans-
12786                  * parent ioctls.  If it's not ours, discard it if we're
12787                  * running as a driver, or pass it on if we're a module.
12788                  */
12789                 iocp = (struct iocblk *)mp->b_rptr;
12790                 ipip = ip_sioctl_lookup(iocp->ioc_cmd);
12791                 if (ipip == NULL) {
12792                         if (q->q_next == NULL) {
12793                                 goto nak;
12794                         } else {
12795                                 putnext(q, mp);
12796                         }
12797                         return (0);
12798                 }
12799                 if ((q->q_next != NULL) && !(ipip->ipi_flags & IPI_MODOK)) {
12800                         /*
12801                          * The ioctl is one we recognise, but is not consumed
12802                          * by IP as a module and we are a module, so we drop
12803                          */
12804                         goto nak;
12805                 }
12806 
12807                 /* IOCTL continuation following copyin or copyout. */
12808                 if (mi_copy_state(q, mp, NULL) == -1) {
12809                         /*
12810                          * The copy operation failed.  mi_copy_state already
12811                          * cleaned up, so we're out of here.
12812                          */
12813                         return (0);
12814                 }
12815                 /*
12816                  * If we just completed a copy in, we become writer and
12817                  * continue processing in ip_sioctl_copyin_done.  If it
12818                  * was a copy out, we call mi_copyout again.  If there is
12819                  * nothing more to copy out, it will complete the IOCTL.
12820                  */
12821                 if (MI_COPY_DIRECTION(mp) == MI_COPY_IN) {
12822                         if (!(mp1 = mp->b_cont) || !(mp1 = mp1->b_cont)) {
12823                                 mi_copy_done(q, mp, EPROTO);
12824                                 return (0);
12825                         }
12826                         /*
12827                          * Check for cases that need more copying.  A return
12828                          * value of 0 means a second copyin has been started,
12829                          * so we return; a return value of 1 means no more
12830                          * copying is needed, so we continue.
12831                          */
12832                         if (ipip->ipi_cmd_type == MSFILT_CMD &&
12833                             MI_COPY_COUNT(mp) == 1) {
12834                                 if (ip_copyin_msfilter(q, mp) == 0)
12835                                         return (0);
12836                         }
12837                         /*
12838                          * Refhold the conn, till the ioctl completes. This is
12839                          * needed in case the ioctl ends up in the pending mp
12840                          * list. Every mp in the ipx_pending_mp list must have
12841                          * a refhold on the conn to resume processing. The
12842                          * refhold is released when the ioctl completes
12843                          * (whether normally or abnormally). An ioctlref is also
12844                          * placed on the conn to prevent TCP from removing the
12845                          * queue needed to send the ioctl reply back.
12846                          * In all cases ip_ioctl_finish is called to finish
12847                          * the ioctl and release the refholds.
12848                          */
12849                         if (connp != NULL) {
12850                                 /* This is not a reentry */
12851                                 CONN_INC_REF(connp);
12852                                 CONN_INC_IOCTLREF(connp);
12853                         } else {
12854                                 if (!(ipip->ipi_flags & IPI_MODOK)) {
12855                                         mi_copy_done(q, mp, EINVAL);
12856                                         return (0);
12857                                 }
12858                         }
12859 
12860                         ip_process_ioctl(NULL, q, mp, ipip);
12861 
12862                 } else {
12863                         mi_copyout(q, mp);
12864                 }
12865                 return (0);
12866 
12867         case M_IOCNAK:
12868                 /*
12869                  * The only way we could get here is if a resolver didn't like
12870                  * an IOCTL we sent it.  This shouldn't happen.
12871                  */
12872                 (void) mi_strlog(q, 1, SL_ERROR|SL_TRACE,
12873                     "ip_wput_nondata: unexpected M_IOCNAK, ioc_cmd 0x%x",
12874                     ((struct iocblk *)mp->b_rptr)->ioc_cmd);
12875                 freemsg(mp);
12876                 return (0);
12877         case M_IOCACK:
12878                 /* /dev/ip shouldn't see this */
12879                 goto nak;
12880         case M_FLUSH:
12881                 if (*mp->b_rptr & FLUSHW)
12882                         flushq(q, FLUSHALL);
12883                 if (q->q_next) {
12884                         putnext(q, mp);
12885                         return (0);
12886                 }
12887                 if (*mp->b_rptr & FLUSHR) {
12888                         *mp->b_rptr &= ~FLUSHW;
12889                         qreply(q, mp);
12890                         return (0);
12891                 }
12892                 freemsg(mp);
12893                 return (0);
12894         case M_CTL:
12895                 break;
12896         case M_PROTO:
12897         case M_PCPROTO:
12898                 /*
12899                  * The only PROTO messages we expect are SNMP-related.
12900                  */
12901                 switch (((union T_primitives *)mp->b_rptr)->type) {
12902                 case T_SVR4_OPTMGMT_REQ:
12903                         ip2dbg(("ip_wput_nondata: T_SVR4_OPTMGMT_REQ "
12904                             "flags %x\n",
12905                             ((struct T_optmgmt_req *)mp->b_rptr)->MGMT_flags));
12906 
12907                         if (connp == NULL) {
12908                                 proto_str = "T_SVR4_OPTMGMT_REQ";
12909                                 goto protonak;
12910                         }
12911 
12912                         /*
12913                          * All Solaris components should pass a db_credp
12914                          * for this TPI message, hence we ASSERT.
12915                          * But in case there is some other M_PROTO that looks
12916                          * like a TPI message sent by some other kernel
12917                          * component, we check and return an error.
12918                          */
12919                         cr = msg_getcred(mp, NULL);
12920                         ASSERT(cr != NULL);
12921                         if (cr == NULL) {
12922                                 mp = mi_tpi_err_ack_alloc(mp, TSYSERR, EINVAL);
12923                                 if (mp != NULL)
12924                                         qreply(q, mp);
12925                                 return (0);
12926                         }
12927 
12928                         if (!snmpcom_req(q, mp, ip_snmp_set, ip_snmp_get, cr)) {
12929                                 proto_str = "Bad SNMPCOM request?";
12930                                 goto protonak;
12931                         }
12932                         return (0);
12933                 default:
12934                         ip1dbg(("ip_wput_nondata: dropping M_PROTO prim %u\n",
12935                             (int)*(uint_t *)mp->b_rptr));
12936                         freemsg(mp);
12937                         return (0);
12938                 }
12939         default:
12940                 break;
12941         }
12942         if (q->q_next) {
12943                 putnext(q, mp);
12944         } else
12945                 freemsg(mp);
12946         return (0);
12947 
12948 nak:
12949         iocp->ioc_error = EINVAL;
12950         mp->b_datap->db_type = M_IOCNAK;
12951         iocp->ioc_count = 0;
12952         qreply(q, mp);
12953         return (0);
12954 
12955 protonak:
12956         cmn_err(CE_NOTE, "IP doesn't process %s as a module", proto_str);
12957         if ((mp = mi_tpi_err_ack_alloc(mp, TPROTO, EINVAL)) != NULL)
12958                 qreply(q, mp);
12959         return (0);
12960 }
12961 
12962 /*
12963  * Process IP options in an outbound packet.  Verify that the nexthop in a
12964  * strict source route is onlink.
12965  * Returns non-zero if something fails in which case an ICMP error has been
12966  * sent and mp freed.
12967  *
12968  * Assumes the ULP has called ip_massage_options to move nexthop into ipha_dst.
12969  */
12970 int
12971 ip_output_options(mblk_t *mp, ipha_t *ipha, ip_xmit_attr_t *ixa, ill_t *ill)
12972 {
12973         ipoptp_t        opts;
12974         uchar_t         *opt;
12975         uint8_t         optval;
12976         uint8_t         optlen;
12977         ipaddr_t        dst;
12978         intptr_t        code = 0;
12979         ire_t           *ire;
12980         ip_stack_t      *ipst = ixa->ixa_ipst;
12981         ip_recv_attr_t  iras;
12982 
12983         ip2dbg(("ip_output_options\n"));
12984 
12985         dst = ipha->ipha_dst;
12986         for (optval = ipoptp_first(&opts, ipha);
12987             optval != IPOPT_EOL;
12988             optval = ipoptp_next(&opts)) {
12989                 opt = opts.ipoptp_cur;
12990                 optlen = opts.ipoptp_len;
12991                 ip2dbg(("ip_output_options: opt %d, len %d\n",
12992                     optval, optlen));
12993                 switch (optval) {
12994                         uint32_t off;
12995                 case IPOPT_SSRR:
12996                 case IPOPT_LSRR:
12997                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
12998                                 ip1dbg((
12999                                     "ip_output_options: bad option offset\n"));
13000                                 code = (char *)&opt[IPOPT_OLEN] -
13001                                     (char *)ipha;
13002                                 goto param_prob;
13003                         }
13004                         off = opt[IPOPT_OFFSET];
13005                         ip1dbg(("ip_output_options: next hop 0x%x\n",
13006                             ntohl(dst)));
13007                         /*
13008                          * For strict: verify that dst is directly
13009                          * reachable.
13010                          */
13011                         if (optval == IPOPT_SSRR) {
13012                                 ire = ire_ftable_lookup_v4(dst, 0, 0,
13013                                     IRE_INTERFACE, NULL, ALL_ZONES,
13014                                     ixa->ixa_tsl,
13015                                     MATCH_IRE_TYPE | MATCH_IRE_SECATTR, 0, ipst,
13016                                     NULL);
13017                                 if (ire == NULL) {
13018                                         ip1dbg(("ip_output_options: SSRR not"
13019                                             " directly reachable: 0x%x\n",
13020                                             ntohl(dst)));
13021                                         goto bad_src_route;
13022                                 }
13023                                 ire_refrele(ire);
13024                         }
13025                         break;
13026                 case IPOPT_RR:
13027                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
13028                                 ip1dbg((
13029                                     "ip_output_options: bad option offset\n"));
13030                                 code = (char *)&opt[IPOPT_OLEN] -
13031                                     (char *)ipha;
13032                                 goto param_prob;
13033                         }
13034                         break;
13035                 case IPOPT_TS:
13036                         /*
13037                          * Verify that length >=5 and that there is either
13038                          * room for another timestamp or that the overflow
13039                          * counter is not maxed out.
13040                          */
13041                         code = (char *)&opt[IPOPT_OLEN] - (char *)ipha;
13042                         if (optlen < IPOPT_MINLEN_IT) {
13043                                 goto param_prob;
13044                         }
13045                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
13046                                 ip1dbg((
13047                                     "ip_output_options: bad option offset\n"));
13048                                 code = (char *)&opt[IPOPT_OFFSET] -
13049                                     (char *)ipha;
13050                                 goto param_prob;
13051                         }
13052                         switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
13053                         case IPOPT_TS_TSONLY:
13054                                 off = IPOPT_TS_TIMELEN;
13055                                 break;
13056                         case IPOPT_TS_TSANDADDR:
13057                         case IPOPT_TS_PRESPEC:
13058                         case IPOPT_TS_PRESPEC_RFC791:
13059                                 off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
13060                                 break;
13061                         default:
13062                                 code = (char *)&opt[IPOPT_POS_OV_FLG] -
13063                                     (char *)ipha;
13064                                 goto param_prob;
13065                         }
13066                         if (opt[IPOPT_OFFSET] - 1 + off > optlen &&
13067                             (opt[IPOPT_POS_OV_FLG] & 0xF0) == 0xF0) {
13068                                 /*
13069                                  * No room and the overflow counter is 15
13070                                  * already.
13071                                  */
13072                                 goto param_prob;
13073                         }
13074                         break;
13075                 }
13076         }
13077 
13078         if ((opts.ipoptp_flags & IPOPTP_ERROR) == 0)
13079                 return (0);
13080 
13081         ip1dbg(("ip_output_options: error processing IP options."));
13082         code = (char *)&opt[IPOPT_OFFSET] - (char *)ipha;
13083 
13084 param_prob:
13085         bzero(&iras, sizeof (iras));
13086         iras.ira_ill = iras.ira_rill = ill;
13087         iras.ira_ruifindex = ill->ill_phyint->phyint_ifindex;
13088         iras.ira_rifindex = iras.ira_ruifindex;
13089         iras.ira_flags = IRAF_IS_IPV4;
13090 
13091         ip_drop_output("ip_output_options", mp, ill);
13092         icmp_param_problem(mp, (uint8_t)code, &iras);
13093         ASSERT(!(iras.ira_flags & IRAF_IPSEC_SECURE));
13094         return (-1);
13095 
13096 bad_src_route:
13097         bzero(&iras, sizeof (iras));
13098         iras.ira_ill = iras.ira_rill = ill;
13099         iras.ira_ruifindex = ill->ill_phyint->phyint_ifindex;
13100         iras.ira_rifindex = iras.ira_ruifindex;
13101         iras.ira_flags = IRAF_IS_IPV4;
13102 
13103         ip_drop_input("ICMP_SOURCE_ROUTE_FAILED", mp, ill);
13104         icmp_unreachable(mp, ICMP_SOURCE_ROUTE_FAILED, &iras);
13105         ASSERT(!(iras.ira_flags & IRAF_IPSEC_SECURE));
13106         return (-1);
13107 }
13108 
13109 /*
13110  * The maximum value of conn_drain_list_cnt is CONN_MAXDRAINCNT.
13111  * conn_drain_list_cnt can be changed by setting conn_drain_nthreads
13112  * thru /etc/system.
13113  */
13114 #define CONN_MAXDRAINCNT        64
13115 
13116 static void
13117 conn_drain_init(ip_stack_t *ipst)
13118 {
13119         int i, j;
13120         idl_tx_list_t *itl_tx;
13121 
13122         ipst->ips_conn_drain_list_cnt = conn_drain_nthreads;
13123 
13124         if ((ipst->ips_conn_drain_list_cnt == 0) ||
13125             (ipst->ips_conn_drain_list_cnt > CONN_MAXDRAINCNT)) {
13126                 /*
13127                  * Default value of the number of drainers is the
13128                  * number of cpus, subject to maximum of 8 drainers.
