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) 2011 Joyent, Inc. All rights reserved.
  26  */
  27 
  28 #include <sys/types.h>
  29 #include <sys/stream.h>
  30 #include <sys/dlpi.h>
  31 #include <sys/stropts.h>
  32 #include <sys/sysmacros.h>
  33 #include <sys/strsubr.h>
  34 #include <sys/strlog.h>
  35 #include <sys/strsun.h>
  36 #include <sys/zone.h>
  37 #define _SUN_TPI_VERSION 2
  38 #include <sys/tihdr.h>
  39 #include <sys/xti_inet.h>
  40 #include <sys/ddi.h>
  41 #include <sys/suntpi.h>
  42 #include <sys/cmn_err.h>
  43 #include <sys/debug.h>
  44 #include <sys/kobj.h>
  45 #include <sys/modctl.h>
  46 #include <sys/atomic.h>
  47 #include <sys/policy.h>
  48 #include <sys/priv.h>
  49 #include <sys/taskq.h>
  50 
  51 #include <sys/systm.h>
  52 #include <sys/param.h>
  53 #include <sys/kmem.h>
  54 #include <sys/sdt.h>
  55 #include <sys/socket.h>
  56 #include <sys/vtrace.h>
  57 #include <sys/isa_defs.h>
  58 #include <sys/mac.h>
  59 #include <net/if.h>
  60 #include <net/if_arp.h>
  61 #include <net/route.h>
  62 #include <sys/sockio.h>
  63 #include <netinet/in.h>
  64 #include <net/if_dl.h>
  65 
  66 #include <inet/common.h>
  67 #include <inet/mi.h>
  68 #include <inet/mib2.h>
  69 #include <inet/nd.h>
  70 #include <inet/arp.h>
  71 #include <inet/snmpcom.h>
  72 #include <inet/optcom.h>
  73 #include <inet/kstatcom.h>
  74 
  75 #include <netinet/igmp_var.h>
  76 #include <netinet/ip6.h>
  77 #include <netinet/icmp6.h>
  78 #include <netinet/sctp.h>
  79 
  80 #include <inet/ip.h>
  81 #include <inet/ip_impl.h>
  82 #include <inet/ip6.h>
  83 #include <inet/ip6_asp.h>
  84 #include <inet/tcp.h>
  85 #include <inet/tcp_impl.h>
  86 #include <inet/ip_multi.h>
  87 #include <inet/ip_if.h>
  88 #include <inet/ip_ire.h>
  89 #include <inet/ip_ftable.h>
  90 #include <inet/ip_rts.h>
  91 #include <inet/ip_ndp.h>
  92 #include <inet/ip_listutils.h>
  93 #include <netinet/igmp.h>
  94 #include <netinet/ip_mroute.h>
  95 #include <inet/ipp_common.h>
  96 
  97 #include <net/pfkeyv2.h>
  98 #include <inet/sadb.h>
  99 #include <inet/ipsec_impl.h>
 100 #include <inet/iptun/iptun_impl.h>
 101 #include <inet/ipdrop.h>
 102 #include <inet/ip_netinfo.h>
 103 #include <inet/ilb_ip.h>
 104 
 105 #include <sys/ethernet.h>
 106 #include <net/if_types.h>
 107 #include <sys/cpuvar.h>
 108 
 109 #include <ipp/ipp.h>
 110 #include <ipp/ipp_impl.h>
 111 #include <ipp/ipgpc/ipgpc.h>
 112 
 113 #include <sys/pattr.h>
 114 #include <inet/dccp/dccp_ip.h>
 115 #include <inet/dccp/dccp_impl.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 int             ip_close(queue_t *, int);
 670 static char     *ip_dot_saddr(uchar_t *, char *);
 671 static void     ip_lrput(queue_t *, mblk_t *);
 672 ipaddr_t        ip_net_mask(ipaddr_t);
 673 char            *ip_nv_lookup(nv_t *, int);
 674 void    ip_rput(queue_t *, mblk_t *);
 675 static void     ip_rput_dlpi_writer(ipsq_t *dummy_sq, queue_t *q, mblk_t *mp,
 676                     void *dummy_arg);
 677 int             ip_snmp_get(queue_t *, mblk_t *, int, boolean_t);
 678 static mblk_t   *ip_snmp_get_mib2_ip(queue_t *, mblk_t *,
 679                     mib2_ipIfStatsEntry_t *, ip_stack_t *, boolean_t);
 680 static mblk_t   *ip_snmp_get_mib2_ip_traffic_stats(queue_t *, mblk_t *,
 681                     ip_stack_t *, boolean_t);
 682 static mblk_t   *ip_snmp_get_mib2_ip6(queue_t *, mblk_t *, ip_stack_t *,
 683                     boolean_t);
 684 static mblk_t   *ip_snmp_get_mib2_icmp(queue_t *, mblk_t *, ip_stack_t *ipst);
 685 static mblk_t   *ip_snmp_get_mib2_icmp6(queue_t *, mblk_t *, ip_stack_t *ipst);
 686 static mblk_t   *ip_snmp_get_mib2_igmp(queue_t *, mblk_t *, ip_stack_t *ipst);
 687 static mblk_t   *ip_snmp_get_mib2_multi(queue_t *, mblk_t *, ip_stack_t *ipst);
 688 static mblk_t   *ip_snmp_get_mib2_ip_addr(queue_t *, mblk_t *,
 689                     ip_stack_t *ipst, boolean_t);
 690 static mblk_t   *ip_snmp_get_mib2_ip6_addr(queue_t *, mblk_t *,
 691                     ip_stack_t *ipst, boolean_t);
 692 static mblk_t   *ip_snmp_get_mib2_ip_group_src(queue_t *, mblk_t *,
 693                     ip_stack_t *ipst);
 694 static mblk_t   *ip_snmp_get_mib2_ip6_group_src(queue_t *, mblk_t *,
 695                     ip_stack_t *ipst);
 696 static mblk_t   *ip_snmp_get_mib2_ip_group_mem(queue_t *, mblk_t *,
 697                     ip_stack_t *ipst);
 698 static mblk_t   *ip_snmp_get_mib2_ip6_group_mem(queue_t *, mblk_t *,
 699                     ip_stack_t *ipst);
 700 static mblk_t   *ip_snmp_get_mib2_virt_multi(queue_t *, mblk_t *,
 701                     ip_stack_t *ipst);
 702 static mblk_t   *ip_snmp_get_mib2_multi_rtable(queue_t *, mblk_t *,
 703                     ip_stack_t *ipst);
 704 static mblk_t   *ip_snmp_get_mib2_ip_route_media(queue_t *, mblk_t *, int,
 705                     ip_stack_t *ipst);
 706 static mblk_t   *ip_snmp_get_mib2_ip6_route_media(queue_t *, mblk_t *, int,
 707                     ip_stack_t *ipst);
 708 static void     ip_snmp_get2_v4(ire_t *, iproutedata_t *);
 709 static void     ip_snmp_get2_v6_route(ire_t *, iproutedata_t *);
 710 static int      ip_snmp_get2_v4_media(ncec_t *, iproutedata_t *);
 711 static int      ip_snmp_get2_v6_media(ncec_t *, iproutedata_t *);
 712 int             ip_snmp_set(queue_t *, int, int, uchar_t *, int);
 713 
 714 static mblk_t   *ip_fragment_copyhdr(uchar_t *, int, int, ip_stack_t *,
 715                     mblk_t *);
 716 
 717 static void     conn_drain_init(ip_stack_t *);
 718 static void     conn_drain_fini(ip_stack_t *);
 719 static void     conn_drain(conn_t *connp, boolean_t closing);
 720 
 721 static void     conn_walk_drain(ip_stack_t *, idl_tx_list_t *);
 722 static void     conn_walk_sctp(pfv_t, void *, zoneid_t, netstack_t *);
 723 
 724 static void     *ip_stack_init(netstackid_t stackid, netstack_t *ns);
 725 static void     ip_stack_shutdown(netstackid_t stackid, void *arg);
 726 static void     ip_stack_fini(netstackid_t stackid, void *arg);
 727 
 728 static int      ip_multirt_apply_membership(int (*fn)(conn_t *, boolean_t,
 729     const in6_addr_t *, ipaddr_t, uint_t, mcast_record_t, const in6_addr_t *),
 730     ire_t *, conn_t *, boolean_t, const in6_addr_t *,  mcast_record_t,
 731     const in6_addr_t *);
 732 
 733 static int      ip_squeue_switch(int);
 734 
 735 static void     *ip_kstat_init(netstackid_t, ip_stack_t *);
 736 static void     ip_kstat_fini(netstackid_t, kstat_t *);
 737 static int      ip_kstat_update(kstat_t *kp, int rw);
 738 static void     *icmp_kstat_init(netstackid_t);
 739 static void     icmp_kstat_fini(netstackid_t, kstat_t *);
 740 static int      icmp_kstat_update(kstat_t *kp, int rw);
 741 static void     *ip_kstat2_init(netstackid_t, ip_stat_t *);
 742 static void     ip_kstat2_fini(netstackid_t, kstat_t *);
 743 
 744 static void     ipobs_init(ip_stack_t *);
 745 static void     ipobs_fini(ip_stack_t *);
 746 
 747 static int      ip_tp_cpu_update(cpu_setup_t, int, void *);
 748 
 749 ipaddr_t        ip_g_all_ones = IP_HOST_MASK;
 750 
 751 static long ip_rput_pullups;
 752 int     dohwcksum = 1;  /* use h/w cksum if supported by the hardware */
 753 
 754 vmem_t *ip_minor_arena_sa; /* for minor nos. from INET_MIN_DEV+2 thru 2^^18-1 */
 755 vmem_t *ip_minor_arena_la; /* for minor nos. from 2^^18 thru 2^^32-1 */
 756 
 757 int     ip_debug;
 758 
 759 /*
 760  * Multirouting/CGTP stuff
 761  */
 762 int     ip_cgtp_filter_rev = CGTP_FILTER_REV;   /* CGTP hooks version */
 763 
 764 /*
 765  * IP tunables related declarations. Definitions are in ip_tunables.c
 766  */
 767 extern mod_prop_info_t ip_propinfo_tbl[];
 768 extern int ip_propinfo_count;
 769 
 770 /*
 771  * Table of IP ioctls encoding the various properties of the ioctl and
 772  * indexed based on the last byte of the ioctl command. Occasionally there
 773  * is a clash, and there is more than 1 ioctl with the same last byte.
 774  * In such a case 1 ioctl is encoded in the ndx table and the remaining
 775  * ioctls are encoded in the misc table. An entry in the ndx table is
 776  * retrieved by indexing on the last byte of the ioctl command and comparing
 777  * the ioctl command with the value in the ndx table. In the event of a
 778  * mismatch the misc table is then searched sequentially for the desired
 779  * ioctl command.
 780  *
 781  * Entry: <command> <copyin_size> <flags> <cmd_type> <function> <restart_func>
 782  */
 783 ip_ioctl_cmd_t ip_ndx_ioctl_table[] = {
 784         /* 000 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 785         /* 001 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 786         /* 002 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 787         /* 003 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 788         /* 004 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 789         /* 005 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 790         /* 006 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 791         /* 007 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 792         /* 008 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 793         /* 009 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 794 
 795         /* 010 */ { SIOCADDRT,  sizeof (struct rtentry), IPI_PRIV,
 796                         MISC_CMD, ip_siocaddrt, NULL },
 797         /* 011 */ { SIOCDELRT,  sizeof (struct rtentry), IPI_PRIV,
 798                         MISC_CMD, ip_siocdelrt, NULL },
 799 
 800         /* 012 */ { SIOCSIFADDR, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
 801                         IF_CMD, ip_sioctl_addr, ip_sioctl_addr_restart },
 802         /* 013 */ { SIOCGIFADDR, sizeof (struct ifreq), IPI_GET_CMD,
 803                         IF_CMD, ip_sioctl_get_addr, NULL },
 804 
 805         /* 014 */ { SIOCSIFDSTADDR, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
 806                         IF_CMD, ip_sioctl_dstaddr, ip_sioctl_dstaddr_restart },
 807         /* 015 */ { SIOCGIFDSTADDR, sizeof (struct ifreq),
 808                         IPI_GET_CMD, IF_CMD, ip_sioctl_get_dstaddr, NULL },
 809 
 810         /* 016 */ { SIOCSIFFLAGS, sizeof (struct ifreq),
 811                         IPI_PRIV | IPI_WR,
 812                         IF_CMD, ip_sioctl_flags, ip_sioctl_flags_restart },
 813         /* 017 */ { SIOCGIFFLAGS, sizeof (struct ifreq),
 814                         IPI_MODOK | IPI_GET_CMD,
 815                         IF_CMD, ip_sioctl_get_flags, NULL },
 816 
 817         /* 018 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 818         /* 019 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 819 
 820         /* copyin size cannot be coded for SIOCGIFCONF */
 821         /* 020 */ { O_SIOCGIFCONF, 0, IPI_GET_CMD,
 822                         MISC_CMD, ip_sioctl_get_ifconf, NULL },
 823 
 824         /* 021 */ { SIOCSIFMTU, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
 825                         IF_CMD, ip_sioctl_mtu, NULL },
 826         /* 022 */ { SIOCGIFMTU, sizeof (struct ifreq), IPI_GET_CMD,
 827                         IF_CMD, ip_sioctl_get_mtu, NULL },
 828         /* 023 */ { SIOCGIFBRDADDR, sizeof (struct ifreq),
 829                         IPI_GET_CMD, IF_CMD, ip_sioctl_get_brdaddr, NULL },
 830         /* 024 */ { SIOCSIFBRDADDR, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
 831                         IF_CMD, ip_sioctl_brdaddr, NULL },
 832         /* 025 */ { SIOCGIFNETMASK, sizeof (struct ifreq),
 833                         IPI_GET_CMD, IF_CMD, ip_sioctl_get_netmask, NULL },
 834         /* 026 */ { SIOCSIFNETMASK, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
 835                         IF_CMD, ip_sioctl_netmask, ip_sioctl_netmask_restart },
 836         /* 027 */ { SIOCGIFMETRIC, sizeof (struct ifreq),
 837                         IPI_GET_CMD, IF_CMD, ip_sioctl_get_metric, NULL },
 838         /* 028 */ { SIOCSIFMETRIC, sizeof (struct ifreq), IPI_PRIV,
 839                         IF_CMD, ip_sioctl_metric, NULL },
 840         /* 029 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 841 
 842         /* See 166-168 below for extended SIOC*XARP ioctls */
 843         /* 030 */ { SIOCSARP, sizeof (struct arpreq), IPI_PRIV | IPI_WR,
 844                         ARP_CMD, ip_sioctl_arp, NULL },
 845         /* 031 */ { SIOCGARP, sizeof (struct arpreq), IPI_GET_CMD,
 846                         ARP_CMD, ip_sioctl_arp, NULL },
 847         /* 032 */ { SIOCDARP, sizeof (struct arpreq), IPI_PRIV | IPI_WR,
 848                         ARP_CMD, ip_sioctl_arp, NULL },
 849 
 850         /* 033 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 851         /* 034 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 852         /* 035 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 853         /* 036 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 854         /* 037 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 855         /* 038 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 856         /* 039 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 857         /* 040 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 858         /* 041 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 859         /* 042 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 860         /* 043 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 861         /* 044 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 862         /* 045 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 863         /* 046 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 864         /* 047 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 865         /* 048 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 866         /* 049 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 867         /* 050 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 868         /* 051 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 869         /* 052 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 870         /* 053 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 871 
 872         /* 054 */ { IF_UNITSEL, sizeof (int), IPI_PRIV | IPI_WR | IPI_MODOK,
 873                         MISC_CMD, if_unitsel, if_unitsel_restart },
 874 
 875         /* 055 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 876         /* 056 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 877         /* 057 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 878         /* 058 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 879         /* 059 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 880         /* 060 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 881         /* 061 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 882         /* 062 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 883         /* 063 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 884         /* 064 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 885         /* 065 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 886         /* 066 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 887         /* 067 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 888         /* 068 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 889         /* 069 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 890         /* 070 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 891         /* 071 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 892         /* 072 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 893 
 894         /* 073 */ { SIOCSIFNAME, sizeof (struct ifreq),
 895                         IPI_PRIV | IPI_WR | IPI_MODOK,
 896                         IF_CMD, ip_sioctl_sifname, NULL },
 897 
 898         /* 074 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 899         /* 075 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 900         /* 076 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 901         /* 077 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 902         /* 078 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 903         /* 079 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 904         /* 080 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 905         /* 081 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 906         /* 082 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 907         /* 083 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 908         /* 084 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 909         /* 085 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 910         /* 086 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 911 
 912         /* 087 */ { SIOCGIFNUM, sizeof (int), IPI_GET_CMD,
 913                         MISC_CMD, ip_sioctl_get_ifnum, NULL },
 914         /* 088 */ { SIOCGIFMUXID, sizeof (struct ifreq), IPI_GET_CMD,
 915                         IF_CMD, ip_sioctl_get_muxid, NULL },
 916         /* 089 */ { SIOCSIFMUXID, sizeof (struct ifreq),
 917                         IPI_PRIV | IPI_WR, IF_CMD, ip_sioctl_muxid, NULL },
 918 
 919         /* Both if and lif variants share same func */
 920         /* 090 */ { SIOCGIFINDEX, sizeof (struct ifreq), IPI_GET_CMD,
 921                         IF_CMD, ip_sioctl_get_lifindex, NULL },
 922         /* Both if and lif variants share same func */
 923         /* 091 */ { SIOCSIFINDEX, sizeof (struct ifreq),
 924                         IPI_PRIV | IPI_WR, IF_CMD, ip_sioctl_slifindex, NULL },
 925 
 926         /* copyin size cannot be coded for SIOCGIFCONF */
 927         /* 092 */ { SIOCGIFCONF, 0, IPI_GET_CMD,
 928                         MISC_CMD, ip_sioctl_get_ifconf, NULL },
 929         /* 093 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 930         /* 094 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 931         /* 095 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 932         /* 096 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 933         /* 097 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 934         /* 098 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 935         /* 099 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 936         /* 100 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 937         /* 101 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 938         /* 102 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 939         /* 103 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 940         /* 104 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 941         /* 105 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 942         /* 106 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 943         /* 107 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 944         /* 108 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 945         /* 109 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 946 
 947         /* 110 */ { SIOCLIFREMOVEIF, sizeof (struct lifreq),
 948                         IPI_PRIV | IPI_WR, LIF_CMD, ip_sioctl_removeif,
 949                         ip_sioctl_removeif_restart },
 950         /* 111 */ { SIOCLIFADDIF, sizeof (struct lifreq),
 951                         IPI_GET_CMD | IPI_PRIV | IPI_WR,
 952                         LIF_CMD, ip_sioctl_addif, NULL },
 953 #define SIOCLIFADDR_NDX 112
 954         /* 112 */ { SIOCSLIFADDR, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
 955                         LIF_CMD, ip_sioctl_addr, ip_sioctl_addr_restart },
 956         /* 113 */ { SIOCGLIFADDR, sizeof (struct lifreq),
 957                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_addr, NULL },
 958         /* 114 */ { SIOCSLIFDSTADDR, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
 959                         LIF_CMD, ip_sioctl_dstaddr, ip_sioctl_dstaddr_restart },
 960         /* 115 */ { SIOCGLIFDSTADDR, sizeof (struct lifreq),
 961                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_dstaddr, NULL },
 962         /* 116 */ { SIOCSLIFFLAGS, sizeof (struct lifreq),
 963                         IPI_PRIV | IPI_WR,
 964                         LIF_CMD, ip_sioctl_flags, ip_sioctl_flags_restart },
 965         /* 117 */ { SIOCGLIFFLAGS, sizeof (struct lifreq),
 966                         IPI_GET_CMD | IPI_MODOK,
 967                         LIF_CMD, ip_sioctl_get_flags, NULL },
 968 
 969         /* 118 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 970         /* 119 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 971 
 972         /* 120 */ { O_SIOCGLIFCONF, 0, IPI_GET_CMD, MISC_CMD,
 973                         ip_sioctl_get_lifconf, NULL },
 974         /* 121 */ { SIOCSLIFMTU, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
 975                         LIF_CMD, ip_sioctl_mtu, NULL },
 976         /* 122 */ { SIOCGLIFMTU, sizeof (struct lifreq), IPI_GET_CMD,
 977                         LIF_CMD, ip_sioctl_get_mtu, NULL },
 978         /* 123 */ { SIOCGLIFBRDADDR, sizeof (struct lifreq),
 979                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_brdaddr, NULL },
 980         /* 124 */ { SIOCSLIFBRDADDR, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
 981                         LIF_CMD, ip_sioctl_brdaddr, NULL },
 982         /* 125 */ { SIOCGLIFNETMASK, sizeof (struct lifreq),
 983                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_netmask, NULL },
 984         /* 126 */ { SIOCSLIFNETMASK, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
 985                         LIF_CMD, ip_sioctl_netmask, ip_sioctl_netmask_restart },
 986         /* 127 */ { SIOCGLIFMETRIC, sizeof (struct lifreq),
 987                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_metric, NULL },
 988         /* 128 */ { SIOCSLIFMETRIC, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
 989                         LIF_CMD, ip_sioctl_metric, NULL },
 990         /* 129 */ { SIOCSLIFNAME, sizeof (struct lifreq),
 991                         IPI_PRIV | IPI_WR | IPI_MODOK,
 992                         LIF_CMD, ip_sioctl_slifname,
 993                         ip_sioctl_slifname_restart },
 994 
 995         /* 130 */ { SIOCGLIFNUM, sizeof (struct lifnum), IPI_GET_CMD,
 996                         MISC_CMD, ip_sioctl_get_lifnum, NULL },
 997         /* 131 */ { SIOCGLIFMUXID, sizeof (struct lifreq),
 998                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_muxid, NULL },
 999         /* 132 */ { SIOCSLIFMUXID, sizeof (struct lifreq),
1000                         IPI_PRIV | IPI_WR, LIF_CMD, ip_sioctl_muxid, NULL },
1001         /* 133 */ { SIOCGLIFINDEX, sizeof (struct lifreq),
1002                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_lifindex, 0 },
1003         /* 134 */ { SIOCSLIFINDEX, sizeof (struct lifreq),
1004                         IPI_PRIV | IPI_WR, LIF_CMD, ip_sioctl_slifindex, 0 },
1005         /* 135 */ { SIOCSLIFTOKEN, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1006                         LIF_CMD, ip_sioctl_token, NULL },
1007         /* 136 */ { SIOCGLIFTOKEN, sizeof (struct lifreq),
1008                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_token, NULL },
1009         /* 137 */ { SIOCSLIFSUBNET, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1010                         LIF_CMD, ip_sioctl_subnet, ip_sioctl_subnet_restart },
1011         /* 138 */ { SIOCGLIFSUBNET, sizeof (struct lifreq),
1012                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_subnet, NULL },
1013         /* 139 */ { SIOCSLIFLNKINFO, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1014                         LIF_CMD, ip_sioctl_lnkinfo, NULL },
1015 
1016         /* 140 */ { SIOCGLIFLNKINFO, sizeof (struct lifreq),
1017                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_lnkinfo, NULL },
1018         /* 141 */ { SIOCLIFDELND, sizeof (struct lifreq), IPI_PRIV,
1019                         LIF_CMD, ip_siocdelndp_v6, NULL },
1020         /* 142 */ { SIOCLIFGETND, sizeof (struct lifreq), IPI_GET_CMD,
1021                         LIF_CMD, ip_siocqueryndp_v6, NULL },
1022         /* 143 */ { SIOCLIFSETND, sizeof (struct lifreq), IPI_PRIV,
1023                         LIF_CMD, ip_siocsetndp_v6, NULL },
1024         /* 144 */ { SIOCTMYADDR, sizeof (struct sioc_addrreq), IPI_GET_CMD,
1025                         MISC_CMD, ip_sioctl_tmyaddr, NULL },
1026         /* 145 */ { SIOCTONLINK, sizeof (struct sioc_addrreq), IPI_GET_CMD,
1027                         MISC_CMD, ip_sioctl_tonlink, NULL },
1028         /* 146 */ { SIOCTMYSITE, sizeof (struct sioc_addrreq), 0,
1029                         MISC_CMD, ip_sioctl_tmysite, NULL },
1030         /* 147 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1031         /* 148 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1032         /* IPSECioctls handled in ip_sioctl_copyin_setup itself */
1033         /* 149 */ { SIOCFIPSECONFIG, 0, IPI_PRIV, MISC_CMD, NULL, NULL },
1034         /* 150 */ { SIOCSIPSECONFIG, 0, IPI_PRIV, MISC_CMD, NULL, NULL },
1035         /* 151 */ { SIOCDIPSECONFIG, 0, IPI_PRIV, MISC_CMD, NULL, NULL },
1036         /* 152 */ { SIOCLIPSECONFIG, 0, IPI_PRIV, MISC_CMD, 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         (pfi_t)ip_rput, NULL, ip_openv4, ip_close, NULL,
1188         &ip_mod_info
1189 };
1190 
1191 struct qinit iprinitv6 = {
1192         (pfi_t)ip_rput_v6, NULL, ip_openv6, ip_close, NULL,
1193         &ip_mod_info
1194 };
1195 
1196 static struct qinit ipwinit = {
1197         (pfi_t)ip_wput_nondata, (pfi_t)ip_wsrv, NULL, NULL, NULL,
1198         &ip_mod_info
1199 };
1200 
1201 static struct qinit iplrinit = {
1202         (pfi_t)ip_lrput, NULL, ip_openv4, ip_close, NULL,
1203         &ip_mod_info
1204 };
1205 
1206 static struct qinit iplwinit = {
1207         (pfi_t)ip_lwput, NULL, NULL, NULL, NULL,
1208         &ip_mod_info
1209 };
1210 
1211 /* For AF_INET aka /dev/ip */
1212 struct streamtab ipinfov4 = {
1213         &iprinitv4, &ipwinit, &iplrinit, &iplwinit
1214 };
1215 
1216 /* For AF_INET6 aka /dev/ip6 */
1217 struct streamtab ipinfov6 = {
1218         &iprinitv6, &ipwinit, &iplrinit, &iplwinit
1219 };
1220 
1221 #ifdef  DEBUG
1222 boolean_t skip_sctp_cksum = B_FALSE;
1223 #endif
1224 
1225 /*
1226  * Generate an ICMP fragmentation needed message.
1227  * When called from ip_output side a minimal ip_recv_attr_t needs to be
1228  * constructed by the caller.
1229  */
1230 void
1231 icmp_frag_needed(mblk_t *mp, int mtu, ip_recv_attr_t *ira)
1232 {
1233         icmph_t icmph;
1234         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
1235 
1236         mp = icmp_pkt_err_ok(mp, ira);
1237         if (mp == NULL)
1238                 return;
1239 
1240         bzero(&icmph, sizeof (icmph_t));
1241         icmph.icmph_type = ICMP_DEST_UNREACHABLE;
1242         icmph.icmph_code = ICMP_FRAGMENTATION_NEEDED;
1243         icmph.icmph_du_mtu = htons((uint16_t)mtu);
1244         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutFragNeeded);
1245         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDestUnreachs);
1246 
1247         icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
1248 }
1249 
1250 /*
1251  * icmp_inbound_v4 deals with ICMP messages that are handled by IP.
1252  * If the ICMP message is consumed by IP, i.e., it should not be delivered
1253  * to any IPPROTO_ICMP raw sockets, then it returns NULL.
1254  * Likewise, if the ICMP error is misformed (too short, etc), then it
1255  * returns NULL. The caller uses this to determine whether or not to send
1256  * to raw sockets.
1257  *
1258  * All error messages are passed to the matching transport stream.
1259  *
1260  * The following cases are handled by icmp_inbound:
1261  * 1) It needs to send a reply back and possibly delivering it
1262  *    to the "interested" upper clients.
1263  * 2) Return the mblk so that the caller can pass it to the RAW socket clients.
1264  * 3) It needs to change some values in IP only.
1265  * 4) It needs to change some values in IP and upper layers e.g TCP
1266  *    by delivering an error to the upper layers.
1267  *
1268  * We handle the above three cases in the context of IPsec in the
1269  * following way :
1270  *
1271  * 1) Send the reply back in the same way as the request came in.
1272  *    If it came in encrypted, it goes out encrypted. If it came in
1273  *    clear, it goes out in clear. Thus, this will prevent chosen
1274  *    plain text attack.
1275  * 2) The client may or may not expect things to come in secure.
1276  *    If it comes in secure, the policy constraints are checked
1277  *    before delivering it to the upper layers. If it comes in
1278  *    clear, ipsec_inbound_accept_clear will decide whether to
1279  *    accept this in clear or not. In both the cases, if the returned
1280  *    message (IP header + 8 bytes) that caused the icmp message has
1281  *    AH/ESP headers, it is sent up to AH/ESP for validation before
1282  *    sending up. If there are only 8 bytes of returned message, then
1283  *    upper client will not be notified.
1284  * 3) Check with global policy to see whether it matches the constaints.
1285  *    But this will be done only if icmp_accept_messages_in_clear is
1286  *    zero.
1287  * 4) If we need to change both in IP and ULP, then the decision taken
1288  *    while affecting the values in IP and while delivering up to TCP
1289  *    should be the same.
1290  *
1291  *      There are two cases.
1292  *
1293  *      a) If we reject data at the IP layer (ipsec_check_global_policy()
1294  *         failed), we will not deliver it to the ULP, even though they
1295  *         are *willing* to accept in *clear*. This is fine as our global
1296  *         disposition to icmp messages asks us reject the datagram.
1297  *
1298  *      b) If we accept data at the IP layer (ipsec_check_global_policy()
1299  *         succeeded or icmp_accept_messages_in_clear is 1), and not able
1300  *         to deliver it to ULP (policy failed), it can lead to
1301  *         consistency problems. The cases known at this time are
1302  *         ICMP_DESTINATION_UNREACHABLE  messages with following code
1303  *         values :
1304  *
1305  *         - ICMP_FRAGMENTATION_NEEDED : IP adapts to the new value
1306  *           and Upper layer rejects. Then the communication will
1307  *           come to a stop. This is solved by making similar decisions
1308  *           at both levels. Currently, when we are unable to deliver
1309  *           to the Upper Layer (due to policy failures) while IP has
1310  *           adjusted dce_pmtu, the next outbound datagram would
1311  *           generate a local ICMP_FRAGMENTATION_NEEDED message - which
1312  *           will be with the right level of protection. Thus the right
1313  *           value will be communicated even if we are not able to
1314  *           communicate when we get from the wire initially. But this
1315  *           assumes there would be at least one outbound datagram after
1316  *           IP has adjusted its dce_pmtu value. To make things
1317  *           simpler, we accept in clear after the validation of
1318  *           AH/ESP headers.
1319  *
1320  *         - Other ICMP ERRORS : We may not be able to deliver it to the
1321  *           upper layer depending on the level of protection the upper
1322  *           layer expects and the disposition in ipsec_inbound_accept_clear().
1323  *           ipsec_inbound_accept_clear() decides whether a given ICMP error
1324  *           should be accepted in clear when the Upper layer expects secure.
1325  *           Thus the communication may get aborted by some bad ICMP
1326  *           packets.
1327  */
1328 mblk_t *
1329 icmp_inbound_v4(mblk_t *mp, ip_recv_attr_t *ira)
1330 {
1331         icmph_t         *icmph;
1332         ipha_t          *ipha;          /* Outer header */
1333         int             ip_hdr_length;  /* Outer header length */
1334         boolean_t       interested;
1335         ipif_t          *ipif;
1336         uint32_t        ts;
1337         uint32_t        *tsp;
1338         timestruc_t     now;
1339         ill_t           *ill = ira->ira_ill;
1340         ip_stack_t      *ipst = ill->ill_ipst;
1341         zoneid_t        zoneid = ira->ira_zoneid;
1342         int             len_needed;
1343         mblk_t          *mp_ret = NULL;
1344 
1345         ipha = (ipha_t *)mp->b_rptr;
1346 
1347         BUMP_MIB(&ipst->ips_icmp_mib, icmpInMsgs);
1348 
1349         ip_hdr_length = ira->ira_ip_hdr_length;
1350         if ((mp->b_wptr - mp->b_rptr) < (ip_hdr_length + ICMPH_SIZE)) {
1351                 if (ira->ira_pktlen < (ip_hdr_length + ICMPH_SIZE)) {
1352                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
1353                         ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
1354                         freemsg(mp);
1355                         return (NULL);
1356                 }
1357                 /* Last chance to get real. */
1358                 ipha = ip_pullup(mp, ip_hdr_length + ICMPH_SIZE, ira);
1359                 if (ipha == NULL) {
1360                         BUMP_MIB(&ipst->ips_icmp_mib, icmpInErrors);
1361                         freemsg(mp);
1362                         return (NULL);
1363                 }
1364         }
1365 
1366         /* The IP header will always be a multiple of four bytes */
1367         icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1368         ip2dbg(("icmp_inbound_v4: type %d code %d\n", icmph->icmph_type,
1369             icmph->icmph_code));
1370 
1371         /*
1372          * We will set "interested" to "true" if we should pass a copy to
1373          * the transport or if we handle the packet locally.
1374          */
1375         interested = B_FALSE;
1376         switch (icmph->icmph_type) {
1377         case ICMP_ECHO_REPLY:
1378                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInEchoReps);
1379                 break;
1380         case ICMP_DEST_UNREACHABLE:
1381                 if (icmph->icmph_code == ICMP_FRAGMENTATION_NEEDED)
1382                         BUMP_MIB(&ipst->ips_icmp_mib, icmpInFragNeeded);
1383                 interested = B_TRUE;    /* Pass up to transport */
1384                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInDestUnreachs);
1385                 break;
1386         case ICMP_SOURCE_QUENCH:
1387                 interested = B_TRUE;    /* Pass up to transport */
1388                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInSrcQuenchs);
1389                 break;
1390         case ICMP_REDIRECT:
1391                 if (!ipst->ips_ip_ignore_redirect)
1392                         interested = B_TRUE;
1393                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInRedirects);
1394                 break;
1395         case ICMP_ECHO_REQUEST:
1396                 /*
1397                  * Whether to respond to echo requests that come in as IP
1398                  * broadcasts or as IP multicast is subject to debate
1399                  * (what isn't?).  We aim to please, you pick it.
1400                  * Default is do it.
1401                  */
1402                 if (ira->ira_flags & IRAF_MULTICAST) {
1403                         /* multicast: respond based on tunable */
1404                         interested = ipst->ips_ip_g_resp_to_echo_mcast;
1405                 } else if (ira->ira_flags & IRAF_BROADCAST) {
1406                         /* broadcast: respond based on tunable */
1407                         interested = ipst->ips_ip_g_resp_to_echo_bcast;
1408                 } else {
1409                         /* unicast: always respond */
1410                         interested = B_TRUE;
1411                 }
1412                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInEchos);
1413                 if (!interested) {
1414                         /* We never pass these to RAW sockets */
1415                         freemsg(mp);
1416                         return (NULL);
1417                 }
1418 
1419                 /* Check db_ref to make sure we can modify the packet. */
1420                 if (mp->b_datap->db_ref > 1) {
1421                         mblk_t  *mp1;
1422 
1423                         mp1 = copymsg(mp);
1424                         freemsg(mp);
1425                         if (!mp1) {
1426                                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
1427                                 return (NULL);
1428                         }
1429                         mp = mp1;
1430                         ipha = (ipha_t *)mp->b_rptr;
1431                         icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1432                 }
1433                 icmph->icmph_type = ICMP_ECHO_REPLY;
1434                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutEchoReps);
1435                 icmp_send_reply_v4(mp, ipha, icmph, ira);
1436                 return (NULL);
1437 
1438         case ICMP_ROUTER_ADVERTISEMENT:
1439         case ICMP_ROUTER_SOLICITATION:
1440                 break;
1441         case ICMP_TIME_EXCEEDED:
1442                 interested = B_TRUE;    /* Pass up to transport */
1443                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInTimeExcds);
1444                 break;
1445         case ICMP_PARAM_PROBLEM:
1446                 interested = B_TRUE;    /* Pass up to transport */
1447                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInParmProbs);
1448                 break;
1449         case ICMP_TIME_STAMP_REQUEST:
1450                 /* Response to Time Stamp Requests is local policy. */
1451                 if (ipst->ips_ip_g_resp_to_timestamp) {
1452                         if (ira->ira_flags & IRAF_MULTIBROADCAST)
1453                                 interested =
1454                                     ipst->ips_ip_g_resp_to_timestamp_bcast;
1455                         else
1456                                 interested = B_TRUE;
1457                 }
1458                 if (!interested) {
1459                         /* We never pass these to RAW sockets */
1460                         freemsg(mp);
1461                         return (NULL);
1462                 }
1463 
1464                 /* Make sure we have enough of the packet */
1465                 len_needed = ip_hdr_length + ICMPH_SIZE +
1466                     3 * sizeof (uint32_t);
1467 
1468                 if (mp->b_wptr - mp->b_rptr < len_needed) {
1469                         ipha = ip_pullup(mp, len_needed, ira);
1470                         if (ipha == NULL) {
1471                                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1472                                 ip_drop_input("ipIfStatsInDiscards - ip_pullup",
1473                                     mp, ill);
1474                                 freemsg(mp);
1475                                 return (NULL);
1476                         }
1477                         /* Refresh following the pullup. */
1478                         icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1479                 }
1480                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInTimestamps);
1481                 /* Check db_ref to make sure we can modify the packet. */
1482                 if (mp->b_datap->db_ref > 1) {
1483                         mblk_t  *mp1;
1484 
1485                         mp1 = copymsg(mp);
1486                         freemsg(mp);
1487                         if (!mp1) {
1488                                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
1489                                 return (NULL);
1490                         }
1491                         mp = mp1;
1492                         ipha = (ipha_t *)mp->b_rptr;
1493                         icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1494                 }
1495                 icmph->icmph_type = ICMP_TIME_STAMP_REPLY;
1496                 tsp = (uint32_t *)&icmph[1];
1497                 tsp++;          /* Skip past 'originate time' */
1498                 /* Compute # of milliseconds since midnight */
1499                 gethrestime(&now);
1500                 ts = (now.tv_sec % (24 * 60 * 60)) * 1000 +
1501                     now.tv_nsec / (NANOSEC / MILLISEC);
1502                 *tsp++ = htonl(ts);     /* Lay in 'receive time' */
1503                 *tsp++ = htonl(ts);     /* Lay in 'send time' */
1504                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutTimestampReps);
1505                 icmp_send_reply_v4(mp, ipha, icmph, ira);
1506                 return (NULL);
1507 
1508         case ICMP_TIME_STAMP_REPLY:
1509                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInTimestampReps);
1510                 break;
1511         case ICMP_INFO_REQUEST:
1512                 /* Per RFC 1122 3.2.2.7, ignore this. */
1513         case ICMP_INFO_REPLY:
1514                 break;
1515         case ICMP_ADDRESS_MASK_REQUEST:
1516                 if (ira->ira_flags & IRAF_MULTIBROADCAST) {
1517                         interested =
1518                             ipst->ips_ip_respond_to_address_mask_broadcast;
1519                 } else {
1520                         interested = B_TRUE;
1521                 }
1522                 if (!interested) {
1523                         /* We never pass these to RAW sockets */
1524                         freemsg(mp);
1525                         return (NULL);
1526                 }
1527                 len_needed = ip_hdr_length + ICMPH_SIZE + IP_ADDR_LEN;
1528                 if (mp->b_wptr - mp->b_rptr < len_needed) {
1529                         ipha = ip_pullup(mp, len_needed, ira);
1530                         if (ipha == NULL) {
1531                                 BUMP_MIB(ill->ill_ip_mib,
1532                                     ipIfStatsInTruncatedPkts);
1533                                 ip_drop_input("ipIfStatsInTruncatedPkts", mp,
1534                                     ill);
1535                                 freemsg(mp);
1536                                 return (NULL);
1537                         }
1538                         /* Refresh following the pullup. */
1539                         icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1540                 }
1541                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInAddrMasks);
1542                 /* Check db_ref to make sure we can modify the packet. */
1543                 if (mp->b_datap->db_ref > 1) {
1544                         mblk_t  *mp1;
1545 
1546                         mp1 = copymsg(mp);
1547                         freemsg(mp);
1548                         if (!mp1) {
1549                                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
1550                                 return (NULL);
1551                         }
1552                         mp = mp1;
1553                         ipha = (ipha_t *)mp->b_rptr;
1554                         icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1555                 }
1556                 /*
1557                  * Need the ipif with the mask be the same as the source
1558                  * address of the mask reply. For unicast we have a specific
1559                  * ipif. For multicast/broadcast we only handle onlink
1560                  * senders, and use the source address to pick an ipif.
1561                  */
1562                 ipif = ipif_lookup_addr(ipha->ipha_dst, ill, zoneid, ipst);
1563                 if (ipif == NULL) {
1564                         /* Broadcast or multicast */
1565                         ipif = ipif_lookup_remote(ill, ipha->ipha_src, zoneid);
1566                         if (ipif == NULL) {
1567                                 freemsg(mp);
1568                                 return (NULL);
1569                         }
1570                 }
1571                 icmph->icmph_type = ICMP_ADDRESS_MASK_REPLY;
1572                 bcopy(&ipif->ipif_net_mask, &icmph[1], IP_ADDR_LEN);
1573                 ipif_refrele(ipif);
1574                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutAddrMaskReps);
1575                 icmp_send_reply_v4(mp, ipha, icmph, ira);
1576                 return (NULL);
1577 
1578         case ICMP_ADDRESS_MASK_REPLY:
1579                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInAddrMaskReps);
1580                 break;
1581         default:
1582                 interested = B_TRUE;    /* Pass up to transport */
1583                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInUnknowns);
1584                 break;
1585         }
1586         /*
1587          * See if there is an ICMP client to avoid an extra copymsg/freemsg
1588          * if there isn't one.
1589          */
1590         if (ipst->ips_ipcl_proto_fanout_v4[IPPROTO_ICMP].connf_head != NULL) {
1591                 /* If there is an ICMP client and we want one too, copy it. */
1592 
1593                 if (!interested) {
1594                         /* Caller will deliver to RAW sockets */
1595                         return (mp);
1596                 }
1597                 mp_ret = copymsg(mp);
1598                 if (mp_ret == NULL) {
1599                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1600                         ip_drop_input("ipIfStatsInDiscards - copymsg", mp, ill);
1601                 }
1602         } else if (!interested) {
1603                 /* Neither we nor raw sockets are interested. Drop packet now */
1604                 freemsg(mp);
1605                 return (NULL);
1606         }
1607 
1608         /*
1609          * ICMP error or redirect packet. Make sure we have enough of
1610          * the header and that db_ref == 1 since we might end up modifying
1611          * the packet.
1612          */
1613         if (mp->b_cont != NULL) {
1614                 if (ip_pullup(mp, -1, ira) == NULL) {
1615                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1616                         ip_drop_input("ipIfStatsInDiscards - ip_pullup",
1617                             mp, ill);
1618                         freemsg(mp);
1619                         return (mp_ret);
1620                 }
1621         }
1622 
1623         if (mp->b_datap->db_ref > 1) {
1624                 mblk_t  *mp1;
1625 
1626                 mp1 = copymsg(mp);
1627                 if (mp1 == NULL) {
1628                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1629                         ip_drop_input("ipIfStatsInDiscards - copymsg", mp, ill);
1630                         freemsg(mp);
1631                         return (mp_ret);
1632                 }
1633                 freemsg(mp);
1634                 mp = mp1;
1635         }
1636 
1637         /*
1638          * In case mp has changed, verify the message before any further
1639          * processes.
1640          */
1641         ipha = (ipha_t *)mp->b_rptr;
1642         icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1643         if (!icmp_inbound_verify_v4(mp, icmph, ira)) {
1644                 freemsg(mp);
1645                 return (mp_ret);
1646         }
1647 
1648         switch (icmph->icmph_type) {
1649         case ICMP_REDIRECT:
1650                 icmp_redirect_v4(mp, ipha, icmph, ira);
1651                 break;
1652         case ICMP_DEST_UNREACHABLE:
1653                 if (icmph->icmph_code == ICMP_FRAGMENTATION_NEEDED) {
1654                         /* Update DCE and adjust MTU is icmp header if needed */
1655                         icmp_inbound_too_big_v4(icmph, ira);
1656                 }
1657                 /* FALLTHRU */
1658         default:
1659                 icmp_inbound_error_fanout_v4(mp, icmph, ira);
1660                 break;
1661         }
1662         return (mp_ret);
1663 }
1664 
1665 /*
1666  * Send an ICMP echo, timestamp or address mask reply.
1667  * The caller has already updated the payload part of the packet.
1668  * We handle the ICMP checksum, IP source address selection and feed
1669  * the packet into ip_output_simple.
1670  */
1671 static void
1672 icmp_send_reply_v4(mblk_t *mp, ipha_t *ipha, icmph_t *icmph,
1673     ip_recv_attr_t *ira)
1674 {
1675         uint_t          ip_hdr_length = ira->ira_ip_hdr_length;
1676         ill_t           *ill = ira->ira_ill;
1677         ip_stack_t      *ipst = ill->ill_ipst;
1678         ip_xmit_attr_t  ixas;
1679 
1680         /* Send out an ICMP packet */
1681         icmph->icmph_checksum = 0;
1682         icmph->icmph_checksum = IP_CSUM(mp, ip_hdr_length, 0);
1683         /* Reset time to live. */
1684         ipha->ipha_ttl = ipst->ips_ip_def_ttl;
1685         {
1686                 /* Swap source and destination addresses */
1687                 ipaddr_t tmp;
1688 
1689                 tmp = ipha->ipha_src;
1690                 ipha->ipha_src = ipha->ipha_dst;
1691                 ipha->ipha_dst = tmp;
1692         }
1693         ipha->ipha_ident = 0;
1694         if (!IS_SIMPLE_IPH(ipha))
1695                 icmp_options_update(ipha);
1696 
1697         bzero(&ixas, sizeof (ixas));
1698         ixas.ixa_flags = IXAF_BASIC_SIMPLE_V4;
1699         ixas.ixa_zoneid = ira->ira_zoneid;
1700         ixas.ixa_cred = kcred;
1701         ixas.ixa_cpid = NOPID;
1702         ixas.ixa_tsl = ira->ira_tsl; /* Behave as a multi-level responder */
1703         ixas.ixa_ifindex = 0;
1704         ixas.ixa_ipst = ipst;
1705         ixas.ixa_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1706 
1707         if (!(ira->ira_flags & IRAF_IPSEC_SECURE)) {
1708                 /*
1709                  * This packet should go out the same way as it
1710                  * came in i.e in clear, independent of the IPsec policy
1711                  * for transmitting packets.
1712                  */
1713                 ixas.ixa_flags |= IXAF_NO_IPSEC;
1714         } else {
1715                 if (!ipsec_in_to_out(ira, &ixas, mp, ipha, NULL)) {
1716                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1717                         /* Note: mp already consumed and ip_drop_packet done */
1718                         return;
1719                 }
1720         }
1721         if (ira->ira_flags & IRAF_MULTIBROADCAST) {
1722                 /*
1723                  * Not one or our addresses (IRE_LOCALs), thus we let
1724                  * ip_output_simple pick the source.
1725                  */
1726                 ipha->ipha_src = INADDR_ANY;
1727                 ixas.ixa_flags |= IXAF_SET_SOURCE;
1728         }
1729         /* Should we send with DF and use dce_pmtu? */
1730         if (ipst->ips_ipv4_icmp_return_pmtu) {
1731                 ixas.ixa_flags |= IXAF_PMTU_DISCOVERY;
1732                 ipha->ipha_fragment_offset_and_flags |= IPH_DF_HTONS;
1733         }
1734 
1735         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutMsgs);
1736 
1737         (void) ip_output_simple(mp, &ixas);
1738         ixa_cleanup(&ixas);
1739 }
1740 
1741 /*
1742  * Verify the ICMP messages for either for ICMP error or redirect packet.
1743  * The caller should have fully pulled up the message. If it's a redirect
1744  * packet, only basic checks on IP header will be done; otherwise, verify
1745  * the packet by looking at the included ULP header.
1746  *
1747  * Called before icmp_inbound_error_fanout_v4 is called.
1748  */
1749 static boolean_t
1750 icmp_inbound_verify_v4(mblk_t *mp, icmph_t *icmph, ip_recv_attr_t *ira)
1751 {
1752         ill_t           *ill = ira->ira_ill;
1753         int             hdr_length;
1754         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
1755         conn_t          *connp;
1756         ipha_t          *ipha;  /* Inner IP header */
1757 
1758         ipha = (ipha_t *)&icmph[1];
1759         if ((uchar_t *)ipha + IP_SIMPLE_HDR_LENGTH > mp->b_wptr)
1760                 goto truncated;
1761 
1762         hdr_length = IPH_HDR_LENGTH(ipha);
1763 
1764         if ((IPH_HDR_VERSION(ipha) != IPV4_VERSION))
1765                 goto discard_pkt;
1766 
1767         if (hdr_length < sizeof (ipha_t))
1768                 goto truncated;
1769 
1770         if ((uchar_t *)ipha + hdr_length > mp->b_wptr)
1771                 goto truncated;
1772 
1773         /*
1774          * Stop here for ICMP_REDIRECT.
1775          */
1776         if (icmph->icmph_type == ICMP_REDIRECT)
1777                 return (B_TRUE);
1778 
1779         /*
1780          * ICMP errors only.
1781          */
1782         switch (ipha->ipha_protocol) {
1783         case IPPROTO_UDP:
1784                 /*
1785                  * Verify we have at least ICMP_MIN_TP_HDR_LEN bytes of
1786                  * transport header.
1787                  */
1788                 if ((uchar_t *)ipha + hdr_length + ICMP_MIN_TP_HDR_LEN >
1789                     mp->b_wptr)
1790                         goto truncated;
1791                 break;
1792         case IPPROTO_TCP: {
1793                 tcpha_t         *tcpha;
1794 
1795                 /*
1796                  * Verify we have at least ICMP_MIN_TP_HDR_LEN bytes of
1797                  * transport header.
1798                  */
1799                 if ((uchar_t *)ipha + hdr_length + ICMP_MIN_TP_HDR_LEN >
1800                     mp->b_wptr)
1801                         goto truncated;
1802 
1803                 tcpha = (tcpha_t *)((uchar_t *)ipha + hdr_length);
1804                 connp = ipcl_tcp_lookup_reversed_ipv4(ipha, tcpha, TCPS_LISTEN,
1805                     ipst);
1806                 if (connp == NULL)
1807                         goto discard_pkt;
1808 
1809                 if ((connp->conn_verifyicmp != NULL) &&
1810                     !connp->conn_verifyicmp(connp, tcpha, icmph, NULL, ira)) {
1811                         CONN_DEC_REF(connp);
1812                         goto discard_pkt;
1813                 }
1814                 CONN_DEC_REF(connp);
1815                 break;
1816         }
1817         case IPPROTO_SCTP:
1818                 /*
1819                  * Verify we have at least ICMP_MIN_TP_HDR_LEN bytes of
1820                  * transport header.
1821                  */
1822                 if ((uchar_t *)ipha + hdr_length + ICMP_MIN_TP_HDR_LEN >
1823                     mp->b_wptr)
1824                         goto truncated;
1825                 break;
1826         case IPPROTO_ESP:
1827         case IPPROTO_AH:
1828                 break;
1829         case IPPROTO_ENCAP:
1830                 if ((uchar_t *)ipha + hdr_length + sizeof (ipha_t) >
1831                     mp->b_wptr)
1832                         goto truncated;
1833                 break;
1834         default:
1835                 break;
1836         }
1837 
1838         return (B_TRUE);
1839 
1840 discard_pkt:
1841         /* Bogus ICMP error. */
1842         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1843         return (B_FALSE);
1844 
1845 truncated:
1846         /* We pulled up everthing already. Must be truncated */
1847         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
1848         ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
1849         return (B_FALSE);
1850 }
1851 
1852 /* Table from RFC 1191 */
1853 static int icmp_frag_size_table[] =
1854 { 32000, 17914, 8166, 4352, 2002, 1496, 1006, 508, 296, 68 };
1855 
1856 /*
1857  * Process received ICMP Packet too big.
1858  * Just handles the DCE create/update, including using the above table of
1859  * PMTU guesses. The caller is responsible for validating the packet before
1860  * passing it in and also to fanout the ICMP error to any matching transport
1861  * conns. Assumes the message has been fully pulled up and verified.
1862  *
1863  * Before getting here, the caller has called icmp_inbound_verify_v4()
1864  * that should have verified with ULP to prevent undoing the changes we're
1865  * going to make to DCE. For example, TCP might have verified that the packet
1866  * which generated error is in the send window.
1867  *
1868  * In some cases modified this MTU in the ICMP header packet; the caller
1869  * should pass to the matching ULP after this returns.
1870  */
1871 static void
1872 icmp_inbound_too_big_v4(icmph_t *icmph, ip_recv_attr_t *ira)
1873 {
1874         dce_t           *dce;
1875         int             old_mtu;
1876         int             mtu, orig_mtu;
1877         ipaddr_t        dst;
1878         boolean_t       disable_pmtud;
1879         ill_t           *ill = ira->ira_ill;
1880         ip_stack_t      *ipst = ill->ill_ipst;
1881         uint_t          hdr_length;
1882         ipha_t          *ipha;
1883 
1884         /* Caller already pulled up everything. */
1885         ipha = (ipha_t *)&icmph[1];
1886         ASSERT(icmph->icmph_type == ICMP_DEST_UNREACHABLE &&
1887             icmph->icmph_code == ICMP_FRAGMENTATION_NEEDED);
1888         ASSERT(ill != NULL);
1889 
1890         hdr_length = IPH_HDR_LENGTH(ipha);
1891 
1892         /*
1893          * We handle path MTU for source routed packets since the DCE
1894          * is looked up using the final destination.
1895          */
1896         dst = ip_get_dst(ipha);
1897 
1898         dce = dce_lookup_and_add_v4(dst, ipst);
1899         if (dce == NULL) {
1900                 /* Couldn't add a unique one - ENOMEM */
1901                 ip1dbg(("icmp_inbound_too_big_v4: no dce for 0x%x\n",
1902                     ntohl(dst)));
1903                 return;
1904         }
1905 
1906         /* Check for MTU discovery advice as described in RFC 1191 */
1907         mtu = ntohs(icmph->icmph_du_mtu);
1908         orig_mtu = mtu;
1909         disable_pmtud = B_FALSE;
1910 
1911         mutex_enter(&dce->dce_lock);
1912         if (dce->dce_flags & DCEF_PMTU)
1913                 old_mtu = dce->dce_pmtu;
1914         else
1915                 old_mtu = ill->ill_mtu;
1916 
1917         if (icmph->icmph_du_zero != 0 || mtu < ipst->ips_ip_pmtu_min) {
1918                 uint32_t length;
1919                 int     i;
1920 
1921                 /*
1922                  * Use the table from RFC 1191 to figure out
1923                  * the next "plateau" based on the length in
1924                  * the original IP packet.
1925                  */
1926                 length = ntohs(ipha->ipha_length);
1927                 DTRACE_PROBE2(ip4__pmtu__guess, dce_t *, dce,
1928                     uint32_t, length);
1929                 if (old_mtu <= length &&
1930                     old_mtu >= length - hdr_length) {
1931                         /*
1932                          * Handle broken BSD 4.2 systems that
1933                          * return the wrong ipha_length in ICMP
1934                          * errors.
1935                          */
1936                         ip1dbg(("Wrong mtu: sent %d, dce %d\n",
1937                             length, old_mtu));
1938                         length -= hdr_length;
1939                 }
1940                 for (i = 0; i < A_CNT(icmp_frag_size_table); i++) {
1941                         if (length > icmp_frag_size_table[i])
1942                                 break;
1943                 }
1944                 if (i == A_CNT(icmp_frag_size_table)) {
1945                         /* Smaller than IP_MIN_MTU! */
1946                         ip1dbg(("Too big for packet size %d\n",
1947                             length));
1948                         disable_pmtud = B_TRUE;
1949                         mtu = ipst->ips_ip_pmtu_min;
1950                 } else {
1951                         mtu = icmp_frag_size_table[i];
1952                         ip1dbg(("Calculated mtu %d, packet size %d, "
1953                             "before %d\n", mtu, length, old_mtu));
1954                         if (mtu < ipst->ips_ip_pmtu_min) {
1955                                 mtu = ipst->ips_ip_pmtu_min;
1956                                 disable_pmtud = B_TRUE;
1957                         }
1958                 }
1959         }
1960         if (disable_pmtud)
1961                 dce->dce_flags |= DCEF_TOO_SMALL_PMTU;
1962         else
1963                 dce->dce_flags &= ~DCEF_TOO_SMALL_PMTU;
1964 
1965         dce->dce_pmtu = MIN(old_mtu, mtu);
1966         /* Prepare to send the new max frag size for the ULP. */
1967         icmph->icmph_du_zero = 0;
1968         icmph->icmph_du_mtu =  htons((uint16_t)dce->dce_pmtu);
1969         DTRACE_PROBE4(ip4__pmtu__change, icmph_t *, icmph, dce_t *,
1970             dce, int, orig_mtu, int, mtu);
1971 
1972         /* We now have a PMTU for sure */
1973         dce->dce_flags |= DCEF_PMTU;
1974         dce->dce_last_change_time = TICK_TO_SEC(ddi_get_lbolt64());
1975         mutex_exit(&dce->dce_lock);
1976         /*
1977          * After dropping the lock the new value is visible to everyone.
1978          * Then we bump the generation number so any cached values reinspect
1979          * the dce_t.
1980          */
1981         dce_increment_generation(dce);
1982         dce_refrele(dce);
1983 }
1984 
1985 /*
1986  * If the packet in error is Self-Encapsulated, icmp_inbound_error_fanout_v4
1987  * calls this function.
1988  */
1989 static mblk_t *
1990 icmp_inbound_self_encap_error_v4(mblk_t *mp, ipha_t *ipha, ipha_t *in_ipha)
1991 {
1992         int length;
1993 
1994         ASSERT(mp->b_datap->db_type == M_DATA);
1995 
1996         /* icmp_inbound_v4 has already pulled up the whole error packet */
1997         ASSERT(mp->b_cont == NULL);
1998 
1999         /*
2000          * The length that we want to overlay is the inner header
2001          * and what follows it.
2002          */
2003         length = msgdsize(mp) - ((uchar_t *)in_ipha - mp->b_rptr);
2004 
2005         /*
2006          * Overlay the inner header and whatever follows it over the
2007          * outer header.
2008          */
2009         bcopy((uchar_t *)in_ipha, (uchar_t *)ipha, length);
2010 
2011         /* Adjust for what we removed */
2012         mp->b_wptr -= (uchar_t *)in_ipha - (uchar_t *)ipha;
2013         return (mp);
2014 }
2015 
2016 /*
2017  * Try to pass the ICMP message upstream in case the ULP cares.
2018  *
2019  * If the packet that caused the ICMP error is secure, we send
2020  * it to AH/ESP to make sure that the attached packet has a
2021  * valid association. ipha in the code below points to the
2022  * IP header of the packet that caused the error.
2023  *
2024  * For IPsec cases, we let the next-layer-up (which has access to
2025  * cached policy on the conn_t, or can query the SPD directly)
2026  * subtract out any IPsec overhead if they must.  We therefore make no
2027  * adjustments here for IPsec overhead.
2028  *
2029  * IFN could have been generated locally or by some router.
2030  *
2031  * LOCAL : ire_send_wire (before calling ipsec_out_process) can call
2032  * icmp_frag_needed/icmp_pkt2big_v6 to generated a local IFN.
2033  *          This happens because IP adjusted its value of MTU on an
2034  *          earlier IFN message and could not tell the upper layer,
2035  *          the new adjusted value of MTU e.g. Packet was encrypted
2036  *          or there was not enough information to fanout to upper
2037  *          layers. Thus on the next outbound datagram, ire_send_wire
2038  *          generates the IFN, where IPsec processing has *not* been
2039  *          done.
2040  *
2041  *          Note that we retain ixa_fragsize across IPsec thus once
2042  *          we have picking ixa_fragsize and entered ipsec_out_process we do
2043  *          no change the fragsize even if the path MTU changes before
2044  *          we reach ip_output_post_ipsec.
2045  *
2046  *          In the local case, IRAF_LOOPBACK will be set indicating
2047  *          that IFN was generated locally.
2048  *
2049  * ROUTER : IFN could be secure or non-secure.
2050  *
2051  *          * SECURE : We use the IPSEC_IN to fanout to AH/ESP if the
2052  *            packet in error has AH/ESP headers to validate the AH/ESP
2053  *            headers. AH/ESP will verify whether there is a valid SA or
2054  *            not and send it back. We will fanout again if we have more
2055  *            data in the packet.
2056  *
2057  *            If the packet in error does not have AH/ESP, we handle it
2058  *            like any other case.
2059  *
2060  *          * NON_SECURE : If the packet in error has AH/ESP headers, we send it
2061  *            up to AH/ESP for validation. AH/ESP will verify whether there is a
2062  *            valid SA or not and send it back. We will fanout again if
2063  *            we have more data in the packet.
2064  *
2065  *            If the packet in error does not have AH/ESP, we handle it
2066  *            like any other case.
2067  *
2068  * The caller must have called icmp_inbound_verify_v4.
2069  */
2070 static void
2071 icmp_inbound_error_fanout_v4(mblk_t *mp, icmph_t *icmph, ip_recv_attr_t *ira)
2072 {
2073         uint16_t        *up;    /* Pointer to ports in ULP header */
2074         uint32_t        ports;  /* reversed ports for fanout */
2075         ipha_t          ripha;  /* With reversed addresses */
2076         ipha_t          *ipha;  /* Inner IP header */
2077         uint_t          hdr_length; /* Inner IP header length */
2078         tcpha_t         *tcpha;
2079         conn_t          *connp;
2080         ill_t           *ill = ira->ira_ill;
2081         ip_stack_t      *ipst = ill->ill_ipst;
2082         ipsec_stack_t   *ipss = ipst->ips_netstack->netstack_ipsec;
2083         ill_t           *rill = ira->ira_rill;
2084 
2085         /* Caller already pulled up everything. */
2086         ipha = (ipha_t *)&icmph[1];
2087         ASSERT((uchar_t *)&ipha[1] <= mp->b_wptr);
2088         ASSERT(mp->b_cont == NULL);
2089 
2090         hdr_length = IPH_HDR_LENGTH(ipha);
2091         ira->ira_protocol = ipha->ipha_protocol;
2092 
2093         /*
2094          * We need a separate IP header with the source and destination
2095          * addresses reversed to do fanout/classification because the ipha in
2096          * the ICMP error is in the form we sent it out.
2097          */
2098         ripha.ipha_src = ipha->ipha_dst;
2099         ripha.ipha_dst = ipha->ipha_src;
2100         ripha.ipha_protocol = ipha->ipha_protocol;
2101         ripha.ipha_version_and_hdr_length = ipha->ipha_version_and_hdr_length;
2102 
2103         ip2dbg(("icmp_inbound_error_v4: proto %d %x to %x: %d/%d\n",
2104             ripha.ipha_protocol, ntohl(ipha->ipha_src),
2105             ntohl(ipha->ipha_dst),
2106             icmph->icmph_type, icmph->icmph_code));
2107 
2108         switch (ipha->ipha_protocol) {
2109         case IPPROTO_UDP:
2110                 up = (uint16_t *)((uchar_t *)ipha + hdr_length);
2111 
2112                 /* Attempt to find a client stream based on port. */
2113                 ip2dbg(("icmp_inbound_error_v4: UDP ports %d to %d\n",
2114                     ntohs(up[0]), ntohs(up[1])));
2115 
2116                 /* Note that we send error to all matches. */
2117                 ira->ira_flags |= IRAF_ICMP_ERROR;
2118                 ip_fanout_udp_multi_v4(mp, &ripha, up[0], up[1], ira);
2119                 ira->ira_flags &= ~IRAF_ICMP_ERROR;
2120                 return;
2121 
2122         case IPPROTO_TCP:
2123                 /*
2124                  * Find a TCP client stream for this packet.
2125                  * Note that we do a reverse lookup since the header is
2126                  * in the form we sent it out.
2127                  */
2128                 tcpha = (tcpha_t *)((uchar_t *)ipha + hdr_length);
2129                 connp = ipcl_tcp_lookup_reversed_ipv4(ipha, tcpha, TCPS_LISTEN,
2130                     ipst);
2131                 if (connp == NULL)
2132                         goto discard_pkt;
2133 
2134                 if (CONN_INBOUND_POLICY_PRESENT(connp, ipss) ||
2135                     (ira->ira_flags & IRAF_IPSEC_SECURE)) {
2136                         mp = ipsec_check_inbound_policy(mp, connp,
2137                             ipha, NULL, ira);
2138                         if (mp == NULL) {
2139                                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
2140                                 /* Note that mp is NULL */
2141                                 ip_drop_input("ipIfStatsInDiscards", mp, ill);
2142                                 CONN_DEC_REF(connp);
2143                                 return;
2144                         }
2145                 }
2146 
2147                 ira->ira_flags |= IRAF_ICMP_ERROR;
2148                 ira->ira_ill = ira->ira_rill = NULL;
2149                 if (IPCL_IS_TCP(connp)) {
2150                         SQUEUE_ENTER_ONE(connp->conn_sqp, mp,
2151                             connp->conn_recvicmp, connp, ira, SQ_FILL,
2152                             SQTAG_TCP_INPUT_ICMP_ERR);
2153                 } else {
2154                         /* Not TCP; must be SOCK_RAW, IPPROTO_TCP */
2155                         (connp->conn_recv)(connp, mp, NULL, ira);
2156                         CONN_DEC_REF(connp);
2157                 }
2158                 ira->ira_ill = ill;
2159                 ira->ira_rill = rill;
2160                 ira->ira_flags &= ~IRAF_ICMP_ERROR;
2161                 return;
2162 
2163         case IPPROTO_SCTP:
2164                 up = (uint16_t *)((uchar_t *)ipha + hdr_length);
2165                 /* Find a SCTP client stream for this packet. */
2166                 ((uint16_t *)&ports)[0] = up[1];
2167                 ((uint16_t *)&ports)[1] = up[0];
2168 
2169                 ira->ira_flags |= IRAF_ICMP_ERROR;
2170                 ip_fanout_sctp(mp, &ripha, NULL, ports, ira);
2171                 ira->ira_flags &= ~IRAF_ICMP_ERROR;
2172                 return;
2173 
2174         case IPPROTO_ESP:
2175         case IPPROTO_AH:
2176                 if (!ipsec_loaded(ipss)) {
2177                         ip_proto_not_sup(mp, ira);
2178                         return;
2179                 }
2180 
2181                 if (ipha->ipha_protocol == IPPROTO_ESP)
2182                         mp = ipsecesp_icmp_error(mp, ira);
2183                 else
2184                         mp = ipsecah_icmp_error(mp, ira);
2185                 if (mp == NULL)
2186                         return;
2187 
2188                 /* Just in case ipsec didn't preserve the NULL b_cont */
2189                 if (mp->b_cont != NULL) {
2190                         if (!pullupmsg(mp, -1))
2191                                 goto discard_pkt;
2192                 }
2193 
2194                 /*
2195                  * Note that ira_pktlen and ira_ip_hdr_length are no longer
2196                  * correct, but we don't use them any more here.
2197                  *
2198                  * If succesful, the mp has been modified to not include
2199                  * the ESP/AH header so we can fanout to the ULP's icmp
2200                  * error handler.
2201                  */
2202                 if (mp->b_wptr - mp->b_rptr < IP_SIMPLE_HDR_LENGTH)
2203                         goto truncated;
2204 
2205                 /* Verify the modified message before any further processes. */
2206                 ipha = (ipha_t *)mp->b_rptr;
2207                 hdr_length = IPH_HDR_LENGTH(ipha);
2208                 icmph = (icmph_t *)&mp->b_rptr[hdr_length];
2209                 if (!icmp_inbound_verify_v4(mp, icmph, ira)) {
2210                         freemsg(mp);
2211                         return;
2212                 }
2213 
2214                 icmp_inbound_error_fanout_v4(mp, icmph, ira);
2215                 return;
2216 
2217         case IPPROTO_ENCAP: {
2218                 /* Look for self-encapsulated packets that caused an error */
2219                 ipha_t *in_ipha;
2220 
2221                 /*
2222                  * Caller has verified that length has to be
2223                  * at least the size of IP header.
2224                  */
2225                 ASSERT(hdr_length >= sizeof (ipha_t));
2226                 /*
2227                  * Check the sanity of the inner IP header like
2228                  * we did for the outer header.
2229                  */
2230                 in_ipha = (ipha_t *)((uchar_t *)ipha + hdr_length);
2231                 if ((IPH_HDR_VERSION(in_ipha) != IPV4_VERSION)) {
2232                         goto discard_pkt;
2233                 }
2234                 if (IPH_HDR_LENGTH(in_ipha) < sizeof (ipha_t)) {
2235                         goto discard_pkt;
2236                 }
2237                 /* Check for Self-encapsulated tunnels */
2238                 if (in_ipha->ipha_src == ipha->ipha_src &&
2239                     in_ipha->ipha_dst == ipha->ipha_dst) {
2240 
2241                         mp = icmp_inbound_self_encap_error_v4(mp, ipha,
2242                             in_ipha);
2243                         if (mp == NULL)
2244                                 goto discard_pkt;
2245 
2246                         /*
2247                          * Just in case self_encap didn't preserve the NULL
2248                          * b_cont
2249                          */
2250                         if (mp->b_cont != NULL) {
2251                                 if (!pullupmsg(mp, -1))
2252                                         goto discard_pkt;
2253                         }
2254                         /*
2255                          * Note that ira_pktlen and ira_ip_hdr_length are no
2256                          * longer correct, but we don't use them any more here.
2257                          */
2258                         if (mp->b_wptr - mp->b_rptr < IP_SIMPLE_HDR_LENGTH)
2259                                 goto truncated;
2260 
2261                         /*
2262                          * Verify the modified message before any further
2263                          * processes.
2264                          */
2265                         ipha = (ipha_t *)mp->b_rptr;
2266                         hdr_length = IPH_HDR_LENGTH(ipha);
2267                         icmph = (icmph_t *)&mp->b_rptr[hdr_length];
2268                         if (!icmp_inbound_verify_v4(mp, icmph, ira)) {
2269                                 freemsg(mp);
2270                                 return;
2271                         }
2272 
2273                         /*
2274                          * The packet in error is self-encapsualted.
2275                          * And we are finding it further encapsulated
2276                          * which we could not have possibly generated.
2277                          */
2278                         if (ipha->ipha_protocol == IPPROTO_ENCAP) {
2279                                 goto discard_pkt;
2280                         }
2281                         icmp_inbound_error_fanout_v4(mp, icmph, ira);
2282                         return;
2283                 }
2284                 /* No self-encapsulated */
2285                 /* FALLTHRU */
2286         }
2287         case IPPROTO_IPV6:
2288                 if ((connp = ipcl_iptun_classify_v4(&ripha.ipha_src,
2289                     &ripha.ipha_dst, ipst)) != NULL) {
2290                         ira->ira_flags |= IRAF_ICMP_ERROR;
2291                         connp->conn_recvicmp(connp, mp, NULL, ira);
2292                         CONN_DEC_REF(connp);
2293                         ira->ira_flags &= ~IRAF_ICMP_ERROR;
2294                         return;
2295                 }
2296                 /*
2297                  * No IP tunnel is interested, fallthrough and see
2298                  * if a raw socket will want it.
2299                  */
2300                 /* FALLTHRU */
2301         default:
2302                 ira->ira_flags |= IRAF_ICMP_ERROR;
2303                 ip_fanout_proto_v4(mp, &ripha, ira);
2304                 ira->ira_flags &= ~IRAF_ICMP_ERROR;
2305                 return;
2306         }
2307         /* NOTREACHED */
2308 discard_pkt:
2309         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
2310         ip1dbg(("icmp_inbound_error_fanout_v4: drop pkt\n"));
2311         ip_drop_input("ipIfStatsInDiscards", mp, ill);
2312         freemsg(mp);
2313         return;
2314 
2315 truncated:
2316         /* We pulled up everthing already. Must be truncated */
2317         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
2318         ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
2319         freemsg(mp);
2320 }
2321 
2322 /*
2323  * Common IP options parser.
2324  *
2325  * Setup routine: fill in *optp with options-parsing state, then
2326  * tail-call ipoptp_next to return the first option.
2327  */
2328 uint8_t
2329 ipoptp_first(ipoptp_t *optp, ipha_t *ipha)
2330 {
2331         uint32_t totallen; /* total length of all options */
2332 
2333         totallen = ipha->ipha_version_and_hdr_length -
2334             (uint8_t)((IP_VERSION << 4) + IP_SIMPLE_HDR_LENGTH_IN_WORDS);
2335         totallen <<= 2;
2336         optp->ipoptp_next = (uint8_t *)(&ipha[1]);
2337         optp->ipoptp_end = optp->ipoptp_next + totallen;
2338         optp->ipoptp_flags = 0;
2339         return (ipoptp_next(optp));
2340 }
2341 
2342 /* Like above but without an ipha_t */
2343 uint8_t
2344 ipoptp_first2(ipoptp_t *optp, uint32_t totallen, uint8_t *opt)
2345 {
2346         optp->ipoptp_next = opt;
2347         optp->ipoptp_end = optp->ipoptp_next + totallen;
2348         optp->ipoptp_flags = 0;
2349         return (ipoptp_next(optp));
2350 }
2351 
2352 /*
2353  * Common IP options parser: extract next option.
2354  */
2355 uint8_t
2356 ipoptp_next(ipoptp_t *optp)
2357 {
2358         uint8_t *end = optp->ipoptp_end;
2359         uint8_t *cur = optp->ipoptp_next;
2360         uint8_t opt, len, pointer;
2361 
2362         /*
2363          * If cur > end already, then the ipoptp_end or ipoptp_next pointer
2364          * has been corrupted.
2365          */
2366         ASSERT(cur <= end);
2367 
2368         if (cur == end)
2369                 return (IPOPT_EOL);
2370 
2371         opt = cur[IPOPT_OPTVAL];
2372 
2373         /*
2374          * Skip any NOP options.
2375          */
2376         while (opt == IPOPT_NOP) {
2377                 cur++;
2378                 if (cur == end)
2379                         return (IPOPT_EOL);
2380                 opt = cur[IPOPT_OPTVAL];
2381         }
2382 
2383         if (opt == IPOPT_EOL)
2384                 return (IPOPT_EOL);
2385 
2386         /*
2387          * Option requiring a length.
2388          */
2389         if ((cur + 1) >= end) {
2390                 optp->ipoptp_flags |= IPOPTP_ERROR;
2391                 return (IPOPT_EOL);
2392         }
2393         len = cur[IPOPT_OLEN];
2394         if (len < 2) {
2395                 optp->ipoptp_flags |= IPOPTP_ERROR;
2396                 return (IPOPT_EOL);
2397         }
2398         optp->ipoptp_cur = cur;
2399         optp->ipoptp_len = len;
2400         optp->ipoptp_next = cur + len;
2401         if (cur + len > end) {
2402                 optp->ipoptp_flags |= IPOPTP_ERROR;
2403                 return (IPOPT_EOL);
2404         }
2405 
2406         /*
2407          * For the options which require a pointer field, make sure
2408          * its there, and make sure it points to either something
2409          * inside this option, or the end of the option.
2410          */
2411         switch (opt) {
2412         case IPOPT_RR:
2413         case IPOPT_TS:
2414         case IPOPT_LSRR:
2415         case IPOPT_SSRR:
2416                 if (len <= IPOPT_OFFSET) {
2417                         optp->ipoptp_flags |= IPOPTP_ERROR;
2418                         return (opt);
2419                 }
2420                 pointer = cur[IPOPT_OFFSET];
2421                 if (pointer - 1 > len) {
2422                         optp->ipoptp_flags |= IPOPTP_ERROR;
2423                         return (opt);
2424                 }
2425                 break;
2426         }
2427 
2428         /*
2429          * Sanity check the pointer field based on the type of the
2430          * option.
2431          */
2432         switch (opt) {
2433         case IPOPT_RR:
2434         case IPOPT_SSRR:
2435         case IPOPT_LSRR:
2436                 if (pointer < IPOPT_MINOFF_SR)
2437                         optp->ipoptp_flags |= IPOPTP_ERROR;
2438                 break;
2439         case IPOPT_TS:
2440                 if (pointer < IPOPT_MINOFF_IT)
2441                         optp->ipoptp_flags |= IPOPTP_ERROR;
2442                 /*
2443                  * Note that the Internet Timestamp option also
2444                  * contains two four bit fields (the Overflow field,
2445                  * and the Flag field), which follow the pointer
2446                  * field.  We don't need to check that these fields
2447                  * fall within the length of the option because this
2448                  * was implicitely done above.  We've checked that the
2449                  * pointer value is at least IPOPT_MINOFF_IT, and that
2450                  * it falls within the option.  Since IPOPT_MINOFF_IT >
2451                  * IPOPT_POS_OV_FLG, we don't need the explicit check.
2452                  */
2453                 ASSERT(len > IPOPT_POS_OV_FLG);
2454                 break;
2455         }
2456 
2457         return (opt);
2458 }
2459 
2460 /*
2461  * Use the outgoing IP header to create an IP_OPTIONS option the way
2462  * it was passed down from the application.
2463  *
2464  * This is compatible with BSD in that it returns
2465  * the reverse source route with the final destination
2466  * as the last entry. The first 4 bytes of the option
2467  * will contain the final destination.
2468  */
2469 int
2470 ip_opt_get_user(conn_t *connp, uchar_t *buf)
2471 {
2472         ipoptp_t        opts;
2473         uchar_t         *opt;
2474         uint8_t         optval;
2475         uint8_t         optlen;
2476         uint32_t        len = 0;
2477         uchar_t         *buf1 = buf;
2478         uint32_t        totallen;
2479         ipaddr_t        dst;
2480         ip_pkt_t        *ipp = &connp->conn_xmit_ipp;
2481 
2482         if (!(ipp->ipp_fields & IPPF_IPV4_OPTIONS))
2483                 return (0);
2484 
2485         totallen = ipp->ipp_ipv4_options_len;
2486         if (totallen & 0x3)
2487                 return (0);
2488 
2489         buf += IP_ADDR_LEN;     /* Leave room for final destination */
2490         len += IP_ADDR_LEN;
2491         bzero(buf1, IP_ADDR_LEN);
2492 
2493         dst = connp->conn_faddr_v4;
2494 
2495         for (optval = ipoptp_first2(&opts, totallen, ipp->ipp_ipv4_options);
2496             optval != IPOPT_EOL;
2497             optval = ipoptp_next(&opts)) {
2498                 int     off;
2499 
2500                 opt = opts.ipoptp_cur;
2501                 if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
2502                         break;
2503                 }
2504                 optlen = opts.ipoptp_len;
2505 
2506                 switch (optval) {
2507                 case IPOPT_SSRR:
2508                 case IPOPT_LSRR:
2509 
2510                         /*
2511                          * Insert destination as the first entry in the source
2512                          * route and move down the entries on step.
2513                          * The last entry gets placed at buf1.
