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 2008 Sun Microsystems, Inc.  All rights reserved.
  24  * Use is subject to license terms.
  25  */
  26 /* Copyright (c) 1990 Mentat Inc. */
  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/sunddi.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 
  50 #include <sys/systm.h>
  51 #include <sys/param.h>
  52 #include <sys/kmem.h>
  53 #include <sys/sdt.h>
  54 #include <sys/socket.h>
  55 #include <sys/vtrace.h>
  56 #include <sys/isa_defs.h>
  57 #include <sys/mac.h>
  58 #include <net/if.h>
  59 #include <net/if_arp.h>
  60 #include <net/route.h>
  61 #include <sys/sockio.h>
  62 #include <netinet/in.h>
  63 #include <net/if_dl.h>
  64 
  65 #include <inet/common.h>
  66 #include <inet/mi.h>
  67 #include <inet/mib2.h>
  68 #include <inet/nd.h>
  69 #include <inet/arp.h>
  70 #include <inet/snmpcom.h>
  71 #include <inet/optcom.h>
  72 #include <inet/kstatcom.h>
  73 
  74 #include <netinet/igmp_var.h>
  75 #include <netinet/ip6.h>
  76 #include <netinet/icmp6.h>
  77 #include <netinet/sctp.h>
  78 
  79 #include <inet/ip.h>
  80 #include <inet/ip_impl.h>
  81 #include <inet/ip6.h>
  82 #include <inet/ip6_asp.h>
  83 #include <inet/tcp.h>
  84 #include <inet/tcp_impl.h>
  85 #include <inet/ip_multi.h>
  86 #include <inet/ip_if.h>
  87 #include <inet/ip_ire.h>
  88 #include <inet/ip_ftable.h>
  89 #include <inet/ip_rts.h>
  90 #include <inet/ip_ndp.h>
  91 #include <inet/ip_listutils.h>
  92 #include <netinet/igmp.h>
  93 #include <netinet/ip_mroute.h>
  94 #include <inet/ipp_common.h>
  95 
  96 #include <net/pfkeyv2.h>
  97 #include <inet/ipsec_info.h>
  98 #include <inet/sadb.h>
  99 #include <inet/ipsec_impl.h>
 100 #include <sys/iphada.h>
 101 #include <inet/tun.h>
 102 #include <inet/ipdrop.h>
 103 #include <inet/ip_netinfo.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/multidata.h>
 114 #include <sys/pattr.h>
 115 
 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 #include <sys/sunddi.h>
 123 
 124 #include <sys/tsol/label.h>
 125 #include <sys/tsol/tnet.h>
 126 
 127 #include <rpc/pmap_prot.h>
 128 
 129 /*
 130  * Values for squeue switch:
 131  * IP_SQUEUE_ENTER_NODRAIN: squeue_enter_nodrain
 132  * IP_SQUEUE_ENTER: squeue_enter
 133  * IP_SQUEUE_FILL: squeue_fill
 134  */
 135 int ip_squeue_enter = 2;        /* Setable in /etc/system */
 136 
 137 squeue_func_t ip_input_proc;
 138 #define SET_BPREV_FLAG(x)       ((mblk_t *)(uintptr_t)(x))
 139 
 140 /*
 141  * Setable in /etc/system
 142  */
 143 int ip_poll_normal_ms = 100;
 144 int ip_poll_normal_ticks = 0;
 145 int ip_modclose_ackwait_ms = 3000;
 146 
 147 /*
 148  * It would be nice to have these present only in DEBUG systems, but the
 149  * current design of the global symbol checking logic requires them to be
 150  * unconditionally present.
 151  */
 152 uint_t ip_thread_data;                  /* TSD key for debug support */
 153 krwlock_t ip_thread_rwlock;
 154 list_t  ip_thread_list;
 155 
 156 /*
 157  * Structure to represent a linked list of msgblks. Used by ip_snmp_ functions.
 158  */
 159 
 160 struct listptr_s {
 161         mblk_t  *lp_head;       /* pointer to the head of the list */
 162         mblk_t  *lp_tail;       /* pointer to the tail of the list */
 163 };
 164 
 165 typedef struct listptr_s listptr_t;
 166 
 167 /*
 168  * This is used by ip_snmp_get_mib2_ip_route_media and
 169  * ip_snmp_get_mib2_ip6_route_media to carry the lists of return data.
 170  */
 171 typedef struct iproutedata_s {
 172         uint_t          ird_idx;
 173         listptr_t       ird_route;      /* ipRouteEntryTable */
 174         listptr_t       ird_netmedia;   /* ipNetToMediaEntryTable */
 175         listptr_t       ird_attrs;      /* ipRouteAttributeTable */
 176 } iproutedata_t;
 177 
 178 /*
 179  * Cluster specific hooks. These should be NULL when booted as a non-cluster
 180  */
 181 
 182 /*
 183  * Hook functions to enable cluster networking
 184  * On non-clustered systems these vectors must always be NULL.
 185  *
 186  * Hook function to Check ip specified ip address is a shared ip address
 187  * in the cluster
 188  *
 189  */
 190 int (*cl_inet_isclusterwide)(uint8_t protocol,
 191     sa_family_t addr_family, uint8_t *laddrp) = NULL;
 192 
 193 /*
 194  * Hook function to generate cluster wide ip fragment identifier
 195  */
 196 uint32_t (*cl_inet_ipident)(uint8_t protocol, sa_family_t addr_family,
 197     uint8_t *laddrp, uint8_t *faddrp) = NULL;
 198 
 199 /*
 200  * Synchronization notes:
 201  *
 202  * IP is a fully D_MP STREAMS module/driver. Thus it does not depend on any
 203  * MT level protection given by STREAMS. IP uses a combination of its own
 204  * internal serialization mechanism and standard Solaris locking techniques.
 205  * The internal serialization is per phyint (no IPMP) or per IPMP group.
 206  * This is used to serialize plumbing operations, IPMP operations, certain
 207  * multicast operations, most set ioctls, igmp/mld timers etc.
 208  *
 209  * Plumbing is a long sequence of operations involving message
 210  * exchanges between IP, ARP and device drivers. Many set ioctls are typically
 211  * involved in plumbing operations. A natural model is to serialize these
 212  * ioctls one per ill. For example plumbing of hme0 and qfe0 can go on in
 213  * parallel without any interference. But various set ioctls on hme0 are best
 214  * serialized. However if the system uses IPMP, the operations are easier if
 215  * they are serialized on a per IPMP group basis since IPMP operations
 216  * happen across ill's of a group. Thus the lowest common denominator is to
 217  * serialize most set ioctls, multicast join/leave operations, IPMP operations
 218  * igmp/mld timer operations, and processing of DLPI control messages received
 219  * from drivers on a per IPMP group basis. If the system does not employ
 220  * IPMP the serialization is on a per phyint basis. This serialization is
 221  * provided by the ipsq_t and primitives operating on this. Details can
 222  * be found in ip_if.c above the core primitives operating on ipsq_t.