13129                  */
13130                 if (boot_max_ncpus != -1)
13131                         ipst->ips_conn_drain_list_cnt = MIN(boot_max_ncpus, 8);
13132                 else
13133                         ipst->ips_conn_drain_list_cnt = MIN(max_ncpus, 8);
13134         }
13135 
13136         ipst->ips_idl_tx_list =
13137             kmem_zalloc(TX_FANOUT_SIZE * sizeof (idl_tx_list_t), KM_SLEEP);
13138         for (i = 0; i < TX_FANOUT_SIZE; i++) {
13139                 itl_tx =  &ipst->ips_idl_tx_list[i];
13140                 itl_tx->txl_drain_list =
13141                     kmem_zalloc(ipst->ips_conn_drain_list_cnt *
13142                     sizeof (idl_t), KM_SLEEP);
13143                 mutex_init(&itl_tx->txl_lock, NULL, MUTEX_DEFAULT, NULL);
13144                 for (j = 0; j < ipst->ips_conn_drain_list_cnt; j++) {
13145                         mutex_init(&itl_tx->txl_drain_list[j].idl_lock, NULL,
13146                             MUTEX_DEFAULT, NULL);
13147                         itl_tx->txl_drain_list[j].idl_itl = itl_tx;
13148                 }
13149         }
13150 }
13151 
13152 static void
13153 conn_drain_fini(ip_stack_t *ipst)
13154 {
13155         int i;
13156         idl_tx_list_t *itl_tx;
13157 
13158         for (i = 0; i < TX_FANOUT_SIZE; i++) {
13159                 itl_tx =  &ipst->ips_idl_tx_list[i];
13160                 kmem_free(itl_tx->txl_drain_list,
13161                     ipst->ips_conn_drain_list_cnt * sizeof (idl_t));
13162         }
13163         kmem_free(ipst->ips_idl_tx_list,
13164             TX_FANOUT_SIZE * sizeof (idl_tx_list_t));
13165         ipst->ips_idl_tx_list = NULL;
13166 }
13167 
13168 /*
13169  * Flow control has blocked us from proceeding.  Insert the given conn in one
13170  * of the conn drain lists.  When flow control is unblocked, either ip_wsrv()
13171  * (STREAMS) or ill_flow_enable() (direct) will be called back, which in turn
13172  * will call conn_walk_drain().  See the flow control notes at the top of this
13173  * file for more details.
13174  */
13175 void
13176 conn_drain_insert(conn_t *connp, idl_tx_list_t *tx_list)
13177 {
13178         idl_t   *idl = tx_list->txl_drain_list;
13179         uint_t  index;
13180         ip_stack_t      *ipst = connp->conn_netstack->netstack_ip;
13181 
13182         mutex_enter(&connp->conn_lock);
13183         if (connp->conn_state_flags & CONN_CLOSING) {
13184                 /*
13185                  * The conn is closing as a result of which CONN_CLOSING
13186                  * is set. Return.
13187                  */
13188                 mutex_exit(&connp->conn_lock);
13189                 return;
13190         } else if (connp->conn_idl == NULL) {
13191                 /*
13192                  * Assign the next drain list round robin. We dont' use
13193                  * a lock, and thus it may not be strictly round robin.
13194                  * Atomicity of load/stores is enough to make sure that
13195                  * conn_drain_list_index is always within bounds.
13196                  */
13197                 index = tx_list->txl_drain_index;
13198                 ASSERT(index < ipst->ips_conn_drain_list_cnt);
13199                 connp->conn_idl = &tx_list->txl_drain_list[index];
13200                 index++;
13201                 if (index == ipst->ips_conn_drain_list_cnt)
13202                         index = 0;
13203                 tx_list->txl_drain_index = index;
13204         } else {
13205                 ASSERT(connp->conn_idl->idl_itl == tx_list);
13206         }
13207         mutex_exit(&connp->conn_lock);
13208 
13209         idl = connp->conn_idl;
13210         mutex_enter(&idl->idl_lock);
13211         if ((connp->conn_drain_prev != NULL) ||
13212             (connp->conn_state_flags & CONN_CLOSING)) {
13213                 /*
13214                  * The conn is either already in the drain list or closing.
13215                  * (We needed to check for CONN_CLOSING again since close can
13216                  * sneak in between dropping conn_lock and acquiring idl_lock.)
13217                  */
13218                 mutex_exit(&idl->idl_lock);
13219                 return;
13220         }
13221 
13222         /*
13223          * The conn is not in the drain list. Insert it at the
13224          * tail of the drain list. The drain list is circular
13225          * and doubly linked. idl_conn points to the 1st element
13226          * in the list.
13227          */
13228         if (idl->idl_conn == NULL) {
13229                 idl->idl_conn = connp;
13230                 connp->conn_drain_next = connp;
13231                 connp->conn_drain_prev = connp;
13232         } else {
13233                 conn_t *head = idl->idl_conn;
13234 
13235                 connp->conn_drain_next = head;
13236                 connp->conn_drain_prev = head->conn_drain_prev;
13237                 head->conn_drain_prev->conn_drain_next = connp;
13238                 head->conn_drain_prev = connp;
13239         }
13240         /*
13241          * For non streams based sockets assert flow control.
13242          */
13243         conn_setqfull(connp, NULL);
13244         mutex_exit(&idl->idl_lock);
13245 }
13246 
13247 static void
13248 conn_drain_remove(conn_t *connp)
13249 {
13250         idl_t *idl = connp->conn_idl;
13251 
13252         if (idl != NULL) {
13253                 /*
13254                  * Remove ourself from the drain list.
13255                  */
13256                 if (connp->conn_drain_next == connp) {
13257                         /* Singleton in the list */
13258                         ASSERT(connp->conn_drain_prev == connp);
13259                         idl->idl_conn = NULL;
13260                 } else {
13261                         connp->conn_drain_prev->conn_drain_next =
13262                             connp->conn_drain_next;
13263                         connp->conn_drain_next->conn_drain_prev =
13264                             connp->conn_drain_prev;
13265                         if (idl->idl_conn == connp)
13266                                 idl->idl_conn = connp->conn_drain_next;
13267                 }
13268 
13269                 /*
13270                  * NOTE: because conn_idl is associated with a specific drain
13271                  * list which in turn is tied to the index the TX ring
13272                  * (txl_cookie) hashes to, and because the TX ring can change
13273                  * over the lifetime of the conn_t, we must clear conn_idl so
13274                  * a subsequent conn_drain_insert() will set conn_idl again
13275                  * based on the latest txl_cookie.
13276                  */
13277                 connp->conn_idl = NULL;
13278         }
13279         connp->conn_drain_next = NULL;
13280         connp->conn_drain_prev = NULL;
13281 
13282         conn_clrqfull(connp, NULL);
13283         /*
13284          * For streams based sockets open up flow control.
13285          */
13286         if (!IPCL_IS_NONSTR(connp))
13287                 enableok(connp->conn_wq);
13288 }
13289 
13290 /*
13291  * This conn is closing, and we are called from ip_close. OR
13292  * this conn is draining because flow-control on the ill has been relieved.
13293  *
13294  * We must also need to remove conn's on this idl from the list, and also
13295  * inform the sockfs upcalls about the change in flow-control.
13296  */
13297 static void
13298 conn_drain(conn_t *connp, boolean_t closing)
13299 {
13300         idl_t *idl;
13301         conn_t *next_connp;
13302 
13303         /*
13304          * connp->conn_idl is stable at this point, and no lock is needed
13305          * to check it. If we are called from ip_close, close has already
13306          * set CONN_CLOSING, thus freezing the value of conn_idl, and
13307          * called us only because conn_idl is non-null. If we are called thru
13308          * service, conn_idl could be null, but it cannot change because
13309          * service is single-threaded per queue, and there cannot be another
13310          * instance of service trying to call conn_drain_insert on this conn
13311          * now.
13312          */
13313         ASSERT(!closing || connp == NULL || connp->conn_idl != NULL);
13314 
13315         /*
13316          * If the conn doesn't exist or is not on a drain list, bail.
13317          */
13318         if (connp == NULL || connp->conn_idl == NULL ||
13319             connp->conn_drain_prev == NULL) {
13320                 return;
13321         }
13322 
13323         idl = connp->conn_idl;
13324         ASSERT(MUTEX_HELD(&idl->idl_lock));
13325 
13326         if (!closing) {
13327                 next_connp = connp->conn_drain_next;
13328                 while (next_connp != connp) {
13329                         conn_t *delconnp = next_connp;
13330 
13331                         next_connp = next_connp->conn_drain_next;
13332                         conn_drain_remove(delconnp);
13333                 }
13334                 ASSERT(connp->conn_drain_next == idl->idl_conn);
13335         }
13336         conn_drain_remove(connp);
13337 }
13338 
13339 /*
13340  * Write service routine. Shared perimeter entry point.
13341  * The device queue's messages has fallen below the low water mark and STREAMS
13342  * has backenabled the ill_wq. Send sockfs notification about flow-control on
13343  * each waiting conn.
13344  */
13345 int
13346 ip_wsrv(queue_t *q)
13347 {
13348         ill_t   *ill;
13349 
13350         ill = (ill_t *)q->q_ptr;
13351         if (ill->ill_state_flags == 0) {
13352                 ip_stack_t *ipst = ill->ill_ipst;
13353 
13354                 /*
13355                  * The device flow control has opened up.
13356                  * Walk through conn drain lists and qenable the
13357                  * first conn in each list. This makes sense only
13358                  * if the stream is fully plumbed and setup.
13359                  * Hence the ill_state_flags check above.
13360                  */
13361                 ip1dbg(("ip_wsrv: walking\n"));
13362                 conn_walk_drain(ipst, &ipst->ips_idl_tx_list[0]);
13363                 enableok(ill->ill_wq);
13364         }
13365         return (0);
13366 }
13367 
13368 /*
13369  * Callback to disable flow control in IP.
13370  *
13371  * This is a mac client callback added when the DLD_CAPAB_DIRECT capability
13372  * is enabled.
13373  *
13374  * When MAC_TX() is not able to send any more packets, dld sets its queue
13375  * to QFULL and enable the STREAMS flow control. Later, when the underlying
13376  * driver is able to continue to send packets, it calls mac_tx_(ring_)update()
13377  * function and wakes up corresponding mac worker threads, which in turn
13378  * calls this callback function, and disables flow control.
13379  */
13380 void
13381 ill_flow_enable(void *arg, ip_mac_tx_cookie_t cookie)
13382 {
13383         ill_t *ill = (ill_t *)arg;
13384         ip_stack_t *ipst = ill->ill_ipst;
13385         idl_tx_list_t *idl_txl;
13386 
13387         idl_txl = &ipst->ips_idl_tx_list[IDLHASHINDEX(cookie)];
13388         mutex_enter(&idl_txl->txl_lock);
13389         /* add code to to set a flag to indicate idl_txl is enabled */
13390         conn_walk_drain(ipst, idl_txl);
13391         mutex_exit(&idl_txl->txl_lock);
13392 }
13393 
13394 /*
13395  * Flow control has been relieved and STREAMS has backenabled us; drain
13396  * all the conn lists on `tx_list'.
13397  */
13398 static void
13399 conn_walk_drain(ip_stack_t *ipst, idl_tx_list_t *tx_list)
13400 {
13401         int i;
13402         idl_t *idl;
13403 
13404         IP_STAT(ipst, ip_conn_walk_drain);
13405 
13406         for (i = 0; i < ipst->ips_conn_drain_list_cnt; i++) {
13407                 idl = &tx_list->txl_drain_list[i];
13408                 mutex_enter(&idl->idl_lock);
13409                 conn_drain(idl->idl_conn, B_FALSE);
13410                 mutex_exit(&idl->idl_lock);
13411         }
13412 }
13413 
13414 /*
13415  * Determine if the ill and multicast aspects of that packets
13416  * "matches" the conn.
13417  */
13418 boolean_t
13419 conn_wantpacket(conn_t *connp, ip_recv_attr_t *ira, ipha_t *ipha)
13420 {
13421         ill_t           *ill = ira->ira_rill;
13422         zoneid_t        zoneid = ira->ira_zoneid;
13423         uint_t          in_ifindex;
13424         ipaddr_t        dst, src;
13425 
13426         dst = ipha->ipha_dst;
13427         src = ipha->ipha_src;
13428 
13429         /*
13430          * conn_incoming_ifindex is set by IP_BOUND_IF which limits
13431          * unicast, broadcast and multicast reception to
13432          * conn_incoming_ifindex.
13433          * conn_wantpacket is called for unicast, broadcast and
13434          * multicast packets.