2514                          */
2515                         buf[IPOPT_OPTVAL] = optval;
2516                         buf[IPOPT_OLEN] = optlen;
2517                         buf[IPOPT_OFFSET] = optlen;
2518 
2519                         off = optlen - IP_ADDR_LEN;
2520                         if (off < 0) {
2521                                 /* No entries in source route */
2522                                 break;
2523                         }
2524                         /* Last entry in source route if not already set */
2525                         if (dst == INADDR_ANY)
2526                                 bcopy(opt + off, buf1, IP_ADDR_LEN);
2527                         off -= IP_ADDR_LEN;
2528 
2529                         while (off > 0) {
2530                                 bcopy(opt + off,
2531                                     buf + off + IP_ADDR_LEN,
2532                                     IP_ADDR_LEN);
2533                                 off -= IP_ADDR_LEN;
2534                         }
2535                         /* ipha_dst into first slot */
2536                         bcopy(&dst, buf + off + IP_ADDR_LEN,
2537                             IP_ADDR_LEN);
2538                         buf += optlen;
2539                         len += optlen;
2540                         break;
2541 
2542                 default:
2543                         bcopy(opt, buf, optlen);
2544                         buf += optlen;
2545                         len += optlen;
2546                         break;
2547                 }
2548         }
2549 done:
2550         /* Pad the resulting options */
2551         while (len & 0x3) {
2552                 *buf++ = IPOPT_EOL;
2553                 len++;
2554         }
2555         return (len);
2556 }
2557 
2558 /*
2559  * Update any record route or timestamp options to include this host.
2560  * Reverse any source route option.
2561  * This routine assumes that the options are well formed i.e. that they
2562  * have already been checked.
2563  */
2564 static void
2565 icmp_options_update(ipha_t *ipha)
2566 {
2567         ipoptp_t        opts;
2568         uchar_t         *opt;
2569         uint8_t         optval;
2570         ipaddr_t        src;            /* Our local address */
2571         ipaddr_t        dst;
2572 
2573         ip2dbg(("icmp_options_update\n"));
2574         src = ipha->ipha_src;
2575         dst = ipha->ipha_dst;
2576 
2577         for (optval = ipoptp_first(&opts, ipha);
2578             optval != IPOPT_EOL;
2579             optval = ipoptp_next(&opts)) {
2580                 ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
2581                 opt = opts.ipoptp_cur;
2582                 ip2dbg(("icmp_options_update: opt %d, len %d\n",
2583                     optval, opts.ipoptp_len));
2584                 switch (optval) {
2585                         int off1, off2;
2586                 case IPOPT_SSRR:
2587                 case IPOPT_LSRR:
2588                         /*
2589                          * Reverse the source route.  The first entry
2590                          * should be the next to last one in the current
2591                          * source route (the last entry is our address).
2592                          * The last entry should be the final destination.
2593                          */
2594                         off1 = IPOPT_MINOFF_SR - 1;
2595                         off2 = opt[IPOPT_OFFSET] - IP_ADDR_LEN - 1;
2596                         if (off2 < 0) {
2597                                 /* No entries in source route */
2598                                 ip1dbg((
2599                                     "icmp_options_update: bad src route\n"));
2600                                 break;
2601                         }
2602                         bcopy((char *)opt + off2, &dst, IP_ADDR_LEN);
2603                         bcopy(&ipha->ipha_dst, (char *)opt + off2, IP_ADDR_LEN);
2604                         bcopy(&dst, &ipha->ipha_dst, IP_ADDR_LEN);
2605                         off2 -= IP_ADDR_LEN;
2606 
2607                         while (off1 < off2) {
2608                                 bcopy((char *)opt + off1, &src, IP_ADDR_LEN);
2609                                 bcopy((char *)opt + off2, (char *)opt + off1,
2610                                     IP_ADDR_LEN);
2611                                 bcopy(&src, (char *)opt + off2, IP_ADDR_LEN);
2612                                 off1 += IP_ADDR_LEN;
2613                                 off2 -= IP_ADDR_LEN;
2614                         }
2615                         opt[IPOPT_OFFSET] = IPOPT_MINOFF_SR;
2616                         break;
2617                 }
2618         }
2619 }
2620 
2621 /*
2622  * Process received ICMP Redirect messages.
2623  * Assumes the caller has verified that the headers are in the pulled up mblk.
2624  * Consumes mp.
2625  */
2626 static void
2627 icmp_redirect_v4(mblk_t *mp, ipha_t *ipha, icmph_t *icmph, ip_recv_attr_t *ira)
2628 {
2629         ire_t           *ire, *nire;
2630         ire_t           *prev_ire;
2631         ipaddr_t        src, dst, gateway;
2632         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
2633         ipha_t          *inner_ipha;    /* Inner IP header */
2634 
2635         /* Caller already pulled up everything. */
2636         inner_ipha = (ipha_t *)&icmph[1];
2637         src = ipha->ipha_src;
2638         dst = inner_ipha->ipha_dst;
2639         gateway = icmph->icmph_rd_gateway;
2640         /* Make sure the new gateway is reachable somehow. */
2641         ire = ire_ftable_lookup_v4(gateway, 0, 0, IRE_ONLINK, NULL,
2642             ALL_ZONES, NULL, MATCH_IRE_TYPE, 0, ipst, NULL);
2643         /*
2644          * Make sure we had a route for the dest in question and that
2645          * that route was pointing to the old gateway (the source of the
2646          * redirect packet.)
2647          * We do longest match and then compare ire_gateway_addr below.
2648          */
2649         prev_ire = ire_ftable_lookup_v4(dst, 0, 0, 0, NULL, ALL_ZONES,
2650             NULL, MATCH_IRE_DSTONLY, 0, ipst, NULL);
2651         /*
2652          * Check that
2653          *      the redirect was not from ourselves
2654          *      the new gateway and the old gateway are directly reachable
2655          */
2656         if (prev_ire == NULL || ire == NULL ||
2657             (prev_ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK)) ||
2658             (prev_ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) ||
2659             !(ire->ire_type & IRE_IF_ALL) ||
2660             prev_ire->ire_gateway_addr != src) {
2661                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInBadRedirects);
2662                 ip_drop_input("icmpInBadRedirects - ire", mp, ira->ira_ill);
2663                 freemsg(mp);
2664                 if (ire != NULL)
2665                         ire_refrele(ire);
2666                 if (prev_ire != NULL)
2667                         ire_refrele(prev_ire);
2668                 return;
2669         }
2670 
2671         ire_refrele(prev_ire);
2672         ire_refrele(ire);
2673 
2674         /*
2675          * TODO: more precise handling for cases 0, 2, 3, the latter two
2676          * require TOS routing
2677          */
2678         switch (icmph->icmph_code) {
2679         case 0:
2680         case 1:
2681                 /* TODO: TOS specificity for cases 2 and 3 */
2682         case 2:
2683         case 3:
2684                 break;
2685         default:
2686                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInBadRedirects);
2687                 ip_drop_input("icmpInBadRedirects - code", mp, ira->ira_ill);
2688                 freemsg(mp);
2689                 return;
2690         }
2691         /*
2692          * Create a Route Association.  This will allow us to remember that
2693          * someone we believe told us to use the particular gateway.
2694          */
2695         ire = ire_create(
2696             (uchar_t *)&dst,                        /* dest addr */
2697             (uchar_t *)&ip_g_all_ones,              /* mask */
2698             (uchar_t *)&gateway,            /* gateway addr */
2699             IRE_HOST,
2700             NULL,                               /* ill */
2701             ALL_ZONES,
2702             (RTF_DYNAMIC | RTF_GATEWAY | RTF_HOST),
2703             NULL,                               /* tsol_gc_t */
2704             ipst);
2705 
2706         if (ire == NULL) {
2707                 freemsg(mp);
2708                 return;
2709         }
2710         nire = ire_add(ire);
2711         /* Check if it was a duplicate entry */
2712         if (nire != NULL && nire != ire) {
2713                 ASSERT(nire->ire_identical_ref > 1);
2714                 ire_delete(nire);
2715                 ire_refrele(nire);
2716                 nire = NULL;
2717         }
2718         ire = nire;
2719         if (ire != NULL) {
2720                 ire_refrele(ire);               /* Held in ire_add */
2721 
2722                 /* tell routing sockets that we received a redirect */
2723                 ip_rts_change(RTM_REDIRECT, dst, gateway, IP_HOST_MASK, 0, src,
2724                     (RTF_DYNAMIC | RTF_GATEWAY | RTF_HOST), 0,
2725                     (RTA_DST | RTA_GATEWAY | RTA_NETMASK | RTA_AUTHOR), ipst);
2726         }
2727 
2728         /*
2729          * Delete any existing IRE_HOST type redirect ires for this destination.
2730          * This together with the added IRE has the effect of
2731          * modifying an existing redirect.
2732          */
2733         prev_ire = ire_ftable_lookup_v4(dst, 0, src, IRE_HOST, NULL,
2734             ALL_ZONES, NULL, (MATCH_IRE_GW | MATCH_IRE_TYPE), 0, ipst, NULL);
2735         if (prev_ire != NULL) {
2736                 if (prev_ire ->ire_flags & RTF_DYNAMIC)
2737                         ire_delete(prev_ire);
2738                 ire_refrele(prev_ire);
2739         }
2740 
2741         freemsg(mp);
2742 }
2743 
2744 /*
2745  * Generate an ICMP parameter problem message.
2746  * When called from ip_output side a minimal ip_recv_attr_t needs to be
2747  * constructed by the caller.
2748  */
2749 static void
2750 icmp_param_problem(mblk_t *mp, uint8_t ptr, ip_recv_attr_t *ira)
2751 {
2752         icmph_t icmph;
2753         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
2754 
2755         mp = icmp_pkt_err_ok(mp, ira);
2756         if (mp == NULL)
2757                 return;
2758 
2759         bzero(&icmph, sizeof (icmph_t));
2760         icmph.icmph_type = ICMP_PARAM_PROBLEM;
2761         icmph.icmph_pp_ptr = ptr;
2762         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutParmProbs);
2763         icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
2764 }
2765 
2766 /*
2767  * Build and ship an IPv4 ICMP message using the packet data in mp, and
2768  * the ICMP header pointed to by "stuff".  (May be called as writer.)
2769  * Note: assumes that icmp_pkt_err_ok has been called to verify that
2770  * an icmp error packet can be sent.
2771  * Assigns an appropriate source address to the packet. If ipha_dst is
2772  * one of our addresses use it for source. Otherwise let ip_output_simple
2773  * pick the source address.
2774  */
2775 static void
2776 icmp_pkt(mblk_t *mp, void *stuff, size_t len, ip_recv_attr_t *ira)
2777 {
2778         ipaddr_t dst;
2779         icmph_t *icmph;
2780         ipha_t  *ipha;
2781         uint_t  len_needed;
2782         size_t  msg_len;
2783         mblk_t  *mp1;
2784         ipaddr_t src;
2785         ire_t   *ire;
2786         ip_xmit_attr_t ixas;
2787         ip_stack_t *ipst = ira->ira_ill->ill_ipst;
2788 
2789         ipha = (ipha_t *)mp->b_rptr;
2790 
2791         bzero(&ixas, sizeof (ixas));
2792         ixas.ixa_flags = IXAF_BASIC_SIMPLE_V4;
2793         ixas.ixa_zoneid = ira->ira_zoneid;
2794         ixas.ixa_ifindex = 0;
2795         ixas.ixa_ipst = ipst;
2796         ixas.ixa_cred = kcred;
2797         ixas.ixa_cpid = NOPID;
2798         ixas.ixa_tsl = ira->ira_tsl; /* Behave as a multi-level responder */
2799         ixas.ixa_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
2800 
2801         if (ira->ira_flags & IRAF_IPSEC_SECURE) {
2802                 /*
2803                  * Apply IPsec based on how IPsec was applied to
2804                  * the packet that had the error.
2805                  *
2806                  * If it was an outbound packet that caused the ICMP
2807                  * error, then the caller will have setup the IRA
2808                  * appropriately.
2809                  */
2810                 if (!ipsec_in_to_out(ira, &ixas, mp, ipha, NULL)) {
2811                         BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsOutDiscards);
2812                         /* Note: mp already consumed and ip_drop_packet done */
2813                         return;
2814                 }
2815         } else {
2816                 /*
2817                  * This is in clear. The icmp message we are building
2818                  * here should go out in clear, independent of our policy.
2819                  */
2820                 ixas.ixa_flags |= IXAF_NO_IPSEC;
2821         }
2822 
2823         /* Remember our eventual destination */
2824         dst = ipha->ipha_src;
2825 
2826         /*
2827          * If the packet was for one of our unicast addresses, make
2828          * sure we respond with that as the source. Otherwise
2829          * have ip_output_simple pick the source address.
2830          */
2831         ire = ire_ftable_lookup_v4(ipha->ipha_dst, 0, 0,
2832             (IRE_LOCAL|IRE_LOOPBACK), NULL, ira->ira_zoneid, NULL,
2833             MATCH_IRE_TYPE|MATCH_IRE_ZONEONLY, 0, ipst, NULL);
2834         if (ire != NULL) {
2835                 ire_refrele(ire);
2836                 src = ipha->ipha_dst;
2837         } else {
2838                 src = INADDR_ANY;
2839                 ixas.ixa_flags |= IXAF_SET_SOURCE;
2840         }
2841 
2842         /*
2843          * Check if we can send back more then 8 bytes in addition to
2844          * the IP header.  We try to send 64 bytes of data and the internal
2845          * header in the special cases of ipv4 encapsulated ipv4 or ipv6.
2846          */
2847         len_needed = IPH_HDR_LENGTH(ipha);
2848         if (ipha->ipha_protocol == IPPROTO_ENCAP ||
2849             ipha->ipha_protocol == IPPROTO_IPV6) {
2850                 if (!pullupmsg(mp, -1)) {
2851                         BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsOutDiscards);
2852                         ip_drop_output("ipIfStatsOutDiscards", mp, NULL);
2853                         freemsg(mp);
2854                         return;
2855                 }
2856                 ipha = (ipha_t *)mp->b_rptr;
2857 
2858                 if (ipha->ipha_protocol == IPPROTO_ENCAP) {
2859                         len_needed += IPH_HDR_LENGTH(((uchar_t *)ipha +
2860                             len_needed));
2861                 } else {
2862                         ip6_t *ip6h = (ip6_t *)((uchar_t *)ipha + len_needed);
2863 
2864                         ASSERT(ipha->ipha_protocol == IPPROTO_IPV6);
2865                         len_needed += ip_hdr_length_v6(mp, ip6h);
2866                 }
2867         }
2868         len_needed += ipst->ips_ip_icmp_return;
2869         msg_len = msgdsize(mp);
2870         if (msg_len > len_needed) {
2871                 (void) adjmsg(mp, len_needed - msg_len);
2872                 msg_len = len_needed;
2873         }
2874         mp1 = allocb(sizeof (icmp_ipha) + len, BPRI_MED);
2875         if (mp1 == NULL) {
2876                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutErrors);
2877                 freemsg(mp);
2878                 return;
2879         }
2880         mp1->b_cont = mp;
2881         mp = mp1;
2882 
2883         /*
2884          * Set IXAF_TRUSTED_ICMP so we can let the ICMP messages this
2885          * node generates be accepted in peace by all on-host destinations.
2886          * If we do NOT assume that all on-host destinations trust
2887          * self-generated ICMP messages, then rework here, ip6.c, and spd.c.
2888          * (Look for IXAF_TRUSTED_ICMP).
2889          */
2890         ixas.ixa_flags |= IXAF_TRUSTED_ICMP;
2891 
2892         ipha = (ipha_t *)mp->b_rptr;
2893         mp1->b_wptr = (uchar_t *)ipha + (sizeof (icmp_ipha) + len);
2894         *ipha = icmp_ipha;
2895         ipha->ipha_src = src;
2896         ipha->ipha_dst = dst;
2897         ipha->ipha_ttl = ipst->ips_ip_def_ttl;
2898         msg_len += sizeof (icmp_ipha) + len;
2899         if (msg_len > IP_MAXPACKET) {
2900                 (void) adjmsg(mp, IP_MAXPACKET - msg_len);
2901                 msg_len = IP_MAXPACKET;
2902         }
2903         ipha->ipha_length = htons((uint16_t)msg_len);
2904         icmph = (icmph_t *)&ipha[1];
2905         bcopy(stuff, icmph, len);
2906         icmph->icmph_checksum = 0;
2907         icmph->icmph_checksum = IP_CSUM(mp, (int32_t)sizeof (ipha_t), 0);
2908         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutMsgs);
2909 
2910         (void) ip_output_simple(mp, &ixas);
2911         ixa_cleanup(&ixas);
2912 }
2913 
2914 /*
2915  * Determine if an ICMP error packet can be sent given the rate limit.
2916  * The limit consists of an average frequency (icmp_pkt_err_interval measured
2917  * in milliseconds) and a burst size. Burst size number of packets can
2918  * be sent arbitrarely closely spaced.
2919  * The state is tracked using two variables to implement an approximate
2920  * token bucket filter:
2921  *      icmp_pkt_err_last - lbolt value when the last burst started
2922  *      icmp_pkt_err_sent - number of packets sent in current burst
2923  */
2924 boolean_t
2925 icmp_err_rate_limit(ip_stack_t *ipst)
2926 {
2927         clock_t now = TICK_TO_MSEC(ddi_get_lbolt());
2928         uint_t refilled; /* Number of packets refilled in tbf since last */
2929         /* Guard against changes by loading into local variable */
2930         uint_t err_interval = ipst->ips_ip_icmp_err_interval;
2931 
2932         if (err_interval == 0)
2933                 return (B_FALSE);
2934 
2935         if (ipst->ips_icmp_pkt_err_last > now) {
2936                 /* 100HZ lbolt in ms for 32bit arch wraps every 49.7 days */
2937                 ipst->ips_icmp_pkt_err_last = 0;
2938                 ipst->ips_icmp_pkt_err_sent = 0;
2939         }
2940         /*
2941          * If we are in a burst update the token bucket filter.
2942          * Update the "last" time to be close to "now" but make sure
2943          * we don't loose precision.
2944          */
2945         if (ipst->ips_icmp_pkt_err_sent != 0) {
2946                 refilled = (now - ipst->ips_icmp_pkt_err_last)/err_interval;
2947                 if (refilled > ipst->ips_icmp_pkt_err_sent) {
2948                         ipst->ips_icmp_pkt_err_sent = 0;
2949                 } else {
2950                         ipst->ips_icmp_pkt_err_sent -= refilled;
2951                         ipst->ips_icmp_pkt_err_last += refilled * err_interval;
2952                 }
2953         }
2954         if (ipst->ips_icmp_pkt_err_sent == 0) {
2955                 /* Start of new burst */
2956                 ipst->ips_icmp_pkt_err_last = now;
2957         }
2958         if (ipst->ips_icmp_pkt_err_sent < ipst->ips_ip_icmp_err_burst) {
2959                 ipst->ips_icmp_pkt_err_sent++;
2960                 ip1dbg(("icmp_err_rate_limit: %d sent in burst\n",
2961                     ipst->ips_icmp_pkt_err_sent));
2962                 return (B_FALSE);
2963         }
2964         ip1dbg(("icmp_err_rate_limit: dropped\n"));
2965         return (B_TRUE);
2966 }
2967 
2968 /*
2969  * Check if it is ok to send an IPv4 ICMP error packet in
2970  * response to the IPv4 packet in mp.
2971  * Free the message and return null if no
2972  * ICMP error packet should be sent.
2973  */
2974 static mblk_t *
2975 icmp_pkt_err_ok(mblk_t *mp, ip_recv_attr_t *ira)
2976 {
2977         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
2978         icmph_t *icmph;
2979         ipha_t  *ipha;
2980         uint_t  len_needed;
2981 
2982         if (!mp)
2983                 return (NULL);
2984         ipha = (ipha_t *)mp->b_rptr;
2985         if (ip_csum_hdr(ipha)) {
2986                 BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsInCksumErrs);
2987                 ip_drop_input("ipIfStatsInCksumErrs", mp, NULL);
2988                 freemsg(mp);
2989                 return (NULL);
2990         }
2991         if (ip_type_v4(ipha->ipha_dst, ipst) == IRE_BROADCAST ||
2992             ip_type_v4(ipha->ipha_src, ipst) == IRE_BROADCAST ||
2993             CLASSD(ipha->ipha_dst) ||
2994             CLASSD(ipha->ipha_src) ||
2995             (ntohs(ipha->ipha_fragment_offset_and_flags) & IPH_OFFSET)) {
2996                 /* Note: only errors to the fragment with offset 0 */
2997                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
2998                 freemsg(mp);
2999                 return (NULL);
3000         }
3001         if (ipha->ipha_protocol == IPPROTO_ICMP) {
3002                 /*
3003                  * Check the ICMP type.  RFC 1122 sez:  don't send ICMP
3004                  * errors in response to any ICMP errors.
3005                  */
3006                 len_needed = IPH_HDR_LENGTH(ipha) + ICMPH_SIZE;
3007                 if (mp->b_wptr - mp->b_rptr < len_needed) {
3008                         if (!pullupmsg(mp, len_needed)) {
3009                                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInErrors);
3010                                 freemsg(mp);
3011                                 return (NULL);
3012                         }
3013                         ipha = (ipha_t *)mp->b_rptr;
3014                 }
3015                 icmph = (icmph_t *)
3016                     (&((char *)ipha)[IPH_HDR_LENGTH(ipha)]);
3017                 switch (icmph->icmph_type) {
3018                 case ICMP_DEST_UNREACHABLE:
3019                 case ICMP_SOURCE_QUENCH:
3020                 case ICMP_TIME_EXCEEDED:
3021                 case ICMP_PARAM_PROBLEM:
3022                 case ICMP_REDIRECT:
3023                         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
3024                         freemsg(mp);
3025                         return (NULL);
3026                 default:
3027                         break;
3028                 }
3029         }
3030         /*
3031          * If this is a labeled system, then check to see if we're allowed to
3032          * send a response to this particular sender.  If not, then just drop.
3033          */
3034         if (is_system_labeled() && !tsol_can_reply_error(mp, ira)) {
3035                 ip2dbg(("icmp_pkt_err_ok: can't respond to packet\n"));
3036                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
3037                 freemsg(mp);
3038                 return (NULL);
3039         }
3040         if (icmp_err_rate_limit(ipst)) {
3041                 /*
3042                  * Only send ICMP error packets every so often.
3043                  * This should be done on a per port/source basis,
3044                  * but for now this will suffice.
3045                  */
3046                 freemsg(mp);
3047                 return (NULL);
3048         }
3049         return (mp);
3050 }
3051 
3052 /*
3053  * Called when a packet was sent out the same link that it arrived on.
3054  * Check if it is ok to send a redirect and then send it.
3055  */
3056 void
3057 ip_send_potential_redirect_v4(mblk_t *mp, ipha_t *ipha, ire_t *ire,
3058     ip_recv_attr_t *ira)
3059 {
3060         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
3061         ipaddr_t        src, nhop;
3062         mblk_t          *mp1;
3063         ire_t           *nhop_ire;
3064 
3065         /*
3066          * Check the source address to see if it originated
3067          * on the same logical subnet it is going back out on.
3068          * If so, we should be able to send it a redirect.
3069          * Avoid sending a redirect if the destination
3070          * is directly connected (i.e., we matched an IRE_ONLINK),
3071          * or if the packet was source routed out this interface.
3072          *
3073          * We avoid sending a redirect if the
3074          * destination is directly connected
3075          * because it is possible that multiple
3076          * IP subnets may have been configured on
3077          * the link, and the source may not
3078          * be on the same subnet as ip destination,
3079          * even though they are on the same
3080          * physical link.
3081          */
3082         if ((ire->ire_type & IRE_ONLINK) ||
3083             ip_source_routed(ipha, ipst))
3084                 return;
3085 
3086         nhop_ire = ire_nexthop(ire);
3087         if (nhop_ire == NULL)
3088                 return;
3089 
3090         nhop = nhop_ire->ire_addr;
3091 
3092         if (nhop_ire->ire_type & IRE_IF_CLONE) {
3093                 ire_t   *ire2;
3094 
3095                 /* Follow ire_dep_parent to find non-clone IRE_INTERFACE */
3096                 mutex_enter(&nhop_ire->ire_lock);
3097                 ire2 = nhop_ire->ire_dep_parent;
3098                 if (ire2 != NULL)
3099                         ire_refhold(ire2);
3100                 mutex_exit(&nhop_ire->ire_lock);
3101                 ire_refrele(nhop_ire);
3102                 nhop_ire = ire2;
3103         }
3104         if (nhop_ire == NULL)
3105                 return;
3106 
3107         ASSERT(!(nhop_ire->ire_type & IRE_IF_CLONE));
3108 
3109         src = ipha->ipha_src;
3110 
3111         /*
3112          * We look at the interface ire for the nexthop,
3113          * to see if ipha_src is in the same subnet
3114          * as the nexthop.
3115          */
3116         if ((src & nhop_ire->ire_mask) == (nhop & nhop_ire->ire_mask)) {
3117                 /*
3118                  * The source is directly connected.
3119                  */
3120                 mp1 = copymsg(mp);
3121                 if (mp1 != NULL) {
3122                         icmp_send_redirect(mp1, nhop, ira);
3123                 }
3124         }
3125         ire_refrele(nhop_ire);
3126 }
3127 
3128 /*
3129  * Generate an ICMP redirect message.
3130  */
3131 static void
3132 icmp_send_redirect(mblk_t *mp, ipaddr_t gateway, ip_recv_attr_t *ira)
3133 {
3134         icmph_t icmph;
3135         ip_stack_t *ipst = ira->ira_ill->ill_ipst;
3136 
3137         mp = icmp_pkt_err_ok(mp, ira);
3138         if (mp == NULL)
3139                 return;
3140 
3141         bzero(&icmph, sizeof (icmph_t));
3142         icmph.icmph_type = ICMP_REDIRECT;
3143         icmph.icmph_code = 1;
3144         icmph.icmph_rd_gateway = gateway;
3145         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutRedirects);
3146         icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
3147 }
3148 
3149 /*
3150  * Generate an ICMP time exceeded message.
3151  */
3152 void
3153 icmp_time_exceeded(mblk_t *mp, uint8_t code, ip_recv_attr_t *ira)
3154 {
3155         icmph_t icmph;
3156         ip_stack_t *ipst = ira->ira_ill->ill_ipst;
3157 
3158         mp = icmp_pkt_err_ok(mp, ira);
3159         if (mp == NULL)
3160                 return;
3161 
3162         bzero(&icmph, sizeof (icmph_t));
3163         icmph.icmph_type = ICMP_TIME_EXCEEDED;
3164         icmph.icmph_code = code;
3165         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutTimeExcds);
3166         icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
3167 }
3168 
3169 /*
3170  * Generate an ICMP unreachable message.
3171  * When called from ip_output side a minimal ip_recv_attr_t needs to be
3172  * constructed by the caller.
3173  */
3174 void
3175 icmp_unreachable(mblk_t *mp, uint8_t code, ip_recv_attr_t *ira)
3176 {
3177         icmph_t icmph;
3178         ip_stack_t *ipst = ira->ira_ill->ill_ipst;
3179 
3180         mp = icmp_pkt_err_ok(mp, ira);
3181         if (mp == NULL)
3182                 return;
3183 
3184         bzero(&icmph, sizeof (icmph_t));
3185         icmph.icmph_type = ICMP_DEST_UNREACHABLE;
3186         icmph.icmph_code = code;
3187         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDestUnreachs);
3188         icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
3189 }
3190 
3191 /*
3192  * Latch in the IPsec state for a stream based the policy in the listener
3193  * and the actions in the ip_recv_attr_t.
3194  * Called directly from TCP and SCTP.
3195  */
3196 boolean_t
3197 ip_ipsec_policy_inherit(conn_t *connp, conn_t *lconnp, ip_recv_attr_t *ira)
3198 {
3199         ASSERT(lconnp->conn_policy != NULL);
3200         ASSERT(connp->conn_policy == NULL);
3201 
3202         IPPH_REFHOLD(lconnp->conn_policy);
3203         connp->conn_policy = lconnp->conn_policy;
3204 
3205         if (ira->ira_ipsec_action != NULL) {
3206                 if (connp->conn_latch == NULL) {
3207                         connp->conn_latch = iplatch_create();
3208                         if (connp->conn_latch == NULL)
3209                                 return (B_FALSE);
3210                 }
3211                 ipsec_latch_inbound(connp, ira);
3212         }
3213         return (B_TRUE);
3214 }
3215 
3216 /*
3217  * Verify whether or not the IP address is a valid local address.
3218  * Could be a unicast, including one for a down interface.
3219  * If allow_mcbc then a multicast or broadcast address is also
3220  * acceptable.
3221  *
3222  * In the case of a broadcast/multicast address, however, the
3223  * upper protocol is expected to reset the src address
3224  * to zero when we return IPVL_MCAST/IPVL_BCAST so that
3225  * no packets are emitted with broadcast/multicast address as
3226  * source address (that violates hosts requirements RFC 1122)
3227  * The addresses valid for bind are:
3228  *      (1) - INADDR_ANY (0)
3229  *      (2) - IP address of an UP interface
3230  *      (3) - IP address of a DOWN interface
3231  *      (4) - valid local IP broadcast addresses. In this case
3232  *      the conn will only receive packets destined to
3233  *      the specified broadcast address.
3234  *      (5) - a multicast address. In this case
3235  *      the conn will only receive packets destined to
3236  *      the specified multicast address. Note: the
3237  *      application still has to issue an
3238  *      IP_ADD_MEMBERSHIP socket option.
3239  *
3240  * In all the above cases, the bound address must be valid in the current zone.
3241  * When the address is loopback, multicast or broadcast, there might be many
3242  * matching IREs so bind has to look up based on the zone.
3243  */
3244 ip_laddr_t
3245 ip_laddr_verify_v4(ipaddr_t src_addr, zoneid_t zoneid,
3246     ip_stack_t *ipst, boolean_t allow_mcbc)
3247 {
3248         ire_t *src_ire;
3249 
3250         ASSERT(src_addr != INADDR_ANY);
3251 
3252         src_ire = ire_ftable_lookup_v4(src_addr, 0, 0, 0,
3253             NULL, zoneid, NULL, MATCH_IRE_ZONEONLY, 0, ipst, NULL);
3254 
3255         /*
3256          * If an address other than in6addr_any is requested,
3257          * we verify that it is a valid address for bind
3258          * Note: Following code is in if-else-if form for
3259          * readability compared to a condition check.
3260          */
3261         if (src_ire != NULL && (src_ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK))) {
3262                 /*
3263                  * (2) Bind to address of local UP interface
3264                  */
3265                 ire_refrele(src_ire);
3266                 return (IPVL_UNICAST_UP);
3267         } else if (src_ire != NULL && src_ire->ire_type & IRE_BROADCAST) {
3268                 /*
3269                  * (4) Bind to broadcast address
3270                  */
3271                 ire_refrele(src_ire);
3272                 if (allow_mcbc)
3273                         return (IPVL_BCAST);
3274                 else
3275                         return (IPVL_BAD);
3276         } else if (CLASSD(src_addr)) {
3277                 /* (5) bind to multicast address. */
3278                 if (src_ire != NULL)
3279                         ire_refrele(src_ire);
3280 
3281                 if (allow_mcbc)
3282                         return (IPVL_MCAST);
3283                 else
3284                         return (IPVL_BAD);
3285         } else {
3286                 ipif_t *ipif;
3287 
3288                 /*
3289                  * (3) Bind to address of local DOWN interface?
3290                  * (ipif_lookup_addr() looks up all interfaces
3291                  * but we do not get here for UP interfaces
3292                  * - case (2) above)
3293                  */
3294                 if (src_ire != NULL)
3295                         ire_refrele(src_ire);
3296 
3297                 ipif = ipif_lookup_addr(src_addr, NULL, zoneid, ipst);
3298                 if (ipif == NULL)
3299                         return (IPVL_BAD);
3300 
3301                 /* Not a useful source? */
3302                 if (ipif->ipif_flags & (IPIF_NOLOCAL | IPIF_ANYCAST)) {
3303                         ipif_refrele(ipif);
3304                         return (IPVL_BAD);
3305                 }
3306                 ipif_refrele(ipif);
3307                 return (IPVL_UNICAST_DOWN);
3308         }
3309 }
3310 
3311 /*
3312  * Insert in the bind fanout for IPv4 and IPv6.
3313  * The caller should already have used ip_laddr_verify_v*() before calling
3314  * this.
3315  */
3316 int
3317 ip_laddr_fanout_insert(conn_t *connp)
3318 {
3319         int             error;
3320 
3321         /*
3322          * Allow setting new policies. For example, disconnects result
3323          * in us being called. As we would have set conn_policy_cached
3324          * to B_TRUE before, we should set it to B_FALSE, so that policy
3325          * can change after the disconnect.
3326          */
3327         connp->conn_policy_cached = B_FALSE;
3328 
3329         error = ipcl_bind_insert(connp);
3330         if (error != 0) {
3331                 if (connp->conn_anon_port) {
3332                         (void) tsol_mlp_anon(crgetzone(connp->conn_cred),
3333                             connp->conn_mlp_type, connp->conn_proto,
3334                             ntohs(connp->conn_lport), B_FALSE);
3335                 }
3336                 connp->conn_mlp_type = mlptSingle;
3337         }
3338         return (error);
3339 }
3340 
3341 /*
3342  * Verify that both the source and destination addresses are valid. If
3343  * IPDF_VERIFY_DST is not set, then the destination address may be unreachable,
3344  * i.e. have no route to it.  Protocols like TCP want to verify destination
3345  * reachability, while tunnels do not.
3346  *
3347  * Determine the route, the interface, and (optionally) the source address
3348  * to use to reach a given destination.
3349  * Note that we allow connect to broadcast and multicast addresses when
3350  * IPDF_ALLOW_MCBC is set.
3351  * first_hop and dst_addr are normally the same, but if source routing
3352  * they will differ; in that case the first_hop is what we'll use for the
3353  * routing lookup but the dce and label checks will be done on dst_addr,
3354  *
3355  * If uinfo is set, then we fill in the best available information
3356  * we have for the destination. This is based on (in priority order) any
3357  * metrics and path MTU stored in a dce_t, route metrics, and finally the
3358  * ill_mtu/ill_mc_mtu.
3359  *
3360  * Tsol note: If we have a source route then dst_addr != firsthop. But we
3361  * always do the label check on dst_addr.
3362  */
3363 int
3364 ip_set_destination_v4(ipaddr_t *src_addrp, ipaddr_t dst_addr, ipaddr_t firsthop,
3365     ip_xmit_attr_t *ixa, iulp_t *uinfo, uint32_t flags, uint_t mac_mode)
3366 {
3367         ire_t           *ire = NULL;
3368         int             error = 0;
3369         ipaddr_t        setsrc;                         /* RTF_SETSRC */
3370         zoneid_t        zoneid = ixa->ixa_zoneid;    /* Honors SO_ALLZONES */
3371         ip_stack_t      *ipst = ixa->ixa_ipst;
3372         dce_t           *dce;
3373         uint_t          pmtu;
3374         uint_t          generation;
3375         nce_t           *nce;
3376         ill_t           *ill = NULL;
3377         boolean_t       multirt = B_FALSE;
3378 
3379         ASSERT(ixa->ixa_flags & IXAF_IS_IPV4);
3380 
3381         /*
3382          * We never send to zero; the ULPs map it to the loopback address.
3383          * We can't allow it since we use zero to mean unitialized in some
3384          * places.
3385          */
3386         ASSERT(dst_addr != INADDR_ANY);
3387 
3388         if (is_system_labeled()) {
3389                 ts_label_t *tsl = NULL;
3390 
3391                 error = tsol_check_dest(ixa->ixa_tsl, &dst_addr, IPV4_VERSION,
3392                     mac_mode, (flags & IPDF_ZONE_IS_GLOBAL) != 0, &tsl);
3393                 if (error != 0)
3394                         return (error);
3395                 if (tsl != NULL) {
3396                         /* Update the label */
3397                         ip_xmit_attr_replace_tsl(ixa, tsl);
3398                 }
3399         }
3400 
3401         setsrc = INADDR_ANY;
3402         /*
3403          * Select a route; For IPMP interfaces, we would only select
3404          * a "hidden" route (i.e., going through a specific under_ill)
3405          * if ixa_ifindex has been specified.
3406          */
3407         ire = ip_select_route_v4(firsthop, *src_addrp, ixa,
3408             &generation, &setsrc, &error, &multirt);
3409         ASSERT(ire != NULL);    /* IRE_NOROUTE if none found */
3410         if (error != 0)
3411                 goto bad_addr;
3412 
3413         /*
3414          * ire can't be a broadcast or multicast unless IPDF_ALLOW_MCBC is set.
3415          * If IPDF_VERIFY_DST is set, the destination must be reachable;
3416          * Otherwise the destination needn't be reachable.
3417          *
3418          * If we match on a reject or black hole, then we've got a
3419          * local failure.  May as well fail out the connect() attempt,
3420          * since it's never going to succeed.
3421          */
3422         if (ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
3423                 /*
3424                  * If we're verifying destination reachability, we always want
3425                  * to complain here.
3426                  *
3427                  * If we're not verifying destination reachability but the
3428                  * destination has a route, we still want to fail on the
3429                  * temporary address and broadcast address tests.
3430                  *
3431                  * In both cases do we let the code continue so some reasonable
3432                  * information is returned to the caller. That enables the
3433                  * caller to use (and even cache) the IRE. conn_ip_ouput will
3434                  * use the generation mismatch path to check for the unreachable
3435                  * case thereby avoiding any specific check in the main path.
3436                  */
3437                 ASSERT(generation == IRE_GENERATION_VERIFY);
3438                 if (flags & IPDF_VERIFY_DST) {
3439                         /*
3440                          * Set errno but continue to set up ixa_ire to be
3441                          * the RTF_REJECT|RTF_BLACKHOLE IRE.
3442                          * That allows callers to use ip_output to get an
3443                          * ICMP error back.
3444                          */
3445                         if (!(ire->ire_type & IRE_HOST))
3446                                 error = ENETUNREACH;
3447                         else
3448                                 error = EHOSTUNREACH;
3449                 }
3450         }
3451 
3452         if ((ire->ire_type & (IRE_BROADCAST|IRE_MULTICAST)) &&
3453             !(flags & IPDF_ALLOW_MCBC)) {
3454                 ire_refrele(ire);
3455                 ire = ire_reject(ipst, B_FALSE);
3456                 generation = IRE_GENERATION_VERIFY;
3457                 error = ENETUNREACH;
3458         }
3459 
3460         /* Cache things */
3461         if (ixa->ixa_ire != NULL)
3462                 ire_refrele_notr(ixa->ixa_ire);
3463 #ifdef DEBUG
3464         ire_refhold_notr(ire);
3465         ire_refrele(ire);
3466 #endif
3467         ixa->ixa_ire = ire;
3468         ixa->ixa_ire_generation = generation;
3469 
3470         /*
3471          * Ensure that ixa_dce is always set any time that ixa_ire is set,
3472          * since some callers will send a packet to conn_ip_output() even if
3473          * there's an error.
3474          */
3475         if (flags & IPDF_UNIQUE_DCE) {
3476                 /* Fallback to the default dce if allocation fails */
3477                 dce = dce_lookup_and_add_v4(dst_addr, ipst);
3478                 if (dce != NULL)
3479                         generation = dce->dce_generation;
3480                 else
3481                         dce = dce_lookup_v4(dst_addr, ipst, &generation);
3482         } else {
3483                 dce = dce_lookup_v4(dst_addr, ipst, &generation);
3484         }
3485         ASSERT(dce != NULL);
3486         if (ixa->ixa_dce != NULL)
3487                 dce_refrele_notr(ixa->ixa_dce);
3488 #ifdef DEBUG
3489         dce_refhold_notr(dce);
3490         dce_refrele(dce);
3491 #endif
3492         ixa->ixa_dce = dce;
3493         ixa->ixa_dce_generation = generation;
3494 
3495         /*
3496          * For multicast with multirt we have a flag passed back from
3497          * ire_lookup_multi_ill_v4 since we don't have an IRE for each
3498          * possible multicast address.
3499          * We also need a flag for multicast since we can't check
3500          * whether RTF_MULTIRT is set in ixa_ire for multicast.
3501          */
3502         if (multirt) {
3503                 ixa->ixa_postfragfn = ip_postfrag_multirt_v4;
3504                 ixa->ixa_flags |= IXAF_MULTIRT_MULTICAST;
3505         } else {
3506                 ixa->ixa_postfragfn = ire->ire_postfragfn;
3507                 ixa->ixa_flags &= ~IXAF_MULTIRT_MULTICAST;
3508         }
3509         if (!(ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE))) {
3510                 /* Get an nce to cache. */
3511                 nce = ire_to_nce(ire, firsthop, NULL);
3512                 if (nce == NULL) {
3513                         /* Allocation failure? */
3514                         ixa->ixa_ire_generation = IRE_GENERATION_VERIFY;
3515                 } else {
3516                         if (ixa->ixa_nce != NULL)
3517                                 nce_refrele(ixa->ixa_nce);
3518                         ixa->ixa_nce = nce;
3519                 }
3520         }
3521 
3522         /*
3523          * If the source address is a loopback address, the
3524          * destination had best be local or multicast.
3525          * If we are sending to an IRE_LOCAL using a loopback source then
3526          * it had better be the same zoneid.
3527          */
3528         if (*src_addrp == htonl(INADDR_LOOPBACK)) {
3529                 if ((ire->ire_type & IRE_LOCAL) && ire->ire_zoneid != zoneid) {
3530                         ire = NULL;     /* Stored in ixa_ire */
3531                         error = EADDRNOTAVAIL;
3532                         goto bad_addr;
3533                 }
3534                 if (!(ire->ire_type & (IRE_LOOPBACK|IRE_LOCAL|IRE_MULTICAST))) {
3535                         ire = NULL;     /* Stored in ixa_ire */
3536                         error = EADDRNOTAVAIL;
3537                         goto bad_addr;
3538                 }
3539         }
3540         if (ire->ire_type & IRE_BROADCAST) {
3541                 /*
3542                  * If the ULP didn't have a specified source, then we
3543                  * make sure we reselect the source when sending
3544                  * broadcasts out different interfaces.
3545                  */
3546                 if (flags & IPDF_SELECT_SRC)
3547                         ixa->ixa_flags |= IXAF_SET_SOURCE;
3548                 else
3549                         ixa->ixa_flags &= ~IXAF_SET_SOURCE;
3550         }
3551 
3552         /*
3553          * Does the caller want us to pick a source address?
3554          */
3555         if (flags & IPDF_SELECT_SRC) {
3556                 ipaddr_t        src_addr;
3557 
3558                 /*
3559                  * We use use ire_nexthop_ill to avoid the under ipmp
3560                  * interface for source address selection. Note that for ipmp
3561                  * probe packets, ixa_ifindex would have been specified, and
3562                  * the ip_select_route() invocation would have picked an ire
3563                  * will ire_ill pointing at an under interface.
3564                  */
3565                 ill = ire_nexthop_ill(ire);
3566 
3567                 /* If unreachable we have no ill but need some source */
3568                 if (ill == NULL) {
3569                         src_addr = htonl(INADDR_LOOPBACK);
3570                         /* Make sure we look for a better source address */
3571                         generation = SRC_GENERATION_VERIFY;
3572                 } else {
3573                         error = ip_select_source_v4(ill, setsrc, dst_addr,
3574                             ixa->ixa_multicast_ifaddr, zoneid,
3575                             ipst, &src_addr, &generation, NULL);
3576                         if (error != 0) {
3577                                 ire = NULL;     /* Stored in ixa_ire */
3578                                 goto bad_addr;
3579                         }
3580                 }
3581 
3582                 /*
3583                  * We allow the source address to to down.
3584                  * However, we check that we don't use the loopback address
3585                  * as a source when sending out on the wire.
3586                  */
3587                 if ((src_addr == htonl(INADDR_LOOPBACK)) &&
3588                     !(ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK|IRE_MULTICAST)) &&
3589                     !(ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE))) {
3590                         ire = NULL;     /* Stored in ixa_ire */
3591                         error = EADDRNOTAVAIL;
3592                         goto bad_addr;
3593                 }
3594 
3595                 *src_addrp = src_addr;
3596                 ixa->ixa_src_generation = generation;
3597         }
3598 
3599         /*
3600          * Make sure we don't leave an unreachable ixa_nce in place
3601          * since ip_select_route is used when we unplumb i.e., remove
3602          * references on ixa_ire, ixa_nce, and ixa_dce.
3603          */
3604         nce = ixa->ixa_nce;
3605         if (nce != NULL && nce->nce_is_condemned) {
3606                 nce_refrele(nce);
3607                 ixa->ixa_nce = NULL;
3608                 ixa->ixa_ire_generation = IRE_GENERATION_VERIFY;
3609         }
3610 
3611         /*
3612          * The caller has set IXAF_PMTU_DISCOVERY if path MTU is desired.
3613          * However, we can't do it for IPv4 multicast or broadcast.
3614          */
3615         if (ire->ire_type & (IRE_BROADCAST|IRE_MULTICAST))
3616                 ixa->ixa_flags &= ~IXAF_PMTU_DISCOVERY;
3617 
3618         /*
3619          * Set initial value for fragmentation limit. Either conn_ip_output
3620          * or ULP might updates it when there are routing changes.
3621          * Handles a NULL ixa_ire->ire_ill or a NULL ixa_nce for RTF_REJECT.
3622          */
3623         pmtu = ip_get_pmtu(ixa);
3624         ixa->ixa_fragsize = pmtu;
3625         /* Make sure ixa_fragsize and ixa_pmtu remain identical */
3626         if (ixa->ixa_flags & IXAF_VERIFY_PMTU)
3627                 ixa->ixa_pmtu = pmtu;
3628 
3629         /*
3630          * Extract information useful for some transports.
3631          * First we look for DCE metrics. Then we take what we have in
3632          * the metrics in the route, where the offlink is used if we have
3633          * one.
3634          */
3635         if (uinfo != NULL) {
3636                 bzero(uinfo, sizeof (*uinfo));
3637 
3638                 if (dce->dce_flags & DCEF_UINFO)
3639                         *uinfo = dce->dce_uinfo;
3640 
3641                 rts_merge_metrics(uinfo, &ire->ire_metrics);
3642 
3643                 /* Allow ire_metrics to decrease the path MTU from above */
3644                 if (uinfo->iulp_mtu == 0 || uinfo->iulp_mtu > pmtu)
3645                         uinfo->iulp_mtu = pmtu;
3646 
3647                 uinfo->iulp_localnet = (ire->ire_type & IRE_ONLINK) != 0;
3648                 uinfo->iulp_loopback = (ire->ire_type & IRE_LOOPBACK) != 0;
3649                 uinfo->iulp_local = (ire->ire_type & IRE_LOCAL) != 0;
3650         }
3651 
3652         if (ill != NULL)
3653                 ill_refrele(ill);
3654 
3655         return (error);
3656 
3657 bad_addr:
3658         if (ire != NULL)
3659                 ire_refrele(ire);
3660 
3661         if (ill != NULL)
3662                 ill_refrele(ill);
3663 
3664         /*
3665          * Make sure we don't leave an unreachable ixa_nce in place
3666          * since ip_select_route is used when we unplumb i.e., remove
3667          * references on ixa_ire, ixa_nce, and ixa_dce.
3668          */
3669         nce = ixa->ixa_nce;
3670         if (nce != NULL && nce->nce_is_condemned) {
3671                 nce_refrele(nce);
3672                 ixa->ixa_nce = NULL;
3673                 ixa->ixa_ire_generation = IRE_GENERATION_VERIFY;
3674         }
3675 
3676         return (error);
3677 }
3678 
3679 
3680 /*
3681  * Get the base MTU for the case when path MTU discovery is not used.
3682  * Takes the MTU of the IRE into account.
3683  */
3684 uint_t
3685 ip_get_base_mtu(ill_t *ill, ire_t *ire)
3686 {
3687         uint_t mtu;
3688         uint_t iremtu = ire->ire_metrics.iulp_mtu;
3689 
3690         if (ire->ire_type & (IRE_MULTICAST|IRE_BROADCAST))
3691                 mtu = ill->ill_mc_mtu;
3692         else
3693                 mtu = ill->ill_mtu;
3694 
3695         if (iremtu != 0 && iremtu < mtu)
3696                 mtu = iremtu;
3697 
3698         return (mtu);
3699 }
3700 
3701 /*
3702  * Get the PMTU for the attributes. Handles both IPv4 and IPv6.
3703  * Assumes that ixa_ire, dce, and nce have already been set up.
3704  *
3705  * The caller has set IXAF_PMTU_DISCOVERY if path MTU discovery is desired.
3706  * We avoid path MTU discovery if it is disabled with ndd.
3707  * Furtermore, if the path MTU is too small, then we don't set DF for IPv4.
3708  *
3709  * NOTE: We also used to turn it off for source routed packets. That
3710  * is no longer required since the dce is per final destination.
3711  */
3712 uint_t
3713 ip_get_pmtu(ip_xmit_attr_t *ixa)
3714 {
3715         ip_stack_t      *ipst = ixa->ixa_ipst;
3716         dce_t           *dce;
3717         nce_t           *nce;
3718         ire_t           *ire;
3719         uint_t          pmtu;
3720 
3721         ire = ixa->ixa_ire;
3722         dce = ixa->ixa_dce;
3723         nce = ixa->ixa_nce;
3724 
3725         /*
3726          * If path MTU discovery has been turned off by ndd, then we ignore
3727          * any dce_pmtu and for IPv4 we will not set DF.
3728          */
3729         if (!ipst->ips_ip_path_mtu_discovery)
3730                 ixa->ixa_flags &= ~IXAF_PMTU_DISCOVERY;
3731 
3732         pmtu = IP_MAXPACKET;
3733         /*
3734          * Decide whether whether IPv4 sets DF
3735          * For IPv6 "no DF" means to use the 1280 mtu
3736          */
3737         if (ixa->ixa_flags & IXAF_PMTU_DISCOVERY) {
3738                 ixa->ixa_flags |= IXAF_PMTU_IPV4_DF;
3739         } else {
3740                 ixa->ixa_flags &= ~IXAF_PMTU_IPV4_DF;
3741                 if (!(ixa->ixa_flags & IXAF_IS_IPV4))
3742                         pmtu = IPV6_MIN_MTU;
3743         }
3744 
3745         /* Check if the PMTU is to old before we use it */
3746         if ((dce->dce_flags & DCEF_PMTU) &&
3747             TICK_TO_SEC(ddi_get_lbolt64()) - dce->dce_last_change_time >
3748             ipst->ips_ip_pathmtu_interval) {
3749                 /*
3750                  * Older than 20 minutes. Drop the path MTU information.
3751                  */
3752                 mutex_enter(&dce->dce_lock);
3753                 dce->dce_flags &= ~(DCEF_PMTU|DCEF_TOO_SMALL_PMTU);
3754                 dce->dce_last_change_time = TICK_TO_SEC(ddi_get_lbolt64());
3755                 mutex_exit(&dce->dce_lock);
3756                 dce_increment_generation(dce);
3757         }
3758 
3759         /* The metrics on the route can lower the path MTU */
3760         if (ire->ire_metrics.iulp_mtu != 0 &&
3761             ire->ire_metrics.iulp_mtu < pmtu)
3762                 pmtu = ire->ire_metrics.iulp_mtu;
3763 
3764         /*
3765          * If the path MTU is smaller than some minimum, we still use dce_pmtu
3766          * above (would be 576 for IPv4 and 1280 for IPv6), but we clear
3767          * IXAF_PMTU_IPV4_DF so that we avoid setting DF for IPv4.
3768          */
3769         if (ixa->ixa_flags & IXAF_PMTU_DISCOVERY) {
3770                 if (dce->dce_flags & DCEF_PMTU) {
3771                         if (dce->dce_pmtu < pmtu)
3772                                 pmtu = dce->dce_pmtu;
3773 
3774                         if (dce->dce_flags & DCEF_TOO_SMALL_PMTU) {
3775                                 ixa->ixa_flags |= IXAF_PMTU_TOO_SMALL;
3776                                 ixa->ixa_flags &= ~IXAF_PMTU_IPV4_DF;
3777                         } else {
3778                                 ixa->ixa_flags &= ~IXAF_PMTU_TOO_SMALL;
3779                                 ixa->ixa_flags |= IXAF_PMTU_IPV4_DF;
3780                         }
3781                 } else {
3782                         ixa->ixa_flags &= ~IXAF_PMTU_TOO_SMALL;
3783                         ixa->ixa_flags |= IXAF_PMTU_IPV4_DF;
3784                 }
3785         }
3786 
3787         /*
3788          * If we have an IRE_LOCAL we use the loopback mtu instead of
3789          * the ill for going out the wire i.e., IRE_LOCAL gets the same
3790          * mtu as IRE_LOOPBACK.
3791          */
3792         if (ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK)) {
3793                 uint_t loopback_mtu;
3794 
3795                 loopback_mtu = (ire->ire_ipversion == IPV6_VERSION) ?
3796                     ip_loopback_mtu_v6plus : ip_loopback_mtuplus;
3797 
3798                 if (loopback_mtu < pmtu)
3799                         pmtu = loopback_mtu;
3800         } else if (nce != NULL) {
3801                 /*
3802                  * Make sure we don't exceed the interface MTU.
3803                  * In the case of RTF_REJECT or RTF_BLACKHOLE we might not have
3804                  * an ill. We'd use the above IP_MAXPACKET in that case just
3805                  * to tell the transport something larger than zero.
3806                  */
3807                 if (ire->ire_type & (IRE_MULTICAST|IRE_BROADCAST)) {
3808                         if (nce->nce_common->ncec_ill->ill_mc_mtu < pmtu)
3809                                 pmtu = nce->nce_common->ncec_ill->ill_mc_mtu;
3810                         if (nce->nce_common->ncec_ill != nce->nce_ill &&
3811                             nce->nce_ill->ill_mc_mtu < pmtu) {
3812                                 /*
3813                                  * for interfaces in an IPMP group, the mtu of
3814                                  * the nce_ill (under_ill) could be different
3815                                  * from the mtu of the ncec_ill, so we take the
3816                                  * min of the two.
3817                                  */
3818                                 pmtu = nce->nce_ill->ill_mc_mtu;
3819                         }
3820                 } else {
3821                         if (nce->nce_common->ncec_ill->ill_mtu < pmtu)
3822                                 pmtu = nce->nce_common->ncec_ill->ill_mtu;
3823                         if (nce->nce_common->ncec_ill != nce->nce_ill &&
3824                             nce->nce_ill->ill_mtu < pmtu) {
3825                                 /*
3826                                  * for interfaces in an IPMP group, the mtu of
3827                                  * the nce_ill (under_ill) could be different
3828                                  * from the mtu of the ncec_ill, so we take the
3829                                  * min of the two.
3830                                  */
3831                                 pmtu = nce->nce_ill->ill_mtu;
3832                         }
3833                 }
3834         }
3835 
3836         /*
3837          * Handle the IPV6_USE_MIN_MTU socket option or ancillary data.
3838          * Only applies to IPv6.
3839          */
3840         if (!(ixa->ixa_flags & IXAF_IS_IPV4)) {
3841                 if (ixa->ixa_flags & IXAF_USE_MIN_MTU) {
3842                         switch (ixa->ixa_use_min_mtu) {
3843                         case IPV6_USE_MIN_MTU_MULTICAST:
3844                                 if (ire->ire_type & IRE_MULTICAST)
3845                                         pmtu = IPV6_MIN_MTU;
3846                                 break;
3847                         case IPV6_USE_MIN_MTU_ALWAYS:
3848                                 pmtu = IPV6_MIN_MTU;
3849                                 break;
3850                         case IPV6_USE_MIN_MTU_NEVER:
3851                                 break;
3852                         }
3853                 } else {
3854                         /* Default is IPV6_USE_MIN_MTU_MULTICAST */
3855                         if (ire->ire_type & IRE_MULTICAST)
3856                                 pmtu = IPV6_MIN_MTU;
3857                 }
3858         }
3859 
3860         /*
3861          * After receiving an ICMPv6 "packet too big" message with a
3862          * MTU < 1280, and for multirouted IPv6 packets, the IP layer
3863          * will insert a 8-byte fragment header in every packet. We compensate
3864          * for those cases by returning a smaller path MTU to the ULP.
3865          *
3866          * In the case of CGTP then ip_output will add a fragment header.
3867          * Make sure there is room for it by telling a smaller number
3868          * to the transport.
3869          *
3870          * When IXAF_IPV6_ADDR_FRAGHDR we subtract the frag hdr here
3871          * so the ULPs consistently see a iulp_pmtu and ip_get_pmtu()
3872          * which is the size of the packets it can send.
3873          */
3874         if (!(ixa->ixa_flags & IXAF_IS_IPV4)) {
3875                 if ((dce->dce_flags & DCEF_TOO_SMALL_PMTU) ||
3876                     (ire->ire_flags & RTF_MULTIRT) ||
3877                     (ixa->ixa_flags & IXAF_MULTIRT_MULTICAST)) {
3878                         pmtu -= sizeof (ip6_frag_t);
3879                         ixa->ixa_flags |= IXAF_IPV6_ADD_FRAGHDR;
3880                 }
3881         }
3882 
3883         return (pmtu);
3884 }
3885 
3886 /*
3887  * Carve "len" bytes out of an mblk chain, consuming any we empty, and duping
3888  * the final piece where we don't.  Return a pointer to the first mblk in the
3889  * result, and update the pointer to the next mblk to chew on.  If anything
3890  * goes wrong (i.e., dupb fails), we waste everything in sight and return a
3891  * NULL pointer.
3892  */
3893 mblk_t *
3894 ip_carve_mp(mblk_t **mpp, ssize_t len)
3895 {
3896         mblk_t  *mp0;
3897         mblk_t  *mp1;
3898         mblk_t  *mp2;
3899 
3900         if (!len || !mpp || !(mp0 = *mpp))
3901                 return (NULL);
3902         /* If we aren't going to consume the first mblk, we need a dup. */
3903         if (mp0->b_wptr - mp0->b_rptr > len) {
3904                 mp1 = dupb(mp0);
3905                 if (mp1) {
3906                         /* Partition the data between the two mblks. */
3907                         mp1->b_wptr = mp1->b_rptr + len;
3908                         mp0->b_rptr = mp1->b_wptr;
3909                         /*
3910                          * after adjustments if mblk not consumed is now
3911                          * unaligned, try to align it. If this fails free
3912                          * all messages and let upper layer recover.
3913                          */
3914                         if (!OK_32PTR(mp0->b_rptr)) {
3915                                 if (!pullupmsg(mp0, -1)) {
3916                                         freemsg(mp0);
3917                                         freemsg(mp1);
3918                                         *mpp = NULL;
3919                                         return (NULL);
3920                                 }
3921                         }
3922                 }
3923                 return (mp1);
3924         }
3925         /* Eat through as many mblks as we need to get len bytes. */
3926         len -= mp0->b_wptr - mp0->b_rptr;
3927         for (mp2 = mp1 = mp0; (mp2 = mp2->b_cont) != 0 && len; mp1 = mp2) {
3928                 if (mp2->b_wptr - mp2->b_rptr > len) {
3929                         /*
3930                          * We won't consume the entire last mblk.  Like
3931                          * above, dup and partition it.
3932                          */
3933                         mp1->b_cont = dupb(mp2);
3934                         mp1 = mp1->b_cont;
3935                         if (!mp1) {
3936                                 /*
3937                                  * Trouble.  Rather than go to a lot of
3938                                  * trouble to clean up, we free the messages.
3939                                  * This won't be any worse than losing it on
3940                                  * the wire.
3941                                  */
3942                                 freemsg(mp0);
3943                                 freemsg(mp2);
3944                                 *mpp = NULL;
3945                                 return (NULL);
3946                         }
3947                         mp1->b_wptr = mp1->b_rptr + len;
3948                         mp2->b_rptr = mp1->b_wptr;
3949                         /*
3950                          * after adjustments if mblk not consumed is now
3951                          * unaligned, try to align it. If this fails free
3952                          * all messages and let upper layer recover.
3953                          */
3954                         if (!OK_32PTR(mp2->b_rptr)) {
3955                                 if (!pullupmsg(mp2, -1)) {
3956                                         freemsg(mp0);
3957                                         freemsg(mp2);
3958                                         *mpp = NULL;
3959                                         return (NULL);
3960                                 }
3961                         }
3962                         *mpp = mp2;
3963                         return (mp0);
3964                 }
3965                 /* Decrement len by the amount we just got. */
3966                 len -= mp2->b_wptr - mp2->b_rptr;
3967         }
3968         /*
3969          * len should be reduced to zero now.  If not our caller has
3970          * screwed up.
3971          */
3972         if (len) {
3973                 /* Shouldn't happen! */
3974                 freemsg(mp0);
3975                 *mpp = NULL;
3976                 return (NULL);
3977         }
3978         /*
3979          * We consumed up to exactly the end of an mblk.  Detach the part
3980          * we are returning from the rest of the chain.
3981          */
3982         mp1->b_cont = NULL;
3983         *mpp = mp2;
3984         return (mp0);
3985 }
3986 
3987 /* The ill stream is being unplumbed. Called from ip_close */
3988 int
3989 ip_modclose(ill_t *ill)
3990 {
3991         boolean_t success;
3992         ipsq_t  *ipsq;
3993         ipif_t  *ipif;
3994         queue_t *q = ill->ill_rq;
3995         ip_stack_t      *ipst = ill->ill_ipst;
3996         int     i;
3997         arl_ill_common_t *ai = ill->ill_common;
3998 
3999         /*
4000          * The punlink prior to this may have initiated a capability
4001          * negotiation. But ipsq_enter will block until that finishes or
4002          * times out.
4003          */
4004         success = ipsq_enter(ill, B_FALSE, NEW_OP);
4005 
4006         /*
4007          * Open/close/push/pop is guaranteed to be single threaded
4008          * per stream by STREAMS. FS guarantees that all references
4009          * from top are gone before close is called. So there can't
4010          * be another close thread that has set CONDEMNED on this ill.
4011          * and cause ipsq_enter to return failure.
4012          */
4013         ASSERT(success);
4014         ipsq = ill->ill_phyint->phyint_ipsq;
4015 
4016         /*
4017          * Mark it condemned. No new reference will be made to this ill.
4018          * Lookup functions will return an error. Threads that try to
4019          * increment the refcnt must check for ILL_CAN_LOOKUP. This ensures
4020          * that the refcnt will drop down to zero.
4021          */
4022         mutex_enter(&ill->ill_lock);
4023         ill->ill_state_flags |= ILL_CONDEMNED;
4024         for (ipif = ill->ill_ipif; ipif != NULL;
4025             ipif = ipif->ipif_next) {
4026                 ipif->ipif_state_flags |= IPIF_CONDEMNED;
4027         }
4028         /*
4029          * Wake up anybody waiting to enter the ipsq. ipsq_enter
4030          * returns  error if ILL_CONDEMNED is set
4031          */
4032         cv_broadcast(&ill->ill_cv);
4033         mutex_exit(&ill->ill_lock);
4034 
4035         /*
4036          * Send all the deferred DLPI messages downstream which came in
4037          * during the small window right before ipsq_enter(). We do this
4038          * without waiting for the ACKs because all the ACKs for M_PROTO
4039          * messages are ignored in ip_rput() when ILL_CONDEMNED is set.
4040          */
4041         ill_dlpi_send_deferred(ill);
4042 
4043         /*
4044          * Shut down fragmentation reassembly.
4045          * ill_frag_timer won't start a timer again.
4046          * Now cancel any existing timer
4047          */
4048         (void) untimeout(ill->ill_frag_timer_id);
4049         (void) ill_frag_timeout(ill, 0);
4050 
4051         /*
4052          * Call ill_delete to bring down the ipifs, ilms and ill on
4053          * this ill. Then wait for the refcnts to drop to zero.
4054          * ill_is_freeable checks whether the ill is really quiescent.
4055          * Then make sure that threads that are waiting to enter the
4056          * ipsq have seen the error returned by ipsq_enter and have
4057          * gone away. Then we call ill_delete_tail which does the
4058          * DL_UNBIND_REQ with the driver and then qprocsoff.
4059          */
4060         ill_delete(ill);
4061         mutex_enter(&ill->ill_lock);
4062         while (!ill_is_freeable(ill))
4063                 cv_wait(&ill->ill_cv, &ill->ill_lock);
4064 
4065         while (ill->ill_waiters)
4066                 cv_wait(&ill->ill_cv, &ill->ill_lock);
4067 
4068         mutex_exit(&ill->ill_lock);
4069 
4070         /*
4071          * ill_delete_tail drops reference on ill_ipst, but we need to keep
4072          * it held until the end of the function since the cleanup
4073          * below needs to be able to use the ip_stack_t.
4074          */
4075         netstack_hold(ipst->ips_netstack);
4076 
4077         /* qprocsoff is done via ill_delete_tail */
4078         ill_delete_tail(ill);
4079         /*
4080          * synchronously wait for arp stream to unbind. After this, we
4081          * cannot get any data packets up from the driver.
4082          */
4083         arp_unbind_complete(ill);
4084         ASSERT(ill->ill_ipst == NULL);
4085 
4086         /*
4087          * Walk through all conns and qenable those that have queued data.
4088          * Close synchronization needs this to
4089          * be done to ensure that all upper layers blocked
4090          * due to flow control to the closing device
4091          * get unblocked.
4092          */
4093         ip1dbg(("ip_wsrv: walking\n"));
4094         for (i = 0; i < TX_FANOUT_SIZE; i++) {
4095                 conn_walk_drain(ipst, &ipst->ips_idl_tx_list[i]);
4096         }
4097 
4098         /*
4099          * ai can be null if this is an IPv6 ill, or if the IPv4
4100          * stream is being torn down before ARP was plumbed (e.g.,
4101          * /sbin/ifconfig plumbing a stream twice, and encountering
4102          * an error
4103          */
4104         if (ai != NULL) {
4105                 ASSERT(!ill->ill_isv6);
4106                 mutex_enter(&ai->ai_lock);
4107                 ai->ai_ill = NULL;
4108                 if (ai->ai_arl == NULL) {
4109                         mutex_destroy(&ai->ai_lock);
4110                         kmem_free(ai, sizeof (*ai));
4111                 } else {
4112                         cv_signal(&ai->ai_ill_unplumb_done);
4113                         mutex_exit(&ai->ai_lock);
4114                 }
4115         }
4116 
4117         mutex_enter(&ipst->ips_ip_mi_lock);
4118         mi_close_unlink(&ipst->ips_ip_g_head, (IDP)ill);
4119         mutex_exit(&ipst->ips_ip_mi_lock);
4120 
4121         /*
4122          * credp could be null if the open didn't succeed and ip_modopen
4123          * itself calls ip_close.
4124          */
4125         if (ill->ill_credp != NULL)
4126                 crfree(ill->ill_credp);
4127 
4128         mutex_destroy(&ill->ill_saved_ire_lock);
4129         mutex_destroy(&ill->ill_lock);
4130         rw_destroy(&ill->ill_mcast_lock);
4131         mutex_destroy(&ill->ill_mcast_serializer);
4132         list_destroy(&ill->ill_nce);
4133 
4134         /*
4135          * Now we are done with the module close pieces that
4136          * need the netstack_t.
4137          */
4138         netstack_rele(ipst->ips_netstack);
4139 
4140         mi_close_free((IDP)ill);
4141         q->q_ptr = WR(q)->q_ptr = NULL;
4142 
4143         ipsq_exit(ipsq);
4144 
4145         return (0);
4146 }
4147 
4148 /*
4149  * This is called as part of close() for IP, UDP, ICMP, and RTS
4150  * in order to quiesce the conn.
4151  */
4152 void
4153 ip_quiesce_conn(conn_t *connp)
4154 {
4155         boolean_t       drain_cleanup_reqd = B_FALSE;
4156         boolean_t       conn_ioctl_cleanup_reqd = B_FALSE;
4157         boolean_t       ilg_cleanup_reqd = B_FALSE;
4158         ip_stack_t      *ipst;
4159 
4160         ASSERT(!IPCL_IS_TCP(connp));
4161         ipst = connp->conn_netstack->netstack_ip;
4162 
4163         /*
4164          * Mark the conn as closing, and this conn must not be
4165          * inserted in future into any list. Eg. conn_drain_insert(),
4166          * won't insert this conn into the conn_drain_list.
4167          *
4168          * conn_idl, and conn_ilg cannot get set henceforth.
4169          */
4170         mutex_enter(&connp->conn_lock);
4171         ASSERT(!(connp->conn_state_flags & CONN_QUIESCED));
4172         connp->conn_state_flags |= CONN_CLOSING;
4173         if (connp->conn_idl != NULL)
4174                 drain_cleanup_reqd = B_TRUE;
4175         if (connp->conn_oper_pending_ill != NULL)
4176                 conn_ioctl_cleanup_reqd = B_TRUE;
4177         if (connp->conn_dhcpinit_ill != NULL) {
4178                 ASSERT(connp->conn_dhcpinit_ill->ill_dhcpinit != 0);
4179                 atomic_dec_32(&connp->conn_dhcpinit_ill->ill_dhcpinit);
4180                 ill_set_inputfn(connp->conn_dhcpinit_ill);
4181                 connp->conn_dhcpinit_ill = NULL;
4182         }
4183         if (connp->conn_ilg != NULL)
4184                 ilg_cleanup_reqd = B_TRUE;
4185         mutex_exit(&connp->conn_lock);
4186 
4187         if (conn_ioctl_cleanup_reqd)
4188                 conn_ioctl_cleanup(connp);
4189 
4190         if (is_system_labeled() && connp->conn_anon_port) {
4191                 (void) tsol_mlp_anon(crgetzone(connp->conn_cred),
4192                     connp->conn_mlp_type, connp->conn_proto,
4193                     ntohs(connp->conn_lport), B_FALSE);
4194                 connp->conn_anon_port = 0;
4195         }
4196         connp->conn_mlp_type = mlptSingle;
4197 
4198         /*
4199          * Remove this conn from any fanout list it is on.
4200          * and then wait for any threads currently operating
4201          * on this endpoint to finish
4202          */
4203         ipcl_hash_remove(connp);
4204 
4205         /*
4206          * Remove this conn from the drain list, and do any other cleanup that
4207          * may be required.  (TCP conns are never flow controlled, and
4208          * conn_idl will be NULL.)
4209          */
4210         if (drain_cleanup_reqd && connp->conn_idl != NULL) {
4211                 idl_t *idl = connp->conn_idl;
4212 
4213                 mutex_enter(&idl->idl_lock);
4214                 conn_drain(connp, B_TRUE);
4215                 mutex_exit(&idl->idl_lock);
4216         }
4217 
4218         if (connp == ipst->ips_ip_g_mrouter)
4219                 (void) ip_mrouter_done(ipst);
4220 
4221         if (ilg_cleanup_reqd)
4222                 ilg_delete_all(connp);
4223 
4224         /*
4225          * Now conn refcnt can increase only thru CONN_INC_REF_LOCKED.
4226          * callers from write side can't be there now because close
4227          * is in progress. The only other caller is ipcl_walk
4228          * which checks for the condemned flag.
4229          */
4230         mutex_enter(&connp->conn_lock);
4231         connp->conn_state_flags |= CONN_CONDEMNED;
4232         while (connp->conn_ref != 1)
4233                 cv_wait(&connp->conn_cv, &connp->conn_lock);
4234         connp->conn_state_flags |= CONN_QUIESCED;
4235         mutex_exit(&connp->conn_lock);
4236 }
4237 
4238 /* ARGSUSED */
4239 int
4240 ip_close(queue_t *q, int flags)
4241 {
4242         conn_t          *connp;
4243 
4244         /*
4245          * Call the appropriate delete routine depending on whether this is
4246          * a module or device.
4247          */
4248         if (WR(q)->q_next != NULL) {
4249                 /* This is a module close */
4250                 return (ip_modclose((ill_t *)q->q_ptr));
4251         }
4252 
4253         connp = q->q_ptr;
4254         ip_quiesce_conn(connp);
4255 
4256         qprocsoff(q);
4257 
4258         /*
4259          * Now we are truly single threaded on this stream, and can
4260          * delete the things hanging off the connp, and finally the connp.
4261          * We removed this connp from the fanout list, it cannot be
4262          * accessed thru the fanouts, and we already waited for the
4263          * conn_ref to drop to 0. We are already in close, so
4264          * there cannot be any other thread from the top. qprocsoff
4265          * has completed, and service has completed or won't run in
4266          * future.
4267          */
4268         ASSERT(connp->conn_ref == 1);
4269 
4270         inet_minor_free(connp->conn_minor_arena, connp->conn_dev);
4271 
4272         connp->conn_ref--;
4273         ipcl_conn_destroy(connp);
4274 
4275         q->q_ptr = WR(q)->q_ptr = NULL;
4276         return (0);
4277 }
4278 
4279 /*
4280  * Wapper around putnext() so that ip_rts_request can merely use
4281  * conn_recv.
4282  */
4283 /*ARGSUSED2*/
4284 static void
4285 ip_conn_input(void *arg1, mblk_t *mp, void *arg2, ip_recv_attr_t *ira)
4286 {
4287         conn_t *connp = (conn_t *)arg1;
4288 
4289         putnext(connp->conn_rq, mp);
4290 }
4291 
4292 /* Dummy in case ICMP error delivery is attempted to a /dev/ip instance */
4293 /* ARGSUSED */
4294 static void
4295 ip_conn_input_icmp(void *arg1, mblk_t *mp, void *arg2, ip_recv_attr_t *ira)
4296 {
4297         freemsg(mp);
4298 }
4299 
4300 /*
4301  * Called when the module is about to be unloaded
4302  */
4303 void
4304 ip_ddi_destroy(void)
4305 {
4306         /* This needs to be called before destroying any transports. */
4307         mutex_enter(&cpu_lock);
4308         unregister_cpu_setup_func(ip_tp_cpu_update, NULL);
4309         mutex_exit(&cpu_lock);
4310 
4311         tnet_fini();
4312 
4313         icmp_ddi_g_destroy();
4314         rts_ddi_g_destroy();
4315         udp_ddi_g_destroy();
4316         dccp_ddi_g_destroy();
4317         sctp_ddi_g_destroy();
4318         tcp_ddi_g_destroy();
4319         ilb_ddi_g_destroy();
4320         dce_g_destroy();
4321         ipsec_policy_g_destroy();
4322         ipcl_g_destroy();
4323         ip_net_g_destroy();
4324         ip_ire_g_fini();
4325         inet_minor_destroy(ip_minor_arena_sa);
4326 #if defined(_LP64)
4327         inet_minor_destroy(ip_minor_arena_la);
4328 #endif
4329 
4330 #ifdef DEBUG
4331         list_destroy(&ip_thread_list);
4332         rw_destroy(&ip_thread_rwlock);
4333         tsd_destroy(&ip_thread_data);
4334 #endif
4335 
4336         netstack_unregister(NS_IP);
4337 }
4338 
4339 /*
4340  * First step in cleanup.
4341  */
4342 /* ARGSUSED */
4343 static void
4344 ip_stack_shutdown(netstackid_t stackid, void *arg)
4345 {
4346         ip_stack_t *ipst = (ip_stack_t *)arg;
4347 
4348 #ifdef NS_DEBUG
4349         printf("ip_stack_shutdown(%p, stack %d)\n", (void *)ipst, stackid);
4350 #endif
4351 
4352         /*
4353          * Perform cleanup for special interfaces (loopback and IPMP).
4354          */
4355         ip_interface_cleanup(ipst);
4356 
4357         /*
4358          * The *_hook_shutdown()s start the process of notifying any
4359          * consumers that things are going away.... nothing is destroyed.
4360          */
4361         ipv4_hook_shutdown(ipst);
4362         ipv6_hook_shutdown(ipst);
4363         arp_hook_shutdown(ipst);
4364 
4365         mutex_enter(&ipst->ips_capab_taskq_lock);
4366         ipst->ips_capab_taskq_quit = B_TRUE;
4367         cv_signal(&ipst->ips_capab_taskq_cv);
4368         mutex_exit(&ipst->ips_capab_taskq_lock);
4369 }
4370 
4371 /*
4372  * Free the IP stack instance.
4373  */
4374 static void
4375 ip_stack_fini(netstackid_t stackid, void *arg)
4376 {
4377         ip_stack_t *ipst = (ip_stack_t *)arg;
4378         int ret;
4379 
4380 #ifdef NS_DEBUG
4381         printf("ip_stack_fini(%p, stack %d)\n", (void *)ipst, stackid);
4382 #endif
4383         /*
4384          * At this point, all of the notifications that the events and
4385          * protocols are going away have been run, meaning that we can
4386          * now set about starting to clean things up.
4387          */
4388         ipobs_fini(ipst);
4389         ipv4_hook_destroy(ipst);
4390         ipv6_hook_destroy(ipst);
4391         arp_hook_destroy(ipst);
4392         ip_net_destroy(ipst);
4393 
4394         ipmp_destroy(ipst);
4395 
4396         ip_kstat_fini(stackid, ipst->ips_ip_mibkp);
4397         ipst->ips_ip_mibkp = NULL;
4398         icmp_kstat_fini(stackid, ipst->ips_icmp_mibkp);
4399         ipst->ips_icmp_mibkp = NULL;
4400         ip_kstat2_fini(stackid, ipst->ips_ip_kstat);
4401         ipst->ips_ip_kstat = NULL;
4402         bzero(&ipst->ips_ip_statistics, sizeof (ipst->ips_ip_statistics));
4403         ip6_kstat_fini(stackid, ipst->ips_ip6_kstat);
4404         ipst->ips_ip6_kstat = NULL;
4405         bzero(&ipst->ips_ip6_statistics, sizeof (ipst->ips_ip6_statistics));
4406 
4407         kmem_free(ipst->ips_propinfo_tbl,
4408             ip_propinfo_count * sizeof (mod_prop_info_t));
4409         ipst->ips_propinfo_tbl = NULL;
4410 
4411         dce_stack_destroy(ipst);
4412         ip_mrouter_stack_destroy(ipst);
4413 
4414         ret = untimeout(ipst->ips_igmp_timeout_id);
4415         if (ret == -1) {
4416                 ASSERT(ipst->ips_igmp_timeout_id == 0);
4417         } else {
4418                 ASSERT(ipst->ips_igmp_timeout_id != 0);
4419                 ipst->ips_igmp_timeout_id = 0;
4420         }
4421         ret = untimeout(ipst->ips_igmp_slowtimeout_id);
4422         if (ret == -1) {
4423                 ASSERT(ipst->ips_igmp_slowtimeout_id == 0);
4424         } else {
4425                 ASSERT(ipst->ips_igmp_slowtimeout_id != 0);
4426                 ipst->ips_igmp_slowtimeout_id = 0;
4427         }
4428         ret = untimeout(ipst->ips_mld_timeout_id);
4429         if (ret == -1) {
4430                 ASSERT(ipst->ips_mld_timeout_id == 0);
4431         } else {
4432                 ASSERT(ipst->ips_mld_timeout_id != 0);
4433                 ipst->ips_mld_timeout_id = 0;
4434         }
4435         ret = untimeout(ipst->ips_mld_slowtimeout_id);
4436         if (ret == -1) {
4437                 ASSERT(ipst->ips_mld_slowtimeout_id == 0);
4438         } else {
4439                 ASSERT(ipst->ips_mld_slowtimeout_id != 0);
4440                 ipst->ips_mld_slowtimeout_id = 0;
4441         }
4442 
4443         ip_ire_fini(ipst);
4444         ip6_asp_free(ipst);
4445         conn_drain_fini(ipst);
4446         ipcl_destroy(ipst);
4447 
4448         mutex_destroy(&ipst->ips_ndp4->ndp_g_lock);
4449         mutex_destroy(&ipst->ips_ndp6->ndp_g_lock);
4450         kmem_free(ipst->ips_ndp4, sizeof (ndp_g_t));
4451         ipst->ips_ndp4 = NULL;
4452         kmem_free(ipst->ips_ndp6, sizeof (ndp_g_t));
4453         ipst->ips_ndp6 = NULL;
4454 
4455         if (ipst->ips_loopback_ksp != NULL) {
4456                 kstat_delete_netstack(ipst->ips_loopback_ksp, stackid);
4457                 ipst->ips_loopback_ksp = NULL;
4458         }
4459 
4460         mutex_destroy(&ipst->ips_capab_taskq_lock);
4461         cv_destroy(&ipst->ips_capab_taskq_cv);
4462 
4463         rw_destroy(&ipst->ips_srcid_lock);
4464 
4465         mutex_destroy(&ipst->ips_ip_mi_lock);
4466         rw_destroy(&ipst->ips_ill_g_usesrc_lock);
4467 
4468         mutex_destroy(&ipst->ips_igmp_timer_lock);
4469         mutex_destroy(&ipst->ips_mld_timer_lock);
4470         mutex_destroy(&ipst->ips_igmp_slowtimeout_lock);
4471         mutex_destroy(&ipst->ips_mld_slowtimeout_lock);
4472         mutex_destroy(&ipst->ips_ip_addr_avail_lock);
4473         rw_destroy(&ipst->ips_ill_g_lock);
4474 
4475         kmem_free(ipst->ips_phyint_g_list, sizeof (phyint_list_t));
4476         ipst->ips_phyint_g_list = NULL;
4477         kmem_free(ipst->ips_ill_g_heads, sizeof (ill_g_head_t) * MAX_G_HEADS);
4478         ipst->ips_ill_g_heads = NULL;
4479 
4480         ldi_ident_release(ipst->ips_ldi_ident);
4481         kmem_free(ipst, sizeof (*ipst));
4482 }
4483 
4484 /*
4485  * This function is called from the TSD destructor, and is used to debug
4486  * reference count issues in IP. See block comment in <inet/ip_if.h> for
4487  * details.