 223  *
 224  * Lookups of an ipif or ill by a thread return a refheld ipif / ill.
 225  * Simiarly lookup of an ire by a thread also returns a refheld ire.
 226  * In addition ipif's and ill's referenced by the ire are also indirectly
 227  * refheld. Thus no ipif or ill can vanish nor can critical parameters like
 228  * the ipif's address or netmask change as long as an ipif is refheld
 229  * directly or indirectly. For example an SIOCLIFADDR ioctl that changes the
 230  * address of an ipif has to go through the ipsq_t. This ensures that only
 231  * 1 such exclusive operation proceeds at any time on the ipif. It then
 232  * deletes all ires associated with this ipif, and waits for all refcnts
 233  * associated with this ipif to come down to zero. The address is changed
 234  * only after the ipif has been quiesced. Then the ipif is brought up again.
 235  * More details are described above the comment in ip_sioctl_flags.
 236  *
 237  * Packet processing is based mostly on IREs and are fully multi-threaded
 238  * using standard Solaris MT techniques.
 239  *
 240  * There are explicit locks in IP to handle:
 241  * - The ip_g_head list maintained by mi_open_link() and friends.
 242  *
 243  * - The reassembly data structures (one lock per hash bucket)
 244  *
 245  * - conn_lock is meant to protect conn_t fields. The fields actually
 246  *   protected by conn_lock are documented in the conn_t definition.
 247  *
 248  * - ire_lock to protect some of the fields of the ire, IRE tables
 249  *   (one lock per hash bucket). Refer to ip_ire.c for details.
 250  *
 251  * - ndp_g_lock and nce_lock for protecting NCEs.
 252  *
 253  * - ill_lock protects fields of the ill and ipif. Details in ip.h
 254  *
 255  * - ill_g_lock: This is a global reader/writer lock. Protects the following
 256  *      * The AVL tree based global multi list of all ills.
 257  *      * The linked list of all ipifs of an ill
 258  *      * The <ill-ipsq> mapping
 259  *      * The ipsq->ipsq_phyint_list threaded by phyint_ipsq_next
 260  *      * The illgroup list threaded by ill_group_next.
 261  *      * <ill-phyint> association
 262  *   Insertion/deletion of an ill in the system, insertion/deletion of an ipif
 263  *   into an ill, changing the <ill-ipsq> mapping of an ill, insertion/deletion
 264  *   of an ill into the illgrp list, changing the <ill-phyint> assoc of an ill
 265  *   will all have to hold the ill_g_lock as writer for the actual duration
 266  *   of the insertion/deletion/change. More details about the <ill-ipsq> mapping
 267  *   may be found in the IPMP section.
 268  *
 269  * - ill_lock:  This is a per ill mutex.
 270  *   It protects some members of the ill and is documented below.
 271  *   It also protects the <ill-ipsq> mapping
 272  *   It also protects the illgroup list threaded by ill_group_next.
 273  *   It also protects the <ill-phyint> assoc.
 274  *   It also protects the list of ipifs hanging off the ill.
 275  *
 276  * - ipsq_lock: This is a per ipsq_t mutex lock.
 277  *   This protects all the other members of the ipsq struct except
 278  *   ipsq_refs and ipsq_phyint_list which are protected by ill_g_lock
 279  *
 280  * - illgrp_lock: This is a per ill_group mutex lock.
 281  *   The only thing it protects is the illgrp_ill_schednext member of ill_group
 282  *   which dictates which is the next ill in an ill_group that is to be chosen
 283  *   for sending outgoing packets, through creation of an IRE_CACHE that
 284  *   references this ill.
 285  *
 286  * - phyint_lock: This is a per phyint mutex lock. Protects just the
 287  *   phyint_flags
 288  *
 289  * - ip_g_nd_lock: This is a global reader/writer lock.
 290  *   Any call to nd_load to load a new parameter to the ND table must hold the
 291  *   lock as writer. ND_GET/ND_SET routines that read the ND table hold the lock
 292  *   as reader.
 293  *
 294  * - ip_addr_avail_lock: This is used to ensure the uniqueness of IP addresses.
 295  *   This lock is held in ipif_up_done and the ipif is marked IPIF_UP and the
 296  *   uniqueness check also done atomically.
 297  *
 298  * - ipsec_capab_ills_lock: This readers/writer lock protects the global
 299  *   lists of IPsec capable ills (ipsec_capab_ills_{ah,esp}). It is taken
 300  *   as a writer when adding or deleting elements from these lists, and
 301  *   as a reader when walking these lists to send a SADB update to the
 302  *   IPsec capable ills.
 303  *
 304  * - ill_g_usesrc_lock: This readers/writer lock protects the usesrc
 305  *   group list linked by ill_usesrc_grp_next. It also protects the
 306  *   ill_usesrc_ifindex field. It is taken as a writer when a member of the
 307  *   group is being added or deleted.  This lock is taken as a reader when
 308  *   walking the list/group(eg: to get the number of members in a usesrc group).
 309  *   Note, it is only necessary to take this lock if the ill_usesrc_grp_next
 310  *   field is changing state i.e from NULL to non-NULL or vice-versa. For
 311  *   example, it is not necessary to take this lock in the initial portion
 312  *   of ip_sioctl_slifusesrc or at all in ip_sioctl_groupname and
 313  *   ip_sioctl_flags since the these operations are executed exclusively and
 314  *   that ensures that the "usesrc group state" cannot change. The "usesrc
 315  *   group state" change can happen only in the latter part of
 316  *   ip_sioctl_slifusesrc and in ill_delete.
 317  *
 318  * Changing <ill-phyint>, <ill-ipsq>, <ill-illgroup> assocications.
 319  *
 320  * To change the <ill-phyint> association, the ill_g_lock must be held
 321  * as writer, and the ill_locks of both the v4 and v6 instance of the ill
 322  * must be held.
 323  *
 324  * To change the <ill-ipsq> association the ill_g_lock must be held as writer
 325  * and the ill_lock of the ill in question must be held.
 326  *
 327  * To change the <ill-illgroup> association the ill_g_lock must be held as
 328  * writer and the ill_lock of the ill in question must be held.
 329  *
 330  * To add or delete an ipif from the list of ipifs hanging off the ill,
 331  * ill_g_lock (writer) and ill_lock must be held and the thread must be
 332  * a writer on the associated ipsq,.
 333  *
 334  * To add or delete an ill to the system, the ill_g_lock must be held as
 335  * writer and the thread must be a writer on the associated ipsq.
 336  *
 337  * To add or delete an ilm to an ill, the ill_lock must be held and the thread
 338  * must be a writer on the associated ipsq.