13435          */
13436         in_ifindex = connp->conn_incoming_ifindex;
13437 
13438         /* mpathd can bind to the under IPMP interface, which we allow */
13439         if (in_ifindex != 0 && in_ifindex != ill->ill_phyint->phyint_ifindex) {
13440                 if (!IS_UNDER_IPMP(ill))
13441                         return (B_FALSE);
13442 
13443                 if (in_ifindex != ipmp_ill_get_ipmp_ifindex(ill))
13444                         return (B_FALSE);
13445         }
13446 
13447         if (!IPCL_ZONE_MATCH(connp, zoneid))
13448                 return (B_FALSE);
13449 
13450         if (!(ira->ira_flags & IRAF_MULTICAST))
13451                 return (B_TRUE);
13452 
13453         if (connp->conn_multi_router) {
13454                 /* multicast packet and multicast router socket: send up */
13455                 return (B_TRUE);
13456         }
13457 
13458         if (ipha->ipha_protocol == IPPROTO_PIM ||
13459             ipha->ipha_protocol == IPPROTO_RSVP)
13460                 return (B_TRUE);
13461 
13462         return (conn_hasmembers_ill_withsrc_v4(connp, dst, src, ira->ira_ill));
13463 }
13464 
13465 void
13466 conn_setqfull(conn_t *connp, boolean_t *flow_stopped)
13467 {
13468         if (IPCL_IS_NONSTR(connp)) {
13469                 (*connp->conn_upcalls->su_txq_full)
13470                     (connp->conn_upper_handle, B_TRUE);
13471                 if (flow_stopped != NULL)
13472                         *flow_stopped = B_TRUE;
13473         } else {
13474                 queue_t *q = connp->conn_wq;
13475 
13476                 ASSERT(q != NULL);
13477                 if (!(q->q_flag & QFULL)) {
13478                         mutex_enter(QLOCK(q));
13479                         if (!(q->q_flag & QFULL)) {
13480                                 /* still need to set QFULL */
13481                                 q->q_flag |= QFULL;
13482                                 /* set flow_stopped to true under QLOCK */
13483                                 if (flow_stopped != NULL)
13484                                         *flow_stopped = B_TRUE;
13485                                 mutex_exit(QLOCK(q));
13486                         } else {
13487                                 /* flow_stopped is left unchanged */
13488                                 mutex_exit(QLOCK(q));
13489                         }
13490                 }
13491         }
13492 }
13493 
13494 void
13495 conn_clrqfull(conn_t *connp, boolean_t *flow_stopped)
13496 {
13497         if (IPCL_IS_NONSTR(connp)) {
13498                 (*connp->conn_upcalls->su_txq_full)
13499                     (connp->conn_upper_handle, B_FALSE);
13500                 if (flow_stopped != NULL)
13501                         *flow_stopped = B_FALSE;
13502         } else {
13503                 queue_t *q = connp->conn_wq;
13504 
13505                 ASSERT(q != NULL);
13506                 if (q->q_flag & QFULL) {
13507                         mutex_enter(QLOCK(q));
13508                         if (q->q_flag & QFULL) {
13509                                 q->q_flag &= ~QFULL;
13510                                 /* set flow_stopped to false under QLOCK */
13511                                 if (flow_stopped != NULL)
13512                                         *flow_stopped = B_FALSE;
13513                                 mutex_exit(QLOCK(q));
13514                                 if (q->q_flag & QWANTW)
13515                                         qbackenable(q, 0);
13516                         } else {
13517                                 /* flow_stopped is left unchanged */
13518                                 mutex_exit(QLOCK(q));
13519                         }
13520                 }
13521         }
13522 
13523         mutex_enter(&connp->conn_lock);
13524         connp->conn_blocked = B_FALSE;
13525         mutex_exit(&connp->conn_lock);
13526 }
13527 
13528 /*
13529  * Return the length in bytes of the IPv4 headers (base header, label, and
13530  * other IP options) that will be needed based on the
13531  * ip_pkt_t structure passed by the caller.
13532  *
13533  * The returned length does not include the length of the upper level
13534  * protocol (ULP) header.
13535  * The caller needs to check that the length doesn't exceed the max for IPv4.
13536  */
13537 int
13538 ip_total_hdrs_len_v4(const ip_pkt_t *ipp)
13539 {
13540         int len;
13541 
13542         len = IP_SIMPLE_HDR_LENGTH;
13543         if (ipp->ipp_fields & IPPF_LABEL_V4) {
13544                 ASSERT(ipp->ipp_label_len_v4 != 0);
13545                 /* We need to round up here */
13546                 len += (ipp->ipp_label_len_v4 + 3) & ~3;
13547         }
13548 
13549         if (ipp->ipp_fields & IPPF_IPV4_OPTIONS) {
13550                 ASSERT(ipp->ipp_ipv4_options_len != 0);
13551                 ASSERT((ipp->ipp_ipv4_options_len & 3) == 0);
13552                 len += ipp->ipp_ipv4_options_len;
13553         }
13554         return (len);
13555 }
13556 
13557 /*
13558  * All-purpose routine to build an IPv4 header with options based
13559  * on the abstract ip_pkt_t.
13560  *
13561  * The caller has to set the source and destination address as well as
13562  * ipha_length. The caller has to massage any source route and compensate
13563  * for the ULP pseudo-header checksum due to the source route.
13564  */
13565 void
13566 ip_build_hdrs_v4(uchar_t *buf, uint_t buf_len, const ip_pkt_t *ipp,
13567     uint8_t protocol)
13568 {
13569         ipha_t  *ipha = (ipha_t *)buf;
13570         uint8_t *cp;
13571 
13572         /* Initialize IPv4 header */
13573         ipha->ipha_type_of_service = ipp->ipp_type_of_service;
13574         ipha->ipha_length = 0;       /* Caller will set later */
13575         ipha->ipha_ident = 0;
13576         ipha->ipha_fragment_offset_and_flags = 0;
13577         ipha->ipha_ttl = ipp->ipp_unicast_hops;
13578         ipha->ipha_protocol = protocol;
13579         ipha->ipha_hdr_checksum = 0;
13580 
13581         if ((ipp->ipp_fields & IPPF_ADDR) &&
13582             IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr))
13583                 ipha->ipha_src = ipp->ipp_addr_v4;
13584 
13585         cp = (uint8_t *)&ipha[1];
13586         if (ipp->ipp_fields & IPPF_LABEL_V4) {
13587                 ASSERT(ipp->ipp_label_len_v4 != 0);
13588                 bcopy(ipp->ipp_label_v4, cp, ipp->ipp_label_len_v4);
13589                 cp += ipp->ipp_label_len_v4;
13590                 /* We need to round up here */
13591                 while ((uintptr_t)cp & 0x3) {
13592                         *cp++ = IPOPT_NOP;
13593                 }
13594         }
13595 
13596         if (ipp->ipp_fields & IPPF_IPV4_OPTIONS) {
13597                 ASSERT(ipp->ipp_ipv4_options_len != 0);
13598                 ASSERT((ipp->ipp_ipv4_options_len & 3) == 0);
13599                 bcopy(ipp->ipp_ipv4_options, cp, ipp->ipp_ipv4_options_len);
13600                 cp += ipp->ipp_ipv4_options_len;
13601         }
13602         ipha->ipha_version_and_hdr_length =
13603             (uint8_t)((IP_VERSION << 4) + buf_len / 4);
13604 
13605         ASSERT((int)(cp - buf) == buf_len);
13606 }
13607 
13608 /* Allocate the private structure */
13609 static int
13610 ip_priv_alloc(void **bufp)
13611 {
13612         void    *buf;
13613 
13614         if ((buf = kmem_alloc(sizeof (ip_priv_t), KM_NOSLEEP)) == NULL)
13615                 return (ENOMEM);
13616 
13617         *bufp = buf;
13618         return (0);
13619 }
13620 
13621 /* Function to delete the private structure */
13622 void
13623 ip_priv_free(void *buf)
13624 {
13625         ASSERT(buf != NULL);
13626         kmem_free(buf, sizeof (ip_priv_t));
13627 }
13628 
13629 /*
13630  * The entry point for IPPF processing.
13631  * If the classifier (IPGPC_CLASSIFY) is not loaded and configured, the
13632  * routine just returns.
13633  *
13634  * When called, ip_process generates an ipp_packet_t structure
13635  * which holds the state information for this packet and invokes the
13636  * the classifier (via ipp_packet_process). The classification, depending on
13637  * configured filters, results in a list of actions for this packet. Invoking
13638  * an action may cause the packet to be dropped, in which case we return NULL.
13639  * proc indicates the callout position for
13640  * this packet and ill is the interface this packet arrived on or will leave
13641  * on (inbound and outbound resp.).
13642  *
13643  * We do the processing on the rill (mapped to the upper if ipmp), but MIB
13644  * on the ill corrsponding to the destination IP address.
13645  */
13646 mblk_t *
13647 ip_process(ip_proc_t proc, mblk_t *mp, ill_t *rill, ill_t *ill)
13648 {
13649         ip_priv_t       *priv;
13650         ipp_action_id_t aid;
13651         int             rc = 0;
13652         ipp_packet_t    *pp;
13653 
13654         /* If the classifier is not loaded, return  */
13655         if ((aid = ipp_action_lookup(IPGPC_CLASSIFY)) == IPP_ACTION_INVAL) {
13656                 return (mp);
13657         }
13658 
13659         ASSERT(mp != NULL);
13660 
13661         /* Allocate the packet structure */
13662         rc = ipp_packet_alloc(&pp, "ip", aid);
13663         if (rc != 0)
13664                 goto drop;
13665 
13666         /* Allocate the private structure */
13667         rc = ip_priv_alloc((void **)&priv);
13668         if (rc != 0) {
13669                 ipp_packet_free(pp);
13670                 goto drop;
13671         }
13672         priv->proc = proc;
13673         priv->ill_index = ill_get_upper_ifindex(rill);
13674 
13675         ipp_packet_set_private(pp, priv, ip_priv_free);
13676         ipp_packet_set_data(pp, mp);
13677 
13678         /* Invoke the classifier */
13679         rc = ipp_packet_process(&pp);
13680         if (pp != NULL) {
13681                 mp = ipp_packet_get_data(pp);
13682                 ipp_packet_free(pp);
13683                 if (rc != 0)
13684                         goto drop;
13685                 return (mp);
13686         } else {
13687                 /* No mp to trace in ip_drop_input/ip_drop_output  */
13688                 mp = NULL;
13689         }
13690 drop:
13691         if (proc == IPP_LOCAL_IN || proc == IPP_FWD_IN) {
13692                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
13693                 ip_drop_input("ip_process", mp, ill);
13694         } else {
13695                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
13696                 ip_drop_output("ip_process", mp, ill);
13697         }
13698         freemsg(mp);
13699         return (NULL);
13700 }
13701 
13702 /*
13703  * Propagate a multicast group membership operation (add/drop) on
13704  * all the interfaces crossed by the related multirt routes.
13705  * The call is considered successful if the operation succeeds
13706  * on at least one interface.
13707  *
13708  * This assumes that a set of IRE_HOST/RTF_MULTIRT has been created for the
13709  * multicast addresses with the ire argument being the first one.
13710  * We walk the bucket to find all the of those.
13711  *
13712  * Common to IPv4 and IPv6.
13713  */
13714 static int
13715 ip_multirt_apply_membership(int (*fn)(conn_t *, boolean_t,
13716     const in6_addr_t *, ipaddr_t, uint_t, mcast_record_t, const in6_addr_t *),
13717     ire_t *ire, conn_t *connp, boolean_t checkonly, const in6_addr_t *v6group,
13718     mcast_record_t fmode, const in6_addr_t *v6src)
13719 {
13720         ire_t           *ire_gw;
13721         irb_t           *irb;
13722         int             ifindex;
13723         int             error = 0;
13724         int             result;
13725         ip_stack_t      *ipst = ire->ire_ipst;
13726         ipaddr_t        group;
13727         boolean_t       isv6;
13728         int             match_flags;
13729 
13730         if (IN6_IS_ADDR_V4MAPPED(v6group)) {
13731                 IN6_V4MAPPED_TO_IPADDR(v6group, group);
13732                 isv6 = B_FALSE;
13733         } else {
13734                 isv6 = B_TRUE;
13735         }
13736 
13737         irb = ire->ire_bucket;
13738         ASSERT(irb != NULL);
13739 
13740         result = 0;
13741         irb_refhold(irb);
13742         for (; ire != NULL; ire = ire->ire_next) {
13743                 if ((ire->ire_flags & RTF_MULTIRT) == 0)
13744                         continue;
13745 
13746                 /* We handle -ifp routes by matching on the ill if set */
13747                 match_flags = MATCH_IRE_TYPE;
13748                 if (ire->ire_ill != NULL)
13749                         match_flags |= MATCH_IRE_ILL;
13750 
13751                 if (isv6) {
13752                         if (!IN6_ARE_ADDR_EQUAL(&ire->ire_addr_v6, v6group))
13753                                 continue;
13754 
13755                         ire_gw = ire_ftable_lookup_v6(&ire->ire_gateway_addr_v6,
13756                             0, 0, IRE_INTERFACE, ire->ire_ill, ALL_ZONES, NULL,
13757                             match_flags, 0, ipst, NULL);
13758                 } else {
13759                         if (ire->ire_addr != group)
13760                                 continue;
13761 
13762                         ire_gw = ire_ftable_lookup_v4(ire->ire_gateway_addr,
13763                             0, 0, IRE_INTERFACE, ire->ire_ill, ALL_ZONES, NULL,
13764                             match_flags, 0, ipst, NULL);
13765                 }
13766                 /* No interface route exists for the gateway; skip this ire. */
13767                 if (ire_gw == NULL)
13768                         continue;
13769                 if (ire_gw->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
13770                         ire_refrele(ire_gw);
13771                         continue;
13772                 }
13773                 ASSERT(ire_gw->ire_ill != NULL);     /* IRE_INTERFACE */
13774                 ifindex = ire_gw->ire_ill->ill_phyint->phyint_ifindex;
13775 
13776                 /*
13777                  * The operation is considered a success if
13778                  * it succeeds at least once on any one interface.
13779                  */
13780                 error = fn(connp, checkonly, v6group, INADDR_ANY, ifindex,
13781                     fmode, v6src);
13782                 if (error == 0)
13783                         result = CGTP_MCAST_SUCCESS;
13784 
13785                 ire_refrele(ire_gw);
13786         }
13787         irb_refrele(irb);
13788         /*
13789          * Consider the call as successful if we succeeded on at least
13790          * one interface. Otherwise, return the last encountered error.
13791          */
13792         return (result == CGTP_MCAST_SUCCESS ? 0 : error);
13793 }
13794 
13795 /*
13796  * Return the expected CGTP hooks version number.
13797  */
13798 int
13799 ip_cgtp_filter_supported(void)
13800 {
13801         return (ip_cgtp_filter_rev);
13802 }
13803 
13804 /*
13805  * CGTP hooks can be registered by invoking this function.
13806  * Checks that the version number matches.
13807  */
13808 int
13809 ip_cgtp_filter_register(netstackid_t stackid, cgtp_filter_ops_t *ops)
13810 {
13811         netstack_t *ns;
13812         ip_stack_t *ipst;
13813 
13814         if (ops->cfo_filter_rev != CGTP_FILTER_REV)
13815                 return (ENOTSUP);
13816 
13817         ns = netstack_find_by_stackid(stackid);
13818         if (ns == NULL)
13819                 return (EINVAL);
13820         ipst = ns->netstack_ip;
13821         ASSERT(ipst != NULL);
13822 
13823         if (ipst->ips_ip_cgtp_filter_ops != NULL) {
13824                 netstack_rele(ns);
13825                 return (EALREADY);
13826         }
13827 
13828         ipst->ips_ip_cgtp_filter_ops = ops;
13829 
13830         ill_set_inputfn_all(ipst);
13831 
13832         netstack_rele(ns);
13833         return (0);
13834 }
13835 
13836 /*
13837  * CGTP hooks can be unregistered by invoking this function.