4488  */
4489 static void
4490 ip_thread_exit(void *phash)
4491 {
4492         th_hash_t *thh = phash;
4493 
4494         rw_enter(&ip_thread_rwlock, RW_WRITER);
4495         list_remove(&ip_thread_list, thh);
4496         rw_exit(&ip_thread_rwlock);
4497         mod_hash_destroy_hash(thh->thh_hash);
4498         kmem_free(thh, sizeof (*thh));
4499 }
4500 
4501 /*
4502  * Called when the IP kernel module is loaded into the kernel
4503  */
4504 void
4505 ip_ddi_init(void)
4506 {
4507         ip_squeue_flag = ip_squeue_switch(ip_squeue_enter);
4508 
4509         /*
4510          * For IP and TCP the minor numbers should start from 2 since we have 4
4511          * initial devices: ip, ip6, tcp, tcp6.
4512          */
4513         /*
4514          * If this is a 64-bit kernel, then create two separate arenas -
4515          * one for TLIs in the range of INET_MIN_DEV+2 through 2^^18-1, and the
4516          * other for socket apps in the range 2^^18 through 2^^32-1.
4517          */
4518         ip_minor_arena_la = NULL;
4519         ip_minor_arena_sa = NULL;
4520 #if defined(_LP64)
4521         if ((ip_minor_arena_sa = inet_minor_create("ip_minor_arena_sa",
4522             INET_MIN_DEV + 2, MAXMIN32, KM_SLEEP)) == NULL) {
4523                 cmn_err(CE_PANIC,
4524                     "ip_ddi_init: ip_minor_arena_sa creation failed\n");
4525         }
4526         if ((ip_minor_arena_la = inet_minor_create("ip_minor_arena_la",
4527             MAXMIN32 + 1, MAXMIN64, KM_SLEEP)) == NULL) {
4528                 cmn_err(CE_PANIC,
4529                     "ip_ddi_init: ip_minor_arena_la creation failed\n");
4530         }
4531 #else
4532         if ((ip_minor_arena_sa = inet_minor_create("ip_minor_arena_sa",
4533             INET_MIN_DEV + 2, MAXMIN, KM_SLEEP)) == NULL) {
4534                 cmn_err(CE_PANIC,
4535                     "ip_ddi_init: ip_minor_arena_sa creation failed\n");
4536         }
4537 #endif
4538         ip_poll_normal_ticks = MSEC_TO_TICK_ROUNDUP(ip_poll_normal_ms);
4539 
4540         ipcl_g_init();
4541         ip_ire_g_init();
4542         ip_net_g_init();
4543 
4544 #ifdef DEBUG
4545         tsd_create(&ip_thread_data, ip_thread_exit);
4546         rw_init(&ip_thread_rwlock, NULL, RW_DEFAULT, NULL);
4547         list_create(&ip_thread_list, sizeof (th_hash_t),
4548             offsetof(th_hash_t, thh_link));
4549 #endif
4550         ipsec_policy_g_init();
4551         tcp_ddi_g_init();
4552         sctp_ddi_g_init();
4553         dccp_ddi_g_init();
4554         dce_g_init();
4555 
4556         /*
4557          * We want to be informed each time a stack is created or
4558          * destroyed in the kernel, so we can maintain the
4559          * set of udp_stack_t's.
4560          */
4561         netstack_register(NS_IP, ip_stack_init, ip_stack_shutdown,
4562             ip_stack_fini);
4563 
4564         tnet_init();
4565 
4566         udp_ddi_g_init();
4567         rts_ddi_g_init();
4568         icmp_ddi_g_init();
4569         ilb_ddi_g_init();
4570 
4571         /* This needs to be called after all transports are initialized. */
4572         mutex_enter(&cpu_lock);
4573         register_cpu_setup_func(ip_tp_cpu_update, NULL);
4574         mutex_exit(&cpu_lock);
4575 }
4576 
4577 /*
4578  * Initialize the IP stack instance.
4579  */
4580 static void *
4581 ip_stack_init(netstackid_t stackid, netstack_t *ns)
4582 {
4583         ip_stack_t      *ipst;
4584         size_t          arrsz;
4585         major_t         major;
4586 
4587 #ifdef NS_DEBUG
4588         printf("ip_stack_init(stack %d)\n", stackid);
4589 #endif
4590 
4591         ipst = (ip_stack_t *)kmem_zalloc(sizeof (*ipst), KM_SLEEP);
4592         ipst->ips_netstack = ns;
4593 
4594         ipst->ips_ill_g_heads = kmem_zalloc(sizeof (ill_g_head_t) * MAX_G_HEADS,
4595             KM_SLEEP);
4596         ipst->ips_phyint_g_list = kmem_zalloc(sizeof (phyint_list_t),
4597             KM_SLEEP);
4598         ipst->ips_ndp4 = kmem_zalloc(sizeof (ndp_g_t), KM_SLEEP);
4599         ipst->ips_ndp6 = kmem_zalloc(sizeof (ndp_g_t), KM_SLEEP);
4600         mutex_init(&ipst->ips_ndp4->ndp_g_lock, NULL, MUTEX_DEFAULT, NULL);
4601         mutex_init(&ipst->ips_ndp6->ndp_g_lock, NULL, MUTEX_DEFAULT, NULL);
4602 
4603         mutex_init(&ipst->ips_igmp_timer_lock, NULL, MUTEX_DEFAULT, NULL);
4604         ipst->ips_igmp_deferred_next = INFINITY;
4605         mutex_init(&ipst->ips_mld_timer_lock, NULL, MUTEX_DEFAULT, NULL);
4606         ipst->ips_mld_deferred_next = INFINITY;
4607         mutex_init(&ipst->ips_igmp_slowtimeout_lock, NULL, MUTEX_DEFAULT, NULL);
4608         mutex_init(&ipst->ips_mld_slowtimeout_lock, NULL, MUTEX_DEFAULT, NULL);
4609         mutex_init(&ipst->ips_ip_mi_lock, NULL, MUTEX_DEFAULT, NULL);
4610         mutex_init(&ipst->ips_ip_addr_avail_lock, NULL, MUTEX_DEFAULT, NULL);
4611         rw_init(&ipst->ips_ill_g_lock, NULL, RW_DEFAULT, NULL);
4612         rw_init(&ipst->ips_ill_g_usesrc_lock, NULL, RW_DEFAULT, NULL);
4613 
4614         ipcl_init(ipst);
4615         ip_ire_init(ipst);
4616         ip6_asp_init(ipst);
4617         ipif_init(ipst);
4618         conn_drain_init(ipst);
4619         ip_mrouter_stack_init(ipst);
4620         dce_stack_init(ipst);
4621 
4622         ipst->ips_ip_multirt_log_interval = 1000;
4623 
4624         ipst->ips_ill_index = 1;
4625 
4626         ipst->ips_saved_ip_forwarding = -1;
4627         ipst->ips_reg_vif_num = ALL_VIFS;    /* Index to Register vif */
4628 
4629         arrsz = ip_propinfo_count * sizeof (mod_prop_info_t);
4630         ipst->ips_propinfo_tbl = (mod_prop_info_t *)kmem_alloc(arrsz, KM_SLEEP);
4631         bcopy(ip_propinfo_tbl, ipst->ips_propinfo_tbl, arrsz);
4632 
4633         ipst->ips_ip_mibkp = ip_kstat_init(stackid, ipst);
4634         ipst->ips_icmp_mibkp = icmp_kstat_init(stackid);
4635         ipst->ips_ip_kstat = ip_kstat2_init(stackid, &ipst->ips_ip_statistics);
4636         ipst->ips_ip6_kstat =
4637             ip6_kstat_init(stackid, &ipst->ips_ip6_statistics);
4638 
4639         ipst->ips_ip_src_id = 1;
4640         rw_init(&ipst->ips_srcid_lock, NULL, RW_DEFAULT, NULL);
4641 
4642         ipst->ips_src_generation = SRC_GENERATION_INITIAL;
4643 
4644         ip_net_init(ipst, ns);
4645         ipv4_hook_init(ipst);
4646         ipv6_hook_init(ipst);
4647         arp_hook_init(ipst);
4648         ipmp_init(ipst);
4649         ipobs_init(ipst);
4650 
4651         /*
4652          * Create the taskq dispatcher thread and initialize related stuff.
4653          */
4654         mutex_init(&ipst->ips_capab_taskq_lock, NULL, MUTEX_DEFAULT, NULL);
4655         cv_init(&ipst->ips_capab_taskq_cv, NULL, CV_DEFAULT, NULL);
4656         ipst->ips_capab_taskq_thread = thread_create(NULL, 0,
4657             ill_taskq_dispatch, ipst, 0, &p0, TS_RUN, minclsyspri);
4658 
4659         major = mod_name_to_major(INET_NAME);
4660         (void) ldi_ident_from_major(major, &ipst->ips_ldi_ident);
4661         return (ipst);
4662 }
4663 
4664 /*
4665  * Allocate and initialize a DLPI template of the specified length.  (May be
4666  * called as writer.)
4667  */
4668 mblk_t *
4669 ip_dlpi_alloc(size_t len, t_uscalar_t prim)
4670 {
4671         mblk_t  *mp;
4672 
4673         mp = allocb(len, BPRI_MED);
4674         if (!mp)
4675                 return (NULL);
4676 
4677         /*
4678          * DLPIv2 says that DL_INFO_REQ and DL_TOKEN_REQ (the latter
4679          * of which we don't seem to use) are sent with M_PCPROTO, and
4680          * that other DLPI are M_PROTO.
4681          */
4682         if (prim == DL_INFO_REQ) {
4683                 mp->b_datap->db_type = M_PCPROTO;
4684         } else {
4685                 mp->b_datap->db_type = M_PROTO;
4686         }
4687 
4688         mp->b_wptr = mp->b_rptr + len;
4689         bzero(mp->b_rptr, len);
4690         ((dl_unitdata_req_t *)mp->b_rptr)->dl_primitive = prim;
4691         return (mp);
4692 }
4693 
4694 /*
4695  * Allocate and initialize a DLPI notification.  (May be called as writer.)
4696  */
4697 mblk_t *
4698 ip_dlnotify_alloc(uint_t notification, uint_t data)
4699 {
4700         dl_notify_ind_t *notifyp;
4701         mblk_t          *mp;
4702 
4703         if ((mp = ip_dlpi_alloc(DL_NOTIFY_IND_SIZE, DL_NOTIFY_IND)) == NULL)
4704                 return (NULL);
4705 
4706         notifyp = (dl_notify_ind_t *)mp->b_rptr;
4707         notifyp->dl_notification = notification;
4708         notifyp->dl_data = data;
4709         return (mp);
4710 }
4711 
4712 mblk_t *
4713 ip_dlnotify_alloc2(uint_t notification, uint_t data1, uint_t data2)
4714 {
4715         dl_notify_ind_t *notifyp;
4716         mblk_t          *mp;
4717 
4718         if ((mp = ip_dlpi_alloc(DL_NOTIFY_IND_SIZE, DL_NOTIFY_IND)) == NULL)
4719                 return (NULL);
4720 
4721         notifyp = (dl_notify_ind_t *)mp->b_rptr;
4722         notifyp->dl_notification = notification;
4723         notifyp->dl_data1 = data1;
4724         notifyp->dl_data2 = data2;
4725         return (mp);
4726 }
4727 
4728 /*
4729  * Debug formatting routine.  Returns a character string representation of the
4730  * addr in buf, of the form xxx.xxx.xxx.xxx.  This routine takes the address
4731  * in the form of a ipaddr_t and calls ip_dot_saddr with a pointer.
4732  *
4733  * Once the ndd table-printing interfaces are removed, this can be changed to
4734  * standard dotted-decimal form.
4735  */
4736 char *
4737 ip_dot_addr(ipaddr_t addr, char *buf)
4738 {
4739         uint8_t *ap = (uint8_t *)&addr;
4740 
4741         (void) mi_sprintf(buf, "%03d.%03d.%03d.%03d",
4742             ap[0] & 0xFF, ap[1] & 0xFF, ap[2] & 0xFF, ap[3] & 0xFF);
4743         return (buf);
4744 }
4745 
4746 /*
4747  * Write the given MAC address as a printable string in the usual colon-
4748  * separated format.
4749  */
4750 const char *
4751 mac_colon_addr(const uint8_t *addr, size_t alen, char *buf, size_t buflen)
4752 {
4753         char *bp;
4754 
4755         if (alen == 0 || buflen < 4)
4756                 return ("?");
4757         bp = buf;
4758         for (;;) {
4759                 /*
4760                  * If there are more MAC address bytes available, but we won't
4761                  * have any room to print them, then add "..." to the string
4762                  * instead.  See below for the 'magic number' explanation.
4763                  */
4764                 if ((alen == 2 && buflen < 6) || (alen > 2 && buflen < 7)) {
4765                         (void) strcpy(bp, "...");
4766                         break;
4767                 }
4768                 (void) sprintf(bp, "%02x", *addr++);
4769                 bp += 2;
4770                 if (--alen == 0)
4771                         break;
4772                 *bp++ = ':';
4773                 buflen -= 3;
4774                 /*
4775                  * At this point, based on the first 'if' statement above,
4776                  * either alen == 1 and buflen >= 3, or alen > 1 and
4777                  * buflen >= 4.  The first case leaves room for the final "xx"
4778                  * number and trailing NUL byte.  The second leaves room for at
4779                  * least "...".  Thus the apparently 'magic' numbers chosen for
4780                  * that statement.
4781                  */
4782         }
4783         return (buf);
4784 }
4785 
4786 /*
4787  * Called when it is conceptually a ULP that would sent the packet
4788  * e.g., port unreachable and protocol unreachable. Check that the packet
4789  * would have passed the IPsec global policy before sending the error.
4790  *
4791  * Send an ICMP error after patching up the packet appropriately.
4792  * Uses ip_drop_input and bumps the appropriate MIB.
4793  */
4794 void
4795 ip_fanout_send_icmp_v4(mblk_t *mp, uint_t icmp_type, uint_t icmp_code,
4796     ip_recv_attr_t *ira)
4797 {
4798         ipha_t          *ipha;
4799         boolean_t       secure;
4800         ill_t           *ill = ira->ira_ill;
4801         ip_stack_t      *ipst = ill->ill_ipst;
4802         netstack_t      *ns = ipst->ips_netstack;
4803         ipsec_stack_t   *ipss = ns->netstack_ipsec;
4804 
4805         secure = ira->ira_flags & IRAF_IPSEC_SECURE;
4806 
4807         /*
4808          * We are generating an icmp error for some inbound packet.
4809          * Called from all ip_fanout_(udp, tcp, proto) functions.
4810          * Before we generate an error, check with global policy
4811          * to see whether this is allowed to enter the system. As
4812          * there is no "conn", we are checking with global policy.
4813          */
4814         ipha = (ipha_t *)mp->b_rptr;
4815         if (secure || ipss->ipsec_inbound_v4_policy_present) {
4816                 mp = ipsec_check_global_policy(mp, NULL, ipha, NULL, ira, ns);
4817                 if (mp == NULL)
4818                         return;
4819         }
4820 
4821         /* We never send errors for protocols that we do implement */
4822         if (ira->ira_protocol == IPPROTO_ICMP ||
4823             ira->ira_protocol == IPPROTO_IGMP) {
4824                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
4825                 ip_drop_input("ip_fanout_send_icmp_v4", mp, ill);
4826                 freemsg(mp);
4827                 return;
4828         }
4829         /*
4830          * Have to correct checksum since
4831          * the packet might have been
4832          * fragmented and the reassembly code in ip_rput
4833          * does not restore the IP checksum.
4834          */
4835         ipha->ipha_hdr_checksum = 0;
4836         ipha->ipha_hdr_checksum = ip_csum_hdr(ipha);
4837 
4838         switch (icmp_type) {
4839         case ICMP_DEST_UNREACHABLE:
4840                 switch (icmp_code) {
4841                 case ICMP_PROTOCOL_UNREACHABLE:
4842                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInUnknownProtos);
4843                         ip_drop_input("ipIfStatsInUnknownProtos", mp, ill);
4844                         break;
4845                 case ICMP_PORT_UNREACHABLE:
4846                         BUMP_MIB(ill->ill_ip_mib, udpIfStatsNoPorts);
4847                         ip_drop_input("ipIfStatsNoPorts", mp, ill);
4848                         break;
4849                 }
4850 
4851                 icmp_unreachable(mp, icmp_code, ira);
4852                 break;
4853         default:
4854 #ifdef DEBUG
4855                 panic("ip_fanout_send_icmp_v4: wrong type");
4856                 /*NOTREACHED*/
4857 #else
4858                 freemsg(mp);
4859                 break;
4860 #endif
4861         }
4862 }
4863 
4864 /*
4865  * Used to send an ICMP error message when a packet is received for
4866  * a protocol that is not supported. The mblk passed as argument
4867  * is consumed by this function.
4868  */
4869 void
4870 ip_proto_not_sup(mblk_t *mp, ip_recv_attr_t *ira)
4871 {
4872         ipha_t          *ipha;
4873 
4874         ipha = (ipha_t *)mp->b_rptr;
4875         if (ira->ira_flags & IRAF_IS_IPV4) {
4876                 ASSERT(IPH_HDR_VERSION(ipha) == IP_VERSION);
4877                 ip_fanout_send_icmp_v4(mp, ICMP_DEST_UNREACHABLE,
4878                     ICMP_PROTOCOL_UNREACHABLE, ira);
4879         } else {
4880                 ASSERT(IPH_HDR_VERSION(ipha) == IPV6_VERSION);
4881                 ip_fanout_send_icmp_v6(mp, ICMP6_PARAM_PROB,
4882                     ICMP6_PARAMPROB_NEXTHEADER, ira);
4883         }
4884 }
4885 
4886 /*
4887  * Deliver a rawip packet to the given conn, possibly applying ipsec policy.
4888  * Handles IPv4 and IPv6.
4889  * We are responsible for disposing of mp, such as by freemsg() or putnext()
4890  * Caller is responsible for dropping references to the conn.
4891  */
4892 void
4893 ip_fanout_proto_conn(conn_t *connp, mblk_t *mp, ipha_t *ipha, ip6_t *ip6h,
4894     ip_recv_attr_t *ira)
4895 {
4896         ill_t           *ill = ira->ira_ill;
4897         ip_stack_t      *ipst = ill->ill_ipst;
4898         ipsec_stack_t   *ipss = ipst->ips_netstack->netstack_ipsec;
4899         boolean_t       secure;
4900         uint_t          protocol = ira->ira_protocol;
4901         iaflags_t       iraflags = ira->ira_flags;
4902         queue_t         *rq;
4903 
4904         secure = iraflags & IRAF_IPSEC_SECURE;
4905 
4906         rq = connp->conn_rq;
4907         if (IPCL_IS_NONSTR(connp) ? connp->conn_flow_cntrld : !canputnext(rq)) {
4908                 switch (protocol) {
4909                 case IPPROTO_ICMPV6:
4910                         BUMP_MIB(ill->ill_icmp6_mib, ipv6IfIcmpInOverflows);
4911                         break;
4912                 case IPPROTO_ICMP:
4913                         BUMP_MIB(&ipst->ips_icmp_mib, icmpInOverflows);
4914                         break;
4915                 default:
4916                         BUMP_MIB(ill->ill_ip_mib, rawipIfStatsInOverflows);
4917                         break;
4918                 }
4919                 freemsg(mp);
4920                 return;
4921         }
4922 
4923         ASSERT(!(IPCL_IS_IPTUN(connp)));
4924 
4925         if (((iraflags & IRAF_IS_IPV4) ?
4926             CONN_INBOUND_POLICY_PRESENT(connp, ipss) :
4927             CONN_INBOUND_POLICY_PRESENT_V6(connp, ipss)) ||
4928             secure) {
4929                 mp = ipsec_check_inbound_policy(mp, connp, ipha,
4930                     ip6h, ira);
4931                 if (mp == NULL) {
4932                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
4933                         /* Note that mp is NULL */
4934                         ip_drop_input("ipIfStatsInDiscards", mp, ill);
4935                         return;
4936                 }
4937         }
4938 
4939         if (iraflags & IRAF_ICMP_ERROR) {
4940                 (connp->conn_recvicmp)(connp, mp, NULL, ira);
4941         } else {
4942                 ill_t *rill = ira->ira_rill;
4943 
4944                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCInDelivers);
4945                 ira->ira_ill = ira->ira_rill = NULL;
4946                 /* Send it upstream */
4947                 (connp->conn_recv)(connp, mp, NULL, ira);
4948                 ira->ira_ill = ill;
4949                 ira->ira_rill = rill;
4950         }
4951 }
4952 
4953 /*
4954  * Handle protocols with which IP is less intimate.  There
4955  * can be more than one stream bound to a particular
4956  * protocol.  When this is the case, normally each one gets a copy
4957  * of any incoming packets.
4958  *
4959  * IPsec NOTE :
4960  *
4961  * Don't allow a secure packet going up a non-secure connection.
4962  * We don't allow this because
4963  *
4964  * 1) Reply might go out in clear which will be dropped at
4965  *    the sending side.
4966  * 2) If the reply goes out in clear it will give the
4967  *    adversary enough information for getting the key in
4968  *    most of the cases.
4969  *
4970  * Moreover getting a secure packet when we expect clear
4971  * implies that SA's were added without checking for
4972  * policy on both ends. This should not happen once ISAKMP
4973  * is used to negotiate SAs as SAs will be added only after
4974  * verifying the policy.
4975  *
4976  * Zones notes:
4977  * Earlier in ip_input on a system with multiple shared-IP zones we
4978  * duplicate the multicast and broadcast packets and send them up
4979  * with each explicit zoneid that exists on that ill.
4980  * This means that here we can match the zoneid with SO_ALLZONES being special.
4981  */
4982 void
4983 ip_fanout_proto_v4(mblk_t *mp, ipha_t *ipha, ip_recv_attr_t *ira)
4984 {
4985         mblk_t          *mp1;
4986         ipaddr_t        laddr;
4987         conn_t          *connp, *first_connp, *next_connp;
4988         connf_t         *connfp;
4989         ill_t           *ill = ira->ira_ill;
4990         ip_stack_t      *ipst = ill->ill_ipst;
4991 
4992         laddr = ipha->ipha_dst;
4993 
4994         connfp = &ipst->ips_ipcl_proto_fanout_v4[ira->ira_protocol];
4995         mutex_enter(&connfp->connf_lock);
4996         connp = connfp->connf_head;
4997         for (connp = connfp->connf_head; connp != NULL;
4998             connp = connp->conn_next) {
4999                 /* Note: IPCL_PROTO_MATCH includes conn_wantpacket */
5000                 if (IPCL_PROTO_MATCH(connp, ira, ipha) &&
5001                     (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5002                     tsol_receive_local(mp, &laddr, IPV4_VERSION, ira, connp))) {
5003                         break;
5004                 }
5005         }
5006 
5007         if (connp == NULL) {
5008                 /*
5009                  * No one bound to these addresses.  Is
5010                  * there a client that wants all
5011                  * unclaimed datagrams?
5012                  */
5013                 mutex_exit(&connfp->connf_lock);
5014                 ip_fanout_send_icmp_v4(mp, ICMP_DEST_UNREACHABLE,
5015                     ICMP_PROTOCOL_UNREACHABLE, ira);
5016                 return;
5017         }
5018 
5019         ASSERT(IPCL_IS_NONSTR(connp) || connp->conn_rq != NULL);
5020 
5021         CONN_INC_REF(connp);
5022         first_connp = connp;
5023         connp = connp->conn_next;
5024 
5025         for (;;) {
5026                 while (connp != NULL) {
5027                         /* Note: IPCL_PROTO_MATCH includes conn_wantpacket */
5028                         if (IPCL_PROTO_MATCH(connp, ira, ipha) &&
5029                             (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5030                             tsol_receive_local(mp, &laddr, IPV4_VERSION,
5031                             ira, connp)))
5032                                 break;
5033                         connp = connp->conn_next;
5034                 }
5035 
5036                 if (connp == NULL) {
5037                         /* No more interested clients */
5038                         connp = first_connp;
5039                         break;
5040                 }
5041                 if (((mp1 = dupmsg(mp)) == NULL) &&
5042                     ((mp1 = copymsg(mp)) == NULL)) {
5043                         /* Memory allocation failed */
5044                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
5045                         ip_drop_input("ipIfStatsInDiscards", mp, ill);
5046                         connp = first_connp;
5047                         break;
5048                 }
5049 
5050                 CONN_INC_REF(connp);
5051                 mutex_exit(&connfp->connf_lock);
5052 
5053                 ip_fanout_proto_conn(connp, mp1, (ipha_t *)mp1->b_rptr, NULL,
5054                     ira);
5055 
5056                 mutex_enter(&connfp->connf_lock);
5057                 /* Follow the next pointer before releasing the conn. */
5058                 next_connp = connp->conn_next;
5059                 CONN_DEC_REF(connp);
5060                 connp = next_connp;
5061         }
5062 
5063         /* Last one.  Send it upstream. */
5064         mutex_exit(&connfp->connf_lock);
5065 
5066         ip_fanout_proto_conn(connp, mp, ipha, NULL, ira);
5067 
5068         CONN_DEC_REF(connp);
5069 }
5070 
5071 /*
5072  * If we have a IPsec NAT-Traversal packet, strip the zero-SPI or
5073  * pass it along to ESP if the SPI is non-zero.  Returns the mblk if the mblk
5074  * is not consumed.
5075  *
5076  * One of three things can happen, all of which affect the passed-in mblk:
5077  *
5078  * 1.) The packet is stock UDP and gets its zero-SPI stripped.  Return mblk..
5079  *
5080  * 2.) The packet is ESP-in-UDP, gets transformed into an equivalent
5081  *     ESP packet, and is passed along to ESP for consumption.  Return NULL.
5082  *
5083  * 3.) The packet is an ESP-in-UDP Keepalive.  Drop it and return NULL.
5084  */
5085 mblk_t *
5086 zero_spi_check(mblk_t *mp, ip_recv_attr_t *ira)
5087 {
5088         int shift, plen, iph_len;
5089         ipha_t *ipha;
5090         udpha_t *udpha;
5091         uint32_t *spi;
5092         uint32_t esp_ports;
5093         uint8_t *orptr;
5094         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
5095         ipsec_stack_t   *ipss = ipst->ips_netstack->netstack_ipsec;
5096 
5097         ipha = (ipha_t *)mp->b_rptr;
5098         iph_len = ira->ira_ip_hdr_length;
5099         plen = ira->ira_pktlen;
5100 
5101         if (plen - iph_len - sizeof (udpha_t) < sizeof (uint32_t)) {
5102                 /*
5103                  * Most likely a keepalive for the benefit of an intervening
5104                  * NAT.  These aren't for us, per se, so drop it.
5105                  *
5106                  * RFC 3947/8 doesn't say for sure what to do for 2-3
5107                  * byte packets (keepalives are 1-byte), but we'll drop them
5108                  * also.
5109                  */
5110                 ip_drop_packet(mp, B_TRUE, ira->ira_ill,
5111                     DROPPER(ipss, ipds_esp_nat_t_ka), &ipss->ipsec_dropper);
5112                 return (NULL);
5113         }
5114 
5115         if (MBLKL(mp) < iph_len + sizeof (udpha_t) + sizeof (*spi)) {
5116                 /* might as well pull it all up - it might be ESP. */
5117                 if (!pullupmsg(mp, -1)) {
5118                         ip_drop_packet(mp, B_TRUE, ira->ira_ill,
5119                             DROPPER(ipss, ipds_esp_nomem),
5120                             &ipss->ipsec_dropper);
5121                         return (NULL);
5122                 }
5123 
5124                 ipha = (ipha_t *)mp->b_rptr;
5125         }
5126         spi = (uint32_t *)(mp->b_rptr + iph_len + sizeof (udpha_t));
5127         if (*spi == 0) {
5128                 /* UDP packet - remove 0-spi. */
5129                 shift = sizeof (uint32_t);
5130         } else {
5131                 /* ESP-in-UDP packet - reduce to ESP. */
5132                 ipha->ipha_protocol = IPPROTO_ESP;
5133                 shift = sizeof (udpha_t);
5134         }
5135 
5136         /* Fix IP header */
5137         ira->ira_pktlen = (plen - shift);
5138         ipha->ipha_length = htons(ira->ira_pktlen);
5139         ipha->ipha_hdr_checksum = 0;
5140 
5141         orptr = mp->b_rptr;
5142         mp->b_rptr += shift;
5143 
5144         udpha = (udpha_t *)(orptr + iph_len);
5145         if (*spi == 0) {
5146                 ASSERT((uint8_t *)ipha == orptr);
5147                 udpha->uha_length = htons(plen - shift - iph_len);
5148                 iph_len += sizeof (udpha_t);    /* For the call to ovbcopy(). */
5149                 esp_ports = 0;
5150         } else {
5151                 esp_ports = *((uint32_t *)udpha);
5152                 ASSERT(esp_ports != 0);
5153         }
5154         ovbcopy(orptr, orptr + shift, iph_len);
5155         if (esp_ports != 0) /* Punt up for ESP processing. */ {
5156                 ipha = (ipha_t *)(orptr + shift);
5157 
5158                 ira->ira_flags |= IRAF_ESP_UDP_PORTS;
5159                 ira->ira_esp_udp_ports = esp_ports;
5160                 ip_fanout_v4(mp, ipha, ira);
5161                 return (NULL);
5162         }
5163         return (mp);
5164 }
5165 
5166 /*
5167  * Deliver a udp packet to the given conn, possibly applying ipsec policy.
5168  * Handles IPv4 and IPv6.
5169  * We are responsible for disposing of mp, such as by freemsg() or putnext()
5170  * Caller is responsible for dropping references to the conn.
5171  */
5172 void
5173 ip_fanout_udp_conn(conn_t *connp, mblk_t *mp, ipha_t *ipha, ip6_t *ip6h,
5174     ip_recv_attr_t *ira)
5175 {
5176         ill_t           *ill = ira->ira_ill;
5177         ip_stack_t      *ipst = ill->ill_ipst;
5178         ipsec_stack_t   *ipss = ipst->ips_netstack->netstack_ipsec;
5179         boolean_t       secure;
5180         iaflags_t       iraflags = ira->ira_flags;
5181 
5182         secure = iraflags & IRAF_IPSEC_SECURE;
5183 
5184         if (IPCL_IS_NONSTR(connp) ? connp->conn_flow_cntrld :
5185             !canputnext(connp->conn_rq)) {
5186                 BUMP_MIB(ill->ill_ip_mib, udpIfStatsInOverflows);
5187                 freemsg(mp);
5188                 return;
5189         }
5190 
5191         if (((iraflags & IRAF_IS_IPV4) ?
5192             CONN_INBOUND_POLICY_PRESENT(connp, ipss) :
5193             CONN_INBOUND_POLICY_PRESENT_V6(connp, ipss)) ||
5194             secure) {
5195                 mp = ipsec_check_inbound_policy(mp, connp, ipha,
5196                     ip6h, ira);
5197                 if (mp == NULL) {
5198                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
5199                         /* Note that mp is NULL */
5200                         ip_drop_input("ipIfStatsInDiscards", mp, ill);
5201                         return;
5202                 }
5203         }
5204 
5205         /*
5206          * Since this code is not used for UDP unicast we don't need a NAT_T
5207          * check. Only ip_fanout_v4 has that check.
5208          */
5209         if (ira->ira_flags & IRAF_ICMP_ERROR) {
5210                 (connp->conn_recvicmp)(connp, mp, NULL, ira);
5211         } else {
5212                 ill_t *rill = ira->ira_rill;
5213 
5214                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCInDelivers);
5215                 ira->ira_ill = ira->ira_rill = NULL;
5216                 /* Send it upstream */
5217                 (connp->conn_recv)(connp, mp, NULL, ira);
5218                 ira->ira_ill = ill;
5219                 ira->ira_rill = rill;
5220         }
5221 }
5222 
5223 /*
5224  * Fanout for UDP packets that are multicast or broadcast, and ICMP errors.
5225  * (Unicast fanout is handled in ip_input_v4.)
5226  *
5227  * If SO_REUSEADDR is set all multicast and broadcast packets
5228  * will be delivered to all conns bound to the same port.
5229  *
5230  * If there is at least one matching AF_INET receiver, then we will
5231  * ignore any AF_INET6 receivers.
5232  * In the special case where an AF_INET socket binds to 0.0.0.0/<port> and an
5233  * AF_INET6 socket binds to ::/<port>, only the AF_INET socket receives the IPv4
5234  * packets.
5235  *
5236  * Zones notes:
5237  * Earlier in ip_input on a system with multiple shared-IP zones we
5238  * duplicate the multicast and broadcast packets and send them up
5239  * with each explicit zoneid that exists on that ill.
5240  * This means that here we can match the zoneid with SO_ALLZONES being special.
5241  */
5242 void
5243 ip_fanout_udp_multi_v4(mblk_t *mp, ipha_t *ipha, uint16_t lport, uint16_t fport,
5244     ip_recv_attr_t *ira)
5245 {
5246         ipaddr_t        laddr;
5247         in6_addr_t      v6faddr;
5248         conn_t          *connp;
5249         connf_t         *connfp;
5250         ipaddr_t        faddr;
5251         ill_t           *ill = ira->ira_ill;
5252         ip_stack_t      *ipst = ill->ill_ipst;
5253 
5254         ASSERT(ira->ira_flags & (IRAF_MULTIBROADCAST|IRAF_ICMP_ERROR));
5255 
5256         laddr = ipha->ipha_dst;
5257         faddr = ipha->ipha_src;
5258 
5259         connfp = &ipst->ips_ipcl_udp_fanout[IPCL_UDP_HASH(lport, ipst)];
5260         mutex_enter(&connfp->connf_lock);
5261         connp = connfp->connf_head;
5262 
5263         /*
5264          * If SO_REUSEADDR has been set on the first we send the
5265          * packet to all clients that have joined the group and
5266          * match the port.
5267          */
5268         while (connp != NULL) {
5269                 if ((IPCL_UDP_MATCH(connp, lport, laddr, fport, faddr)) &&
5270                     conn_wantpacket(connp, ira, ipha) &&
5271                     (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5272                     tsol_receive_local(mp, &laddr, IPV4_VERSION, ira, connp)))
5273                         break;
5274                 connp = connp->conn_next;
5275         }
5276 
5277         if (connp == NULL)
5278                 goto notfound;
5279 
5280         CONN_INC_REF(connp);
5281 
5282         if (connp->conn_reuseaddr) {
5283                 conn_t          *first_connp = connp;
5284                 conn_t          *next_connp;
5285                 mblk_t          *mp1;
5286 
5287                 connp = connp->conn_next;
5288                 for (;;) {
5289                         while (connp != NULL) {
5290                                 if (IPCL_UDP_MATCH(connp, lport, laddr,
5291                                     fport, faddr) &&
5292                                     conn_wantpacket(connp, ira, ipha) &&
5293                                     (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5294                                     tsol_receive_local(mp, &laddr, IPV4_VERSION,
5295                                     ira, connp)))
5296                                         break;
5297                                 connp = connp->conn_next;
5298                         }
5299                         if (connp == NULL) {
5300                                 /* No more interested clients */
5301                                 connp = first_connp;
5302                                 break;
5303                         }
5304                         if (((mp1 = dupmsg(mp)) == NULL) &&
5305                             ((mp1 = copymsg(mp)) == NULL)) {
5306                                 /* Memory allocation failed */
5307                                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
5308                                 ip_drop_input("ipIfStatsInDiscards", mp, ill);
5309                                 connp = first_connp;
5310                                 break;
5311                         }
5312                         CONN_INC_REF(connp);
5313                         mutex_exit(&connfp->connf_lock);
5314 
5315                         IP_STAT(ipst, ip_udp_fanmb);
5316                         ip_fanout_udp_conn(connp, mp1, (ipha_t *)mp1->b_rptr,
5317                             NULL, ira);
5318                         mutex_enter(&connfp->connf_lock);
5319                         /* Follow the next pointer before releasing the conn */
5320                         next_connp = connp->conn_next;
5321                         CONN_DEC_REF(connp);
5322                         connp = next_connp;
5323                 }
5324         }
5325 
5326         /* Last one.  Send it upstream. */
5327         mutex_exit(&connfp->connf_lock);
5328         IP_STAT(ipst, ip_udp_fanmb);
5329         ip_fanout_udp_conn(connp, mp, ipha, NULL, ira);
5330         CONN_DEC_REF(connp);
5331         return;
5332 
5333 notfound:
5334         mutex_exit(&connfp->connf_lock);
5335         /*
5336          * IPv6 endpoints bound to multicast IPv4-mapped addresses
5337          * have already been matched above, since they live in the IPv4
5338          * fanout tables. This implies we only need to
5339          * check for IPv6 in6addr_any endpoints here.
5340          * Thus we compare using ipv6_all_zeros instead of the destination
5341          * address, except for the multicast group membership lookup which
5342          * uses the IPv4 destination.
5343          */
5344         IN6_IPADDR_TO_V4MAPPED(ipha->ipha_src, &v6faddr);
5345         connfp = &ipst->ips_ipcl_udp_fanout[IPCL_UDP_HASH(lport, ipst)];
5346         mutex_enter(&connfp->connf_lock);
5347         connp = connfp->connf_head;
5348         /*
5349          * IPv4 multicast packet being delivered to an AF_INET6
5350          * in6addr_any endpoint.
5351          * Need to check conn_wantpacket(). Note that we use conn_wantpacket()
5352          * and not conn_wantpacket_v6() since any multicast membership is
5353          * for an IPv4-mapped multicast address.
5354          */
5355         while (connp != NULL) {
5356                 if (IPCL_UDP_MATCH_V6(connp, lport, ipv6_all_zeros,
5357                     fport, v6faddr) &&
5358                     conn_wantpacket(connp, ira, ipha) &&
5359                     (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5360                     tsol_receive_local(mp, &laddr, IPV4_VERSION, ira, connp)))
5361                         break;
5362                 connp = connp->conn_next;
5363         }
5364 
5365         if (connp == NULL) {
5366                 /*
5367                  * No one bound to this port.  Is
5368                  * there a client that wants all
5369                  * unclaimed datagrams?
5370                  */
5371                 mutex_exit(&connfp->connf_lock);
5372 
5373                 if (ipst->ips_ipcl_proto_fanout_v4[IPPROTO_UDP].connf_head !=
5374                     NULL) {
5375                         ASSERT(ira->ira_protocol == IPPROTO_UDP);
5376                         ip_fanout_proto_v4(mp, ipha, ira);
5377                 } else {
5378                         /*
5379                          * We used to attempt to send an icmp error here, but
5380                          * since this is known to be a multicast packet
5381                          * and we don't send icmp errors in response to
5382                          * multicast, just drop the packet and give up sooner.
5383                          */
5384                         BUMP_MIB(ill->ill_ip_mib, udpIfStatsNoPorts);
5385                         freemsg(mp);
5386                 }
5387                 return;
5388         }
5389         ASSERT(IPCL_IS_NONSTR(connp) || connp->conn_rq != NULL);
5390 
5391         /*
5392          * If SO_REUSEADDR has been set on the first we send the
5393          * packet to all clients that have joined the group and
5394          * match the port.
5395          */
5396         if (connp->conn_reuseaddr) {
5397                 conn_t          *first_connp = connp;
5398                 conn_t          *next_connp;
5399                 mblk_t          *mp1;
5400 
5401                 CONN_INC_REF(connp);
5402                 connp = connp->conn_next;
5403                 for (;;) {
5404                         while (connp != NULL) {
5405                                 if (IPCL_UDP_MATCH_V6(connp, lport,
5406                                     ipv6_all_zeros, fport, v6faddr) &&
5407                                     conn_wantpacket(connp, ira, ipha) &&
5408                                     (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5409                                     tsol_receive_local(mp, &laddr, IPV4_VERSION,
5410                                     ira, connp)))
5411                                         break;
5412                                 connp = connp->conn_next;
5413                         }
5414                         if (connp == NULL) {
5415                                 /* No more interested clients */
5416                                 connp = first_connp;
5417                                 break;
5418                         }
5419                         if (((mp1 = dupmsg(mp)) == NULL) &&
5420                             ((mp1 = copymsg(mp)) == NULL)) {
5421                                 /* Memory allocation failed */
5422                                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
5423                                 ip_drop_input("ipIfStatsInDiscards", mp, ill);
5424                                 connp = first_connp;
5425                                 break;
5426                         }
5427                         CONN_INC_REF(connp);
5428                         mutex_exit(&connfp->connf_lock);
5429 
5430                         IP_STAT(ipst, ip_udp_fanmb);
5431                         ip_fanout_udp_conn(connp, mp1, (ipha_t *)mp1->b_rptr,
5432                             NULL, ira);
5433                         mutex_enter(&connfp->connf_lock);
5434                         /* Follow the next pointer before releasing the conn */
5435                         next_connp = connp->conn_next;
5436                         CONN_DEC_REF(connp);
5437                         connp = next_connp;
5438                 }
5439         }
5440 
5441         /* Last one.  Send it upstream. */
5442         mutex_exit(&connfp->connf_lock);
5443         IP_STAT(ipst, ip_udp_fanmb);
5444         ip_fanout_udp_conn(connp, mp, ipha, NULL, ira);
5445         CONN_DEC_REF(connp);
5446 }
5447 
5448 /*
5449  * Split an incoming packet's IPv4 options into the label and the other options.
5450  * If 'allocate' is set it does memory allocation for the ip_pkt_t, including
5451  * clearing out any leftover label or options.
5452  * Otherwise it just makes ipp point into the packet.
5453  *
5454  * Returns zero if ok; ENOMEM if the buffer couldn't be allocated.
5455  */
5456 int
5457 ip_find_hdr_v4(ipha_t *ipha, ip_pkt_t *ipp, boolean_t allocate)
5458 {
5459         uchar_t         *opt;
5460         uint32_t        totallen;
5461         uint32_t        optval;
5462         uint32_t        optlen;
5463 
5464         ipp->ipp_fields |= IPPF_HOPLIMIT | IPPF_TCLASS | IPPF_ADDR;
5465         ipp->ipp_hoplimit = ipha->ipha_ttl;
5466         ipp->ipp_type_of_service = ipha->ipha_type_of_service;
5467         IN6_IPADDR_TO_V4MAPPED(ipha->ipha_dst, &ipp->ipp_addr);
5468 
5469         /*
5470          * Get length (in 4 byte octets) of IP header options.
5471          */
5472         totallen = ipha->ipha_version_and_hdr_length -
5473             (uint8_t)((IP_VERSION << 4) + IP_SIMPLE_HDR_LENGTH_IN_WORDS);
5474 
5475         if (totallen == 0) {
5476                 if (!allocate)
5477                         return (0);
5478 
5479                 /* Clear out anything from a previous packet */
5480                 if (ipp->ipp_fields & IPPF_IPV4_OPTIONS) {
5481                         kmem_free(ipp->ipp_ipv4_options,
5482                             ipp->ipp_ipv4_options_len);
5483                         ipp->ipp_ipv4_options = NULL;
5484                         ipp->ipp_ipv4_options_len = 0;
5485                         ipp->ipp_fields &= ~IPPF_IPV4_OPTIONS;
5486                 }
5487                 if (ipp->ipp_fields & IPPF_LABEL_V4) {
5488                         kmem_free(ipp->ipp_label_v4, ipp->ipp_label_len_v4);
5489                         ipp->ipp_label_v4 = NULL;
5490                         ipp->ipp_label_len_v4 = 0;
5491                         ipp->ipp_fields &= ~IPPF_LABEL_V4;
5492                 }
5493                 return (0);
5494         }
5495 
5496         totallen <<= 2;
5497         opt = (uchar_t *)&ipha[1];
5498         if (!is_system_labeled()) {
5499 
5500         copyall:
5501                 if (!allocate) {
5502                         if (totallen != 0) {
5503                                 ipp->ipp_ipv4_options = opt;
5504                                 ipp->ipp_ipv4_options_len = totallen;
5505                                 ipp->ipp_fields |= IPPF_IPV4_OPTIONS;
5506                         }
5507                         return (0);
5508                 }
5509                 /* Just copy all of options */
5510                 if (ipp->ipp_fields & IPPF_IPV4_OPTIONS) {
5511                         if (totallen == ipp->ipp_ipv4_options_len) {
5512                                 bcopy(opt, ipp->ipp_ipv4_options, totallen);
5513                                 return (0);
5514                         }
5515                         kmem_free(ipp->ipp_ipv4_options,
5516                             ipp->ipp_ipv4_options_len);
5517                         ipp->ipp_ipv4_options = NULL;
5518                         ipp->ipp_ipv4_options_len = 0;
5519                         ipp->ipp_fields &= ~IPPF_IPV4_OPTIONS;
5520                 }
5521                 if (totallen == 0)
5522                         return (0);
5523 
5524                 ipp->ipp_ipv4_options = kmem_alloc(totallen, KM_NOSLEEP);
5525                 if (ipp->ipp_ipv4_options == NULL)
5526                         return (ENOMEM);
5527                 ipp->ipp_ipv4_options_len = totallen;
5528                 ipp->ipp_fields |= IPPF_IPV4_OPTIONS;
5529                 bcopy(opt, ipp->ipp_ipv4_options, totallen);
5530                 return (0);
5531         }
5532 
5533         if (allocate && (ipp->ipp_fields & IPPF_LABEL_V4)) {
5534                 kmem_free(ipp->ipp_label_v4, ipp->ipp_label_len_v4);
5535                 ipp->ipp_label_v4 = NULL;
5536                 ipp->ipp_label_len_v4 = 0;
5537                 ipp->ipp_fields &= ~IPPF_LABEL_V4;
5538         }
5539 
5540         /*
5541          * Search for CIPSO option.
5542          * We assume CIPSO is first in options if it is present.
5543          * If it isn't, then ipp_opt_ipv4_options will not include the options
5544          * prior to the CIPSO option.
5545          */
5546         while (totallen != 0) {
5547                 switch (optval = opt[IPOPT_OPTVAL]) {
5548                 case IPOPT_EOL:
5549                         return (0);
5550                 case IPOPT_NOP:
5551                         optlen = 1;
5552                         break;
5553                 default:
5554                         if (totallen <= IPOPT_OLEN)
5555                                 return (EINVAL);
5556                         optlen = opt[IPOPT_OLEN];
5557                         if (optlen < 2)
5558                                 return (EINVAL);
5559                 }
5560                 if (optlen > totallen)
5561                         return (EINVAL);
5562 
5563                 switch (optval) {
5564                 case IPOPT_COMSEC:
5565                         if (!allocate) {
5566                                 ipp->ipp_label_v4 = opt;
5567                                 ipp->ipp_label_len_v4 = optlen;
5568                                 ipp->ipp_fields |= IPPF_LABEL_V4;
5569                         } else {
5570                                 ipp->ipp_label_v4 = kmem_alloc(optlen,
5571                                     KM_NOSLEEP);
5572                                 if (ipp->ipp_label_v4 == NULL)
5573                                         return (ENOMEM);
5574                                 ipp->ipp_label_len_v4 = optlen;
5575                                 ipp->ipp_fields |= IPPF_LABEL_V4;
5576                                 bcopy(opt, ipp->ipp_label_v4, optlen);
5577                         }
5578                         totallen -= optlen;
5579                         opt += optlen;
5580 
5581                         /* Skip padding bytes until we get to a multiple of 4 */
5582                         while ((totallen & 3) != 0 && opt[0] == IPOPT_NOP) {
5583                                 totallen--;
5584                                 opt++;
5585                         }
5586                         /* Remaining as ipp_ipv4_options */
5587                         goto copyall;
5588                 }
5589                 totallen -= optlen;
5590                 opt += optlen;
5591         }
5592         /* No CIPSO found; return everything as ipp_ipv4_options */
5593         totallen = ipha->ipha_version_and_hdr_length -
5594             (uint8_t)((IP_VERSION << 4) + IP_SIMPLE_HDR_LENGTH_IN_WORDS);
5595         totallen <<= 2;
5596         opt = (uchar_t *)&ipha[1];
5597         goto copyall;
5598 }
5599 
5600 /*
5601  * Efficient versions of lookup for an IRE when we only
5602  * match the address.
5603  * For RTF_REJECT or BLACKHOLE we return IRE_NOROUTE.
5604  * Does not handle multicast addresses.
5605  */
5606 uint_t
5607 ip_type_v4(ipaddr_t addr, ip_stack_t *ipst)
5608 {
5609         ire_t *ire;
5610         uint_t result;
5611 
5612         ire = ire_ftable_lookup_simple_v4(addr, 0, ipst, NULL);
5613         ASSERT(ire != NULL);
5614         if (ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE))
5615                 result = IRE_NOROUTE;
5616         else
5617                 result = ire->ire_type;
5618         ire_refrele(ire);
5619         return (result);
5620 }
5621 
5622 /*
5623  * Efficient versions of lookup for an IRE when we only
5624  * match the address.
5625  * For RTF_REJECT or BLACKHOLE we return IRE_NOROUTE.
5626  * Does not handle multicast addresses.
5627  */
5628 uint_t
5629 ip_type_v6(const in6_addr_t *addr, ip_stack_t *ipst)
5630 {
5631         ire_t *ire;
5632         uint_t result;
5633 
5634         ire = ire_ftable_lookup_simple_v6(addr, 0, ipst, NULL);
5635         ASSERT(ire != NULL);
5636         if (ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE))
5637                 result = IRE_NOROUTE;
5638         else
5639                 result = ire->ire_type;
5640         ire_refrele(ire);
5641         return (result);
5642 }
5643 
5644 /*
5645  * Nobody should be sending
5646  * packets up this stream
5647  */
5648 static void
5649 ip_lrput(queue_t *q, mblk_t *mp)
5650 {
5651         switch (mp->b_datap->db_type) {
5652         case M_FLUSH:
5653                 /* Turn around */
5654                 if (*mp->b_rptr & FLUSHW) {
5655                         *mp->b_rptr &= ~FLUSHR;
5656                         qreply(q, mp);
5657                         return;
5658                 }
5659                 break;
5660         }
5661         freemsg(mp);
5662 }
5663 
5664 /* Nobody should be sending packets down this stream */
5665 /* ARGSUSED */
5666 void
5667 ip_lwput(queue_t *q, mblk_t *mp)
5668 {
5669         freemsg(mp);
5670 }
5671 
5672 /*
5673  * Move the first hop in any source route to ipha_dst and remove that part of
5674  * the source route.  Called by other protocols.  Errors in option formatting
5675  * are ignored - will be handled by ip_output_options. Return the final
5676  * destination (either ipha_dst or the last entry in a source route.)
5677  */
5678 ipaddr_t
5679 ip_massage_options(ipha_t *ipha, netstack_t *ns)
5680 {
5681         ipoptp_t        opts;
5682         uchar_t         *opt;
5683         uint8_t         optval;
5684         uint8_t         optlen;
5685         ipaddr_t        dst;
5686         int             i;
5687         ip_stack_t      *ipst = ns->netstack_ip;
5688 
5689         ip2dbg(("ip_massage_options\n"));
5690         dst = ipha->ipha_dst;
5691         for (optval = ipoptp_first(&opts, ipha);
5692             optval != IPOPT_EOL;
5693             optval = ipoptp_next(&opts)) {
5694                 opt = opts.ipoptp_cur;
5695                 switch (optval) {
5696                         uint8_t off;
5697                 case IPOPT_SSRR:
5698                 case IPOPT_LSRR:
5699                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
5700                                 ip1dbg(("ip_massage_options: bad src route\n"));
5701                                 break;
5702                         }
5703                         optlen = opts.ipoptp_len;
5704                         off = opt[IPOPT_OFFSET];
5705                         off--;
5706                 redo_srr:
5707                         if (optlen < IP_ADDR_LEN ||
5708                             off > optlen - IP_ADDR_LEN) {
5709                                 /* End of source route */
5710                                 ip1dbg(("ip_massage_options: end of SR\n"));
5711                                 break;
5712                         }
5713                         bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
5714                         ip1dbg(("ip_massage_options: next hop 0x%x\n",
5715                             ntohl(dst)));
5716                         /*
5717                          * Check if our address is present more than
5718                          * once as consecutive hops in source route.
5719                          * XXX verify per-interface ip_forwarding
5720                          * for source route?
5721                          */
5722                         if (ip_type_v4(dst, ipst) == IRE_LOCAL) {
5723                                 off += IP_ADDR_LEN;
5724                                 goto redo_srr;
5725                         }
5726                         if (dst == htonl(INADDR_LOOPBACK)) {
5727                                 ip1dbg(("ip_massage_options: loopback addr in "
5728                                     "source route!\n"));
5729                                 break;
5730                         }
5731                         /*
5732                          * Update ipha_dst to be the first hop and remove the
5733                          * first hop from the source route (by overwriting
5734                          * part of the option with NOP options).
5735                          */
5736                         ipha->ipha_dst = dst;
5737                         /* Put the last entry in dst */
5738                         off = ((optlen - IP_ADDR_LEN - 3) & ~(IP_ADDR_LEN-1)) +
5739                             3;
5740                         bcopy(&opt[off], &dst, IP_ADDR_LEN);
5741 
5742                         ip1dbg(("ip_massage_options: last hop 0x%x\n",
5743                             ntohl(dst)));
5744                         /* Move down and overwrite */
5745                         opt[IP_ADDR_LEN] = opt[0];
5746                         opt[IP_ADDR_LEN+1] = opt[IPOPT_OLEN] - IP_ADDR_LEN;
5747                         opt[IP_ADDR_LEN+2] = opt[IPOPT_OFFSET];
5748                         for (i = 0; i < IP_ADDR_LEN; i++)
5749                                 opt[i] = IPOPT_NOP;
5750                         break;
5751                 }
5752         }
5753         return (dst);
5754 }
5755 
5756 /*
5757  * Return the network mask
5758  * associated with the specified address.
5759  */
5760 ipaddr_t
5761 ip_net_mask(ipaddr_t addr)
5762 {
5763         uchar_t *up = (uchar_t *)&addr;
5764         ipaddr_t mask = 0;
5765         uchar_t *maskp = (uchar_t *)&mask;
5766 
5767 #if defined(__i386) || defined(__amd64)
5768 #define TOTALLY_BRAIN_DAMAGED_C_COMPILER
5769 #endif
5770 #ifdef  TOTALLY_BRAIN_DAMAGED_C_COMPILER
5771         maskp[0] = maskp[1] = maskp[2] = maskp[3] = 0;
5772 #endif
5773         if (CLASSD(addr)) {
5774                 maskp[0] = 0xF0;
5775                 return (mask);
5776         }
5777 
5778         /* We assume Class E default netmask to be 32 */
5779         if (CLASSE(addr))
5780                 return (0xffffffffU);
5781 
5782         if (addr == 0)
5783                 return (0);
5784         maskp[0] = 0xFF;
5785         if ((up[0] & 0x80) == 0)
5786                 return (mask);
5787 
5788         maskp[1] = 0xFF;
5789         if ((up[0] & 0xC0) == 0x80)
5790                 return (mask);
5791 
5792         maskp[2] = 0xFF;
5793         if ((up[0] & 0xE0) == 0xC0)
5794                 return (mask);
5795 
5796         /* Otherwise return no mask */
5797         return ((ipaddr_t)0);
5798 }
5799 
5800 /* Name/Value Table Lookup Routine */
5801 char *
5802 ip_nv_lookup(nv_t *nv, int value)
5803 {
5804         if (!nv)
5805                 return (NULL);
5806         for (; nv->nv_name; nv++) {
5807                 if (nv->nv_value == value)
5808                         return (nv->nv_name);
5809         }
5810         return ("unknown");
5811 }
5812 
5813 static int
5814 ip_wait_for_info_ack(ill_t *ill)
5815 {
5816         int err;
5817 
5818         mutex_enter(&ill->ill_lock);
5819         while (ill->ill_state_flags & ILL_LL_SUBNET_PENDING) {
5820                 /*
5821                  * Return value of 0 indicates a pending signal.
5822                  */
5823                 err = cv_wait_sig(&ill->ill_cv, &ill->ill_lock);
5824                 if (err == 0) {
5825                         mutex_exit(&ill->ill_lock);
5826                         return (EINTR);
5827                 }
5828         }
5829         mutex_exit(&ill->ill_lock);
5830         /*
5831          * ip_rput_other could have set an error  in ill_error on
5832          * receipt of M_ERROR.
5833          */
5834         return (ill->ill_error);
5835 }
5836 
5837 /*
5838  * This is a module open, i.e. this is a control stream for access
5839  * to a DLPI device.  We allocate an ill_t as the instance data in
5840  * this case.
5841  */
5842 static int
5843 ip_modopen(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
5844 {
5845         ill_t   *ill;
5846         int     err;
5847         zoneid_t zoneid;
5848         netstack_t *ns;
5849         ip_stack_t *ipst;
5850 
5851         /*
5852          * Prevent unprivileged processes from pushing IP so that
5853          * they can't send raw IP.
5854          */
5855         if (secpolicy_net_rawaccess(credp) != 0)
5856                 return (EPERM);
5857 
5858         ns = netstack_find_by_cred(credp);
5859         ASSERT(ns != NULL);
5860         ipst = ns->netstack_ip;
5861         ASSERT(ipst != NULL);
5862 
5863         /*
5864          * For exclusive stacks we set the zoneid to zero
5865          * to make IP operate as if in the global zone.
5866          */
5867         if (ipst->ips_netstack->netstack_stackid != GLOBAL_NETSTACKID)
5868                 zoneid = GLOBAL_ZONEID;
5869         else
5870                 zoneid = crgetzoneid(credp);
5871 
5872         ill = (ill_t *)mi_open_alloc_sleep(sizeof (ill_t));
5873         q->q_ptr = WR(q)->q_ptr = ill;
5874         ill->ill_ipst = ipst;
5875         ill->ill_zoneid = zoneid;
5876 
5877         /*
5878          * ill_init initializes the ill fields and then sends down
5879          * down a DL_INFO_REQ after calling qprocson.
5880          */
5881         err = ill_init(q, ill);
5882 
5883         if (err != 0) {
5884                 mi_free(ill);
5885                 netstack_rele(ipst->ips_netstack);
5886                 q->q_ptr = NULL;
5887                 WR(q)->q_ptr = NULL;
5888                 return (err);
5889         }
5890 
5891         /*
5892          * Wait for the DL_INFO_ACK if a DL_INFO_REQ was sent.
5893          *
5894          * ill_init initializes the ipsq marking this thread as
5895          * writer
5896          */
5897         ipsq_exit(ill->ill_phyint->phyint_ipsq);
5898         err = ip_wait_for_info_ack(ill);
5899         if (err == 0)
5900                 ill->ill_credp = credp;
5901         else
5902                 goto fail;
5903 
5904         crhold(credp);
5905 
5906         mutex_enter(&ipst->ips_ip_mi_lock);
5907         err = mi_open_link(&ipst->ips_ip_g_head, (IDP)q->q_ptr, devp, flag,
5908             sflag, credp);
5909         mutex_exit(&ipst->ips_ip_mi_lock);
5910 fail:
5911         if (err) {
5912                 (void) ip_close(q, 0);
5913                 return (err);
5914         }
5915         return (0);
5916 }
5917 
5918 /* For /dev/ip aka AF_INET open */
5919 int
5920 ip_openv4(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
5921 {
5922         return (ip_open(q, devp, flag, sflag, credp, B_FALSE));
5923 }
5924 
5925 /* For /dev/ip6 aka AF_INET6 open */
5926 int
5927 ip_openv6(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
5928 {
5929         return (ip_open(q, devp, flag, sflag, credp, B_TRUE));
5930 }
5931 
5932 /* IP open routine. */
5933 int
5934 ip_open(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp,
5935     boolean_t isv6)
5936 {
5937         conn_t          *connp;
5938         major_t         maj;
5939         zoneid_t        zoneid;
5940         netstack_t      *ns;
5941         ip_stack_t      *ipst;
5942 
5943         /* Allow reopen. */
5944         if (q->q_ptr != NULL)
5945                 return (0);
5946 
5947         if (sflag & MODOPEN) {
5948                 /* This is a module open */
5949                 return (ip_modopen(q, devp, flag, sflag, credp));
5950         }
5951 
5952         if ((flag & ~(FKLYR)) == IP_HELPER_STR) {
5953                 /*
5954                  * Non streams based socket looking for a stream
5955                  * to access IP
5956                  */
5957                 return (ip_helper_stream_setup(q, devp, flag, sflag,
5958                     credp, isv6));
5959         }
5960 
5961         ns = netstack_find_by_cred(credp);
5962         ASSERT(ns != NULL);
5963         ipst = ns->netstack_ip;
5964         ASSERT(ipst != NULL);
5965 
5966         /*
5967          * For exclusive stacks we set the zoneid to zero
5968          * to make IP operate as if in the global zone.
5969          */
5970         if (ipst->ips_netstack->netstack_stackid != GLOBAL_NETSTACKID)
5971                 zoneid = GLOBAL_ZONEID;
5972         else
5973                 zoneid = crgetzoneid(credp);
5974 
5975         /*
5976          * We are opening as a device. This is an IP client stream, and we
5977          * allocate an conn_t as the instance data.
5978          */
5979         connp = ipcl_conn_create(IPCL_IPCCONN, KM_SLEEP, ipst->ips_netstack);
5980 
5981         /*
5982          * ipcl_conn_create did a netstack_hold. Undo the hold that was
5983          * done by netstack_find_by_cred()
5984          */
5985         netstack_rele(ipst->ips_netstack);
5986 
5987         connp->conn_ixa->ixa_flags |= IXAF_MULTICAST_LOOP | IXAF_SET_ULP_CKSUM;
5988         /* conn_allzones can not be set this early, hence no IPCL_ZONEID */
5989         connp->conn_ixa->ixa_zoneid = zoneid;
5990         connp->conn_zoneid = zoneid;
5991 
5992         connp->conn_rq = q;
5993         q->q_ptr = WR(q)->q_ptr = connp;
5994 
5995         /* Minor tells us which /dev entry was opened */
5996         if (isv6) {
5997                 connp->conn_family = AF_INET6;
5998                 connp->conn_ipversion = IPV6_VERSION;
5999                 connp->conn_ixa->ixa_flags &= ~IXAF_IS_IPV4;
6000                 connp->conn_ixa->ixa_src_preferences = IPV6_PREFER_SRC_DEFAULT;
6001         } else {
6002                 connp->conn_family = AF_INET;
6003                 connp->conn_ipversion = IPV4_VERSION;
6004                 connp->conn_ixa->ixa_flags |= IXAF_IS_IPV4;
6005         }
6006 
6007         if ((ip_minor_arena_la != NULL) && (flag & SO_SOCKSTR) &&
6008             ((connp->conn_dev = inet_minor_alloc(ip_minor_arena_la)) != 0)) {
6009                 connp->conn_minor_arena = ip_minor_arena_la;
6010         } else {
6011                 /*
6012                  * Either minor numbers in the large arena were exhausted
6013                  * or a non socket application is doing the open.
6014                  * Try to allocate from the small arena.
6015                  */
6016                 if ((connp->conn_dev =
6017                     inet_minor_alloc(ip_minor_arena_sa)) == 0) {
6018                         /* CONN_DEC_REF takes care of netstack_rele() */
6019                         q->q_ptr = WR(q)->q_ptr = NULL;
6020                         CONN_DEC_REF(connp);
6021                         return (EBUSY);
6022                 }
6023                 connp->conn_minor_arena = ip_minor_arena_sa;
6024         }
6025 
6026         maj = getemajor(*devp);
6027         *devp = makedevice(maj, (minor_t)connp->conn_dev);
6028 
6029         /*
6030          * connp->conn_cred is crfree()ed in ipcl_conn_destroy()
6031          */
6032         connp->conn_cred = credp;
6033         connp->conn_cpid = curproc->p_pid;
6034         /* Cache things in ixa without an extra refhold */
6035         ASSERT(!(connp->conn_ixa->ixa_free_flags & IXA_FREE_CRED));
6036         connp->conn_ixa->ixa_cred = connp->conn_cred;
6037         connp->conn_ixa->ixa_cpid = connp->conn_cpid;
6038         if (is_system_labeled())
6039                 connp->conn_ixa->ixa_tsl = crgetlabel(connp->conn_cred);
6040 
6041         /*
6042          * Handle IP_IOC_RTS_REQUEST and other ioctls which use conn_recv
6043          */
6044         connp->conn_recv = ip_conn_input;
6045         connp->conn_recvicmp = ip_conn_input_icmp;
6046 
6047         crhold(connp->conn_cred);
6048 
6049         /*
6050          * If the caller has the process-wide flag set, then default to MAC
6051          * exempt mode.  This allows read-down to unlabeled hosts.
6052          */
6053         if (getpflags(NET_MAC_AWARE, credp) != 0)
6054                 connp->conn_mac_mode = CONN_MAC_AWARE;
6055 
6056         connp->conn_zone_is_global = (crgetzoneid(credp) == GLOBAL_ZONEID);
6057 
6058         connp->conn_rq = q;
6059         connp->conn_wq = WR(q);
6060 
6061         /* Non-zero default values */
6062         connp->conn_ixa->ixa_flags |= IXAF_MULTICAST_LOOP;
6063 
6064         /*
6065          * Make the conn globally visible to walkers
6066          */
6067         ASSERT(connp->conn_ref == 1);
6068         mutex_enter(&connp->conn_lock);
6069         connp->conn_state_flags &= ~CONN_INCIPIENT;
6070         mutex_exit(&connp->conn_lock);
6071 
6072         qprocson(q);
6073 
6074         return (0);
6075 }
6076 
6077 /*
6078  * Set IPsec policy from an ipsec_req_t. If the req is not "zero" and valid,
6079  * all of them are copied to the conn_t. If the req is "zero", the policy is
6080  * zeroed out. A "zero" policy has zero ipsr_{ah,req,self_encap}_req
6081  * fields.
6082  * We keep only the latest setting of the policy and thus policy setting
6083  * is not incremental/cumulative.
6084  *
6085  * Requests to set policies with multiple alternative actions will
6086  * go through a different API.
6087  */
6088 int
6089 ipsec_set_req(cred_t *cr, conn_t *connp, ipsec_req_t *req)
6090 {
6091         uint_t ah_req = 0;
6092         uint_t esp_req = 0;
6093         uint_t se_req = 0;
6094         ipsec_act_t *actp = NULL;
6095         uint_t nact;
6096         ipsec_policy_head_t *ph;
6097         boolean_t is_pol_reset, is_pol_inserted = B_FALSE;
6098         int error = 0;
6099         netstack_t      *ns = connp->conn_netstack;
6100         ip_stack_t      *ipst = ns->netstack_ip;
6101         ipsec_stack_t   *ipss = ns->netstack_ipsec;
6102 
6103 #define REQ_MASK (IPSEC_PREF_REQUIRED|IPSEC_PREF_NEVER)
6104 
6105         /*
6106          * The IP_SEC_OPT option does not allow variable length parameters,
6107          * hence a request cannot be NULL.
6108          */
6109         if (req == NULL)
6110                 return (EINVAL);
6111 
6112         ah_req = req->ipsr_ah_req;
6113         esp_req = req->ipsr_esp_req;
6114         se_req = req->ipsr_self_encap_req;
6115 
6116         /* Don't allow setting self-encap without one or more of AH/ESP. */
6117         if (se_req != 0 && esp_req == 0 && ah_req == 0)
6118                 return (EINVAL);
6119 
6120         /*
6121          * Are we dealing with a request to reset the policy (i.e.
6122          * zero requests).
6123          */
6124         is_pol_reset = ((ah_req & REQ_MASK) == 0 &&
6125             (esp_req & REQ_MASK) == 0 &&
6126             (se_req & REQ_MASK) == 0);
6127 
6128         if (!is_pol_reset) {
6129                 /*
6130                  * If we couldn't load IPsec, fail with "protocol
6131                  * not supported".
6132                  * IPsec may not have been loaded for a request with zero
6133                  * policies, so we don't fail in this case.
6134                  */
6135                 mutex_enter(&ipss->ipsec_loader_lock);
6136                 if (ipss->ipsec_loader_state != IPSEC_LOADER_SUCCEEDED) {
6137                         mutex_exit(&ipss->ipsec_loader_lock);
6138                         return (EPROTONOSUPPORT);
6139                 }
6140                 mutex_exit(&ipss->ipsec_loader_lock);
6141 
6142                 /*
6143                  * Test for valid requests. Invalid algorithms
6144                  * need to be tested by IPsec code because new
6145                  * algorithms can be added dynamically.
6146                  */
6147                 if ((ah_req & ~(REQ_MASK|IPSEC_PREF_UNIQUE)) != 0 ||
6148                     (esp_req & ~(REQ_MASK|IPSEC_PREF_UNIQUE)) != 0 ||
6149                     (se_req & ~(REQ_MASK|IPSEC_PREF_UNIQUE)) != 0) {
6150                         return (EINVAL);
6151                 }
6152 
6153                 /*
6154                  * Only privileged users can issue these
6155                  * requests.
6156                  */
6157                 if (((ah_req & IPSEC_PREF_NEVER) ||
6158                     (esp_req & IPSEC_PREF_NEVER) ||
6159                     (se_req & IPSEC_PREF_NEVER)) &&
6160                     secpolicy_ip_config(cr, B_FALSE) != 0) {
6161                         return (EPERM);
6162                 }
6163 
6164                 /*
6165                  * The IPSEC_PREF_REQUIRED and IPSEC_PREF_NEVER
6166                  * are mutually exclusive.
6167                  */
6168                 if (((ah_req & REQ_MASK) == REQ_MASK) ||
6169                     ((esp_req & REQ_MASK) == REQ_MASK) ||
6170                     ((se_req & REQ_MASK) == REQ_MASK)) {
6171                         /* Both of them are set */
6172                         return (EINVAL);
6173                 }
6174         }
6175 
6176         ASSERT(MUTEX_HELD(&connp->conn_lock));
6177 
6178         /*
6179          * If we have already cached policies in conn_connect(), don't
6180          * let them change now. We cache policies for connections
6181          * whose src,dst [addr, port] is known.
6182          */
6183         if (connp->conn_policy_cached) {
6184                 return (EINVAL);
6185         }
6186 
6187         /*
6188          * We have a zero policies, reset the connection policy if already
6189          * set. This will cause the connection to inherit the
6190          * global policy, if any.
6191          */
6192         if (is_pol_reset) {
6193                 if (connp->conn_policy != NULL) {
6194                         IPPH_REFRELE(connp->conn_policy, ipst->ips_netstack);
6195                         connp->conn_policy = NULL;
6196                 }
6197                 connp->conn_in_enforce_policy = B_FALSE;
6198                 connp->conn_out_enforce_policy = B_FALSE;
6199                 return (0);
6200         }
6201 
6202         ph = connp->conn_policy = ipsec_polhead_split(connp->conn_policy,
6203             ipst->ips_netstack);
6204         if (ph == NULL)
6205                 goto enomem;
6206 
6207         ipsec_actvec_from_req(req, &actp, &nact, ipst->ips_netstack);
6208         if (actp == NULL)
6209                 goto enomem;
6210 
6211         /*
6212          * Always insert IPv4 policy entries, since they can also apply to
6213          * ipv6 sockets being used in ipv4-compat mode.
6214          */
6215         if (!ipsec_polhead_insert(ph, actp, nact, IPSEC_AF_V4,
6216             IPSEC_TYPE_INBOUND, ns))
6217                 goto enomem;
6218         is_pol_inserted = B_TRUE;
6219         if (!ipsec_polhead_insert(ph, actp, nact, IPSEC_AF_V4,
6220             IPSEC_TYPE_OUTBOUND, ns))
6221                 goto enomem;
6222 
6223         /*
6224          * We're looking at a v6 socket, also insert the v6-specific
6225          * entries.
6226          */
6227         if (connp->conn_family == AF_INET6) {
6228                 if (!ipsec_polhead_insert(ph, actp, nact, IPSEC_AF_V6,
6229                     IPSEC_TYPE_INBOUND, ns))
6230                         goto enomem;
6231                 if (!ipsec_polhead_insert(ph, actp, nact, IPSEC_AF_V6,
6232                     IPSEC_TYPE_OUTBOUND, ns))
6233                         goto enomem;
6234         }
6235 
6236         ipsec_actvec_free(actp, nact);
6237 
6238         /*
6239          * If the requests need security, set enforce_policy.
6240          * If the requests are IPSEC_PREF_NEVER, one should
6241          * still set conn_out_enforce_policy so that ip_set_destination
6242          * marks the ip_xmit_attr_t appropriatly. This is needed so that
6243          * for connections that we don't cache policy in at connect time,
6244          * if global policy matches in ip_output_attach_policy, we
6245          * don't wrongly inherit global policy. Similarly, we need
6246          * to set conn_in_enforce_policy also so that we don't verify
6247          * policy wrongly.
6248          */
6249         if ((ah_req & REQ_MASK) != 0 ||
6250             (esp_req & REQ_MASK) != 0 ||
6251             (se_req & REQ_MASK) != 0) {
6252                 connp->conn_in_enforce_policy = B_TRUE;
6253                 connp->conn_out_enforce_policy = B_TRUE;
6254         }
6255 
6256         return (error);
6257 #undef REQ_MASK
6258 
6259         /*
6260          * Common memory-allocation-failure exit path.
6261          */
6262 enomem:
6263         if (actp != NULL)
6264                 ipsec_actvec_free(actp, nact);
6265         if (is_pol_inserted)
6266                 ipsec_polhead_flush(ph, ns);
6267         return (ENOMEM);
6268 }
6269 
6270 /*
6271  * Set socket options for joining and leaving multicast groups.
6272  * Common to IPv4 and IPv6; inet6 indicates the type of socket.
6273  * The caller has already check that the option name is consistent with
6274  * the address family of the socket.
6275  */
6276 int
6277 ip_opt_set_multicast_group(conn_t *connp, t_scalar_t name,
6278     uchar_t *invalp, boolean_t inet6, boolean_t checkonly)
6279 {
6280         int             *i1 = (int *)invalp;
6281         int             error = 0;
6282         ip_stack_t      *ipst = connp->conn_netstack->netstack_ip;
6283         struct ip_mreq  *v4_mreqp;
6284         struct ipv6_mreq *v6_mreqp;
6285         struct group_req *greqp;
6286         ire_t *ire;
6287         boolean_t done = B_FALSE;
6288         ipaddr_t ifaddr;
6289         in6_addr_t v6group;
6290         uint_t ifindex;
6291         boolean_t mcast_opt = B_TRUE;
6292         mcast_record_t fmode;
6293         int (*optfn)(conn_t *, boolean_t, const in6_addr_t *,
6294             ipaddr_t, uint_t, mcast_record_t, const in6_addr_t *);
6295 
6296         switch (name) {
6297         case IP_ADD_MEMBERSHIP:
6298         case IPV6_JOIN_GROUP:
6299                 mcast_opt = B_FALSE;
6300                 /* FALLTHRU */
6301         case MCAST_JOIN_GROUP:
6302                 fmode = MODE_IS_EXCLUDE;
6303                 optfn = ip_opt_add_group;
6304                 break;
6305 
6306         case IP_DROP_MEMBERSHIP:
6307         case IPV6_LEAVE_GROUP:
6308                 mcast_opt = B_FALSE;
6309                 /* FALLTHRU */
6310         case MCAST_LEAVE_GROUP:
6311                 fmode = MODE_IS_INCLUDE;
6312                 optfn = ip_opt_delete_group;
6313                 break;
6314         default:
6315                 ASSERT(0);
6316         }
6317 
6318         if (mcast_opt) {
6319                 struct sockaddr_in *sin;
6320                 struct sockaddr_in6 *sin6;
6321 
6322                 greqp = (struct group_req *)i1;
6323                 if (greqp->gr_group.ss_family == AF_INET) {
6324                         sin = (struct sockaddr_in *)&(greqp->gr_group);
6325                         IN6_INADDR_TO_V4MAPPED(&sin->sin_addr, &v6group);
6326                 } else {
6327                         if (!inet6)
6328                                 return (EINVAL);        /* Not on INET socket */
6329 
6330                         sin6 = (struct sockaddr_in6 *)&(greqp->gr_group);
6331                         v6group = sin6->sin6_addr;
6332                 }
6333                 ifaddr = INADDR_ANY;
6334                 ifindex = greqp->gr_interface;
6335         } else if (inet6) {
6336                 v6_mreqp = (struct ipv6_mreq *)i1;
6337                 v6group = v6_mreqp->ipv6mr_multiaddr;
6338                 ifaddr = INADDR_ANY;
6339                 ifindex = v6_mreqp->ipv6mr_interface;
6340         } else {
6341                 v4_mreqp = (struct ip_mreq *)i1;
6342                 IN6_INADDR_TO_V4MAPPED(&v4_mreqp->imr_multiaddr, &v6group);
6343                 ifaddr = (ipaddr_t)v4_mreqp->imr_interface.s_addr;
6344                 ifindex = 0;
6345         }
6346 
6347         /*
6348          * In the multirouting case, we need to replicate
6349          * the request on all interfaces that will take part
6350          * in replication.  We do so because multirouting is
6351          * reflective, thus we will probably receive multi-
6352          * casts on those interfaces.
6353          * The ip_multirt_apply_membership() succeeds if
6354          * the operation succeeds on at least one interface.