 339  *
 340  * Lock hierarchy
 341  *
 342  * Some lock hierarchy scenarios are listed below.
 343  *
 344  * ill_g_lock -> conn_lock -> ill_lock -> ipsq_lock
 345  * ill_g_lock -> illgrp_lock -> ill_lock
 346  * ill_g_lock -> ill_lock(s) -> phyint_lock
 347  * ill_g_lock -> ndp_g_lock -> ill_lock -> nce_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  *
 352  * When more than 1 ill lock is needed to be held, all ill lock addresses
 353  * are sorted on address and locked starting from highest addressed lock
 354  * downward.
 355  *
 356  * IPsec scenarios
 357  *
 358  * ipsa_lock -> ill_g_lock -> ill_lock
 359  * ipsec_capab_ills_lock -> ill_g_lock -> ill_lock
 360  * ipsec_capab_ills_lock -> ipsa_lock
 361  * ill_g_usesrc_lock -> ill_g_lock -> ill_lock
 362  *
 363  * Trusted Solaris scenarios
 364  *
 365  * igsa_lock -> gcgrp_rwlock -> gcgrp_lock
 366  * igsa_lock -> gcdb_lock
 367  * gcgrp_rwlock -> ire_lock
 368  * gcgrp_rwlock -> gcdb_lock
 369  *
 370  *
 371  * Routing/forwarding table locking notes:
 372  *
 373  * Lock acquisition order: Radix tree lock, irb_lock.
 374  * Requirements:
 375  * i.  Walker must not hold any locks during the walker callback.
 376  * ii  Walker must not see a truncated tree during the walk because of any node
 377  *     deletion.
 378  * iii Existing code assumes ire_bucket is valid if it is non-null and is used
 379  *     in many places in the code to walk the irb list. Thus even if all the
 380  *     ires in a bucket have been deleted, we still can't free the radix node
 381  *     until the ires have actually been inactive'd (freed).
 382  *
 383  * Tree traversal - Need to hold the global tree lock in read mode.
 384  * Before dropping the global tree lock, need to either increment the ire_refcnt
 385  * to ensure that the radix node can't be deleted.
 386  *
 387  * Tree add - Need to hold the global tree lock in write mode to add a
 388  * radix node. To prevent the node from being deleted, increment the
 389  * irb_refcnt, after the node is added to the tree. The ire itself is
 390  * added later while holding the irb_lock, but not the tree lock.
 391  *
 392  * Tree delete - Need to hold the global tree lock and irb_lock in write mode.
 393  * All associated ires must be inactive (i.e. freed), and irb_refcnt
 394  * must be zero.
 395  *
 396  * Walker - Increment irb_refcnt before calling the walker callback. Hold the
 397  * global tree lock (read mode) for traversal.
 398  *
 399  * IPsec notes :
 400  *
 401  * IP interacts with the IPsec code (AH/ESP) by tagging a M_CTL message
 402  * in front of the actual packet. For outbound datagrams, the M_CTL
 403  * contains a ipsec_out_t (defined in ipsec_info.h), which has the
 404  * information used by the IPsec code for applying the right level of
 405  * protection. The information initialized by IP in the ipsec_out_t
 406  * is determined by the per-socket policy or global policy in the system.
 407  * For inbound datagrams, the M_CTL contains a ipsec_in_t (defined in
 408  * ipsec_info.h) which starts out with nothing in it. It gets filled
 409  * with the right information if it goes through the AH/ESP code, which
 410  * happens if the incoming packet is secure. The information initialized
 411  * by AH/ESP, is later used by IP(during fanouts to ULP) to see whether
 412  * the policy requirements needed by per-socket policy or global policy
 413  * is met or not.
 414  *
 415  * If there is both per-socket policy (set using setsockopt) and there
 416  * is also global policy match for the 5 tuples of the socket,
 417  * ipsec_override_policy() makes the decision of which one to use.
 418  *
 419  * For fully connected sockets i.e dst, src [addr, port] is known,
 420  * conn_policy_cached is set indicating that policy has been cached.
 421  * conn_in_enforce_policy may or may not be set depending on whether
 422  * there is a global policy match or per-socket policy match.
 423  * Policy inheriting happpens in ip_bind during the ipa_conn_t bind.
 424  * Once the right policy is set on the conn_t, policy cannot change for
 425  * this socket. This makes life simpler for TCP (UDP ?) where
 426  * re-transmissions go out with the same policy. For symmetry, policy
 427  * is cached for fully connected UDP sockets also. Thus if policy is cached,
 428  * it also implies that policy is latched i.e policy cannot change
 429  * on these sockets. As we have the right policy on the conn, we don't
 430  * have to lookup global policy for every outbound and inbound datagram
 431  * and thus serving as an optimization. Note that a global policy change
 432  * does not affect fully connected sockets if they have policy. If fully
 433  * connected sockets did not have any policy associated with it, global
 434  * policy change may affect them.
 435  *
 436  * IP Flow control notes:
 437  *
 438  * Non-TCP streams are flow controlled by IP. On the send side, if the packet
 439  * cannot be sent down to the driver by IP, because of a canput failure, IP
 440  * does a putq on the conn_wq. This will cause ip_wsrv to run on the conn_wq.
 441  * ip_wsrv in turn, inserts the conn in a list of conn's that need to be drained
 442  * when the flowcontrol condition subsides. Ultimately STREAMS backenables the
 443  * ip_wsrv on the IP module, which in turn does a qenable of the conn_wq of the
 444  * first conn in the list of conn's to be drained. ip_wsrv on this conn drains
 445  * the queued messages, and removes the conn from the drain list, if all
 446  * messages were drained. It also qenables the next conn in the drain list to
 447  * continue the drain process.
 448  *
 449  * In reality the drain list is not a single list, but a configurable number
 450  * of lists. The ip_wsrv on the IP module, qenables the first conn in each
 451  * list. If the ip_wsrv of the next qenabled conn does not run, because the
 452  * stream closes, ip_close takes responsibility to qenable the next conn in
 453  * the drain list. The directly called ip_wput path always does a putq, if
 454  * it cannot putnext. Thus synchronization problems are handled between
 455  * ip_wsrv and ip_close. conn_drain_insert and conn_drain_tail are the only
 456  * functions that manipulate this drain list. Furthermore conn_drain_insert
 457  * is called only from ip_wsrv, and there can be only 1 instance of ip_wsrv
 458  * running on a queue at any time. conn_drain_tail can be simultaneously called
 459  * from both ip_wsrv and ip_close.
 460  *
 461  * IPQOS notes:
 462  *
 463  * IPQoS Policies are applied to packets using IPPF (IP Policy framework)
 464  * and IPQoS modules. IPPF includes hooks in IP at different control points
 465  * (callout positions) which direct packets to IPQoS modules for policy
 466  * processing. Policies, if present, are global.