13838  * Returns ENXIO if there was no registration.
13839  * Returns EBUSY if the ndd variable has not been turned off.
13840  */
13841 int
13842 ip_cgtp_filter_unregister(netstackid_t stackid)
13843 {
13844         netstack_t *ns;
13845         ip_stack_t *ipst;
13846 
13847         ns = netstack_find_by_stackid(stackid);
13848         if (ns == NULL)
13849                 return (EINVAL);
13850         ipst = ns->netstack_ip;
13851         ASSERT(ipst != NULL);
13852 
13853         if (ipst->ips_ip_cgtp_filter) {
13854                 netstack_rele(ns);
13855                 return (EBUSY);
13856         }
13857 
13858         if (ipst->ips_ip_cgtp_filter_ops == NULL) {
13859                 netstack_rele(ns);
13860                 return (ENXIO);
13861         }
13862         ipst->ips_ip_cgtp_filter_ops = NULL;
13863 
13864         ill_set_inputfn_all(ipst);
13865 
13866         netstack_rele(ns);
13867         return (0);
13868 }
13869 
13870 /*
13871  * Check whether there is a CGTP filter registration.
13872  * Returns non-zero if there is a registration, otherwise returns zero.
13873  * Note: returns zero if bad stackid.
13874  */
13875 int
13876 ip_cgtp_filter_is_registered(netstackid_t stackid)
13877 {
13878         netstack_t *ns;
13879         ip_stack_t *ipst;
13880         int ret;
13881 
13882         ns = netstack_find_by_stackid(stackid);
13883         if (ns == NULL)
13884                 return (0);
13885         ipst = ns->netstack_ip;
13886         ASSERT(ipst != NULL);
13887 
13888         if (ipst->ips_ip_cgtp_filter_ops != NULL)
13889                 ret = 1;
13890         else
13891                 ret = 0;
13892 
13893         netstack_rele(ns);
13894         return (ret);
13895 }
13896 
13897 static int
13898 ip_squeue_switch(int val)
13899 {
13900         int rval;
13901 
13902         switch (val) {
13903         case IP_SQUEUE_ENTER_NODRAIN:
13904                 rval = SQ_NODRAIN;
13905                 break;
13906         case IP_SQUEUE_ENTER:
13907                 rval = SQ_PROCESS;
13908                 break;
13909         case IP_SQUEUE_FILL:
13910         default:
13911                 rval = SQ_FILL;
13912                 break;
13913         }
13914         return (rval);
13915 }
13916 
13917 static void *
13918 ip_kstat2_init(netstackid_t stackid, ip_stat_t *ip_statisticsp)
13919 {
13920         kstat_t *ksp;
13921 
13922         ip_stat_t template = {
13923                 { "ip_udp_fannorm",             KSTAT_DATA_UINT64 },
13924                 { "ip_udp_fanmb",               KSTAT_DATA_UINT64 },
13925                 { "ip_recv_pullup",             KSTAT_DATA_UINT64 },
13926                 { "ip_db_ref",                  KSTAT_DATA_UINT64 },
13927                 { "ip_notaligned",              KSTAT_DATA_UINT64 },
13928                 { "ip_multimblk",               KSTAT_DATA_UINT64 },
13929                 { "ip_opt",                     KSTAT_DATA_UINT64 },
13930                 { "ipsec_proto_ahesp",          KSTAT_DATA_UINT64 },
13931                 { "ip_conn_flputbq",            KSTAT_DATA_UINT64 },
13932                 { "ip_conn_walk_drain",         KSTAT_DATA_UINT64 },
13933                 { "ip_out_sw_cksum",            KSTAT_DATA_UINT64 },
13934                 { "ip_out_sw_cksum_bytes",      KSTAT_DATA_UINT64 },
13935                 { "ip_in_sw_cksum",             KSTAT_DATA_UINT64 },
13936                 { "ip_ire_reclaim_calls",       KSTAT_DATA_UINT64 },
13937                 { "ip_ire_reclaim_deleted",     KSTAT_DATA_UINT64 },
13938                 { "ip_nce_reclaim_calls",       KSTAT_DATA_UINT64 },
13939                 { "ip_nce_reclaim_deleted",     KSTAT_DATA_UINT64 },
13940                 { "ip_dce_reclaim_calls",       KSTAT_DATA_UINT64 },
13941                 { "ip_dce_reclaim_deleted",     KSTAT_DATA_UINT64 },
13942                 { "ip_tcp_in_full_hw_cksum_err",        KSTAT_DATA_UINT64 },
13943                 { "ip_tcp_in_part_hw_cksum_err",        KSTAT_DATA_UINT64 },
13944                 { "ip_tcp_in_sw_cksum_err",             KSTAT_DATA_UINT64 },
13945                 { "ip_udp_in_full_hw_cksum_err",        KSTAT_DATA_UINT64 },
13946                 { "ip_udp_in_part_hw_cksum_err",        KSTAT_DATA_UINT64 },
13947                 { "ip_udp_in_sw_cksum_err",     KSTAT_DATA_UINT64 },
13948                 { "conn_in_recvdstaddr",        KSTAT_DATA_UINT64 },
13949                 { "conn_in_recvopts",           KSTAT_DATA_UINT64 },
13950                 { "conn_in_recvif",             KSTAT_DATA_UINT64 },
13951                 { "conn_in_recvslla",           KSTAT_DATA_UINT64 },
13952                 { "conn_in_recvucred",          KSTAT_DATA_UINT64 },
13953                 { "conn_in_recvttl",            KSTAT_DATA_UINT64 },
13954                 { "conn_in_recvhopopts",        KSTAT_DATA_UINT64 },
13955                 { "conn_in_recvhoplimit",       KSTAT_DATA_UINT64 },
13956                 { "conn_in_recvdstopts",        KSTAT_DATA_UINT64 },
13957                 { "conn_in_recvrthdrdstopts",   KSTAT_DATA_UINT64 },
13958                 { "conn_in_recvrthdr",          KSTAT_DATA_UINT64 },
13959                 { "conn_in_recvpktinfo",        KSTAT_DATA_UINT64 },
13960                 { "conn_in_recvtclass",         KSTAT_DATA_UINT64 },
13961                 { "conn_in_timestamp",          KSTAT_DATA_UINT64 },
13962         };
13963 
13964         ksp = kstat_create_netstack("ip", 0, "ipstat", "net",
13965             KSTAT_TYPE_NAMED, sizeof (template) / sizeof (kstat_named_t),
13966             KSTAT_FLAG_VIRTUAL, stackid);
13967 
13968         if (ksp == NULL)
13969                 return (NULL);
13970 
13971         bcopy(&template, ip_statisticsp, sizeof (template));
13972         ksp->ks_data = (void *)ip_statisticsp;
13973         ksp->ks_private = (void *)(uintptr_t)stackid;
13974 
13975         kstat_install(ksp);
13976         return (ksp);
13977 }
13978 
13979 static void
13980 ip_kstat2_fini(netstackid_t stackid, kstat_t *ksp)
13981 {
13982         if (ksp != NULL) {
13983                 ASSERT(stackid == (netstackid_t)(uintptr_t)ksp->ks_private);
13984                 kstat_delete_netstack(ksp, stackid);
13985         }
13986 }
13987 
13988 static void *
13989 ip_kstat_init(netstackid_t stackid, ip_stack_t *ipst)
13990 {
13991         kstat_t *ksp;
13992 
13993         ip_named_kstat_t template = {
13994                 { "forwarding",         KSTAT_DATA_UINT32, 0 },
13995                 { "defaultTTL",         KSTAT_DATA_UINT32, 0 },
13996                 { "inReceives",         KSTAT_DATA_UINT64, 0 },
13997                 { "inHdrErrors",        KSTAT_DATA_UINT32, 0 },
13998                 { "inAddrErrors",       KSTAT_DATA_UINT32, 0 },
13999                 { "forwDatagrams",      KSTAT_DATA_UINT64, 0 },
14000                 { "inUnknownProtos",    KSTAT_DATA_UINT32, 0 },
14001                 { "inDiscards",         KSTAT_DATA_UINT32, 0 },
14002                 { "inDelivers",         KSTAT_DATA_UINT64, 0 },
14003                 { "outRequests",        KSTAT_DATA_UINT64, 0 },
14004                 { "outDiscards",        KSTAT_DATA_UINT32, 0 },
14005                 { "outNoRoutes",        KSTAT_DATA_UINT32, 0 },
14006                 { "reasmTimeout",       KSTAT_DATA_UINT32, 0 },
14007                 { "reasmReqds",         KSTAT_DATA_UINT32, 0 },
14008                 { "reasmOKs",           KSTAT_DATA_UINT32, 0 },
14009                 { "reasmFails",         KSTAT_DATA_UINT32, 0 },
14010                 { "fragOKs",            KSTAT_DATA_UINT32, 0 },
14011                 { "fragFails",          KSTAT_DATA_UINT32, 0 },
14012                 { "fragCreates",        KSTAT_DATA_UINT32, 0 },
14013                 { "addrEntrySize",      KSTAT_DATA_INT32, 0 },
14014                 { "routeEntrySize",     KSTAT_DATA_INT32, 0 },
14015                 { "netToMediaEntrySize",        KSTAT_DATA_INT32, 0 },
14016                 { "routingDiscards",    KSTAT_DATA_UINT32, 0 },
14017                 { "inErrs",             KSTAT_DATA_UINT32, 0 },
14018                 { "noPorts",            KSTAT_DATA_UINT32, 0 },
14019                 { "inCksumErrs",        KSTAT_DATA_UINT32, 0 },
14020                 { "reasmDuplicates",    KSTAT_DATA_UINT32, 0 },
14021                 { "reasmPartDups",      KSTAT_DATA_UINT32, 0 },
14022                 { "forwProhibits",      KSTAT_DATA_UINT32, 0 },
14023                 { "udpInCksumErrs",     KSTAT_DATA_UINT32, 0 },
14024                 { "udpInOverflows",     KSTAT_DATA_UINT32, 0 },
14025                 { "rawipInOverflows",   KSTAT_DATA_UINT32, 0 },
14026                 { "ipsecInSucceeded",   KSTAT_DATA_UINT32, 0 },
14027                 { "ipsecInFailed",      KSTAT_DATA_INT32, 0 },
14028                 { "memberEntrySize",    KSTAT_DATA_INT32, 0 },
14029                 { "inIPv6",             KSTAT_DATA_UINT32, 0 },
14030                 { "outIPv6",            KSTAT_DATA_UINT32, 0 },
14031                 { "outSwitchIPv6",      KSTAT_DATA_UINT32, 0 },
14032         };
14033 
14034         ksp = kstat_create_netstack("ip", 0, "ip", "mib2", KSTAT_TYPE_NAMED,
14035             NUM_OF_FIELDS(ip_named_kstat_t), 0, stackid);
14036         if (ksp == NULL || ksp->ks_data == NULL)
14037                 return (NULL);
14038 
14039         template.forwarding.value.ui32 = WE_ARE_FORWARDING(ipst) ? 1:2;
14040         template.defaultTTL.value.ui32 = (uint32_t)ipst->ips_ip_def_ttl;
14041         template.reasmTimeout.value.ui32 = ipst->ips_ip_reassembly_timeout;
14042         template.addrEntrySize.value.i32 = sizeof (mib2_ipAddrEntry_t);
14043         template.routeEntrySize.value.i32 = sizeof (mib2_ipRouteEntry_t);
14044 
14045         template.netToMediaEntrySize.value.i32 =
14046             sizeof (mib2_ipNetToMediaEntry_t);
14047 
14048         template.memberEntrySize.value.i32 = sizeof (ipv6_member_t);
14049 
14050         bcopy(&template, ksp->ks_data, sizeof (template));
14051         ksp->ks_update = ip_kstat_update;
14052         ksp->ks_private = (void *)(uintptr_t)stackid;
14053 
14054         kstat_install(ksp);
14055         return (ksp);
14056 }
14057 
14058 static void
14059 ip_kstat_fini(netstackid_t stackid, kstat_t *ksp)
14060 {
14061         if (ksp != NULL) {
14062                 ASSERT(stackid == (netstackid_t)(uintptr_t)ksp->ks_private);
14063                 kstat_delete_netstack(ksp, stackid);
14064         }
14065 }
14066 
14067 static int
14068 ip_kstat_update(kstat_t *kp, int rw)
14069 {
14070         ip_named_kstat_t *ipkp;
14071         mib2_ipIfStatsEntry_t ipmib;
14072         ill_walk_context_t ctx;
14073         ill_t *ill;
14074         netstackid_t    stackid = (zoneid_t)(uintptr_t)kp->ks_private;
14075         netstack_t      *ns;
14076         ip_stack_t      *ipst;
14077 
14078         if (kp->ks_data == NULL)
14079                 return (EIO);
14080 
14081         if (rw == KSTAT_WRITE)
14082                 return (EACCES);
14083 
14084         ns = netstack_find_by_stackid(stackid);
14085         if (ns == NULL)
14086                 return (-1);
14087         ipst = ns->netstack_ip;
14088         if (ipst == NULL) {
14089                 netstack_rele(ns);
14090                 return (-1);
14091         }
14092         ipkp = (ip_named_kstat_t *)kp->ks_data;
14093 
14094         bcopy(&ipst->ips_ip_mib, &ipmib, sizeof (ipmib));
14095         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
14096         ill = ILL_START_WALK_V4(&ctx, ipst);
14097         for (; ill != NULL; ill = ill_next(&ctx, ill))