6355          */
6356         if (IN6_IS_ADDR_V4MAPPED(&v6group)) {
6357                 ipaddr_t group;
6358 
6359                 IN6_V4MAPPED_TO_IPADDR(&v6group, group);
6360 
6361                 ire = ire_ftable_lookup_v4(group, IP_HOST_MASK, 0,
6362                     IRE_HOST | IRE_INTERFACE, NULL, ALL_ZONES, NULL,
6363                     MATCH_IRE_MASK | MATCH_IRE_TYPE, 0, ipst, NULL);
6364         } else {
6365                 ire = ire_ftable_lookup_v6(&v6group, &ipv6_all_ones, 0,
6366                     IRE_HOST | IRE_INTERFACE, NULL, ALL_ZONES, NULL,
6367                     MATCH_IRE_MASK | MATCH_IRE_TYPE, 0, ipst, NULL);
6368         }
6369         if (ire != NULL) {
6370                 if (ire->ire_flags & RTF_MULTIRT) {
6371                         error = ip_multirt_apply_membership(optfn, ire, connp,
6372                             checkonly, &v6group, fmode, &ipv6_all_zeros);
6373                         done = B_TRUE;
6374                 }
6375                 ire_refrele(ire);
6376         }
6377 
6378         if (!done) {
6379                 error = optfn(connp, checkonly, &v6group, ifaddr, ifindex,
6380                     fmode, &ipv6_all_zeros);
6381         }
6382         return (error);
6383 }
6384 
6385 /*
6386  * Set socket options for joining and leaving multicast groups
6387  * for specific sources.
6388  * Common to IPv4 and IPv6; inet6 indicates the type of socket.
6389  * The caller has already check that the option name is consistent with
6390  * the address family of the socket.
6391  */
6392 int
6393 ip_opt_set_multicast_sources(conn_t *connp, t_scalar_t name,
6394     uchar_t *invalp, boolean_t inet6, boolean_t checkonly)
6395 {
6396         int             *i1 = (int *)invalp;
6397         int             error = 0;
6398         ip_stack_t      *ipst = connp->conn_netstack->netstack_ip;
6399         struct ip_mreq_source *imreqp;
6400         struct group_source_req *gsreqp;
6401         in6_addr_t v6group, v6src;
6402         uint32_t ifindex;
6403         ipaddr_t ifaddr;
6404         boolean_t mcast_opt = B_TRUE;
6405         mcast_record_t fmode;
6406         ire_t *ire;
6407         boolean_t done = B_FALSE;
6408         int (*optfn)(conn_t *, boolean_t, const in6_addr_t *,
6409             ipaddr_t, uint_t, mcast_record_t, const in6_addr_t *);
6410 
6411         switch (name) {
6412         case IP_BLOCK_SOURCE:
6413                 mcast_opt = B_FALSE;
6414                 /* FALLTHRU */
6415         case MCAST_BLOCK_SOURCE:
6416                 fmode = MODE_IS_EXCLUDE;
6417                 optfn = ip_opt_add_group;
6418                 break;
6419 
6420         case IP_UNBLOCK_SOURCE:
6421                 mcast_opt = B_FALSE;
6422                 /* FALLTHRU */
6423         case MCAST_UNBLOCK_SOURCE:
6424                 fmode = MODE_IS_EXCLUDE;
6425                 optfn = ip_opt_delete_group;
6426                 break;
6427 
6428         case IP_ADD_SOURCE_MEMBERSHIP:
6429                 mcast_opt = B_FALSE;
6430                 /* FALLTHRU */
6431         case MCAST_JOIN_SOURCE_GROUP:
6432                 fmode = MODE_IS_INCLUDE;
6433                 optfn = ip_opt_add_group;
6434                 break;
6435 
6436         case IP_DROP_SOURCE_MEMBERSHIP:
6437                 mcast_opt = B_FALSE;
6438                 /* FALLTHRU */
6439         case MCAST_LEAVE_SOURCE_GROUP:
6440                 fmode = MODE_IS_INCLUDE;
6441                 optfn = ip_opt_delete_group;
6442                 break;
6443         default:
6444                 ASSERT(0);
6445         }
6446 
6447         if (mcast_opt) {
6448                 gsreqp = (struct group_source_req *)i1;
6449                 ifindex = gsreqp->gsr_interface;
6450                 if (gsreqp->gsr_group.ss_family == AF_INET) {
6451                         struct sockaddr_in *s;
6452                         s = (struct sockaddr_in *)&gsreqp->gsr_group;
6453                         IN6_INADDR_TO_V4MAPPED(&s->sin_addr, &v6group);
6454                         s = (struct sockaddr_in *)&gsreqp->gsr_source;
6455                         IN6_INADDR_TO_V4MAPPED(&s->sin_addr, &v6src);
6456                 } else {
6457                         struct sockaddr_in6 *s6;
6458 
6459                         if (!inet6)
6460                                 return (EINVAL);        /* Not on INET socket */
6461 
6462                         s6 = (struct sockaddr_in6 *)&gsreqp->gsr_group;
6463                         v6group = s6->sin6_addr;
6464                         s6 = (struct sockaddr_in6 *)&gsreqp->gsr_source;
6465                         v6src = s6->sin6_addr;
6466                 }
6467                 ifaddr = INADDR_ANY;
6468         } else {
6469                 imreqp = (struct ip_mreq_source *)i1;
6470                 IN6_INADDR_TO_V4MAPPED(&imreqp->imr_multiaddr, &v6group);
6471                 IN6_INADDR_TO_V4MAPPED(&imreqp->imr_sourceaddr, &v6src);
6472                 ifaddr = (ipaddr_t)imreqp->imr_interface.s_addr;
6473                 ifindex = 0;
6474         }
6475 
6476         /*
6477          * Handle src being mapped INADDR_ANY by changing it to unspecified.
6478          */
6479         if (IN6_IS_ADDR_V4MAPPED_ANY(&v6src))
6480                 v6src = ipv6_all_zeros;
6481 
6482         /*
6483          * In the multirouting case, we need to replicate
6484          * the request as noted in the mcast cases above.
6485          */
6486         if (IN6_IS_ADDR_V4MAPPED(&v6group)) {
6487                 ipaddr_t group;
6488 
6489                 IN6_V4MAPPED_TO_IPADDR(&v6group, group);
6490 
6491                 ire = ire_ftable_lookup_v4(group, IP_HOST_MASK, 0,
6492                     IRE_HOST | IRE_INTERFACE, NULL, ALL_ZONES, NULL,
6493                     MATCH_IRE_MASK | MATCH_IRE_TYPE, 0, ipst, NULL);
6494         } else {
6495                 ire = ire_ftable_lookup_v6(&v6group, &ipv6_all_ones, 0,
6496                     IRE_HOST | IRE_INTERFACE, NULL, ALL_ZONES, NULL,
6497                     MATCH_IRE_MASK | MATCH_IRE_TYPE, 0, ipst, NULL);
6498         }
6499         if (ire != NULL) {
6500                 if (ire->ire_flags & RTF_MULTIRT) {
6501                         error = ip_multirt_apply_membership(optfn, ire, connp,
6502                             checkonly, &v6group, fmode, &v6src);
6503                         done = B_TRUE;
6504                 }
6505                 ire_refrele(ire);
6506         }
6507         if (!done) {
6508                 error = optfn(connp, checkonly, &v6group, ifaddr, ifindex,
6509                     fmode, &v6src);
6510         }
6511         return (error);
6512 }
6513 
6514 /*
6515  * Given a destination address and a pointer to where to put the information
6516  * this routine fills in the mtuinfo.
6517  * The socket must be connected.
6518  * For sctp conn_faddr is the primary address.
6519  */
6520 int
6521 ip_fill_mtuinfo(conn_t *connp, ip_xmit_attr_t *ixa, struct ip6_mtuinfo *mtuinfo)
6522 {
6523         uint32_t        pmtu = IP_MAXPACKET;
6524         uint_t          scopeid;
6525 
6526         if (IN6_IS_ADDR_UNSPECIFIED(&connp->conn_faddr_v6))
6527                 return (-1);
6528 
6529         /* In case we never sent or called ip_set_destination_v4/v6 */
6530         if (ixa->ixa_ire != NULL)
6531                 pmtu = ip_get_pmtu(ixa);
6532 
6533         if (ixa->ixa_flags & IXAF_SCOPEID_SET)
6534                 scopeid = ixa->ixa_scopeid;
6535         else
6536                 scopeid = 0;
6537 
6538         bzero(mtuinfo, sizeof (*mtuinfo));
6539         mtuinfo->ip6m_addr.sin6_family = AF_INET6;
6540         mtuinfo->ip6m_addr.sin6_port = connp->conn_fport;
6541         mtuinfo->ip6m_addr.sin6_addr = connp->conn_faddr_v6;
6542         mtuinfo->ip6m_addr.sin6_scope_id = scopeid;
6543         mtuinfo->ip6m_mtu = pmtu;
6544 
6545         return (sizeof (struct ip6_mtuinfo));
6546 }
6547 
6548 /*
6549  * When the src multihoming is changed from weak to [strong, preferred]
6550  * ip_ire_rebind_walker is called to walk the list of all ire_t entries
6551  * and identify routes that were created by user-applications in the
6552  * unbound state (i.e., without RTA_IFP), and for which an ire_ill is not
6553  * currently defined. These routes are then 'rebound', i.e., their ire_ill
6554  * is selected by finding an interface route for the gateway.
6555  */
6556 /* ARGSUSED */
6557 void
6558 ip_ire_rebind_walker(ire_t *ire, void *notused)
6559 {
6560         if (!ire->ire_unbound || ire->ire_ill != NULL)
6561                 return;
6562         ire_rebind(ire);
6563         ire_delete(ire);
6564 }
6565 
6566 /*
6567  * When the src multihoming is changed from  [strong, preferred] to weak,
6568  * ip_ire_unbind_walker is called to walk the list of all ire_t entries, and
6569  * set any entries that were created by user-applications in the unbound state
6570  * (i.e., without RTA_IFP) back to having a NULL ire_ill.
6571  */
6572 /* ARGSUSED */
6573 void
6574 ip_ire_unbind_walker(ire_t *ire, void *notused)
6575 {
6576         ire_t *new_ire;
6577 
6578         if (!ire->ire_unbound || ire->ire_ill == NULL)
6579                 return;
6580         if (ire->ire_ipversion == IPV6_VERSION) {
6581                 new_ire = ire_create_v6(&ire->ire_addr_v6, &ire->ire_mask_v6,
6582                     &ire->ire_gateway_addr_v6, ire->ire_type, NULL,
6583                     ire->ire_zoneid, ire->ire_flags, NULL, ire->ire_ipst);
6584         } else {
6585                 new_ire = ire_create((uchar_t *)&ire->ire_addr,
6586                     (uchar_t *)&ire->ire_mask,
6587                     (uchar_t *)&ire->ire_gateway_addr, ire->ire_type, NULL,
6588                     ire->ire_zoneid, ire->ire_flags, NULL, ire->ire_ipst);
6589         }
6590         if (new_ire == NULL)
6591                 return;
6592         new_ire->ire_unbound = B_TRUE;
6593         /*
6594          * The bound ire must first be deleted so that we don't return
6595          * the existing one on the attempt to add the unbound new_ire.
6596          */
6597         ire_delete(ire);
6598         new_ire = ire_add(new_ire);
6599         if (new_ire != NULL)
6600                 ire_refrele(new_ire);
6601 }
6602 
6603 /*
6604  * When the settings of ip*_strict_src_multihoming tunables are changed,
6605  * all cached routes need to be recomputed. This recomputation needs to be
6606  * done when going from weaker to stronger modes so that the cached ire
6607  * for the connection does not violate the current ip*_strict_src_multihoming
6608  * setting. It also needs to be done when going from stronger to weaker modes,
6609  * so that we fall back to matching on the longest-matching-route (as opposed
6610  * to a shorter match that may have been selected in the strong mode
6611  * to satisfy src_multihoming settings).
6612  *
6613  * The cached ixa_ire entires for all conn_t entries are marked as
6614  * "verify" so that they will be recomputed for the next packet.
6615  */
6616 void
6617 conn_ire_revalidate(conn_t *connp, void *arg)
6618 {
6619         boolean_t isv6 = (boolean_t)arg;
6620 
6621         if ((isv6 && connp->conn_ipversion != IPV6_VERSION) ||
6622             (!isv6 && connp->conn_ipversion != IPV4_VERSION))
6623                 return;
6624         connp->conn_ixa->ixa_ire_generation = IRE_GENERATION_VERIFY;
6625 }
6626 
6627 /*
6628  * Handles both IPv4 and IPv6 reassembly - doing the out-of-order cases,
6629  * When an ipf is passed here for the first time, if
6630  * we already have in-order fragments on the queue, we convert from the fast-
6631  * path reassembly scheme to the hard-case scheme.  From then on, additional
6632  * fragments are reassembled here.  We keep track of the start and end offsets
6633  * of each piece, and the number of holes in the chain.  When the hole count
6634  * goes to zero, we are done!
6635  *
6636  * The ipf_count will be updated to account for any mblk(s) added (pointed to
6637  * by mp) or subtracted (freeb()ed dups), upon return the caller must update
6638  * ipfb_count and ill_frag_count by the difference of ipf_count before and
6639  * after the call to ip_reassemble().
6640  */
6641 int
6642 ip_reassemble(mblk_t *mp, ipf_t *ipf, uint_t start, boolean_t more, ill_t *ill,
6643     size_t msg_len)
6644 {
6645         uint_t  end;
6646         mblk_t  *next_mp;
6647         mblk_t  *mp1;
6648         uint_t  offset;
6649         boolean_t incr_dups = B_TRUE;
6650         boolean_t offset_zero_seen = B_FALSE;
6651         boolean_t pkt_boundary_checked = B_FALSE;
6652 
6653         /* If start == 0 then ipf_nf_hdr_len has to be set. */
6654         ASSERT(start != 0 || ipf->ipf_nf_hdr_len != 0);
6655 
6656         /* Add in byte count */
6657         ipf->ipf_count += msg_len;
6658         if (ipf->ipf_end) {
6659                 /*
6660                  * We were part way through in-order reassembly, but now there
6661                  * is a hole.  We walk through messages already queued, and
6662                  * mark them for hard case reassembly.  We know that up till
6663                  * now they were in order starting from offset zero.
6664                  */
6665                 offset = 0;
6666                 for (mp1 = ipf->ipf_mp->b_cont; mp1; mp1 = mp1->b_cont) {
6667                         IP_REASS_SET_START(mp1, offset);
6668                         if (offset == 0) {
6669                                 ASSERT(ipf->ipf_nf_hdr_len != 0);
6670                                 offset = -ipf->ipf_nf_hdr_len;
6671                         }
6672                         offset += mp1->b_wptr - mp1->b_rptr;
6673                         IP_REASS_SET_END(mp1, offset);
6674                 }
6675                 /* One hole at the end. */
6676                 ipf->ipf_hole_cnt = 1;
6677                 /* Brand it as a hard case, forever. */
6678                 ipf->ipf_end = 0;
6679         }
6680         /* Walk through all the new pieces. */
6681         do {
6682                 end = start + (mp->b_wptr - mp->b_rptr);
6683                 /*
6684                  * If start is 0, decrease 'end' only for the first mblk of
6685                  * the fragment. Otherwise 'end' can get wrong value in the
6686                  * second pass of the loop if first mblk is exactly the
6687                  * size of ipf_nf_hdr_len.
6688                  */
6689                 if (start == 0 && !offset_zero_seen) {
6690                         /* First segment */
6691                         ASSERT(ipf->ipf_nf_hdr_len != 0);
6692                         end -= ipf->ipf_nf_hdr_len;
6693                         offset_zero_seen = B_TRUE;
6694                 }
6695                 next_mp = mp->b_cont;
6696                 /*
6697                  * We are checking to see if there is any interesing data
6698                  * to process.  If there isn't and the mblk isn't the
6699                  * one which carries the unfragmentable header then we
6700                  * drop it.  It's possible to have just the unfragmentable
6701                  * header come through without any data.  That needs to be
6702                  * saved.
6703                  *
6704                  * If the assert at the top of this function holds then the
6705                  * term "ipf->ipf_nf_hdr_len != 0" isn't needed.  This code
6706                  * is infrequently traveled enough that the test is left in
6707                  * to protect against future code changes which break that
6708                  * invariant.
6709                  */
6710                 if (start == end && start != 0 && ipf->ipf_nf_hdr_len != 0) {
6711                         /* Empty.  Blast it. */
6712                         IP_REASS_SET_START(mp, 0);
6713                         IP_REASS_SET_END(mp, 0);
6714                         /*
6715                          * If the ipf points to the mblk we are about to free,
6716                          * update ipf to point to the next mblk (or NULL
6717                          * if none).
6718                          */
6719                         if (ipf->ipf_mp->b_cont == mp)
6720                                 ipf->ipf_mp->b_cont = next_mp;
6721                         freeb(mp);
6722                         continue;
6723                 }
6724                 mp->b_cont = NULL;
6725                 IP_REASS_SET_START(mp, start);
6726                 IP_REASS_SET_END(mp, end);
6727                 if (!ipf->ipf_tail_mp) {
6728                         ipf->ipf_tail_mp = mp;
6729                         ipf->ipf_mp->b_cont = mp;
6730                         if (start == 0 || !more) {
6731                                 ipf->ipf_hole_cnt = 1;
6732                                 /*
6733                                  * if the first fragment comes in more than one
6734                                  * mblk, this loop will be executed for each
6735                                  * mblk. Need to adjust hole count so exiting
6736                                  * this routine will leave hole count at 1.
6737                                  */
6738                                 if (next_mp)
6739                                         ipf->ipf_hole_cnt++;
6740                         } else
6741                                 ipf->ipf_hole_cnt = 2;
6742                         continue;
6743                 } else if (ipf->ipf_last_frag_seen && !more &&
6744                     !pkt_boundary_checked) {
6745                         /*
6746                          * We check datagram boundary only if this fragment
6747                          * claims to be the last fragment and we have seen a
6748                          * last fragment in the past too. We do this only
6749                          * once for a given fragment.
6750                          *
6751                          * start cannot be 0 here as fragments with start=0
6752                          * and MF=0 gets handled as a complete packet. These
6753                          * fragments should not reach here.
6754                          */
6755 
6756                         if (start + msgdsize(mp) !=
6757                             IP_REASS_END(ipf->ipf_tail_mp)) {
6758                                 /*
6759                                  * We have two fragments both of which claim
6760                                  * to be the last fragment but gives conflicting
6761                                  * information about the whole datagram size.
6762                                  * Something fishy is going on. Drop the
6763                                  * fragment and free up the reassembly list.
6764                                  */
6765                                 return (IP_REASS_FAILED);
6766                         }
6767 
6768                         /*
6769                          * We shouldn't come to this code block again for this
6770                          * particular fragment.
6771                          */
6772                         pkt_boundary_checked = B_TRUE;
6773                 }
6774 
6775                 /* New stuff at or beyond tail? */
6776                 offset = IP_REASS_END(ipf->ipf_tail_mp);
6777                 if (start >= offset) {
6778                         if (ipf->ipf_last_frag_seen) {
6779                                 /* current fragment is beyond last fragment */
6780                                 return (IP_REASS_FAILED);
6781                         }
6782                         /* Link it on end. */
6783                         ipf->ipf_tail_mp->b_cont = mp;
6784                         ipf->ipf_tail_mp = mp;
6785                         if (more) {
6786                                 if (start != offset)
6787                                         ipf->ipf_hole_cnt++;
6788                         } else if (start == offset && next_mp == NULL)
6789                                         ipf->ipf_hole_cnt--;
6790                         continue;
6791                 }
6792                 mp1 = ipf->ipf_mp->b_cont;
6793                 offset = IP_REASS_START(mp1);
6794                 /* New stuff at the front? */
6795                 if (start < offset) {
6796                         if (start == 0) {
6797                                 if (end >= offset) {
6798                                         /* Nailed the hole at the begining. */
6799                                         ipf->ipf_hole_cnt--;
6800                                 }
6801                         } else if (end < offset) {
6802                                 /*
6803                                  * A hole, stuff, and a hole where there used
6804                                  * to be just a hole.
6805                                  */
6806                                 ipf->ipf_hole_cnt++;
6807                         }
6808                         mp->b_cont = mp1;
6809                         /* Check for overlap. */
6810                         while (end > offset) {
6811                                 if (end < IP_REASS_END(mp1)) {
6812                                         mp->b_wptr -= end - offset;
6813                                         IP_REASS_SET_END(mp, offset);
6814                                         BUMP_MIB(ill->ill_ip_mib,
6815                                             ipIfStatsReasmPartDups);
6816                                         break;
6817                                 }
6818                                 /* Did we cover another hole? */
6819                                 if ((mp1->b_cont &&
6820                                     IP_REASS_END(mp1) !=
6821                                     IP_REASS_START(mp1->b_cont) &&
6822                                     end >= IP_REASS_START(mp1->b_cont)) ||
6823                                     (!ipf->ipf_last_frag_seen && !more)) {
6824                                         ipf->ipf_hole_cnt--;
6825                                 }
6826                                 /* Clip out mp1. */
6827                                 if ((mp->b_cont = mp1->b_cont) == NULL) {
6828                                         /*
6829                                          * After clipping out mp1, this guy
6830                                          * is now hanging off the end.
6831                                          */
6832                                         ipf->ipf_tail_mp = mp;
6833                                 }
6834                                 IP_REASS_SET_START(mp1, 0);
6835                                 IP_REASS_SET_END(mp1, 0);
6836                                 /* Subtract byte count */
6837                                 ipf->ipf_count -= mp1->b_datap->db_lim -
6838                                     mp1->b_datap->db_base;
6839                                 freeb(mp1);
6840                                 BUMP_MIB(ill->ill_ip_mib,
6841                                     ipIfStatsReasmPartDups);
6842                                 mp1 = mp->b_cont;
6843                                 if (!mp1)
6844                                         break;
6845                                 offset = IP_REASS_START(mp1);
6846                         }
6847                         ipf->ipf_mp->b_cont = mp;
6848                         continue;
6849                 }
6850                 /*
6851                  * The new piece starts somewhere between the start of the head
6852                  * and before the end of the tail.
6853                  */
6854                 for (; mp1; mp1 = mp1->b_cont) {
6855                         offset = IP_REASS_END(mp1);
6856                         if (start < offset) {
6857                                 if (end <= offset) {
6858                                         /* Nothing new. */
6859                                         IP_REASS_SET_START(mp, 0);
6860                                         IP_REASS_SET_END(mp, 0);
6861                                         /* Subtract byte count */
6862                                         ipf->ipf_count -= mp->b_datap->db_lim -
6863                                             mp->b_datap->db_base;
6864                                         if (incr_dups) {
6865                                                 ipf->ipf_num_dups++;
6866                                                 incr_dups = B_FALSE;
6867                                         }
6868                                         freeb(mp);
6869                                         BUMP_MIB(ill->ill_ip_mib,
6870                                             ipIfStatsReasmDuplicates);
6871                                         break;
6872                                 }
6873                                 /*
6874                                  * Trim redundant stuff off beginning of new
6875                                  * piece.
6876                                  */
6877                                 IP_REASS_SET_START(mp, offset);
6878                                 mp->b_rptr += offset - start;
6879                                 BUMP_MIB(ill->ill_ip_mib,
6880                                     ipIfStatsReasmPartDups);
6881                                 start = offset;
6882                                 if (!mp1->b_cont) {
6883                                         /*
6884                                          * After trimming, this guy is now
6885                                          * hanging off the end.
6886                                          */
6887                                         mp1->b_cont = mp;
6888                                         ipf->ipf_tail_mp = mp;
6889                                         if (!more) {
6890                                                 ipf->ipf_hole_cnt--;
6891                                         }
6892                                         break;
6893                                 }
6894                         }
6895                         if (start >= IP_REASS_START(mp1->b_cont))
6896                                 continue;
6897                         /* Fill a hole */
6898                         if (start > offset)
6899                                 ipf->ipf_hole_cnt++;
6900                         mp->b_cont = mp1->b_cont;
6901                         mp1->b_cont = mp;
6902                         mp1 = mp->b_cont;
6903                         offset = IP_REASS_START(mp1);
6904                         if (end >= offset) {
6905                                 ipf->ipf_hole_cnt--;
6906                                 /* Check for overlap. */
6907                                 while (end > offset) {
6908                                         if (end < IP_REASS_END(mp1)) {
6909                                                 mp->b_wptr -= end - offset;
6910                                                 IP_REASS_SET_END(mp, offset);
6911                                                 /*
6912                                                  * TODO we might bump
6913                                                  * this up twice if there is
6914                                                  * overlap at both ends.
6915                                                  */
6916                                                 BUMP_MIB(ill->ill_ip_mib,
6917                                                     ipIfStatsReasmPartDups);
6918                                                 break;
6919                                         }
6920                                         /* Did we cover another hole? */
6921                                         if ((mp1->b_cont &&
6922                                             IP_REASS_END(mp1)
6923                                             != IP_REASS_START(mp1->b_cont) &&
6924                                             end >=
6925                                             IP_REASS_START(mp1->b_cont)) ||
6926                                             (!ipf->ipf_last_frag_seen &&
6927                                             !more)) {
6928                                                 ipf->ipf_hole_cnt--;
6929                                         }
6930                                         /* Clip out mp1. */
6931                                         if ((mp->b_cont = mp1->b_cont) ==
6932                                             NULL) {
6933                                                 /*
6934                                                  * After clipping out mp1,
6935                                                  * this guy is now hanging
6936                                                  * off the end.
6937                                                  */
6938                                                 ipf->ipf_tail_mp = mp;
6939                                         }
6940                                         IP_REASS_SET_START(mp1, 0);
6941                                         IP_REASS_SET_END(mp1, 0);
6942                                         /* Subtract byte count */
6943                                         ipf->ipf_count -=
6944                                             mp1->b_datap->db_lim -
6945                                             mp1->b_datap->db_base;
6946                                         freeb(mp1);
6947                                         BUMP_MIB(ill->ill_ip_mib,
6948                                             ipIfStatsReasmPartDups);
6949                                         mp1 = mp->b_cont;
6950                                         if (!mp1)
6951                                                 break;
6952                                         offset = IP_REASS_START(mp1);
6953                                 }
6954                         }
6955                         break;
6956                 }
6957         } while (start = end, mp = next_mp);
6958 
6959         /* Fragment just processed could be the last one. Remember this fact */
6960         if (!more)
6961                 ipf->ipf_last_frag_seen = B_TRUE;
6962 
6963         /* Still got holes? */
6964         if (ipf->ipf_hole_cnt)
6965                 return (IP_REASS_PARTIAL);
6966         /* Clean up overloaded fields to avoid upstream disasters. */
6967         for (mp1 = ipf->ipf_mp->b_cont; mp1; mp1 = mp1->b_cont) {
6968                 IP_REASS_SET_START(mp1, 0);
6969                 IP_REASS_SET_END(mp1, 0);
6970         }
6971         return (IP_REASS_COMPLETE);
6972 }
6973 
6974 /*
6975  * Fragmentation reassembly.  Each ILL has a hash table for
6976  * queuing packets undergoing reassembly for all IPIFs
6977  * associated with the ILL.  The hash is based on the packet
6978  * IP ident field.  The ILL frag hash table was allocated
6979  * as a timer block at the time the ILL was created.  Whenever
6980  * there is anything on the reassembly queue, the timer will
6981  * be running.  Returns the reassembled packet if reassembly completes.
6982  */
6983 mblk_t *
6984 ip_input_fragment(mblk_t *mp, ipha_t *ipha, ip_recv_attr_t *ira)
6985 {
6986         uint32_t        frag_offset_flags;
6987         mblk_t          *t_mp;
6988         ipaddr_t        dst;
6989         uint8_t         proto = ipha->ipha_protocol;
6990         uint32_t        sum_val;
6991         uint16_t        sum_flags;
6992         ipf_t           *ipf;
6993         ipf_t           **ipfp;
6994         ipfb_t          *ipfb;
6995         uint16_t        ident;
6996         uint32_t        offset;
6997         ipaddr_t        src;
6998         uint_t          hdr_length;
6999         uint32_t        end;
7000         mblk_t          *mp1;
7001         mblk_t          *tail_mp;
7002         size_t          count;
7003         size_t          msg_len;
7004         uint8_t         ecn_info = 0;
7005         uint32_t        packet_size;
7006         boolean_t       pruned = B_FALSE;
7007         ill_t           *ill = ira->ira_ill;
7008         ip_stack_t      *ipst = ill->ill_ipst;
7009 
7010         /*
7011          * Drop the fragmented as early as possible, if
7012          * we don't have resource(s) to re-assemble.
7013          */
7014         if (ipst->ips_ip_reass_queue_bytes == 0) {
7015                 freemsg(mp);
7016                 return (NULL);
7017         }
7018 
7019         /* Check for fragmentation offset; return if there's none */
7020         if ((frag_offset_flags = ntohs(ipha->ipha_fragment_offset_and_flags) &
7021             (IPH_MF | IPH_OFFSET)) == 0)
7022                 return (mp);
7023 
7024         /*
7025          * We utilize hardware computed checksum info only for UDP since
7026          * IP fragmentation is a normal occurrence for the protocol.  In
7027          * addition, checksum offload support for IP fragments carrying
7028          * UDP payload is commonly implemented across network adapters.
7029          */
7030         ASSERT(ira->ira_rill != NULL);
7031         if (proto == IPPROTO_UDP && dohwcksum &&
7032             ILL_HCKSUM_CAPABLE(ira->ira_rill) &&
7033             (DB_CKSUMFLAGS(mp) & (HCK_FULLCKSUM | HCK_PARTIALCKSUM))) {
7034                 mblk_t *mp1 = mp->b_cont;
7035                 int32_t len;
7036 
7037                 /* Record checksum information from the packet */
7038                 sum_val = (uint32_t)DB_CKSUM16(mp);
7039                 sum_flags = DB_CKSUMFLAGS(mp);
7040 
7041                 /* IP payload offset from beginning of mblk */
7042                 offset = ((uchar_t *)ipha + IPH_HDR_LENGTH(ipha)) - mp->b_rptr;
7043 
7044                 if ((sum_flags & HCK_PARTIALCKSUM) &&
7045                     (mp1 == NULL || mp1->b_cont == NULL) &&
7046                     offset >= DB_CKSUMSTART(mp) &&
7047                     ((len = offset - DB_CKSUMSTART(mp)) & 1) == 0) {
7048                         uint32_t adj;
7049                         /*
7050                          * Partial checksum has been calculated by hardware
7051                          * and attached to the packet; in addition, any
7052                          * prepended extraneous data is even byte aligned.
7053                          * If any such data exists, we adjust the checksum;
7054                          * this would also handle any postpended data.
7055                          */
7056                         IP_ADJCKSUM_PARTIAL(mp->b_rptr + DB_CKSUMSTART(mp),
7057                             mp, mp1, len, adj);
7058 
7059                         /* One's complement subtract extraneous checksum */
7060                         if (adj >= sum_val)
7061                                 sum_val = ~(adj - sum_val) & 0xFFFF;
7062                         else
7063                                 sum_val -= adj;
7064                 }
7065         } else {
7066                 sum_val = 0;
7067                 sum_flags = 0;
7068         }
7069 
7070         /* Clear hardware checksumming flag */
7071         DB_CKSUMFLAGS(mp) = 0;
7072 
7073         ident = ipha->ipha_ident;
7074         offset = (frag_offset_flags << 3) & 0xFFFF;
7075         src = ipha->ipha_src;
7076         dst = ipha->ipha_dst;
7077         hdr_length = IPH_HDR_LENGTH(ipha);
7078         end = ntohs(ipha->ipha_length) - hdr_length;
7079 
7080         /* If end == 0 then we have a packet with no data, so just free it */
7081         if (end == 0) {
7082                 freemsg(mp);
7083                 return (NULL);
7084         }
7085 
7086         /* Record the ECN field info. */
7087         ecn_info = (ipha->ipha_type_of_service & 0x3);
7088         if (offset != 0) {
7089                 /*
7090                  * If this isn't the first piece, strip the header, and
7091                  * add the offset to the end value.
7092                  */
7093                 mp->b_rptr += hdr_length;
7094                 end += offset;
7095         }
7096 
7097         /* Handle vnic loopback of fragments */
7098         if (mp->b_datap->db_ref > 2)
7099                 msg_len = 0;
7100         else
7101                 msg_len = MBLKSIZE(mp);
7102 
7103         tail_mp = mp;
7104         while (tail_mp->b_cont != NULL) {
7105                 tail_mp = tail_mp->b_cont;
7106                 if (tail_mp->b_datap->db_ref <= 2)
7107                         msg_len += MBLKSIZE(tail_mp);
7108         }
7109 
7110         /* If the reassembly list for this ILL will get too big, prune it */
7111         if ((msg_len + sizeof (*ipf) + ill->ill_frag_count) >=
7112             ipst->ips_ip_reass_queue_bytes) {
7113                 DTRACE_PROBE3(ip_reass_queue_bytes, uint_t, msg_len,
7114                     uint_t, ill->ill_frag_count,
7115                     uint_t, ipst->ips_ip_reass_queue_bytes);
7116                 ill_frag_prune(ill,
7117                     (ipst->ips_ip_reass_queue_bytes < msg_len) ? 0 :
7118                     (ipst->ips_ip_reass_queue_bytes - msg_len));
7119                 pruned = B_TRUE;
7120         }
7121 
7122         ipfb = &ill->ill_frag_hash_tbl[ILL_FRAG_HASH(src, ident)];
7123         mutex_enter(&ipfb->ipfb_lock);
7124 
7125         ipfp = &ipfb->ipfb_ipf;
7126         /* Try to find an existing fragment queue for this packet. */
7127         for (;;) {
7128                 ipf = ipfp[0];
7129                 if (ipf != NULL) {
7130                         /*
7131                          * It has to match on ident and src/dst address.
7132                          */
7133                         if (ipf->ipf_ident == ident &&
7134                             ipf->ipf_src == src &&
7135                             ipf->ipf_dst == dst &&
7136                             ipf->ipf_protocol == proto) {
7137                                 /*
7138                                  * If we have received too many
7139                                  * duplicate fragments for this packet
7140                                  * free it.
7141                                  */
7142                                 if (ipf->ipf_num_dups > ip_max_frag_dups) {
7143                                         ill_frag_free_pkts(ill, ipfb, ipf, 1);
7144                                         freemsg(mp);
7145                                         mutex_exit(&ipfb->ipfb_lock);
7146                                         return (NULL);
7147                                 }
7148                                 /* Found it. */
7149                                 break;
7150                         }
7151                         ipfp = &ipf->ipf_hash_next;
7152                         continue;
7153                 }
7154 
7155                 /*
7156                  * If we pruned the list, do we want to store this new
7157                  * fragment?. We apply an optimization here based on the
7158                  * fact that most fragments will be received in order.
7159                  * So if the offset of this incoming fragment is zero,
7160                  * it is the first fragment of a new packet. We will
7161                  * keep it.  Otherwise drop the fragment, as we have
7162                  * probably pruned the packet already (since the
7163                  * packet cannot be found).
7164                  */
7165                 if (pruned && offset != 0) {
7166                         mutex_exit(&ipfb->ipfb_lock);
7167                         freemsg(mp);
7168                         return (NULL);
7169                 }
7170 
7171                 if (ipfb->ipfb_frag_pkts >= MAX_FRAG_PKTS(ipst))  {
7172                         /*
7173                          * Too many fragmented packets in this hash
7174                          * bucket. Free the oldest.
7175                          */
7176                         ill_frag_free_pkts(ill, ipfb, ipfb->ipfb_ipf, 1);
7177                 }
7178 
7179                 /* New guy.  Allocate a frag message. */
7180                 mp1 = allocb(sizeof (*ipf), BPRI_MED);
7181                 if (mp1 == NULL) {
7182                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
7183                         ip_drop_input("ipIfStatsInDiscards", mp, ill);
7184                         freemsg(mp);
7185 reass_done:
7186                         mutex_exit(&ipfb->ipfb_lock);
7187                         return (NULL);
7188                 }
7189 
7190                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsReasmReqds);
7191                 mp1->b_cont = mp;
7192 
7193                 /* Initialize the fragment header. */
7194                 ipf = (ipf_t *)mp1->b_rptr;
7195                 ipf->ipf_mp = mp1;
7196                 ipf->ipf_ptphn = ipfp;
7197                 ipfp[0] = ipf;
7198                 ipf->ipf_hash_next = NULL;
7199                 ipf->ipf_ident = ident;
7200                 ipf->ipf_protocol = proto;
7201                 ipf->ipf_src = src;
7202                 ipf->ipf_dst = dst;
7203                 ipf->ipf_nf_hdr_len = 0;
7204                 /* Record reassembly start time. */
7205                 ipf->ipf_timestamp = gethrestime_sec();
7206                 /* Record ipf generation and account for frag header */
7207                 ipf->ipf_gen = ill->ill_ipf_gen++;
7208                 ipf->ipf_count = MBLKSIZE(mp1);
7209                 ipf->ipf_last_frag_seen = B_FALSE;
7210                 ipf->ipf_ecn = ecn_info;
7211                 ipf->ipf_num_dups = 0;
7212                 ipfb->ipfb_frag_pkts++;
7213                 ipf->ipf_checksum = 0;
7214                 ipf->ipf_checksum_flags = 0;
7215 
7216                 /* Store checksum value in fragment header */
7217                 if (sum_flags != 0) {
7218                         sum_val = (sum_val & 0xFFFF) + (sum_val >> 16);
7219                         sum_val = (sum_val & 0xFFFF) + (sum_val >> 16);
7220                         ipf->ipf_checksum = sum_val;
7221                         ipf->ipf_checksum_flags = sum_flags;
7222                 }
7223 
7224                 /*
7225                  * We handle reassembly two ways.  In the easy case,
7226                  * where all the fragments show up in order, we do
7227                  * minimal bookkeeping, and just clip new pieces on
7228                  * the end.  If we ever see a hole, then we go off
7229                  * to ip_reassemble which has to mark the pieces and
7230                  * keep track of the number of holes, etc.  Obviously,
7231                  * the point of having both mechanisms is so we can
7232                  * handle the easy case as efficiently as possible.
7233                  */
7234                 if (offset == 0) {
7235                         /* Easy case, in-order reassembly so far. */
7236                         ipf->ipf_count += msg_len;
7237                         ipf->ipf_tail_mp = tail_mp;
7238                         /*
7239                          * Keep track of next expected offset in
7240                          * ipf_end.
7241                          */
7242                         ipf->ipf_end = end;
7243                         ipf->ipf_nf_hdr_len = hdr_length;
7244                 } else {
7245                         /* Hard case, hole at the beginning. */
7246                         ipf->ipf_tail_mp = NULL;
7247                         /*
7248                          * ipf_end == 0 means that we have given up
7249                          * on easy reassembly.
7250                          */
7251                         ipf->ipf_end = 0;
7252 
7253                         /* Forget checksum offload from now on */
7254                         ipf->ipf_checksum_flags = 0;
7255 
7256                         /*
7257                          * ipf_hole_cnt is set by ip_reassemble.
7258                          * ipf_count is updated by ip_reassemble.
7259                          * No need to check for return value here
7260                          * as we don't expect reassembly to complete
7261                          * or fail for the first fragment itself.
7262                          */
7263                         (void) ip_reassemble(mp, ipf,
7264                             (frag_offset_flags & IPH_OFFSET) << 3,
7265                             (frag_offset_flags & IPH_MF), ill, msg_len);
7266                 }
7267                 /* Update per ipfb and ill byte counts */
7268                 ipfb->ipfb_count += ipf->ipf_count;
7269                 ASSERT(ipfb->ipfb_count > 0);     /* Wraparound */
7270                 atomic_add_32(&ill->ill_frag_count, ipf->ipf_count);
7271                 /* If the frag timer wasn't already going, start it. */
7272                 mutex_enter(&ill->ill_lock);
7273                 ill_frag_timer_start(ill);
7274                 mutex_exit(&ill->ill_lock);
7275                 goto reass_done;
7276         }
7277 
7278         /*
7279          * If the packet's flag has changed (it could be coming up
7280          * from an interface different than the previous, therefore
7281          * possibly different checksum capability), then forget about
7282          * any stored checksum states.  Otherwise add the value to
7283          * the existing one stored in the fragment header.
7284          */
7285         if (sum_flags != 0 && sum_flags == ipf->ipf_checksum_flags) {
7286                 sum_val += ipf->ipf_checksum;
7287                 sum_val = (sum_val & 0xFFFF) + (sum_val >> 16);
7288                 sum_val = (sum_val & 0xFFFF) + (sum_val >> 16);
7289                 ipf->ipf_checksum = sum_val;
7290         } else if (ipf->ipf_checksum_flags != 0) {
7291                 /* Forget checksum offload from now on */
7292                 ipf->ipf_checksum_flags = 0;
7293         }
7294 
7295         /*
7296          * We have a new piece of a datagram which is already being
7297          * reassembled.  Update the ECN info if all IP fragments
7298          * are ECN capable.  If there is one which is not, clear
7299          * all the info.  If there is at least one which has CE
7300          * code point, IP needs to report that up to transport.
7301          */
7302         if (ecn_info != IPH_ECN_NECT && ipf->ipf_ecn != IPH_ECN_NECT) {
7303                 if (ecn_info == IPH_ECN_CE)
7304                         ipf->ipf_ecn = IPH_ECN_CE;
7305         } else {
7306                 ipf->ipf_ecn = IPH_ECN_NECT;
7307         }
7308         if (offset && ipf->ipf_end == offset) {
7309                 /* The new fragment fits at the end */
7310                 ipf->ipf_tail_mp->b_cont = mp;
7311                 /* Update the byte count */
7312                 ipf->ipf_count += msg_len;
7313                 /* Update per ipfb and ill byte counts */
7314                 ipfb->ipfb_count += msg_len;
7315                 ASSERT(ipfb->ipfb_count > 0);     /* Wraparound */
7316                 atomic_add_32(&ill->ill_frag_count, msg_len);
7317                 if (frag_offset_flags & IPH_MF) {
7318                         /* More to come. */
7319                         ipf->ipf_end = end;
7320                         ipf->ipf_tail_mp = tail_mp;
7321                         goto reass_done;
7322                 }
7323         } else {
7324                 /* Go do the hard cases. */
7325                 int ret;
7326 
7327                 if (offset == 0)
7328                         ipf->ipf_nf_hdr_len = hdr_length;
7329 
7330                 /* Save current byte count */
7331                 count = ipf->ipf_count;
7332                 ret = ip_reassemble(mp, ipf,
7333                     (frag_offset_flags & IPH_OFFSET) << 3,
7334                     (frag_offset_flags & IPH_MF), ill, msg_len);
7335                 /* Count of bytes added and subtracted (freeb()ed) */
7336                 count = ipf->ipf_count - count;
7337                 if (count) {
7338                         /* Update per ipfb and ill byte counts */
7339                         ipfb->ipfb_count += count;
7340                         ASSERT(ipfb->ipfb_count > 0); /* Wraparound */
7341                         atomic_add_32(&ill->ill_frag_count, count);
7342                 }
7343                 if (ret == IP_REASS_PARTIAL) {
7344                         goto reass_done;
7345                 } else if (ret == IP_REASS_FAILED) {
7346                         /* Reassembly failed. Free up all resources */
7347                         ill_frag_free_pkts(ill, ipfb, ipf, 1);
7348                         for (t_mp = mp; t_mp != NULL; t_mp = t_mp->b_cont) {
7349                                 IP_REASS_SET_START(t_mp, 0);
7350                                 IP_REASS_SET_END(t_mp, 0);
7351                         }
7352                         freemsg(mp);
7353                         goto reass_done;
7354                 }
7355                 /* We will reach here iff 'ret' is IP_REASS_COMPLETE */
7356         }
7357         /*
7358          * We have completed reassembly.  Unhook the frag header from
7359          * the reassembly list.
7360          *
7361          * Before we free the frag header, record the ECN info
7362          * to report back to the transport.
7363          */
7364         ecn_info = ipf->ipf_ecn;
7365         BUMP_MIB(ill->ill_ip_mib, ipIfStatsReasmOKs);
7366         ipfp = ipf->ipf_ptphn;
7367 
7368         /* We need to supply these to caller */
7369         if ((sum_flags = ipf->ipf_checksum_flags) != 0)
7370                 sum_val = ipf->ipf_checksum;
7371         else
7372                 sum_val = 0;
7373 
7374         mp1 = ipf->ipf_mp;
7375         count = ipf->ipf_count;
7376         ipf = ipf->ipf_hash_next;
7377         if (ipf != NULL)
7378                 ipf->ipf_ptphn = ipfp;
7379         ipfp[0] = ipf;
7380         atomic_add_32(&ill->ill_frag_count, -count);
7381         ASSERT(ipfb->ipfb_count >= count);
7382         ipfb->ipfb_count -= count;
7383         ipfb->ipfb_frag_pkts--;
7384         mutex_exit(&ipfb->ipfb_lock);
7385         /* Ditch the frag header. */
7386         mp = mp1->b_cont;
7387 
7388         freeb(mp1);
7389 
7390         /* Restore original IP length in header. */
7391         packet_size = (uint32_t)msgdsize(mp);
7392         if (packet_size > IP_MAXPACKET) {
7393                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7394                 ip_drop_input("Reassembled packet too large", mp, ill);
7395                 freemsg(mp);
7396                 return (NULL);
7397         }
7398 
7399         if (DB_REF(mp) > 1) {
7400                 mblk_t *mp2 = copymsg(mp);
7401 
7402                 if (mp2 == NULL) {
7403                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
7404                         ip_drop_input("ipIfStatsInDiscards", mp, ill);
7405                         freemsg(mp);
7406                         return (NULL);
7407                 }
7408                 freemsg(mp);
7409                 mp = mp2;
7410         }
7411         ipha = (ipha_t *)mp->b_rptr;
7412 
7413         ipha->ipha_length = htons((uint16_t)packet_size);
7414         /* We're now complete, zip the frag state */
7415         ipha->ipha_fragment_offset_and_flags = 0;
7416         /* Record the ECN info. */
7417         ipha->ipha_type_of_service &= 0xFC;
7418         ipha->ipha_type_of_service |= ecn_info;
7419 
7420         /* Update the receive attributes */
7421         ira->ira_pktlen = packet_size;
7422         ira->ira_ip_hdr_length = IPH_HDR_LENGTH(ipha);
7423 
7424         /* Reassembly is successful; set checksum information in packet */
7425         DB_CKSUM16(mp) = (uint16_t)sum_val;
7426         DB_CKSUMFLAGS(mp) = sum_flags;
7427         DB_CKSUMSTART(mp) = ira->ira_ip_hdr_length;
7428 
7429         return (mp);
7430 }
7431 
7432 /*
7433  * Pullup function that should be used for IP input in order to
7434  * ensure we do not loose the L2 source address; we need the l2 source
7435  * address for IP_RECVSLLA and for ndp_input.
7436  *
7437  * We return either NULL or b_rptr.
7438  */
7439 void *
7440 ip_pullup(mblk_t *mp, ssize_t len, ip_recv_attr_t *ira)
7441 {
7442         ill_t           *ill = ira->ira_ill;
7443 
7444         if (ip_rput_pullups++ == 0) {
7445                 (void) mi_strlog(ill->ill_rq, 1, SL_ERROR|SL_TRACE,
7446                     "ip_pullup: %s forced us to "
7447                     " pullup pkt, hdr len %ld, hdr addr %p",
7448                     ill->ill_name, len, (void *)mp->b_rptr);
7449         }
7450         if (!(ira->ira_flags & IRAF_L2SRC_SET))
7451                 ip_setl2src(mp, ira, ira->ira_rill);
7452         ASSERT(ira->ira_flags & IRAF_L2SRC_SET);
7453         if (!pullupmsg(mp, len))
7454                 return (NULL);
7455         else
7456                 return (mp->b_rptr);
7457 }
7458 
7459 /*
7460  * Make sure ira_l2src has an address. If we don't have one fill with zeros.
7461  * When called from the ULP ira_rill will be NULL hence the caller has to
7462  * pass in the ill.
7463  */
7464 /* ARGSUSED */
7465 void
7466 ip_setl2src(mblk_t *mp, ip_recv_attr_t *ira, ill_t *ill)
7467 {
7468         const uchar_t *addr;
7469         int alen;
7470 
7471         if (ira->ira_flags & IRAF_L2SRC_SET)
7472                 return;
7473 
7474         ASSERT(ill != NULL);
7475         alen = ill->ill_phys_addr_length;
7476         ASSERT(alen <= sizeof (ira->ira_l2src));
7477         if (ira->ira_mhip != NULL &&
7478             (addr = ira->ira_mhip->mhi_saddr) != NULL) {
7479                 bcopy(addr, ira->ira_l2src, alen);
7480         } else if ((ira->ira_flags & IRAF_L2SRC_LOOPBACK) &&
7481             (addr = ill->ill_phys_addr) != NULL) {
7482                 bcopy(addr, ira->ira_l2src, alen);
7483         } else {
7484                 bzero(ira->ira_l2src, alen);
7485         }
7486         ira->ira_flags |= IRAF_L2SRC_SET;
7487 }
7488 
7489 /*
7490  * check ip header length and align it.
7491  */
7492 mblk_t *
7493 ip_check_and_align_header(mblk_t *mp, uint_t min_size, ip_recv_attr_t *ira)
7494 {
7495         ill_t   *ill = ira->ira_ill;
7496         ssize_t len;
7497 
7498         len = MBLKL(mp);
7499 
7500         if (!OK_32PTR(mp->b_rptr))
7501                 IP_STAT(ill->ill_ipst, ip_notaligned);
7502         else
7503                 IP_STAT(ill->ill_ipst, ip_recv_pullup);
7504 
7505         /* Guard against bogus device drivers */
7506         if (len < 0) {
7507                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7508                 ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7509                 freemsg(mp);
7510                 return (NULL);
7511         }
7512 
7513         if (len == 0) {
7514                 /* GLD sometimes sends up mblk with b_rptr == b_wptr! */
7515                 mblk_t *mp1 = mp->b_cont;
7516 
7517                 if (!(ira->ira_flags & IRAF_L2SRC_SET))
7518                         ip_setl2src(mp, ira, ira->ira_rill);
7519                 ASSERT(ira->ira_flags & IRAF_L2SRC_SET);
7520 
7521                 freeb(mp);
7522                 mp = mp1;
7523                 if (mp == NULL)
7524                         return (NULL);
7525 
7526                 if (OK_32PTR(mp->b_rptr) && MBLKL(mp) >= min_size)
7527                         return (mp);
7528         }
7529         if (ip_pullup(mp, min_size, ira) == NULL) {
7530                 if (msgdsize(mp) < min_size) {
7531                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7532                         ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7533                 } else {
7534                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
7535                         ip_drop_input("ipIfStatsInDiscards", mp, ill);
7536                 }
7537                 freemsg(mp);
7538                 return (NULL);
7539         }
7540         return (mp);
7541 }
7542 
7543 /*
7544  * Common code for IPv4 and IPv6 to check and pullup multi-mblks
7545  */
7546 mblk_t *
7547 ip_check_length(mblk_t *mp, uchar_t *rptr, ssize_t len, uint_t pkt_len,
7548     uint_t min_size, ip_recv_attr_t *ira)
7549 {
7550         ill_t   *ill = ira->ira_ill;
7551 
7552         /*
7553          * Make sure we have data length consistent
7554          * with the IP header.
7555          */
7556         if (mp->b_cont == NULL) {
7557                 /* pkt_len is based on ipha_len, not the mblk length */
7558                 if (pkt_len < min_size) {
7559                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7560                         ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7561                         freemsg(mp);
7562                         return (NULL);
7563                 }
7564                 if (len < 0) {
7565                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
7566                         ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
7567                         freemsg(mp);
7568                         return (NULL);
7569                 }
7570                 /* Drop any pad */
7571                 mp->b_wptr = rptr + pkt_len;
7572         } else if ((len += msgdsize(mp->b_cont)) != 0) {
7573                 ASSERT(pkt_len >= min_size);
7574                 if (pkt_len < min_size) {
7575                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7576                         ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7577                         freemsg(mp);
7578                         return (NULL);
7579                 }
7580                 if (len < 0) {
7581                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
7582                         ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
7583                         freemsg(mp);
7584                         return (NULL);
7585                 }
7586                 /* Drop any pad */
7587                 (void) adjmsg(mp, -len);
7588                 /*
7589                  * adjmsg may have freed an mblk from the chain, hence
7590                  * invalidate any hw checksum here. This will force IP to
7591                  * calculate the checksum in sw, but only for this packet.
7592                  */
7593                 DB_CKSUMFLAGS(mp) = 0;
7594                 IP_STAT(ill->ill_ipst, ip_multimblk);
7595         }
7596         return (mp);
7597 }
7598 
7599 /*
7600  * Check that the IPv4 opt_len is consistent with the packet and pullup
7601  * the options.
7602  */
7603 mblk_t *
7604 ip_check_optlen(mblk_t *mp, ipha_t *ipha, uint_t opt_len, uint_t pkt_len,
7605     ip_recv_attr_t *ira)
7606 {
7607         ill_t   *ill = ira->ira_ill;
7608         ssize_t len;
7609 
7610         /* Assume no IPv6 packets arrive over the IPv4 queue */
7611         if (IPH_HDR_VERSION(ipha) != IPV4_VERSION) {
7612                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7613                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInWrongIPVersion);
7614                 ip_drop_input("IPvN packet on IPv4 ill", mp, ill);
7615                 freemsg(mp);
7616                 return (NULL);
7617         }
7618 
7619         if (opt_len > (15 - IP_SIMPLE_HDR_LENGTH_IN_WORDS)) {
7620                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7621                 ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7622                 freemsg(mp);
7623                 return (NULL);
7624         }
7625         /*
7626          * Recompute complete header length and make sure we
7627          * have access to all of it.
7628          */
7629         len = ((size_t)opt_len + IP_SIMPLE_HDR_LENGTH_IN_WORDS) << 2;
7630         if (len > (mp->b_wptr - mp->b_rptr)) {
7631                 if (len > pkt_len) {
7632                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7633                         ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7634                         freemsg(mp);
7635                         return (NULL);
7636                 }
7637                 if (ip_pullup(mp, len, ira) == NULL) {
7638                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
7639                         ip_drop_input("ipIfStatsInDiscards", mp, ill);
7640                         freemsg(mp);
7641                         return (NULL);
7642                 }
7643         }
7644         return (mp);
7645 }
7646 
7647 /*
7648  * Returns a new ire, or the same ire, or NULL.
7649  * If a different IRE is returned, then it is held; the caller
7650  * needs to release it.
7651  * In no case is there any hold/release on the ire argument.
7652  */
7653 ire_t *
7654 ip_check_multihome(void *addr, ire_t *ire, ill_t *ill)
7655 {
7656         ire_t           *new_ire;
7657         ill_t           *ire_ill;
7658         uint_t          ifindex;
7659         ip_stack_t      *ipst = ill->ill_ipst;
7660         boolean_t       strict_check = B_FALSE;
7661 
7662         /*
7663          * IPMP common case: if IRE and ILL are in the same group, there's no
7664          * issue (e.g. packet received on an underlying interface matched an
7665          * IRE_LOCAL on its associated group interface).
7666          */
7667         ASSERT(ire->ire_ill != NULL);
7668         if (IS_IN_SAME_ILLGRP(ill, ire->ire_ill))
7669                 return (ire);
7670 
7671         /*
7672          * Do another ire lookup here, using the ingress ill, to see if the
7673          * interface is in a usesrc group.
7674          * As long as the ills belong to the same group, we don't consider
7675          * them to be arriving on the wrong interface. Thus, if the switch
7676          * is doing inbound load spreading, we won't drop packets when the
7677          * ip*_strict_dst_multihoming switch is on.
7678          * We also need to check for IPIF_UNNUMBERED point2point interfaces
7679          * where the local address may not be unique. In this case we were
7680          * at the mercy of the initial ire lookup and the IRE_LOCAL it
7681          * actually returned. The new lookup, which is more specific, should
7682          * only find the IRE_LOCAL associated with the ingress ill if one
7683          * exists.
7684          */
7685         if (ire->ire_ipversion == IPV4_VERSION) {
7686                 if (ipst->ips_ip_strict_dst_multihoming)
7687                         strict_check = B_TRUE;
7688                 new_ire = ire_ftable_lookup_v4(*((ipaddr_t *)addr), 0, 0,
7689                     IRE_LOCAL, ill, ALL_ZONES, NULL,
7690                     (MATCH_IRE_TYPE|MATCH_IRE_ILL), 0, ipst, NULL);
7691         } else {
7692                 ASSERT(!IN6_IS_ADDR_MULTICAST((in6_addr_t *)addr));
7693                 if (ipst->ips_ipv6_strict_dst_multihoming)
7694                         strict_check = B_TRUE;
7695                 new_ire = ire_ftable_lookup_v6((in6_addr_t *)addr, NULL, NULL,
7696                     IRE_LOCAL, ill, ALL_ZONES, NULL,
7697                     (MATCH_IRE_TYPE|MATCH_IRE_ILL), 0, ipst, NULL);
7698         }
7699         /*
7700          * If the same ire that was returned in ip_input() is found then this
7701          * is an indication that usesrc groups are in use. The packet
7702          * arrived on a different ill in the group than the one associated with
7703          * the destination address.  If a different ire was found then the same
7704          * IP address must be hosted on multiple ills. This is possible with
7705          * unnumbered point2point interfaces. We switch to use this new ire in
7706          * order to have accurate interface statistics.
7707          */
7708         if (new_ire != NULL) {
7709                 /* Note: held in one case but not the other? Caller handles */
7710                 if (new_ire != ire)
7711                         return (new_ire);
7712                 /* Unchanged */
7713                 ire_refrele(new_ire);
7714                 return (ire);
7715         }
7716 
7717         /*
7718          * Chase pointers once and store locally.
7719          */
7720         ASSERT(ire->ire_ill != NULL);
7721         ire_ill = ire->ire_ill;
7722         ifindex = ill->ill_usesrc_ifindex;
7723 
7724         /*
7725          * Check if it's a legal address on the 'usesrc' interface.
7726          * For IPMP data addresses the IRE_LOCAL is the upper, hence we
7727          * can just check phyint_ifindex.
7728          */
7729         if (ifindex != 0 && ifindex == ire_ill->ill_phyint->phyint_ifindex) {
7730                 return (ire);
7731         }
7732 
7733         /*
7734          * If the ip*_strict_dst_multihoming switch is on then we can
7735          * only accept this packet if the interface is marked as routing.
7736          */
7737         if (!(strict_check))
7738                 return (ire);
7739 
7740         if ((ill->ill_flags & ire->ire_ill->ill_flags & ILLF_ROUTER) != 0) {
7741                 return (ire);
7742         }
7743         return (NULL);
7744 }
7745 
7746 /*
7747  * This function is used to construct a mac_header_info_s from a
7748  * DL_UNITDATA_IND message.
7749  * The address fields in the mhi structure points into the message,
7750  * thus the caller can't use those fields after freeing the message.
7751  *
7752  * We determine whether the packet received is a non-unicast packet
7753  * and in doing so, determine whether or not it is broadcast vs multicast.
7754  * For it to be a broadcast packet, we must have the appropriate mblk_t
7755  * hanging off the ill_t.  If this is either not present or doesn't match
7756  * the destination mac address in the DL_UNITDATA_IND, the packet is deemed
7757  * to be multicast.  Thus NICs that have no broadcast address (or no
7758  * capability for one, such as point to point links) cannot return as
7759  * the packet being broadcast.
7760  */
7761 void
7762 ip_dlur_to_mhi(ill_t *ill, mblk_t *mb, struct mac_header_info_s *mhip)
7763 {
7764         dl_unitdata_ind_t *ind = (dl_unitdata_ind_t *)mb->b_rptr;
7765         mblk_t *bmp;
7766         uint_t extra_offset;
7767 
7768         bzero(mhip, sizeof (struct mac_header_info_s));
7769 
7770         mhip->mhi_dsttype = MAC_ADDRTYPE_UNICAST;
7771 
7772         if (ill->ill_sap_length < 0)
7773                 extra_offset = 0;
7774         else
7775                 extra_offset = ill->ill_sap_length;
7776 
7777         mhip->mhi_daddr = (uchar_t *)ind + ind->dl_dest_addr_offset +
7778             extra_offset;
7779         mhip->mhi_saddr = (uchar_t *)ind + ind->dl_src_addr_offset +
7780             extra_offset;
7781 
7782         if (!ind->dl_group_address)
7783                 return;
7784 
7785         /* Multicast or broadcast */
7786         mhip->mhi_dsttype = MAC_ADDRTYPE_MULTICAST;
7787 
7788         if (ind->dl_dest_addr_offset > sizeof (*ind) &&
7789             ind->dl_dest_addr_offset + ind->dl_dest_addr_length < MBLKL(mb) &&
7790             (bmp = ill->ill_bcast_mp) != NULL) {
7791                 dl_unitdata_req_t *dlur;
7792                 uint8_t *bphys_addr;
7793 
7794                 dlur = (dl_unitdata_req_t *)bmp->b_rptr;
7795                 bphys_addr = (uchar_t *)dlur + dlur->dl_dest_addr_offset +
7796                     extra_offset;
7797 
7798                 if (bcmp(mhip->mhi_daddr, bphys_addr,
7799                     ind->dl_dest_addr_length) == 0)
7800                         mhip->mhi_dsttype = MAC_ADDRTYPE_BROADCAST;
7801         }
7802 }
7803 
7804 /*
7805  * This function is used to construct a mac_header_info_s from a
7806  * M_DATA fastpath message from a DLPI driver.
7807  * The address fields in the mhi structure points into the message,
7808  * thus the caller can't use those fields after freeing the message.
7809  *
7810  * We determine whether the packet received is a non-unicast packet
7811  * and in doing so, determine whether or not it is broadcast vs multicast.
7812  * For it to be a broadcast packet, we must have the appropriate mblk_t
7813  * hanging off the ill_t.  If this is either not present or doesn't match
7814  * the destination mac address in the DL_UNITDATA_IND, the packet is deemed
7815  * to be multicast.  Thus NICs that have no broadcast address (or no
7816  * capability for one, such as point to point links) cannot return as
7817  * the packet being broadcast.
7818  */
7819 void
7820 ip_mdata_to_mhi(ill_t *ill, mblk_t *mp, struct mac_header_info_s *mhip)
7821 {
7822         mblk_t *bmp;
7823         struct ether_header *pether;
7824 
7825         bzero(mhip, sizeof (struct mac_header_info_s));
7826 
7827         mhip->mhi_dsttype = MAC_ADDRTYPE_UNICAST;
7828 
7829         pether = (struct ether_header *)((char *)mp->b_rptr
7830             - sizeof (struct ether_header));
7831 
7832         /*
7833          * Make sure the interface is an ethernet type, since we don't
7834          * know the header format for anything but Ethernet. Also make
7835          * sure we are pointing correctly above db_base.
7836          */
7837         if (ill->ill_type != IFT_ETHER)
7838                 return;
7839 
7840 retry:
7841         if ((uchar_t *)pether < mp->b_datap->db_base)
7842                 return;
7843 
7844         /* Is there a VLAN tag? */
7845         if (ill->ill_isv6) {
7846                 if (pether->ether_type != htons(ETHERTYPE_IPV6)) {
7847                         pether = (struct ether_header *)((char *)pether - 4);
7848                         goto retry;
7849                 }
7850         } else {
7851                 if (pether->ether_type != htons(ETHERTYPE_IP)) {
7852                         pether = (struct ether_header *)((char *)pether - 4);
7853                         goto retry;
7854                 }
7855         }
7856         mhip->mhi_daddr = (uchar_t *)&pether->ether_dhost;
7857         mhip->mhi_saddr = (uchar_t *)&pether->ether_shost;
7858 
7859         if (!(mhip->mhi_daddr[0] & 0x01))
7860                 return;
7861 
7862         /* Multicast or broadcast */
7863         mhip->mhi_dsttype = MAC_ADDRTYPE_MULTICAST;
7864 
7865         if ((bmp = ill->ill_bcast_mp) != NULL) {
7866                 dl_unitdata_req_t *dlur;
7867                 uint8_t *bphys_addr;
7868                 uint_t  addrlen;
7869 
7870                 dlur = (dl_unitdata_req_t *)bmp->b_rptr;
7871                 addrlen = dlur->dl_dest_addr_length;
7872                 if (ill->ill_sap_length < 0) {
7873                         bphys_addr = (uchar_t *)dlur +
7874                             dlur->dl_dest_addr_offset;
7875                         addrlen += ill->ill_sap_length;
7876                 } else {
7877                         bphys_addr = (uchar_t *)dlur +
7878                             dlur->dl_dest_addr_offset +
7879                             ill->ill_sap_length;
7880                         addrlen -= ill->ill_sap_length;
7881                 }
7882                 if (bcmp(mhip->mhi_daddr, bphys_addr, addrlen) == 0)
7883                         mhip->mhi_dsttype = MAC_ADDRTYPE_BROADCAST;
7884         }
7885 }
7886 
7887 /*
7888  * Handle anything but M_DATA messages
7889  * We see the DL_UNITDATA_IND which are part
7890  * of the data path, and also the other messages from the driver.
7891  */
7892 void
7893 ip_rput_notdata(ill_t *ill, mblk_t *mp)
7894 {
7895         mblk_t          *first_mp;
7896         struct iocblk   *iocp;
7897         struct mac_header_info_s mhi;
7898 
7899         switch (DB_TYPE(mp)) {
7900         case M_PROTO:
7901         case M_PCPROTO: {
7902                 if (((dl_unitdata_ind_t *)mp->b_rptr)->dl_primitive !=
7903                     DL_UNITDATA_IND) {
7904                         /* Go handle anything other than data elsewhere. */
7905                         ip_rput_dlpi(ill, mp);
7906                         return;
7907                 }
7908 
7909                 first_mp = mp;
7910                 mp = first_mp->b_cont;
7911                 first_mp->b_cont = NULL;
7912 
7913                 if (mp == NULL) {
7914                         freeb(first_mp);
7915                         return;
7916                 }
7917                 ip_dlur_to_mhi(ill, first_mp, &mhi);
7918                 if (ill->ill_isv6)
7919                         ip_input_v6(ill, NULL, mp, &mhi);
7920                 else
7921                         ip_input(ill, NULL, mp, &mhi);
7922 
7923                 /* Ditch the DLPI header. */
7924                 freeb(first_mp);
7925                 return;
7926         }
7927         case M_IOCACK:
7928                 iocp = (struct iocblk *)mp->b_rptr;
7929                 switch (iocp->ioc_cmd) {
7930                 case DL_IOC_HDR_INFO:
7931                         ill_fastpath_ack(ill, mp);
7932                         return;
7933                 default:
7934                         putnext(ill->ill_rq, mp);
7935                         return;
7936                 }
7937                 /* FALLTHRU */
7938         case M_ERROR:
7939         case M_HANGUP:
7940                 mutex_enter(&ill->ill_lock);
7941                 if (ill->ill_state_flags & ILL_CONDEMNED) {
7942                         mutex_exit(&ill->ill_lock);
7943                         freemsg(mp);
7944                         return;
7945                 }
7946                 ill_refhold_locked(ill);
7947                 mutex_exit(&ill->ill_lock);
7948                 qwriter_ip(ill, ill->ill_rq, mp, ip_rput_other, CUR_OP,
7949                     B_FALSE);
7950                 return;
7951         case M_CTL:
7952                 putnext(ill->ill_rq, mp);
7953                 return;
7954         case M_IOCNAK:
7955                 ip1dbg(("got iocnak "));
7956                 iocp = (struct iocblk *)mp->b_rptr;
7957                 switch (iocp->ioc_cmd) {
7958                 case DL_IOC_HDR_INFO:
7959                         ip_rput_other(NULL, ill->ill_rq, mp, NULL);
7960                         return;
7961                 default:
7962                         break;
7963                 }
7964                 /* FALLTHRU */
7965         default:
7966                 putnext(ill->ill_rq, mp);
7967                 return;
7968         }
7969 }
7970 
7971 /* Read side put procedure.  Packets coming from the wire arrive here. */
7972 void
7973 ip_rput(queue_t *q, mblk_t *mp)
7974 {
7975         ill_t   *ill;
7976         union DL_primitives *dl;
7977 
7978         ill = (ill_t *)q->q_ptr;
7979 
7980         if (ill->ill_state_flags & (ILL_CONDEMNED | ILL_LL_SUBNET_PENDING)) {
7981                 /*
7982                  * If things are opening or closing, only accept high-priority
7983                  * DLPI messages.  (On open ill->ill_ipif has not yet been
7984                  * created; on close, things hanging off the ill may have been
7985                  * freed already.)
7986                  */
7987                 dl = (union DL_primitives *)mp->b_rptr;
7988                 if (DB_TYPE(mp) != M_PCPROTO ||
7989                     dl->dl_primitive == DL_UNITDATA_IND) {
7990                         inet_freemsg(mp);
7991                         return;
7992                 }
7993         }
7994         if (DB_TYPE(mp) == M_DATA) {
7995                 struct mac_header_info_s mhi;
7996 
7997                 ip_mdata_to_mhi(ill, mp, &mhi);
7998                 ip_input(ill, NULL, mp, &mhi);
7999         } else {
8000                 ip_rput_notdata(ill, mp);
8001         }
8002 }
8003 
8004 /*
8005  * Move the information to a copy.
8006  */
8007 mblk_t *
8008 ip_fix_dbref(mblk_t *mp, ip_recv_attr_t *ira)
8009 {
8010         mblk_t          *mp1;
8011         ill_t           *ill = ira->ira_ill;
8012         ip_stack_t      *ipst = ill->ill_ipst;
8013 
8014         IP_STAT(ipst, ip_db_ref);
8015 
8016         /* Make sure we have ira_l2src before we loose the original mblk */
8017         if (!(ira->ira_flags & IRAF_L2SRC_SET))
8018                 ip_setl2src(mp, ira, ira->ira_rill);
8019 
8020         mp1 = copymsg(mp);
8021         if (mp1 == NULL) {
8022                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
8023                 ip_drop_input("ipIfStatsInDiscards", mp, ill);
8024                 freemsg(mp);
8025                 return (NULL);
8026         }
8027         /* preserve the hardware checksum flags and data, if present */
8028         if (DB_CKSUMFLAGS(mp) != 0) {
8029                 DB_CKSUMFLAGS(mp1) = DB_CKSUMFLAGS(mp);
8030                 DB_CKSUMSTART(mp1) = DB_CKSUMSTART(mp);
8031                 DB_CKSUMSTUFF(mp1) = DB_CKSUMSTUFF(mp);
8032                 DB_CKSUMEND(mp1) = DB_CKSUMEND(mp);
8033                 DB_CKSUM16(mp1) = DB_CKSUM16(mp);
8034         }
8035         freemsg(mp);
8036         return (mp1);
8037 }
8038 
8039 static void
8040 ip_dlpi_error(ill_t *ill, t_uscalar_t prim, t_uscalar_t dl_err,
8041     t_uscalar_t err)
8042 {
8043         if (dl_err == DL_SYSERR) {
8044                 (void) mi_strlog(ill->ill_rq, 1, SL_CONSOLE|SL_ERROR|SL_TRACE,
8045                     "%s: %s failed: DL_SYSERR (errno %u)\n",
8046                     ill->ill_name, dl_primstr(prim), err);
8047                 return;
8048         }
8049 
8050         (void) mi_strlog(ill->ill_rq, 1, SL_CONSOLE|SL_ERROR|SL_TRACE,
8051             "%s: %s failed: %s\n", ill->ill_name, dl_primstr(prim),
8052             dl_errstr(dl_err));
8053 }
8054 
8055 /*
8056  * ip_rput_dlpi is called by ip_rput to handle all DLPI messages other
8057  * than DL_UNITDATA_IND messages. If we need to process this message
8058  * exclusively, we call qwriter_ip, in which case we also need to call
8059  * ill_refhold before that, since qwriter_ip does an ill_refrele.
8060  */
8061 void
8062 ip_rput_dlpi(ill_t *ill, mblk_t *mp)
8063 {
8064         dl_ok_ack_t     *dloa = (dl_ok_ack_t *)mp->b_rptr;
8065         dl_error_ack_t  *dlea = (dl_error_ack_t *)dloa;
8066         queue_t         *q = ill->ill_rq;
8067         t_uscalar_t     prim = dloa->dl_primitive;
8068         t_uscalar_t     reqprim = DL_PRIM_INVAL;
8069 
8070         DTRACE_PROBE3(ill__dlpi, char *, "ip_rput_dlpi",
8071             char *, dl_primstr(prim), ill_t *, ill);
8072         ip1dbg(("ip_rput_dlpi"));
8073 
8074         /*
8075          * If we received an ACK but didn't send a request for it, then it
8076          * can't be part of any pending operation; discard up-front.
8077          */
8078         switch (prim) {
8079         case DL_ERROR_ACK:
8080                 reqprim = dlea->dl_error_primitive;
8081                 ip2dbg(("ip_rput_dlpi(%s): DL_ERROR_ACK for %s (0x%x): %s "
8082                     "(0x%x), unix %u\n", ill->ill_name, dl_primstr(reqprim),
8083                     reqprim, dl_errstr(dlea->dl_errno), dlea->dl_errno,
8084                     dlea->dl_unix_errno));
8085                 break;
8086         case DL_OK_ACK:
8087                 reqprim = dloa->dl_correct_primitive;
8088                 break;
8089         case DL_INFO_ACK:
8090                 reqprim = DL_INFO_REQ;
8091                 break;
8092         case DL_BIND_ACK:
8093                 reqprim = DL_BIND_REQ;
8094                 break;
8095         case DL_PHYS_ADDR_ACK:
8096                 reqprim = DL_PHYS_ADDR_REQ;
8097                 break;
8098         case DL_NOTIFY_ACK:
8099                 reqprim = DL_NOTIFY_REQ;
8100                 break;
8101         case DL_CAPABILITY_ACK:
8102                 reqprim = DL_CAPABILITY_REQ;
8103                 break;
8104         }
8105 
8106         if (prim != DL_NOTIFY_IND) {
8107                 if (reqprim == DL_PRIM_INVAL ||
8108                     !ill_dlpi_pending(ill, reqprim)) {
8109                         /* Not a DLPI message we support or expected */
8110                         freemsg(mp);
8111                         return;
8112                 }
8113                 ip1dbg(("ip_rput: received %s for %s\n", dl_primstr(prim),
8114                     dl_primstr(reqprim)));
8115         }
8116 
8117         switch (reqprim) {
8118         case DL_UNBIND_REQ:
8119                 /*
8120                  * NOTE: we mark the unbind as complete even if we got a
8121                  * DL_ERROR_ACK, since there's not much else we can do.
8122                  */
8123                 mutex_enter(&ill->ill_lock);
8124                 ill->ill_state_flags &= ~ILL_DL_UNBIND_IN_PROGRESS;
8125                 cv_signal(&ill->ill_cv);
8126                 mutex_exit(&ill->ill_lock);
8127                 break;
8128 
8129         case DL_ENABMULTI_REQ:
8130                 if (prim == DL_OK_ACK) {
8131                         if (ill->ill_dlpi_multicast_state == IDS_INPROGRESS)
8132                                 ill->ill_dlpi_multicast_state = IDS_OK;
8133                 }
8134                 break;
8135         }
8136 
8137         /*
8138          * The message is one we're waiting for (or DL_NOTIFY_IND), but we
8139          * need to become writer to continue to process it.  Because an
8140          * exclusive operation doesn't complete until replies to all queued
8141          * DLPI messages have been received, we know we're in the middle of an
8142          * exclusive operation and pass CUR_OP (except for DL_NOTIFY_IND).
8143          *
8144          * As required by qwriter_ip(), we refhold the ill; it will refrele.
8145          * Since this is on the ill stream we unconditionally bump up the
8146          * refcount without doing ILL_CAN_LOOKUP().
8147          */
8148         ill_refhold(ill);
8149         if (prim == DL_NOTIFY_IND)
8150                 qwriter_ip(ill, q, mp, ip_rput_dlpi_writer, NEW_OP, B_FALSE);
8151         else
8152                 qwriter_ip(ill, q, mp, ip_rput_dlpi_writer, CUR_OP, B_FALSE);
8153 }
8154 
8155 /*
8156  * Handling of DLPI messages that require exclusive access to the ipsq.
8157  *
8158  * Need to do ipsq_pending_mp_get on ioctl completion, which could
8159  * happen here. (along with mi_copy_done)
8160  */
8161 /* ARGSUSED */
8162 static void
8163 ip_rput_dlpi_writer(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *dummy_arg)
8164 {
8165         dl_ok_ack_t     *dloa = (dl_ok_ack_t *)mp->b_rptr;
8166         dl_error_ack_t  *dlea = (dl_error_ack_t *)dloa;
8167         int             err = 0;
8168         ill_t           *ill = (ill_t *)q->q_ptr;
8169         ipif_t          *ipif = NULL;
8170         mblk_t          *mp1 = NULL;
8171         conn_t          *connp = NULL;
8172         t_uscalar_t     paddrreq;
8173         mblk_t          *mp_hw;
8174         boolean_t       success;
8175         boolean_t       ioctl_aborted = B_FALSE;
8176         boolean_t       log = B_TRUE;
8177 
8178         DTRACE_PROBE3(ill__dlpi, char *, "ip_rput_dlpi_writer",
8179             char *, dl_primstr(dloa->dl_primitive), ill_t *, ill);
8180 
8181         ip1dbg(("ip_rput_dlpi_writer .."));
8182         ASSERT(ipsq->ipsq_xop == ill->ill_phyint->phyint_ipsq->ipsq_xop);
8183         ASSERT(IAM_WRITER_ILL(ill));
8184 
8185         ipif = ipsq->ipsq_xop->ipx_pending_ipif;
8186         /*
8187          * The current ioctl could have been aborted by the user and a new
8188          * ioctl to bring up another ill could have started. We could still
8189          * get a response from the driver later.
8190          */
8191         if (ipif != NULL && ipif->ipif_ill != ill)
8192                 ioctl_aborted = B_TRUE;
8193 
8194         switch (dloa->dl_primitive) {
8195         case DL_ERROR_ACK:
8196                 ip1dbg(("ip_rput_dlpi_writer: got DL_ERROR_ACK for %s\n",
8197                     dl_primstr(dlea->dl_error_primitive)));
8198 
8199                 DTRACE_PROBE3(ill__dlpi, char *, "ip_rput_dlpi_writer error",
8200                     char *, dl_primstr(dlea->dl_error_primitive),
8201                     ill_t *, ill);
8202 
8203                 switch (dlea->dl_error_primitive) {
8204                 case DL_DISABMULTI_REQ:
8205                         ill_dlpi_done(ill, dlea->dl_error_primitive);
8206                         break;
8207                 case DL_PROMISCON_REQ:
8208                 case DL_PROMISCOFF_REQ:
8209                 case DL_UNBIND_REQ:
8210                 case DL_ATTACH_REQ:
8211                 case DL_INFO_REQ:
8212                         ill_dlpi_done(ill, dlea->dl_error_primitive);
8213                         break;
8214                 case DL_NOTIFY_REQ:
8215                         ill_dlpi_done(ill, DL_NOTIFY_REQ);
8216                         log = B_FALSE;
8217                         break;
8218                 case DL_PHYS_ADDR_REQ:
8219                         /*
8220                          * For IPv6 only, there are two additional
8221                          * phys_addr_req's sent to the driver to get the
8222                          * IPv6 token and lla. This allows IP to acquire
8223                          * the hardware address format for a given interface
8224                          * without having built in knowledge of the hardware
8225                          * address. ill_phys_addr_pend keeps track of the last
8226                          * DL_PAR sent so we know which response we are
8227                          * dealing with. ill_dlpi_done will update
8228                          * ill_phys_addr_pend when it sends the next req.