 467  *
 468  * The callout positions are located in the following paths:
 469  *              o local_in (packets destined for this host)
 470  *              o local_out (packets orginating from this host )
 471  *              o fwd_in  (packets forwarded by this m/c - inbound)
 472  *              o fwd_out (packets forwarded by this m/c - outbound)
 473  * Hooks at these callout points can be enabled/disabled using the ndd variable
 474  * ip_policy_mask (a bit mask with the 4 LSB indicating the callout positions).
 475  * By default all the callout positions are enabled.
 476  *
 477  * Outbound (local_out)
 478  * Hooks are placed in ip_wput_ire and ipsec_out_process.
 479  *
 480  * Inbound (local_in)
 481  * Hooks are placed in ip_proto_input, icmp_inbound, ip_fanout_proto and
 482  * TCP and UDP fanout routines.
 483  *
 484  * Forwarding (in and out)
 485  * Hooks are placed in ip_rput_forward.
 486  *
 487  * IP Policy Framework processing (IPPF processing)
 488  * Policy processing for a packet is initiated by ip_process, which ascertains
 489  * that the classifier (ipgpc) is loaded and configured, failing which the
 490  * packet resumes normal processing in IP. If the clasifier is present, the
 491  * packet is acted upon by one or more IPQoS modules (action instances), per
 492  * filters configured in ipgpc and resumes normal IP processing thereafter.
 493  * An action instance can drop a packet in course of its processing.
 494  *
 495  * A boolean variable, ip_policy, is used in all the fanout routines that can
 496  * invoke ip_process for a packet. This variable indicates if the packet should
 497  * to be sent for policy processing. The variable is set to B_TRUE by default,
 498  * i.e. when the routines are invoked in the normal ip procesing path for a
 499  * packet. The two exceptions being ip_wput_local and icmp_inbound_error_fanout;
 500  * ip_policy is set to B_FALSE for all the routines called in these two
 501  * functions because, in the former case,  we don't process loopback traffic
 502  * currently while in the latter, the packets have already been processed in
 503  * icmp_inbound.
 504  *
 505  * Zones notes:
 506  *
 507  * The partitioning rules for networking are as follows:
 508  * 1) Packets coming from a zone must have a source address belonging to that
 509  * zone.
 510  * 2) Packets coming from a zone can only be sent on a physical interface on
 511  * which the zone has an IP address.
 512  * 3) Between two zones on the same machine, packet delivery is only allowed if
 513  * there's a matching route for the destination and zone in the forwarding
 514  * table.
 515  * 4) The TCP and UDP port spaces are per-zone; that is, two processes in
 516  * different zones can bind to the same port with the wildcard address
 517  * (INADDR_ANY).
 518  *
 519  * The granularity of interface partitioning is at the logical interface level.
 520  * Therefore, every zone has its own IP addresses, and incoming packets can be
 521  * attributed to a zone unambiguously. A logical interface is placed into a zone
 522  * using the SIOCSLIFZONE ioctl; this sets the ipif_zoneid field in the ipif_t
 523  * structure. Rule (1) is implemented by modifying the source address selection
 524  * algorithm so that the list of eligible addresses is filtered based on the
 525  * sending process zone.
 526  *
 527  * The Internet Routing Entries (IREs) are either exclusive to a zone or shared
 528  * across all zones, depending on their type. Here is the break-up:
 529  *
 530  * IRE type                             Shared/exclusive
 531  * --------                             ----------------
 532  * IRE_BROADCAST                        Exclusive
 533  * IRE_DEFAULT (default routes)         Shared (*)
 534  * IRE_LOCAL                            Exclusive (x)
 535  * IRE_LOOPBACK                         Exclusive
 536  * IRE_PREFIX (net routes)              Shared (*)
 537  * IRE_CACHE                            Exclusive
 538  * IRE_IF_NORESOLVER (interface routes) Exclusive
 539  * IRE_IF_RESOLVER (interface routes)   Exclusive
 540  * IRE_HOST (host routes)               Shared (*)
 541  *
 542  * (*) A zone can only use a default or off-subnet route if the gateway is
 543  * directly reachable from the zone, that is, if the gateway's address matches
 544  * one of the zone's logical interfaces.
 545  *
 546  * (x) IRE_LOCAL are handled a bit differently, since for all other entries
 547  * in ire_ctable and IRE_INTERFACE, ire_src_addr is what can be used as source
 548  * when sending packets using the IRE. For IRE_LOCAL ire_src_addr is the IP
 549  * address of the zone itself (the destination). Since IRE_LOCAL is used
 550  * for communication between zones, ip_wput_ire has special logic to set
 551  * the right source address when sending using an IRE_LOCAL.
 552  *
 553  * Furthermore, when ip_restrict_interzone_loopback is set (the default),
 554  * ire_cache_lookup restricts loopback using an IRE_LOCAL
 555  * between zone to the case when L2 would have conceptually looped the packet
 556  * back, i.e. the loopback which is required since neither Ethernet drivers
 557  * nor Ethernet hardware loops them back. This is the case when the normal
 558  * routes (ignoring IREs with different zoneids) would send out the packet on
 559  * the same ill (or ill group) as the ill with which is IRE_LOCAL is
 560  * associated.
 561  *
 562  * Multiple zones can share a common broadcast address; typically all zones
 563  * share the 255.255.255.255 address. Incoming as well as locally originated
 564  * broadcast packets must be dispatched to all the zones on the broadcast
 565  * network. For directed broadcasts (e.g. 10.16.72.255) this is not trivial
 566  * since some zones may not be on the 10.16.72/24 network. To handle this, each
 567  * zone has its own set of IRE_BROADCAST entries; then, broadcast packets are
 568  * sent to every zone that has an IRE_BROADCAST entry for the destination
 569  * address on the input ill, see conn_wantpacket().
 570  *
 571  * Applications in different zones can join the same multicast group address.
 572  * For IPv4, group memberships are per-logical interface, so they're already
 573  * inherently part of a zone. For IPv6, group memberships are per-physical
 574  * interface, so we distinguish IPv6 group memberships based on group address,
 575  * interface and zoneid. In both cases, received multicast packets are sent to
 576  * every zone for which a group membership entry exists. On IPv6 we need to
 577  * check that the target zone still has an address on the receiving physical
 578  * interface; it could have been removed since the application issued the
 579  * IPV6_JOIN_GROUP.
 580  */
 581 
 582 /*
 583  * Squeue Fanout flags:
 584  *      0: No fanout.
 585  *      1: Fanout across all squeues
 586  */
 587 boolean_t       ip_squeue_fanout = 0;
 588 
 589 /*
 590  * Maximum dups allowed per packet.