14098                 ip_mib2_add_ip_stats(&ipmib, ill->ill_ip_mib);
14099         rw_exit(&ipst->ips_ill_g_lock);
14100 
14101         ipkp->forwarding.value.ui32 =                ipmib.ipIfStatsForwarding;
14102         ipkp->defaultTTL.value.ui32 =                ipmib.ipIfStatsDefaultTTL;
14103         ipkp->inReceives.value.ui64 =                ipmib.ipIfStatsHCInReceives;
14104         ipkp->inHdrErrors.value.ui32 =               ipmib.ipIfStatsInHdrErrors;
14105         ipkp->inAddrErrors.value.ui32 =              ipmib.ipIfStatsInAddrErrors;
14106         ipkp->forwDatagrams.value.ui64 = ipmib.ipIfStatsHCOutForwDatagrams;
14107         ipkp->inUnknownProtos.value.ui32 =   ipmib.ipIfStatsInUnknownProtos;
14108         ipkp->inDiscards.value.ui32 =                ipmib.ipIfStatsInDiscards;
14109         ipkp->inDelivers.value.ui64 =                ipmib.ipIfStatsHCInDelivers;
14110         ipkp->outRequests.value.ui64 =               ipmib.ipIfStatsHCOutRequests;
14111         ipkp->outDiscards.value.ui32 =               ipmib.ipIfStatsOutDiscards;
14112         ipkp->outNoRoutes.value.ui32 =               ipmib.ipIfStatsOutNoRoutes;
14113         ipkp->reasmTimeout.value.ui32 =              ipst->ips_ip_reassembly_timeout;
14114         ipkp->reasmReqds.value.ui32 =                ipmib.ipIfStatsReasmReqds;
14115         ipkp->reasmOKs.value.ui32 =          ipmib.ipIfStatsReasmOKs;
14116         ipkp->reasmFails.value.ui32 =                ipmib.ipIfStatsReasmFails;
14117         ipkp->fragOKs.value.ui32 =           ipmib.ipIfStatsOutFragOKs;
14118         ipkp->fragFails.value.ui32 =         ipmib.ipIfStatsOutFragFails;
14119         ipkp->fragCreates.value.ui32 =               ipmib.ipIfStatsOutFragCreates;
14120 
14121         ipkp->routingDiscards.value.ui32 =   0;
14122         ipkp->inErrs.value.ui32 =            ipmib.tcpIfStatsInErrs;
14123         ipkp->noPorts.value.ui32 =           ipmib.udpIfStatsNoPorts;
14124         ipkp->inCksumErrs.value.ui32 =               ipmib.ipIfStatsInCksumErrs;
14125         ipkp->reasmDuplicates.value.ui32 =   ipmib.ipIfStatsReasmDuplicates;
14126         ipkp->reasmPartDups.value.ui32 =     ipmib.ipIfStatsReasmPartDups;
14127         ipkp->forwProhibits.value.ui32 =     ipmib.ipIfStatsForwProhibits;
14128         ipkp->udpInCksumErrs.value.ui32 =    ipmib.udpIfStatsInCksumErrs;
14129         ipkp->udpInOverflows.value.ui32 =    ipmib.udpIfStatsInOverflows;
14130         ipkp->rawipInOverflows.value.ui32 =  ipmib.rawipIfStatsInOverflows;
14131         ipkp->ipsecInSucceeded.value.ui32 =  ipmib.ipsecIfStatsInSucceeded;
14132         ipkp->ipsecInFailed.value.i32 =              ipmib.ipsecIfStatsInFailed;
14133 
14134         ipkp->inIPv6.value.ui32 =    ipmib.ipIfStatsInWrongIPVersion;
14135         ipkp->outIPv6.value.ui32 =   ipmib.ipIfStatsOutWrongIPVersion;
14136         ipkp->outSwitchIPv6.value.ui32 = ipmib.ipIfStatsOutSwitchIPVersion;
14137 
14138         netstack_rele(ns);
14139 
14140         return (0);
14141 }
14142 
14143 static void *
14144 icmp_kstat_init(netstackid_t stackid)
14145 {
14146         kstat_t *ksp;
14147 
14148         icmp_named_kstat_t template = {
14149                 { "inMsgs",             KSTAT_DATA_UINT32 },
14150                 { "inErrors",           KSTAT_DATA_UINT32 },
14151                 { "inDestUnreachs",     KSTAT_DATA_UINT32 },
14152                 { "inTimeExcds",        KSTAT_DATA_UINT32 },
14153                 { "inParmProbs",        KSTAT_DATA_UINT32 },
14154                 { "inSrcQuenchs",       KSTAT_DATA_UINT32 },
14155                 { "inRedirects",        KSTAT_DATA_UINT32 },
14156                 { "inEchos",            KSTAT_DATA_UINT32 },
14157                 { "inEchoReps",         KSTAT_DATA_UINT32 },
14158                 { "inTimestamps",       KSTAT_DATA_UINT32 },
14159                 { "inTimestampReps",    KSTAT_DATA_UINT32 },
14160                 { "inAddrMasks",        KSTAT_DATA_UINT32 },
14161                 { "inAddrMaskReps",     KSTAT_DATA_UINT32 },
14162                 { "outMsgs",            KSTAT_DATA_UINT32 },
14163                 { "outErrors",          KSTAT_DATA_UINT32 },
14164                 { "outDestUnreachs",    KSTAT_DATA_UINT32 },
14165                 { "outTimeExcds",       KSTAT_DATA_UINT32 },
14166                 { "outParmProbs",       KSTAT_DATA_UINT32 },
14167                 { "outSrcQuenchs",      KSTAT_DATA_UINT32 },
14168                 { "outRedirects",       KSTAT_DATA_UINT32 },
14169                 { "outEchos",           KSTAT_DATA_UINT32 },
14170                 { "outEchoReps",        KSTAT_DATA_UINT32 },
14171                 { "outTimestamps",      KSTAT_DATA_UINT32 },
14172                 { "outTimestampReps",   KSTAT_DATA_UINT32 },
14173                 { "outAddrMasks",       KSTAT_DATA_UINT32 },
14174                 { "outAddrMaskReps",    KSTAT_DATA_UINT32 },
14175                 { "inChksumErrs",       KSTAT_DATA_UINT32 },
14176                 { "inUnknowns",         KSTAT_DATA_UINT32 },
14177                 { "inFragNeeded",       KSTAT_DATA_UINT32 },
14178                 { "outFragNeeded",      KSTAT_DATA_UINT32 },
14179                 { "outDrops",           KSTAT_DATA_UINT32 },
14180                 { "inOverFlows",        KSTAT_DATA_UINT32 },
14181                 { "inBadRedirects",     KSTAT_DATA_UINT32 },
14182         };
14183 
14184         ksp = kstat_create_netstack("ip", 0, "icmp", "mib2", KSTAT_TYPE_NAMED,
14185             NUM_OF_FIELDS(icmp_named_kstat_t), 0, stackid);
14186         if (ksp == NULL || ksp->ks_data == NULL)
14187                 return (NULL);
14188 
14189         bcopy(&template, ksp->ks_data, sizeof (template));
14190 
14191         ksp->ks_update = icmp_kstat_update;
14192         ksp->ks_private = (void *)(uintptr_t)stackid;
14193 
14194         kstat_install(ksp);
14195         return (ksp);
14196 }
14197 
14198 static void
14199 icmp_kstat_fini(netstackid_t stackid, kstat_t *ksp)
14200 {
14201         if (ksp != NULL) {
14202                 ASSERT(stackid == (netstackid_t)(uintptr_t)ksp->ks_private);
14203                 kstat_delete_netstack(ksp, stackid);
14204         }
14205 }
14206 
14207 static int
14208 icmp_kstat_update(kstat_t *kp, int rw)
14209 {
14210         icmp_named_kstat_t *icmpkp;
14211         netstackid_t    stackid = (zoneid_t)(uintptr_t)kp->ks_private;
14212         netstack_t      *ns;
14213         ip_stack_t      *ipst;
14214 
14215         if (kp->ks_data == NULL)
14216                 return (EIO);
14217 
14218         if (rw == KSTAT_WRITE)
14219                 return (EACCES);
14220 
14221         ns = netstack_find_by_stackid(stackid);
14222         if (ns == NULL)
14223                 return (-1);
14224         ipst = ns->netstack_ip;
14225         if (ipst == NULL) {
14226                 netstack_rele(ns);
14227                 return (-1);
14228         }
14229         icmpkp = (icmp_named_kstat_t *)kp->ks_data;
14230 
14231         icmpkp->inMsgs.value.ui32 =      ipst->ips_icmp_mib.icmpInMsgs;
14232         icmpkp->inErrors.value.ui32 =            ipst->ips_icmp_mib.icmpInErrors;
14233         icmpkp->inDestUnreachs.value.ui32 =
14234             ipst->ips_icmp_mib.icmpInDestUnreachs;
14235         icmpkp->inTimeExcds.value.ui32 =    ipst->ips_icmp_mib.icmpInTimeExcds;
14236         icmpkp->inParmProbs.value.ui32 =    ipst->ips_icmp_mib.icmpInParmProbs;
14237         icmpkp->inSrcQuenchs.value.ui32 =   ipst->ips_icmp_mib.icmpInSrcQuenchs;
14238         icmpkp->inRedirects.value.ui32 =    ipst->ips_icmp_mib.icmpInRedirects;
14239         icmpkp->inEchos.value.ui32 =     ipst->ips_icmp_mib.icmpInEchos;
14240         icmpkp->inEchoReps.value.ui32 =          ipst->ips_icmp_mib.icmpInEchoReps;
14241         icmpkp->inTimestamps.value.ui32 =   ipst->ips_icmp_mib.icmpInTimestamps;
14242         icmpkp->inTimestampReps.value.ui32 =
14243             ipst->ips_icmp_mib.icmpInTimestampReps;
14244         icmpkp->inAddrMasks.value.ui32 =    ipst->ips_icmp_mib.icmpInAddrMasks;
14245         icmpkp->inAddrMaskReps.value.ui32 =
14246             ipst->ips_icmp_mib.icmpInAddrMaskReps;
14247         icmpkp->outMsgs.value.ui32 =     ipst->ips_icmp_mib.icmpOutMsgs;
14248         icmpkp->outErrors.value.ui32 =           ipst->ips_icmp_mib.icmpOutErrors;
14249         icmpkp->outDestUnreachs.value.ui32 =
14250             ipst->ips_icmp_mib.icmpOutDestUnreachs;
14251         icmpkp->outTimeExcds.value.ui32 =   ipst->ips_icmp_mib.icmpOutTimeExcds;
14252         icmpkp->outParmProbs.value.ui32 =   ipst->ips_icmp_mib.icmpOutParmProbs;
14253         icmpkp->outSrcQuenchs.value.ui32 =
14254             ipst->ips_icmp_mib.icmpOutSrcQuenchs;
14255         icmpkp->outRedirects.value.ui32 =   ipst->ips_icmp_mib.icmpOutRedirects;
14256         icmpkp->outEchos.value.ui32 =            ipst->ips_icmp_mib.icmpOutEchos;
14257         icmpkp->outEchoReps.value.ui32 =    ipst->ips_icmp_mib.icmpOutEchoReps;
14258         icmpkp->outTimestamps.value.ui32 =
14259             ipst->ips_icmp_mib.icmpOutTimestamps;
14260         icmpkp->outTimestampReps.value.ui32 =
14261             ipst->ips_icmp_mib.icmpOutTimestampReps;
14262         icmpkp->outAddrMasks.value.ui32 =
14263             ipst->ips_icmp_mib.icmpOutAddrMasks;
14264         icmpkp->outAddrMaskReps.value.ui32 =
14265             ipst->ips_icmp_mib.icmpOutAddrMaskReps;
14266         icmpkp->inCksumErrs.value.ui32 =    ipst->ips_icmp_mib.icmpInCksumErrs;
14267         icmpkp->inUnknowns.value.ui32 =          ipst->ips_icmp_mib.icmpInUnknowns;
14268         icmpkp->inFragNeeded.value.ui32 =   ipst->ips_icmp_mib.icmpInFragNeeded;
14269         icmpkp->outFragNeeded.value.ui32 =
14270             ipst->ips_icmp_mib.icmpOutFragNeeded;
14271         icmpkp->outDrops.value.ui32 =            ipst->ips_icmp_mib.icmpOutDrops;
14272         icmpkp->inOverflows.value.ui32 =    ipst->ips_icmp_mib.icmpInOverflows;
14273         icmpkp->inBadRedirects.value.ui32 =
14274             ipst->ips_icmp_mib.icmpInBadRedirects;
14275 
14276         netstack_rele(ns);
14277         return (0);
14278 }
14279 
14280 /*
14281  * This is the fanout function for raw socket opened for SCTP.  Note
14282  * that it is called after SCTP checks that there is no socket which
14283  * wants a packet.  Then before SCTP handles this out of the blue packet,
14284  * this function is called to see if there is any raw socket for SCTP.
14285  * If there is and it is bound to the correct address, the packet will
14286  * be sent to that socket.  Note that only one raw socket can be bound to
14287  * a port.  This is assured in ipcl_sctp_hash_insert();
14288  */
14289 void
14290 ip_fanout_sctp_raw(mblk_t *mp, ipha_t *ipha, ip6_t *ip6h, uint32_t ports,
14291     ip_recv_attr_t *ira)
14292 {
14293         conn_t          *connp;
14294         queue_t         *rq;
14295         boolean_t       secure;
14296         ill_t           *ill = ira->ira_ill;
14297         ip_stack_t      *ipst = ill->ill_ipst;
14298         ipsec_stack_t   *ipss = ipst->ips_netstack->netstack_ipsec;
14299         sctp_stack_t    *sctps = ipst->ips_netstack->netstack_sctp;
14300         iaflags_t       iraflags = ira->ira_flags;
14301         ill_t           *rill = ira->ira_rill;
14302 
14303         secure = iraflags & IRAF_IPSEC_SECURE;
14304 
14305         connp = ipcl_classify_raw(mp, IPPROTO_SCTP, ports, ipha, ip6h,
14306             ira, ipst);
14307         if (connp == NULL) {
14308                 /*
14309                  * Although raw sctp is not summed, OOB chunks must be.
14310                  * Drop the packet here if the sctp checksum failed.