8229                          * We don't complete the IOCTL until all three DL_PARs
8230                          * have been attempted, so set *_len to 0 and break.
8231                          */
8232                         paddrreq = ill->ill_phys_addr_pend;
8233                         ill_dlpi_done(ill, DL_PHYS_ADDR_REQ);
8234                         if (paddrreq == DL_IPV6_TOKEN) {
8235                                 ill->ill_token_length = 0;
8236                                 log = B_FALSE;
8237                                 break;
8238                         } else if (paddrreq == DL_IPV6_LINK_LAYER_ADDR) {
8239                                 ill->ill_nd_lla_len = 0;
8240                                 log = B_FALSE;
8241                                 break;
8242                         }
8243                         /*
8244                          * Something went wrong with the DL_PHYS_ADDR_REQ.
8245                          * We presumably have an IOCTL hanging out waiting
8246                          * for completion. Find it and complete the IOCTL
8247                          * with the error noted.
8248                          * However, ill_dl_phys was called on an ill queue
8249                          * (from SIOCSLIFNAME), thus conn_pending_ill is not
8250                          * set. But the ioctl is known to be pending on ill_wq.
8251                          */
8252                         if (!ill->ill_ifname_pending)
8253                                 break;
8254                         ill->ill_ifname_pending = 0;
8255                         if (!ioctl_aborted)
8256                                 mp1 = ipsq_pending_mp_get(ipsq, &connp);
8257                         if (mp1 != NULL) {
8258                                 /*
8259                                  * This operation (SIOCSLIFNAME) must have
8260                                  * happened on the ill. Assert there is no conn
8261                                  */
8262                                 ASSERT(connp == NULL);
8263                                 q = ill->ill_wq;
8264                         }
8265                         break;
8266                 case DL_BIND_REQ:
8267                         ill_dlpi_done(ill, DL_BIND_REQ);
8268                         if (ill->ill_ifname_pending)
8269                                 break;
8270                         mutex_enter(&ill->ill_lock);
8271                         ill->ill_state_flags &= ~ILL_DOWN_IN_PROGRESS;
8272                         mutex_exit(&ill->ill_lock);
8273                         /*
8274                          * Something went wrong with the bind.  We presumably
8275                          * have an IOCTL hanging out waiting for completion.
8276                          * Find it, take down the interface that was coming
8277                          * up, and complete the IOCTL with the error noted.
8278                          */
8279                         if (!ioctl_aborted)
8280                                 mp1 = ipsq_pending_mp_get(ipsq, &connp);
8281                         if (mp1 != NULL) {
8282                                 /*
8283                                  * This might be a result of a DL_NOTE_REPLUMB
8284                                  * notification. In that case, connp is NULL.
8285                                  */
8286                                 if (connp != NULL)
8287                                         q = CONNP_TO_WQ(connp);
8288 
8289                                 (void) ipif_down(ipif, NULL, NULL);
8290                                 /* error is set below the switch */
8291                         }
8292                         break;
8293                 case DL_ENABMULTI_REQ:
8294                         ill_dlpi_done(ill, DL_ENABMULTI_REQ);
8295 
8296                         if (ill->ill_dlpi_multicast_state == IDS_INPROGRESS)
8297                                 ill->ill_dlpi_multicast_state = IDS_FAILED;
8298                         if (ill->ill_dlpi_multicast_state == IDS_FAILED) {
8299 
8300                                 printf("ip: joining multicasts failed (%d)"
8301                                     " on %s - will use link layer "
8302                                     "broadcasts for multicast\n",
8303                                     dlea->dl_errno, ill->ill_name);
8304 
8305                                 /*
8306                                  * Set up for multi_bcast; We are the
8307                                  * writer, so ok to access ill->ill_ipif
8308                                  * without any lock.
8309                                  */
8310                                 mutex_enter(&ill->ill_phyint->phyint_lock);
8311                                 ill->ill_phyint->phyint_flags |=
8312                                     PHYI_MULTI_BCAST;
8313                                 mutex_exit(&ill->ill_phyint->phyint_lock);
8314 
8315                         }
8316                         freemsg(mp);    /* Don't want to pass this up */
8317                         return;
8318                 case DL_CAPABILITY_REQ:
8319                         ip1dbg(("ip_rput_dlpi_writer: got DL_ERROR_ACK for "
8320                             "DL_CAPABILITY REQ\n"));
8321                         if (ill->ill_dlpi_capab_state == IDCS_PROBE_SENT)
8322                                 ill->ill_dlpi_capab_state = IDCS_FAILED;
8323                         ill_capability_done(ill);
8324                         freemsg(mp);
8325                         return;
8326                 }
8327                 /*
8328                  * Note the error for IOCTL completion (mp1 is set when
8329                  * ready to complete ioctl). If ill_ifname_pending_err is
8330                  * set, an error occured during plumbing (ill_ifname_pending),
8331                  * so we want to report that error.
8332                  *
8333                  * NOTE: there are two addtional DL_PHYS_ADDR_REQ's
8334                  * (DL_IPV6_TOKEN and DL_IPV6_LINK_LAYER_ADDR) that are
8335                  * expected to get errack'd if the driver doesn't support
8336                  * these flags (e.g. ethernet). log will be set to B_FALSE
8337                  * if these error conditions are encountered.
8338                  */
8339                 if (mp1 != NULL) {
8340                         if (ill->ill_ifname_pending_err != 0)  {
8341                                 err = ill->ill_ifname_pending_err;
8342                                 ill->ill_ifname_pending_err = 0;
8343                         } else {
8344                                 err = dlea->dl_unix_errno ?
8345                                     dlea->dl_unix_errno : ENXIO;
8346                         }
8347                 /*
8348                  * If we're plumbing an interface and an error hasn't already
8349                  * been saved, set ill_ifname_pending_err to the error passed
8350                  * up. Ignore the error if log is B_FALSE (see comment above).
8351                  */
8352                 } else if (log && ill->ill_ifname_pending &&
8353                     ill->ill_ifname_pending_err == 0) {
8354                         ill->ill_ifname_pending_err = dlea->dl_unix_errno ?
8355                             dlea->dl_unix_errno : ENXIO;
8356                 }
8357 
8358                 if (log)
8359                         ip_dlpi_error(ill, dlea->dl_error_primitive,
8360                             dlea->dl_errno, dlea->dl_unix_errno);
8361                 break;
8362         case DL_CAPABILITY_ACK:
8363                 ill_capability_ack(ill, mp);
8364                 /*
8365                  * The message has been handed off to ill_capability_ack
8366                  * and must not be freed below
8367                  */
8368                 mp = NULL;
8369                 break;
8370 
8371         case DL_INFO_ACK:
8372                 /* Call a routine to handle this one. */
8373                 ill_dlpi_done(ill, DL_INFO_REQ);
8374                 ip_ll_subnet_defaults(ill, mp);
8375                 ASSERT(!MUTEX_HELD(&ill->ill_phyint->phyint_ipsq->ipsq_lock));
8376                 return;
8377         case DL_BIND_ACK:
8378                 /*
8379                  * We should have an IOCTL waiting on this unless
8380                  * sent by ill_dl_phys, in which case just return
8381                  */
8382                 ill_dlpi_done(ill, DL_BIND_REQ);
8383 
8384                 if (ill->ill_ifname_pending) {
8385                         DTRACE_PROBE2(ip__rput__dlpi__ifname__pending,
8386                             ill_t *, ill, mblk_t *, mp);
8387                         break;
8388                 }
8389                 mutex_enter(&ill->ill_lock);
8390                 ill->ill_dl_up = 1;
8391                 ill->ill_state_flags &= ~ILL_DOWN_IN_PROGRESS;
8392                 mutex_exit(&ill->ill_lock);
8393 
8394                 if (!ioctl_aborted)
8395                         mp1 = ipsq_pending_mp_get(ipsq, &connp);
8396                 if (mp1 == NULL) {
8397                         DTRACE_PROBE1(ip__rput__dlpi__no__mblk, ill_t *, ill);
8398                         break;
8399                 }
8400                 /*
8401                  * mp1 was added by ill_dl_up(). if that is a result of
8402                  * a DL_NOTE_REPLUMB notification, connp could be NULL.
8403                  */
8404                 if (connp != NULL)
8405                         q = CONNP_TO_WQ(connp);
8406                 /*
8407                  * We are exclusive. So nothing can change even after
8408                  * we get the pending mp.
8409                  */
8410                 ip1dbg(("ip_rput_dlpi: bind_ack %s\n", ill->ill_name));
8411                 DTRACE_PROBE1(ip__rput__dlpi__bind__ack, ill_t *, ill);
8412                 ill_nic_event_dispatch(ill, 0, NE_UP, NULL, 0);
8413 
8414                 /*
8415                  * Now bring up the resolver; when that is complete, we'll
8416                  * create IREs.  Note that we intentionally mirror what
8417                  * ipif_up() would have done, because we got here by way of
8418                  * ill_dl_up(), which stopped ipif_up()'s processing.
8419                  */
8420                 if (ill->ill_isv6) {
8421                         /*
8422                          * v6 interfaces.
8423                          * Unlike ARP which has to do another bind
8424                          * and attach, once we get here we are
8425                          * done with NDP
8426                          */
8427                         (void) ipif_resolver_up(ipif, Res_act_initial);
8428                         if ((err = ipif_ndp_up(ipif, B_TRUE)) == 0)
8429                                 err = ipif_up_done_v6(ipif);
8430                 } else if (ill->ill_net_type == IRE_IF_RESOLVER) {
8431                         /*
8432                          * ARP and other v4 external resolvers.
8433                          * Leave the pending mblk intact so that
8434                          * the ioctl completes in ip_rput().
8435                          */
8436                         if (connp != NULL)
8437                                 mutex_enter(&connp->conn_lock);
8438                         mutex_enter(&ill->ill_lock);
8439                         success = ipsq_pending_mp_add(connp, ipif, q, mp1, 0);
8440                         mutex_exit(&ill->ill_lock);
8441                         if (connp != NULL)
8442                                 mutex_exit(&connp->conn_lock);
8443                         if (success) {
8444                                 err = ipif_resolver_up(ipif, Res_act_initial);
8445                                 if (err == EINPROGRESS) {
8446                                         freemsg(mp);
8447                                         return;
8448                                 }
8449                                 mp1 = ipsq_pending_mp_get(ipsq, &connp);
8450                         } else {
8451                                 /* The conn has started closing */
8452                                 err = EINTR;
8453                         }
8454                 } else {
8455                         /*
8456                          * This one is complete. Reply to pending ioctl.
8457                          */
8458                         (void) ipif_resolver_up(ipif, Res_act_initial);
8459                         err = ipif_up_done(ipif);
8460                 }
8461 
8462                 if ((err == 0) && (ill->ill_up_ipifs)) {
8463                         err = ill_up_ipifs(ill, q, mp1);
8464                         if (err == EINPROGRESS) {
8465                                 freemsg(mp);
8466                                 return;
8467                         }
8468                 }
8469 
8470                 /*
8471                  * If we have a moved ipif to bring up, and everything has
8472                  * succeeded to this point, bring it up on the IPMP ill.
8473                  * Otherwise, leave it down -- the admin can try to bring it
8474                  * up by hand if need be.
8475                  */
8476                 if (ill->ill_move_ipif != NULL) {
8477                         if (err != 0) {
8478                                 ill->ill_move_ipif = NULL;
8479                         } else {
8480                                 ipif = ill->ill_move_ipif;
8481                                 ill->ill_move_ipif = NULL;
8482                                 err = ipif_up(ipif, q, mp1);
8483                                 if (err == EINPROGRESS) {
8484                                         freemsg(mp);
8485                                         return;
8486                                 }
8487                         }
8488                 }
8489                 break;
8490 
8491         case DL_NOTIFY_IND: {
8492                 dl_notify_ind_t *notify = (dl_notify_ind_t *)mp->b_rptr;
8493                 uint_t orig_mtu, orig_mc_mtu;
8494 
8495                 switch (notify->dl_notification) {
8496                 case DL_NOTE_PHYS_ADDR:
8497                         err = ill_set_phys_addr(ill, mp);
8498                         break;
8499 
8500                 case DL_NOTE_REPLUMB:
8501                         /*
8502                          * Directly return after calling ill_replumb().
8503                          * Note that we should not free mp as it is reused
8504                          * in the ill_replumb() function.
8505                          */
8506                         err = ill_replumb(ill, mp);
8507                         return;
8508 
8509                 case DL_NOTE_FASTPATH_FLUSH:
8510                         nce_flush(ill, B_FALSE);
8511                         break;
8512 
8513                 case DL_NOTE_SDU_SIZE:
8514                 case DL_NOTE_SDU_SIZE2:
8515                         /*
8516                          * The dce and fragmentation code can cope with
8517                          * this changing while packets are being sent.
8518                          * When packets are sent ip_output will discover
8519                          * a change.
8520                          *
8521                          * Change the MTU size of the interface.
8522                          */
8523                         mutex_enter(&ill->ill_lock);
8524                         orig_mtu = ill->ill_mtu;
8525                         orig_mc_mtu = ill->ill_mc_mtu;
8526                         switch (notify->dl_notification) {
8527                         case DL_NOTE_SDU_SIZE:
8528                                 ill->ill_current_frag =
8529                                     (uint_t)notify->dl_data;
8530                                 ill->ill_mc_mtu = (uint_t)notify->dl_data;
8531                                 break;
8532                         case DL_NOTE_SDU_SIZE2:
8533                                 ill->ill_current_frag =
8534                                     (uint_t)notify->dl_data1;
8535                                 ill->ill_mc_mtu = (uint_t)notify->dl_data2;
8536                                 break;
8537                         }
8538                         if (ill->ill_current_frag > ill->ill_max_frag)
8539                                 ill->ill_max_frag = ill->ill_current_frag;
8540 
8541                         if (!(ill->ill_flags & ILLF_FIXEDMTU)) {
8542                                 ill->ill_mtu = ill->ill_current_frag;
8543 
8544                                 /*
8545                                  * If ill_user_mtu was set (via
8546                                  * SIOCSLIFLNKINFO), clamp ill_mtu at it.
8547                                  */
8548                                 if (ill->ill_user_mtu != 0 &&
8549                                     ill->ill_user_mtu < ill->ill_mtu)
8550                                         ill->ill_mtu = ill->ill_user_mtu;
8551 
8552                                 if (ill->ill_user_mtu != 0 &&
8553                                     ill->ill_user_mtu < ill->ill_mc_mtu)
8554                                         ill->ill_mc_mtu = ill->ill_user_mtu;
8555 
8556                                 if (ill->ill_isv6) {
8557                                         if (ill->ill_mtu < IPV6_MIN_MTU)
8558                                                 ill->ill_mtu = IPV6_MIN_MTU;
8559                                         if (ill->ill_mc_mtu < IPV6_MIN_MTU)
8560                                                 ill->ill_mc_mtu = IPV6_MIN_MTU;
8561                                 } else {
8562                                         if (ill->ill_mtu < IP_MIN_MTU)
8563                                                 ill->ill_mtu = IP_MIN_MTU;
8564                                         if (ill->ill_mc_mtu < IP_MIN_MTU)
8565                                                 ill->ill_mc_mtu = IP_MIN_MTU;
8566                                 }
8567                         } else if (ill->ill_mc_mtu > ill->ill_mtu) {
8568                                 ill->ill_mc_mtu = ill->ill_mtu;
8569                         }
8570 
8571                         mutex_exit(&ill->ill_lock);
8572                         /*
8573                          * Make sure all dce_generation checks find out
8574                          * that ill_mtu/ill_mc_mtu has changed.
8575                          */
8576                         if (orig_mtu != ill->ill_mtu ||
8577                             orig_mc_mtu != ill->ill_mc_mtu) {
8578                                 dce_increment_all_generations(ill->ill_isv6,
8579                                     ill->ill_ipst);
8580                         }
8581 
8582                         /*
8583                          * Refresh IPMP meta-interface MTU if necessary.
8584                          */
8585                         if (IS_UNDER_IPMP(ill))
8586                                 ipmp_illgrp_refresh_mtu(ill->ill_grp);
8587                         break;
8588 
8589                 case DL_NOTE_LINK_UP:
8590                 case DL_NOTE_LINK_DOWN: {
8591                         /*
8592                          * We are writer. ill / phyint / ipsq assocs stable.
8593                          * The RUNNING flag reflects the state of the link.
8594                          */
8595                         phyint_t *phyint = ill->ill_phyint;
8596                         uint64_t new_phyint_flags;
8597                         boolean_t changed = B_FALSE;
8598                         boolean_t went_up;
8599 
8600                         went_up = notify->dl_notification == DL_NOTE_LINK_UP;
8601                         mutex_enter(&phyint->phyint_lock);
8602 
8603                         new_phyint_flags = went_up ?
8604                             phyint->phyint_flags | PHYI_RUNNING :
8605                             phyint->phyint_flags & ~PHYI_RUNNING;
8606 
8607                         if (IS_IPMP(ill)) {
8608                                 new_phyint_flags = went_up ?
8609                                     new_phyint_flags & ~PHYI_FAILED :
8610                                     new_phyint_flags | PHYI_FAILED;
8611                         }
8612 
8613                         if (new_phyint_flags != phyint->phyint_flags) {
8614                                 phyint->phyint_flags = new_phyint_flags;
8615                                 changed = B_TRUE;
8616                         }
8617                         mutex_exit(&phyint->phyint_lock);
8618                         /*
8619                          * ill_restart_dad handles the DAD restart and routing
8620                          * socket notification logic.
8621                          */
8622                         if (changed) {
8623                                 ill_restart_dad(phyint->phyint_illv4, went_up);
8624                                 ill_restart_dad(phyint->phyint_illv6, went_up);
8625                         }
8626                         break;
8627                 }
8628                 case DL_NOTE_PROMISC_ON_PHYS: {
8629                         phyint_t *phyint = ill->ill_phyint;
8630 
8631                         mutex_enter(&phyint->phyint_lock);
8632                         phyint->phyint_flags |= PHYI_PROMISC;
8633                         mutex_exit(&phyint->phyint_lock);
8634                         break;
8635                 }
8636                 case DL_NOTE_PROMISC_OFF_PHYS: {
8637                         phyint_t *phyint = ill->ill_phyint;
8638 
8639                         mutex_enter(&phyint->phyint_lock);
8640                         phyint->phyint_flags &= ~PHYI_PROMISC;
8641                         mutex_exit(&phyint->phyint_lock);
8642                         break;
8643                 }
8644                 case DL_NOTE_CAPAB_RENEG:
8645                         /*
8646                          * Something changed on the driver side.
8647                          * It wants us to renegotiate the capabilities
8648                          * on this ill. One possible cause is the aggregation
8649                          * interface under us where a port got added or
8650                          * went away.
8651                          *
8652                          * If the capability negotiation is already done
8653                          * or is in progress, reset the capabilities and
8654                          * mark the ill's ill_capab_reneg to be B_TRUE,
8655                          * so that when the ack comes back, we can start
8656                          * the renegotiation process.
8657                          *
8658                          * Note that if ill_capab_reneg is already B_TRUE
8659                          * (ill_dlpi_capab_state is IDS_UNKNOWN in this case),
8660                          * the capability resetting request has been sent
8661                          * and the renegotiation has not been started yet;
8662                          * nothing needs to be done in this case.
8663                          */
8664                         ipsq_current_start(ipsq, ill->ill_ipif, 0);
8665                         ill_capability_reset(ill, B_TRUE);
8666                         ipsq_current_finish(ipsq);
8667                         break;
8668 
8669                 case DL_NOTE_ALLOWED_IPS:
8670                         ill_set_allowed_ips(ill, mp);
8671                         break;
8672                 default:
8673                         ip0dbg(("ip_rput_dlpi_writer: unknown notification "
8674                             "type 0x%x for DL_NOTIFY_IND\n",
8675                             notify->dl_notification));
8676                         break;
8677                 }
8678 
8679                 /*
8680                  * As this is an asynchronous operation, we
8681                  * should not call ill_dlpi_done
8682                  */
8683                 break;
8684         }
8685         case DL_NOTIFY_ACK: {
8686                 dl_notify_ack_t *noteack = (dl_notify_ack_t *)mp->b_rptr;
8687 
8688                 if (noteack->dl_notifications & DL_NOTE_LINK_UP)
8689                         ill->ill_note_link = 1;
8690                 ill_dlpi_done(ill, DL_NOTIFY_REQ);
8691                 break;
8692         }
8693         case DL_PHYS_ADDR_ACK: {
8694                 /*
8695                  * As part of plumbing the interface via SIOCSLIFNAME,
8696                  * ill_dl_phys() will queue a series of DL_PHYS_ADDR_REQs,
8697                  * whose answers we receive here.  As each answer is received,
8698                  * we call ill_dlpi_done() to dispatch the next request as
8699                  * we're processing the current one.  Once all answers have
8700                  * been received, we use ipsq_pending_mp_get() to dequeue the
8701                  * outstanding IOCTL and reply to it.  (Because ill_dl_phys()
8702                  * is invoked from an ill queue, conn_oper_pending_ill is not
8703                  * available, but we know the ioctl is pending on ill_wq.)
8704                  */
8705                 uint_t  paddrlen, paddroff;
8706                 uint8_t *addr;
8707 
8708                 paddrreq = ill->ill_phys_addr_pend;
8709                 paddrlen = ((dl_phys_addr_ack_t *)mp->b_rptr)->dl_addr_length;
8710                 paddroff = ((dl_phys_addr_ack_t *)mp->b_rptr)->dl_addr_offset;
8711                 addr = mp->b_rptr + paddroff;
8712 
8713                 ill_dlpi_done(ill, DL_PHYS_ADDR_REQ);
8714                 if (paddrreq == DL_IPV6_TOKEN) {
8715                         /*
8716                          * bcopy to low-order bits of ill_token
8717                          *
8718                          * XXX Temporary hack - currently, all known tokens
8719                          * are 64 bits, so I'll cheat for the moment.
8720                          */
8721                         bcopy(addr, &ill->ill_token.s6_addr32[2], paddrlen);
8722                         ill->ill_token_length = paddrlen;
8723                         break;
8724                 } else if (paddrreq == DL_IPV6_LINK_LAYER_ADDR) {
8725                         ASSERT(ill->ill_nd_lla_mp == NULL);
8726                         ill_set_ndmp(ill, mp, paddroff, paddrlen);
8727                         mp = NULL;
8728                         break;
8729                 } else if (paddrreq == DL_CURR_DEST_ADDR) {
8730                         ASSERT(ill->ill_dest_addr_mp == NULL);
8731                         ill->ill_dest_addr_mp = mp;
8732                         ill->ill_dest_addr = addr;
8733                         mp = NULL;
8734                         if (ill->ill_isv6) {
8735                                 ill_setdesttoken(ill);
8736                                 ipif_setdestlinklocal(ill->ill_ipif);
8737                         }
8738                         break;
8739                 }
8740 
8741                 ASSERT(paddrreq == DL_CURR_PHYS_ADDR);
8742                 ASSERT(ill->ill_phys_addr_mp == NULL);
8743                 if (!ill->ill_ifname_pending)
8744                         break;
8745                 ill->ill_ifname_pending = 0;
8746                 if (!ioctl_aborted)
8747                         mp1 = ipsq_pending_mp_get(ipsq, &connp);
8748                 if (mp1 != NULL) {
8749                         ASSERT(connp == NULL);
8750                         q = ill->ill_wq;
8751                 }
8752                 /*
8753                  * If any error acks received during the plumbing sequence,
8754                  * ill_ifname_pending_err will be set. Break out and send up
8755                  * the error to the pending ioctl.
8756                  */
8757                 if (ill->ill_ifname_pending_err != 0) {
8758                         err = ill->ill_ifname_pending_err;
8759                         ill->ill_ifname_pending_err = 0;
8760                         break;
8761                 }
8762 
8763                 ill->ill_phys_addr_mp = mp;
8764                 ill->ill_phys_addr = (paddrlen == 0 ? NULL : addr);
8765                 mp = NULL;
8766 
8767                 /*
8768                  * If paddrlen or ill_phys_addr_length is zero, the DLPI
8769                  * provider doesn't support physical addresses.  We check both
8770                  * paddrlen and ill_phys_addr_length because sppp (PPP) does
8771                  * not have physical addresses, but historically adversises a
8772                  * physical address length of 0 in its DL_INFO_ACK, but 6 in
8773                  * its DL_PHYS_ADDR_ACK.
8774                  */
8775                 if (paddrlen == 0 || ill->ill_phys_addr_length == 0) {
8776                         ill->ill_phys_addr = NULL;
8777                 } else if (paddrlen != ill->ill_phys_addr_length) {
8778                         ip0dbg(("DL_PHYS_ADDR_ACK: got addrlen %d, expected %d",
8779                             paddrlen, ill->ill_phys_addr_length));
8780                         err = EINVAL;
8781                         break;
8782                 }
8783 
8784                 if (ill->ill_nd_lla_mp == NULL) {
8785                         if ((mp_hw = copyb(ill->ill_phys_addr_mp)) == NULL) {
8786                                 err = ENOMEM;
8787                                 break;
8788                         }
8789                         ill_set_ndmp(ill, mp_hw, paddroff, paddrlen);
8790                 }
8791 
8792                 if (ill->ill_isv6) {
8793                         ill_setdefaulttoken(ill);
8794                         ipif_setlinklocal(ill->ill_ipif);
8795                 }
8796                 break;
8797         }
8798         case DL_OK_ACK:
8799                 ip2dbg(("DL_OK_ACK %s (0x%x)\n",
8800                     dl_primstr((int)dloa->dl_correct_primitive),
8801                     dloa->dl_correct_primitive));
8802                 DTRACE_PROBE3(ill__dlpi, char *, "ip_rput_dlpi_writer ok",
8803                     char *, dl_primstr(dloa->dl_correct_primitive),
8804                     ill_t *, ill);
8805 
8806                 switch (dloa->dl_correct_primitive) {
8807                 case DL_ENABMULTI_REQ:
8808                 case DL_DISABMULTI_REQ:
8809                         ill_dlpi_done(ill, dloa->dl_correct_primitive);
8810                         break;
8811                 case DL_PROMISCON_REQ:
8812                 case DL_PROMISCOFF_REQ:
8813                 case DL_UNBIND_REQ:
8814                 case DL_ATTACH_REQ:
8815                         ill_dlpi_done(ill, dloa->dl_correct_primitive);
8816                         break;
8817                 }
8818                 break;
8819         default:
8820                 break;
8821         }
8822 
8823         freemsg(mp);
8824         if (mp1 == NULL)
8825                 return;
8826 
8827         /*
8828          * The operation must complete without EINPROGRESS since
8829          * ipsq_pending_mp_get() has removed the mblk (mp1).  Otherwise,
8830          * the operation will be stuck forever inside the IPSQ.
8831          */
8832         ASSERT(err != EINPROGRESS);
8833 
8834         DTRACE_PROBE4(ipif__ioctl, char *, "ip_rput_dlpi_writer finish",
8835             int, ipsq->ipsq_xop->ipx_current_ioctl, ill_t *, ill,
8836             ipif_t *, NULL);
8837 
8838         switch (ipsq->ipsq_xop->ipx_current_ioctl) {
8839         case 0:
8840                 ipsq_current_finish(ipsq);
8841                 break;
8842 
8843         case SIOCSLIFNAME:
8844         case IF_UNITSEL: {
8845                 ill_t *ill_other = ILL_OTHER(ill);
8846 
8847                 /*
8848                  * If SIOCSLIFNAME or IF_UNITSEL is about to succeed, and the
8849                  * ill has a peer which is in an IPMP group, then place ill
8850                  * into the same group.  One catch: although ifconfig plumbs
8851                  * the appropriate IPMP meta-interface prior to plumbing this
8852                  * ill, it is possible for multiple ifconfig applications to
8853                  * race (or for another application to adjust plumbing), in
8854                  * which case the IPMP meta-interface we need will be missing.
8855                  * If so, kick the phyint out of the group.
8856                  */
8857                 if (err == 0 && ill_other != NULL && IS_UNDER_IPMP(ill_other)) {
8858                         ipmp_grp_t      *grp = ill->ill_phyint->phyint_grp;
8859                         ipmp_illgrp_t   *illg;
8860 
8861                         illg = ill->ill_isv6 ? grp->gr_v6 : grp->gr_v4;
8862                         if (illg == NULL)
8863                                 ipmp_phyint_leave_grp(ill->ill_phyint);
8864                         else
8865                                 ipmp_ill_join_illgrp(ill, illg);
8866                 }
8867 
8868                 if (ipsq->ipsq_xop->ipx_current_ioctl == IF_UNITSEL)
8869                         ip_ioctl_finish(q, mp1, err, NO_COPYOUT, ipsq);
8870                 else
8871                         ip_ioctl_finish(q, mp1, err, COPYOUT, ipsq);
8872                 break;
8873         }
8874         case SIOCLIFADDIF:
8875                 ip_ioctl_finish(q, mp1, err, COPYOUT, ipsq);
8876                 break;
8877 
8878         default:
8879                 ip_ioctl_finish(q, mp1, err, NO_COPYOUT, ipsq);
8880                 break;
8881         }
8882 }
8883 
8884 /*
8885  * ip_rput_other is called by ip_rput to handle messages modifying the global
8886  * state in IP.  If 'ipsq' is non-NULL, caller is writer on it.
8887  */
8888 /* ARGSUSED */
8889 void
8890 ip_rput_other(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *dummy_arg)
8891 {
8892         ill_t           *ill = q->q_ptr;
8893         struct iocblk   *iocp;
8894 
8895         ip1dbg(("ip_rput_other "));
8896         if (ipsq != NULL) {
8897                 ASSERT(IAM_WRITER_IPSQ(ipsq));
8898                 ASSERT(ipsq->ipsq_xop ==
8899                     ill->ill_phyint->phyint_ipsq->ipsq_xop);
8900         }
8901 
8902         switch (mp->b_datap->db_type) {
8903         case M_ERROR:
8904         case M_HANGUP:
8905                 /*
8906                  * The device has a problem.  We force the ILL down.  It can
8907                  * be brought up again manually using SIOCSIFFLAGS (via
8908                  * ifconfig or equivalent).
8909                  */
8910                 ASSERT(ipsq != NULL);
8911                 if (mp->b_rptr < mp->b_wptr)
8912                         ill->ill_error = (int)(*mp->b_rptr & 0xFF);
8913                 if (ill->ill_error == 0)
8914                         ill->ill_error = ENXIO;
8915                 if (!ill_down_start(q, mp))
8916                         return;
8917                 ipif_all_down_tail(ipsq, q, mp, NULL);
8918                 break;
8919         case M_IOCNAK: {
8920                 iocp = (struct iocblk *)mp->b_rptr;
8921 
8922                 ASSERT(iocp->ioc_cmd == DL_IOC_HDR_INFO);
8923                 /*
8924                  * If this was the first attempt, turn off the fastpath
8925                  * probing.
8926                  */
8927                 mutex_enter(&ill->ill_lock);
8928                 if (ill->ill_dlpi_fastpath_state == IDS_INPROGRESS) {
8929                         ill->ill_dlpi_fastpath_state = IDS_FAILED;
8930                         mutex_exit(&ill->ill_lock);
8931                         /*
8932                          * don't flush the nce_t entries: we use them
8933                          * as an index to the ncec itself.
8934                          */
8935                         ip1dbg(("ip_rput: DLPI fastpath off on interface %s\n",
8936                             ill->ill_name));
8937                 } else {
8938                         mutex_exit(&ill->ill_lock);
8939                 }
8940                 freemsg(mp);
8941                 break;
8942         }
8943         default:
8944                 ASSERT(0);
8945                 break;
8946         }
8947 }
8948 
8949 /*
8950  * Update any source route, record route or timestamp options
8951  * When it fails it has consumed the message and BUMPed the MIB.
8952  */
8953 boolean_t
8954 ip_forward_options(mblk_t *mp, ipha_t *ipha, ill_t *dst_ill,
8955     ip_recv_attr_t *ira)
8956 {
8957         ipoptp_t        opts;
8958         uchar_t         *opt;
8959         uint8_t         optval;
8960         uint8_t         optlen;
8961         ipaddr_t        dst;
8962         ipaddr_t        ifaddr;
8963         uint32_t        ts;
8964         timestruc_t     now;
8965         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
8966 
8967         ip2dbg(("ip_forward_options\n"));
8968         dst = ipha->ipha_dst;
8969         for (optval = ipoptp_first(&opts, ipha);
8970             optval != IPOPT_EOL;
8971             optval = ipoptp_next(&opts)) {
8972                 ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
8973                 opt = opts.ipoptp_cur;
8974                 optlen = opts.ipoptp_len;
8975                 ip2dbg(("ip_forward_options: opt %d, len %d\n",
8976                     optval, opts.ipoptp_len));
8977                 switch (optval) {
8978                         uint32_t off;
8979                 case IPOPT_SSRR:
8980                 case IPOPT_LSRR:
8981                         /* Check if adminstratively disabled */
8982                         if (!ipst->ips_ip_forward_src_routed) {
8983                                 BUMP_MIB(dst_ill->ill_ip_mib,
8984                                     ipIfStatsForwProhibits);
8985                                 ip_drop_input("ICMP_SOURCE_ROUTE_FAILED",
8986                                     mp, dst_ill);
8987                                 icmp_unreachable(mp, ICMP_SOURCE_ROUTE_FAILED,
8988                                     ira);
8989                                 return (B_FALSE);
8990                         }
8991                         if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
8992                                 /*
8993                                  * Must be partial since ip_input_options
8994                                  * checked for strict.
8995                                  */
8996                                 break;
8997                         }
8998                         off = opt[IPOPT_OFFSET];
8999                         off--;
9000                 redo_srr:
9001                         if (optlen < IP_ADDR_LEN ||
9002                             off > optlen - IP_ADDR_LEN) {
9003                                 /* End of source route */
9004                                 ip1dbg((
9005                                     "ip_forward_options: end of SR\n"));
9006                                 break;
9007                         }
9008                         /* Pick a reasonable address on the outbound if */
9009                         ASSERT(dst_ill != NULL);
9010                         if (ip_select_source_v4(dst_ill, INADDR_ANY, dst,
9011                             INADDR_ANY, ALL_ZONES, ipst, &ifaddr, NULL,
9012                             NULL) != 0) {
9013                                 /* No source! Shouldn't happen */
9014                                 ifaddr = INADDR_ANY;
9015                         }
9016                         bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
9017                         bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9018                         ip1dbg(("ip_forward_options: next hop 0x%x\n",
9019                             ntohl(dst)));
9020 
9021                         /*
9022                          * Check if our address is present more than
9023                          * once as consecutive hops in source route.
9024                          */
9025                         if (ip_type_v4(dst, ipst) == IRE_LOCAL) {
9026                                 off += IP_ADDR_LEN;
9027                                 opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9028                                 goto redo_srr;
9029                         }
9030                         ipha->ipha_dst = dst;
9031                         opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9032                         break;
9033                 case IPOPT_RR:
9034                         off = opt[IPOPT_OFFSET];
9035                         off--;
9036                         if (optlen < IP_ADDR_LEN ||
9037                             off > optlen - IP_ADDR_LEN) {
9038                                 /* No more room - ignore */
9039                                 ip1dbg((
9040                                     "ip_forward_options: end of RR\n"));
9041                                 break;
9042                         }
9043                         /* Pick a reasonable address on the outbound if */
9044                         ASSERT(dst_ill != NULL);
9045                         if (ip_select_source_v4(dst_ill, INADDR_ANY, dst,
9046                             INADDR_ANY, ALL_ZONES, ipst, &ifaddr, NULL,
9047                             NULL) != 0) {
9048                                 /* No source! Shouldn't happen */
9049                                 ifaddr = INADDR_ANY;
9050                         }
9051                         bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9052                         opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9053                         break;
9054                 case IPOPT_TS:
9055                         /* Insert timestamp if there is room */
9056                         switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9057                         case IPOPT_TS_TSONLY:
9058                                 off = IPOPT_TS_TIMELEN;
9059                                 break;
9060                         case IPOPT_TS_PRESPEC:
9061                         case IPOPT_TS_PRESPEC_RFC791:
9062                                 /* Verify that the address matched */
9063                                 off = opt[IPOPT_OFFSET] - 1;
9064                                 bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
9065                                 if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
9066                                         /* Not for us */
9067                                         break;
9068                                 }
9069                                 /* FALLTHRU */
9070                         case IPOPT_TS_TSANDADDR:
9071                                 off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
9072                                 break;
9073                         default:
9074                                 /*
9075                                  * ip_*put_options should have already
9076                                  * dropped this packet.
9077                                  */
9078                                 cmn_err(CE_PANIC, "ip_forward_options: "
9079                                     "unknown IT - bug in ip_input_options?\n");
9080                                 return (B_TRUE);        /* Keep "lint" happy */
9081                         }
9082                         if (opt[IPOPT_OFFSET] - 1 + off > optlen) {
9083                                 /* Increase overflow counter */
9084                                 off = (opt[IPOPT_POS_OV_FLG] >> 4) + 1;
9085                                 opt[IPOPT_POS_OV_FLG] =
9086                                     (uint8_t)((opt[IPOPT_POS_OV_FLG] & 0x0F) |
9087                                     (off << 4));
9088                                 break;
9089                         }
9090                         off = opt[IPOPT_OFFSET] - 1;
9091                         switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9092                         case IPOPT_TS_PRESPEC:
9093                         case IPOPT_TS_PRESPEC_RFC791:
9094                         case IPOPT_TS_TSANDADDR:
9095                                 /* Pick a reasonable addr on the outbound if */
9096                                 ASSERT(dst_ill != NULL);
9097                                 if (ip_select_source_v4(dst_ill, INADDR_ANY,
9098                                     dst, INADDR_ANY, ALL_ZONES, ipst, &ifaddr,
9099                                     NULL, NULL) != 0) {
9100                                         /* No source! Shouldn't happen */
9101                                         ifaddr = INADDR_ANY;
9102                                 }
9103                                 bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9104                                 opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9105                                 /* FALLTHRU */
9106                         case IPOPT_TS_TSONLY:
9107                                 off = opt[IPOPT_OFFSET] - 1;
9108                                 /* Compute # of milliseconds since midnight */
9109                                 gethrestime(&now);
9110                                 ts = (now.tv_sec % (24 * 60 * 60)) * 1000 +
9111                                     now.tv_nsec / (NANOSEC / MILLISEC);
9112                                 bcopy(&ts, (char *)opt + off, IPOPT_TS_TIMELEN);
9113                                 opt[IPOPT_OFFSET] += IPOPT_TS_TIMELEN;
9114                                 break;
9115                         }
9116                         break;
9117                 }
9118         }
9119         return (B_TRUE);
9120 }
9121 
9122 /*
9123  * Call ill_frag_timeout to do garbage collection. ill_frag_timeout
9124  * returns 'true' if there are still fragments left on the queue, in
9125  * which case we restart the timer.
9126  */
9127 void
9128 ill_frag_timer(void *arg)
9129 {
9130         ill_t   *ill = (ill_t *)arg;
9131         boolean_t frag_pending;
9132         ip_stack_t *ipst = ill->ill_ipst;
9133         time_t  timeout;
9134 
9135         mutex_enter(&ill->ill_lock);
9136         ASSERT(!ill->ill_fragtimer_executing);
9137         if (ill->ill_state_flags & ILL_CONDEMNED) {
9138                 ill->ill_frag_timer_id = 0;
9139                 mutex_exit(&ill->ill_lock);
9140                 return;
9141         }
9142         ill->ill_fragtimer_executing = 1;
9143         mutex_exit(&ill->ill_lock);
9144 
9145         timeout = (ill->ill_isv6 ? ipst->ips_ipv6_reassembly_timeout :
9146             ipst->ips_ip_reassembly_timeout);
9147 
9148         frag_pending = ill_frag_timeout(ill, timeout);
9149 
9150         /*
9151          * Restart the timer, if we have fragments pending or if someone
9152          * wanted us to be scheduled again.
9153          */
9154         mutex_enter(&ill->ill_lock);
9155         ill->ill_fragtimer_executing = 0;
9156         ill->ill_frag_timer_id = 0;
9157         if (frag_pending || ill->ill_fragtimer_needrestart)
9158                 ill_frag_timer_start(ill);
9159         mutex_exit(&ill->ill_lock);
9160 }
9161 
9162 void
9163 ill_frag_timer_start(ill_t *ill)
9164 {
9165         ip_stack_t *ipst = ill->ill_ipst;
9166         clock_t timeo_ms;
9167 
9168         ASSERT(MUTEX_HELD(&ill->ill_lock));
9169 
9170         /* If the ill is closing or opening don't proceed */
9171         if (ill->ill_state_flags & ILL_CONDEMNED)
9172                 return;
9173 
9174         if (ill->ill_fragtimer_executing) {
9175                 /*
9176                  * ill_frag_timer is currently executing. Just record the
9177                  * the fact that we want the timer to be restarted.
9178                  * ill_frag_timer will post a timeout before it returns,
9179                  * ensuring it will be called again.
9180                  */
9181                 ill->ill_fragtimer_needrestart = 1;
9182                 return;
9183         }
9184 
9185         if (ill->ill_frag_timer_id == 0) {
9186                 timeo_ms = (ill->ill_isv6 ? ipst->ips_ipv6_reassembly_timeout :
9187                     ipst->ips_ip_reassembly_timeout) * SECONDS;
9188 
9189                 /*
9190                  * The timer is neither running nor is the timeout handler
9191                  * executing. Post a timeout so that ill_frag_timer will be
9192                  * called
9193                  */
9194                 ill->ill_frag_timer_id = timeout(ill_frag_timer, ill,
9195                     MSEC_TO_TICK(timeo_ms >> 1));
9196                 ill->ill_fragtimer_needrestart = 0;
9197         }
9198 }
9199 
9200 /*
9201  * Update any source route, record route or timestamp options.
9202  * Check that we are at end of strict source route.
9203  * The options have already been checked for sanity in ip_input_options().
9204  */
9205 boolean_t
9206 ip_input_local_options(mblk_t *mp, ipha_t *ipha, ip_recv_attr_t *ira)
9207 {
9208         ipoptp_t        opts;
9209         uchar_t         *opt;
9210         uint8_t         optval;
9211         uint8_t         optlen;
9212         ipaddr_t        dst;
9213         ipaddr_t        ifaddr;
9214         uint32_t        ts;
9215         timestruc_t     now;
9216         ill_t           *ill = ira->ira_ill;
9217         ip_stack_t      *ipst = ill->ill_ipst;
9218 
9219         ip2dbg(("ip_input_local_options\n"));
9220 
9221         for (optval = ipoptp_first(&opts, ipha);
9222             optval != IPOPT_EOL;
9223             optval = ipoptp_next(&opts)) {
9224                 ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
9225                 opt = opts.ipoptp_cur;
9226                 optlen = opts.ipoptp_len;
9227                 ip2dbg(("ip_input_local_options: opt %d, len %d\n",
9228                     optval, optlen));
9229                 switch (optval) {
9230                         uint32_t off;
9231                 case IPOPT_SSRR:
9232                 case IPOPT_LSRR:
9233                         off = opt[IPOPT_OFFSET];
9234                         off--;
9235                         if (optlen < IP_ADDR_LEN ||
9236                             off > optlen - IP_ADDR_LEN) {
9237                                 /* End of source route */
9238                                 ip1dbg(("ip_input_local_options: end of SR\n"));
9239                                 break;
9240                         }
9241                         /*
9242                          * This will only happen if two consecutive entries
9243                          * in the source route contains our address or if
9244                          * it is a packet with a loose source route which
9245                          * reaches us before consuming the whole source route
9246                          */
9247                         ip1dbg(("ip_input_local_options: not end of SR\n"));
9248                         if (optval == IPOPT_SSRR) {
9249                                 goto bad_src_route;
9250                         }
9251                         /*
9252                          * Hack: instead of dropping the packet truncate the
9253                          * source route to what has been used by filling the
9254                          * rest with IPOPT_NOP.
9255                          */
9256                         opt[IPOPT_OLEN] = (uint8_t)off;
9257                         while (off < optlen) {
9258                                 opt[off++] = IPOPT_NOP;
9259                         }
9260                         break;
9261                 case IPOPT_RR:
9262                         off = opt[IPOPT_OFFSET];
9263                         off--;
9264                         if (optlen < IP_ADDR_LEN ||
9265                             off > optlen - IP_ADDR_LEN) {
9266                                 /* No more room - ignore */
9267                                 ip1dbg((
9268                                     "ip_input_local_options: end of RR\n"));
9269                                 break;
9270                         }
9271                         /* Pick a reasonable address on the outbound if */
9272                         if (ip_select_source_v4(ill, INADDR_ANY, ipha->ipha_dst,
9273                             INADDR_ANY, ALL_ZONES, ipst, &ifaddr, NULL,
9274                             NULL) != 0) {
9275                                 /* No source! Shouldn't happen */
9276                                 ifaddr = INADDR_ANY;
9277                         }
9278                         bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9279                         opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9280                         break;
9281                 case IPOPT_TS:
9282                         /* Insert timestamp if there is romm */
9283                         switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9284                         case IPOPT_TS_TSONLY:
9285                                 off = IPOPT_TS_TIMELEN;
9286                                 break;
9287                         case IPOPT_TS_PRESPEC:
9288                         case IPOPT_TS_PRESPEC_RFC791:
9289                                 /* Verify that the address matched */
9290                                 off = opt[IPOPT_OFFSET] - 1;
9291                                 bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
9292                                 if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
9293                                         /* Not for us */
9294                                         break;
9295                                 }
9296                                 /* FALLTHRU */
9297                         case IPOPT_TS_TSANDADDR:
9298                                 off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
9299                                 break;
9300                         default:
9301                                 /*
9302                                  * ip_*put_options should have already
9303                                  * dropped this packet.
9304                                  */
9305                                 cmn_err(CE_PANIC, "ip_input_local_options: "
9306                                     "unknown IT - bug in ip_input_options?\n");
9307                                 return (B_TRUE);        /* Keep "lint" happy */
9308                         }
9309                         if (opt[IPOPT_OFFSET] - 1 + off > optlen) {
9310                                 /* Increase overflow counter */
9311                                 off = (opt[IPOPT_POS_OV_FLG] >> 4) + 1;
9312                                 opt[IPOPT_POS_OV_FLG] =
9313                                     (uint8_t)((opt[IPOPT_POS_OV_FLG] & 0x0F) |
9314                                     (off << 4));
9315                                 break;
9316                         }
9317                         off = opt[IPOPT_OFFSET] - 1;
9318                         switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9319                         case IPOPT_TS_PRESPEC:
9320                         case IPOPT_TS_PRESPEC_RFC791:
9321                         case IPOPT_TS_TSANDADDR:
9322                                 /* Pick a reasonable addr on the outbound if */
9323                                 if (ip_select_source_v4(ill, INADDR_ANY,
9324                                     ipha->ipha_dst, INADDR_ANY, ALL_ZONES, ipst,
9325                                     &ifaddr, NULL, NULL) != 0) {
9326                                         /* No source! Shouldn't happen */
9327                                         ifaddr = INADDR_ANY;
9328                                 }
9329                                 bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9330                                 opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9331                                 /* FALLTHRU */
9332                         case IPOPT_TS_TSONLY:
9333                                 off = opt[IPOPT_OFFSET] - 1;
9334                                 /* Compute # of milliseconds since midnight */
9335                                 gethrestime(&now);
9336                                 ts = (now.tv_sec % (24 * 60 * 60)) * 1000 +
9337                                     now.tv_nsec / (NANOSEC / MILLISEC);
9338                                 bcopy(&ts, (char *)opt + off, IPOPT_TS_TIMELEN);
9339                                 opt[IPOPT_OFFSET] += IPOPT_TS_TIMELEN;
9340                                 break;
9341                         }
9342                         break;
9343                 }
9344         }
9345         return (B_TRUE);
9346 
9347 bad_src_route:
9348         /* make sure we clear any indication of a hardware checksum */
9349         DB_CKSUMFLAGS(mp) = 0;
9350         ip_drop_input("ICMP_SOURCE_ROUTE_FAILED", mp, ill);
9351         icmp_unreachable(mp, ICMP_SOURCE_ROUTE_FAILED, ira);
9352         return (B_FALSE);
9353 
9354 }
9355 
9356 /*
9357  * Process IP options in an inbound packet.  Always returns the nexthop.
9358  * Normally this is the passed in nexthop, but if there is an option
9359  * that effects the nexthop (such as a source route) that will be returned.
9360  * Sets *errorp if there is an error, in which case an ICMP error has been sent
9361  * and mp freed.
9362  */
9363 ipaddr_t
9364 ip_input_options(ipha_t *ipha, ipaddr_t dst, mblk_t *mp,
9365     ip_recv_attr_t *ira, int *errorp)
9366 {
9367         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
9368         ipoptp_t        opts;
9369         uchar_t         *opt;
9370         uint8_t         optval;
9371         uint8_t         optlen;
9372         intptr_t        code = 0;
9373         ire_t           *ire;
9374 
9375         ip2dbg(("ip_input_options\n"));
9376         *errorp = 0;
9377         for (optval = ipoptp_first(&opts, ipha);
9378             optval != IPOPT_EOL;
9379             optval = ipoptp_next(&opts)) {
9380                 opt = opts.ipoptp_cur;
9381                 optlen = opts.ipoptp_len;
9382                 ip2dbg(("ip_input_options: opt %d, len %d\n",
9383                     optval, optlen));
9384                 /*
9385                  * Note: we need to verify the checksum before we
9386                  * modify anything thus this routine only extracts the next
9387                  * hop dst from any source route.
9388                  */
9389                 switch (optval) {
9390                         uint32_t off;
9391                 case IPOPT_SSRR:
9392                 case IPOPT_LSRR:
9393                         if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
9394                                 if (optval == IPOPT_SSRR) {
9395                                         ip1dbg(("ip_input_options: not next"
9396                                             " strict source route 0x%x\n",
9397                                             ntohl(dst)));
9398                                         code = (char *)&ipha->ipha_dst -
9399                                             (char *)ipha;
9400                                         goto param_prob; /* RouterReq's */
9401                                 }
9402                                 ip2dbg(("ip_input_options: "
9403                                     "not next source route 0x%x\n",
9404                                     ntohl(dst)));
9405                                 break;
9406                         }
9407 
9408                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
9409                                 ip1dbg((
9410                                     "ip_input_options: bad option offset\n"));
9411                                 code = (char *)&opt[IPOPT_OLEN] -
9412                                     (char *)ipha;
9413                                 goto param_prob;
9414                         }
9415                         off = opt[IPOPT_OFFSET];
9416                         off--;
9417                 redo_srr:
9418                         if (optlen < IP_ADDR_LEN ||
9419                             off > optlen - IP_ADDR_LEN) {
9420                                 /* End of source route */
9421                                 ip1dbg(("ip_input_options: end of SR\n"));
9422                                 break;
9423                         }
9424                         bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
9425                         ip1dbg(("ip_input_options: next hop 0x%x\n",
9426                             ntohl(dst)));
9427 
9428                         /*
9429                          * Check if our address is present more than
9430                          * once as consecutive hops in source route.
9431                          * XXX verify per-interface ip_forwarding
9432                          * for source route?
9433                          */
9434                         if (ip_type_v4(dst, ipst) == IRE_LOCAL) {
9435                                 off += IP_ADDR_LEN;
9436                                 goto redo_srr;
9437                         }
9438 
9439                         if (dst == htonl(INADDR_LOOPBACK)) {
9440                                 ip1dbg(("ip_input_options: loopback addr in "
9441                                     "source route!\n"));
9442                                 goto bad_src_route;
9443                         }
9444                         /*
9445                          * For strict: verify that dst is directly
9446                          * reachable.
9447                          */
9448                         if (optval == IPOPT_SSRR) {
9449                                 ire = ire_ftable_lookup_v4(dst, 0, 0,
9450                                     IRE_INTERFACE, NULL, ALL_ZONES,
9451                                     ira->ira_tsl,
9452                                     MATCH_IRE_TYPE | MATCH_IRE_SECATTR, 0, ipst,
9453                                     NULL);
9454                                 if (ire == NULL) {
9455                                         ip1dbg(("ip_input_options: SSRR not "
9456                                             "directly reachable: 0x%x\n",
9457                                             ntohl(dst)));
9458                                         goto bad_src_route;
9459                                 }
9460                                 ire_refrele(ire);
9461                         }
9462                         /*
9463                          * Defer update of the offset and the record route
9464                          * until the packet is forwarded.
9465                          */
9466                         break;
9467                 case IPOPT_RR:
9468                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
9469                                 ip1dbg((
9470                                     "ip_input_options: bad option offset\n"));
9471                                 code = (char *)&opt[IPOPT_OLEN] -
9472                                     (char *)ipha;
9473                                 goto param_prob;
9474                         }
9475                         break;
9476                 case IPOPT_TS:
9477                         /*
9478                          * Verify that length >= 5 and that there is either
9479                          * room for another timestamp or that the overflow
9480                          * counter is not maxed out.
9481                          */
9482                         code = (char *)&opt[IPOPT_OLEN] - (char *)ipha;
9483                         if (optlen < IPOPT_MINLEN_IT) {
9484                                 goto param_prob;
9485                         }
9486                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
9487                                 ip1dbg((
9488                                     "ip_input_options: bad option offset\n"));
9489                                 code = (char *)&opt[IPOPT_OFFSET] -
9490                                     (char *)ipha;
9491                                 goto param_prob;
9492                         }
9493                         switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9494                         case IPOPT_TS_TSONLY:
9495                                 off = IPOPT_TS_TIMELEN;
9496                                 break;
9497                         case IPOPT_TS_TSANDADDR:
9498                         case IPOPT_TS_PRESPEC:
9499                         case IPOPT_TS_PRESPEC_RFC791:
9500                                 off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
9501                                 break;
9502                         default:
9503                                 code = (char *)&opt[IPOPT_POS_OV_FLG] -
9504                                     (char *)ipha;
9505                                 goto param_prob;
9506                         }
9507                         if (opt[IPOPT_OFFSET] - 1 + off > optlen &&
9508                             (opt[IPOPT_POS_OV_FLG] & 0xF0) == 0xF0) {
9509                                 /*
9510                                  * No room and the overflow counter is 15
9511                                  * already.
9512                                  */
9513                                 goto param_prob;
9514                         }
9515                         break;
9516                 }
9517         }
9518 
9519         if ((opts.ipoptp_flags & IPOPTP_ERROR) == 0) {
9520                 return (dst);
9521         }
9522 
9523         ip1dbg(("ip_input_options: error processing IP options."));
9524         code = (char *)&opt[IPOPT_OFFSET] - (char *)ipha;
9525 
9526 param_prob:
9527         /* make sure we clear any indication of a hardware checksum */
9528         DB_CKSUMFLAGS(mp) = 0;
9529         ip_drop_input("ICMP_PARAM_PROBLEM", mp, ira->ira_ill);
9530         icmp_param_problem(mp, (uint8_t)code, ira);
9531         *errorp = -1;
9532         return (dst);
9533 
9534 bad_src_route:
9535         /* make sure we clear any indication of a hardware checksum */
9536         DB_CKSUMFLAGS(mp) = 0;
9537         ip_drop_input("ICMP_SOURCE_ROUTE_FAILED", mp, ira->ira_ill);
9538         icmp_unreachable(mp, ICMP_SOURCE_ROUTE_FAILED, ira);
9539         *errorp = -1;
9540         return (dst);
9541 }
9542 
9543 /*
9544  * IP & ICMP info in >=14 msg's ...
9545  *  - ip fixed part (mib2_ip_t)
9546  *  - icmp fixed part (mib2_icmp_t)
9547  *  - ipAddrEntryTable (ip 20)          all IPv4 ipifs
9548  *  - ipRouteEntryTable (ip 21)         all IPv4 IREs
9549  *  - ipNetToMediaEntryTable (ip 22)    all IPv4 Neighbor Cache entries
9550  *  - ipRouteAttributeTable (ip 102)    labeled routes
9551  *  - ip multicast membership (ip_member_t)
9552  *  - ip multicast source filtering (ip_grpsrc_t)
9553  *  - igmp fixed part (struct igmpstat)
9554  *  - multicast routing stats (struct mrtstat)
9555  *  - multicast routing vifs (array of struct vifctl)
9556  *  - multicast routing routes (array of struct mfcctl)
9557  *  - ip6 fixed part (mib2_ipv6IfStatsEntry_t)
9558  *                                      One per ill plus one generic
9559  *  - icmp6 fixed part (mib2_ipv6IfIcmpEntry_t)
9560  *                                      One per ill plus one generic
9561  *  - ipv6RouteEntry                    all IPv6 IREs
9562  *  - ipv6RouteAttributeTable (ip6 102) labeled routes
9563  *  - ipv6NetToMediaEntry               all IPv6 Neighbor Cache entries
9564  *  - ipv6AddrEntry                     all IPv6 ipifs
9565  *  - ipv6 multicast membership (ipv6_member_t)
9566  *  - ipv6 multicast source filtering (ipv6_grpsrc_t)
9567  *
9568  * NOTE: original mpctl is copied for msg's 2..N, since its ctl part is
9569  * already filled in by the caller.
9570  * If legacy_req is true then MIB structures needs to be truncated to their
9571  * legacy sizes before being returned.
9572  * Return value of 0 indicates that no messages were sent and caller
9573  * should free mpctl.
9574  */
9575 int
9576 ip_snmp_get(queue_t *q, mblk_t *mpctl, int level, boolean_t legacy_req)
9577 {
9578         ip_stack_t *ipst;
9579         sctp_stack_t *sctps;
9580 
9581         if (q->q_next != NULL) {
9582                 ipst = ILLQ_TO_IPST(q);
9583         } else {
9584                 ipst = CONNQ_TO_IPST(q);
9585         }
9586         ASSERT(ipst != NULL);
9587         sctps = ipst->ips_netstack->netstack_sctp;
9588 
9589         if (mpctl == NULL || mpctl->b_cont == NULL) {
9590                 return (0);
9591         }
9592 
9593         /*
9594          * For the purposes of the (broken) packet shell use
9595          * of the level we make sure MIB2_TCP/MIB2_UDP can be used
9596          * to make TCP and UDP appear first in the list of mib items.
9597          * TBD: We could expand this and use it in netstat so that
9598          * the kernel doesn't have to produce large tables (connections,
9599          * routes, etc) when netstat only wants the statistics or a particular
9600          * table.
9601          */
9602         if (!(level == MIB2_TCP || level == MIB2_UDP)) {
9603                 if ((mpctl = icmp_snmp_get(q, mpctl)) == NULL) {
9604                         return (1);
9605                 }
9606         }
9607 
9608         if (level != MIB2_TCP) {
9609                 if ((mpctl = udp_snmp_get(q, mpctl, legacy_req)) == NULL) {
9610                         return (1);
9611                 }
9612         }
9613 
9614         if (level != MIB2_UDP) {
9615                 if ((mpctl = tcp_snmp_get(q, mpctl, legacy_req)) == NULL) {
9616                         return (1);
9617                 }
9618         }
9619 
9620         if ((mpctl = ip_snmp_get_mib2_ip_traffic_stats(q, mpctl,
9621             ipst, legacy_req)) == NULL) {
9622                 return (1);
9623         }
9624 
9625         if ((mpctl = ip_snmp_get_mib2_ip6(q, mpctl, ipst,
9626             legacy_req)) == NULL) {
9627                 return (1);
9628         }
9629 
9630         if ((mpctl = ip_snmp_get_mib2_icmp(q, mpctl, ipst)) == NULL) {
9631                 return (1);
9632         }
9633 
9634         if ((mpctl = ip_snmp_get_mib2_icmp6(q, mpctl, ipst)) == NULL) {
9635                 return (1);
9636         }
9637 
9638         if ((mpctl = ip_snmp_get_mib2_igmp(q, mpctl, ipst)) == NULL) {
9639                 return (1);
9640         }
9641 
9642         if ((mpctl = ip_snmp_get_mib2_multi(q, mpctl, ipst)) == NULL) {
9643                 return (1);
9644         }
9645 
9646         if ((mpctl = ip_snmp_get_mib2_ip_addr(q, mpctl, ipst,
9647             legacy_req)) == NULL) {
9648                 return (1);
9649         }
9650 
9651         if ((mpctl = ip_snmp_get_mib2_ip6_addr(q, mpctl, ipst,
9652             legacy_req)) == NULL) {
9653                 return (1);
9654         }
9655 
9656         if ((mpctl = ip_snmp_get_mib2_ip_group_mem(q, mpctl, ipst)) == NULL) {
9657                 return (1);
9658         }
9659 
9660         if ((mpctl = ip_snmp_get_mib2_ip6_group_mem(q, mpctl, ipst)) == NULL) {
9661                 return (1);
9662         }
9663 
9664         if ((mpctl = ip_snmp_get_mib2_ip_group_src(q, mpctl, ipst)) == NULL) {
9665                 return (1);
9666         }
9667 
9668         if ((mpctl = ip_snmp_get_mib2_ip6_group_src(q, mpctl, ipst)) == NULL) {
9669                 return (1);
9670         }
9671 
9672         if ((mpctl = ip_snmp_get_mib2_virt_multi(q, mpctl, ipst)) == NULL) {
9673                 return (1);
9674         }
9675 
9676         if ((mpctl = ip_snmp_get_mib2_multi_rtable(q, mpctl, ipst)) == NULL) {
9677                 return (1);
9678         }
9679 
9680         mpctl = ip_snmp_get_mib2_ip_route_media(q, mpctl, level, ipst);
9681         if (mpctl == NULL)
9682                 return (1);
9683 
9684         mpctl = ip_snmp_get_mib2_ip6_route_media(q, mpctl, level, ipst);
9685         if (mpctl == NULL)
9686                 return (1);
9687 
9688         if ((mpctl = sctp_snmp_get_mib2(q, mpctl, sctps)) == NULL) {
9689                 return (1);
9690         }
9691 
9692         if ((mpctl = ip_snmp_get_mib2_ip_dce(q, mpctl, ipst)) == NULL) {
9693                 return (1);
9694         }
9695 
9696         if ((mpctl = dccp_snmp_get(q, mpctl, legacy_req)) == NULL) {
9697                 return (1);
9698         }
9699 
9700         freemsg(mpctl);
9701         return (1);
9702 }
9703 
9704 /* Get global (legacy) IPv4 statistics */
9705 static mblk_t *
9706 ip_snmp_get_mib2_ip(queue_t *q, mblk_t *mpctl, mib2_ipIfStatsEntry_t *ipmib,
9707     ip_stack_t *ipst, boolean_t legacy_req)
9708 {
9709         mib2_ip_t               old_ip_mib;
9710         struct opthdr           *optp;
9711         mblk_t                  *mp2ctl;
9712         mib2_ipAddrEntry_t      mae;
9713 
9714         /*
9715          * make a copy of the original message
9716          */
9717         mp2ctl = copymsg(mpctl);
9718 
9719         /* fixed length IP structure... */
9720         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9721         optp->level = MIB2_IP;
9722         optp->name = 0;
9723         SET_MIB(old_ip_mib.ipForwarding,
9724             (WE_ARE_FORWARDING(ipst) ? 1 : 2));
9725         SET_MIB(old_ip_mib.ipDefaultTTL,
9726             (uint32_t)ipst->ips_ip_def_ttl);
9727         SET_MIB(old_ip_mib.ipReasmTimeout,
9728             ipst->ips_ip_reassembly_timeout);
9729         SET_MIB(old_ip_mib.ipAddrEntrySize,
9730             (legacy_req) ? LEGACY_MIB_SIZE(&mae, mib2_ipAddrEntry_t) :
9731             sizeof (mib2_ipAddrEntry_t));
9732         SET_MIB(old_ip_mib.ipRouteEntrySize,
9733             sizeof (mib2_ipRouteEntry_t));
9734         SET_MIB(old_ip_mib.ipNetToMediaEntrySize,
9735             sizeof (mib2_ipNetToMediaEntry_t));
9736         SET_MIB(old_ip_mib.ipMemberEntrySize, sizeof (ip_member_t));
9737         SET_MIB(old_ip_mib.ipGroupSourceEntrySize, sizeof (ip_grpsrc_t));
9738         SET_MIB(old_ip_mib.ipRouteAttributeSize,
9739             sizeof (mib2_ipAttributeEntry_t));
9740         SET_MIB(old_ip_mib.transportMLPSize, sizeof (mib2_transportMLPEntry_t));
9741         SET_MIB(old_ip_mib.ipDestEntrySize, sizeof (dest_cache_entry_t));
9742 
9743         /*
9744          * Grab the statistics from the new IP MIB
9745          */
9746         SET_MIB(old_ip_mib.ipInReceives,
9747             (uint32_t)ipmib->ipIfStatsHCInReceives);
9748         SET_MIB(old_ip_mib.ipInHdrErrors, ipmib->ipIfStatsInHdrErrors);
9749         SET_MIB(old_ip_mib.ipInAddrErrors, ipmib->ipIfStatsInAddrErrors);
9750         SET_MIB(old_ip_mib.ipForwDatagrams,
9751             (uint32_t)ipmib->ipIfStatsHCOutForwDatagrams);
9752         SET_MIB(old_ip_mib.ipInUnknownProtos,
9753             ipmib->ipIfStatsInUnknownProtos);
9754         SET_MIB(old_ip_mib.ipInDiscards, ipmib->ipIfStatsInDiscards);
9755         SET_MIB(old_ip_mib.ipInDelivers,
9756             (uint32_t)ipmib->ipIfStatsHCInDelivers);
9757         SET_MIB(old_ip_mib.ipOutRequests,
9758             (uint32_t)ipmib->ipIfStatsHCOutRequests);
9759         SET_MIB(old_ip_mib.ipOutDiscards, ipmib->ipIfStatsOutDiscards);
9760         SET_MIB(old_ip_mib.ipOutNoRoutes, ipmib->ipIfStatsOutNoRoutes);
9761         SET_MIB(old_ip_mib.ipReasmReqds, ipmib->ipIfStatsReasmReqds);
9762         SET_MIB(old_ip_mib.ipReasmOKs, ipmib->ipIfStatsReasmOKs);
9763         SET_MIB(old_ip_mib.ipReasmFails, ipmib->ipIfStatsReasmFails);
9764         SET_MIB(old_ip_mib.ipFragOKs, ipmib->ipIfStatsOutFragOKs);
9765         SET_MIB(old_ip_mib.ipFragFails, ipmib->ipIfStatsOutFragFails);
9766         SET_MIB(old_ip_mib.ipFragCreates, ipmib->ipIfStatsOutFragCreates);
9767 
9768         /* ipRoutingDiscards is not being used */
9769         SET_MIB(old_ip_mib.ipRoutingDiscards, 0);
9770         SET_MIB(old_ip_mib.tcpInErrs, ipmib->tcpIfStatsInErrs);
9771         SET_MIB(old_ip_mib.udpNoPorts, ipmib->udpIfStatsNoPorts);
9772         SET_MIB(old_ip_mib.ipInCksumErrs, ipmib->ipIfStatsInCksumErrs);
9773         SET_MIB(old_ip_mib.ipReasmDuplicates,
9774             ipmib->ipIfStatsReasmDuplicates);
9775         SET_MIB(old_ip_mib.ipReasmPartDups, ipmib->ipIfStatsReasmPartDups);
9776         SET_MIB(old_ip_mib.ipForwProhibits, ipmib->ipIfStatsForwProhibits);
9777         SET_MIB(old_ip_mib.udpInCksumErrs, ipmib->udpIfStatsInCksumErrs);
9778         SET_MIB(old_ip_mib.udpInOverflows, ipmib->udpIfStatsInOverflows);
9779         SET_MIB(old_ip_mib.rawipInOverflows,
9780             ipmib->rawipIfStatsInOverflows);
9781 
9782         SET_MIB(old_ip_mib.ipsecInSucceeded, ipmib->ipsecIfStatsInSucceeded);
9783         SET_MIB(old_ip_mib.ipsecInFailed, ipmib->ipsecIfStatsInFailed);
9784         SET_MIB(old_ip_mib.ipInIPv6, ipmib->ipIfStatsInWrongIPVersion);
9785         SET_MIB(old_ip_mib.ipOutIPv6, ipmib->ipIfStatsOutWrongIPVersion);
9786         SET_MIB(old_ip_mib.ipOutSwitchIPv6,
9787             ipmib->ipIfStatsOutSwitchIPVersion);
9788 
9789         if (!snmp_append_data(mpctl->b_cont, (char *)&old_ip_mib,
9790             (int)sizeof (old_ip_mib))) {
9791                 ip1dbg(("ip_snmp_get_mib2_ip: failed to allocate %u bytes\n",
9792                     (uint_t)sizeof (old_ip_mib)));
9793         }
9794 
9795         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
9796         ip3dbg(("ip_snmp_get_mib2_ip: level %d, name %d, len %d\n",
9797             (int)optp->level, (int)optp->name, (int)optp->len));
9798         qreply(q, mpctl);
9799         return (mp2ctl);
9800 }
9801 
9802 /* Per interface IPv4 statistics */
9803 static mblk_t *
9804 ip_snmp_get_mib2_ip_traffic_stats(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst,
9805     boolean_t legacy_req)
9806 {
9807         struct opthdr           *optp;
9808         mblk_t                  *mp2ctl;
9809         ill_t                   *ill;
9810         ill_walk_context_t      ctx;
9811         mblk_t                  *mp_tail = NULL;
9812         mib2_ipIfStatsEntry_t   global_ip_mib;
9813         mib2_ipAddrEntry_t      mae;
9814 
9815         /*
9816          * Make a copy of the original message
9817          */
9818         mp2ctl = copymsg(mpctl);
9819 
9820         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9821         optp->level = MIB2_IP;
9822         optp->name = MIB2_IP_TRAFFIC_STATS;
9823         /* Include "unknown interface" ip_mib */
9824         ipst->ips_ip_mib.ipIfStatsIPVersion = MIB2_INETADDRESSTYPE_ipv4;
9825         ipst->ips_ip_mib.ipIfStatsIfIndex =
9826             MIB2_UNKNOWN_INTERFACE; /* Flag to netstat */
9827         SET_MIB(ipst->ips_ip_mib.ipIfStatsForwarding,
9828             (ipst->ips_ip_forwarding ? 1 : 2));
9829         SET_MIB(ipst->ips_ip_mib.ipIfStatsDefaultTTL,
9830             (uint32_t)ipst->ips_ip_def_ttl);
9831         SET_MIB(ipst->ips_ip_mib.ipIfStatsEntrySize,
9832             sizeof (mib2_ipIfStatsEntry_t));
9833         SET_MIB(ipst->ips_ip_mib.ipIfStatsAddrEntrySize,
9834             sizeof (mib2_ipAddrEntry_t));
9835         SET_MIB(ipst->ips_ip_mib.ipIfStatsRouteEntrySize,
9836             sizeof (mib2_ipRouteEntry_t));
9837         SET_MIB(ipst->ips_ip_mib.ipIfStatsNetToMediaEntrySize,
9838             sizeof (mib2_ipNetToMediaEntry_t));
9839         SET_MIB(ipst->ips_ip_mib.ipIfStatsMemberEntrySize,
9840             sizeof (ip_member_t));
9841         SET_MIB(ipst->ips_ip_mib.ipIfStatsGroupSourceEntrySize,
9842             sizeof (ip_grpsrc_t));
9843 
9844         bcopy(&ipst->ips_ip_mib, &global_ip_mib, sizeof (global_ip_mib));
9845 
9846         if (legacy_req) {
9847                 SET_MIB(global_ip_mib.ipIfStatsAddrEntrySize,
9848                     LEGACY_MIB_SIZE(&mae, mib2_ipAddrEntry_t));
9849         }
9850 
9851         if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
9852             (char *)&global_ip_mib, (int)sizeof (global_ip_mib))) {
9853                 ip1dbg(("ip_snmp_get_mib2_ip_traffic_stats: "
9854                     "failed to allocate %u bytes\n",
9855                     (uint_t)sizeof (global_ip_mib)));
9856         }
9857 
9858         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
9859         ill = ILL_START_WALK_V4(&ctx, ipst);
9860         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
9861                 ill->ill_ip_mib->ipIfStatsIfIndex =
9862                     ill->ill_phyint->phyint_ifindex;
9863                 SET_MIB(ill->ill_ip_mib->ipIfStatsForwarding,
9864                     (ipst->ips_ip_forwarding ? 1 : 2));
9865                 SET_MIB(ill->ill_ip_mib->ipIfStatsDefaultTTL,
9866                     (uint32_t)ipst->ips_ip_def_ttl);
9867 
9868                 ip_mib2_add_ip_stats(&global_ip_mib, ill->ill_ip_mib);
9869                 if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
9870                     (char *)ill->ill_ip_mib,
9871                     (int)sizeof (*ill->ill_ip_mib))) {
9872                         ip1dbg(("ip_snmp_get_mib2_ip_traffic_stats: "
9873                             "failed to allocate %u bytes\n",
9874                             (uint_t)sizeof (*ill->ill_ip_mib)));
9875                 }
9876         }
9877         rw_exit(&ipst->ips_ill_g_lock);
9878 
9879         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
9880         ip3dbg(("ip_snmp_get_mib2_ip_traffic_stats: "
9881             "level %d, name %d, len %d\n",
9882             (int)optp->level, (int)optp->name, (int)optp->len));
9883         qreply(q, mpctl);
9884 
9885         if (mp2ctl == NULL)
9886                 return (NULL);
9887 
9888         return (ip_snmp_get_mib2_ip(q, mp2ctl, &global_ip_mib, ipst,
9889             legacy_req));
9890 }
9891 
9892 /* Global IPv4 ICMP statistics */
9893 static mblk_t *
9894 ip_snmp_get_mib2_icmp(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
9895 {
9896         struct opthdr           *optp;
9897         mblk_t                  *mp2ctl;
9898 
9899         /*
9900          * Make a copy of the original message
9901          */
9902         mp2ctl = copymsg(mpctl);
9903 
9904         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9905         optp->level = MIB2_ICMP;
9906         optp->name = 0;
9907         if (!snmp_append_data(mpctl->b_cont, (char *)&ipst->ips_icmp_mib,
9908             (int)sizeof (ipst->ips_icmp_mib))) {
9909                 ip1dbg(("ip_snmp_get_mib2_icmp: failed to allocate %u bytes\n",
9910                     (uint_t)sizeof (ipst->ips_icmp_mib)));
9911         }
9912         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
9913         ip3dbg(("ip_snmp_get_mib2_icmp: level %d, name %d, len %d\n",
9914             (int)optp->level, (int)optp->name, (int)optp->len));
9915         qreply(q, mpctl);
9916         return (mp2ctl);
9917 }
9918 
9919 /* Global IPv4 IGMP statistics */
9920 static mblk_t *
9921 ip_snmp_get_mib2_igmp(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
9922 {
9923         struct opthdr           *optp;
9924         mblk_t                  *mp2ctl;
9925 
9926         /*
9927          * make a copy of the original message
9928          */
9929         mp2ctl = copymsg(mpctl);
9930 
9931         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9932         optp->level = EXPER_IGMP;
9933         optp->name = 0;
9934         if (!snmp_append_data(mpctl->b_cont, (char *)&ipst->ips_igmpstat,
9935             (int)sizeof (ipst->ips_igmpstat))) {
9936                 ip1dbg(("ip_snmp_get_mib2_igmp: failed to allocate %u bytes\n",
9937                     (uint_t)sizeof (ipst->ips_igmpstat)));
9938         }
9939         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
9940         ip3dbg(("ip_snmp_get_mib2_igmp: level %d, name %d, len %d\n",
9941             (int)optp->level, (int)optp->name, (int)optp->len));
9942         qreply(q, mpctl);
9943         return (mp2ctl);
9944 }
9945 
9946 /* Global IPv4 Multicast Routing statistics */
9947 static mblk_t *
9948 ip_snmp_get_mib2_multi(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
9949 {
9950         struct opthdr           *optp;
9951         mblk_t                  *mp2ctl;
9952 
9953         /*
9954          * make a copy of the original message
9955          */
9956         mp2ctl = copymsg(mpctl);
9957 
9958         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9959         optp->level = EXPER_DVMRP;
9960         optp->name = 0;
9961         if (!ip_mroute_stats(mpctl->b_cont, ipst)) {
9962                 ip0dbg(("ip_mroute_stats: failed\n"));
9963         }
9964         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
9965         ip3dbg(("ip_snmp_get_mib2_multi: level %d, name %d, len %d\n",
9966             (int)optp->level, (int)optp->name, (int)optp->len));
9967         qreply(q, mpctl);
9968         return (mp2ctl);
9969 }
9970 
9971 /* IPv4 address information */
9972 static mblk_t *
9973 ip_snmp_get_mib2_ip_addr(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst,
9974     boolean_t legacy_req)
9975 {
9976         struct opthdr           *optp;
9977         mblk_t                  *mp2ctl;
9978         mblk_t                  *mp_tail = NULL;
9979         ill_t                   *ill;
9980         ipif_t                  *ipif;
9981         uint_t                  bitval;
9982         mib2_ipAddrEntry_t      mae;
9983         size_t                  mae_size;
9984         zoneid_t                zoneid;
9985         ill_walk_context_t      ctx;
9986 
9987         /*
9988          * make a copy of the original message
9989          */
9990         mp2ctl = copymsg(mpctl);
9991 
9992         mae_size = (legacy_req) ? LEGACY_MIB_SIZE(&mae, mib2_ipAddrEntry_t) :
9993             sizeof (mib2_ipAddrEntry_t);
9994 
9995         /* ipAddrEntryTable */
9996 
9997         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9998         optp->level = MIB2_IP;
9999         optp->name = MIB2_IP_ADDR;
10000         zoneid = Q_TO_CONN(q)->conn_zoneid;
10001 
10002         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10003         ill = ILL_START_WALK_V4(&ctx, ipst);
10004         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10005                 for (ipif = ill->ill_ipif; ipif != NULL;
10006                     ipif = ipif->ipif_next) {
10007                         if (ipif->ipif_zoneid != zoneid &&
10008                             ipif->ipif_zoneid != ALL_ZONES)
10009                                 continue;
10010                         /* Sum of count from dead IRE_LO* and our current */
10011                         mae.ipAdEntInfo.ae_ibcnt = ipif->ipif_ib_pkt_count;
10012                         if (ipif->ipif_ire_local != NULL) {
10013                                 mae.ipAdEntInfo.ae_ibcnt +=
10014                                     ipif->ipif_ire_local->ire_ib_pkt_count;
10015                         }
10016                         mae.ipAdEntInfo.ae_obcnt = 0;
10017                         mae.ipAdEntInfo.ae_focnt = 0;
10018 
10019                         ipif_get_name(ipif, mae.ipAdEntIfIndex.o_bytes,
10020                             OCTET_LENGTH);
10021                         mae.ipAdEntIfIndex.o_length =
10022                             mi_strlen(mae.ipAdEntIfIndex.o_bytes);
10023                         mae.ipAdEntAddr = ipif->ipif_lcl_addr;
10024                         mae.ipAdEntNetMask = ipif->ipif_net_mask;
10025                         mae.ipAdEntInfo.ae_subnet = ipif->ipif_subnet;
10026                         mae.ipAdEntInfo.ae_subnet_len =
10027                             ip_mask_to_plen(ipif->ipif_net_mask);
10028                         mae.ipAdEntInfo.ae_src_addr = ipif->ipif_lcl_addr;
10029                         for (bitval = 1;
10030                             bitval &&
10031                             !(bitval & ipif->ipif_brd_addr);
10032                             bitval <<= 1)
10033                                 noop;
10034                         mae.ipAdEntBcastAddr = bitval;
10035                         mae.ipAdEntReasmMaxSize = IP_MAXPACKET;
10036                         mae.ipAdEntInfo.ae_mtu = ipif->ipif_ill->ill_mtu;
10037                         mae.ipAdEntInfo.ae_metric  = ipif->ipif_ill->ill_metric;
10038                         mae.ipAdEntInfo.ae_broadcast_addr =
10039                             ipif->ipif_brd_addr;
10040                         mae.ipAdEntInfo.ae_pp_dst_addr =
10041                             ipif->ipif_pp_dst_addr;
10042                         mae.ipAdEntInfo.ae_flags = ipif->ipif_flags |
10043                             ill->ill_flags | ill->ill_phyint->phyint_flags;
10044                         mae.ipAdEntRetransmitTime =
10045                             ill->ill_reachable_retrans_time;
10046 
10047                         if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10048                             (char *)&mae, (int)mae_size)) {
10049                                 ip1dbg(("ip_snmp_get_mib2_ip_addr: failed to "
10050                                     "allocate %u bytes\n", (uint_t)mae_size));
10051                         }
10052                 }
10053         }
10054         rw_exit(&ipst->ips_ill_g_lock);
10055 
10056         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10057         ip3dbg(("ip_snmp_get_mib2_ip_addr: level %d, name %d, len %d\n",
10058             (int)optp->level, (int)optp->name, (int)optp->len));
10059         qreply(q, mpctl);
10060         return (mp2ctl);
10061 }
10062 
10063 /* IPv6 address information */
10064 static mblk_t *
10065 ip_snmp_get_mib2_ip6_addr(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst,
10066     boolean_t legacy_req)
10067 {
10068         struct opthdr           *optp;
10069         mblk_t                  *mp2ctl;
10070         mblk_t                  *mp_tail = NULL;
10071         ill_t                   *ill;
10072         ipif_t                  *ipif;
10073         mib2_ipv6AddrEntry_t    mae6;
10074         size_t                  mae6_size;
10075         zoneid_t                zoneid;
10076         ill_walk_context_t      ctx;
10077 
10078         /*
10079          * make a copy of the original message
10080          */
10081         mp2ctl = copymsg(mpctl);
10082 
10083         mae6_size = (legacy_req) ?