 591  */
 592 uint_t ip_max_frag_dups = 10;
 593 
 594 #define IS_SIMPLE_IPH(ipha)                                             \
 595         ((ipha)->ipha_version_and_hdr_length == IP_SIMPLE_HDR_VERSION)
 596 
 597 /* RFC1122 Conformance */
 598 #define IP_FORWARD_DEFAULT      IP_FORWARD_NEVER
 599 
 600 #define ILL_MAX_NAMELEN                 LIFNAMSIZ
 601 
 602 static int      conn_set_held_ipif(conn_t *, ipif_t **, ipif_t *);
 603 
 604 static int      ip_open(queue_t *q, dev_t *devp, int flag, int sflag,
 605                     cred_t *credp, boolean_t isv6);
 606 static mblk_t   *ip_wput_attach_llhdr(mblk_t *, ire_t *, ip_proc_t, uint32_t,
 607                     ipha_t **);
 608 
 609 static void     icmp_frag_needed(queue_t *, mblk_t *, int, zoneid_t,
 610                     ip_stack_t *);
 611 static void     icmp_inbound(queue_t *, mblk_t *, boolean_t, ill_t *, int,
 612                     uint32_t, boolean_t, boolean_t, ill_t *, zoneid_t);
 613 static ipaddr_t icmp_get_nexthop_addr(ipha_t *, ill_t *, zoneid_t, mblk_t *mp);
 614 static boolean_t icmp_inbound_too_big(icmph_t *, ipha_t *, ill_t *, zoneid_t,
 615                     mblk_t *, int, ip_stack_t *);
 616 static void     icmp_inbound_error_fanout(queue_t *, ill_t *, mblk_t *,
 617                     icmph_t *, ipha_t *, int, int, boolean_t, boolean_t,
 618                     ill_t *, zoneid_t);
 619 static void     icmp_options_update(ipha_t *);
 620 static void     icmp_param_problem(queue_t *, mblk_t *, uint8_t, zoneid_t,
 621                     ip_stack_t *);
 622 static void     icmp_pkt(queue_t *, mblk_t *, void *, size_t, boolean_t,
 623                     zoneid_t zoneid, ip_stack_t *);
 624 static mblk_t   *icmp_pkt_err_ok(mblk_t *, ip_stack_t *);
 625 static void     icmp_redirect(ill_t *, mblk_t *);
 626 static void     icmp_send_redirect(queue_t *, mblk_t *, ipaddr_t,
 627                     ip_stack_t *);
 628 
 629 static void     ip_arp_news(queue_t *, mblk_t *);
 630 static boolean_t ip_bind_insert_ire(mblk_t *, ire_t *, iulp_t *,
 631                     ip_stack_t *);
 632 mblk_t          *ip_dlpi_alloc(size_t, t_uscalar_t);
 633 char            *ip_dot_addr(ipaddr_t, char *);
 634 mblk_t          *ip_carve_mp(mblk_t **, ssize_t);
 635 int             ip_close(queue_t *, int);
 636 static char     *ip_dot_saddr(uchar_t *, char *);
 637 static void     ip_fanout_proto(queue_t *, mblk_t *, ill_t *, ipha_t *, uint_t,
 638                     boolean_t, boolean_t, ill_t *, zoneid_t);
 639 static void     ip_fanout_tcp(queue_t *, mblk_t *, ill_t *, ipha_t *, uint_t,
 640                     boolean_t, boolean_t, zoneid_t);
 641 static void     ip_fanout_udp(queue_t *, mblk_t *, ill_t *, ipha_t *, uint32_t,
 642                     boolean_t, uint_t, boolean_t, boolean_t, ill_t *, zoneid_t);
 643 static void     ip_lrput(queue_t *, mblk_t *);
 644 ipaddr_t        ip_net_mask(ipaddr_t);
 645 void            ip_newroute(queue_t *, mblk_t *, ipaddr_t, conn_t *, zoneid_t,
 646                     ip_stack_t *);
 647 static void     ip_newroute_ipif(queue_t *, mblk_t *, ipif_t *, ipaddr_t,
 648                     conn_t *, uint32_t, zoneid_t, ip_opt_info_t *);
 649 char            *ip_nv_lookup(nv_t *, int);
 650 static boolean_t        ip_check_for_ipsec_opt(queue_t *, mblk_t *);
 651 static int      ip_param_get(queue_t *, mblk_t *, caddr_t, cred_t *);
 652 static int      ip_param_generic_get(queue_t *, mblk_t *, caddr_t, cred_t *);
 653 static boolean_t        ip_param_register(IDP *ndp, ipparam_t *, size_t,
 654     ipndp_t *, size_t);
 655 static int      ip_param_set(queue_t *, mblk_t *, char *, caddr_t, cred_t *);
 656 void    ip_rput(queue_t *, mblk_t *);
 657 static void     ip_rput_dlpi_writer(ipsq_t *dummy_sq, queue_t *q, mblk_t *mp,
 658                     void *dummy_arg);
 659 void    ip_rput_forward(ire_t *, ipha_t *, mblk_t *, ill_t *);
 660 static int      ip_rput_forward_options(mblk_t *, ipha_t *, ire_t *,
 661     ip_stack_t *);
 662 static boolean_t        ip_rput_local_options(queue_t *, mblk_t *, ipha_t *,
 663                             ire_t *, ip_stack_t *);
 664 static boolean_t        ip_rput_multimblk_ipoptions(queue_t *, ill_t *,
 665                             mblk_t *, ipha_t **, ipaddr_t *, ip_stack_t *);
 666 static int      ip_rput_options(queue_t *, mblk_t *, ipha_t *, ipaddr_t *,
 667     ip_stack_t *);
 668 static boolean_t ip_rput_fragment(queue_t *, mblk_t **, ipha_t *, uint32_t *,
 669                     uint16_t *);
 670 int             ip_snmp_get(queue_t *, mblk_t *, int);
 671 static mblk_t   *ip_snmp_get_mib2_ip(queue_t *, mblk_t *,
 672                     mib2_ipIfStatsEntry_t *, ip_stack_t *);
 673 static mblk_t   *ip_snmp_get_mib2_ip_traffic_stats(queue_t *, mblk_t *,
 674                     ip_stack_t *);
 675 static mblk_t   *ip_snmp_get_mib2_ip6(queue_t *, mblk_t *, ip_stack_t *);
 676 static mblk_t   *ip_snmp_get_mib2_icmp(queue_t *, mblk_t *, ip_stack_t *ipst);
 677 static mblk_t   *ip_snmp_get_mib2_icmp6(queue_t *, mblk_t *, ip_stack_t *ipst);
 678 static mblk_t   *ip_snmp_get_mib2_igmp(queue_t *, mblk_t *, ip_stack_t *ipst);