14311                  */
14312                 if (iraflags & IRAF_SCTP_CSUM_ERR) {
14313                         SCTPS_BUMP_MIB(sctps, sctpChecksumError);
14314                         freemsg(mp);
14315                         return;
14316                 }
14317                 ira->ira_ill = ira->ira_rill = NULL;
14318                 sctp_ootb_input(mp, ira, ipst);
14319                 ira->ira_ill = ill;
14320                 ira->ira_rill = rill;
14321                 return;
14322         }
14323         rq = connp->conn_rq;
14324         if (IPCL_IS_NONSTR(connp) ? connp->conn_flow_cntrld : !canputnext(rq)) {
14325                 CONN_DEC_REF(connp);
14326                 BUMP_MIB(ill->ill_ip_mib, rawipIfStatsInOverflows);
14327                 freemsg(mp);
14328                 return;
14329         }
14330         if (((iraflags & IRAF_IS_IPV4) ?
14331             CONN_INBOUND_POLICY_PRESENT(connp, ipss) :
14332             CONN_INBOUND_POLICY_PRESENT_V6(connp, ipss)) ||
14333             secure) {
14334                 mp = ipsec_check_inbound_policy(mp, connp, ipha,
14335                     ip6h, ira);
14336                 if (mp == NULL) {
14337                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
14338                         /* Note that mp is NULL */
14339                         ip_drop_input("ipIfStatsInDiscards", mp, ill);
14340                         CONN_DEC_REF(connp);
14341                         return;
14342                 }
14343         }
14344 
14345         if (iraflags & IRAF_ICMP_ERROR) {
14346                 (connp->conn_recvicmp)(connp, mp, NULL, ira);
14347         } else {
14348                 ill_t *rill = ira->ira_rill;
14349 
14350                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCInDelivers);
14351                 /* This is the SOCK_RAW, IPPROTO_SCTP case. */
14352                 ira->ira_ill = ira->ira_rill = NULL;
14353                 (connp->conn_recv)(connp, mp, NULL, ira);
14354                 ira->ira_ill = ill;
14355                 ira->ira_rill = rill;
14356         }
14357         CONN_DEC_REF(connp);
14358 }
14359 
14360 /*
14361  * Free a packet that has the link-layer dl_unitdata_req_t or fast-path
14362  * header before the ip payload.
14363  */
14364 static void
14365 ip_xmit_flowctl_drop(ill_t *ill, mblk_t *mp, boolean_t is_fp_mp, int fp_mp_len)
14366 {
14367         int len = (mp->b_wptr - mp->b_rptr);
14368         mblk_t *ip_mp;
14369 
14370         BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
14371         if (is_fp_mp || len != fp_mp_len) {
14372                 if (len > fp_mp_len) {
14373                         /*
14374                          * fastpath header and ip header in the first mblk
14375                          */
14376                         mp->b_rptr += fp_mp_len;
14377                 } else {
14378                         /*
14379                          * ip_xmit_attach_llhdr had to prepend an mblk to
14380                          * attach the fastpath header before ip header.
14381                          */
14382                         ip_mp = mp->b_cont;
14383                         freeb(mp);
14384                         mp = ip_mp;
14385                         mp->b_rptr += (fp_mp_len - len);
14386                 }
14387         } else {
14388                 ip_mp = mp->b_cont;
14389                 freeb(mp);
14390                 mp = ip_mp;
14391         }
14392         ip_drop_output("ipIfStatsOutDiscards - flow ctl", mp, ill);
14393         freemsg(mp);
14394 }
14395 
14396 /*
14397  * Normal post fragmentation function.
14398  *
14399  * Send a packet using the passed in nce. This handles both IPv4 and IPv6
14400  * using the same state machine.
14401  *
14402  * We return an error on failure. In particular we return EWOULDBLOCK
14403  * when the driver flow controls. In that case this ensures that ip_wsrv runs
14404  * (currently by canputnext failure resulting in backenabling from GLD.)
14405  * This allows the callers of conn_ip_output() to use EWOULDBLOCK as an
14406  * indication that they can flow control until ip_wsrv() tells then to restart.
14407  *
14408  * If the nce passed by caller is incomplete, this function
14409  * queues the packet and if necessary, sends ARP request and bails.
14410  * If the Neighbor Cache passed is fully resolved, we simply prepend
14411  * the link-layer header to the packet, do ipsec hw acceleration
14412  * work if necessary, and send the packet out on the wire.
14413  */
14414 /* ARGSUSED6 */
14415 int
14416 ip_xmit(mblk_t *mp, nce_t *nce, iaflags_t ixaflags, uint_t pkt_len,
14417     uint32_t xmit_hint, zoneid_t szone, zoneid_t nolzid, uintptr_t *ixacookie)
14418 {
14419         queue_t         *wq;
14420         ill_t           *ill = nce->nce_ill;
14421         ip_stack_t      *ipst = ill->ill_ipst;
14422         uint64_t        delta;
14423         boolean_t       isv6 = ill->ill_isv6;
14424         boolean_t       fp_mp;
14425         ncec_t          *ncec = nce->nce_common;
14426         int64_t         now = LBOLT_FASTPATH64;
14427         boolean_t       is_probe;
14428 
14429         DTRACE_PROBE1(ip__xmit, nce_t *, nce);
14430 
14431         ASSERT(mp != NULL);
14432         ASSERT(mp->b_datap->db_type == M_DATA);
14433         ASSERT(pkt_len == msgdsize(mp));
14434 
14435         /*
14436          * If we have already been here and are coming back after ARP/ND.
14437          * the IXAF_NO_TRACE flag is set. We skip FW_HOOKS, DTRACE and ipobs
14438          * in that case since they have seen the packet when it came here
14439          * the first time.
14440          */
14441         if (ixaflags & IXAF_NO_TRACE)
14442                 goto sendit;
14443 
14444         if (ixaflags & IXAF_IS_IPV4) {
14445                 ipha_t *ipha = (ipha_t *)mp->b_rptr;
14446 
14447                 ASSERT(!isv6);
14448                 ASSERT(pkt_len == ntohs(((ipha_t *)mp->b_rptr)->ipha_length));
14449                 if (HOOKS4_INTERESTED_PHYSICAL_OUT(ipst) &&
14450                     !(ixaflags & IXAF_NO_PFHOOK)) {
14451                         int     error;
14452 
14453                         FW_HOOKS(ipst->ips_ip4_physical_out_event,
14454                             ipst->ips_ipv4firewall_physical_out,
14455                             NULL, ill, ipha, mp, mp, 0, ipst, error);
14456                         DTRACE_PROBE1(ip4__physical__out__end,
14457                             mblk_t *, mp);
14458                         if (mp == NULL)
14459                                 return (error);
14460 
14461                         /* The length could have changed */
14462                         pkt_len = msgdsize(mp);
14463                 }
14464                 if (ipst->ips_ip4_observe.he_interested) {
14465                         /*
14466                          * Note that for TX the zoneid is the sending
14467                          * zone, whether or not MLP is in play.
14468                          * Since the szone argument is the IP zoneid (i.e.,
14469                          * zero for exclusive-IP zones) and ipobs wants
14470                          * the system zoneid, we map it here.
14471                          */
14472                         szone = IP_REAL_ZONEID(szone, ipst);
14473 
14474                         /*
14475                          * On the outbound path the destination zone will be
14476                          * unknown as we're sending this packet out on the
14477                          * wire.
14478                          */
14479                         ipobs_hook(mp, IPOBS_HOOK_OUTBOUND, szone, ALL_ZONES,
14480                             ill, ipst);
14481                 }
14482                 DTRACE_IP7(send, mblk_t *, mp,  conn_t *, NULL,
14483                     void_ip_t *, ipha,  __dtrace_ipsr_ill_t *, ill,
14484                     ipha_t *, ipha, ip6_t *, NULL, int, 0);
14485         } else {
14486                 ip6_t *ip6h = (ip6_t *)mp->b_rptr;
14487 
14488                 ASSERT(isv6);
14489                 ASSERT(pkt_len ==
14490                     ntohs(((ip6_t *)mp->b_rptr)->ip6_plen) + IPV6_HDR_LEN);
14491                 if (HOOKS6_INTERESTED_PHYSICAL_OUT(ipst) &&
14492                     !(ixaflags & IXAF_NO_PFHOOK)) {
14493                         int     error;
14494 
14495                         FW_HOOKS6(ipst->ips_ip6_physical_out_event,
14496                             ipst->ips_ipv6firewall_physical_out,
14497                             NULL, ill, ip6h, mp, mp, 0, ipst, error);
14498                         DTRACE_PROBE1(ip6__physical__out__end,
14499                             mblk_t *, mp);
14500                         if (mp == NULL)
14501                                 return (error);
14502 
14503                         /* The length could have changed */
14504                         pkt_len = msgdsize(mp);
14505                 }
14506                 if (ipst->ips_ip6_observe.he_interested) {
14507                         /* See above */
14508                         szone = IP_REAL_ZONEID(szone, ipst);
14509 
14510                         ipobs_hook(mp, IPOBS_HOOK_OUTBOUND, szone, ALL_ZONES,
14511                             ill, ipst);
14512                 }
14513                 DTRACE_IP7(send, mblk_t *, mp,  conn_t *, NULL,
14514                     void_ip_t *, ip6h,  __dtrace_ipsr_ill_t *, ill,
14515                     ipha_t *, NULL, ip6_t *, ip6h, int, 0);
14516         }
14517 
14518 sendit:
14519         /*
14520          * We check the state without a lock because the state can never
14521          * move "backwards" to initial or incomplete.
14522          */
14523         switch (ncec->ncec_state) {
14524         case ND_REACHABLE:
14525         case ND_STALE:
14526         case ND_DELAY:
14527         case ND_PROBE:
14528                 mp = ip_xmit_attach_llhdr(mp, nce);
14529                 if (mp == NULL) {
14530                         /*
14531                          * ip_xmit_attach_llhdr has increased
14532                          * ipIfStatsOutDiscards and called ip_drop_output()
14533                          */
14534                         return (ENOBUFS);
14535                 }
14536                 /*
14537                  * check if nce_fastpath completed and we tagged on a
14538                  * copy of nce_fp_mp in ip_xmit_attach_llhdr().
14539                  */
14540                 fp_mp = (mp->b_datap->db_type == M_DATA);
14541 
14542                 if (fp_mp &&
14543                     (ill->ill_capabilities & ILL_CAPAB_DLD_DIRECT)) {
14544                         ill_dld_direct_t *idd;
14545 
14546                         idd = &ill->ill_dld_capab->idc_direct;
14547                         /*
14548                          * Send the packet directly to DLD, where it
14549                          * may be queued depending on the availability
14550                          * of transmit resources at the media layer.
14551                          * Return value should be taken into
14552                          * account and flow control the TCP.
14553                          */
14554                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCOutTransmits);
14555                         UPDATE_MIB(ill->ill_ip_mib, ipIfStatsHCOutOctets,
14556                             pkt_len);
14557 
14558                         if (ixaflags & IXAF_NO_DEV_FLOW_CTL) {
14559                                 (void) idd->idd_tx_df(idd->idd_tx_dh, mp,
14560                                     (uintptr_t)xmit_hint, IP_DROP_ON_NO_DESC);
14561                         } else {
14562                                 uintptr_t cookie;
14563 
14564                                 if ((cookie = idd->idd_tx_df(idd->idd_tx_dh,
14565                                     mp, (uintptr_t)xmit_hint, 0)) != 0) {
14566                                         if (ixacookie != NULL)
14567                                                 *ixacookie = cookie;
14568                                         return (EWOULDBLOCK);
14569                                 }
14570                         }
14571                 } else {
14572                         wq = ill->ill_wq;
14573 
14574                         if (!(ixaflags & IXAF_NO_DEV_FLOW_CTL) &&
14575                             !canputnext(wq)) {
14576                                 if (ixacookie != NULL)
14577                                         *ixacookie = 0;
14578                                 ip_xmit_flowctl_drop(ill, mp, fp_mp,
14579                                     nce->nce_fp_mp != NULL ?
14580                                     MBLKL(nce->nce_fp_mp) : 0);
14581                                 return (EWOULDBLOCK);
14582                         }
14583                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCOutTransmits);
14584                         UPDATE_MIB(ill->ill_ip_mib, ipIfStatsHCOutOctets,
14585                             pkt_len);
14586                         putnext(wq, mp);
14587                 }
14588 
14589                 /*
14590                  * The rest of this function implements Neighbor Unreachability
14591                  * detection. Determine if the ncec is eligible for NUD.
14592                  */
14593                 if (ncec->ncec_flags & NCE_F_NONUD)
14594                         return (0);
14595 
14596                 ASSERT(ncec->ncec_state != ND_INCOMPLETE);
14597 
14598                 /*
14599                  * Check for upper layer advice
14600                  */
14601                 if (ixaflags & IXAF_REACH_CONF) {
14602                         timeout_id_t tid;
14603 
14604                         /*
14605                          * It should be o.k. to check the state without
14606                          * a lock here, at most we lose an advice.
14607                          */
14608                         ncec->ncec_last = TICK_TO_MSEC(now);
14609                         if (ncec->ncec_state != ND_REACHABLE) {
14610                                 mutex_enter(&ncec->ncec_lock);
14611                                 ncec->ncec_state = ND_REACHABLE;
14612                                 tid = ncec->ncec_timeout_id;
14613                                 ncec->ncec_timeout_id = 0;
14614                                 mutex_exit(&ncec->ncec_lock);
14615                                 (void) untimeout(tid);
14616                                 if (ip_debug > 2) {
14617                                         /* ip1dbg */
14618                                         pr_addr_dbg("ip_xmit: state"
14619                                             " for %s changed to"
14620                                             " REACHABLE\n", AF_INET6,
14621                                             &ncec->ncec_addr);
14622                                 }
14623                         }
14624                         return (0);
14625                 }
14626 
14627                 delta =  TICK_TO_MSEC(now) - ncec->ncec_last;
14628                 ip1dbg(("ip_xmit: delta = %" PRId64
14629                     " ill_reachable_time = %d \n", delta,
14630                     ill->ill_reachable_time));
14631                 if (delta > (uint64_t)ill->ill_reachable_time) {
14632                         mutex_enter(&ncec->ncec_lock);
14633                         switch (ncec->ncec_state) {
14634                         case ND_REACHABLE:
14635                                 ASSERT((ncec->ncec_flags & NCE_F_NONUD) == 0);
14636                                 /* FALLTHROUGH */
14637                         case ND_STALE:
14638                                 /*
14639                                  * ND_REACHABLE is identical to
14640                                  * ND_STALE in this specific case. If
14641                                  * reachable time has expired for this
14642                                  * neighbor (delta is greater than
14643                                  * reachable time), conceptually, the
14644                                  * neighbor cache is no longer in
14645                                  * REACHABLE state, but already in
14646                                  * STALE state.  So the correct
14647                                  * transition here is to ND_DELAY.