10084             LEGACY_MIB_SIZE(&mae6, mib2_ipv6AddrEntry_t) :
10085             sizeof (mib2_ipv6AddrEntry_t);
10086 
10087         /* ipv6AddrEntryTable */
10088 
10089         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10090         optp->level = MIB2_IP6;
10091         optp->name = MIB2_IP6_ADDR;
10092         zoneid = Q_TO_CONN(q)->conn_zoneid;
10093 
10094         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10095         ill = ILL_START_WALK_V6(&ctx, ipst);
10096         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10097                 for (ipif = ill->ill_ipif; ipif != NULL;
10098                     ipif = ipif->ipif_next) {
10099                         if (ipif->ipif_zoneid != zoneid &&
10100                             ipif->ipif_zoneid != ALL_ZONES)
10101                                 continue;
10102                         /* Sum of count from dead IRE_LO* and our current */
10103                         mae6.ipv6AddrInfo.ae_ibcnt = ipif->ipif_ib_pkt_count;
10104                         if (ipif->ipif_ire_local != NULL) {
10105                                 mae6.ipv6AddrInfo.ae_ibcnt +=
10106                                     ipif->ipif_ire_local->ire_ib_pkt_count;
10107                         }
10108                         mae6.ipv6AddrInfo.ae_obcnt = 0;
10109                         mae6.ipv6AddrInfo.ae_focnt = 0;
10110 
10111                         ipif_get_name(ipif, mae6.ipv6AddrIfIndex.o_bytes,
10112                             OCTET_LENGTH);
10113                         mae6.ipv6AddrIfIndex.o_length =
10114                             mi_strlen(mae6.ipv6AddrIfIndex.o_bytes);
10115                         mae6.ipv6AddrAddress = ipif->ipif_v6lcl_addr;
10116                         mae6.ipv6AddrPfxLength =
10117                             ip_mask_to_plen_v6(&ipif->ipif_v6net_mask);
10118                         mae6.ipv6AddrInfo.ae_subnet = ipif->ipif_v6subnet;
10119                         mae6.ipv6AddrInfo.ae_subnet_len =
10120                             mae6.ipv6AddrPfxLength;
10121                         mae6.ipv6AddrInfo.ae_src_addr = ipif->ipif_v6lcl_addr;
10122 
10123                         /* Type: stateless(1), stateful(2), unknown(3) */
10124                         if (ipif->ipif_flags & IPIF_ADDRCONF)
10125                                 mae6.ipv6AddrType = 1;
10126                         else
10127                                 mae6.ipv6AddrType = 2;
10128                         /* Anycast: true(1), false(2) */
10129                         if (ipif->ipif_flags & IPIF_ANYCAST)
10130                                 mae6.ipv6AddrAnycastFlag = 1;
10131                         else
10132                                 mae6.ipv6AddrAnycastFlag = 2;
10133 
10134                         /*
10135                          * Address status: preferred(1), deprecated(2),
10136                          * invalid(3), inaccessible(4), unknown(5)
10137                          */
10138                         if (ipif->ipif_flags & IPIF_NOLOCAL)
10139                                 mae6.ipv6AddrStatus = 3;
10140                         else if (ipif->ipif_flags & IPIF_DEPRECATED)
10141                                 mae6.ipv6AddrStatus = 2;
10142                         else
10143                                 mae6.ipv6AddrStatus = 1;
10144                         mae6.ipv6AddrInfo.ae_mtu = ipif->ipif_ill->ill_mtu;
10145                         mae6.ipv6AddrInfo.ae_metric  =
10146                             ipif->ipif_ill->ill_metric;
10147                         mae6.ipv6AddrInfo.ae_pp_dst_addr =
10148                             ipif->ipif_v6pp_dst_addr;
10149                         mae6.ipv6AddrInfo.ae_flags = ipif->ipif_flags |
10150                             ill->ill_flags | ill->ill_phyint->phyint_flags;
10151                         mae6.ipv6AddrReasmMaxSize = IP_MAXPACKET;
10152                         mae6.ipv6AddrIdentifier = ill->ill_token;
10153                         mae6.ipv6AddrIdentifierLen = ill->ill_token_length;
10154                         mae6.ipv6AddrReachableTime = ill->ill_reachable_time;
10155                         mae6.ipv6AddrRetransmitTime =
10156                             ill->ill_reachable_retrans_time;
10157                         if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10158                             (char *)&mae6, (int)mae6_size)) {
10159                                 ip1dbg(("ip_snmp_get_mib2_ip6_addr: failed to "
10160                                     "allocate %u bytes\n",
10161                                     (uint_t)mae6_size));
10162                         }
10163                 }
10164         }
10165         rw_exit(&ipst->ips_ill_g_lock);
10166 
10167         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10168         ip3dbg(("ip_snmp_get_mib2_ip6_addr: level %d, name %d, len %d\n",
10169             (int)optp->level, (int)optp->name, (int)optp->len));
10170         qreply(q, mpctl);
10171         return (mp2ctl);
10172 }
10173 
10174 /* IPv4 multicast group membership. */
10175 static mblk_t *
10176 ip_snmp_get_mib2_ip_group_mem(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10177 {
10178         struct opthdr           *optp;
10179         mblk_t                  *mp2ctl;
10180         ill_t                   *ill;
10181         ipif_t                  *ipif;
10182         ilm_t                   *ilm;
10183         ip_member_t             ipm;
10184         mblk_t                  *mp_tail = NULL;
10185         ill_walk_context_t      ctx;
10186         zoneid_t                zoneid;
10187 
10188         /*
10189          * make a copy of the original message
10190          */
10191         mp2ctl = copymsg(mpctl);
10192         zoneid = Q_TO_CONN(q)->conn_zoneid;
10193 
10194         /* ipGroupMember table */
10195         optp = (struct opthdr *)&mpctl->b_rptr[
10196             sizeof (struct T_optmgmt_ack)];
10197         optp->level = MIB2_IP;
10198         optp->name = EXPER_IP_GROUP_MEMBERSHIP;
10199 
10200         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10201         ill = ILL_START_WALK_V4(&ctx, ipst);
10202         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10203                 /* Make sure the ill isn't going away. */
10204                 if (!ill_check_and_refhold(ill))
10205                         continue;
10206                 rw_exit(&ipst->ips_ill_g_lock);
10207                 rw_enter(&ill->ill_mcast_lock, RW_READER);
10208                 for (ilm = ill->ill_ilm; ilm; ilm = ilm->ilm_next) {
10209                         if (ilm->ilm_zoneid != zoneid &&
10210                             ilm->ilm_zoneid != ALL_ZONES)
10211                                 continue;
10212 
10213                         /* Is there an ipif for ilm_ifaddr? */
10214                         for (ipif = ill->ill_ipif; ipif != NULL;
10215                             ipif = ipif->ipif_next) {
10216                                 if (!IPIF_IS_CONDEMNED(ipif) &&
10217                                     ipif->ipif_lcl_addr == ilm->ilm_ifaddr &&
10218                                     ilm->ilm_ifaddr != INADDR_ANY)
10219                                         break;
10220                         }
10221                         if (ipif != NULL) {
10222                                 ipif_get_name(ipif,
10223                                     ipm.ipGroupMemberIfIndex.o_bytes,
10224                                     OCTET_LENGTH);
10225                         } else {
10226                                 ill_get_name(ill,
10227                                     ipm.ipGroupMemberIfIndex.o_bytes,
10228                                     OCTET_LENGTH);
10229                         }
10230                         ipm.ipGroupMemberIfIndex.o_length =
10231                             mi_strlen(ipm.ipGroupMemberIfIndex.o_bytes);
10232 
10233                         ipm.ipGroupMemberAddress = ilm->ilm_addr;
10234                         ipm.ipGroupMemberRefCnt = ilm->ilm_refcnt;
10235                         ipm.ipGroupMemberFilterMode = ilm->ilm_fmode;
10236                         if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10237                             (char *)&ipm, (int)sizeof (ipm))) {
10238                                 ip1dbg(("ip_snmp_get_mib2_ip_group: "
10239                                     "failed to allocate %u bytes\n",
10240                                     (uint_t)sizeof (ipm)));
10241                         }
10242                 }
10243                 rw_exit(&ill->ill_mcast_lock);
10244                 ill_refrele(ill);
10245                 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10246         }
10247         rw_exit(&ipst->ips_ill_g_lock);
10248         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10249         ip3dbg(("ip_snmp_get: level %d, name %d, len %d\n",
10250             (int)optp->level, (int)optp->name, (int)optp->len));
10251         qreply(q, mpctl);
10252         return (mp2ctl);
10253 }
10254 
10255 /* IPv6 multicast group membership. */
10256 static mblk_t *
10257 ip_snmp_get_mib2_ip6_group_mem(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10258 {
10259         struct opthdr           *optp;
10260         mblk_t                  *mp2ctl;
10261         ill_t                   *ill;
10262         ilm_t                   *ilm;
10263         ipv6_member_t           ipm6;
10264         mblk_t                  *mp_tail = NULL;
10265         ill_walk_context_t      ctx;
10266         zoneid_t                zoneid;
10267 
10268         /*
10269          * make a copy of the original message
10270          */
10271         mp2ctl = copymsg(mpctl);
10272         zoneid = Q_TO_CONN(q)->conn_zoneid;
10273 
10274         /* ip6GroupMember table */
10275         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10276         optp->level = MIB2_IP6;
10277         optp->name = EXPER_IP6_GROUP_MEMBERSHIP;
10278 
10279         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10280         ill = ILL_START_WALK_V6(&ctx, ipst);
10281         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10282                 /* Make sure the ill isn't going away. */
10283                 if (!ill_check_and_refhold(ill))
10284                         continue;
10285                 rw_exit(&ipst->ips_ill_g_lock);
10286                 /*
10287                  * Normally we don't have any members on under IPMP interfaces.
10288                  * We report them as a debugging aid.
10289                  */
10290                 rw_enter(&ill->ill_mcast_lock, RW_READER);
10291                 ipm6.ipv6GroupMemberIfIndex = ill->ill_phyint->phyint_ifindex;
10292                 for (ilm = ill->ill_ilm; ilm; ilm = ilm->ilm_next) {
10293                         if (ilm->ilm_zoneid != zoneid &&
10294                             ilm->ilm_zoneid != ALL_ZONES)
10295                                 continue;       /* not this zone */
10296                         ipm6.ipv6GroupMemberAddress = ilm->ilm_v6addr;
10297                         ipm6.ipv6GroupMemberRefCnt = ilm->ilm_refcnt;
10298                         ipm6.ipv6GroupMemberFilterMode = ilm->ilm_fmode;
10299                         if (!snmp_append_data2(mpctl->b_cont,
10300                             &mp_tail,
10301                             (char *)&ipm6, (int)sizeof (ipm6))) {
10302                                 ip1dbg(("ip_snmp_get_mib2_ip6_group: "
10303                                     "failed to allocate %u bytes\n",
10304                                     (uint_t)sizeof (ipm6)));
10305                         }
10306                 }
10307                 rw_exit(&ill->ill_mcast_lock);
10308                 ill_refrele(ill);
10309                 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10310         }
10311         rw_exit(&ipst->ips_ill_g_lock);
10312 
10313         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10314         ip3dbg(("ip_snmp_get: level %d, name %d, len %d\n",
10315             (int)optp->level, (int)optp->name, (int)optp->len));
10316         qreply(q, mpctl);
10317         return (mp2ctl);
10318 }
10319 
10320 /* IP multicast filtered sources */
10321 static mblk_t *
10322 ip_snmp_get_mib2_ip_group_src(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10323 {
10324         struct opthdr           *optp;
10325         mblk_t                  *mp2ctl;
10326         ill_t                   *ill;
10327         ipif_t                  *ipif;
10328         ilm_t                   *ilm;
10329         ip_grpsrc_t             ips;
10330         mblk_t                  *mp_tail = NULL;
10331         ill_walk_context_t      ctx;
10332         zoneid_t                zoneid;
10333         int                     i;
10334         slist_t                 *sl;
10335 
10336         /*
10337          * make a copy of the original message
10338          */
10339         mp2ctl = copymsg(mpctl);
10340         zoneid = Q_TO_CONN(q)->conn_zoneid;
10341 
10342         /* ipGroupSource table */
10343         optp = (struct opthdr *)&mpctl->b_rptr[
10344             sizeof (struct T_optmgmt_ack)];
10345         optp->level = MIB2_IP;
10346         optp->name = EXPER_IP_GROUP_SOURCES;
10347 
10348         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10349         ill = ILL_START_WALK_V4(&ctx, ipst);
10350         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10351                 /* Make sure the ill isn't going away. */
10352                 if (!ill_check_and_refhold(ill))
10353                         continue;
10354                 rw_exit(&ipst->ips_ill_g_lock);
10355                 rw_enter(&ill->ill_mcast_lock, RW_READER);
10356                 for (ilm = ill->ill_ilm; ilm; ilm = ilm->ilm_next) {
10357                         sl = ilm->ilm_filter;
10358                         if (ilm->ilm_zoneid != zoneid &&
10359                             ilm->ilm_zoneid != ALL_ZONES)
10360                                 continue;
10361                         if (SLIST_IS_EMPTY(sl))
10362                                 continue;
10363 
10364                         /* Is there an ipif for ilm_ifaddr? */
10365                         for (ipif = ill->ill_ipif; ipif != NULL;
10366                             ipif = ipif->ipif_next) {
10367                                 if (!IPIF_IS_CONDEMNED(ipif) &&
10368                                     ipif->ipif_lcl_addr == ilm->ilm_ifaddr &&
10369                                     ilm->ilm_ifaddr != INADDR_ANY)
10370                                         break;
10371                         }
10372                         if (ipif != NULL) {
10373                                 ipif_get_name(ipif,
10374                                     ips.ipGroupSourceIfIndex.o_bytes,
10375                                     OCTET_LENGTH);
10376                         } else {
10377                                 ill_get_name(ill,
10378                                     ips.ipGroupSourceIfIndex.o_bytes,
10379                                     OCTET_LENGTH);
10380                         }
10381                         ips.ipGroupSourceIfIndex.o_length =
10382                             mi_strlen(ips.ipGroupSourceIfIndex.o_bytes);
10383 
10384                         ips.ipGroupSourceGroup = ilm->ilm_addr;
10385                         for (i = 0; i < sl->sl_numsrc; i++) {
10386                                 if (!IN6_IS_ADDR_V4MAPPED(&sl->sl_addr[i]))
10387                                         continue;
10388                                 IN6_V4MAPPED_TO_IPADDR(&sl->sl_addr[i],
10389                                     ips.ipGroupSourceAddress);
10390                                 if (snmp_append_data2(mpctl->b_cont, &mp_tail,
10391                                     (char *)&ips, (int)sizeof (ips)) == 0) {
10392                                         ip1dbg(("ip_snmp_get_mib2_ip_group_src:"
10393                                             " failed to allocate %u bytes\n",
10394                                             (uint_t)sizeof (ips)));
10395                                 }
10396                         }
10397                 }
10398                 rw_exit(&ill->ill_mcast_lock);
10399                 ill_refrele(ill);
10400                 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10401         }
10402         rw_exit(&ipst->ips_ill_g_lock);
10403         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10404         ip3dbg(("ip_snmp_get: level %d, name %d, len %d\n",
10405             (int)optp->level, (int)optp->name, (int)optp->len));
10406         qreply(q, mpctl);
10407         return (mp2ctl);
10408 }
10409 
10410 /* IPv6 multicast filtered sources. */
10411 static mblk_t *
10412 ip_snmp_get_mib2_ip6_group_src(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10413 {
10414         struct opthdr           *optp;
10415         mblk_t                  *mp2ctl;
10416         ill_t                   *ill;
10417         ilm_t                   *ilm;
10418         ipv6_grpsrc_t           ips6;
10419         mblk_t                  *mp_tail = NULL;
10420         ill_walk_context_t      ctx;
10421         zoneid_t                zoneid;
10422         int                     i;
10423         slist_t                 *sl;
10424 
10425         /*
10426          * make a copy of the original message
10427          */
10428         mp2ctl = copymsg(mpctl);
10429         zoneid = Q_TO_CONN(q)->conn_zoneid;
10430 
10431         /* ip6GroupMember table */
10432         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10433         optp->level = MIB2_IP6;
10434         optp->name = EXPER_IP6_GROUP_SOURCES;
10435 
10436         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10437         ill = ILL_START_WALK_V6(&ctx, ipst);
10438         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10439                 /* Make sure the ill isn't going away. */
10440                 if (!ill_check_and_refhold(ill))
10441                         continue;
10442                 rw_exit(&ipst->ips_ill_g_lock);
10443                 /*
10444                  * Normally we don't have any members on under IPMP interfaces.
10445                  * We report them as a debugging aid.
10446                  */
10447                 rw_enter(&ill->ill_mcast_lock, RW_READER);
10448                 ips6.ipv6GroupSourceIfIndex = ill->ill_phyint->phyint_ifindex;
10449                 for (ilm = ill->ill_ilm; ilm; ilm = ilm->ilm_next) {
10450                         sl = ilm->ilm_filter;
10451                         if (ilm->ilm_zoneid != zoneid &&
10452                             ilm->ilm_zoneid != ALL_ZONES)
10453                                 continue;
10454                         if (SLIST_IS_EMPTY(sl))
10455                                 continue;
10456                         ips6.ipv6GroupSourceGroup = ilm->ilm_v6addr;
10457                         for (i = 0; i < sl->sl_numsrc; i++) {
10458                                 ips6.ipv6GroupSourceAddress = sl->sl_addr[i];
10459                                 if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10460                                     (char *)&ips6, (int)sizeof (ips6))) {
10461                                         ip1dbg(("ip_snmp_get_mib2_ip6_"
10462                                             "group_src: failed to allocate "
10463                                             "%u bytes\n",
10464                                             (uint_t)sizeof (ips6)));
10465                                 }
10466                         }
10467                 }
10468                 rw_exit(&ill->ill_mcast_lock);
10469                 ill_refrele(ill);
10470                 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10471         }
10472         rw_exit(&ipst->ips_ill_g_lock);
10473 
10474         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10475         ip3dbg(("ip_snmp_get: level %d, name %d, len %d\n",
10476             (int)optp->level, (int)optp->name, (int)optp->len));
10477         qreply(q, mpctl);
10478         return (mp2ctl);
10479 }
10480 
10481 /* Multicast routing virtual interface table. */
10482 static mblk_t *
10483 ip_snmp_get_mib2_virt_multi(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10484 {
10485         struct opthdr           *optp;
10486         mblk_t                  *mp2ctl;
10487 
10488         /*
10489          * make a copy of the original message
10490          */
10491         mp2ctl = copymsg(mpctl);
10492 
10493         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10494         optp->level = EXPER_DVMRP;
10495         optp->name = EXPER_DVMRP_VIF;
10496         if (!ip_mroute_vif(mpctl->b_cont, ipst)) {
10497                 ip0dbg(("ip_mroute_vif: failed\n"));
10498         }
10499         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10500         ip3dbg(("ip_snmp_get_mib2_virt_multi: level %d, name %d, len %d\n",
10501             (int)optp->level, (int)optp->name, (int)optp->len));
10502         qreply(q, mpctl);
10503         return (mp2ctl);
10504 }
10505 
10506 /* Multicast routing table. */
10507 static mblk_t *
10508 ip_snmp_get_mib2_multi_rtable(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10509 {
10510         struct opthdr           *optp;
10511         mblk_t                  *mp2ctl;
10512 
10513         /*
10514          * make a copy of the original message
10515          */
10516         mp2ctl = copymsg(mpctl);
10517 
10518         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10519         optp->level = EXPER_DVMRP;
10520         optp->name = EXPER_DVMRP_MRT;
10521         if (!ip_mroute_mrt(mpctl->b_cont, ipst)) {
10522                 ip0dbg(("ip_mroute_mrt: failed\n"));
10523         }
10524         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10525         ip3dbg(("ip_snmp_get_mib2_multi_rtable: level %d, name %d, len %d\n",
10526             (int)optp->level, (int)optp->name, (int)optp->len));
10527         qreply(q, mpctl);
10528         return (mp2ctl);
10529 }
10530 
10531 /*
10532  * Return ipRouteEntryTable, ipNetToMediaEntryTable, and ipRouteAttributeTable
10533  * in one IRE walk.
10534  */
10535 static mblk_t *
10536 ip_snmp_get_mib2_ip_route_media(queue_t *q, mblk_t *mpctl, int level,
10537     ip_stack_t *ipst)
10538 {
10539         struct opthdr   *optp;
10540         mblk_t          *mp2ctl;        /* Returned */
10541         mblk_t          *mp3ctl;        /* nettomedia */
10542         mblk_t          *mp4ctl;        /* routeattrs */
10543         iproutedata_t   ird;
10544         zoneid_t        zoneid;
10545 
10546         /*
10547          * make copies of the original message
10548          *      - mp2ctl is returned unchanged to the caller for his use
10549          *      - mpctl is sent upstream as ipRouteEntryTable
10550          *      - mp3ctl is sent upstream as ipNetToMediaEntryTable
10551          *      - mp4ctl is sent upstream as ipRouteAttributeTable
10552          */
10553         mp2ctl = copymsg(mpctl);
10554         mp3ctl = copymsg(mpctl);
10555         mp4ctl = copymsg(mpctl);
10556         if (mp3ctl == NULL || mp4ctl == NULL) {
10557                 freemsg(mp4ctl);
10558                 freemsg(mp3ctl);
10559                 freemsg(mp2ctl);
10560                 freemsg(mpctl);
10561                 return (NULL);
10562         }
10563 
10564         bzero(&ird, sizeof (ird));
10565 
10566         ird.ird_route.lp_head = mpctl->b_cont;
10567         ird.ird_netmedia.lp_head = mp3ctl->b_cont;
10568         ird.ird_attrs.lp_head = mp4ctl->b_cont;
10569         /*
10570          * If the level has been set the special EXPER_IP_AND_ALL_IRES value,
10571          * then also include ire_testhidden IREs and IRE_IF_CLONE.  This is
10572          * intended a temporary solution until a proper MIB API is provided
10573          * that provides complete filtering/caller-opt-in.
10574          */
10575         if (level == EXPER_IP_AND_ALL_IRES)
10576                 ird.ird_flags |= IRD_REPORT_ALL;
10577 
10578         zoneid = Q_TO_CONN(q)->conn_zoneid;
10579         ire_walk_v4(ip_snmp_get2_v4, &ird, zoneid, ipst);
10580 
10581         /* ipRouteEntryTable in mpctl */
10582         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10583         optp->level = MIB2_IP;
10584         optp->name = MIB2_IP_ROUTE;
10585         optp->len = msgdsize(ird.ird_route.lp_head);
10586         ip3dbg(("ip_snmp_get_mib2_ip_route_media: level %d, name %d, len %d\n",
10587             (int)optp->level, (int)optp->name, (int)optp->len));
10588         qreply(q, mpctl);
10589 
10590         /* ipNetToMediaEntryTable in mp3ctl */
10591         ncec_walk(NULL, ip_snmp_get2_v4_media, &ird, ipst);
10592 
10593         optp = (struct opthdr *)&mp3ctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10594         optp->level = MIB2_IP;
10595         optp->name = MIB2_IP_MEDIA;
10596         optp->len = msgdsize(ird.ird_netmedia.lp_head);
10597         ip3dbg(("ip_snmp_get_mib2_ip_route_media: level %d, name %d, len %d\n",
10598             (int)optp->level, (int)optp->name, (int)optp->len));
10599         qreply(q, mp3ctl);
10600 
10601         /* ipRouteAttributeTable in mp4ctl */
10602         optp = (struct opthdr *)&mp4ctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10603         optp->level = MIB2_IP;
10604         optp->name = EXPER_IP_RTATTR;
10605         optp->len = msgdsize(ird.ird_attrs.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         if (optp->len == 0)
10609                 freemsg(mp4ctl);
10610         else
10611                 qreply(q, mp4ctl);
10612 
10613         return (mp2ctl);
10614 }
10615 
10616 /*
10617  * Return ipv6RouteEntryTable and ipv6RouteAttributeTable in one IRE walk, and
10618  * ipv6NetToMediaEntryTable in an NDP walk.
10619  */
10620 static mblk_t *
10621 ip_snmp_get_mib2_ip6_route_media(queue_t *q, mblk_t *mpctl, int level,
10622     ip_stack_t *ipst)
10623 {
10624         struct opthdr   *optp;
10625         mblk_t          *mp2ctl;        /* Returned */
10626         mblk_t          *mp3ctl;        /* nettomedia */
10627         mblk_t          *mp4ctl;        /* routeattrs */
10628         iproutedata_t   ird;
10629         zoneid_t        zoneid;
10630 
10631         /*
10632          * make copies of the original message
10633          *      - mp2ctl is returned unchanged to the caller for his use
10634          *      - mpctl is sent upstream as ipv6RouteEntryTable
10635          *      - mp3ctl is sent upstream as ipv6NetToMediaEntryTable
10636          *      - mp4ctl is sent upstream as ipv6RouteAttributeTable
10637          */
10638         mp2ctl = copymsg(mpctl);
10639         mp3ctl = copymsg(mpctl);
10640         mp4ctl = copymsg(mpctl);
10641         if (mp3ctl == NULL || mp4ctl == NULL) {
10642                 freemsg(mp4ctl);
10643                 freemsg(mp3ctl);
10644                 freemsg(mp2ctl);
10645                 freemsg(mpctl);
10646                 return (NULL);
10647         }
10648 
10649         bzero(&ird, sizeof (ird));
10650 
10651         ird.ird_route.lp_head = mpctl->b_cont;
10652         ird.ird_netmedia.lp_head = mp3ctl->b_cont;
10653         ird.ird_attrs.lp_head = mp4ctl->b_cont;
10654         /*
10655          * If the level has been set the special EXPER_IP_AND_ALL_IRES value,
10656          * then also include ire_testhidden IREs and IRE_IF_CLONE.  This is
10657          * intended a temporary solution until a proper MIB API is provided
10658          * that provides complete filtering/caller-opt-in.
10659          */
10660         if (level == EXPER_IP_AND_ALL_IRES)
10661                 ird.ird_flags |= IRD_REPORT_ALL;
10662 
10663         zoneid = Q_TO_CONN(q)->conn_zoneid;
10664         ire_walk_v6(ip_snmp_get2_v6_route, &ird, zoneid, ipst);
10665 
10666         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10667         optp->level = MIB2_IP6;
10668         optp->name = MIB2_IP6_ROUTE;
10669         optp->len = msgdsize(ird.ird_route.lp_head);
10670         ip3dbg(("ip_snmp_get_mib2_ip6_route_media: level %d, name %d, len %d\n",
10671             (int)optp->level, (int)optp->name, (int)optp->len));
10672         qreply(q, mpctl);
10673 
10674         /* ipv6NetToMediaEntryTable in mp3ctl */
10675         ncec_walk(NULL, ip_snmp_get2_v6_media, &ird, ipst);
10676 
10677         optp = (struct opthdr *)&mp3ctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10678         optp->level = MIB2_IP6;
10679         optp->name = MIB2_IP6_MEDIA;
10680         optp->len = msgdsize(ird.ird_netmedia.lp_head);
10681         ip3dbg(("ip_snmp_get_mib2_ip6_route_media: level %d, name %d, len %d\n",
10682             (int)optp->level, (int)optp->name, (int)optp->len));
10683         qreply(q, mp3ctl);
10684 
10685         /* ipv6RouteAttributeTable in mp4ctl */
10686         optp = (struct opthdr *)&mp4ctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10687         optp->level = MIB2_IP6;
10688         optp->name = EXPER_IP_RTATTR;
10689         optp->len = msgdsize(ird.ird_attrs.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         if (optp->len == 0)
10693                 freemsg(mp4ctl);
10694         else
10695                 qreply(q, mp4ctl);
10696 
10697         return (mp2ctl);
10698 }
10699 
10700 /*
10701  * IPv6 mib: One per ill
10702  */
10703 static mblk_t *
10704 ip_snmp_get_mib2_ip6(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst,
10705     boolean_t legacy_req)
10706 {
10707         struct opthdr           *optp;
10708         mblk_t                  *mp2ctl;
10709         ill_t                   *ill;
10710         ill_walk_context_t      ctx;
10711         mblk_t                  *mp_tail = NULL;
10712         mib2_ipv6AddrEntry_t    mae6;
10713         mib2_ipIfStatsEntry_t   *ise;
10714         size_t                  ise_size, iae_size;
10715 
10716         /*
10717          * Make a copy of the original message
10718          */
10719         mp2ctl = copymsg(mpctl);
10720 
10721         /* fixed length IPv6 structure ... */
10722 
10723         if (legacy_req) {
10724                 ise_size = LEGACY_MIB_SIZE(&ipst->ips_ip6_mib,
10725                     mib2_ipIfStatsEntry_t);
10726                 iae_size = LEGACY_MIB_SIZE(&mae6, mib2_ipv6AddrEntry_t);
10727         } else {
10728                 ise_size = sizeof (mib2_ipIfStatsEntry_t);
10729                 iae_size = sizeof (mib2_ipv6AddrEntry_t);
10730         }
10731 
10732         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10733         optp->level = MIB2_IP6;
10734         optp->name = 0;
10735         /* Include "unknown interface" ip6_mib */
10736         ipst->ips_ip6_mib.ipIfStatsIPVersion = MIB2_INETADDRESSTYPE_ipv6;
10737         ipst->ips_ip6_mib.ipIfStatsIfIndex =
10738             MIB2_UNKNOWN_INTERFACE; /* Flag to netstat */
10739         SET_MIB(ipst->ips_ip6_mib.ipIfStatsForwarding,
10740             ipst->ips_ipv6_forwarding ? 1 : 2);
10741         SET_MIB(ipst->ips_ip6_mib.ipIfStatsDefaultHopLimit,
10742             ipst->ips_ipv6_def_hops);
10743         SET_MIB(ipst->ips_ip6_mib.ipIfStatsEntrySize,
10744             sizeof (mib2_ipIfStatsEntry_t));
10745         SET_MIB(ipst->ips_ip6_mib.ipIfStatsAddrEntrySize,
10746             sizeof (mib2_ipv6AddrEntry_t));
10747         SET_MIB(ipst->ips_ip6_mib.ipIfStatsRouteEntrySize,
10748             sizeof (mib2_ipv6RouteEntry_t));
10749         SET_MIB(ipst->ips_ip6_mib.ipIfStatsNetToMediaEntrySize,
10750             sizeof (mib2_ipv6NetToMediaEntry_t));
10751         SET_MIB(ipst->ips_ip6_mib.ipIfStatsMemberEntrySize,
10752             sizeof (ipv6_member_t));
10753         SET_MIB(ipst->ips_ip6_mib.ipIfStatsGroupSourceEntrySize,
10754             sizeof (ipv6_grpsrc_t));
10755 
10756         /*
10757          * Synchronize 64- and 32-bit counters
10758          */
10759         SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsInReceives,
10760             ipIfStatsHCInReceives);
10761         SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsInDelivers,
10762             ipIfStatsHCInDelivers);
10763         SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsOutRequests,
10764             ipIfStatsHCOutRequests);
10765         SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsOutForwDatagrams,
10766             ipIfStatsHCOutForwDatagrams);
10767         SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsOutMcastPkts,
10768             ipIfStatsHCOutMcastPkts);
10769         SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsInMcastPkts,
10770             ipIfStatsHCInMcastPkts);
10771 
10772         if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10773             (char *)&ipst->ips_ip6_mib, (int)ise_size)) {
10774                 ip1dbg(("ip_snmp_get_mib2_ip6: failed to allocate %u bytes\n",
10775                     (uint_t)ise_size));
10776         } else if (legacy_req) {
10777                 /* Adjust the EntrySize fields for legacy requests. */
10778                 ise =
10779                     (mib2_ipIfStatsEntry_t *)(mp_tail->b_wptr - (int)ise_size);
10780                 SET_MIB(ise->ipIfStatsEntrySize, ise_size);
10781                 SET_MIB(ise->ipIfStatsAddrEntrySize, iae_size);
10782         }
10783 
10784         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10785         ill = ILL_START_WALK_V6(&ctx, ipst);
10786         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10787                 ill->ill_ip_mib->ipIfStatsIfIndex =
10788                     ill->ill_phyint->phyint_ifindex;
10789                 SET_MIB(ill->ill_ip_mib->ipIfStatsForwarding,
10790                     ipst->ips_ipv6_forwarding ? 1 : 2);
10791                 SET_MIB(ill->ill_ip_mib->ipIfStatsDefaultHopLimit,
10792                     ill->ill_max_hops);
10793 
10794                 /*
10795                  * Synchronize 64- and 32-bit counters
10796                  */
10797                 SYNC32_MIB(ill->ill_ip_mib, ipIfStatsInReceives,
10798                     ipIfStatsHCInReceives);
10799                 SYNC32_MIB(ill->ill_ip_mib, ipIfStatsInDelivers,
10800                     ipIfStatsHCInDelivers);
10801                 SYNC32_MIB(ill->ill_ip_mib, ipIfStatsOutRequests,
10802                     ipIfStatsHCOutRequests);
10803                 SYNC32_MIB(ill->ill_ip_mib, ipIfStatsOutForwDatagrams,
10804                     ipIfStatsHCOutForwDatagrams);
10805                 SYNC32_MIB(ill->ill_ip_mib, ipIfStatsOutMcastPkts,
10806                     ipIfStatsHCOutMcastPkts);
10807                 SYNC32_MIB(ill->ill_ip_mib, ipIfStatsInMcastPkts,
10808                     ipIfStatsHCInMcastPkts);
10809 
10810                 if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10811                     (char *)ill->ill_ip_mib, (int)ise_size)) {
10812                         ip1dbg(("ip_snmp_get_mib2_ip6: failed to allocate "
10813                         "%u bytes\n", (uint_t)ise_size));
10814                 } else if (legacy_req) {
10815                         /* Adjust the EntrySize fields for legacy requests. */
10816                         ise = (mib2_ipIfStatsEntry_t *)(mp_tail->b_wptr -
10817                             (int)ise_size);
10818                         SET_MIB(ise->ipIfStatsEntrySize, ise_size);
10819                         SET_MIB(ise->ipIfStatsAddrEntrySize, iae_size);
10820                 }
10821         }
10822         rw_exit(&ipst->ips_ill_g_lock);
10823 
10824         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10825         ip3dbg(("ip_snmp_get_mib2_ip6: level %d, name %d, len %d\n",
10826             (int)optp->level, (int)optp->name, (int)optp->len));
10827         qreply(q, mpctl);
10828         return (mp2ctl);
10829 }
10830 
10831 /*
10832  * ICMPv6 mib: One per ill
10833  */
10834 static mblk_t *
10835 ip_snmp_get_mib2_icmp6(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10836 {
10837         struct opthdr           *optp;
10838         mblk_t                  *mp2ctl;
10839         ill_t                   *ill;
10840         ill_walk_context_t      ctx;
10841         mblk_t                  *mp_tail = NULL;
10842         /*
10843          * Make a copy of the original message
10844          */
10845         mp2ctl = copymsg(mpctl);
10846 
10847         /* fixed length ICMPv6 structure ... */
10848 
10849         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10850         optp->level = MIB2_ICMP6;
10851         optp->name = 0;
10852         /* Include "unknown interface" icmp6_mib */
10853         ipst->ips_icmp6_mib.ipv6IfIcmpIfIndex =
10854             MIB2_UNKNOWN_INTERFACE; /* netstat flag */
10855         ipst->ips_icmp6_mib.ipv6IfIcmpEntrySize =
10856             sizeof (mib2_ipv6IfIcmpEntry_t);
10857         if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10858             (char *)&ipst->ips_icmp6_mib,
10859             (int)sizeof (ipst->ips_icmp6_mib))) {
10860                 ip1dbg(("ip_snmp_get_mib2_icmp6: failed to allocate %u bytes\n",
10861                     (uint_t)sizeof (ipst->ips_icmp6_mib)));
10862         }
10863 
10864         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10865         ill = ILL_START_WALK_V6(&ctx, ipst);
10866         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10867                 ill->ill_icmp6_mib->ipv6IfIcmpIfIndex =
10868                     ill->ill_phyint->phyint_ifindex;
10869                 if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10870                     (char *)ill->ill_icmp6_mib,
10871                     (int)sizeof (*ill->ill_icmp6_mib))) {
10872                         ip1dbg(("ip_snmp_get_mib2_icmp6: failed to allocate "
10873                             "%u bytes\n",
10874                             (uint_t)sizeof (*ill->ill_icmp6_mib)));
10875                 }
10876         }
10877         rw_exit(&ipst->ips_ill_g_lock);
10878 
10879         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10880         ip3dbg(("ip_snmp_get_mib2_icmp6: level %d, name %d, len %d\n",
10881             (int)optp->level, (int)optp->name, (int)optp->len));
10882         qreply(q, mpctl);
10883         return (mp2ctl);
10884 }
10885 
10886 /*
10887  * ire_walk routine to create both ipRouteEntryTable and
10888  * ipRouteAttributeTable in one IRE walk
10889  */
10890 static void
10891 ip_snmp_get2_v4(ire_t *ire, iproutedata_t *ird)
10892 {
10893         ill_t                           *ill;
10894         mib2_ipRouteEntry_t             *re;
10895         mib2_ipAttributeEntry_t         iaes;
10896         tsol_ire_gw_secattr_t           *attrp;
10897         tsol_gc_t                       *gc = NULL;
10898         tsol_gcgrp_t                    *gcgrp = NULL;
10899         ip_stack_t                      *ipst = ire->ire_ipst;
10900 
10901         ASSERT(ire->ire_ipversion == IPV4_VERSION);
10902 
10903         if (!(ird->ird_flags & IRD_REPORT_ALL)) {
10904                 if (ire->ire_testhidden)
10905                         return;
10906                 if (ire->ire_type & IRE_IF_CLONE)
10907                         return;
10908         }
10909 
10910         if ((re = kmem_zalloc(sizeof (*re), KM_NOSLEEP)) == NULL)
10911                 return;
10912 
10913         if ((attrp = ire->ire_gw_secattr) != NULL) {
10914                 mutex_enter(&attrp->igsa_lock);
10915                 if ((gc = attrp->igsa_gc) != NULL) {
10916                         gcgrp = gc->gc_grp;
10917                         ASSERT(gcgrp != NULL);
10918                         rw_enter(&gcgrp->gcgrp_rwlock, RW_READER);
10919                 }
10920                 mutex_exit(&attrp->igsa_lock);
10921         }
10922         /*
10923          * Return all IRE types for route table... let caller pick and choose
10924          */
10925         re->ipRouteDest = ire->ire_addr;
10926         ill = ire->ire_ill;
10927         re->ipRouteIfIndex.o_length = 0;
10928         if (ill != NULL) {
10929                 ill_get_name(ill, re->ipRouteIfIndex.o_bytes, OCTET_LENGTH);
10930                 re->ipRouteIfIndex.o_length =
10931                     mi_strlen(re->ipRouteIfIndex.o_bytes);
10932         }
10933         re->ipRouteMetric1 = -1;
10934         re->ipRouteMetric2 = -1;
10935         re->ipRouteMetric3 = -1;
10936         re->ipRouteMetric4 = -1;
10937 
10938         re->ipRouteNextHop = ire->ire_gateway_addr;
10939         /* indirect(4), direct(3), or invalid(2) */
10940         if (ire->ire_flags & (RTF_REJECT | RTF_BLACKHOLE))
10941                 re->ipRouteType = 2;
10942         else if (ire->ire_type & IRE_ONLINK)
10943                 re->ipRouteType = 3;
10944         else
10945                 re->ipRouteType = 4;
10946 
10947         re->ipRouteProto = -1;
10948         re->ipRouteAge = gethrestime_sec() - ire->ire_create_time;
10949         re->ipRouteMask = ire->ire_mask;
10950         re->ipRouteMetric5 = -1;
10951         re->ipRouteInfo.re_max_frag = ire->ire_metrics.iulp_mtu;
10952         if (ire->ire_ill != NULL && re->ipRouteInfo.re_max_frag == 0)
10953                 re->ipRouteInfo.re_max_frag = ire->ire_ill->ill_mtu;
10954 
10955         re->ipRouteInfo.re_frag_flag = 0;
10956         re->ipRouteInfo.re_rtt               = 0;
10957         re->ipRouteInfo.re_src_addr  = 0;
10958         re->ipRouteInfo.re_ref               = ire->ire_refcnt;
10959         re->ipRouteInfo.re_obpkt     = ire->ire_ob_pkt_count;
10960         re->ipRouteInfo.re_ibpkt     = ire->ire_ib_pkt_count;
10961         re->ipRouteInfo.re_flags     = ire->ire_flags;
10962 
10963         /* Add the IRE_IF_CLONE's counters to their parent IRE_INTERFACE */
10964         if (ire->ire_type & IRE_INTERFACE) {
10965                 ire_t *child;
10966 
10967                 rw_enter(&ipst->ips_ire_dep_lock, RW_READER);
10968                 child = ire->ire_dep_children;
10969                 while (child != NULL) {
10970                         re->ipRouteInfo.re_obpkt += child->ire_ob_pkt_count;
10971                         re->ipRouteInfo.re_ibpkt += child->ire_ib_pkt_count;
10972                         child = child->ire_dep_sib_next;
10973                 }
10974                 rw_exit(&ipst->ips_ire_dep_lock);
10975         }
10976 
10977         if (ire->ire_flags & RTF_DYNAMIC) {
10978                 re->ipRouteInfo.re_ire_type  = IRE_HOST_REDIRECT;
10979         } else {
10980                 re->ipRouteInfo.re_ire_type  = ire->ire_type;
10981         }
10982 
10983         if (!snmp_append_data2(ird->ird_route.lp_head, &ird->ird_route.lp_tail,
10984             (char *)re, (int)sizeof (*re))) {
10985                 ip1dbg(("ip_snmp_get2_v4: failed to allocate %u bytes\n",
10986                     (uint_t)sizeof (*re)));
10987         }
10988 
10989         if (gc != NULL) {
10990                 iaes.iae_routeidx = ird->ird_idx;
10991                 iaes.iae_doi = gc->gc_db->gcdb_doi;
10992                 iaes.iae_slrange = gc->gc_db->gcdb_slrange;
10993 
10994                 if (!snmp_append_data2(ird->ird_attrs.lp_head,
10995                     &ird->ird_attrs.lp_tail, (char *)&iaes, sizeof (iaes))) {
10996                         ip1dbg(("ip_snmp_get2_v4: failed to allocate %u "
10997                             "bytes\n", (uint_t)sizeof (iaes)));
10998                 }
10999         }
11000 
11001         /* bump route index for next pass */
11002         ird->ird_idx++;
11003 
11004         kmem_free(re, sizeof (*re));
11005         if (gcgrp != NULL)
11006                 rw_exit(&gcgrp->gcgrp_rwlock);
11007 }
11008 
11009 /*
11010  * ire_walk routine to create ipv6RouteEntryTable and ipRouteEntryTable.
11011  */
11012 static void
11013 ip_snmp_get2_v6_route(ire_t *ire, iproutedata_t *ird)
11014 {
11015         ill_t                           *ill;
11016         mib2_ipv6RouteEntry_t           *re;
11017         mib2_ipAttributeEntry_t         iaes;
11018         tsol_ire_gw_secattr_t           *attrp;
11019         tsol_gc_t                       *gc = NULL;
11020         tsol_gcgrp_t                    *gcgrp = NULL;
11021         ip_stack_t                      *ipst = ire->ire_ipst;
11022 
11023         ASSERT(ire->ire_ipversion == IPV6_VERSION);
11024 
11025         if (!(ird->ird_flags & IRD_REPORT_ALL)) {
11026                 if (ire->ire_testhidden)
11027                         return;
11028                 if (ire->ire_type & IRE_IF_CLONE)
11029                         return;
11030         }
11031 
11032         if ((re = kmem_zalloc(sizeof (*re), KM_NOSLEEP)) == NULL)
11033                 return;
11034 
11035         if ((attrp = ire->ire_gw_secattr) != NULL) {
11036                 mutex_enter(&attrp->igsa_lock);
11037                 if ((gc = attrp->igsa_gc) != NULL) {
11038                         gcgrp = gc->gc_grp;
11039                         ASSERT(gcgrp != NULL);
11040                         rw_enter(&gcgrp->gcgrp_rwlock, RW_READER);
11041                 }
11042                 mutex_exit(&attrp->igsa_lock);
11043         }
11044         /*
11045          * Return all IRE types for route table... let caller pick and choose
11046          */
11047         re->ipv6RouteDest = ire->ire_addr_v6;
11048         re->ipv6RoutePfxLength = ip_mask_to_plen_v6(&ire->ire_mask_v6);
11049         re->ipv6RouteIndex = 0;      /* Unique when multiple with same dest/plen */
11050         re->ipv6RouteIfIndex.o_length = 0;
11051         ill = ire->ire_ill;
11052         if (ill != NULL) {
11053                 ill_get_name(ill, re->ipv6RouteIfIndex.o_bytes, OCTET_LENGTH);
11054                 re->ipv6RouteIfIndex.o_length =
11055                     mi_strlen(re->ipv6RouteIfIndex.o_bytes);
11056         }
11057 
11058         ASSERT(!(ire->ire_type & IRE_BROADCAST));
11059 
11060         mutex_enter(&ire->ire_lock);
11061         re->ipv6RouteNextHop = ire->ire_gateway_addr_v6;
11062         mutex_exit(&ire->ire_lock);
11063 
11064         /* remote(4), local(3), or discard(2) */
11065         if (ire->ire_flags & (RTF_REJECT | RTF_BLACKHOLE))
11066                 re->ipv6RouteType = 2;
11067         else if (ire->ire_type & IRE_ONLINK)
11068                 re->ipv6RouteType = 3;
11069         else
11070                 re->ipv6RouteType = 4;
11071 
11072         re->ipv6RouteProtocol        = -1;
11073         re->ipv6RoutePolicy  = 0;
11074         re->ipv6RouteAge     = gethrestime_sec() - ire->ire_create_time;
11075         re->ipv6RouteNextHopRDI      = 0;
11076         re->ipv6RouteWeight  = 0;
11077         re->ipv6RouteMetric  = 0;
11078         re->ipv6RouteInfo.re_max_frag = ire->ire_metrics.iulp_mtu;
11079         if (ire->ire_ill != NULL && re->ipv6RouteInfo.re_max_frag == 0)
11080                 re->ipv6RouteInfo.re_max_frag = ire->ire_ill->ill_mtu;
11081 
11082         re->ipv6RouteInfo.re_frag_flag       = 0;
11083         re->ipv6RouteInfo.re_rtt     = 0;
11084         re->ipv6RouteInfo.re_src_addr        = ipv6_all_zeros;
11085         re->ipv6RouteInfo.re_obpkt   = ire->ire_ob_pkt_count;
11086         re->ipv6RouteInfo.re_ibpkt   = ire->ire_ib_pkt_count;
11087         re->ipv6RouteInfo.re_ref     = ire->ire_refcnt;
11088         re->ipv6RouteInfo.re_flags   = ire->ire_flags;
11089 
11090         /* Add the IRE_IF_CLONE's counters to their parent IRE_INTERFACE */
11091         if (ire->ire_type & IRE_INTERFACE) {
11092                 ire_t *child;
11093 
11094                 rw_enter(&ipst->ips_ire_dep_lock, RW_READER);
11095                 child = ire->ire_dep_children;
11096                 while (child != NULL) {
11097                         re->ipv6RouteInfo.re_obpkt += child->ire_ob_pkt_count;
11098                         re->ipv6RouteInfo.re_ibpkt += child->ire_ib_pkt_count;
11099                         child = child->ire_dep_sib_next;
11100                 }
11101                 rw_exit(&ipst->ips_ire_dep_lock);
11102         }
11103         if (ire->ire_flags & RTF_DYNAMIC) {
11104                 re->ipv6RouteInfo.re_ire_type        = IRE_HOST_REDIRECT;
11105         } else {
11106                 re->ipv6RouteInfo.re_ire_type        = ire->ire_type;
11107         }
11108 
11109         if (!snmp_append_data2(ird->ird_route.lp_head, &ird->ird_route.lp_tail,
11110             (char *)re, (int)sizeof (*re))) {
11111                 ip1dbg(("ip_snmp_get2_v6: failed to allocate %u bytes\n",
11112                     (uint_t)sizeof (*re)));
11113         }
11114 
11115         if (gc != NULL) {
11116                 iaes.iae_routeidx = ird->ird_idx;
11117                 iaes.iae_doi = gc->gc_db->gcdb_doi;
11118                 iaes.iae_slrange = gc->gc_db->gcdb_slrange;
11119 
11120                 if (!snmp_append_data2(ird->ird_attrs.lp_head,
11121                     &ird->ird_attrs.lp_tail, (char *)&iaes, sizeof (iaes))) {
11122                         ip1dbg(("ip_snmp_get2_v6: failed to allocate %u "
11123                             "bytes\n", (uint_t)sizeof (iaes)));
11124                 }
11125         }
11126 
11127         /* bump route index for next pass */
11128         ird->ird_idx++;
11129 
11130         kmem_free(re, sizeof (*re));
11131         if (gcgrp != NULL)
11132                 rw_exit(&gcgrp->gcgrp_rwlock);
11133 }
11134 
11135 /*
11136  * ncec_walk routine to create ipv6NetToMediaEntryTable
11137  */
11138 static int
11139 ip_snmp_get2_v6_media(ncec_t *ncec, iproutedata_t *ird)
11140 {
11141         ill_t                           *ill;
11142         mib2_ipv6NetToMediaEntry_t      ntme;
11143 
11144         ill = ncec->ncec_ill;
11145         /* skip arpce entries, and loopback ncec entries */
11146         if (ill->ill_isv6 == B_FALSE || ill->ill_net_type == IRE_LOOPBACK)
11147                 return (0);
11148         /*
11149          * Neighbor cache entry attached to IRE with on-link
11150          * destination.
11151          * We report all IPMP groups on ncec_ill which is normally the upper.
11152          */
11153         ntme.ipv6NetToMediaIfIndex = ill->ill_phyint->phyint_ifindex;
11154         ntme.ipv6NetToMediaNetAddress = ncec->ncec_addr;
11155         ntme.ipv6NetToMediaPhysAddress.o_length = ill->ill_phys_addr_length;
11156         if (ncec->ncec_lladdr != NULL) {
11157                 bcopy(ncec->ncec_lladdr, ntme.ipv6NetToMediaPhysAddress.o_bytes,
11158                     ntme.ipv6NetToMediaPhysAddress.o_length);
11159         }
11160         /*
11161          * Note: Returns ND_* states. Should be:
11162          * reachable(1), stale(2), delay(3), probe(4),
11163          * invalid(5), unknown(6)
11164          */
11165         ntme.ipv6NetToMediaState = ncec->ncec_state;
11166         ntme.ipv6NetToMediaLastUpdated = 0;
11167 
11168         /* other(1), dynamic(2), static(3), local(4) */
11169         if (NCE_MYADDR(ncec)) {
11170                 ntme.ipv6NetToMediaType = 4;
11171         } else if (ncec->ncec_flags & NCE_F_PUBLISH) {
11172                 ntme.ipv6NetToMediaType = 1; /* proxy */
11173         } else if (ncec->ncec_flags & NCE_F_STATIC) {
11174                 ntme.ipv6NetToMediaType = 3;
11175         } else if (ncec->ncec_flags & (NCE_F_MCAST|NCE_F_BCAST)) {
11176                 ntme.ipv6NetToMediaType = 1;
11177         } else {
11178                 ntme.ipv6NetToMediaType = 2;
11179         }
11180 
11181         if (!snmp_append_data2(ird->ird_netmedia.lp_head,
11182             &ird->ird_netmedia.lp_tail, (char *)&ntme, sizeof (ntme))) {
11183                 ip1dbg(("ip_snmp_get2_v6_media: failed to allocate %u bytes\n",
11184                     (uint_t)sizeof (ntme)));
11185         }
11186         return (0);
11187 }
11188 
11189 int
11190 nce2ace(ncec_t *ncec)
11191 {
11192         int flags = 0;
11193 
11194         if (NCE_ISREACHABLE(ncec))
11195                 flags |= ACE_F_RESOLVED;
11196         if (ncec->ncec_flags & NCE_F_AUTHORITY)
11197                 flags |= ACE_F_AUTHORITY;
11198         if (ncec->ncec_flags & NCE_F_PUBLISH)
11199                 flags |= ACE_F_PUBLISH;
11200         if ((ncec->ncec_flags & NCE_F_NONUD) != 0)
11201                 flags |= ACE_F_PERMANENT;
11202         if (NCE_MYADDR(ncec))
11203                 flags |= (ACE_F_MYADDR | ACE_F_AUTHORITY);
11204         if (ncec->ncec_flags & NCE_F_UNVERIFIED)
11205                 flags |= ACE_F_UNVERIFIED;
11206         if (ncec->ncec_flags & NCE_F_AUTHORITY)
11207                 flags |= ACE_F_AUTHORITY;
11208         if (ncec->ncec_flags & NCE_F_DELAYED)
11209                 flags |= ACE_F_DELAYED;
11210         return (flags);
11211 }
11212 
11213 /*
11214  * ncec_walk routine to create ipNetToMediaEntryTable
11215  */
11216 static int
11217 ip_snmp_get2_v4_media(ncec_t *ncec, iproutedata_t *ird)
11218 {
11219         ill_t                           *ill;
11220         mib2_ipNetToMediaEntry_t        ntme;
11221         const char                      *name = "unknown";
11222         ipaddr_t                        ncec_addr;
11223 
11224         ill = ncec->ncec_ill;
11225         if (ill->ill_isv6 || (ncec->ncec_flags & NCE_F_BCAST) ||
11226             ill->ill_net_type == IRE_LOOPBACK)
11227                 return (0);
11228 
11229         /* We report all IPMP groups on ncec_ill which is normally the upper. */
11230         name = ill->ill_name;
11231         /* Based on RFC 4293: other(1), inval(2), dyn(3), stat(4) */
11232         if (NCE_MYADDR(ncec)) {
11233                 ntme.ipNetToMediaType = 4;
11234         } else if (ncec->ncec_flags & (NCE_F_MCAST|NCE_F_BCAST|NCE_F_PUBLISH)) {
11235                 ntme.ipNetToMediaType = 1;
11236         } else {
11237                 ntme.ipNetToMediaType = 3;
11238         }
11239         ntme.ipNetToMediaIfIndex.o_length = MIN(OCTET_LENGTH, strlen(name));
11240         bcopy(name, ntme.ipNetToMediaIfIndex.o_bytes,
11241             ntme.ipNetToMediaIfIndex.o_length);
11242 
11243         IN6_V4MAPPED_TO_IPADDR(&ncec->ncec_addr, ncec_addr);
11244         bcopy(&ncec_addr, &ntme.ipNetToMediaNetAddress, sizeof (ncec_addr));
11245 
11246         ntme.ipNetToMediaInfo.ntm_mask.o_length = sizeof (ipaddr_t);
11247         ncec_addr = INADDR_BROADCAST;
11248         bcopy(&ncec_addr, ntme.ipNetToMediaInfo.ntm_mask.o_bytes,
11249             sizeof (ncec_addr));
11250         /*
11251          * map all the flags to the ACE counterpart.
11252          */
11253         ntme.ipNetToMediaInfo.ntm_flags = nce2ace(ncec);
11254 
11255         ntme.ipNetToMediaPhysAddress.o_length =
11256             MIN(OCTET_LENGTH, ill->ill_phys_addr_length);
11257 
11258         if (!NCE_ISREACHABLE(ncec))
11259                 ntme.ipNetToMediaPhysAddress.o_length = 0;
11260         else {
11261                 if (ncec->ncec_lladdr != NULL) {
11262                         bcopy(ncec->ncec_lladdr,
11263                             ntme.ipNetToMediaPhysAddress.o_bytes,
11264                             ntme.ipNetToMediaPhysAddress.o_length);
11265                 }
11266         }
11267 
11268         if (!snmp_append_data2(ird->ird_netmedia.lp_head,
11269             &ird->ird_netmedia.lp_tail, (char *)&ntme, sizeof (ntme))) {
11270                 ip1dbg(("ip_snmp_get2_v4_media: failed to allocate %u bytes\n",
11271                     (uint_t)sizeof (ntme)));
11272         }
11273         return (0);
11274 }
11275 
11276 /*
11277  * return (0) if invalid set request, 1 otherwise, including non-tcp requests
11278  */
11279 /* ARGSUSED */
11280 int
11281 ip_snmp_set(queue_t *q, int level, int name, uchar_t *ptr, int len)
11282 {
11283         switch (level) {
11284         case MIB2_IP:
11285         case MIB2_ICMP:
11286                 switch (name) {
11287                 default:
11288                         break;
11289                 }
11290                 return (1);
11291         default:
11292                 return (1);
11293         }
11294 }
11295 
11296 /*
11297  * When there exists both a 64- and 32-bit counter of a particular type
11298  * (i.e., InReceives), only the 64-bit counters are added.
11299  */
11300 void
11301 ip_mib2_add_ip_stats(mib2_ipIfStatsEntry_t *o1, mib2_ipIfStatsEntry_t *o2)
11302 {
11303         UPDATE_MIB(o1, ipIfStatsInHdrErrors, o2->ipIfStatsInHdrErrors);
11304         UPDATE_MIB(o1, ipIfStatsInTooBigErrors, o2->ipIfStatsInTooBigErrors);
11305         UPDATE_MIB(o1, ipIfStatsInNoRoutes, o2->ipIfStatsInNoRoutes);
11306         UPDATE_MIB(o1, ipIfStatsInAddrErrors, o2->ipIfStatsInAddrErrors);
11307         UPDATE_MIB(o1, ipIfStatsInUnknownProtos, o2->ipIfStatsInUnknownProtos);
11308         UPDATE_MIB(o1, ipIfStatsInTruncatedPkts, o2->ipIfStatsInTruncatedPkts);
11309         UPDATE_MIB(o1, ipIfStatsInDiscards, o2->ipIfStatsInDiscards);
11310         UPDATE_MIB(o1, ipIfStatsOutDiscards, o2->ipIfStatsOutDiscards);
11311         UPDATE_MIB(o1, ipIfStatsOutFragOKs, o2->ipIfStatsOutFragOKs);
11312         UPDATE_MIB(o1, ipIfStatsOutFragFails, o2->ipIfStatsOutFragFails);
11313         UPDATE_MIB(o1, ipIfStatsOutFragCreates, o2->ipIfStatsOutFragCreates);
11314         UPDATE_MIB(o1, ipIfStatsReasmReqds, o2->ipIfStatsReasmReqds);
11315         UPDATE_MIB(o1, ipIfStatsReasmOKs, o2->ipIfStatsReasmOKs);
11316         UPDATE_MIB(o1, ipIfStatsReasmFails, o2->ipIfStatsReasmFails);
11317         UPDATE_MIB(o1, ipIfStatsOutNoRoutes, o2->ipIfStatsOutNoRoutes);
11318         UPDATE_MIB(o1, ipIfStatsReasmDuplicates, o2->ipIfStatsReasmDuplicates);
11319         UPDATE_MIB(o1, ipIfStatsReasmPartDups, o2->ipIfStatsReasmPartDups);
11320         UPDATE_MIB(o1, ipIfStatsForwProhibits, o2->ipIfStatsForwProhibits);
11321         UPDATE_MIB(o1, udpInCksumErrs, o2->udpInCksumErrs);
11322         UPDATE_MIB(o1, udpInOverflows, o2->udpInOverflows);
11323         UPDATE_MIB(o1, rawipInOverflows, o2->rawipInOverflows);
11324         UPDATE_MIB(o1, ipIfStatsInWrongIPVersion,
11325             o2->ipIfStatsInWrongIPVersion);
11326         UPDATE_MIB(o1, ipIfStatsOutWrongIPVersion,
11327             o2->ipIfStatsInWrongIPVersion);
11328         UPDATE_MIB(o1, ipIfStatsOutSwitchIPVersion,
11329             o2->ipIfStatsOutSwitchIPVersion);
11330         UPDATE_MIB(o1, ipIfStatsHCInReceives, o2->ipIfStatsHCInReceives);
11331         UPDATE_MIB(o1, ipIfStatsHCInOctets, o2->ipIfStatsHCInOctets);
11332         UPDATE_MIB(o1, ipIfStatsHCInForwDatagrams,
11333             o2->ipIfStatsHCInForwDatagrams);
11334         UPDATE_MIB(o1, ipIfStatsHCInDelivers, o2->ipIfStatsHCInDelivers);
11335         UPDATE_MIB(o1, ipIfStatsHCOutRequests, o2->ipIfStatsHCOutRequests);
11336         UPDATE_MIB(o1, ipIfStatsHCOutForwDatagrams,
11337             o2->ipIfStatsHCOutForwDatagrams);
11338         UPDATE_MIB(o1, ipIfStatsOutFragReqds, o2->ipIfStatsOutFragReqds);
11339         UPDATE_MIB(o1, ipIfStatsHCOutTransmits, o2->ipIfStatsHCOutTransmits);
11340         UPDATE_MIB(o1, ipIfStatsHCOutOctets, o2->ipIfStatsHCOutOctets);
11341         UPDATE_MIB(o1, ipIfStatsHCInMcastPkts, o2->ipIfStatsHCInMcastPkts);
11342         UPDATE_MIB(o1, ipIfStatsHCInMcastOctets, o2->ipIfStatsHCInMcastOctets);
11343         UPDATE_MIB(o1, ipIfStatsHCOutMcastPkts, o2->ipIfStatsHCOutMcastPkts);
11344         UPDATE_MIB(o1, ipIfStatsHCOutMcastOctets,
11345             o2->ipIfStatsHCOutMcastOctets);
11346         UPDATE_MIB(o1, ipIfStatsHCInBcastPkts, o2->ipIfStatsHCInBcastPkts);
11347         UPDATE_MIB(o1, ipIfStatsHCOutBcastPkts, o2->ipIfStatsHCOutBcastPkts);
11348         UPDATE_MIB(o1, ipsecInSucceeded, o2->ipsecInSucceeded);
11349         UPDATE_MIB(o1, ipsecInFailed, o2->ipsecInFailed);
11350         UPDATE_MIB(o1, ipInCksumErrs, o2->ipInCksumErrs);
11351         UPDATE_MIB(o1, tcpInErrs, o2->tcpInErrs);
11352         UPDATE_MIB(o1, udpNoPorts, o2->udpNoPorts);
11353 }
11354 
11355 void
11356 ip_mib2_add_icmp6_stats(mib2_ipv6IfIcmpEntry_t *o1, mib2_ipv6IfIcmpEntry_t *o2)
11357 {
11358         UPDATE_MIB(o1, ipv6IfIcmpInMsgs, o2->ipv6IfIcmpInMsgs);
11359         UPDATE_MIB(o1, ipv6IfIcmpInErrors, o2->ipv6IfIcmpInErrors);
11360         UPDATE_MIB(o1, ipv6IfIcmpInDestUnreachs, o2->ipv6IfIcmpInDestUnreachs);
11361         UPDATE_MIB(o1, ipv6IfIcmpInAdminProhibs, o2->ipv6IfIcmpInAdminProhibs);
11362         UPDATE_MIB(o1, ipv6IfIcmpInTimeExcds, o2->ipv6IfIcmpInTimeExcds);
11363         UPDATE_MIB(o1, ipv6IfIcmpInParmProblems, o2->ipv6IfIcmpInParmProblems);
11364         UPDATE_MIB(o1, ipv6IfIcmpInPktTooBigs, o2->ipv6IfIcmpInPktTooBigs);
11365         UPDATE_MIB(o1, ipv6IfIcmpInEchos, o2->ipv6IfIcmpInEchos);
11366         UPDATE_MIB(o1, ipv6IfIcmpInEchoReplies, o2->ipv6IfIcmpInEchoReplies);
11367         UPDATE_MIB(o1, ipv6IfIcmpInRouterSolicits,
11368             o2->ipv6IfIcmpInRouterSolicits);
11369         UPDATE_MIB(o1, ipv6IfIcmpInRouterAdvertisements,
11370             o2->ipv6IfIcmpInRouterAdvertisements);
11371         UPDATE_MIB(o1, ipv6IfIcmpInNeighborSolicits,
11372             o2->ipv6IfIcmpInNeighborSolicits);
11373         UPDATE_MIB(o1, ipv6IfIcmpInNeighborAdvertisements,
11374             o2->ipv6IfIcmpInNeighborAdvertisements);
11375         UPDATE_MIB(o1, ipv6IfIcmpInRedirects, o2->ipv6IfIcmpInRedirects);
11376         UPDATE_MIB(o1, ipv6IfIcmpInGroupMembQueries,
11377             o2->ipv6IfIcmpInGroupMembQueries);
11378         UPDATE_MIB(o1, ipv6IfIcmpInGroupMembResponses,
11379             o2->ipv6IfIcmpInGroupMembResponses);
11380         UPDATE_MIB(o1, ipv6IfIcmpInGroupMembReductions,
11381             o2->ipv6IfIcmpInGroupMembReductions);
11382         UPDATE_MIB(o1, ipv6IfIcmpOutMsgs, o2->ipv6IfIcmpOutMsgs);
11383         UPDATE_MIB(o1, ipv6IfIcmpOutErrors, o2->ipv6IfIcmpOutErrors);
11384         UPDATE_MIB(o1, ipv6IfIcmpOutDestUnreachs,
11385             o2->ipv6IfIcmpOutDestUnreachs);
11386         UPDATE_MIB(o1, ipv6IfIcmpOutAdminProhibs,
11387             o2->ipv6IfIcmpOutAdminProhibs);
11388         UPDATE_MIB(o1, ipv6IfIcmpOutTimeExcds, o2->ipv6IfIcmpOutTimeExcds);
11389         UPDATE_MIB(o1, ipv6IfIcmpOutParmProblems,
11390             o2->ipv6IfIcmpOutParmProblems);
11391         UPDATE_MIB(o1, ipv6IfIcmpOutPktTooBigs, o2->ipv6IfIcmpOutPktTooBigs);
11392         UPDATE_MIB(o1, ipv6IfIcmpOutEchos, o2->ipv6IfIcmpOutEchos);
11393         UPDATE_MIB(o1, ipv6IfIcmpOutEchoReplies, o2->ipv6IfIcmpOutEchoReplies);
11394         UPDATE_MIB(o1, ipv6IfIcmpOutRouterSolicits,
11395             o2->ipv6IfIcmpOutRouterSolicits);
11396         UPDATE_MIB(o1, ipv6IfIcmpOutRouterAdvertisements,
11397             o2->ipv6IfIcmpOutRouterAdvertisements);
11398         UPDATE_MIB(o1, ipv6IfIcmpOutNeighborSolicits,
11399             o2->ipv6IfIcmpOutNeighborSolicits);
11400         UPDATE_MIB(o1, ipv6IfIcmpOutNeighborAdvertisements,
11401             o2->ipv6IfIcmpOutNeighborAdvertisements);
11402         UPDATE_MIB(o1, ipv6IfIcmpOutRedirects, o2->ipv6IfIcmpOutRedirects);
11403         UPDATE_MIB(o1, ipv6IfIcmpOutGroupMembQueries,
11404             o2->ipv6IfIcmpOutGroupMembQueries);
11405         UPDATE_MIB(o1, ipv6IfIcmpOutGroupMembResponses,
11406             o2->ipv6IfIcmpOutGroupMembResponses);
11407         UPDATE_MIB(o1, ipv6IfIcmpOutGroupMembReductions,
11408             o2->ipv6IfIcmpOutGroupMembReductions);
11409         UPDATE_MIB(o1, ipv6IfIcmpInOverflows, o2->ipv6IfIcmpInOverflows);
11410         UPDATE_MIB(o1, ipv6IfIcmpBadHoplimit, o2->ipv6IfIcmpBadHoplimit);
11411         UPDATE_MIB(o1, ipv6IfIcmpInBadNeighborAdvertisements,
11412             o2->ipv6IfIcmpInBadNeighborAdvertisements);
11413         UPDATE_MIB(o1, ipv6IfIcmpInBadNeighborSolicitations,
11414             o2->ipv6IfIcmpInBadNeighborSolicitations);
11415         UPDATE_MIB(o1, ipv6IfIcmpInBadRedirects, o2->ipv6IfIcmpInBadRedirects);
11416         UPDATE_MIB(o1, ipv6IfIcmpInGroupMembTotal,
11417             o2->ipv6IfIcmpInGroupMembTotal);
11418         UPDATE_MIB(o1, ipv6IfIcmpInGroupMembBadQueries,
11419             o2->ipv6IfIcmpInGroupMembBadQueries);
11420         UPDATE_MIB(o1, ipv6IfIcmpInGroupMembBadReports,
11421             o2->ipv6IfIcmpInGroupMembBadReports);
11422         UPDATE_MIB(o1, ipv6IfIcmpInGroupMembOurReports,
11423             o2->ipv6IfIcmpInGroupMembOurReports);
11424 }
11425 
11426 /*
11427  * Called before the options are updated to check if this packet will
11428  * be source routed from here.
11429  * This routine assumes that the options are well formed i.e. that they
11430  * have already been checked.
11431  */
11432 boolean_t
11433 ip_source_routed(ipha_t *ipha, ip_stack_t *ipst)
11434 {
11435         ipoptp_t        opts;
11436         uchar_t         *opt;
11437         uint8_t         optval;
11438         uint8_t         optlen;
11439         ipaddr_t        dst;
11440 
11441         if (IS_SIMPLE_IPH(ipha)) {
11442                 ip2dbg(("not source routed\n"));
11443                 return (B_FALSE);
11444         }
11445         dst = ipha->ipha_dst;
11446         for (optval = ipoptp_first(&opts, ipha);
11447             optval != IPOPT_EOL;
11448             optval = ipoptp_next(&opts)) {
11449                 ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
11450                 opt = opts.ipoptp_cur;
11451                 optlen = opts.ipoptp_len;
11452                 ip2dbg(("ip_source_routed: opt %d, len %d\n",
11453                     optval, optlen));
11454                 switch (optval) {
11455                         uint32_t off;
11456                 case IPOPT_SSRR:
11457                 case IPOPT_LSRR:
11458                         /*
11459                          * If dst is one of our addresses and there are some
11460                          * entries left in the source route return (true).
11461                          */
11462                         if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
11463                                 ip2dbg(("ip_source_routed: not next"
11464                                     " source route 0x%x\n",
11465                                     ntohl(dst)));
11466                                 return (B_FALSE);
11467                         }
11468                         off = opt[IPOPT_OFFSET];
11469                         off--;
11470                         if (optlen < IP_ADDR_LEN ||
11471                             off > optlen - IP_ADDR_LEN) {
11472                                 /* End of source route */
11473                                 ip1dbg(("ip_source_routed: end of SR\n"));
11474                                 return (B_FALSE);
11475                         }
11476                         return (B_TRUE);
11477                 }
11478         }
11479         ip2dbg(("not source routed\n"));
11480         return (B_FALSE);
11481 }
11482 
11483 /*
11484  * ip_unbind is called by the transports to remove a conn from
11485  * the fanout table.
11486  */
11487 void
11488 ip_unbind(conn_t *connp)
11489 {
11490 
11491         ASSERT(!MUTEX_HELD(&connp->conn_lock));
11492 
11493         if (is_system_labeled() && connp->conn_anon_port) {
11494                 (void) tsol_mlp_anon(crgetzone(connp->conn_cred),
11495                     connp->conn_mlp_type, connp->conn_proto,
11496                     ntohs(connp->conn_lport), B_FALSE);
11497                 connp->conn_anon_port = 0;
11498         }
11499         connp->conn_mlp_type = mlptSingle;
11500 
11501         ipcl_hash_remove(connp);
11502 }
11503 
11504 /*
11505  * Used for deciding the MSS size for the upper layer. Thus
11506  * we need to check the outbound policy values in the conn.
11507  */
11508 int
11509 conn_ipsec_length(conn_t *connp)
11510 {
11511         ipsec_latch_t *ipl;
11512 
11513         ipl = connp->conn_latch;
11514         if (ipl == NULL)
11515                 return (0);
11516 
11517         if (connp->conn_ixa->ixa_ipsec_policy == NULL)
11518                 return (0);
11519 
11520         return (connp->conn_ixa->ixa_ipsec_policy->ipsp_act->ipa_ovhd);
11521 }
11522 
11523 /*
11524  * Returns an estimate of the IPsec headers size. This is used if
11525  * we don't want to call into IPsec to get the exact size.
11526  */
11527 int
11528 ipsec_out_extra_length(ip_xmit_attr_t *ixa)
11529 {
11530         ipsec_action_t *a;
11531 
11532         if (!(ixa->ixa_flags & IXAF_IPSEC_SECURE))
11533                 return (0);
11534 
11535         a = ixa->ixa_ipsec_action;
11536         if (a == NULL) {
11537                 ASSERT(ixa->ixa_ipsec_policy != NULL);
11538                 a = ixa->ixa_ipsec_policy->ipsp_act;
11539         }
11540         ASSERT(a != NULL);
11541 
11542         return (a->ipa_ovhd);
11543 }
11544 
11545 /*
11546  * If there are any source route options, return the true final
11547  * destination. Otherwise, return the destination.
11548  */
11549 ipaddr_t
11550 ip_get_dst(ipha_t *ipha)
11551 {
11552         ipoptp_t        opts;
11553         uchar_t         *opt;
11554         uint8_t         optval;
11555         uint8_t         optlen;
11556         ipaddr_t        dst;
11557         uint32_t off;
11558 
11559         dst = ipha->ipha_dst;
11560 
11561         if (IS_SIMPLE_IPH(ipha))
11562                 return (dst);
11563 
11564         for (optval = ipoptp_first(&opts, ipha);
11565             optval != IPOPT_EOL;
11566             optval = ipoptp_next(&opts)) {
11567                 opt = opts.ipoptp_cur;
11568                 optlen = opts.ipoptp_len;
11569                 ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
11570                 switch (optval) {
11571                 case IPOPT_SSRR:
11572                 case IPOPT_LSRR:
11573                         off = opt[IPOPT_OFFSET];
11574                         /*
11575                          * If one of the conditions is true, it means
11576                          * end of options and dst already has the right
11577                          * value.
11578                          */
11579                         if (!(optlen < IP_ADDR_LEN || off > optlen - 3)) {
11580                                 off = optlen - IP_ADDR_LEN;
11581                                 bcopy(&opt[off], &dst, IP_ADDR_LEN);
11582                         }
11583                         return (dst);
11584                 default:
11585                         break;
11586                 }
11587         }
11588 
11589         return (dst);
11590 }
11591 
11592 /*
11593  * Outbound IP fragmentation routine.
11594  * Assumes the caller has checked whether or not fragmentation should
11595  * be allowed. Here we copy the DF bit from the header to all the generated
11596  * fragments.
11597  */
11598 int
11599 ip_fragment_v4(mblk_t *mp_orig, nce_t *nce, iaflags_t ixaflags,
11600     uint_t pkt_len, uint32_t max_frag, uint32_t xmit_hint, zoneid_t szone,
11601     zoneid_t nolzid, pfirepostfrag_t postfragfn, uintptr_t *ixa_cookie)
11602 {
11603         int             i1;
11604         int             hdr_len;
11605         mblk_t          *hdr_mp;
11606         ipha_t          *ipha;
11607         int             ip_data_end;
11608         int             len;
11609         mblk_t          *mp = mp_orig;
11610         int             offset;
11611         ill_t           *ill = nce->nce_ill;
11612         ip_stack_t      *ipst = ill->ill_ipst;
11613         mblk_t          *carve_mp;
11614         uint32_t        frag_flag;
11615         uint_t          priority = mp->b_band;
11616         int             error = 0;
11617 
11618         BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragReqds);
11619 
11620         if (pkt_len != msgdsize(mp)) {
11621                 ip0dbg(("Packet length mismatch: %d, %ld\n",
11622                     pkt_len, msgdsize(mp)));
11623                 freemsg(mp);
11624                 return (EINVAL);
11625         }
11626 
11627         if (max_frag == 0) {
11628                 ip1dbg(("ip_fragment_v4: max_frag is zero. Dropping packet\n"));
11629                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11630                 ip_drop_output("FragFails: zero max_frag", mp, ill);
11631                 freemsg(mp);
11632                 return (EINVAL);
11633         }
11634 
11635         ASSERT(MBLKL(mp) >= sizeof (ipha_t));
11636         ipha = (ipha_t *)mp->b_rptr;
11637         ASSERT(ntohs(ipha->ipha_length) == pkt_len);
11638         frag_flag = ntohs(ipha->ipha_fragment_offset_and_flags) & IPH_DF;
11639 
11640         /*
11641          * Establish the starting offset.  May not be zero if we are fragging
11642          * a fragment that is being forwarded.