 679 static mblk_t   *ip_snmp_get_mib2_multi(queue_t *, mblk_t *, ip_stack_t *ipst);
 680 static mblk_t   *ip_snmp_get_mib2_ip_addr(queue_t *, mblk_t *,
 681                     ip_stack_t *ipst);
 682 static mblk_t   *ip_snmp_get_mib2_ip6_addr(queue_t *, mblk_t *,
 683                     ip_stack_t *ipst);
 684 static mblk_t   *ip_snmp_get_mib2_ip_group_src(queue_t *, mblk_t *,
 685                     ip_stack_t *ipst);
 686 static mblk_t   *ip_snmp_get_mib2_ip6_group_src(queue_t *, mblk_t *,
 687                     ip_stack_t *ipst);
 688 static mblk_t   *ip_snmp_get_mib2_ip_group_mem(queue_t *, mblk_t *,
 689                     ip_stack_t *ipst);
 690 static mblk_t   *ip_snmp_get_mib2_ip6_group_mem(queue_t *, mblk_t *,
 691                     ip_stack_t *ipst);
 692 static mblk_t   *ip_snmp_get_mib2_virt_multi(queue_t *, mblk_t *,
 693                     ip_stack_t *ipst);
 694 static mblk_t   *ip_snmp_get_mib2_multi_rtable(queue_t *, mblk_t *,
 695                     ip_stack_t *ipst);
 696 static mblk_t   *ip_snmp_get_mib2_ip_route_media(queue_t *, mblk_t *,
 697                     ip_stack_t *ipst);
 698 static mblk_t   *ip_snmp_get_mib2_ip6_route_media(queue_t *, mblk_t *,
 699                     ip_stack_t *ipst);
 700 static void     ip_snmp_get2_v4(ire_t *, iproutedata_t *);
 701 static void     ip_snmp_get2_v6_route(ire_t *, iproutedata_t *);
 702 static int      ip_snmp_get2_v6_media(nce_t *, iproutedata_t *);
 703 int             ip_snmp_set(queue_t *, int, int, uchar_t *, int);
 704 static boolean_t        ip_source_routed(ipha_t *, ip_stack_t *);
 705 static boolean_t        ip_source_route_included(ipha_t *);
 706 static void     ip_trash_ire_reclaim_stack(ip_stack_t *);
 707 
 708 static void     ip_wput_frag(ire_t *, mblk_t *, ip_pkt_t, uint32_t, uint32_t,
 709                     zoneid_t, ip_stack_t *);
 710 static mblk_t   *ip_wput_frag_copyhdr(uchar_t *, int, int, ip_stack_t *);
 711 static void     ip_wput_local_options(ipha_t *, ip_stack_t *);
 712 static int      ip_wput_options(queue_t *, mblk_t *, ipha_t *, boolean_t,
 713                     zoneid_t, ip_stack_t *);
 714 
 715 static void     conn_drain_init(ip_stack_t *);
 716 static void     conn_drain_fini(ip_stack_t *);
 717 static void     conn_drain_tail(conn_t *connp, boolean_t closing);
 718 
 719 static void     conn_walk_drain(ip_stack_t *);
 720 static void     conn_walk_fanout_table(connf_t *, uint_t, pfv_t, void *,
 721     zoneid_t);
 722 
 723 static void     *ip_stack_init(netstackid_t stackid, netstack_t *ns);
 724 static void     ip_stack_shutdown(netstackid_t stackid, void *arg);
 725 static void     ip_stack_fini(netstackid_t stackid, void *arg);
 726 
 727 static boolean_t        conn_wantpacket(conn_t *, ill_t *, ipha_t *, int,
 728     zoneid_t);
 729 static void     ip_arp_done(ipsq_t *dummy_sq, queue_t *q, mblk_t *mp,
 730     void *dummy_arg);
 731 
 732 static int      ip_forward_set(queue_t *, mblk_t *, char *, caddr_t, cred_t *);
 733 
 734 static int      ip_multirt_apply_membership(int (*fn)(conn_t *, boolean_t,
 735     ipaddr_t, ipaddr_t, uint_t *, mcast_record_t, ipaddr_t, mblk_t *), ire_t *,
 736     conn_t *, boolean_t, ipaddr_t, mcast_record_t, ipaddr_t, mblk_t *);
 737 static void     ip_multirt_bad_mtu(ire_t *, uint32_t);
 738 
 739 static int      ip_cgtp_filter_get(queue_t *, mblk_t *, caddr_t, cred_t *);
 740 static int      ip_cgtp_filter_set(queue_t *, mblk_t *, char *,
 741     caddr_t, cred_t *);
 742 extern int      ip_squeue_bind_set(queue_t *q, mblk_t *mp, char *value,
 743     caddr_t cp, cred_t *cr);
 744 extern int      ip_squeue_profile_set(queue_t *, mblk_t *, char *, caddr_t,
 745     cred_t *);
 746 static int      ip_input_proc_set(queue_t *q, mblk_t *mp, char *value,
 747     caddr_t cp, cred_t *cr);
 748 static int      ip_int_set(queue_t *, mblk_t *, char *, caddr_t,
 749     cred_t *);
 750 static int      ipmp_hook_emulation_set(queue_t *, mblk_t *, char *, caddr_t,
 751     cred_t *);
 752 static squeue_func_t ip_squeue_switch(int);
 753 
 754 static void     *ip_kstat_init(netstackid_t, ip_stack_t *);
 755 static void     ip_kstat_fini(netstackid_t, kstat_t *);
 756 static int      ip_kstat_update(kstat_t *kp, int rw);
 757 static void     *icmp_kstat_init(netstackid_t);
 758 static void     icmp_kstat_fini(netstackid_t, kstat_t *);
 759 static int      icmp_kstat_update(kstat_t *kp, int rw);
 760 static void     *ip_kstat2_init(netstackid_t, ip_stat_t *);
 761 static void     ip_kstat2_fini(netstackid_t, kstat_t *);
 762 
 763 static int      ip_conn_report(queue_t *, mblk_t *, caddr_t, cred_t *);
 764 
 765 static mblk_t   *ip_tcp_input(mblk_t *, ipha_t *, ill_t *, boolean_t,
 766     ire_t *, mblk_t *, uint_t, queue_t *, ill_rx_ring_t *);
 767 
 768 static void     ip_rput_process_forward(queue_t *, mblk_t *, ire_t *,
 769     ipha_t *, ill_t *, boolean_t);
 770 ipaddr_t        ip_g_all_ones = IP_HOST_MASK;
 771 
 772 /* How long, in seconds, we allow frags to hang around. */
 773 #define IP_FRAG_TIMEOUT 60
 774 
 775 /*
 776  * Threshold which determines whether MDT should be used when
 777  * generating IP fragments; payload size must be greater than
 778  * this threshold for MDT to take place.