14648                                  */
14649                                 ncec->ncec_state = ND_DELAY;
14650                                 mutex_exit(&ncec->ncec_lock);
14651                                 nce_restart_timer(ncec,
14652                                     ipst->ips_delay_first_probe_time);
14653                                 if (ip_debug > 3) {
14654                                         /* ip2dbg */
14655                                         pr_addr_dbg("ip_xmit: state"
14656                                             " for %s changed to"
14657                                             " DELAY\n", AF_INET6,
14658                                             &ncec->ncec_addr);
14659                                 }
14660                                 break;
14661                         case ND_DELAY:
14662                         case ND_PROBE:
14663                                 mutex_exit(&ncec->ncec_lock);
14664                                 /* Timers have already started */
14665                                 break;
14666                         case ND_UNREACHABLE:
14667                                 /*
14668                                  * nce_timer has detected that this ncec
14669                                  * is unreachable and initiated deleting
14670                                  * this ncec.
14671                                  * This is a harmless race where we found the
14672                                  * ncec before it was deleted and have
14673                                  * just sent out a packet using this
14674                                  * unreachable ncec.
14675                                  */
14676                                 mutex_exit(&ncec->ncec_lock);
14677                                 break;
14678                         default:
14679                                 ASSERT(0);
14680                                 mutex_exit(&ncec->ncec_lock);
14681                         }
14682                 }
14683                 return (0);
14684 
14685         case ND_INCOMPLETE:
14686                 /*
14687                  * the state could have changed since we didn't hold the lock.
14688                  * Re-verify state under lock.
14689                  */
14690                 is_probe = ipmp_packet_is_probe(mp, nce->nce_ill);
14691                 mutex_enter(&ncec->ncec_lock);
14692                 if (NCE_ISREACHABLE(ncec)) {
14693                         mutex_exit(&ncec->ncec_lock);
14694                         goto sendit;
14695                 }
14696                 /* queue the packet */
14697                 nce_queue_mp(ncec, mp, is_probe);
14698                 mutex_exit(&ncec->ncec_lock);
14699                 DTRACE_PROBE2(ip__xmit__incomplete,
14700                     (ncec_t *), ncec, (mblk_t *), mp);
14701                 return (0);
14702 
14703         case ND_INITIAL:
14704                 /*
14705                  * State could have changed since we didn't hold the lock, so
14706                  * re-verify state.
14707                  */
14708                 is_probe = ipmp_packet_is_probe(mp, nce->nce_ill);
14709                 mutex_enter(&ncec->ncec_lock);
14710                 if (NCE_ISREACHABLE(ncec))  {
14711                         mutex_exit(&ncec->ncec_lock);
14712                         goto sendit;
14713                 }
14714                 nce_queue_mp(ncec, mp, is_probe);
14715                 if (ncec->ncec_state == ND_INITIAL) {
14716                         ncec->ncec_state = ND_INCOMPLETE;
14717                         mutex_exit(&ncec->ncec_lock);
14718                         /*
14719                          * figure out the source we want to use
14720                          * and resolve it.
14721                          */
14722                         ip_ndp_resolve(ncec);
14723                 } else  {
14724                         mutex_exit(&ncec->ncec_lock);
14725                 }
14726                 return (0);
14727 
14728         case ND_UNREACHABLE:
14729                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
14730                 ip_drop_output("ipIfStatsOutDiscards - ND_UNREACHABLE",
14731                     mp, ill);
14732                 freemsg(mp);
14733                 return (0);
14734 
14735         default:
14736                 ASSERT(0);
14737                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
14738                 ip_drop_output("ipIfStatsOutDiscards - ND_other",
14739                     mp, ill);
14740                 freemsg(mp);
14741                 return (ENETUNREACH);
14742         }
14743 }
14744 
14745 /*
14746  * Return B_TRUE if the buffers differ in length or content.
14747  * This is used for comparing extension header buffers.
14748  * Note that an extension header would be declared different
14749  * even if all that changed was the next header value in that header i.e.
14750  * what really changed is the next extension header.
14751  */
14752 boolean_t
14753 ip_cmpbuf(const void *abuf, uint_t alen, boolean_t b_valid, const void *bbuf,
14754     uint_t blen)
14755 {
14756         if (!b_valid)
14757                 blen = 0;
14758 
14759         if (alen != blen)
14760                 return (B_TRUE);
14761         if (alen == 0)
14762                 return (B_FALSE);       /* Both zero length */
14763         return (bcmp(abuf, bbuf, alen));
14764 }
14765 
14766 /*
14767  * Preallocate memory for ip_savebuf(). Returns B_TRUE if ok.
14768  * Return B_FALSE if memory allocation fails - don't change any state!
14769  */
14770 boolean_t
14771 ip_allocbuf(void **dstp, uint_t *dstlenp, boolean_t src_valid,
14772     const void *src, uint_t srclen)
14773 {
14774         void *dst;
14775 
14776         if (!src_valid)
14777                 srclen = 0;
14778 
14779         ASSERT(*dstlenp == 0);
14780         if (src != NULL && srclen != 0) {
14781                 dst = mi_alloc(srclen, BPRI_MED);
14782                 if (dst == NULL)
14783                         return (B_FALSE);
14784         } else {
14785                 dst = NULL;
14786         }
14787         if (*dstp != NULL)
14788                 mi_free(*dstp);
14789         *dstp = dst;
14790         *dstlenp = dst == NULL ? 0 : srclen;
14791         return (B_TRUE);
14792 }
14793 
14794 /*
14795  * Replace what is in *dst, *dstlen with the source.
14796  * Assumes ip_allocbuf has already been called.
14797  */
14798 void
14799 ip_savebuf(void **dstp, uint_t *dstlenp, boolean_t src_valid,
14800     const void *src, uint_t srclen)
14801 {
14802         if (!src_valid)
14803                 srclen = 0;
14804 
14805         ASSERT(*dstlenp == srclen);
14806         if (src != NULL && srclen != 0)
14807                 bcopy(src, *dstp, srclen);
14808 }
14809 
14810 /*
14811  * Free the storage pointed to by the members of an ip_pkt_t.
14812  */
14813 void
14814 ip_pkt_free(ip_pkt_t *ipp)
14815 {
14816         uint_t  fields = ipp->ipp_fields;
14817 
14818         if (fields & IPPF_HOPOPTS) {
14819                 kmem_free(ipp->ipp_hopopts, ipp->ipp_hopoptslen);
14820                 ipp->ipp_hopopts = NULL;
14821                 ipp->ipp_hopoptslen = 0;
14822         }
14823         if (fields & IPPF_RTHDRDSTOPTS) {
14824                 kmem_free(ipp->ipp_rthdrdstopts, ipp->ipp_rthdrdstoptslen);
14825                 ipp->ipp_rthdrdstopts = NULL;
14826                 ipp->ipp_rthdrdstoptslen = 0;
14827         }
14828         if (fields & IPPF_DSTOPTS) {
14829                 kmem_free(ipp->ipp_dstopts, ipp->ipp_dstoptslen);
14830                 ipp->ipp_dstopts = NULL;
14831                 ipp->ipp_dstoptslen = 0;
14832         }
14833         if (fields & IPPF_RTHDR) {
14834                 kmem_free(ipp->ipp_rthdr, ipp->ipp_rthdrlen);
14835                 ipp->ipp_rthdr = NULL;
14836                 ipp->ipp_rthdrlen = 0;
14837         }
14838         if (fields & IPPF_IPV4_OPTIONS) {
14839                 kmem_free(ipp->ipp_ipv4_options, ipp->ipp_ipv4_options_len);
14840                 ipp->ipp_ipv4_options = NULL;
14841                 ipp->ipp_ipv4_options_len = 0;
14842         }
14843         if (fields & IPPF_LABEL_V4) {
14844                 kmem_free(ipp->ipp_label_v4, ipp->ipp_label_len_v4);
14845                 ipp->ipp_label_v4 = NULL;
14846                 ipp->ipp_label_len_v4 = 0;
14847         }
14848         if (fields & IPPF_LABEL_V6) {
14849                 kmem_free(ipp->ipp_label_v6, ipp->ipp_label_len_v6);
14850                 ipp->ipp_label_v6 = NULL;
14851                 ipp->ipp_label_len_v6 = 0;
14852         }
14853         ipp->ipp_fields &= ~(IPPF_HOPOPTS | IPPF_RTHDRDSTOPTS | IPPF_DSTOPTS |
14854             IPPF_RTHDR | IPPF_IPV4_OPTIONS | IPPF_LABEL_V4 | IPPF_LABEL_V6);
14855 }
14856 
14857 /*
14858  * Copy from src to dst and allocate as needed.
14859  * Returns zero or ENOMEM.
14860  *
14861  * The caller must initialize dst to zero.
14862  */
14863 int
14864 ip_pkt_copy(ip_pkt_t *src, ip_pkt_t *dst, int kmflag)
14865 {
14866         uint_t  fields = src->ipp_fields;
14867 
14868         /* Start with fields that don't require memory allocation */
14869         dst->ipp_fields = fields &
14870             ~(IPPF_HOPOPTS | IPPF_RTHDRDSTOPTS | IPPF_DSTOPTS |
14871             IPPF_RTHDR | IPPF_IPV4_OPTIONS | IPPF_LABEL_V4 | IPPF_LABEL_V6);
14872 
14873         dst->ipp_addr = src->ipp_addr;
14874         dst->ipp_unicast_hops = src->ipp_unicast_hops;
14875         dst->ipp_hoplimit = src->ipp_hoplimit;
14876         dst->ipp_tclass = src->ipp_tclass;
14877         dst->ipp_type_of_service = src->ipp_type_of_service;
14878 
14879         if (!(fields & (IPPF_HOPOPTS | IPPF_RTHDRDSTOPTS | IPPF_DSTOPTS |
14880             IPPF_RTHDR | IPPF_IPV4_OPTIONS | IPPF_LABEL_V4 | IPPF_LABEL_V6)))
14881                 return (0);
14882 
14883         if (fields & IPPF_HOPOPTS) {
14884                 dst->ipp_hopopts = kmem_alloc(src->ipp_hopoptslen, kmflag);
14885                 if (dst->ipp_hopopts == NULL) {
14886                         ip_pkt_free(dst);
14887                         return (ENOMEM);
14888                 }
14889                 dst->ipp_fields |= IPPF_HOPOPTS;
14890                 bcopy(src->ipp_hopopts, dst->ipp_hopopts,
14891                     src->ipp_hopoptslen);
14892                 dst->ipp_hopoptslen = src->ipp_hopoptslen;
14893         }
14894         if (fields & IPPF_RTHDRDSTOPTS) {
14895                 dst->ipp_rthdrdstopts = kmem_alloc(src->ipp_rthdrdstoptslen,
14896                     kmflag);
14897                 if (dst->ipp_rthdrdstopts == NULL) {
14898                         ip_pkt_free(dst);
14899                         return (ENOMEM);
14900                 }
14901                 dst->ipp_fields |= IPPF_RTHDRDSTOPTS;
14902                 bcopy(src->ipp_rthdrdstopts, dst->ipp_rthdrdstopts,
14903                     src->ipp_rthdrdstoptslen);
14904                 dst->ipp_rthdrdstoptslen = src->ipp_rthdrdstoptslen;
14905         }
14906         if (fields & IPPF_DSTOPTS) {
14907                 dst->ipp_dstopts = kmem_alloc(src->ipp_dstoptslen, kmflag);
14908                 if (dst->ipp_dstopts == NULL) {
14909                         ip_pkt_free(dst);
14910                         return (ENOMEM);
14911                 }
14912                 dst->ipp_fields |= IPPF_DSTOPTS;
14913                 bcopy(src->ipp_dstopts, dst->ipp_dstopts,
14914                     src->ipp_dstoptslen);
14915                 dst->ipp_dstoptslen = src->ipp_dstoptslen;
14916         }
14917         if (fields & IPPF_RTHDR) {
14918                 dst->ipp_rthdr = kmem_alloc(src->ipp_rthdrlen, kmflag);
14919                 if (dst->ipp_rthdr == NULL) {
14920                         ip_pkt_free(dst);
14921                         return (ENOMEM);
14922                 }
14923                 dst->ipp_fields |= IPPF_RTHDR;
14924                 bcopy(src->ipp_rthdr, dst->ipp_rthdr,
14925                     src->ipp_rthdrlen);
14926                 dst->ipp_rthdrlen = src->ipp_rthdrlen;
14927         }
14928         if (fields & IPPF_IPV4_OPTIONS) {
14929                 dst->ipp_ipv4_options = kmem_alloc(src->ipp_ipv4_options_len,
14930                     kmflag);
14931                 if (dst->ipp_ipv4_options == NULL) {
14932                         ip_pkt_free(dst);
14933                         return (ENOMEM);
14934                 }
14935                 dst->ipp_fields |= IPPF_IPV4_OPTIONS;
14936                 bcopy(src->ipp_ipv4_options, dst->ipp_ipv4_options,
14937                     src->ipp_ipv4_options_len);
14938                 dst->ipp_ipv4_options_len = src->ipp_ipv4_options_len;
14939         }
14940         if (fields & IPPF_LABEL_V4) {
14941                 dst->ipp_label_v4 = kmem_alloc(src->ipp_label_len_v4, kmflag);
14942                 if (dst->ipp_label_v4 == NULL) {
14943                         ip_pkt_free(dst);
14944                         return (ENOMEM);
14945                 }
14946                 dst->ipp_fields |= IPPF_LABEL_V4;
14947                 bcopy(src->ipp_label_v4, dst->ipp_label_v4,
14948                     src->ipp_label_len_v4);
14949                 dst->ipp_label_len_v4 = src->ipp_label_len_v4;
14950         }
14951         if (fields & IPPF_LABEL_V6) {
14952                 dst->ipp_label_v6 = kmem_alloc(src->ipp_label_len_v6, kmflag);
14953                 if (dst->ipp_label_v6 == NULL) {
14954                         ip_pkt_free(dst);
14955                         return (ENOMEM);
14956                 }
14957                 dst->ipp_fields |= IPPF_LABEL_V6;
14958                 bcopy(src->ipp_label_v6, dst->ipp_label_v6,
14959                     src->ipp_label_len_v6);
14960                 dst->ipp_label_len_v6 = src->ipp_label_len_v6;
14961         }
14962         if (fields & IPPF_FRAGHDR) {
14963                 dst->ipp_fraghdr = kmem_alloc(src->ipp_fraghdrlen, kmflag);
14964                 if (dst->ipp_fraghdr == NULL) {
14965                         ip_pkt_free(dst);
14966                         return (ENOMEM);
14967                 }
14968                 dst->ipp_fields |= IPPF_FRAGHDR;
14969                 bcopy(src->ipp_fraghdr, dst->ipp_fraghdr,
14970                     src->ipp_fraghdrlen);
14971                 dst->ipp_fraghdrlen = src->ipp_fraghdrlen;
14972         }
14973         return (0);
14974 }
14975 
14976 /*
14977  * Returns INADDR_ANY if no source route
14978  */
14979 ipaddr_t
14980 ip_pkt_source_route_v4(const ip_pkt_t *ipp)
14981 {
14982         ipaddr_t        nexthop = INADDR_ANY;
14983         ipoptp_t        opts;
14984         uchar_t         *opt;
14985         uint8_t         optval;
14986         uint8_t         optlen;
14987         uint32_t        totallen;
14988 
14989         if (!(ipp->ipp_fields & IPPF_IPV4_OPTIONS))
14990                 return (INADDR_ANY);
14991 
14992         totallen = ipp->ipp_ipv4_options_len;
14993         if (totallen & 0x3)
14994                 return (INADDR_ANY);
14995 
14996         for (optval = ipoptp_first2(&opts, totallen, ipp->ipp_ipv4_options);
14997             optval != IPOPT_EOL;
14998             optval = ipoptp_next(&opts)) {
14999                 opt = opts.ipoptp_cur;
15000                 switch (optval) {
15001                         uint8_t off;
15002                 case IPOPT_SSRR:
15003                 case IPOPT_LSRR:
15004                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
15005                                 break;
15006                         }
15007                         optlen = opts.ipoptp_len;
15008                         off = opt[IPOPT_OFFSET];
15009                         off--;
15010                         if (optlen < IP_ADDR_LEN ||
15011                             off > optlen - IP_ADDR_LEN) {
15012                                 /* End of source route */
15013                                 break;
15014                         }
15015                         bcopy((char *)opt + off, &nexthop, IP_ADDR_LEN);
15016                         if (nexthop == htonl(INADDR_LOOPBACK)) {
15017                                 /* Ignore */
15018                                 nexthop = INADDR_ANY;
15019                                 break;
15020                         }
15021                         break;
15022                 }
15023         }
15024         return (nexthop);
15025 }
15026 
15027 /*
15028  * Reverse a source route.