11643          */
11644         offset = ntohs(ipha->ipha_fragment_offset_and_flags) & IPH_OFFSET;
11645 
11646         /* TODO why is this test needed? */
11647         if (((max_frag - ntohs(ipha->ipha_length)) & ~7) < 8) {
11648                 /* TODO: notify ulp somehow */
11649                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11650                 ip_drop_output("FragFails: bad starting offset", mp, ill);
11651                 freemsg(mp);
11652                 return (EINVAL);
11653         }
11654 
11655         hdr_len = IPH_HDR_LENGTH(ipha);
11656         ipha->ipha_hdr_checksum = 0;
11657 
11658         /*
11659          * Establish the number of bytes maximum per frag, after putting
11660          * in the header.
11661          */
11662         len = (max_frag - hdr_len) & ~7;
11663 
11664         /* Get a copy of the header for the trailing frags */
11665         hdr_mp = ip_fragment_copyhdr((uchar_t *)ipha, hdr_len, offset, ipst,
11666             mp);
11667         if (hdr_mp == NULL) {
11668                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11669                 ip_drop_output("FragFails: no hdr_mp", mp, ill);
11670                 freemsg(mp);
11671                 return (ENOBUFS);
11672         }
11673 
11674         /* Store the starting offset, with the MoreFrags flag. */
11675         i1 = offset | IPH_MF | frag_flag;
11676         ipha->ipha_fragment_offset_and_flags = htons((uint16_t)i1);
11677 
11678         /* Establish the ending byte offset, based on the starting offset. */
11679         offset <<= 3;
11680         ip_data_end = offset + ntohs(ipha->ipha_length) - hdr_len;
11681 
11682         /* Store the length of the first fragment in the IP header. */
11683         i1 = len + hdr_len;
11684         ASSERT(i1 <= IP_MAXPACKET);
11685         ipha->ipha_length = htons((uint16_t)i1);
11686 
11687         /*
11688          * Compute the IP header checksum for the first frag.  We have to
11689          * watch out that we stop at the end of the header.
11690          */
11691         ipha->ipha_hdr_checksum = ip_csum_hdr(ipha);
11692 
11693         /*
11694          * Now carve off the first frag.  Note that this will include the
11695          * original IP header.
11696          */
11697         if (!(mp = ip_carve_mp(&mp_orig, i1))) {
11698                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11699                 ip_drop_output("FragFails: could not carve mp", mp_orig, ill);
11700                 freeb(hdr_mp);
11701                 freemsg(mp_orig);
11702                 return (ENOBUFS);
11703         }
11704 
11705         BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragCreates);
11706 
11707         error = postfragfn(mp, nce, ixaflags, i1, xmit_hint, szone, nolzid,
11708             ixa_cookie);
11709         if (error != 0 && error != EWOULDBLOCK) {
11710                 /* No point in sending the other fragments */
11711                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11712                 ip_drop_output("FragFails: postfragfn failed", mp_orig, ill);
11713                 freeb(hdr_mp);
11714                 freemsg(mp_orig);
11715                 return (error);
11716         }
11717 
11718         /* No need to redo state machine in loop */
11719         ixaflags &= ~IXAF_REACH_CONF;
11720 
11721         /* Advance the offset to the second frag starting point. */
11722         offset += len;
11723         /*
11724          * Update hdr_len from the copied header - there might be less options
11725          * in the later fragments.
11726          */
11727         hdr_len = IPH_HDR_LENGTH(hdr_mp->b_rptr);
11728         /* Loop until done. */
11729         for (;;) {
11730                 uint16_t        offset_and_flags;
11731                 uint16_t        ip_len;
11732 
11733                 if (ip_data_end - offset > len) {
11734                         /*
11735                          * Carve off the appropriate amount from the original
11736                          * datagram.
11737                          */
11738                         if (!(carve_mp = ip_carve_mp(&mp_orig, len))) {
11739                                 mp = NULL;
11740                                 break;
11741                         }
11742                         /*
11743                          * More frags after this one.  Get another copy
11744                          * of the header.
11745                          */
11746                         if (carve_mp->b_datap->db_ref == 1 &&
11747                             hdr_mp->b_wptr - hdr_mp->b_rptr <
11748                             carve_mp->b_rptr - carve_mp->b_datap->db_base) {
11749                                 /* Inline IP header */
11750                                 carve_mp->b_rptr -= hdr_mp->b_wptr -
11751                                     hdr_mp->b_rptr;
11752                                 bcopy(hdr_mp->b_rptr, carve_mp->b_rptr,
11753                                     hdr_mp->b_wptr - hdr_mp->b_rptr);
11754                                 mp = carve_mp;
11755                         } else {
11756                                 if (!(mp = copyb(hdr_mp))) {
11757                                         freemsg(carve_mp);
11758                                         break;
11759                                 }
11760                                 /* Get priority marking, if any. */
11761                                 mp->b_band = priority;
11762                                 mp->b_cont = carve_mp;
11763                         }
11764                         ipha = (ipha_t *)mp->b_rptr;
11765                         offset_and_flags = IPH_MF;
11766                 } else {
11767                         /*
11768                          * Last frag.  Consume the header. Set len to
11769                          * the length of this last piece.
11770                          */
11771                         len = ip_data_end - offset;
11772 
11773                         /*
11774                          * Carve off the appropriate amount from the original
11775                          * datagram.
11776                          */
11777                         if (!(carve_mp = ip_carve_mp(&mp_orig, len))) {
11778                                 mp = NULL;
11779                                 break;
11780                         }
11781                         if (carve_mp->b_datap->db_ref == 1 &&
11782                             hdr_mp->b_wptr - hdr_mp->b_rptr <
11783                             carve_mp->b_rptr - carve_mp->b_datap->db_base) {
11784                                 /* Inline IP header */
11785                                 carve_mp->b_rptr -= hdr_mp->b_wptr -
11786                                     hdr_mp->b_rptr;
11787                                 bcopy(hdr_mp->b_rptr, carve_mp->b_rptr,
11788                                     hdr_mp->b_wptr - hdr_mp->b_rptr);
11789                                 mp = carve_mp;
11790                                 freeb(hdr_mp);
11791                                 hdr_mp = mp;
11792                         } else {
11793                                 mp = hdr_mp;
11794                                 /* Get priority marking, if any. */
11795                                 mp->b_band = priority;
11796                                 mp->b_cont = carve_mp;
11797                         }
11798                         ipha = (ipha_t *)mp->b_rptr;
11799                         /* A frag of a frag might have IPH_MF non-zero */
11800                         offset_and_flags =
11801                             ntohs(ipha->ipha_fragment_offset_and_flags) &
11802                             IPH_MF;
11803                 }
11804                 offset_and_flags |= (uint16_t)(offset >> 3);
11805                 offset_and_flags |= (uint16_t)frag_flag;
11806                 /* Store the offset and flags in the IP header. */
11807                 ipha->ipha_fragment_offset_and_flags = htons(offset_and_flags);
11808 
11809                 /* Store the length in the IP header. */
11810                 ip_len = (uint16_t)(len + hdr_len);
11811                 ipha->ipha_length = htons(ip_len);
11812 
11813                 /*
11814                  * Set the IP header checksum.  Note that mp is just
11815                  * the header, so this is easy to pass to ip_csum.
11816                  */
11817                 ipha->ipha_hdr_checksum = ip_csum_hdr(ipha);
11818 
11819                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragCreates);
11820 
11821                 error = postfragfn(mp, nce, ixaflags, ip_len, xmit_hint, szone,
11822                     nolzid, ixa_cookie);
11823                 /* All done if we just consumed the hdr_mp. */
11824                 if (mp == hdr_mp) {
11825                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragOKs);
11826                         return (error);
11827                 }
11828                 if (error != 0 && error != EWOULDBLOCK) {
11829                         DTRACE_PROBE2(ip__xmit__frag__fail, ill_t *, ill,
11830                             mblk_t *, hdr_mp);
11831                         /* No point in sending the other fragments */
11832                         break;
11833                 }
11834 
11835                 /* Otherwise, advance and loop. */
11836                 offset += len;
11837         }
11838         /* Clean up following allocation failure. */
11839         BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11840         ip_drop_output("FragFails: loop ended", NULL, ill);
11841         if (mp != hdr_mp)
11842                 freeb(hdr_mp);
11843         if (mp != mp_orig)
11844                 freemsg(mp_orig);
11845         return (error);
11846 }
11847 
11848 /*
11849  * Copy the header plus those options which have the copy bit set
11850  */
11851 static mblk_t *
11852 ip_fragment_copyhdr(uchar_t *rptr, int hdr_len, int offset, ip_stack_t *ipst,
11853     mblk_t *src)
11854 {
11855         mblk_t  *mp;
11856         uchar_t *up;
11857 
11858         /*
11859          * Quick check if we need to look for options without the copy bit
11860          * set
11861          */
11862         mp = allocb_tmpl(ipst->ips_ip_wroff_extra + hdr_len, src);
11863         if (!mp)
11864                 return (mp);
11865         mp->b_rptr += ipst->ips_ip_wroff_extra;
11866         if (hdr_len == IP_SIMPLE_HDR_LENGTH || offset != 0) {
11867                 bcopy(rptr, mp->b_rptr, hdr_len);
11868                 mp->b_wptr += hdr_len + ipst->ips_ip_wroff_extra;
11869                 return (mp);
11870         }
11871         up  = mp->b_rptr;
11872         bcopy(rptr, up, IP_SIMPLE_HDR_LENGTH);
11873         up += IP_SIMPLE_HDR_LENGTH;
11874         rptr += IP_SIMPLE_HDR_LENGTH;
11875         hdr_len -= IP_SIMPLE_HDR_LENGTH;
11876         while (hdr_len > 0) {
11877                 uint32_t optval;
11878                 uint32_t optlen;
11879 
11880                 optval = *rptr;
11881                 if (optval == IPOPT_EOL)
11882                         break;
11883                 if (optval == IPOPT_NOP)
11884                         optlen = 1;
11885                 else
11886                         optlen = rptr[1];
11887                 if (optval & IPOPT_COPY) {
11888                         bcopy(rptr, up, optlen);
11889                         up += optlen;
11890                 }
11891                 rptr += optlen;
11892                 hdr_len -= optlen;
11893         }
11894         /*
11895          * Make sure that we drop an even number of words by filling
11896          * with EOL to the next word boundary.
11897          */
11898         for (hdr_len = up - (mp->b_rptr + IP_SIMPLE_HDR_LENGTH);
11899             hdr_len & 0x3; hdr_len++)
11900                 *up++ = IPOPT_EOL;
11901         mp->b_wptr = up;
11902         /* Update header length */
11903         mp->b_rptr[0] = (uint8_t)((IP_VERSION << 4) | ((up - mp->b_rptr) >> 2));
11904         return (mp);
11905 }
11906 
11907 /*
11908  * Update any source route, record route, or timestamp options when
11909  * sending a packet back to ourselves.
11910  * Check that we are at end of strict source route.
11911  * The options have been sanity checked by ip_output_options().
11912  */
11913 void
11914 ip_output_local_options(ipha_t *ipha, ip_stack_t *ipst)
11915 {
11916         ipoptp_t        opts;
11917         uchar_t         *opt;
11918         uint8_t         optval;
11919         uint8_t         optlen;
11920         ipaddr_t        dst;
11921         uint32_t        ts;
11922         timestruc_t     now;
11923 
11924         for (optval = ipoptp_first(&opts, ipha);
11925             optval != IPOPT_EOL;
11926             optval = ipoptp_next(&opts)) {
11927                 opt = opts.ipoptp_cur;
11928                 optlen = opts.ipoptp_len;
11929                 ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
11930                 switch (optval) {
11931                         uint32_t off;
11932                 case IPOPT_SSRR:
11933                 case IPOPT_LSRR:
11934                         off = opt[IPOPT_OFFSET];
11935                         off--;
11936                         if (optlen < IP_ADDR_LEN ||
11937                             off > optlen - IP_ADDR_LEN) {
11938                                 /* End of source route */
11939                                 break;
11940                         }
11941                         /*
11942                          * This will only happen if two consecutive entries
11943                          * in the source route contains our address or if
11944                          * it is a packet with a loose source route which
11945                          * reaches us before consuming the whole source route
11946                          */
11947 
11948                         if (optval == IPOPT_SSRR) {
11949                                 return;
11950                         }
11951                         /*
11952                          * Hack: instead of dropping the packet truncate the
11953                          * source route to what has been used by filling the
11954                          * rest with IPOPT_NOP.
11955                          */
11956                         opt[IPOPT_OLEN] = (uint8_t)off;
11957                         while (off < optlen) {
11958                                 opt[off++] = IPOPT_NOP;
11959                         }
11960                         break;
11961                 case IPOPT_RR:
11962                         off = opt[IPOPT_OFFSET];
11963                         off--;
11964                         if (optlen < IP_ADDR_LEN ||
11965                             off > optlen - IP_ADDR_LEN) {
11966                                 /* No more room - ignore */
11967                                 ip1dbg((
11968                                     "ip_output_local_options: end of RR\n"));
11969                                 break;
11970                         }
11971                         dst = htonl(INADDR_LOOPBACK);
11972                         bcopy(&dst, (char *)opt + off, IP_ADDR_LEN);
11973                         opt[IPOPT_OFFSET] += IP_ADDR_LEN;
11974                         break;
11975                 case IPOPT_TS:
11976                         /* Insert timestamp if there is romm */
11977                         switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
11978                         case IPOPT_TS_TSONLY:
11979                                 off = IPOPT_TS_TIMELEN;
11980                                 break;
11981                         case IPOPT_TS_PRESPEC:
11982                         case IPOPT_TS_PRESPEC_RFC791:
11983                                 /* Verify that the address matched */
11984                                 off = opt[IPOPT_OFFSET] - 1;
11985                                 bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
11986                                 if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
11987                                         /* Not for us */
11988                                         break;
11989                                 }
11990                                 /* FALLTHRU */
11991                         case IPOPT_TS_TSANDADDR:
11992                                 off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
11993                                 break;
11994                         default:
11995                                 /*
11996                                  * ip_*put_options should have already
11997                                  * dropped this packet.
11998                                  */
11999                                 cmn_err(CE_PANIC, "ip_output_local_options: "
12000                                     "unknown IT - bug in ip_output_options?\n");
12001                                 return; /* Keep "lint" happy */
12002                         }
12003                         if (opt[IPOPT_OFFSET] - 1 + off > optlen) {
12004                                 /* Increase overflow counter */
12005                                 off = (opt[IPOPT_POS_OV_FLG] >> 4) + 1;
12006                                 opt[IPOPT_POS_OV_FLG] = (uint8_t)
12007                                     (opt[IPOPT_POS_OV_FLG] & 0x0F) |
12008                                     (off << 4);
12009                                 break;
12010                         }
12011                         off = opt[IPOPT_OFFSET] - 1;
12012                         switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
12013                         case IPOPT_TS_PRESPEC:
12014                         case IPOPT_TS_PRESPEC_RFC791:
12015                         case IPOPT_TS_TSANDADDR:
12016                                 dst = htonl(INADDR_LOOPBACK);
12017                                 bcopy(&dst, (char *)opt + off, IP_ADDR_LEN);
12018                                 opt[IPOPT_OFFSET] += IP_ADDR_LEN;
12019                                 /* FALLTHRU */
12020                         case IPOPT_TS_TSONLY:
12021                                 off = opt[IPOPT_OFFSET] - 1;
12022                                 /* Compute # of milliseconds since midnight */
12023                                 gethrestime(&now);
12024                                 ts = (now.tv_sec % (24 * 60 * 60)) * 1000 +
12025                                     now.tv_nsec / (NANOSEC / MILLISEC);
12026                                 bcopy(&ts, (char *)opt + off, IPOPT_TS_TIMELEN);
12027                                 opt[IPOPT_OFFSET] += IPOPT_TS_TIMELEN;
12028                                 break;
12029                         }
12030                         break;
12031                 }
12032         }
12033 }
12034 
12035 /*
12036  * Prepend an M_DATA fastpath header, and if none present prepend a
12037  * DL_UNITDATA_REQ. Frees the mblk on failure.
12038  *
12039  * nce_dlur_mp and nce_fp_mp can not disappear once they have been set.
12040  * If there is a change to them, the nce will be deleted (condemned) and
12041  * a new nce_t will be created when packets are sent. Thus we need no locks
12042  * to access those fields.
12043  *
12044  * We preserve b_band to support IPQoS. If a DL_UNITDATA_REQ is prepended
12045  * we place b_band in dl_priority.dl_max.
12046  */
12047 static mblk_t *
12048 ip_xmit_attach_llhdr(mblk_t *mp, nce_t *nce)
12049 {
12050         uint_t  hlen;
12051         mblk_t *mp1;
12052         uint_t  priority;
12053         uchar_t *rptr;
12054 
12055         rptr = mp->b_rptr;
12056 
12057         ASSERT(DB_TYPE(mp) == M_DATA);
12058         priority = mp->b_band;
12059 
12060         ASSERT(nce != NULL);
12061         if ((mp1 = nce->nce_fp_mp) != NULL) {
12062                 hlen = MBLKL(mp1);
12063                 /*
12064                  * Check if we have enough room to prepend fastpath
12065                  * header
12066                  */
12067                 if (hlen != 0 && (rptr - mp->b_datap->db_base) >= hlen) {
12068                         rptr -= hlen;
12069                         bcopy(mp1->b_rptr, rptr, hlen);
12070                         /*
12071                          * Set the b_rptr to the start of the link layer
12072                          * header
12073                          */
12074                         mp->b_rptr = rptr;
12075                         return (mp);
12076                 }
12077                 mp1 = copyb(mp1);
12078                 if (mp1 == NULL) {
12079                         ill_t *ill = nce->nce_ill;
12080 
12081                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
12082                         ip_drop_output("ipIfStatsOutDiscards", mp, ill);
12083                         freemsg(mp);
12084                         return (NULL);
12085                 }
12086                 mp1->b_band = priority;
12087                 mp1->b_cont = mp;
12088                 DB_CKSUMSTART(mp1) = DB_CKSUMSTART(mp);
12089                 DB_CKSUMSTUFF(mp1) = DB_CKSUMSTUFF(mp);
12090                 DB_CKSUMEND(mp1) = DB_CKSUMEND(mp);
12091                 DB_CKSUMFLAGS(mp1) = DB_CKSUMFLAGS(mp);
12092                 DB_LSOMSS(mp1) = DB_LSOMSS(mp);
12093                 DTRACE_PROBE1(ip__xmit__copyb, (mblk_t *), mp1);
12094                 /*
12095                  * XXX disable ICK_VALID and compute checksum
12096                  * here; can happen if nce_fp_mp changes and
12097                  * it can't be copied now due to insufficient
12098                  * space. (unlikely, fp mp can change, but it
12099                  * does not increase in length)
12100                  */
12101                 return (mp1);
12102         }
12103         mp1 = copyb(nce->nce_dlur_mp);
12104 
12105         if (mp1 == NULL) {
12106                 ill_t *ill = nce->nce_ill;
12107 
12108                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
12109                 ip_drop_output("ipIfStatsOutDiscards", mp, ill);
12110                 freemsg(mp);
12111                 return (NULL);
12112         }
12113         mp1->b_cont = mp;
12114         if (priority != 0) {
12115                 mp1->b_band = priority;
12116                 ((dl_unitdata_req_t *)(mp1->b_rptr))->dl_priority.dl_max =
12117                     priority;
12118         }
12119         return (mp1);
12120 #undef rptr
12121 }
12122 
12123 /*
12124  * Finish the outbound IPsec processing. This function is called from
12125  * ipsec_out_process() if the IPsec packet was processed
12126  * synchronously, or from {ah,esp}_kcf_callback_outbound() if it was processed
12127  * asynchronously.
12128  *
12129  * This is common to IPv4 and IPv6.
12130  */
12131 int
12132 ip_output_post_ipsec(mblk_t *mp, ip_xmit_attr_t *ixa)
12133 {
12134         iaflags_t       ixaflags = ixa->ixa_flags;
12135         uint_t          pktlen;
12136 
12137 
12138         /* AH/ESP don't update ixa_pktlen when they modify the packet */
12139         if (ixaflags & IXAF_IS_IPV4) {
12140                 ipha_t          *ipha = (ipha_t *)mp->b_rptr;
12141 
12142                 ASSERT(IPH_HDR_VERSION(ipha) == IPV4_VERSION);
12143                 pktlen = ntohs(ipha->ipha_length);
12144         } else {
12145                 ip6_t           *ip6h = (ip6_t *)mp->b_rptr;
12146 
12147                 ASSERT(IPH_HDR_VERSION(mp->b_rptr) == IPV6_VERSION);
12148                 pktlen = ntohs(ip6h->ip6_plen) + IPV6_HDR_LEN;
12149         }
12150 
12151         /*
12152          * We release any hard reference on the SAs here to make
12153          * sure the SAs can be garbage collected. ipsr_sa has a soft reference
12154          * on the SAs.
12155          * If in the future we want the hard latching of the SAs in the
12156          * ip_xmit_attr_t then we should remove this.
12157          */
12158         if (ixa->ixa_ipsec_esp_sa != NULL) {
12159                 IPSA_REFRELE(ixa->ixa_ipsec_esp_sa);
12160                 ixa->ixa_ipsec_esp_sa = NULL;
12161         }
12162         if (ixa->ixa_ipsec_ah_sa != NULL) {
12163                 IPSA_REFRELE(ixa->ixa_ipsec_ah_sa);
12164                 ixa->ixa_ipsec_ah_sa = NULL;
12165         }
12166 
12167         /* Do we need to fragment? */
12168         if ((ixa->ixa_flags & IXAF_IPV6_ADD_FRAGHDR) ||
12169             pktlen > ixa->ixa_fragsize) {
12170                 if (ixaflags & IXAF_IS_IPV4) {
12171                         ASSERT(!(ixa->ixa_flags & IXAF_IPV6_ADD_FRAGHDR));
12172                         /*
12173                          * We check for the DF case in ipsec_out_process
12174                          * hence this only handles the non-DF case.
12175                          */
12176                         return (ip_fragment_v4(mp, ixa->ixa_nce, ixa->ixa_flags,
12177                             pktlen, ixa->ixa_fragsize,
12178                             ixa->ixa_xmit_hint, ixa->ixa_zoneid,
12179                             ixa->ixa_no_loop_zoneid, ixa->ixa_postfragfn,
12180                             &ixa->ixa_cookie));
12181                 } else {
12182                         mp = ip_fraghdr_add_v6(mp, ixa->ixa_ident, ixa);
12183                         if (mp == NULL) {
12184                                 /* MIB and ip_drop_output already done */
12185                                 return (ENOMEM);
12186                         }
12187                         pktlen += sizeof (ip6_frag_t);
12188                         if (pktlen > ixa->ixa_fragsize) {
12189                                 return (ip_fragment_v6(mp, ixa->ixa_nce,
12190                                     ixa->ixa_flags, pktlen,
12191                                     ixa->ixa_fragsize, ixa->ixa_xmit_hint,
12192                                     ixa->ixa_zoneid, ixa->ixa_no_loop_zoneid,
12193                                     ixa->ixa_postfragfn, &ixa->ixa_cookie));
12194                         }
12195                 }
12196         }
12197         return ((ixa->ixa_postfragfn)(mp, ixa->ixa_nce, ixa->ixa_flags,
12198             pktlen, ixa->ixa_xmit_hint, ixa->ixa_zoneid,
12199             ixa->ixa_no_loop_zoneid, NULL));
12200 }
12201 
12202 /*
12203  * Finish the inbound IPsec processing. This function is called from
12204  * ipsec_out_process() if the IPsec packet was processed
12205  * synchronously, or from {ah,esp}_kcf_callback_outbound() if it was processed
12206  * asynchronously.
12207  *
12208  * This is common to IPv4 and IPv6.
12209  */
12210 void
12211 ip_input_post_ipsec(mblk_t *mp, ip_recv_attr_t *ira)
12212 {
12213         iaflags_t       iraflags = ira->ira_flags;
12214 
12215         /* Length might have changed */
12216         if (iraflags & IRAF_IS_IPV4) {
12217                 ipha_t          *ipha = (ipha_t *)mp->b_rptr;
12218 
12219                 ASSERT(IPH_HDR_VERSION(ipha) == IPV4_VERSION);
12220                 ira->ira_pktlen = ntohs(ipha->ipha_length);
12221                 ira->ira_ip_hdr_length = IPH_HDR_LENGTH(ipha);
12222                 ira->ira_protocol = ipha->ipha_protocol;
12223 
12224                 ip_fanout_v4(mp, ipha, ira);
12225         } else {
12226                 ip6_t           *ip6h = (ip6_t *)mp->b_rptr;
12227                 uint8_t         *nexthdrp;
12228 
12229                 ASSERT(IPH_HDR_VERSION(mp->b_rptr) == IPV6_VERSION);
12230                 ira->ira_pktlen = ntohs(ip6h->ip6_plen) + IPV6_HDR_LEN;
12231                 if (!ip_hdr_length_nexthdr_v6(mp, ip6h, &ira->ira_ip_hdr_length,
12232                     &nexthdrp)) {
12233                         /* Malformed packet */
12234                         BUMP_MIB(ira->ira_ill->ill_ip_mib, ipIfStatsInDiscards);
12235                         ip_drop_input("ipIfStatsInDiscards", mp, ira->ira_ill);
12236                         freemsg(mp);
12237                         return;
12238                 }
12239                 ira->ira_protocol = *nexthdrp;
12240                 ip_fanout_v6(mp, ip6h, ira);
12241         }
12242 }
12243 
12244 /*
12245  * Select which AH & ESP SA's to use (if any) for the outbound packet.
12246  *
12247  * If this function returns B_TRUE, the requested SA's have been filled
12248  * into the ixa_ipsec_*_sa pointers.
12249  *
12250  * If the function returns B_FALSE, the packet has been "consumed", most
12251  * likely by an ACQUIRE sent up via PF_KEY to a key management daemon.
12252  *
12253  * The SA references created by the protocol-specific "select"
12254  * function will be released in ip_output_post_ipsec.
12255  */
12256 static boolean_t
12257 ipsec_out_select_sa(mblk_t *mp, ip_xmit_attr_t *ixa)
12258 {
12259         boolean_t need_ah_acquire = B_FALSE, need_esp_acquire = B_FALSE;
12260         ipsec_policy_t *pp;
12261         ipsec_action_t *ap;
12262 
12263         ASSERT(ixa->ixa_flags & IXAF_IPSEC_SECURE);
12264         ASSERT((ixa->ixa_ipsec_policy != NULL) ||
12265             (ixa->ixa_ipsec_action != NULL));
12266 
12267         ap = ixa->ixa_ipsec_action;
12268         if (ap == NULL) {
12269                 pp = ixa->ixa_ipsec_policy;
12270                 ASSERT(pp != NULL);
12271                 ap = pp->ipsp_act;
12272                 ASSERT(ap != NULL);
12273         }
12274 
12275         /*
12276          * We have an action.  now, let's select SA's.
12277          * A side effect of setting ixa_ipsec_*_sa is that it will
12278          * be cached in the conn_t.
12279          */
12280         if (ap->ipa_want_esp) {
12281                 if (ixa->ixa_ipsec_esp_sa == NULL) {
12282                         need_esp_acquire = !ipsec_outbound_sa(mp, ixa,
12283                             IPPROTO_ESP);
12284                 }
12285                 ASSERT(need_esp_acquire || ixa->ixa_ipsec_esp_sa != NULL);
12286         }
12287 
12288         if (ap->ipa_want_ah) {
12289                 if (ixa->ixa_ipsec_ah_sa == NULL) {
12290                         need_ah_acquire = !ipsec_outbound_sa(mp, ixa,
12291                             IPPROTO_AH);
12292                 }
12293                 ASSERT(need_ah_acquire || ixa->ixa_ipsec_ah_sa != NULL);
12294                 /*
12295                  * The ESP and AH processing order needs to be preserved
12296                  * when both protocols are required (ESP should be applied
12297                  * before AH for an outbound packet). Force an ESP ACQUIRE
12298                  * when both ESP and AH are required, and an AH ACQUIRE
12299                  * is needed.
12300                  */
12301                 if (ap->ipa_want_esp && need_ah_acquire)
12302                         need_esp_acquire = B_TRUE;
12303         }
12304 
12305         /*
12306          * Send an ACQUIRE (extended, regular, or both) if we need one.
12307          * Release SAs that got referenced, but will not be used until we
12308          * acquire _all_ of the SAs we need.
12309          */
12310         if (need_ah_acquire || need_esp_acquire) {
12311                 if (ixa->ixa_ipsec_ah_sa != NULL) {
12312                         IPSA_REFRELE(ixa->ixa_ipsec_ah_sa);
12313                         ixa->ixa_ipsec_ah_sa = NULL;
12314                 }
12315                 if (ixa->ixa_ipsec_esp_sa != NULL) {
12316                         IPSA_REFRELE(ixa->ixa_ipsec_esp_sa);
12317                         ixa->ixa_ipsec_esp_sa = NULL;
12318                 }
12319 
12320                 sadb_acquire(mp, ixa, need_ah_acquire, need_esp_acquire);
12321                 return (B_FALSE);
12322         }
12323 
12324         return (B_TRUE);
12325 }
12326 
12327 /*
12328  * Handle IPsec output processing.
12329  * This function is only entered once for a given packet.
12330  * We try to do things synchronously, but if we need to have user-level
12331  * set up SAs, or ESP or AH uses asynchronous kEF, then the operation
12332  * will be completed
12333  *  - when the SAs are added in esp_add_sa_finish/ah_add_sa_finish
12334  *  - when asynchronous ESP is done it will do AH
12335  *
12336  * In all cases we come back in ip_output_post_ipsec() to fragment and
12337  * send out the packet.
12338  */
12339 int
12340 ipsec_out_process(mblk_t *mp, ip_xmit_attr_t *ixa)
12341 {
12342         ill_t           *ill = ixa->ixa_nce->nce_ill;
12343         ip_stack_t      *ipst = ixa->ixa_ipst;
12344         ipsec_stack_t   *ipss;
12345         ipsec_policy_t  *pp;
12346         ipsec_action_t  *ap;
12347 
12348         ASSERT(ixa->ixa_flags & IXAF_IPSEC_SECURE);
12349 
12350         ASSERT((ixa->ixa_ipsec_policy != NULL) ||
12351             (ixa->ixa_ipsec_action != NULL));
12352 
12353         ipss = ipst->ips_netstack->netstack_ipsec;
12354         if (!ipsec_loaded(ipss)) {
12355                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
12356                 ip_drop_packet(mp, B_TRUE, ill,
12357                     DROPPER(ipss, ipds_ip_ipsec_not_loaded),
12358                     &ipss->ipsec_dropper);
12359                 return (ENOTSUP);
12360         }
12361 
12362         ap = ixa->ixa_ipsec_action;
12363         if (ap == NULL) {
12364                 pp = ixa->ixa_ipsec_policy;
12365                 ASSERT(pp != NULL);
12366                 ap = pp->ipsp_act;
12367                 ASSERT(ap != NULL);
12368         }
12369 
12370         /* Handle explicit drop action and bypass. */
12371         switch (ap->ipa_act.ipa_type) {
12372         case IPSEC_ACT_DISCARD:
12373         case IPSEC_ACT_REJECT:
12374                 ip_drop_packet(mp, B_FALSE, ill,
12375                     DROPPER(ipss, ipds_spd_explicit), &ipss->ipsec_spd_dropper);
12376                 return (EHOSTUNREACH);  /* IPsec policy failure */
12377         case IPSEC_ACT_BYPASS:
12378                 return (ip_output_post_ipsec(mp, ixa));
12379         }
12380 
12381         /*
12382          * The order of processing is first insert a IP header if needed.
12383          * Then insert the ESP header and then the AH header.
12384          */
12385         if ((ixa->ixa_flags & IXAF_IS_IPV4) && ap->ipa_want_se) {
12386                 /*
12387                  * First get the outer IP header before sending
12388                  * it to ESP.
12389                  */
12390                 ipha_t *oipha, *iipha;
12391                 mblk_t *outer_mp, *inner_mp;
12392 
12393                 if ((outer_mp = allocb(sizeof (ipha_t), BPRI_HI)) == NULL) {
12394                         (void) mi_strlog(ill->ill_rq, 0,
12395                             SL_ERROR|SL_TRACE|SL_CONSOLE,
12396                             "ipsec_out_process: "
12397                             "Self-Encapsulation failed: Out of memory\n");
12398                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
12399                         ip_drop_output("ipIfStatsOutDiscards", mp, ill);
12400                         freemsg(mp);
12401                         return (ENOBUFS);
12402                 }
12403                 inner_mp = mp;
12404                 ASSERT(inner_mp->b_datap->db_type == M_DATA);
12405                 oipha = (ipha_t *)outer_mp->b_rptr;
12406                 iipha = (ipha_t *)inner_mp->b_rptr;
12407                 *oipha = *iipha;
12408                 outer_mp->b_wptr += sizeof (ipha_t);
12409                 oipha->ipha_length = htons(ntohs(iipha->ipha_length) +
12410                     sizeof (ipha_t));
12411                 oipha->ipha_protocol = IPPROTO_ENCAP;
12412                 oipha->ipha_version_and_hdr_length =
12413                     IP_SIMPLE_HDR_VERSION;
12414                 oipha->ipha_hdr_checksum = 0;
12415                 oipha->ipha_hdr_checksum = ip_csum_hdr(oipha);
12416                 outer_mp->b_cont = inner_mp;
12417                 mp = outer_mp;
12418 
12419                 ixa->ixa_flags |= IXAF_IPSEC_TUNNEL;
12420         }
12421 
12422         /* If we need to wait for a SA then we can't return any errno */
12423         if (((ap->ipa_want_ah && (ixa->ixa_ipsec_ah_sa == NULL)) ||
12424             (ap->ipa_want_esp && (ixa->ixa_ipsec_esp_sa == NULL))) &&
12425             !ipsec_out_select_sa(mp, ixa))
12426                 return (0);
12427 
12428         /*
12429          * By now, we know what SA's to use.  Toss over to ESP & AH
12430          * to do the heavy lifting.
12431          */
12432         if (ap->ipa_want_esp) {
12433                 ASSERT(ixa->ixa_ipsec_esp_sa != NULL);
12434 
12435                 mp = ixa->ixa_ipsec_esp_sa->ipsa_output_func(mp, ixa);
12436                 if (mp == NULL) {
12437                         /*
12438                          * Either it failed or is pending. In the former case
12439                          * ipIfStatsInDiscards was increased.
12440                          */
12441                         return (0);
12442                 }
12443         }
12444 
12445         if (ap->ipa_want_ah) {
12446                 ASSERT(ixa->ixa_ipsec_ah_sa != NULL);
12447 
12448                 mp = ixa->ixa_ipsec_ah_sa->ipsa_output_func(mp, ixa);
12449                 if (mp == NULL) {
12450                         /*
12451                          * Either it failed or is pending. In the former case
12452                          * ipIfStatsInDiscards was increased.
12453                          */
12454                         return (0);
12455                 }
12456         }
12457         /*
12458          * We are done with IPsec processing. Send it over
12459          * the wire.
12460          */
12461         return (ip_output_post_ipsec(mp, ixa));
12462 }
12463 
12464 /*
12465  * ioctls that go through a down/up sequence may need to wait for the down
12466  * to complete. This involves waiting for the ire and ipif refcnts to go down
12467  * to zero. Subsequently the ioctl is restarted from ipif_ill_refrele_tail.
12468  */
12469 /* ARGSUSED */
12470 void
12471 ip_reprocess_ioctl(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *dummy_arg)
12472 {
12473         struct iocblk *iocp;
12474         mblk_t *mp1;
12475         ip_ioctl_cmd_t *ipip;
12476         int err;
12477         sin_t   *sin;
12478         struct lifreq *lifr;
12479         struct ifreq *ifr;
12480 
12481         iocp = (struct iocblk *)mp->b_rptr;
12482         ASSERT(ipsq != NULL);
12483         /* Existence of mp1 verified in ip_wput_nondata */
12484         mp1 = mp->b_cont->b_cont;
12485         ipip = ip_sioctl_lookup(iocp->ioc_cmd);
12486         if (ipip->ipi_cmd == SIOCSLIFNAME || ipip->ipi_cmd == IF_UNITSEL) {
12487                 /*
12488                  * Special case where ipx_current_ipif is not set:
12489                  * ill_phyint_reinit merged the v4 and v6 into a single ipsq.
12490                  * We are here as were not able to complete the operation in
12491                  * ipif_set_values because we could not become exclusive on
12492                  * the new ipsq.
12493                  */
12494                 ill_t *ill = q->q_ptr;
12495                 ipsq_current_start(ipsq, ill->ill_ipif, ipip->ipi_cmd);
12496         }
12497         ASSERT(ipsq->ipsq_xop->ipx_current_ipif != NULL);
12498 
12499         if (ipip->ipi_cmd_type == IF_CMD) {
12500                 /* This a old style SIOC[GS]IF* command */
12501                 ifr = (struct ifreq *)mp1->b_rptr;
12502                 sin = (sin_t *)&ifr->ifr_addr;
12503         } else if (ipip->ipi_cmd_type == LIF_CMD) {
12504                 /* This a new style SIOC[GS]LIF* command */
12505                 lifr = (struct lifreq *)mp1->b_rptr;
12506                 sin = (sin_t *)&lifr->lifr_addr;
12507         } else {
12508                 sin = NULL;
12509         }
12510 
12511         err = (*ipip->ipi_func_restart)(ipsq->ipsq_xop->ipx_current_ipif, sin,
12512             q, mp, ipip, mp1->b_rptr);
12513 
12514         DTRACE_PROBE4(ipif__ioctl, char *, "ip_reprocess_ioctl finish",
12515             int, ipip->ipi_cmd,
12516             ill_t *, ipsq->ipsq_xop->ipx_current_ipif->ipif_ill,
12517             ipif_t *, ipsq->ipsq_xop->ipx_current_ipif);
12518 
12519         ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), ipsq);
12520 }
12521 
12522 /*
12523  * ioctl processing
12524  *
12525  * ioctl processing starts with ip_sioctl_copyin_setup(), which looks up
12526  * the ioctl command in the ioctl tables, determines the copyin data size
12527  * from the ipi_copyin_size field, and does an mi_copyin() of that size.
12528  *
12529  * ioctl processing then continues when the M_IOCDATA makes its way down to
12530  * ip_wput_nondata().  The ioctl is looked up again in the ioctl table, its
12531  * associated 'conn' is refheld till the end of the ioctl and the general
12532  * ioctl processing function ip_process_ioctl() is called to extract the
12533  * arguments and process the ioctl.  To simplify extraction, ioctl commands
12534  * are "typed" based on the arguments they take (e.g., LIF_CMD which takes a
12535  * `struct lifreq'), and a common extract function (e.g., ip_extract_lifreq())
12536  * is used to extract the ioctl's arguments.
12537  *
12538  * ip_process_ioctl determines if the ioctl needs to be serialized, and if
12539  * so goes thru the serialization primitive ipsq_try_enter. Then the
12540  * appropriate function to handle the ioctl is called based on the entry in
12541  * the ioctl table. ioctl completion is encapsulated in ip_ioctl_finish
12542  * which also refreleases the 'conn' that was refheld at the start of the
12543  * ioctl. Finally ipsq_exit is called if needed to exit the ipsq.
12544  *
12545  * Many exclusive ioctls go thru an internal down up sequence as part of
12546  * the operation. For example an attempt to change the IP address of an
12547  * ipif entails ipif_down, set address, ipif_up. Bringing down the interface
12548  * does all the cleanup such as deleting all ires that use this address.
12549  * Then we need to wait till all references to the interface go away.
12550  */
12551 void
12552 ip_process_ioctl(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *arg)
12553 {
12554         struct iocblk *iocp = (struct iocblk *)mp->b_rptr;
12555         ip_ioctl_cmd_t *ipip = arg;
12556         ip_extract_func_t *extract_funcp;
12557         cmd_info_t ci;
12558         int err;
12559         boolean_t entered_ipsq = B_FALSE;
12560 
12561         ip3dbg(("ip_process_ioctl: ioctl %X\n", iocp->ioc_cmd));
12562 
12563         if (ipip == NULL)
12564                 ipip = ip_sioctl_lookup(iocp->ioc_cmd);
12565 
12566         /*
12567          * SIOCLIFADDIF needs to go thru a special path since the
12568          * ill may not exist yet. This happens in the case of lo0
12569          * which is created using this ioctl.
12570          */
12571         if (ipip->ipi_cmd == SIOCLIFADDIF) {
12572                 err = ip_sioctl_addif(NULL, NULL, q, mp, NULL, NULL);
12573                 DTRACE_PROBE4(ipif__ioctl, char *, "ip_process_ioctl finish",
12574                     int, ipip->ipi_cmd, ill_t *, NULL, ipif_t *, NULL);
12575                 ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), NULL);
12576                 return;
12577         }
12578 
12579         ci.ci_ipif = NULL;
12580         switch (ipip->ipi_cmd_type) {
12581         case MISC_CMD:
12582         case MSFILT_CMD:
12583                 /*
12584                  * All MISC_CMD ioctls come in here -- e.g. SIOCGLIFCONF.
12585                  */
12586                 if (ipip->ipi_cmd == IF_UNITSEL) {
12587                         /* ioctl comes down the ill */
12588                         ci.ci_ipif = ((ill_t *)q->q_ptr)->ill_ipif;
12589                         ipif_refhold(ci.ci_ipif);
12590                 }
12591                 err = 0;
12592                 ci.ci_sin = NULL;
12593                 ci.ci_sin6 = NULL;
12594                 ci.ci_lifr = NULL;
12595                 extract_funcp = NULL;
12596                 break;
12597 
12598         case IF_CMD:
12599         case LIF_CMD:
12600                 extract_funcp = ip_extract_lifreq;
12601                 break;
12602 
12603         case ARP_CMD:
12604         case XARP_CMD:
12605                 extract_funcp = ip_extract_arpreq;
12606                 break;
12607 
12608         default:
12609                 ASSERT(0);
12610         }
12611 
12612         if (extract_funcp != NULL) {
12613                 err = (*extract_funcp)(q, mp, ipip, &ci);
12614                 if (err != 0) {
12615                         DTRACE_PROBE4(ipif__ioctl,
12616                             char *, "ip_process_ioctl finish err",
12617                             int, ipip->ipi_cmd, ill_t *, NULL, ipif_t *, NULL);
12618                         ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), NULL);
12619                         return;
12620                 }
12621 
12622                 /*
12623                  * All of the extraction functions return a refheld ipif.
12624                  */
12625                 ASSERT(ci.ci_ipif != NULL);
12626         }
12627 
12628         if (!(ipip->ipi_flags & IPI_WR)) {
12629                 /*
12630                  * A return value of EINPROGRESS means the ioctl is
12631                  * either queued and waiting for some reason or has
12632                  * already completed.
12633                  */
12634                 err = (*ipip->ipi_func)(ci.ci_ipif, ci.ci_sin, q, mp, ipip,
12635                     ci.ci_lifr);
12636                 if (ci.ci_ipif != NULL) {
12637                         DTRACE_PROBE4(ipif__ioctl,
12638                             char *, "ip_process_ioctl finish RD",
12639                             int, ipip->ipi_cmd, ill_t *, ci.ci_ipif->ipif_ill,
12640                             ipif_t *, ci.ci_ipif);
12641                         ipif_refrele(ci.ci_ipif);
12642                 } else {
12643                         DTRACE_PROBE4(ipif__ioctl,
12644                             char *, "ip_process_ioctl finish RD",
12645                             int, ipip->ipi_cmd, ill_t *, NULL, ipif_t *, NULL);
12646                 }
12647                 ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), NULL);
12648                 return;
12649         }
12650 
12651         ASSERT(ci.ci_ipif != NULL);
12652 
12653         /*
12654          * If ipsq is non-NULL, we are already being called exclusively
12655          */
12656         ASSERT(ipsq == NULL || IAM_WRITER_IPSQ(ipsq));
12657         if (ipsq == NULL) {
12658                 ipsq = ipsq_try_enter(ci.ci_ipif, NULL, q, mp, ip_process_ioctl,
12659                     NEW_OP, B_TRUE);
12660                 if (ipsq == NULL) {
12661                         ipif_refrele(ci.ci_ipif);
12662                         return;
12663                 }
12664                 entered_ipsq = B_TRUE;
12665         }
12666         /*
12667          * Release the ipif so that ipif_down and friends that wait for
12668          * references to go away are not misled about the current ipif_refcnt
12669          * values. We are writer so we can access the ipif even after releasing
12670          * the ipif.
12671          */
12672         ipif_refrele(ci.ci_ipif);
12673 
12674         ipsq_current_start(ipsq, ci.ci_ipif, ipip->ipi_cmd);
12675 
12676         /*
12677          * A return value of EINPROGRESS means the ioctl is
12678          * either queued and waiting for some reason or has
12679          * already completed.
12680          */
12681         err = (*ipip->ipi_func)(ci.ci_ipif, ci.ci_sin, q, mp, ipip, ci.ci_lifr);
12682 
12683         DTRACE_PROBE4(ipif__ioctl, char *, "ip_process_ioctl finish WR",
12684             int, ipip->ipi_cmd,
12685             ill_t *, ci.ci_ipif == NULL ? NULL : ci.ci_ipif->ipif_ill,
12686             ipif_t *, ci.ci_ipif);
12687         ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), ipsq);
12688 
12689         if (entered_ipsq)
12690                 ipsq_exit(ipsq);
12691 }
12692 
12693 /*
12694  * Complete the ioctl. Typically ioctls use the mi package and need to
12695  * do mi_copyout/mi_copy_done.
12696  */
12697 void
12698 ip_ioctl_finish(queue_t *q, mblk_t *mp, int err, int mode, ipsq_t *ipsq)
12699 {
12700         conn_t  *connp = NULL;
12701 
12702         if (err == EINPROGRESS)
12703                 return;
12704 
12705         if (CONN_Q(q)) {
12706                 connp = Q_TO_CONN(q);
12707                 ASSERT(connp->conn_ref >= 2);
12708         }
12709 
12710         switch (mode) {
12711         case COPYOUT:
12712                 if (err == 0)
12713                         mi_copyout(q, mp);
12714                 else
12715                         mi_copy_done(q, mp, err);
12716                 break;
12717 
12718         case NO_COPYOUT:
12719                 mi_copy_done(q, mp, err);
12720                 break;
12721 
12722         default:
12723                 ASSERT(mode == CONN_CLOSE);     /* aborted through CONN_CLOSE */
12724                 break;
12725         }
12726 
12727         /*
12728          * The conn refhold and ioctlref placed on the conn at the start of the
12729          * ioctl are released here.
12730          */
12731         if (connp != NULL) {
12732                 CONN_DEC_IOCTLREF(connp);
12733                 CONN_OPER_PENDING_DONE(connp);
12734         }
12735 
12736         if (ipsq != NULL)
12737                 ipsq_current_finish(ipsq);
12738 }
12739 
12740 /* Handles all non data messages */
12741 void
12742 ip_wput_nondata(queue_t *q, mblk_t *mp)
12743 {
12744         mblk_t          *mp1;
12745         struct iocblk   *iocp;
12746         ip_ioctl_cmd_t  *ipip;
12747         conn_t          *connp;
12748         cred_t          *cr;
12749         char            *proto_str;
12750 
12751         if (CONN_Q(q))
12752                 connp = Q_TO_CONN(q);
12753         else
12754                 connp = NULL;
12755 
12756         switch (DB_TYPE(mp)) {
12757         case M_IOCTL:
12758                 /*
12759                  * IOCTL processing begins in ip_sioctl_copyin_setup which
12760                  * will arrange to copy in associated control structures.
12761                  */
12762                 ip_sioctl_copyin_setup(q, mp);
12763                 return;
12764         case M_IOCDATA:
12765                 /*
12766                  * Ensure that this is associated with one of our trans-
12767                  * parent ioctls.  If it's not ours, discard it if we're
12768                  * running as a driver, or pass it on if we're a module.
12769                  */
12770                 iocp = (struct iocblk *)mp->b_rptr;
12771                 ipip = ip_sioctl_lookup(iocp->ioc_cmd);
12772                 if (ipip == NULL) {
12773                         if (q->q_next == NULL) {
12774                                 goto nak;
12775                         } else {
12776                                 putnext(q, mp);
12777                         }
12778                         return;
12779                 }
12780                 if ((q->q_next != NULL) && !(ipip->ipi_flags & IPI_MODOK)) {
12781                         /*
12782                          * The ioctl is one we recognise, but is not consumed
12783                          * by IP as a module and we are a module, so we drop
12784                          */
12785                         goto nak;
12786                 }
12787 
12788                 /* IOCTL continuation following copyin or copyout. */
12789                 if (mi_copy_state(q, mp, NULL) == -1) {
12790                         /*
12791                          * The copy operation failed.  mi_copy_state already
12792                          * cleaned up, so we're out of here.
12793                          */
12794                         return;
12795                 }
12796                 /*
12797                  * If we just completed a copy in, we become writer and
12798                  * continue processing in ip_sioctl_copyin_done.  If it
12799                  * was a copy out, we call mi_copyout again.  If there is
12800                  * nothing more to copy out, it will complete the IOCTL.
12801                  */
12802                 if (MI_COPY_DIRECTION(mp) == MI_COPY_IN) {
12803                         if (!(mp1 = mp->b_cont) || !(mp1 = mp1->b_cont)) {
12804                                 mi_copy_done(q, mp, EPROTO);
12805                                 return;
12806                         }
12807                         /*
12808                          * Check for cases that need more copying.  A return
12809                          * value of 0 means a second copyin has been started,
12810                          * so we return; a return value of 1 means no more
12811                          * copying is needed, so we continue.
12812                          */
12813                         if (ipip->ipi_cmd_type == MSFILT_CMD &&
12814                             MI_COPY_COUNT(mp) == 1) {
12815                                 if (ip_copyin_msfilter(q, mp) == 0)
12816                                         return;
12817                         }
12818                         /*
12819                          * Refhold the conn, till the ioctl completes. This is
12820                          * needed in case the ioctl ends up in the pending mp
12821                          * list. Every mp in the ipx_pending_mp list must have
12822                          * a refhold on the conn to resume processing. The
12823                          * refhold is released when the ioctl completes
12824                          * (whether normally or abnormally). An ioctlref is also
12825                          * placed on the conn to prevent TCP from removing the
12826                          * queue needed to send the ioctl reply back.
12827                          * In all cases ip_ioctl_finish is called to finish
12828                          * the ioctl and release the refholds.
12829                          */
12830                         if (connp != NULL) {
12831                                 /* This is not a reentry */
12832                                 CONN_INC_REF(connp);
12833                                 CONN_INC_IOCTLREF(connp);
12834                         } else {
12835                                 if (!(ipip->ipi_flags & IPI_MODOK)) {
12836                                         mi_copy_done(q, mp, EINVAL);
12837                                         return;
12838                                 }
12839                         }
12840 
12841                         ip_process_ioctl(NULL, q, mp, ipip);
12842 
12843                 } else {
12844                         mi_copyout(q, mp);
12845                 }
12846                 return;
12847 
12848         case M_IOCNAK:
12849                 /*
12850                  * The only way we could get here is if a resolver didn't like
12851                  * an IOCTL we sent it.  This shouldn't happen.
12852                  */
12853                 (void) mi_strlog(q, 1, SL_ERROR|SL_TRACE,
12854                     "ip_wput_nondata: unexpected M_IOCNAK, ioc_cmd 0x%x",
12855                     ((struct iocblk *)mp->b_rptr)->ioc_cmd);
12856                 freemsg(mp);
12857                 return;
12858         case M_IOCACK:
12859                 /* /dev/ip shouldn't see this */
12860                 goto nak;
12861         case M_FLUSH:
12862                 if (*mp->b_rptr & FLUSHW)
12863                         flushq(q, FLUSHALL);
12864                 if (q->q_next) {
12865                         putnext(q, mp);
12866                         return;
12867                 }
12868                 if (*mp->b_rptr & FLUSHR) {
12869                         *mp->b_rptr &= ~FLUSHW;
12870                         qreply(q, mp);
12871                         return;
12872                 }
12873                 freemsg(mp);
12874                 return;
12875         case M_CTL:
12876                 break;
12877         case M_PROTO:
12878         case M_PCPROTO:
12879                 /*
12880                  * The only PROTO messages we expect are SNMP-related.
12881                  */
12882                 switch (((union T_primitives *)mp->b_rptr)->type) {
12883                 case T_SVR4_OPTMGMT_REQ:
12884                         ip2dbg(("ip_wput_nondata: T_SVR4_OPTMGMT_REQ "
12885                             "flags %x\n",
12886                             ((struct T_optmgmt_req *)mp->b_rptr)->MGMT_flags));
12887 
12888                         if (connp == NULL) {
12889                                 proto_str = "T_SVR4_OPTMGMT_REQ";
12890                                 goto protonak;
12891                         }
12892 
12893                         /*
12894                          * All Solaris components should pass a db_credp
12895                          * for this TPI message, hence we ASSERT.
12896                          * But in case there is some other M_PROTO that looks
12897                          * like a TPI message sent by some other kernel
12898                          * component, we check and return an error.
12899                          */
12900                         cr = msg_getcred(mp, NULL);
12901                         ASSERT(cr != NULL);
12902                         if (cr == NULL) {
12903                                 mp = mi_tpi_err_ack_alloc(mp, TSYSERR, EINVAL);
12904                                 if (mp != NULL)
12905                                         qreply(q, mp);
12906                                 return;
12907                         }
12908 
12909                         if (!snmpcom_req(q, mp, ip_snmp_set, ip_snmp_get, cr)) {
12910                                 proto_str = "Bad SNMPCOM request?";
12911                                 goto protonak;
12912                         }
12913                         return;
12914                 default:
12915                         ip1dbg(("ip_wput_nondata: dropping M_PROTO prim %u\n",
12916                             (int)*(uint_t *)mp->b_rptr));
12917                         freemsg(mp);
12918                         return;
12919                 }
12920         default:
12921                 break;
12922         }
12923         if (q->q_next) {
12924                 putnext(q, mp);
12925         } else
12926                 freemsg(mp);
12927         return;
12928 
12929 nak:
12930         iocp->ioc_error = EINVAL;
12931         mp->b_datap->db_type = M_IOCNAK;
12932         iocp->ioc_count = 0;
12933         qreply(q, mp);
12934         return;
12935 
12936 protonak:
12937         cmn_err(CE_NOTE, "IP doesn't process %s as a module", proto_str);
12938         if ((mp = mi_tpi_err_ack_alloc(mp, TPROTO, EINVAL)) != NULL)
12939                 qreply(q, mp);
12940 }
12941 
12942 /*
12943  * Process IP options in an outbound packet.  Verify that the nexthop in a
12944  * strict source route is onlink.
12945  * Returns non-zero if something fails in which case an ICMP error has been
12946  * sent and mp freed.
12947  *
12948  * Assumes the ULP has called ip_massage_options to move nexthop into ipha_dst.
12949  */
12950 int
12951 ip_output_options(mblk_t *mp, ipha_t *ipha, ip_xmit_attr_t *ixa, ill_t *ill)
12952 {
12953         ipoptp_t        opts;
12954         uchar_t         *opt;
12955         uint8_t         optval;
12956         uint8_t         optlen;
12957         ipaddr_t        dst;
12958         intptr_t        code = 0;
12959         ire_t           *ire;
12960         ip_stack_t      *ipst = ixa->ixa_ipst;
12961         ip_recv_attr_t  iras;
12962 
12963         ip2dbg(("ip_output_options\n"));
12964 
12965         dst = ipha->ipha_dst;
12966         for (optval = ipoptp_first(&opts, ipha);
12967             optval != IPOPT_EOL;
12968             optval = ipoptp_next(&opts)) {
12969                 opt = opts.ipoptp_cur;
12970                 optlen = opts.ipoptp_len;
12971                 ip2dbg(("ip_output_options: opt %d, len %d\n",
12972                     optval, optlen));
12973                 switch (optval) {
12974                         uint32_t off;
12975                 case IPOPT_SSRR:
12976                 case IPOPT_LSRR:
12977                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
12978                                 ip1dbg((
12979                                     "ip_output_options: bad option offset\n"));
12980                                 code = (char *)&opt[IPOPT_OLEN] -
12981                                     (char *)ipha;
12982                                 goto param_prob;
12983                         }
12984                         off = opt[IPOPT_OFFSET];
12985                         ip1dbg(("ip_output_options: next hop 0x%x\n",
12986                             ntohl(dst)));
12987                         /*
12988                          * For strict: verify that dst is directly
12989                          * reachable.
12990                          */
12991                         if (optval == IPOPT_SSRR) {
12992                                 ire = ire_ftable_lookup_v4(dst, 0, 0,
12993                                     IRE_INTERFACE, NULL, ALL_ZONES,
12994                                     ixa->ixa_tsl,
12995                                     MATCH_IRE_TYPE | MATCH_IRE_SECATTR, 0, ipst,
12996                                     NULL);
12997                                 if (ire == NULL) {
12998                                         ip1dbg(("ip_output_options: SSRR not"
12999                                             " directly reachable: 0x%x\n",
13000                                             ntohl(dst)));
13001                                         goto bad_src_route;
13002                                 }
13003                                 ire_refrele(ire);
13004                         }
13005                         break;
13006                 case IPOPT_RR:
13007                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
13008                                 ip1dbg((
13009                                     "ip_output_options: bad option offset\n"));
13010                                 code = (char *)&opt[IPOPT_OLEN] -
13011                                     (char *)ipha;
13012                                 goto param_prob;
13013                         }
13014                         break;
13015                 case IPOPT_TS:
13016                         /*
13017                          * Verify that length >=5 and that there is either
13018                          * room for another timestamp or that the overflow
13019                          * counter is not maxed out.
13020                          */
13021                         code = (char *)&opt[IPOPT_OLEN] - (char *)ipha;
13022                         if (optlen < IPOPT_MINLEN_IT) {
13023                                 goto param_prob;
13024                         }
13025                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
13026                                 ip1dbg((
13027                                     "ip_output_options: bad option offset\n"));
13028                                 code = (char *)&opt[IPOPT_OFFSET] -
13029                                     (char *)ipha;
13030                                 goto param_prob;
13031                         }
13032                         switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
13033                         case IPOPT_TS_TSONLY:
13034                                 off = IPOPT_TS_TIMELEN;
13035                                 break;
13036                         case IPOPT_TS_TSANDADDR:
13037                         case IPOPT_TS_PRESPEC:
13038                         case IPOPT_TS_PRESPEC_RFC791:
13039                                 off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
13040                                 break;
13041                         default:
13042                                 code = (char *)&opt[IPOPT_POS_OV_FLG] -
13043                                     (char *)ipha;
13044                                 goto param_prob;
13045                         }
13046                         if (opt[IPOPT_OFFSET] - 1 + off > optlen &&
13047                             (opt[IPOPT_POS_OV_FLG] & 0xF0) == 0xF0) {
13048                                 /*
13049                                  * No room and the overflow counter is 15
13050                                  * already.
13051                                  */
13052                                 goto param_prob;
13053                         }
13054                         break;
13055                 }
13056         }
13057 
13058         if ((opts.ipoptp_flags & IPOPTP_ERROR) == 0)
13059                 return (0);
13060 
13061         ip1dbg(("ip_output_options: error processing IP options."));
13062         code = (char *)&opt[IPOPT_OFFSET] - (char *)ipha;
13063 
13064 param_prob:
13065         bzero(&iras, sizeof (iras));
13066         iras.ira_ill = iras.ira_rill = ill;
13067         iras.ira_ruifindex = ill->ill_phyint->phyint_ifindex;
13068         iras.ira_rifindex = iras.ira_ruifindex;
13069         iras.ira_flags = IRAF_IS_IPV4;
13070 
13071         ip_drop_output("ip_output_options", mp, ill);
13072         icmp_param_problem(mp, (uint8_t)code, &iras);
13073         ASSERT(!(iras.ira_flags & IRAF_IPSEC_SECURE));
13074         return (-1);
13075 
13076 bad_src_route:
13077         bzero(&iras, sizeof (iras));
13078         iras.ira_ill = iras.ira_rill = ill;
13079         iras.ira_ruifindex = ill->ill_phyint->phyint_ifindex;
13080         iras.ira_rifindex = iras.ira_ruifindex;
13081         iras.ira_flags = IRAF_IS_IPV4;
13082 
13083         ip_drop_input("ICMP_SOURCE_ROUTE_FAILED", mp, ill);
13084         icmp_unreachable(mp, ICMP_SOURCE_ROUTE_FAILED, &iras);
13085         ASSERT(!(iras.ira_flags & IRAF_IPSEC_SECURE));
13086         return (-1);
13087 }
13088 
13089 /*
13090  * The maximum value of conn_drain_list_cnt is CONN_MAXDRAINCNT.
13091  * conn_drain_list_cnt can be changed by setting conn_drain_nthreads
13092  * thru /etc/system.
13093  */
13094 #define CONN_MAXDRAINCNT        64
13095 
13096 static void
13097 conn_drain_init(ip_stack_t *ipst)
13098 {
13099         int i, j;
13100         idl_tx_list_t *itl_tx;
13101 
13102         ipst->ips_conn_drain_list_cnt = conn_drain_nthreads;
13103 
13104         if ((ipst->ips_conn_drain_list_cnt == 0) ||
13105             (ipst->ips_conn_drain_list_cnt > CONN_MAXDRAINCNT)) {
13106                 /*
13107                  * Default value of the number of drainers is the
13108                  * number of cpus, subject to maximum of 8 drainers.
13109                  */
13110                 if (boot_max_ncpus != -1)
13111                         ipst->ips_conn_drain_list_cnt = MIN(boot_max_ncpus, 8);
13112                 else
13113                         ipst->ips_conn_drain_list_cnt = MIN(max_ncpus, 8);
13114         }
13115 
13116         ipst->ips_idl_tx_list =
13117             kmem_zalloc(TX_FANOUT_SIZE * sizeof (idl_tx_list_t), KM_SLEEP);
13118         for (i = 0; i < TX_FANOUT_SIZE; i++) {
13119                 itl_tx =  &ipst->ips_idl_tx_list[i];
13120                 itl_tx->txl_drain_list =
13121                     kmem_zalloc(ipst->ips_conn_drain_list_cnt *
13122                     sizeof (idl_t), KM_SLEEP);
13123                 mutex_init(&itl_tx->txl_lock, NULL, MUTEX_DEFAULT, NULL);
13124                 for (j = 0; j < ipst->ips_conn_drain_list_cnt; j++) {
13125                         mutex_init(&itl_tx->txl_drain_list[j].idl_lock, NULL,
13126                             MUTEX_DEFAULT, NULL);
13127                         itl_tx->txl_drain_list[j].idl_itl = itl_tx;
13128                 }
13129         }
13130 }
13131 
13132 static void
13133 conn_drain_fini(ip_stack_t *ipst)
13134 {
13135         int i;
13136         idl_tx_list_t *itl_tx;
13137 
13138         for (i = 0; i < TX_FANOUT_SIZE; i++) {
13139                 itl_tx =  &ipst->ips_idl_tx_list[i];
13140                 kmem_free(itl_tx->txl_drain_list,
13141                     ipst->ips_conn_drain_list_cnt * sizeof (idl_t));
13142         }
13143         kmem_free(ipst->ips_idl_tx_list,
13144             TX_FANOUT_SIZE * sizeof (idl_tx_list_t));
13145         ipst->ips_idl_tx_list = NULL;
13146 }
13147 
13148 /*
13149  * Flow control has blocked us from proceeding.  Insert the given conn in one
13150  * of the conn drain lists.  When flow control is unblocked, either ip_wsrv()
13151  * (STREAMS) or ill_flow_enable() (direct) will be called back, which in turn
13152  * will call conn_walk_drain().  See the flow control notes at the top of this
13153  * file for more details.
13154  */
13155 void
13156 conn_drain_insert(conn_t *connp, idl_tx_list_t *tx_list)
13157 {
13158         idl_t   *idl = tx_list->txl_drain_list;
13159         uint_t  index;
13160         ip_stack_t      *ipst = connp->conn_netstack->netstack_ip;
13161 
13162         mutex_enter(&connp->conn_lock);
13163         if (connp->conn_state_flags & CONN_CLOSING) {
13164                 /*
13165                  * The conn is closing as a result of which CONN_CLOSING
13166                  * is set. Return.
13167                  */
13168                 mutex_exit(&connp->conn_lock);
13169                 return;
13170         } else if (connp->conn_idl == NULL) {
13171                 /*
13172                  * Assign the next drain list round robin. We dont' use
13173                  * a lock, and thus it may not be strictly round robin.
13174                  * Atomicity of load/stores is enough to make sure that
13175                  * conn_drain_list_index is always within bounds.
13176                  */
13177                 index = tx_list->txl_drain_index;
13178                 ASSERT(index < ipst->ips_conn_drain_list_cnt);
13179                 connp->conn_idl = &tx_list->txl_drain_list[index];
13180                 index++;
13181                 if (index == ipst->ips_conn_drain_list_cnt)
13182                         index = 0;
13183                 tx_list->txl_drain_index = index;
13184         } else {
13185                 ASSERT(connp->conn_idl->idl_itl == tx_list);
13186         }
13187         mutex_exit(&connp->conn_lock);
13188 
13189         idl = connp->conn_idl;
13190         mutex_enter(&idl->idl_lock);
13191         if ((connp->conn_drain_prev != NULL) ||
13192             (connp->conn_state_flags & CONN_CLOSING)) {
13193                 /*
13194                  * The conn is either already in the drain list or closing.
13195                  * (We needed to check for CONN_CLOSING again since close can
13196                  * sneak in between dropping conn_lock and acquiring idl_lock.)
13197                  */
13198                 mutex_exit(&idl->idl_lock);
13199                 return;
13200         }
13201 
13202         /*
13203          * The conn is not in the drain list. Insert it at the
13204          * tail of the drain list. The drain list is circular
13205          * and doubly linked. idl_conn points to the 1st element
13206          * in the list.
13207          */
13208         if (idl->idl_conn == NULL) {
13209                 idl->idl_conn = connp;
13210                 connp->conn_drain_next = connp;
13211                 connp->conn_drain_prev = connp;
13212         } else {
13213                 conn_t *head = idl->idl_conn;
13214 
13215                 connp->conn_drain_next = head;
13216                 connp->conn_drain_prev = head->conn_drain_prev;
13217                 head->conn_drain_prev->conn_drain_next = connp;
13218                 head->conn_drain_prev = connp;
13219         }
13220         /*
13221          * For non streams based sockets assert flow control.
13222          */
13223         conn_setqfull(connp, NULL);
13224         mutex_exit(&idl->idl_lock);
13225 }
13226 
13227 static void
13228 conn_drain_remove(conn_t *connp)
13229 {
13230         idl_t *idl = connp->conn_idl;
13231 
13232         if (idl != NULL) {
13233                 /*
13234                  * Remove ourself from the drain list.
13235                  */
13236                 if (connp->conn_drain_next == connp) {
13237                         /* Singleton in the list */
13238                         ASSERT(connp->conn_drain_prev == connp);
13239                         idl->idl_conn = NULL;
13240                 } else {
13241                         connp->conn_drain_prev->conn_drain_next =
13242                             connp->conn_drain_next;
13243                         connp->conn_drain_next->conn_drain_prev =
13244                             connp->conn_drain_prev;
13245                         if (idl->idl_conn == connp)
13246                                 idl->idl_conn = connp->conn_drain_next;
13247                 }
13248 
13249                 /*
13250                  * NOTE: because conn_idl is associated with a specific drain
13251                  * list which in turn is tied to the index the TX ring
13252                  * (txl_cookie) hashes to, and because the TX ring can change
13253                  * over the lifetime of the conn_t, we must clear conn_idl so
13254                  * a subsequent conn_drain_insert() will set conn_idl again
13255                  * based on the latest txl_cookie.
13256                  */
13257                 connp->conn_idl = NULL;
13258         }
13259         connp->conn_drain_next = NULL;
13260         connp->conn_drain_prev = NULL;
13261 
13262         conn_clrqfull(connp, NULL);
13263         /*
13264          * For streams based sockets open up flow control.
13265          */
13266         if (!IPCL_IS_NONSTR(connp))
13267                 enableok(connp->conn_wq);
13268 }
13269 
13270 /*
13271  * This conn is closing, and we are called from ip_close. OR
13272  * this conn is draining because flow-control on the ill has been relieved.
13273  *
13274  * We must also need to remove conn's on this idl from the list, and also
13275  * inform the sockfs upcalls about the change in flow-control.
13276  */
13277 static void
13278 conn_drain(conn_t *connp, boolean_t closing)
13279 {
13280         idl_t *idl;
13281         conn_t *next_connp;
13282 
13283         /*
13284          * connp->conn_idl is stable at this point, and no lock is needed
13285          * to check it. If we are called from ip_close, close has already
13286          * set CONN_CLOSING, thus freezing the value of conn_idl, and
13287          * called us only because conn_idl is non-null. If we are called thru
13288          * service, conn_idl could be null, but it cannot change because
13289          * service is single-threaded per queue, and there cannot be another
13290          * instance of service trying to call conn_drain_insert on this conn
13291          * now.
13292          */
13293         ASSERT(!closing || connp == NULL || connp->conn_idl != NULL);
13294 
13295         /*
13296          * If the conn doesn't exist or is not on a drain list, bail.
13297          */
13298         if (connp == NULL || connp->conn_idl == NULL ||
13299             connp->conn_drain_prev == NULL) {
13300                 return;
13301         }
13302 
13303         idl = connp->conn_idl;
13304         ASSERT(MUTEX_HELD(&idl->idl_lock));
13305 
13306         if (!closing) {
13307                 next_connp = connp->conn_drain_next;
13308                 while (next_connp != connp) {
13309                         conn_t *delconnp = next_connp;
13310 
13311                         next_connp = next_connp->conn_drain_next;
13312                         conn_drain_remove(delconnp);
13313                 }
13314                 ASSERT(connp->conn_drain_next == idl->idl_conn);
13315         }
13316         conn_drain_remove(connp);
13317 }
13318 
13319 /*
13320  * Write service routine. Shared perimeter entry point.
13321  * The device queue's messages has fallen below the low water mark and STREAMS
13322  * has backenabled the ill_wq. Send sockfs notification about flow-control on
13323  * each waiting conn.
13324  */
13325 void
13326 ip_wsrv(queue_t *q)
13327 {
13328         ill_t   *ill;
13329 
13330         ill = (ill_t *)q->q_ptr;
13331         if (ill->ill_state_flags == 0) {
13332                 ip_stack_t *ipst = ill->ill_ipst;
13333 
13334                 /*
13335                  * The device flow control has opened up.
13336                  * Walk through conn drain lists and qenable the
13337                  * first conn in each list. This makes sense only
13338                  * if the stream is fully plumbed and setup.
13339                  * Hence the ill_state_flags check above.
13340                  */
13341                 ip1dbg(("ip_wsrv: walking\n"));
13342                 conn_walk_drain(ipst, &ipst->ips_idl_tx_list[0]);
13343                 enableok(ill->ill_wq);
13344         }
13345 }
13346 
13347 /*
13348  * Callback to disable flow control in IP.
13349  *
13350  * This is a mac client callback added when the DLD_CAPAB_DIRECT capability
13351  * is enabled.
13352  *
13353  * When MAC_TX() is not able to send any more packets, dld sets its queue
13354  * to QFULL and enable the STREAMS flow control. Later, when the underlying
13355  * driver is able to continue to send packets, it calls mac_tx_(ring_)update()
13356  * function and wakes up corresponding mac worker threads, which in turn
13357  * calls this callback function, and disables flow control.
13358  */
13359 void
13360 ill_flow_enable(void *arg, ip_mac_tx_cookie_t cookie)
13361 {
13362         ill_t *ill = (ill_t *)arg;
13363         ip_stack_t *ipst = ill->ill_ipst;
13364         idl_tx_list_t *idl_txl;
13365 
13366         idl_txl = &ipst->ips_idl_tx_list[IDLHASHINDEX(cookie)];
13367         mutex_enter(&idl_txl->txl_lock);
13368         /* add code to to set a flag to indicate idl_txl is enabled */
13369         conn_walk_drain(ipst, idl_txl);
13370         mutex_exit(&idl_txl->txl_lock);
13371 }
13372 
13373 /*
13374  * Flow control has been relieved and STREAMS has backenabled us; drain
13375  * all the conn lists on `tx_list'.
13376  */
13377 static void
13378 conn_walk_drain(ip_stack_t *ipst, idl_tx_list_t *tx_list)
13379 {
13380         int i;
13381         idl_t *idl;
13382 
13383         IP_STAT(ipst, ip_conn_walk_drain);
13384 
13385         for (i = 0; i < ipst->ips_conn_drain_list_cnt; i++) {
13386                 idl = &tx_list->txl_drain_list[i];
13387                 mutex_enter(&idl->idl_lock);
13388                 conn_drain(idl->idl_conn, B_FALSE);
13389                 mutex_exit(&idl->idl_lock);
13390         }
13391 }
13392 
13393 /*
13394  * Determine if the ill and multicast aspects of that packets
13395  * "matches" the conn.
13396  */
13397 boolean_t
13398 conn_wantpacket(conn_t *connp, ip_recv_attr_t *ira, ipha_t *ipha)
13399 {
13400         ill_t           *ill = ira->ira_rill;
13401         zoneid_t        zoneid = ira->ira_zoneid;
13402         uint_t          in_ifindex;
13403         ipaddr_t        dst, src;
13404 
13405         dst = ipha->ipha_dst;
13406         src = ipha->ipha_src;
13407 
13408         /*
13409          * conn_incoming_ifindex is set by IP_BOUND_IF which limits
13410          * unicast, broadcast and multicast reception to
13411          * conn_incoming_ifindex.
13412          * conn_wantpacket is called for unicast, broadcast and
13413          * multicast packets.
13414          */
13415         in_ifindex = connp->conn_incoming_ifindex;
13416 
13417         /* mpathd can bind to the under IPMP interface, which we allow */
13418         if (in_ifindex != 0 && in_ifindex != ill->ill_phyint->phyint_ifindex) {
13419                 if (!IS_UNDER_IPMP(ill))
13420                         return (B_FALSE);
13421 
13422                 if (in_ifindex != ipmp_ill_get_ipmp_ifindex(ill))
13423                         return (B_FALSE);
13424         }
13425 
13426         if (!IPCL_ZONE_MATCH(connp, zoneid))
13427                 return (B_FALSE);
13428 
13429         if (!(ira->ira_flags & IRAF_MULTICAST))
13430                 return (B_TRUE);
13431 
13432         if (connp->conn_multi_router) {
13433                 /* multicast packet and multicast router socket: send up */
13434                 return (B_TRUE);
13435         }
13436 
13437         if (ipha->ipha_protocol == IPPROTO_PIM ||
13438             ipha->ipha_protocol == IPPROTO_RSVP)
13439                 return (B_TRUE);
13440 
13441         return (conn_hasmembers_ill_withsrc_v4(connp, dst, src, ira->ira_ill));
13442 }
13443 
13444 void
13445 conn_setqfull(conn_t *connp, boolean_t *flow_stopped)
13446 {
13447         if (IPCL_IS_NONSTR(connp)) {
13448                 (*connp->conn_upcalls->su_txq_full)
13449                     (connp->conn_upper_handle, B_TRUE);
13450                 if (flow_stopped != NULL)
13451                         *flow_stopped = B_TRUE;
13452         } else {
13453                 queue_t *q = connp->conn_wq;
13454 
13455                 ASSERT(q != NULL);
13456                 if (!(q->q_flag & QFULL)) {
13457                         mutex_enter(QLOCK(q));
13458                         if (!(q->q_flag & QFULL)) {
13459                                 /* still need to set QFULL */
13460                                 q->q_flag |= QFULL;
13461                                 /* set flow_stopped to true under QLOCK */
13462                                 if (flow_stopped != NULL)
13463                                         *flow_stopped = B_TRUE;
13464                                 mutex_exit(QLOCK(q));
13465                         } else {
13466                                 /* flow_stopped is left unchanged */
13467                                 mutex_exit(QLOCK(q));
13468                         }
13469                 }
13470         }
13471 }
13472 
13473 void
13474 conn_clrqfull(conn_t *connp, boolean_t *flow_stopped)
13475 {
13476         if (IPCL_IS_NONSTR(connp)) {
13477                 (*connp->conn_upcalls->su_txq_full)
13478                     (connp->conn_upper_handle, B_FALSE);
13479                 if (flow_stopped != NULL)
13480                         *flow_stopped = B_FALSE;
13481         } else {
13482                 queue_t *q = connp->conn_wq;
13483 
13484                 ASSERT(q != NULL);
13485                 if (q->q_flag & QFULL) {
13486                         mutex_enter(QLOCK(q));
13487                         if (q->q_flag & QFULL) {
13488                                 q->q_flag &= ~QFULL;
13489                                 /* set flow_stopped to false under QLOCK */
13490                                 if (flow_stopped != NULL)
13491                                         *flow_stopped = B_FALSE;
13492                                 mutex_exit(QLOCK(q));
13493                                 if (q->q_flag & QWANTW)
13494                                         qbackenable(q, 0);
13495                         } else {
13496                                 /* flow_stopped is left unchanged */
13497                                 mutex_exit(QLOCK(q));
13498                         }
13499                 }
13500         }
13501 
13502         mutex_enter(&connp->conn_lock);
13503         connp->conn_blocked = B_FALSE;
13504         mutex_exit(&connp->conn_lock);
13505 }
13506 
13507 /*
13508  * Return the length in bytes of the IPv4 headers (base header, label, and
13509  * other IP options) that will be needed based on the
13510  * ip_pkt_t structure passed by the caller.
13511  *
13512  * The returned length does not include the length of the upper level
13513  * protocol (ULP) header.
13514  * The caller needs to check that the length doesn't exceed the max for IPv4.
13515  */
13516 int
13517 ip_total_hdrs_len_v4(const ip_pkt_t *ipp)
13518 {
13519         int len;
13520 
13521         len = IP_SIMPLE_HDR_LENGTH;
13522         if (ipp->ipp_fields & IPPF_LABEL_V4) {
13523                 ASSERT(ipp->ipp_label_len_v4 != 0);
13524                 /* We need to round up here */
13525                 len += (ipp->ipp_label_len_v4 + 3) & ~3;
13526         }
13527 
13528         if (ipp->ipp_fields & IPPF_IPV4_OPTIONS) {
13529                 ASSERT(ipp->ipp_ipv4_options_len != 0);
13530                 ASSERT((ipp->ipp_ipv4_options_len & 3) == 0);
13531                 len += ipp->ipp_ipv4_options_len;
13532         }
13533         return (len);
13534 }
13535 
13536 /*
13537  * All-purpose routine to build an IPv4 header with options based
13538  * on the abstract ip_pkt_t.
13539  *
13540  * The caller has to set the source and destination address as well as
13541  * ipha_length. The caller has to massage any source route and compensate
13542  * for the ULP pseudo-header checksum due to the source route.