 779  */
 780 #define IP_WPUT_FRAG_MDT_MIN    32768
 781 
 782 /* Setable in /etc/system only */
 783 int     ip_wput_frag_mdt_min = IP_WPUT_FRAG_MDT_MIN;
 784 
 785 static long ip_rput_pullups;
 786 int     dohwcksum = 1;  /* use h/w cksum if supported by the hardware */
 787 
 788 vmem_t *ip_minor_arena_sa; /* for minor nos. from INET_MIN_DEV+2 thru 2^^18-1 */
 789 vmem_t *ip_minor_arena_la; /* for minor nos. from 2^^18 thru 2^^32-1 */
 790 
 791 int     ip_debug;
 792 
 793 #ifdef DEBUG
 794 uint32_t ipsechw_debug = 0;
 795 #endif
 796 
 797 /*
 798  * Multirouting/CGTP stuff
 799  */
 800 int     ip_cgtp_filter_rev = CGTP_FILTER_REV;   /* CGTP hooks version */
 801 
 802 /*
 803  * XXX following really should only be in a header. Would need more
 804  * header and .c clean up first.
 805  */
 806 extern optdb_obj_t      ip_opt_obj;
 807 
 808 ulong_t ip_squeue_enter_unbound = 0;
 809 
 810 /*
 811  * Named Dispatch Parameter Table.
 812  * All of these are alterable, within the min/max values given, at run time.
 813  */
 814 static ipparam_t        lcl_param_arr[] = {
 815         /* min  max     value   name */
 816         {  0,   1,      0,      "ip_respond_to_address_mask_broadcast"},
 817         {  0,   1,      1,      "ip_respond_to_echo_broadcast"},
 818         {  0,   1,      1,      "ip_respond_to_echo_multicast"},
 819         {  0,   1,      0,      "ip_respond_to_timestamp"},
 820         {  0,   1,      0,      "ip_respond_to_timestamp_broadcast"},
 821         {  0,   1,      1,      "ip_send_redirects"},
 822         {  0,   1,      0,      "ip_forward_directed_broadcasts"},
 823         {  0,   10,     0,      "ip_mrtdebug"},
 824         {  5000, 999999999,     60000, "ip_ire_timer_interval" },
 825         {  60000, 999999999,    1200000, "ip_ire_arp_interval" },
 826         {  60000, 999999999,    60000, "ip_ire_redirect_interval" },
 827         {  1,   255,    255,    "ip_def_ttl" },
 828         {  0,   1,      0,      "ip_forward_src_routed"},
 829         {  0,   256,    32,     "ip_wroff_extra" },
 830         {  5000, 999999999, 600000, "ip_ire_pathmtu_interval" },
 831         {  8,   65536,  64,     "ip_icmp_return_data_bytes" },
 832         {  0,   1,      1,      "ip_path_mtu_discovery" },
 833         {  0,   240,    30,     "ip_ignore_delete_time" },
 834         {  0,   1,      0,      "ip_ignore_redirect" },
 835         {  0,   1,      1,      "ip_output_queue" },
 836         {  1,   254,    1,      "ip_broadcast_ttl" },
 837         {  0,   99999,  100,    "ip_icmp_err_interval" },
 838         {  1,   99999,  10,     "ip_icmp_err_burst" },
 839         {  0,   999999999,      1000000, "ip_reass_queue_bytes" },
 840         {  0,   1,      0,      "ip_strict_dst_multihoming" },
 841         {  1,   MAX_ADDRS_PER_IF,       256,    "ip_addrs_per_if"},
 842         {  0,   1,      0,      "ipsec_override_persocket_policy" },
 843         {  0,   1,      1,      "icmp_accept_clear_messages" },
 844         {  0,   1,      1,      "igmp_accept_clear_messages" },
 845         {  2,   999999999, ND_DELAY_FIRST_PROBE_TIME,
 846                                 "ip_ndp_delay_first_probe_time"},
 847         {  1,   999999999, ND_MAX_UNICAST_SOLICIT,
 848                                 "ip_ndp_max_unicast_solicit"},
 849         {  1,   255,    IPV6_MAX_HOPS,  "ip6_def_hops" },
 850         {  8,   IPV6_MIN_MTU,   IPV6_MIN_MTU, "ip6_icmp_return_data_bytes" },
 851         {  0,   1,      0,      "ip6_forward_src_routed"},
 852         {  0,   1,      1,      "ip6_respond_to_echo_multicast"},
 853         {  0,   1,      1,      "ip6_send_redirects"},
 854         {  0,   1,      0,      "ip6_ignore_redirect" },
 855         {  0,   1,      0,      "ip6_strict_dst_multihoming" },
 856 
 857         {  1,   8,      3,      "ip_ire_reclaim_fraction" },
 858 
 859         {  0,   999999, 1000,   "ipsec_policy_log_interval" },
 860 
 861         {  0,   1,      1,      "pim_accept_clear_messages" },
 862         {  1000, 20000, 2000,   "ip_ndp_unsolicit_interval" },
 863         {  1,   20,     3,      "ip_ndp_unsolicit_count" },
 864         {  0,   1,      1,      "ip6_ignore_home_address_opt" },
 865         {  0,   15,     0,      "ip_policy_mask" },
 866         {  1000, 60000, 1000,   "ip_multirt_resolution_interval" },
 867         {  0,   255,    1,      "ip_multirt_ttl" },
 868         {  0,   1,      1,      "ip_multidata_outbound" },
 869         {  0,   3600000, 300000, "ip_ndp_defense_interval" },
 870         {  0,   999999, 60*60*24, "ip_max_temp_idle" },
 871         {  0,   1000,   1,      "ip_max_temp_defend" },
 872         {  0,   1000,   3,      "ip_max_defend" },
 873         {  0,   999999, 30,     "ip_defend_interval" },
 874         {  0,   3600000, 300000, "ip_dup_recovery" },
 875         {  0,   1,      1,      "ip_restrict_interzone_loopback" },
 876         {  0,   1,      1,      "ip_lso_outbound" },
 877         {  IGMP_V1_ROUTER, IGMP_V3_ROUTER, IGMP_V3_ROUTER, "igmp_max_version" },
 878         {  MLD_V1_ROUTER, MLD_V2_ROUTER, MLD_V2_ROUTER, "mld_max_version" },
 879 #ifdef DEBUG
 880         {  0,   1,      0,      "ip6_drop_inbound_icmpv6" },
 881 #else
 882         {  0,   0,      0,      "" },
 883 #endif
 884 };
 885 
 886 /*
 887  * Extended NDP table
 888  * The addresses for the first two are filled in to be ips_ip_g_forward
 889  * and ips_ipv6_forward at init time.