15029  */
15030 void
15031 ip_pkt_source_route_reverse_v4(ip_pkt_t *ipp)
15032 {
15033         ipaddr_t        tmp;
15034         ipoptp_t        opts;
15035         uchar_t         *opt;
15036         uint8_t         optval;
15037         uint32_t        totallen;
15038 
15039         if (!(ipp->ipp_fields & IPPF_IPV4_OPTIONS))
15040                 return;
15041 
15042         totallen = ipp->ipp_ipv4_options_len;
15043         if (totallen & 0x3)
15044                 return;
15045 
15046         for (optval = ipoptp_first2(&opts, totallen, ipp->ipp_ipv4_options);
15047             optval != IPOPT_EOL;
15048             optval = ipoptp_next(&opts)) {
15049                 uint8_t off1, off2;
15050 
15051                 opt = opts.ipoptp_cur;
15052                 switch (optval) {
15053                 case IPOPT_SSRR:
15054                 case IPOPT_LSRR:
15055                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
15056                                 break;
15057                         }
15058                         off1 = IPOPT_MINOFF_SR - 1;
15059                         off2 = opt[IPOPT_OFFSET] - IP_ADDR_LEN - 1;
15060                         while (off2 > off1) {
15061                                 bcopy(opt + off2, &tmp, IP_ADDR_LEN);
15062                                 bcopy(opt + off1, opt + off2, IP_ADDR_LEN);
15063                                 bcopy(&tmp, opt + off2, IP_ADDR_LEN);
15064                                 off2 -= IP_ADDR_LEN;
15065                                 off1 += IP_ADDR_LEN;
15066                         }
15067                         opt[IPOPT_OFFSET] = IPOPT_MINOFF_SR;
15068                         break;
15069                 }
15070         }
15071 }
15072 
15073 /*
15074  * Returns NULL if no routing header
15075  */
15076 in6_addr_t *
15077 ip_pkt_source_route_v6(const ip_pkt_t *ipp)
15078 {
15079         in6_addr_t      *nexthop = NULL;
15080         ip6_rthdr0_t    *rthdr;
15081 
15082         if (!(ipp->ipp_fields & IPPF_RTHDR))
15083                 return (NULL);
15084 
15085         rthdr = (ip6_rthdr0_t *)ipp->ipp_rthdr;
15086         if (rthdr->ip6r0_segleft == 0)
15087                 return (NULL);
15088 
15089         nexthop = (in6_addr_t *)((char *)rthdr + sizeof (*rthdr));
15090         return (nexthop);
15091 }
15092 
15093 zoneid_t
15094 ip_get_zoneid_v4(ipaddr_t addr, mblk_t *mp, ip_recv_attr_t *ira,
15095     zoneid_t lookup_zoneid)
15096 {
15097         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
15098         ire_t           *ire;
15099         int             ire_flags = MATCH_IRE_TYPE;
15100         zoneid_t        zoneid = ALL_ZONES;
15101 
15102         if (is_system_labeled() && !tsol_can_accept_raw(mp, ira, B_FALSE))
15103                 return (ALL_ZONES);
15104 
15105         if (lookup_zoneid != ALL_ZONES)
15106                 ire_flags |= MATCH_IRE_ZONEONLY;
15107         ire = ire_ftable_lookup_v4(addr, NULL, NULL, IRE_LOCAL | IRE_LOOPBACK,
15108             NULL, lookup_zoneid, NULL, ire_flags, 0, ipst, NULL);
15109         if (ire != NULL) {
15110                 zoneid = IP_REAL_ZONEID(ire->ire_zoneid, ipst);
15111                 ire_refrele(ire);
15112         }
15113         return (zoneid);
15114 }
15115 
15116 zoneid_t
15117 ip_get_zoneid_v6(in6_addr_t *addr, mblk_t *mp, const ill_t *ill,
15118     ip_recv_attr_t *ira, zoneid_t lookup_zoneid)
15119 {
15120         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
15121         ire_t           *ire;
15122         int             ire_flags = MATCH_IRE_TYPE;
15123         zoneid_t        zoneid = ALL_ZONES;
15124 
15125         if (is_system_labeled() && !tsol_can_accept_raw(mp, ira, B_FALSE))
15126                 return (ALL_ZONES);
15127 
15128         if (IN6_IS_ADDR_LINKLOCAL(addr))
15129                 ire_flags |= MATCH_IRE_ILL;
15130 
15131         if (lookup_zoneid != ALL_ZONES)
15132                 ire_flags |= MATCH_IRE_ZONEONLY;
15133         ire = ire_ftable_lookup_v6(addr, NULL, NULL, IRE_LOCAL | IRE_LOOPBACK,
15134             ill, lookup_zoneid, NULL, ire_flags, 0, ipst, NULL);
15135         if (ire != NULL) {
15136                 zoneid = IP_REAL_ZONEID(ire->ire_zoneid, ipst);
15137                 ire_refrele(ire);
15138         }
15139         return (zoneid);
15140 }
15141 
15142 /*
15143  * IP obserability hook support functions.
15144  */
15145 static void
15146 ipobs_init(ip_stack_t *ipst)
15147 {
15148         netid_t id;
15149 
15150         id = net_getnetidbynetstackid(ipst->ips_netstack->netstack_stackid);
15151 
15152         ipst->ips_ip4_observe_pr = net_protocol_lookup(id, NHF_INET);
15153         VERIFY(ipst->ips_ip4_observe_pr != NULL);
15154 
15155         ipst->ips_ip6_observe_pr = net_protocol_lookup(id, NHF_INET6);
15156         VERIFY(ipst->ips_ip6_observe_pr != NULL);
15157 }
15158 
15159 static void
15160 ipobs_fini(ip_stack_t *ipst)
15161 {
15162 
15163         VERIFY(net_protocol_release(ipst->ips_ip4_observe_pr) == 0);
15164         VERIFY(net_protocol_release(ipst->ips_ip6_observe_pr) == 0);
15165 }
15166 
15167 /*
15168  * hook_pkt_observe_t is composed in network byte order so that the
15169  * entire mblk_t chain handed into hook_run can be used as-is.
15170  * The caveat is that use of the fields, such as the zone fields,
15171  * requires conversion into host byte order first.
15172  */
15173 void
15174 ipobs_hook(mblk_t *mp, int htype, zoneid_t zsrc, zoneid_t zdst,
15175     const ill_t *ill, ip_stack_t *ipst)
15176 {
15177         hook_pkt_observe_t *hdr;
15178         uint64_t grifindex;
15179         mblk_t *imp;
15180 
15181         imp = allocb(sizeof (*hdr), BPRI_HI);
15182         if (imp == NULL)
15183                 return;
15184 
15185         hdr = (hook_pkt_observe_t *)imp->b_rptr;
15186         /*
15187          * b_wptr is set to make the apparent size of the data in the mblk_t
15188          * to exclude the pointers at the end of hook_pkt_observer_t.
15189          */
15190         imp->b_wptr = imp->b_rptr + sizeof (dl_ipnetinfo_t);
15191         imp->b_cont = mp;
15192 
15193         ASSERT(DB_TYPE(mp) == M_DATA);
15194 
15195         if (IS_UNDER_IPMP(ill))
15196                 grifindex = ipmp_ill_get_ipmp_ifindex(ill);
15197         else
15198                 grifindex = 0;
15199 
15200         hdr->hpo_version = 1;
15201         hdr->hpo_htype = htons(htype);
15202         hdr->hpo_pktlen = htonl((ulong_t)msgdsize(mp));
15203         hdr->hpo_ifindex = htonl(ill->ill_phyint->phyint_ifindex);
15204         hdr->hpo_grifindex = htonl(grifindex);
15205         hdr->hpo_zsrc = htonl(zsrc);
15206         hdr->hpo_zdst = htonl(zdst);
15207         hdr->hpo_pkt = imp;
15208         hdr->hpo_ctx = ipst->ips_netstack;
15209 
15210         if (ill->ill_isv6) {
15211                 hdr->hpo_family = AF_INET6;
15212                 (void) hook_run(ipst->ips_ipv6_net_data->netd_hooks,
15213                     ipst->ips_ipv6observing, (hook_data_t)hdr);
15214         } else {
15215                 hdr->hpo_family = AF_INET;
15216                 (void) hook_run(ipst->ips_ipv4_net_data->netd_hooks,
15217                     ipst->ips_ipv4observing, (hook_data_t)hdr);
15218         }
15219 
15220         imp->b_cont = NULL;
15221         freemsg(imp);
15222 }
15223 
15224 /*
15225  * Utility routine that checks if `v4srcp' is a valid address on underlying
15226  * interface `ill'.  If `ipifp' is non-NULL, it's set to a held ipif
15227  * associated with `v4srcp' on success.  NOTE: if this is not called from
15228  * inside the IPSQ (ill_g_lock is not held), `ill' may be removed from the
15229  * group during or after this lookup.
15230  */
15231 boolean_t
15232 ipif_lookup_testaddr_v4(ill_t *ill, const in_addr_t *v4srcp, ipif_t **ipifp)
15233 {
15234         ipif_t *ipif;
15235 
15236         ipif = ipif_lookup_addr_exact(*v4srcp, ill, ill->ill_ipst);
15237         if (ipif != NULL) {
15238                 if (ipifp != NULL)
15239                         *ipifp = ipif;
15240                 else
15241                         ipif_refrele(ipif);
15242                 return (B_TRUE);
15243         }
15244 
15245         ip1dbg(("ipif_lookup_testaddr_v4: cannot find ipif for src %x\n",
15246             *v4srcp));
15247         return (B_FALSE);
15248 }
15249 
15250 /*
15251  * Transport protocol call back function for CPU state change.
15252  */
15253 /* ARGSUSED */
15254 static int
15255 ip_tp_cpu_update(cpu_setup_t what, int id, void *arg)
15256 {
15257         processorid_t cpu_seqid;
15258         netstack_handle_t nh;
15259         netstack_t *ns;
15260 
15261         ASSERT(MUTEX_HELD(&cpu_lock));
15262 
15263         switch (what) {
15264         case CPU_CONFIG:
15265         case CPU_ON:
15266         case CPU_INIT:
15267         case CPU_CPUPART_IN:
15268                 cpu_seqid = cpu[id]->cpu_seqid;
15269                 netstack_next_init(&nh);
15270                 while ((ns = netstack_next(&nh)) != NULL) {
15271                         tcp_stack_cpu_add(ns->netstack_tcp, cpu_seqid);
15272                         sctp_stack_cpu_add(ns->netstack_sctp, cpu_seqid);
15273                         udp_stack_cpu_add(ns->netstack_udp, cpu_seqid);
15274                         netstack_rele(ns);
15275                 }
15276                 netstack_next_fini(&nh);
15277                 break;
15278         case CPU_UNCONFIG:
15279         case CPU_OFF:
15280         case CPU_CPUPART_OUT:
15281                 /*
15282                  * Nothing to do.  We don't remove the per CPU stats from
15283                  * the IP stack even when the CPU goes offline.
15284                  */
15285                 break;
15286         default:
15287                 break;
15288         }
15289         return (0);
15290 }