13543  */
13544 void
13545 ip_build_hdrs_v4(uchar_t *buf, uint_t buf_len, const ip_pkt_t *ipp,
13546     uint8_t protocol)
13547 {
13548         ipha_t  *ipha = (ipha_t *)buf;
13549         uint8_t *cp;
13550 
13551         /* Initialize IPv4 header */
13552         ipha->ipha_type_of_service = ipp->ipp_type_of_service;
13553         ipha->ipha_length = 0;       /* Caller will set later */
13554         ipha->ipha_ident = 0;
13555         ipha->ipha_fragment_offset_and_flags = 0;
13556         ipha->ipha_ttl = ipp->ipp_unicast_hops;
13557         ipha->ipha_protocol = protocol;
13558         ipha->ipha_hdr_checksum = 0;
13559 
13560         if ((ipp->ipp_fields & IPPF_ADDR) &&
13561             IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr))
13562                 ipha->ipha_src = ipp->ipp_addr_v4;
13563 
13564         cp = (uint8_t *)&ipha[1];
13565         if (ipp->ipp_fields & IPPF_LABEL_V4) {
13566                 ASSERT(ipp->ipp_label_len_v4 != 0);
13567                 bcopy(ipp->ipp_label_v4, cp, ipp->ipp_label_len_v4);
13568                 cp += ipp->ipp_label_len_v4;
13569                 /* We need to round up here */
13570                 while ((uintptr_t)cp & 0x3) {
13571                         *cp++ = IPOPT_NOP;
13572                 }
13573         }
13574 
13575         if (ipp->ipp_fields & IPPF_IPV4_OPTIONS) {
13576                 ASSERT(ipp->ipp_ipv4_options_len != 0);
13577                 ASSERT((ipp->ipp_ipv4_options_len & 3) == 0);
13578                 bcopy(ipp->ipp_ipv4_options, cp, ipp->ipp_ipv4_options_len);
13579                 cp += ipp->ipp_ipv4_options_len;
13580         }
13581         ipha->ipha_version_and_hdr_length =
13582             (uint8_t)((IP_VERSION << 4) + buf_len / 4);
13583 
13584         ASSERT((int)(cp - buf) == buf_len);
13585 }
13586 
13587 /* Allocate the private structure */
13588 static int
13589 ip_priv_alloc(void **bufp)
13590 {
13591         void    *buf;
13592 
13593         if ((buf = kmem_alloc(sizeof (ip_priv_t), KM_NOSLEEP)) == NULL)
13594                 return (ENOMEM);
13595 
13596         *bufp = buf;
13597         return (0);
13598 }
13599 
13600 /* Function to delete the private structure */
13601 void
13602 ip_priv_free(void *buf)
13603 {
13604         ASSERT(buf != NULL);
13605         kmem_free(buf, sizeof (ip_priv_t));
13606 }
13607 
13608 /*
13609  * The entry point for IPPF processing.
13610  * If the classifier (IPGPC_CLASSIFY) is not loaded and configured, the
13611  * routine just returns.
13612  *
13613  * When called, ip_process generates an ipp_packet_t structure
13614  * which holds the state information for this packet and invokes the
13615  * the classifier (via ipp_packet_process). The classification, depending on
13616  * configured filters, results in a list of actions for this packet. Invoking
13617  * an action may cause the packet to be dropped, in which case we return NULL.
13618  * proc indicates the callout position for
13619  * this packet and ill is the interface this packet arrived on or will leave
13620  * on (inbound and outbound resp.).
13621  *
13622  * We do the processing on the rill (mapped to the upper if ipmp), but MIB
13623  * on the ill corrsponding to the destination IP address.
13624  */
13625 mblk_t *
13626 ip_process(ip_proc_t proc, mblk_t *mp, ill_t *rill, ill_t *ill)
13627 {
13628         ip_priv_t       *priv;
13629         ipp_action_id_t aid;
13630         int             rc = 0;
13631         ipp_packet_t    *pp;
13632 
13633         /* If the classifier is not loaded, return  */
13634         if ((aid = ipp_action_lookup(IPGPC_CLASSIFY)) == IPP_ACTION_INVAL) {
13635                 return (mp);
13636         }
13637 
13638         ASSERT(mp != NULL);
13639 
13640         /* Allocate the packet structure */
13641         rc = ipp_packet_alloc(&pp, "ip", aid);
13642         if (rc != 0)
13643                 goto drop;
13644 
13645         /* Allocate the private structure */
13646         rc = ip_priv_alloc((void **)&priv);
13647         if (rc != 0) {
13648                 ipp_packet_free(pp);
13649                 goto drop;
13650         }
13651         priv->proc = proc;
13652         priv->ill_index = ill_get_upper_ifindex(rill);
13653 
13654         ipp_packet_set_private(pp, priv, ip_priv_free);
13655         ipp_packet_set_data(pp, mp);
13656 
13657         /* Invoke the classifier */
13658         rc = ipp_packet_process(&pp);
13659         if (pp != NULL) {
13660                 mp = ipp_packet_get_data(pp);
13661                 ipp_packet_free(pp);
13662                 if (rc != 0)
13663                         goto drop;
13664                 return (mp);
13665         } else {
13666                 /* No mp to trace in ip_drop_input/ip_drop_output  */
13667                 mp = NULL;
13668         }
13669 drop:
13670         if (proc == IPP_LOCAL_IN || proc == IPP_FWD_IN) {
13671                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
13672                 ip_drop_input("ip_process", mp, ill);
13673         } else {
13674                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
13675                 ip_drop_output("ip_process", mp, ill);
13676         }
13677         freemsg(mp);
13678         return (NULL);
13679 }
13680 
13681 /*
13682  * Propagate a multicast group membership operation (add/drop) on
13683  * all the interfaces crossed by the related multirt routes.
13684  * The call is considered successful if the operation succeeds
13685  * on at least one interface.
13686  *
13687  * This assumes that a set of IRE_HOST/RTF_MULTIRT has been created for the
13688  * multicast addresses with the ire argument being the first one.
13689  * We walk the bucket to find all the of those.
13690  *
13691  * Common to IPv4 and IPv6.
13692  */
13693 static int
13694 ip_multirt_apply_membership(int (*fn)(conn_t *, boolean_t,
13695     const in6_addr_t *, ipaddr_t, uint_t, mcast_record_t, const in6_addr_t *),
13696     ire_t *ire, conn_t *connp, boolean_t checkonly, const in6_addr_t *v6group,
13697     mcast_record_t fmode, const in6_addr_t *v6src)
13698 {
13699         ire_t           *ire_gw;
13700         irb_t           *irb;
13701         int             ifindex;
13702         int             error = 0;
13703         int             result;
13704         ip_stack_t      *ipst = ire->ire_ipst;
13705         ipaddr_t        group;
13706         boolean_t       isv6;
13707         int             match_flags;
13708 
13709         if (IN6_IS_ADDR_V4MAPPED(v6group)) {
13710                 IN6_V4MAPPED_TO_IPADDR(v6group, group);
13711                 isv6 = B_FALSE;
13712         } else {
13713                 isv6 = B_TRUE;
13714         }
13715 
13716         irb = ire->ire_bucket;
13717         ASSERT(irb != NULL);
13718 
13719         result = 0;
13720         irb_refhold(irb);
13721         for (; ire != NULL; ire = ire->ire_next) {
13722                 if ((ire->ire_flags & RTF_MULTIRT) == 0)
13723                         continue;
13724 
13725                 /* We handle -ifp routes by matching on the ill if set */
13726                 match_flags = MATCH_IRE_TYPE;
13727                 if (ire->ire_ill != NULL)
13728                         match_flags |= MATCH_IRE_ILL;
13729 
13730                 if (isv6) {
13731                         if (!IN6_ARE_ADDR_EQUAL(&ire->ire_addr_v6, v6group))
13732                                 continue;
13733 
13734                         ire_gw = ire_ftable_lookup_v6(&ire->ire_gateway_addr_v6,
13735                             0, 0, IRE_INTERFACE, ire->ire_ill, ALL_ZONES, NULL,
13736                             match_flags, 0, ipst, NULL);
13737                 } else {
13738                         if (ire->ire_addr != group)
13739                                 continue;
13740 
13741                         ire_gw = ire_ftable_lookup_v4(ire->ire_gateway_addr,
13742                             0, 0, IRE_INTERFACE, ire->ire_ill, ALL_ZONES, NULL,
13743                             match_flags, 0, ipst, NULL);
13744                 }
13745                 /* No interface route exists for the gateway; skip this ire. */
13746                 if (ire_gw == NULL)
13747                         continue;
13748                 if (ire_gw->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
13749                         ire_refrele(ire_gw);
13750                         continue;
13751                 }
13752                 ASSERT(ire_gw->ire_ill != NULL);     /* IRE_INTERFACE */
13753                 ifindex = ire_gw->ire_ill->ill_phyint->phyint_ifindex;
13754 
13755                 /*
13756                  * The operation is considered a success if
13757                  * it succeeds at least once on any one interface.
13758                  */
13759                 error = fn(connp, checkonly, v6group, INADDR_ANY, ifindex,
13760                     fmode, v6src);
13761                 if (error == 0)
13762                         result = CGTP_MCAST_SUCCESS;
13763 
13764                 ire_refrele(ire_gw);
13765         }
13766         irb_refrele(irb);
13767         /*
13768          * Consider the call as successful if we succeeded on at least
13769          * one interface. Otherwise, return the last encountered error.
13770          */
13771         return (result == CGTP_MCAST_SUCCESS ? 0 : error);
13772 }
13773 
13774 /*
13775  * Return the expected CGTP hooks version number.
13776  */
13777 int
13778 ip_cgtp_filter_supported(void)
13779 {
13780         return (ip_cgtp_filter_rev);
13781 }
13782 
13783 /*
13784  * CGTP hooks can be registered by invoking this function.
13785  * Checks that the version number matches.
13786  */
13787 int
13788 ip_cgtp_filter_register(netstackid_t stackid, cgtp_filter_ops_t *ops)
13789 {
13790         netstack_t *ns;
13791         ip_stack_t *ipst;
13792 
13793         if (ops->cfo_filter_rev != CGTP_FILTER_REV)
13794                 return (ENOTSUP);
13795 
13796         ns = netstack_find_by_stackid(stackid);
13797         if (ns == NULL)
13798                 return (EINVAL);
13799         ipst = ns->netstack_ip;
13800         ASSERT(ipst != NULL);
13801 
13802         if (ipst->ips_ip_cgtp_filter_ops != NULL) {
13803                 netstack_rele(ns);
13804                 return (EALREADY);
13805         }
13806 
13807         ipst->ips_ip_cgtp_filter_ops = ops;
13808 
13809         ill_set_inputfn_all(ipst);
13810 
13811         netstack_rele(ns);
13812         return (0);
13813 }
13814 
13815 /*
13816  * CGTP hooks can be unregistered by invoking this function.
13817  * Returns ENXIO if there was no registration.
13818  * Returns EBUSY if the ndd variable has not been turned off.
13819  */
13820 int
13821 ip_cgtp_filter_unregister(netstackid_t stackid)
13822 {
13823         netstack_t *ns;
13824         ip_stack_t *ipst;
13825 
13826         ns = netstack_find_by_stackid(stackid);
13827         if (ns == NULL)
13828                 return (EINVAL);
13829         ipst = ns->netstack_ip;
13830         ASSERT(ipst != NULL);
13831 
13832         if (ipst->ips_ip_cgtp_filter) {
13833                 netstack_rele(ns);
13834                 return (EBUSY);
13835         }
13836 
13837         if (ipst->ips_ip_cgtp_filter_ops == NULL) {
13838                 netstack_rele(ns);
13839                 return (ENXIO);
13840         }
13841         ipst->ips_ip_cgtp_filter_ops = NULL;
13842 
13843         ill_set_inputfn_all(ipst);
13844 
13845         netstack_rele(ns);
13846         return (0);
13847 }
13848 
13849 /*
13850  * Check whether there is a CGTP filter registration.
13851  * Returns non-zero if there is a registration, otherwise returns zero.
13852  * Note: returns zero if bad stackid.
13853  */
13854 int
13855 ip_cgtp_filter_is_registered(netstackid_t stackid)
13856 {
13857         netstack_t *ns;
13858         ip_stack_t *ipst;
13859         int ret;
13860 
13861         ns = netstack_find_by_stackid(stackid);
13862         if (ns == NULL)
13863                 return (0);
13864         ipst = ns->netstack_ip;
13865         ASSERT(ipst != NULL);
13866 
13867         if (ipst->ips_ip_cgtp_filter_ops != NULL)
13868                 ret = 1;
13869         else
13870                 ret = 0;
13871 
13872         netstack_rele(ns);
13873         return (ret);
13874 }
13875 
13876 static int
13877 ip_squeue_switch(int val)
13878 {
13879         int rval;
13880 
13881         switch (val) {
13882         case IP_SQUEUE_ENTER_NODRAIN:
13883                 rval = SQ_NODRAIN;
13884                 break;
13885         case IP_SQUEUE_ENTER:
13886                 rval = SQ_PROCESS;
13887                 break;
13888         case IP_SQUEUE_FILL:
13889         default:
13890                 rval = SQ_FILL;
13891                 break;
13892         }
13893         return (rval);
13894 }
13895 
13896 static void *
13897 ip_kstat2_init(netstackid_t stackid, ip_stat_t *ip_statisticsp)
13898 {
13899         kstat_t *ksp;
13900 
13901         ip_stat_t template = {
13902                 { "ip_udp_fannorm",             KSTAT_DATA_UINT64 },
13903                 { "ip_udp_fanmb",               KSTAT_DATA_UINT64 },
13904                 { "ip_recv_pullup",             KSTAT_DATA_UINT64 },
13905                 { "ip_db_ref",                  KSTAT_DATA_UINT64 },
13906                 { "ip_notaligned",              KSTAT_DATA_UINT64 },
13907                 { "ip_multimblk",               KSTAT_DATA_UINT64 },
13908                 { "ip_opt",                     KSTAT_DATA_UINT64 },
13909                 { "ipsec_proto_ahesp",          KSTAT_DATA_UINT64 },
13910                 { "ip_conn_flputbq",            KSTAT_DATA_UINT64 },
13911                 { "ip_conn_walk_drain",         KSTAT_DATA_UINT64 },
13912                 { "ip_out_sw_cksum",            KSTAT_DATA_UINT64 },
13913                 { "ip_out_sw_cksum_bytes",      KSTAT_DATA_UINT64 },
13914                 { "ip_in_sw_cksum",             KSTAT_DATA_UINT64 },
13915                 { "ip_ire_reclaim_calls",       KSTAT_DATA_UINT64 },
13916                 { "ip_ire_reclaim_deleted",     KSTAT_DATA_UINT64 },
13917                 { "ip_nce_reclaim_calls",       KSTAT_DATA_UINT64 },
13918                 { "ip_nce_reclaim_deleted",     KSTAT_DATA_UINT64 },
13919                 { "ip_dce_reclaim_calls",       KSTAT_DATA_UINT64 },
13920                 { "ip_dce_reclaim_deleted",     KSTAT_DATA_UINT64 },
13921                 { "ip_tcp_in_full_hw_cksum_err",        KSTAT_DATA_UINT64 },
13922                 { "ip_tcp_in_part_hw_cksum_err",        KSTAT_DATA_UINT64 },
13923                 { "ip_tcp_in_sw_cksum_err",             KSTAT_DATA_UINT64 },
13924                 { "ip_udp_in_full_hw_cksum_err",        KSTAT_DATA_UINT64 },
13925                 { "ip_udp_in_part_hw_cksum_err",        KSTAT_DATA_UINT64 },
13926                 { "ip_udp_in_sw_cksum_err",     KSTAT_DATA_UINT64 },
13927                 { "conn_in_recvdstaddr",        KSTAT_DATA_UINT64 },
13928                 { "conn_in_recvopts",           KSTAT_DATA_UINT64 },
13929                 { "conn_in_recvif",             KSTAT_DATA_UINT64 },
13930                 { "conn_in_recvslla",           KSTAT_DATA_UINT64 },
13931                 { "conn_in_recvucred",          KSTAT_DATA_UINT64 },
13932                 { "conn_in_recvttl",            KSTAT_DATA_UINT64 },
13933                 { "conn_in_recvhopopts",        KSTAT_DATA_UINT64 },
13934                 { "conn_in_recvhoplimit",       KSTAT_DATA_UINT64 },
13935                 { "conn_in_recvdstopts",        KSTAT_DATA_UINT64 },
13936                 { "conn_in_recvrthdrdstopts",   KSTAT_DATA_UINT64 },
13937                 { "conn_in_recvrthdr",          KSTAT_DATA_UINT64 },
13938                 { "conn_in_recvpktinfo",        KSTAT_DATA_UINT64 },
13939                 { "conn_in_recvtclass",         KSTAT_DATA_UINT64 },
13940                 { "conn_in_timestamp",          KSTAT_DATA_UINT64 },
13941         };
13942 
13943         ksp = kstat_create_netstack("ip", 0, "ipstat", "net",
13944             KSTAT_TYPE_NAMED, sizeof (template) / sizeof (kstat_named_t),
13945             KSTAT_FLAG_VIRTUAL, stackid);
13946 
13947         if (ksp == NULL)
13948                 return (NULL);
13949 
13950         bcopy(&template, ip_statisticsp, sizeof (template));
13951         ksp->ks_data = (void *)ip_statisticsp;
13952         ksp->ks_private = (void *)(uintptr_t)stackid;
13953 
13954         kstat_install(ksp);
13955         return (ksp);
13956 }
13957 
13958 static void
13959 ip_kstat2_fini(netstackid_t stackid, kstat_t *ksp)
13960 {
13961         if (ksp != NULL) {
13962                 ASSERT(stackid == (netstackid_t)(uintptr_t)ksp->ks_private);
13963                 kstat_delete_netstack(ksp, stackid);
13964         }
13965 }
13966 
13967 static void *
13968 ip_kstat_init(netstackid_t stackid, ip_stack_t *ipst)
13969 {
13970         kstat_t *ksp;
13971 
13972         ip_named_kstat_t template = {
13973                 { "forwarding",         KSTAT_DATA_UINT32, 0 },
13974                 { "defaultTTL",         KSTAT_DATA_UINT32, 0 },
13975                 { "inReceives",         KSTAT_DATA_UINT64, 0 },
13976                 { "inHdrErrors",        KSTAT_DATA_UINT32, 0 },
13977                 { "inAddrErrors",       KSTAT_DATA_UINT32, 0 },
13978                 { "forwDatagrams",      KSTAT_DATA_UINT64, 0 },
13979                 { "inUnknownProtos",    KSTAT_DATA_UINT32, 0 },
13980                 { "inDiscards",         KSTAT_DATA_UINT32, 0 },
13981                 { "inDelivers",         KSTAT_DATA_UINT64, 0 },
13982                 { "outRequests",        KSTAT_DATA_UINT64, 0 },
13983                 { "outDiscards",        KSTAT_DATA_UINT32, 0 },
13984                 { "outNoRoutes",        KSTAT_DATA_UINT32, 0 },
13985                 { "reasmTimeout",       KSTAT_DATA_UINT32, 0 },
13986                 { "reasmReqds",         KSTAT_DATA_UINT32, 0 },
13987                 { "reasmOKs",           KSTAT_DATA_UINT32, 0 },
13988                 { "reasmFails",         KSTAT_DATA_UINT32, 0 },
13989                 { "fragOKs",            KSTAT_DATA_UINT32, 0 },
13990                 { "fragFails",          KSTAT_DATA_UINT32, 0 },
13991                 { "fragCreates",        KSTAT_DATA_UINT32, 0 },
13992                 { "addrEntrySize",      KSTAT_DATA_INT32, 0 },
13993                 { "routeEntrySize",     KSTAT_DATA_INT32, 0 },
13994                 { "netToMediaEntrySize",        KSTAT_DATA_INT32, 0 },
13995                 { "routingDiscards",    KSTAT_DATA_UINT32, 0 },
13996                 { "inErrs",             KSTAT_DATA_UINT32, 0 },
13997                 { "noPorts",            KSTAT_DATA_UINT32, 0 },
13998                 { "inCksumErrs",        KSTAT_DATA_UINT32, 0 },
13999                 { "reasmDuplicates",    KSTAT_DATA_UINT32, 0 },
14000                 { "reasmPartDups",      KSTAT_DATA_UINT32, 0 },
14001                 { "forwProhibits",      KSTAT_DATA_UINT32, 0 },
14002                 { "udpInCksumErrs",     KSTAT_DATA_UINT32, 0 },
14003                 { "udpInOverflows",     KSTAT_DATA_UINT32, 0 },
14004                 { "rawipInOverflows",   KSTAT_DATA_UINT32, 0 },
14005                 { "ipsecInSucceeded",   KSTAT_DATA_UINT32, 0 },
14006                 { "ipsecInFailed",      KSTAT_DATA_INT32, 0 },
14007                 { "memberEntrySize",    KSTAT_DATA_INT32, 0 },
14008                 { "inIPv6",             KSTAT_DATA_UINT32, 0 },
14009                 { "outIPv6",            KSTAT_DATA_UINT32, 0 },
14010                 { "outSwitchIPv6",      KSTAT_DATA_UINT32, 0 },
14011         };
14012 
14013         ksp = kstat_create_netstack("ip", 0, "ip", "mib2", KSTAT_TYPE_NAMED,
14014             NUM_OF_FIELDS(ip_named_kstat_t), 0, stackid);
14015         if (ksp == NULL || ksp->ks_data == NULL)
14016                 return (NULL);
14017 
14018         template.forwarding.value.ui32 = WE_ARE_FORWARDING(ipst) ? 1:2;
14019         template.defaultTTL.value.ui32 = (uint32_t)ipst->ips_ip_def_ttl;
14020         template.reasmTimeout.value.ui32 = ipst->ips_ip_reassembly_timeout;
14021         template.addrEntrySize.value.i32 = sizeof (mib2_ipAddrEntry_t);
14022         template.routeEntrySize.value.i32 = sizeof (mib2_ipRouteEntry_t);
14023 
14024         template.netToMediaEntrySize.value.i32 =
14025             sizeof (mib2_ipNetToMediaEntry_t);
14026 
14027         template.memberEntrySize.value.i32 = sizeof (ipv6_member_t);
14028 
14029         bcopy(&template, ksp->ks_data, sizeof (template));
14030         ksp->ks_update = ip_kstat_update;
14031         ksp->ks_private = (void *)(uintptr_t)stackid;
14032 
14033         kstat_install(ksp);
14034         return (ksp);
14035 }
14036 
14037 static void
14038 ip_kstat_fini(netstackid_t stackid, kstat_t *ksp)
14039 {
14040         if (ksp != NULL) {
14041                 ASSERT(stackid == (netstackid_t)(uintptr_t)ksp->ks_private);
14042                 kstat_delete_netstack(ksp, stackid);
14043         }
14044 }
14045 
14046 static int
14047 ip_kstat_update(kstat_t *kp, int rw)
14048 {
14049         ip_named_kstat_t *ipkp;
14050         mib2_ipIfStatsEntry_t ipmib;
14051         ill_walk_context_t ctx;
14052         ill_t *ill;
14053         netstackid_t    stackid = (zoneid_t)(uintptr_t)kp->ks_private;
14054         netstack_t      *ns;
14055         ip_stack_t      *ipst;
14056 
14057         if (kp == NULL || kp->ks_data == NULL)
14058                 return (EIO);
14059 
14060         if (rw == KSTAT_WRITE)
14061                 return (EACCES);
14062 
14063         ns = netstack_find_by_stackid(stackid);
14064         if (ns == NULL)
14065                 return (-1);
14066         ipst = ns->netstack_ip;
14067         if (ipst == NULL) {
14068                 netstack_rele(ns);
14069                 return (-1);
14070         }
14071         ipkp = (ip_named_kstat_t *)kp->ks_data;
14072 
14073         bcopy(&ipst->ips_ip_mib, &ipmib, sizeof (ipmib));
14074         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
14075         ill = ILL_START_WALK_V4(&ctx, ipst);
14076         for (; ill != NULL; ill = ill_next(&ctx, ill))
14077                 ip_mib2_add_ip_stats(&ipmib, ill->ill_ip_mib);
14078         rw_exit(&ipst->ips_ill_g_lock);
14079 
14080         ipkp->forwarding.value.ui32 =                ipmib.ipIfStatsForwarding;
14081         ipkp->defaultTTL.value.ui32 =                ipmib.ipIfStatsDefaultTTL;
14082         ipkp->inReceives.value.ui64 =                ipmib.ipIfStatsHCInReceives;
14083         ipkp->inHdrErrors.value.ui32 =               ipmib.ipIfStatsInHdrErrors;
14084         ipkp->inAddrErrors.value.ui32 =              ipmib.ipIfStatsInAddrErrors;
14085         ipkp->forwDatagrams.value.ui64 = ipmib.ipIfStatsHCOutForwDatagrams;
14086         ipkp->inUnknownProtos.value.ui32 =   ipmib.ipIfStatsInUnknownProtos;
14087         ipkp->inDiscards.value.ui32 =                ipmib.ipIfStatsInDiscards;
14088         ipkp->inDelivers.value.ui64 =                ipmib.ipIfStatsHCInDelivers;
14089         ipkp->outRequests.value.ui64 =               ipmib.ipIfStatsHCOutRequests;
14090         ipkp->outDiscards.value.ui32 =               ipmib.ipIfStatsOutDiscards;
14091         ipkp->outNoRoutes.value.ui32 =               ipmib.ipIfStatsOutNoRoutes;
14092         ipkp->reasmTimeout.value.ui32 =              ipst->ips_ip_reassembly_timeout;
14093         ipkp->reasmReqds.value.ui32 =                ipmib.ipIfStatsReasmReqds;
14094         ipkp->reasmOKs.value.ui32 =          ipmib.ipIfStatsReasmOKs;
14095         ipkp->reasmFails.value.ui32 =                ipmib.ipIfStatsReasmFails;
14096         ipkp->fragOKs.value.ui32 =           ipmib.ipIfStatsOutFragOKs;
14097         ipkp->fragFails.value.ui32 =         ipmib.ipIfStatsOutFragFails;
14098         ipkp->fragCreates.value.ui32 =               ipmib.ipIfStatsOutFragCreates;
14099 
14100         ipkp->routingDiscards.value.ui32 =   0;
14101         ipkp->inErrs.value.ui32 =            ipmib.tcpIfStatsInErrs;
14102         ipkp->noPorts.value.ui32 =           ipmib.udpIfStatsNoPorts;
14103         ipkp->inCksumErrs.value.ui32 =               ipmib.ipIfStatsInCksumErrs;
14104         ipkp->reasmDuplicates.value.ui32 =   ipmib.ipIfStatsReasmDuplicates;
14105         ipkp->reasmPartDups.value.ui32 =     ipmib.ipIfStatsReasmPartDups;
14106         ipkp->forwProhibits.value.ui32 =     ipmib.ipIfStatsForwProhibits;
14107         ipkp->udpInCksumErrs.value.ui32 =    ipmib.udpIfStatsInCksumErrs;
14108         ipkp->udpInOverflows.value.ui32 =    ipmib.udpIfStatsInOverflows;
14109         ipkp->rawipInOverflows.value.ui32 =  ipmib.rawipIfStatsInOverflows;
14110         ipkp->ipsecInSucceeded.value.ui32 =  ipmib.ipsecIfStatsInSucceeded;
14111         ipkp->ipsecInFailed.value.i32 =              ipmib.ipsecIfStatsInFailed;
14112 
14113         ipkp->inIPv6.value.ui32 =    ipmib.ipIfStatsInWrongIPVersion;
14114         ipkp->outIPv6.value.ui32 =   ipmib.ipIfStatsOutWrongIPVersion;
14115         ipkp->outSwitchIPv6.value.ui32 = ipmib.ipIfStatsOutSwitchIPVersion;
14116 
14117         netstack_rele(ns);
14118 
14119         return (0);
14120 }
14121 
14122 static void *
14123 icmp_kstat_init(netstackid_t stackid)
14124 {
14125         kstat_t *ksp;
14126 
14127         icmp_named_kstat_t template = {
14128                 { "inMsgs",             KSTAT_DATA_UINT32 },
14129                 { "inErrors",           KSTAT_DATA_UINT32 },
14130                 { "inDestUnreachs",     KSTAT_DATA_UINT32 },
14131                 { "inTimeExcds",        KSTAT_DATA_UINT32 },
14132                 { "inParmProbs",        KSTAT_DATA_UINT32 },
14133                 { "inSrcQuenchs",       KSTAT_DATA_UINT32 },
14134                 { "inRedirects",        KSTAT_DATA_UINT32 },
14135                 { "inEchos",            KSTAT_DATA_UINT32 },
14136                 { "inEchoReps",         KSTAT_DATA_UINT32 },
14137                 { "inTimestamps",       KSTAT_DATA_UINT32 },
14138                 { "inTimestampReps",    KSTAT_DATA_UINT32 },
14139                 { "inAddrMasks",        KSTAT_DATA_UINT32 },
14140                 { "inAddrMaskReps",     KSTAT_DATA_UINT32 },
14141                 { "outMsgs",            KSTAT_DATA_UINT32 },
14142                 { "outErrors",          KSTAT_DATA_UINT32 },
14143                 { "outDestUnreachs",    KSTAT_DATA_UINT32 },
14144                 { "outTimeExcds",       KSTAT_DATA_UINT32 },
14145                 { "outParmProbs",       KSTAT_DATA_UINT32 },
14146                 { "outSrcQuenchs",      KSTAT_DATA_UINT32 },
14147                 { "outRedirects",       KSTAT_DATA_UINT32 },
14148                 { "outEchos",           KSTAT_DATA_UINT32 },
14149                 { "outEchoReps",        KSTAT_DATA_UINT32 },
14150                 { "outTimestamps",      KSTAT_DATA_UINT32 },
14151                 { "outTimestampReps",   KSTAT_DATA_UINT32 },
14152                 { "outAddrMasks",       KSTAT_DATA_UINT32 },
14153                 { "outAddrMaskReps",    KSTAT_DATA_UINT32 },
14154                 { "inChksumErrs",       KSTAT_DATA_UINT32 },
14155                 { "inUnknowns",         KSTAT_DATA_UINT32 },
14156                 { "inFragNeeded",       KSTAT_DATA_UINT32 },
14157                 { "outFragNeeded",      KSTAT_DATA_UINT32 },
14158                 { "outDrops",           KSTAT_DATA_UINT32 },
14159                 { "inOverFlows",        KSTAT_DATA_UINT32 },
14160                 { "inBadRedirects",     KSTAT_DATA_UINT32 },
14161         };
14162 
14163         ksp = kstat_create_netstack("ip", 0, "icmp", "mib2", KSTAT_TYPE_NAMED,
14164             NUM_OF_FIELDS(icmp_named_kstat_t), 0, stackid);
14165         if (ksp == NULL || ksp->ks_data == NULL)
14166                 return (NULL);
14167 
14168         bcopy(&template, ksp->ks_data, sizeof (template));
14169 
14170         ksp->ks_update = icmp_kstat_update;
14171         ksp->ks_private = (void *)(uintptr_t)stackid;
14172 
14173         kstat_install(ksp);
14174         return (ksp);
14175 }
14176 
14177 static void
14178 icmp_kstat_fini(netstackid_t stackid, kstat_t *ksp)
14179 {
14180         if (ksp != NULL) {
14181                 ASSERT(stackid == (netstackid_t)(uintptr_t)ksp->ks_private);
14182                 kstat_delete_netstack(ksp, stackid);
14183         }
14184 }
14185 
14186 static int
14187 icmp_kstat_update(kstat_t *kp, int rw)
14188 {
14189         icmp_named_kstat_t *icmpkp;
14190         netstackid_t    stackid = (zoneid_t)(uintptr_t)kp->ks_private;
14191         netstack_t      *ns;
14192         ip_stack_t      *ipst;
14193 
14194         if ((kp == NULL) || (kp->ks_data == NULL))
14195                 return (EIO);
14196 
14197         if (rw == KSTAT_WRITE)
14198                 return (EACCES);
14199 
14200         ns = netstack_find_by_stackid(stackid);
14201         if (ns == NULL)
14202                 return (-1);
14203         ipst = ns->netstack_ip;
14204         if (ipst == NULL) {
14205                 netstack_rele(ns);
14206                 return (-1);
14207         }
14208         icmpkp = (icmp_named_kstat_t *)kp->ks_data;
14209 
14210         icmpkp->inMsgs.value.ui32 =      ipst->ips_icmp_mib.icmpInMsgs;
14211         icmpkp->inErrors.value.ui32 =            ipst->ips_icmp_mib.icmpInErrors;
14212         icmpkp->inDestUnreachs.value.ui32 =
14213             ipst->ips_icmp_mib.icmpInDestUnreachs;
14214         icmpkp->inTimeExcds.value.ui32 =    ipst->ips_icmp_mib.icmpInTimeExcds;
14215         icmpkp->inParmProbs.value.ui32 =    ipst->ips_icmp_mib.icmpInParmProbs;
14216         icmpkp->inSrcQuenchs.value.ui32 =   ipst->ips_icmp_mib.icmpInSrcQuenchs;
14217         icmpkp->inRedirects.value.ui32 =    ipst->ips_icmp_mib.icmpInRedirects;
14218         icmpkp->inEchos.value.ui32 =     ipst->ips_icmp_mib.icmpInEchos;
14219         icmpkp->inEchoReps.value.ui32 =          ipst->ips_icmp_mib.icmpInEchoReps;
14220         icmpkp->inTimestamps.value.ui32 =   ipst->ips_icmp_mib.icmpInTimestamps;
14221         icmpkp->inTimestampReps.value.ui32 =
14222             ipst->ips_icmp_mib.icmpInTimestampReps;
14223         icmpkp->inAddrMasks.value.ui32 =    ipst->ips_icmp_mib.icmpInAddrMasks;
14224         icmpkp->inAddrMaskReps.value.ui32 =
14225             ipst->ips_icmp_mib.icmpInAddrMaskReps;
14226         icmpkp->outMsgs.value.ui32 =     ipst->ips_icmp_mib.icmpOutMsgs;
14227         icmpkp->outErrors.value.ui32 =           ipst->ips_icmp_mib.icmpOutErrors;
14228         icmpkp->outDestUnreachs.value.ui32 =
14229             ipst->ips_icmp_mib.icmpOutDestUnreachs;
14230         icmpkp->outTimeExcds.value.ui32 =   ipst->ips_icmp_mib.icmpOutTimeExcds;
14231         icmpkp->outParmProbs.value.ui32 =   ipst->ips_icmp_mib.icmpOutParmProbs;
14232         icmpkp->outSrcQuenchs.value.ui32 =
14233             ipst->ips_icmp_mib.icmpOutSrcQuenchs;
14234         icmpkp->outRedirects.value.ui32 =   ipst->ips_icmp_mib.icmpOutRedirects;
14235         icmpkp->outEchos.value.ui32 =            ipst->ips_icmp_mib.icmpOutEchos;
14236         icmpkp->outEchoReps.value.ui32 =    ipst->ips_icmp_mib.icmpOutEchoReps;
14237         icmpkp->outTimestamps.value.ui32 =
14238             ipst->ips_icmp_mib.icmpOutTimestamps;
14239         icmpkp->outTimestampReps.value.ui32 =
14240             ipst->ips_icmp_mib.icmpOutTimestampReps;
14241         icmpkp->outAddrMasks.value.ui32 =
14242             ipst->ips_icmp_mib.icmpOutAddrMasks;
14243         icmpkp->outAddrMaskReps.value.ui32 =
14244             ipst->ips_icmp_mib.icmpOutAddrMaskReps;
14245         icmpkp->inCksumErrs.value.ui32 =    ipst->ips_icmp_mib.icmpInCksumErrs;
14246         icmpkp->inUnknowns.value.ui32 =          ipst->ips_icmp_mib.icmpInUnknowns;
14247         icmpkp->inFragNeeded.value.ui32 =   ipst->ips_icmp_mib.icmpInFragNeeded;
14248         icmpkp->outFragNeeded.value.ui32 =
14249             ipst->ips_icmp_mib.icmpOutFragNeeded;
14250         icmpkp->outDrops.value.ui32 =            ipst->ips_icmp_mib.icmpOutDrops;
14251         icmpkp->inOverflows.value.ui32 =    ipst->ips_icmp_mib.icmpInOverflows;
14252         icmpkp->inBadRedirects.value.ui32 =
14253             ipst->ips_icmp_mib.icmpInBadRedirects;
14254 
14255         netstack_rele(ns);
14256         return (0);
14257 }
14258 
14259 /*
14260  * This is the fanout function for raw socket opened for SCTP.  Note
14261  * that it is called after SCTP checks that there is no socket which
14262  * wants a packet.  Then before SCTP handles this out of the blue packet,
14263  * this function is called to see if there is any raw socket for SCTP.
14264  * If there is and it is bound to the correct address, the packet will
14265  * be sent to that socket.  Note that only one raw socket can be bound to
14266  * a port.  This is assured in ipcl_sctp_hash_insert();
14267  */
14268 void
14269 ip_fanout_sctp_raw(mblk_t *mp, ipha_t *ipha, ip6_t *ip6h, uint32_t ports,
14270     ip_recv_attr_t *ira)
14271 {
14272         conn_t          *connp;
14273         queue_t         *rq;
14274         boolean_t       secure;
14275         ill_t           *ill = ira->ira_ill;
14276         ip_stack_t      *ipst = ill->ill_ipst;
14277         ipsec_stack_t   *ipss = ipst->ips_netstack->netstack_ipsec;
14278         sctp_stack_t    *sctps = ipst->ips_netstack->netstack_sctp;
14279         iaflags_t       iraflags = ira->ira_flags;
14280         ill_t           *rill = ira->ira_rill;
14281 
14282         secure = iraflags & IRAF_IPSEC_SECURE;
14283 
14284         connp = ipcl_classify_raw(mp, IPPROTO_SCTP, ports, ipha, ip6h,
14285             ira, ipst);
14286         if (connp == NULL) {
14287                 /*
14288                  * Although raw sctp is not summed, OOB chunks must be.
14289                  * Drop the packet here if the sctp checksum failed.
14290                  */
14291                 if (iraflags & IRAF_SCTP_CSUM_ERR) {
14292                         SCTPS_BUMP_MIB(sctps, sctpChecksumError);
14293                         freemsg(mp);
14294                         return;
14295                 }
14296                 ira->ira_ill = ira->ira_rill = NULL;
14297                 sctp_ootb_input(mp, ira, ipst);
14298                 ira->ira_ill = ill;
14299                 ira->ira_rill = rill;
14300                 return;
14301         }
14302         rq = connp->conn_rq;
14303         if (IPCL_IS_NONSTR(connp) ? connp->conn_flow_cntrld : !canputnext(rq)) {
14304                 CONN_DEC_REF(connp);
14305                 BUMP_MIB(ill->ill_ip_mib, rawipIfStatsInOverflows);
14306                 freemsg(mp);
14307                 return;
14308         }
14309         if (((iraflags & IRAF_IS_IPV4) ?
14310             CONN_INBOUND_POLICY_PRESENT(connp, ipss) :
14311             CONN_INBOUND_POLICY_PRESENT_V6(connp, ipss)) ||
14312             secure) {
14313                 mp = ipsec_check_inbound_policy(mp, connp, ipha,
14314                     ip6h, ira);
14315                 if (mp == NULL) {
14316                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
14317                         /* Note that mp is NULL */
14318                         ip_drop_input("ipIfStatsInDiscards", mp, ill);
14319                         CONN_DEC_REF(connp);
14320                         return;
14321                 }
14322         }
14323 
14324         if (iraflags & IRAF_ICMP_ERROR) {
14325                 (connp->conn_recvicmp)(connp, mp, NULL, ira);
14326         } else {
14327                 ill_t *rill = ira->ira_rill;
14328 
14329                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCInDelivers);
14330                 /* This is the SOCK_RAW, IPPROTO_SCTP case. */
14331                 ira->ira_ill = ira->ira_rill = NULL;
14332                 (connp->conn_recv)(connp, mp, NULL, ira);
14333                 ira->ira_ill = ill;
14334                 ira->ira_rill = rill;
14335         }
14336         CONN_DEC_REF(connp);
14337 }
14338 
14339 /*
14340  * Free a packet that has the link-layer dl_unitdata_req_t or fast-path
14341  * header before the ip payload.
14342  */
14343 static void
14344 ip_xmit_flowctl_drop(ill_t *ill, mblk_t *mp, boolean_t is_fp_mp, int fp_mp_len)
14345 {
14346         int len = (mp->b_wptr - mp->b_rptr);
14347         mblk_t *ip_mp;
14348 
14349         BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
14350         if (is_fp_mp || len != fp_mp_len) {
14351                 if (len > fp_mp_len) {
14352                         /*
14353                          * fastpath header and ip header in the first mblk
14354                          */
14355                         mp->b_rptr += fp_mp_len;
14356                 } else {
14357                         /*
14358                          * ip_xmit_attach_llhdr had to prepend an mblk to
14359                          * attach the fastpath header before ip header.
14360                          */
14361                         ip_mp = mp->b_cont;
14362                         freeb(mp);
14363                         mp = ip_mp;
14364                         mp->b_rptr += (fp_mp_len - len);
14365                 }
14366         } else {
14367                 ip_mp = mp->b_cont;
14368                 freeb(mp);
14369                 mp = ip_mp;
14370         }
14371         ip_drop_output("ipIfStatsOutDiscards - flow ctl", mp, ill);
14372         freemsg(mp);
14373 }
14374 
14375 /*
14376  * Normal post fragmentation function.
14377  *
14378  * Send a packet using the passed in nce. This handles both IPv4 and IPv6
14379  * using the same state machine.
14380  *
14381  * We return an error on failure. In particular we return EWOULDBLOCK
14382  * when the driver flow controls. In that case this ensures that ip_wsrv runs
14383  * (currently by canputnext failure resulting in backenabling from GLD.)
14384  * This allows the callers of conn_ip_output() to use EWOULDBLOCK as an
14385  * indication that they can flow control until ip_wsrv() tells then to restart.
14386  *
14387  * If the nce passed by caller is incomplete, this function
14388  * queues the packet and if necessary, sends ARP request and bails.
14389  * If the Neighbor Cache passed is fully resolved, we simply prepend
14390  * the link-layer header to the packet, do ipsec hw acceleration
14391  * work if necessary, and send the packet out on the wire.
14392  */
14393 /* ARGSUSED6 */
14394 int
14395 ip_xmit(mblk_t *mp, nce_t *nce, iaflags_t ixaflags, uint_t pkt_len,
14396     uint32_t xmit_hint, zoneid_t szone, zoneid_t nolzid, uintptr_t *ixacookie)
14397 {
14398         queue_t         *wq;
14399         ill_t           *ill = nce->nce_ill;
14400         ip_stack_t      *ipst = ill->ill_ipst;
14401         uint64_t        delta;
14402         boolean_t       isv6 = ill->ill_isv6;
14403         boolean_t       fp_mp;
14404         ncec_t          *ncec = nce->nce_common;
14405         int64_t         now = LBOLT_FASTPATH64;
14406         boolean_t       is_probe;
14407 
14408         DTRACE_PROBE1(ip__xmit, nce_t *, nce);
14409 
14410         ASSERT(mp != NULL);
14411         ASSERT(mp->b_datap->db_type == M_DATA);
14412         ASSERT(pkt_len == msgdsize(mp));
14413 
14414         /*
14415          * If we have already been here and are coming back after ARP/ND.
14416          * the IXAF_NO_TRACE flag is set. We skip FW_HOOKS, DTRACE and ipobs
14417          * in that case since they have seen the packet when it came here
14418          * the first time.
14419          */
14420         if (ixaflags & IXAF_NO_TRACE)
14421                 goto sendit;
14422 
14423         if (ixaflags & IXAF_IS_IPV4) {
14424                 ipha_t *ipha = (ipha_t *)mp->b_rptr;
14425 
14426                 ASSERT(!isv6);
14427                 ASSERT(pkt_len == ntohs(((ipha_t *)mp->b_rptr)->ipha_length));
14428                 if (HOOKS4_INTERESTED_PHYSICAL_OUT(ipst) &&
14429                     !(ixaflags & IXAF_NO_PFHOOK)) {
14430                         int     error;
14431 
14432                         FW_HOOKS(ipst->ips_ip4_physical_out_event,
14433                             ipst->ips_ipv4firewall_physical_out,
14434                             NULL, ill, ipha, mp, mp, 0, ipst, error);
14435                         DTRACE_PROBE1(ip4__physical__out__end,
14436                             mblk_t *, mp);
14437                         if (mp == NULL)
14438                                 return (error);
14439 
14440                         /* The length could have changed */
14441                         pkt_len = msgdsize(mp);
14442                 }
14443                 if (ipst->ips_ip4_observe.he_interested) {
14444                         /*
14445                          * Note that for TX the zoneid is the sending
14446                          * zone, whether or not MLP is in play.
14447                          * Since the szone argument is the IP zoneid (i.e.,
14448                          * zero for exclusive-IP zones) and ipobs wants
14449                          * the system zoneid, we map it here.
14450                          */
14451                         szone = IP_REAL_ZONEID(szone, ipst);
14452 
14453                         /*
14454                          * On the outbound path the destination zone will be
14455                          * unknown as we're sending this packet out on the
14456                          * wire.
14457                          */
14458                         ipobs_hook(mp, IPOBS_HOOK_OUTBOUND, szone, ALL_ZONES,
14459                             ill, ipst);
14460                 }
14461                 DTRACE_IP7(send, mblk_t *, mp,  conn_t *, NULL,
14462                     void_ip_t *, ipha,  __dtrace_ipsr_ill_t *, ill,
14463                     ipha_t *, ipha, ip6_t *, NULL, int, 0);
14464         } else {
14465                 ip6_t *ip6h = (ip6_t *)mp->b_rptr;
14466 
14467                 ASSERT(isv6);
14468                 ASSERT(pkt_len ==
14469                     ntohs(((ip6_t *)mp->b_rptr)->ip6_plen) + IPV6_HDR_LEN);
14470                 if (HOOKS6_INTERESTED_PHYSICAL_OUT(ipst) &&
14471                     !(ixaflags & IXAF_NO_PFHOOK)) {
14472                         int     error;
14473 
14474                         FW_HOOKS6(ipst->ips_ip6_physical_out_event,
14475                             ipst->ips_ipv6firewall_physical_out,
14476                             NULL, ill, ip6h, mp, mp, 0, ipst, error);
14477                         DTRACE_PROBE1(ip6__physical__out__end,
14478                             mblk_t *, mp);
14479                         if (mp == NULL)
14480                                 return (error);
14481 
14482                         /* The length could have changed */
14483                         pkt_len = msgdsize(mp);
14484                 }
14485                 if (ipst->ips_ip6_observe.he_interested) {
14486                         /* See above */
14487                         szone = IP_REAL_ZONEID(szone, ipst);
14488 
14489                         ipobs_hook(mp, IPOBS_HOOK_OUTBOUND, szone, ALL_ZONES,
14490                             ill, ipst);
14491                 }
14492                 DTRACE_IP7(send, mblk_t *, mp,  conn_t *, NULL,
14493                     void_ip_t *, ip6h,  __dtrace_ipsr_ill_t *, ill,
14494                     ipha_t *, NULL, ip6_t *, ip6h, int, 0);
14495         }
14496 
14497 sendit:
14498         /*
14499          * We check the state without a lock because the state can never
14500          * move "backwards" to initial or incomplete.
14501          */
14502         switch (ncec->ncec_state) {
14503         case ND_REACHABLE:
14504         case ND_STALE:
14505         case ND_DELAY:
14506         case ND_PROBE:
14507                 mp = ip_xmit_attach_llhdr(mp, nce);
14508                 if (mp == NULL) {
14509                         /*
14510                          * ip_xmit_attach_llhdr has increased
14511                          * ipIfStatsOutDiscards and called ip_drop_output()
14512                          */
14513                         return (ENOBUFS);
14514                 }
14515                 /*
14516                  * check if nce_fastpath completed and we tagged on a
14517                  * copy of nce_fp_mp in ip_xmit_attach_llhdr().
14518                  */
14519                 fp_mp = (mp->b_datap->db_type == M_DATA);
14520 
14521                 if (fp_mp &&
14522                     (ill->ill_capabilities & ILL_CAPAB_DLD_DIRECT)) {
14523                         ill_dld_direct_t *idd;
14524 
14525                         idd = &ill->ill_dld_capab->idc_direct;
14526                         /*
14527                          * Send the packet directly to DLD, where it
14528                          * may be queued depending on the availability
14529                          * of transmit resources at the media layer.
14530                          * Return value should be taken into
14531                          * account and flow control the TCP.
14532                          */
14533                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCOutTransmits);
14534                         UPDATE_MIB(ill->ill_ip_mib, ipIfStatsHCOutOctets,
14535                             pkt_len);
14536 
14537                         if (ixaflags & IXAF_NO_DEV_FLOW_CTL) {
14538                                 (void) idd->idd_tx_df(idd->idd_tx_dh, mp,
14539                                     (uintptr_t)xmit_hint, IP_DROP_ON_NO_DESC);
14540                         } else {
14541                                 uintptr_t cookie;
14542 
14543                                 if ((cookie = idd->idd_tx_df(idd->idd_tx_dh,
14544                                     mp, (uintptr_t)xmit_hint, 0)) != 0) {
14545                                         if (ixacookie != NULL)
14546                                                 *ixacookie = cookie;
14547                                         return (EWOULDBLOCK);
14548                                 }
14549                         }
14550                 } else {
14551                         wq = ill->ill_wq;
14552 
14553                         if (!(ixaflags & IXAF_NO_DEV_FLOW_CTL) &&
14554                             !canputnext(wq)) {
14555                                 if (ixacookie != NULL)
14556                                         *ixacookie = 0;
14557                                 ip_xmit_flowctl_drop(ill, mp, fp_mp,
14558                                     nce->nce_fp_mp != NULL ?
14559                                     MBLKL(nce->nce_fp_mp) : 0);
14560                                 return (EWOULDBLOCK);
14561                         }
14562                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCOutTransmits);
14563                         UPDATE_MIB(ill->ill_ip_mib, ipIfStatsHCOutOctets,
14564                             pkt_len);
14565                         putnext(wq, mp);
14566                 }
14567 
14568                 /*
14569                  * The rest of this function implements Neighbor Unreachability
14570                  * detection. Determine if the ncec is eligible for NUD.
14571                  */
14572                 if (ncec->ncec_flags & NCE_F_NONUD)
14573                         return (0);
14574 
14575                 ASSERT(ncec->ncec_state != ND_INCOMPLETE);
14576 
14577                 /*
14578                  * Check for upper layer advice
14579                  */
14580                 if (ixaflags & IXAF_REACH_CONF) {
14581                         timeout_id_t tid;
14582 
14583                         /*
14584                          * It should be o.k. to check the state without
14585                          * a lock here, at most we lose an advice.
14586                          */
14587                         ncec->ncec_last = TICK_TO_MSEC(now);
14588                         if (ncec->ncec_state != ND_REACHABLE) {
14589                                 mutex_enter(&ncec->ncec_lock);
14590                                 ncec->ncec_state = ND_REACHABLE;
14591                                 tid = ncec->ncec_timeout_id;
14592                                 ncec->ncec_timeout_id = 0;
14593                                 mutex_exit(&ncec->ncec_lock);
14594                                 (void) untimeout(tid);
14595                                 if (ip_debug > 2) {
14596                                         /* ip1dbg */
14597                                         pr_addr_dbg("ip_xmit: state"
14598                                             " for %s changed to"
14599                                             " REACHABLE\n", AF_INET6,
14600                                             &ncec->ncec_addr);
14601                                 }
14602                         }
14603                         return (0);
14604                 }
14605 
14606                 delta =  TICK_TO_MSEC(now) - ncec->ncec_last;
14607                 ip1dbg(("ip_xmit: delta = %" PRId64
14608                     " ill_reachable_time = %d \n", delta,
14609                     ill->ill_reachable_time));
14610                 if (delta > (uint64_t)ill->ill_reachable_time) {
14611                         mutex_enter(&ncec->ncec_lock);
14612                         switch (ncec->ncec_state) {
14613                         case ND_REACHABLE:
14614                                 ASSERT((ncec->ncec_flags & NCE_F_NONUD) == 0);
14615                                 /* FALLTHROUGH */
14616                         case ND_STALE:
14617                                 /*
14618                                  * ND_REACHABLE is identical to
14619                                  * ND_STALE in this specific case. If
14620                                  * reachable time has expired for this
14621                                  * neighbor (delta is greater than
14622                                  * reachable time), conceptually, the
14623                                  * neighbor cache is no longer in
14624                                  * REACHABLE state, but already in
14625                                  * STALE state.  So the correct
14626                                  * transition here is to ND_DELAY.
14627                                  */
14628                                 ncec->ncec_state = ND_DELAY;
14629                                 mutex_exit(&ncec->ncec_lock);
14630                                 nce_restart_timer(ncec,
14631                                     ipst->ips_delay_first_probe_time);
14632                                 if (ip_debug > 3) {
14633                                         /* ip2dbg */
14634                                         pr_addr_dbg("ip_xmit: state"
14635                                             " for %s changed to"
14636                                             " DELAY\n", AF_INET6,
14637                                             &ncec->ncec_addr);
14638                                 }
14639                                 break;
14640                         case ND_DELAY:
14641                         case ND_PROBE:
14642                                 mutex_exit(&ncec->ncec_lock);
14643                                 /* Timers have already started */
14644                                 break;
14645                         case ND_UNREACHABLE:
14646                                 /*
14647                                  * nce_timer has detected that this ncec
14648                                  * is unreachable and initiated deleting
14649                                  * this ncec.
14650                                  * This is a harmless race where we found the
14651                                  * ncec before it was deleted and have
14652                                  * just sent out a packet using this
14653                                  * unreachable ncec.
14654                                  */
14655                                 mutex_exit(&ncec->ncec_lock);
14656                                 break;
14657                         default:
14658                                 ASSERT(0);
14659                                 mutex_exit(&ncec->ncec_lock);
14660                         }
14661                 }
14662                 return (0);
14663 
14664         case ND_INCOMPLETE:
14665                 /*
14666                  * the state could have changed since we didn't hold the lock.
14667                  * Re-verify state under lock.
14668                  */
14669                 is_probe = ipmp_packet_is_probe(mp, nce->nce_ill);
14670                 mutex_enter(&ncec->ncec_lock);
14671                 if (NCE_ISREACHABLE(ncec)) {
14672                         mutex_exit(&ncec->ncec_lock);
14673                         goto sendit;
14674                 }
14675                 /* queue the packet */
14676                 nce_queue_mp(ncec, mp, is_probe);
14677                 mutex_exit(&ncec->ncec_lock);
14678                 DTRACE_PROBE2(ip__xmit__incomplete,
14679                     (ncec_t *), ncec, (mblk_t *), mp);
14680                 return (0);
14681 
14682         case ND_INITIAL:
14683                 /*
14684                  * State could have changed since we didn't hold the lock, so
14685                  * re-verify state.
14686                  */
14687                 is_probe = ipmp_packet_is_probe(mp, nce->nce_ill);
14688                 mutex_enter(&ncec->ncec_lock);
14689                 if (NCE_ISREACHABLE(ncec))  {
14690                         mutex_exit(&ncec->ncec_lock);
14691                         goto sendit;
14692                 }
14693                 nce_queue_mp(ncec, mp, is_probe);
14694                 if (ncec->ncec_state == ND_INITIAL) {
14695                         ncec->ncec_state = ND_INCOMPLETE;
14696                         mutex_exit(&ncec->ncec_lock);
14697                         /*
14698                          * figure out the source we want to use
14699                          * and resolve it.
14700                          */
14701                         ip_ndp_resolve(ncec);
14702                 } else  {
14703                         mutex_exit(&ncec->ncec_lock);
14704                 }
14705                 return (0);
14706 
14707         case ND_UNREACHABLE:
14708                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
14709                 ip_drop_output("ipIfStatsOutDiscards - ND_UNREACHABLE",
14710                     mp, ill);
14711                 freemsg(mp);
14712                 return (0);
14713 
14714         default:
14715                 ASSERT(0);
14716                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
14717                 ip_drop_output("ipIfStatsOutDiscards - ND_other",
14718                     mp, ill);
14719                 freemsg(mp);
14720                 return (ENETUNREACH);
14721         }
14722 }
14723 
14724 /*
14725  * Return B_TRUE if the buffers differ in length or content.
14726  * This is used for comparing extension header buffers.
14727  * Note that an extension header would be declared different
14728  * even if all that changed was the next header value in that header i.e.
14729  * what really changed is the next extension header.
14730  */
14731 boolean_t
14732 ip_cmpbuf(const void *abuf, uint_t alen, boolean_t b_valid, const void *bbuf,
14733     uint_t blen)
14734 {
14735         if (!b_valid)
14736                 blen = 0;
14737 
14738         if (alen != blen)
14739                 return (B_TRUE);
14740         if (alen == 0)
14741                 return (B_FALSE);       /* Both zero length */
14742         return (bcmp(abuf, bbuf, alen));
14743 }
14744 
14745 /*
14746  * Preallocate memory for ip_savebuf(). Returns B_TRUE if ok.
14747  * Return B_FALSE if memory allocation fails - don't change any state!
14748  */
14749 boolean_t
14750 ip_allocbuf(void **dstp, uint_t *dstlenp, boolean_t src_valid,
14751     const void *src, uint_t srclen)
14752 {
14753         void *dst;
14754 
14755         if (!src_valid)
14756                 srclen = 0;
14757 
14758         ASSERT(*dstlenp == 0);
14759         if (src != NULL && srclen != 0) {
14760                 dst = mi_alloc(srclen, BPRI_MED);
14761                 if (dst == NULL)
14762                         return (B_FALSE);
14763         } else {
14764                 dst = NULL;
14765         }
14766         if (*dstp != NULL)
14767                 mi_free(*dstp);
14768         *dstp = dst;
14769         *dstlenp = dst == NULL ? 0 : srclen;
14770         return (B_TRUE);
14771 }
14772 
14773 /*
14774  * Replace what is in *dst, *dstlen with the source.
14775  * Assumes ip_allocbuf has already been called.
14776  */
14777 void
14778 ip_savebuf(void **dstp, uint_t *dstlenp, boolean_t src_valid,
14779     const void *src, uint_t srclen)
14780 {
14781         if (!src_valid)
14782                 srclen = 0;
14783 
14784         ASSERT(*dstlenp == srclen);
14785         if (src != NULL && srclen != 0)
14786                 bcopy(src, *dstp, srclen);
14787 }
14788 
14789 /*
14790  * Free the storage pointed to by the members of an ip_pkt_t.
14791  */
14792 void
14793 ip_pkt_free(ip_pkt_t *ipp)
14794 {
14795         uint_t  fields = ipp->ipp_fields;
14796 
14797         if (fields & IPPF_HOPOPTS) {
14798                 kmem_free(ipp->ipp_hopopts, ipp->ipp_hopoptslen);
14799                 ipp->ipp_hopopts = NULL;
14800                 ipp->ipp_hopoptslen = 0;
14801         }
14802         if (fields & IPPF_RTHDRDSTOPTS) {
14803                 kmem_free(ipp->ipp_rthdrdstopts, ipp->ipp_rthdrdstoptslen);
14804                 ipp->ipp_rthdrdstopts = NULL;
14805                 ipp->ipp_rthdrdstoptslen = 0;
14806         }
14807         if (fields & IPPF_DSTOPTS) {
14808                 kmem_free(ipp->ipp_dstopts, ipp->ipp_dstoptslen);
14809                 ipp->ipp_dstopts = NULL;
14810                 ipp->ipp_dstoptslen = 0;
14811         }
14812         if (fields & IPPF_RTHDR) {
14813                 kmem_free(ipp->ipp_rthdr, ipp->ipp_rthdrlen);
14814                 ipp->ipp_rthdr = NULL;
14815                 ipp->ipp_rthdrlen = 0;
14816         }
14817         if (fields & IPPF_IPV4_OPTIONS) {
14818                 kmem_free(ipp->ipp_ipv4_options, ipp->ipp_ipv4_options_len);
14819                 ipp->ipp_ipv4_options = NULL;
14820                 ipp->ipp_ipv4_options_len = 0;
14821         }
14822         if (fields & IPPF_LABEL_V4) {
14823                 kmem_free(ipp->ipp_label_v4, ipp->ipp_label_len_v4);
14824                 ipp->ipp_label_v4 = NULL;
14825                 ipp->ipp_label_len_v4 = 0;
14826         }
14827         if (fields & IPPF_LABEL_V6) {
14828                 kmem_free(ipp->ipp_label_v6, ipp->ipp_label_len_v6);
14829                 ipp->ipp_label_v6 = NULL;
14830                 ipp->ipp_label_len_v6 = 0;
14831         }
14832         ipp->ipp_fields &= ~(IPPF_HOPOPTS | IPPF_RTHDRDSTOPTS | IPPF_DSTOPTS |
14833             IPPF_RTHDR | IPPF_IPV4_OPTIONS | IPPF_LABEL_V4 | IPPF_LABEL_V6);
14834 }
14835 
14836 /*
14837  * Copy from src to dst and allocate as needed.
14838  * Returns zero or ENOMEM.
14839  *
14840  * The caller must initialize dst to zero.
14841  */
14842 int
14843 ip_pkt_copy(ip_pkt_t *src, ip_pkt_t *dst, int kmflag)
14844 {
14845         uint_t  fields = src->ipp_fields;
14846 
14847         /* Start with fields that don't require memory allocation */
14848         dst->ipp_fields = fields &
14849             ~(IPPF_HOPOPTS | IPPF_RTHDRDSTOPTS | IPPF_DSTOPTS |
14850             IPPF_RTHDR | IPPF_IPV4_OPTIONS | IPPF_LABEL_V4 | IPPF_LABEL_V6);
14851 
14852         dst->ipp_addr = src->ipp_addr;
14853         dst->ipp_unicast_hops = src->ipp_unicast_hops;
14854         dst->ipp_hoplimit = src->ipp_hoplimit;
14855         dst->ipp_tclass = src->ipp_tclass;
14856         dst->ipp_type_of_service = src->ipp_type_of_service;
14857 
14858         if (!(fields & (IPPF_HOPOPTS | IPPF_RTHDRDSTOPTS | IPPF_DSTOPTS |
14859             IPPF_RTHDR | IPPF_IPV4_OPTIONS | IPPF_LABEL_V4 | IPPF_LABEL_V6)))
14860                 return (0);
14861 
14862         if (fields & IPPF_HOPOPTS) {
14863                 dst->ipp_hopopts = kmem_alloc(src->ipp_hopoptslen, kmflag);
14864                 if (dst->ipp_hopopts == NULL) {
14865                         ip_pkt_free(dst);
14866                         return (ENOMEM);
14867                 }
14868                 dst->ipp_fields |= IPPF_HOPOPTS;
14869                 bcopy(src->ipp_hopopts, dst->ipp_hopopts,
14870                     src->ipp_hopoptslen);
14871                 dst->ipp_hopoptslen = src->ipp_hopoptslen;
14872         }
14873         if (fields & IPPF_RTHDRDSTOPTS) {
14874                 dst->ipp_rthdrdstopts = kmem_alloc(src->ipp_rthdrdstoptslen,
14875                     kmflag);
14876                 if (dst->ipp_rthdrdstopts == NULL) {
14877                         ip_pkt_free(dst);
14878                         return (ENOMEM);
14879                 }
14880                 dst->ipp_fields |= IPPF_RTHDRDSTOPTS;
14881                 bcopy(src->ipp_rthdrdstopts, dst->ipp_rthdrdstopts,
14882                     src->ipp_rthdrdstoptslen);
14883                 dst->ipp_rthdrdstoptslen = src->ipp_rthdrdstoptslen;
14884         }
14885         if (fields & IPPF_DSTOPTS) {
14886                 dst->ipp_dstopts = kmem_alloc(src->ipp_dstoptslen, kmflag);
14887                 if (dst->ipp_dstopts == NULL) {
14888                         ip_pkt_free(dst);
14889                         return (ENOMEM);
14890                 }
14891                 dst->ipp_fields |= IPPF_DSTOPTS;
14892                 bcopy(src->ipp_dstopts, dst->ipp_dstopts,
14893                     src->ipp_dstoptslen);
14894                 dst->ipp_dstoptslen = src->ipp_dstoptslen;
14895         }
14896         if (fields & IPPF_RTHDR) {
14897                 dst->ipp_rthdr = kmem_alloc(src->ipp_rthdrlen, kmflag);
14898                 if (dst->ipp_rthdr == NULL) {
14899                         ip_pkt_free(dst);
14900                         return (ENOMEM);
14901                 }
14902                 dst->ipp_fields |= IPPF_RTHDR;
14903                 bcopy(src->ipp_rthdr, dst->ipp_rthdr,
14904                     src->ipp_rthdrlen);
14905                 dst->ipp_rthdrlen = src->ipp_rthdrlen;
14906         }
14907         if (fields & IPPF_IPV4_OPTIONS) {
14908                 dst->ipp_ipv4_options = kmem_alloc(src->ipp_ipv4_options_len,
14909                     kmflag);
14910                 if (dst->ipp_ipv4_options == NULL) {
14911                         ip_pkt_free(dst);
14912                         return (ENOMEM);
14913                 }
14914                 dst->ipp_fields |= IPPF_IPV4_OPTIONS;
14915                 bcopy(src->ipp_ipv4_options, dst->ipp_ipv4_options,
14916                     src->ipp_ipv4_options_len);
14917                 dst->ipp_ipv4_options_len = src->ipp_ipv4_options_len;
14918         }
14919         if (fields & IPPF_LABEL_V4) {
14920                 dst->ipp_label_v4 = kmem_alloc(src->ipp_label_len_v4, kmflag);
14921                 if (dst->ipp_label_v4 == NULL) {
14922                         ip_pkt_free(dst);
14923                         return (ENOMEM);
14924                 }
14925                 dst->ipp_fields |= IPPF_LABEL_V4;
14926                 bcopy(src->ipp_label_v4, dst->ipp_label_v4,
14927                     src->ipp_label_len_v4);
14928                 dst->ipp_label_len_v4 = src->ipp_label_len_v4;
14929         }
14930         if (fields & IPPF_LABEL_V6) {
14931                 dst->ipp_label_v6 = kmem_alloc(src->ipp_label_len_v6, kmflag);
14932                 if (dst->ipp_label_v6 == NULL) {
14933                         ip_pkt_free(dst);
14934                         return (ENOMEM);
14935                 }
14936                 dst->ipp_fields |= IPPF_LABEL_V6;
14937                 bcopy(src->ipp_label_v6, dst->ipp_label_v6,
14938                     src->ipp_label_len_v6);
14939                 dst->ipp_label_len_v6 = src->ipp_label_len_v6;
14940         }
14941         if (fields & IPPF_FRAGHDR) {
14942                 dst->ipp_fraghdr = kmem_alloc(src->ipp_fraghdrlen, kmflag);
14943                 if (dst->ipp_fraghdr == NULL) {
14944                         ip_pkt_free(dst);
14945                         return (ENOMEM);
14946                 }
14947                 dst->ipp_fields |= IPPF_FRAGHDR;
14948                 bcopy(src->ipp_fraghdr, dst->ipp_fraghdr,
14949                     src->ipp_fraghdrlen);
14950                 dst->ipp_fraghdrlen = src->ipp_fraghdrlen;
14951         }
14952         return (0);
14953 }
14954 
14955 /*
14956  * Returns INADDR_ANY if no source route
14957  */
14958 ipaddr_t
14959 ip_pkt_source_route_v4(const ip_pkt_t *ipp)
14960 {
14961         ipaddr_t        nexthop = INADDR_ANY;
14962         ipoptp_t        opts;
14963         uchar_t         *opt;
14964         uint8_t         optval;
14965         uint8_t         optlen;
14966         uint32_t        totallen;
14967 
14968         if (!(ipp->ipp_fields & IPPF_IPV4_OPTIONS))
14969                 return (INADDR_ANY);
14970 
14971         totallen = ipp->ipp_ipv4_options_len;
14972         if (totallen & 0x3)
14973                 return (INADDR_ANY);
14974 
14975         for (optval = ipoptp_first2(&opts, totallen, ipp->ipp_ipv4_options);
14976             optval != IPOPT_EOL;
14977             optval = ipoptp_next(&opts)) {
14978                 opt = opts.ipoptp_cur;
14979                 switch (optval) {
14980                         uint8_t off;
14981                 case IPOPT_SSRR:
14982                 case IPOPT_LSRR:
14983                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
14984                                 break;
14985                         }
14986                         optlen = opts.ipoptp_len;
14987                         off = opt[IPOPT_OFFSET];
14988                         off--;
14989                         if (optlen < IP_ADDR_LEN ||
14990                             off > optlen - IP_ADDR_LEN) {
14991                                 /* End of source route */
14992                                 break;
14993                         }
14994                         bcopy((char *)opt + off, &nexthop, IP_ADDR_LEN);
14995                         if (nexthop == htonl(INADDR_LOOPBACK)) {
14996                                 /* Ignore */
14997                                 nexthop = INADDR_ANY;
14998                                 break;
14999                         }
15000                         break;
15001                 }
15002         }
15003         return (nexthop);
15004 }
15005 
15006 /*
15007  * Reverse a source route.
15008  */
15009 void
15010 ip_pkt_source_route_reverse_v4(ip_pkt_t *ipp)
15011 {
15012         ipaddr_t        tmp;
15013         ipoptp_t        opts;
15014         uchar_t         *opt;
15015         uint8_t         optval;
15016         uint32_t        totallen;
15017 
15018         if (!(ipp->ipp_fields & IPPF_IPV4_OPTIONS))
15019                 return;
15020 
15021         totallen = ipp->ipp_ipv4_options_len;
15022         if (totallen & 0x3)
15023                 return;
15024 
15025         for (optval = ipoptp_first2(&opts, totallen, ipp->ipp_ipv4_options);
15026             optval != IPOPT_EOL;
15027             optval = ipoptp_next(&opts)) {
15028                 uint8_t off1, off2;
15029 
15030                 opt = opts.ipoptp_cur;
15031                 switch (optval) {
15032                 case IPOPT_SSRR:
15033                 case IPOPT_LSRR:
15034                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
15035                                 break;
15036                         }
15037                         off1 = IPOPT_MINOFF_SR - 1;
15038                         off2 = opt[IPOPT_OFFSET] - IP_ADDR_LEN - 1;
15039                         while (off2 > off1) {
15040                                 bcopy(opt + off2, &tmp, IP_ADDR_LEN);
15041                                 bcopy(opt + off1, opt + off2, IP_ADDR_LEN);
15042                                 bcopy(&tmp, opt + off2, IP_ADDR_LEN);
15043                                 off2 -= IP_ADDR_LEN;
15044                                 off1 += IP_ADDR_LEN;
15045                         }
15046                         opt[IPOPT_OFFSET] = IPOPT_MINOFF_SR;
15047                         break;
15048                 }
15049         }
15050 }
15051 
15052 /*
15053  * Returns NULL if no routing header
15054  */
15055 in6_addr_t *
15056 ip_pkt_source_route_v6(const ip_pkt_t *ipp)
15057 {
15058         in6_addr_t      *nexthop = NULL;
15059         ip6_rthdr0_t    *rthdr;
15060 
15061         if (!(ipp->ipp_fields & IPPF_RTHDR))
15062                 return (NULL);
15063 
15064         rthdr = (ip6_rthdr0_t *)ipp->ipp_rthdr;
15065         if (rthdr->ip6r0_segleft == 0)
15066                 return (NULL);
15067 
15068         nexthop = (in6_addr_t *)((char *)rthdr + sizeof (*rthdr));
15069         return (nexthop);
15070 }
15071 
15072 zoneid_t
15073 ip_get_zoneid_v4(ipaddr_t addr, mblk_t *mp, ip_recv_attr_t *ira,
15074     zoneid_t lookup_zoneid)
15075 {
15076         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
15077         ire_t           *ire;
15078         int             ire_flags = MATCH_IRE_TYPE;
15079         zoneid_t        zoneid = ALL_ZONES;
15080 
15081         if (is_system_labeled() && !tsol_can_accept_raw(mp, ira, B_FALSE))
15082                 return (ALL_ZONES);
15083 
15084         if (lookup_zoneid != ALL_ZONES)
15085                 ire_flags |= MATCH_IRE_ZONEONLY;
15086         ire = ire_ftable_lookup_v4(addr, NULL, NULL, IRE_LOCAL | IRE_LOOPBACK,
15087             NULL, lookup_zoneid, NULL, ire_flags, 0, ipst, NULL);
15088         if (ire != NULL) {
15089                 zoneid = IP_REAL_ZONEID(ire->ire_zoneid, ipst);
15090                 ire_refrele(ire);
15091         }
15092         return (zoneid);
15093 }
15094 
15095 zoneid_t
15096 ip_get_zoneid_v6(in6_addr_t *addr, mblk_t *mp, const ill_t *ill,
15097     ip_recv_attr_t *ira, zoneid_t lookup_zoneid)
15098 {
15099         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
15100         ire_t           *ire;
15101         int             ire_flags = MATCH_IRE_TYPE;
15102         zoneid_t        zoneid = ALL_ZONES;
15103 
15104         if (is_system_labeled() && !tsol_can_accept_raw(mp, ira, B_FALSE))
15105                 return (ALL_ZONES);
15106 
15107         if (IN6_IS_ADDR_LINKLOCAL(addr))
15108                 ire_flags |= MATCH_IRE_ILL;
15109 
15110         if (lookup_zoneid != ALL_ZONES)
15111                 ire_flags |= MATCH_IRE_ZONEONLY;
15112         ire = ire_ftable_lookup_v6(addr, NULL, NULL, IRE_LOCAL | IRE_LOOPBACK,
15113             ill, lookup_zoneid, NULL, ire_flags, 0, ipst, NULL);
15114         if (ire != NULL) {
15115                 zoneid = IP_REAL_ZONEID(ire->ire_zoneid, ipst);
15116                 ire_refrele(ire);
15117         }
15118         return (zoneid);
15119 }
15120 
15121 /*
15122  * IP obserability hook support functions.
15123  */
15124 static void
15125 ipobs_init(ip_stack_t *ipst)
15126 {
15127         netid_t id;
15128 
15129         id = net_getnetidbynetstackid(ipst->ips_netstack->netstack_stackid);
15130 
15131         ipst->ips_ip4_observe_pr = net_protocol_lookup(id, NHF_INET);
15132         VERIFY(ipst->ips_ip4_observe_pr != NULL);
15133 
15134         ipst->ips_ip6_observe_pr = net_protocol_lookup(id, NHF_INET6);
15135         VERIFY(ipst->ips_ip6_observe_pr != NULL);
15136 }
15137 
15138 static void
15139 ipobs_fini(ip_stack_t *ipst)
15140 {
15141 
15142         VERIFY(net_protocol_release(ipst->ips_ip4_observe_pr) == 0);
15143         VERIFY(net_protocol_release(ipst->ips_ip6_observe_pr) == 0);
15144 }
15145 
15146 /*
15147  * hook_pkt_observe_t is composed in network byte order so that the
15148  * entire mblk_t chain handed into hook_run can be used as-is.
15149  * The caveat is that use of the fields, such as the zone fields,
15150  * requires conversion into host byte order first.
15151  */
15152 void
15153 ipobs_hook(mblk_t *mp, int htype, zoneid_t zsrc, zoneid_t zdst,
15154     const ill_t *ill, ip_stack_t *ipst)
15155 {
15156         hook_pkt_observe_t *hdr;
15157         uint64_t grifindex;
15158         mblk_t *imp;
15159 
15160         imp = allocb(sizeof (*hdr), BPRI_HI);
15161         if (imp == NULL)
15162                 return;
15163 
15164         hdr = (hook_pkt_observe_t *)imp->b_rptr;
15165         /*
15166          * b_wptr is set to make the apparent size of the data in the mblk_t
15167          * to exclude the pointers at the end of hook_pkt_observer_t.
15168          */
15169         imp->b_wptr = imp->b_rptr + sizeof (dl_ipnetinfo_t);
15170         imp->b_cont = mp;
15171 
15172         ASSERT(DB_TYPE(mp) == M_DATA);
15173 
15174         if (IS_UNDER_IPMP(ill))
15175                 grifindex = ipmp_ill_get_ipmp_ifindex(ill);
15176         else
15177                 grifindex = 0;
15178 
15179         hdr->hpo_version = 1;
15180         hdr->hpo_htype = htons(htype);
15181         hdr->hpo_pktlen = htonl((ulong_t)msgdsize(mp));
15182         hdr->hpo_ifindex = htonl(ill->ill_phyint->phyint_ifindex);
15183         hdr->hpo_grifindex = htonl(grifindex);
15184         hdr->hpo_zsrc = htonl(zsrc);
15185         hdr->hpo_zdst = htonl(zdst);
15186         hdr->hpo_pkt = imp;
15187         hdr->hpo_ctx = ipst->ips_netstack;
15188 
15189         if (ill->ill_isv6) {
15190                 hdr->hpo_family = AF_INET6;
15191                 (void) hook_run(ipst->ips_ipv6_net_data->netd_hooks,
15192                     ipst->ips_ipv6observing, (hook_data_t)hdr);
15193         } else {
15194                 hdr->hpo_family = AF_INET;
15195                 (void) hook_run(ipst->ips_ipv4_net_data->netd_hooks,
15196                     ipst->ips_ipv4observing, (hook_data_t)hdr);
15197         }
15198 
15199         imp->b_cont = NULL;
15200         freemsg(imp);
15201 }
15202 
15203 /*
15204  * Utility routine that checks if `v4srcp' is a valid address on underlying
15205  * interface `ill'.  If `ipifp' is non-NULL, it's set to a held ipif
15206  * associated with `v4srcp' on success.  NOTE: if this is not called from
15207  * inside the IPSQ (ill_g_lock is not held), `ill' may be removed from the
15208  * group during or after this lookup.
15209  */
15210 boolean_t
15211 ipif_lookup_testaddr_v4(ill_t *ill, const in_addr_t *v4srcp, ipif_t **ipifp)
15212 {
15213         ipif_t *ipif;
15214 
15215         ipif = ipif_lookup_addr_exact(*v4srcp, ill, ill->ill_ipst);
15216         if (ipif != NULL) {
15217                 if (ipifp != NULL)
15218                         *ipifp = ipif;
15219                 else
15220                         ipif_refrele(ipif);
15221                 return (B_TRUE);
15222         }
15223 
15224         ip1dbg(("ipif_lookup_testaddr_v4: cannot find ipif for src %x\n",
15225             *v4srcp));
15226         return (B_FALSE);
15227 }
15228 
15229 /*
15230  * Transport protocol call back function for CPU state change.
15231  */
15232 /* ARGSUSED */
15233 static int
15234 ip_tp_cpu_update(cpu_setup_t what, int id, void *arg)
15235 {
15236         processorid_t cpu_seqid;
15237         netstack_handle_t nh;
15238         netstack_t *ns;
15239 
15240         ASSERT(MUTEX_HELD(&cpu_lock));
15241 
15242         switch (what) {
15243         case CPU_CONFIG:
15244         case CPU_ON:
15245         case CPU_INIT:
15246         case CPU_CPUPART_IN:
15247                 cpu_seqid = cpu[id]->cpu_seqid;
15248                 netstack_next_init(&nh);
15249                 while ((ns = netstack_next(&nh)) != NULL) {
15250                         dccp_stack_cpu_add(ns->netstack_dccp, cpu_seqid);
15251                         tcp_stack_cpu_add(ns->netstack_tcp, cpu_seqid);
15252                         sctp_stack_cpu_add(ns->netstack_sctp, cpu_seqid);
15253                         udp_stack_cpu_add(ns->netstack_udp, cpu_seqid);
15254                         netstack_rele(ns);
15255                 }
15256                 netstack_next_fini(&nh);
15257                 break;
15258         case CPU_UNCONFIG:
15259         case CPU_OFF:
15260         case CPU_CPUPART_OUT:
15261                 /*
15262                  * Nothing to do.  We don't remove the per CPU stats from
15263                  * the IP stack even when the CPU goes offline.
15264                  */
15265                 break;
15266         default:
15267                 break;
15268         }
15269         return (0);
15270 }