 890  */
 891 static ipndp_t  lcl_ndp_arr[] = {
 892         /* getf                 setf            data                    name */
 893 #define IPNDP_IP_FORWARDING_OFFSET      0
 894         {  ip_param_generic_get,        ip_forward_set, NULL,
 895             "ip_forwarding" },
 896 #define IPNDP_IP6_FORWARDING_OFFSET     1
 897         {  ip_param_generic_get,        ip_forward_set, NULL,
 898             "ip6_forwarding" },
 899         {  ip_ill_report,       NULL,           NULL,
 900             "ip_ill_status" },
 901         {  ip_ipif_report,      NULL,           NULL,
 902             "ip_ipif_status" },
 903         {  ip_conn_report,      NULL,           NULL,
 904             "ip_conn_status" },
 905         {  nd_get_long,         nd_set_long,    (caddr_t)&ip_rput_pullups,
 906             "ip_rput_pullups" },
 907         {  ip_srcid_report,     NULL,           NULL,
 908             "ip_srcid_status" },
 909         { ip_param_generic_get, ip_squeue_profile_set,
 910             (caddr_t)&ip_squeue_profile, "ip_squeue_profile" },
 911         { ip_param_generic_get, ip_squeue_bind_set,
 912             (caddr_t)&ip_squeue_bind, "ip_squeue_bind" },
 913         { ip_param_generic_get, ip_input_proc_set,
 914             (caddr_t)&ip_squeue_enter, "ip_squeue_enter" },
 915         { ip_param_generic_get, ip_int_set,
 916             (caddr_t)&ip_squeue_fanout, "ip_squeue_fanout" },
 917 #define IPNDP_CGTP_FILTER_OFFSET        11
 918         {  ip_cgtp_filter_get,  ip_cgtp_filter_set, NULL,
 919             "ip_cgtp_filter" },
 920         { ip_param_generic_get, ip_int_set,
 921             (caddr_t)&ip_soft_rings_cnt, "ip_soft_rings_cnt" },
 922 #define IPNDP_IPMP_HOOK_OFFSET  13
 923         {  ip_param_generic_get, ipmp_hook_emulation_set, NULL,
 924             "ipmp_hook_emulation" },
 925         {  ip_param_generic_get, ip_int_set, (caddr_t)&ip_debug,
 926             "ip_debug" },
 927 };
 928 
 929 /*
 930  * Table of IP ioctls encoding the various properties of the ioctl and
 931  * indexed based on the last byte of the ioctl command. Occasionally there
 932  * is a clash, and there is more than 1 ioctl with the same last byte.
 933  * In such a case 1 ioctl is encoded in the ndx table and the remaining
 934  * ioctls are encoded in the misc table. An entry in the ndx table is
 935  * retrieved by indexing on the last byte of the ioctl command and comparing
 936  * the ioctl command with the value in the ndx table. In the event of a
 937  * mismatch the misc table is then searched sequentially for the desired
 938  * ioctl command.
 939  *
 940  * Entry: <command> <copyin_size> <flags> <cmd_type> <function> <restart_func>
 941  */
 942 ip_ioctl_cmd_t ip_ndx_ioctl_table[] = {
 943         /* 000 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 944         /* 001 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 945         /* 002 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 946         /* 003 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 947         /* 004 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 948         /* 005 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 949         /* 006 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 950         /* 007 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 951         /* 008 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 952         /* 009 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 953 
 954         /* 010 */ { SIOCADDRT,  sizeof (struct rtentry), IPI_PRIV,
 955                         MISC_CMD, ip_siocaddrt, NULL },
 956         /* 011 */ { SIOCDELRT,  sizeof (struct rtentry), IPI_PRIV,
 957                         MISC_CMD, ip_siocdelrt, NULL },
 958 
 959         /* 012 */ { SIOCSIFADDR, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
 960                         IF_CMD, ip_sioctl_addr, ip_sioctl_addr_restart },
 961         /* 013 */ { SIOCGIFADDR, sizeof (struct ifreq), IPI_GET_CMD | IPI_REPL,
 962                         IF_CMD, ip_sioctl_get_addr, NULL },
 963 
 964         /* 014 */ { SIOCSIFDSTADDR, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
 965                         IF_CMD, ip_sioctl_dstaddr, ip_sioctl_dstaddr_restart },
 966         /* 015 */ { SIOCGIFDSTADDR, sizeof (struct ifreq),
 967                         IPI_GET_CMD | IPI_REPL,
 968                         IF_CMD, ip_sioctl_get_dstaddr, NULL },
 969 
 970         /* 016 */ { SIOCSIFFLAGS, sizeof (struct ifreq),
 971                         IPI_PRIV | IPI_WR | IPI_REPL,
 972                         IF_CMD, ip_sioctl_flags, ip_sioctl_flags_restart },
 973         /* 017 */ { SIOCGIFFLAGS, sizeof (struct ifreq),
 974                         IPI_MODOK | IPI_GET_CMD | IPI_REPL,
 975                         IF_CMD, ip_sioctl_get_flags, NULL },
 976 
 977         /* 018 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 978         /* 019 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 979 
 980         /* copyin size cannot be coded for SIOCGIFCONF */
 981         /* 020 */ { O_SIOCGIFCONF, 0, IPI_GET_CMD,
 982                         MISC_CMD, ip_sioctl_get_ifconf, NULL },
 983 
 984         /* 021 */ { SIOCSIFMTU, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
 985                         IF_CMD, ip_sioctl_mtu, NULL },
 986         /* 022 */ { SIOCGIFMTU, sizeof (struct ifreq), IPI_GET_CMD | IPI_REPL,
 987                         IF_CMD, ip_sioctl_get_mtu, NULL },
 988         /* 023 */ { SIOCGIFBRDADDR, sizeof (struct ifreq),
 989                         IPI_GET_CMD | IPI_REPL,
 990                         IF_CMD, ip_sioctl_get_brdaddr, NULL },
 991         /* 024 */ { SIOCSIFBRDADDR, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
 992                         IF_CMD, ip_sioctl_brdaddr, NULL },
 993         /* 025 */ { SIOCGIFNETMASK, sizeof (struct ifreq),
 994                         IPI_GET_CMD | IPI_REPL,
 995                         IF_CMD, ip_sioctl_get_netmask, NULL },
 996         /* 026 */ { SIOCSIFNETMASK, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
 997                         IF_CMD, ip_sioctl_netmask, ip_sioctl_netmask_restart },
 998         /* 027 */ { SIOCGIFMETRIC, sizeof (struct ifreq),
 999                         IPI_GET_CMD | IPI_REPL,
1000                         IF_CMD, ip_sioctl_get_metric, NULL },
1001         /* 028 */ { SIOCSIFMETRIC, sizeof (struct ifreq), IPI_PRIV,
1002                         IF_CMD, ip_sioctl_metric, NULL },
1003         /* 029 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1004 
1005         /* See 166-168 below for extended SIOC*XARP ioctls */
1006         /* 030 */ { SIOCSARP, sizeof (struct arpreq), IPI_PRIV,
1007                         ARP_CMD, ip_sioctl_arp, NULL },
1008         /* 031 */ { SIOCGARP, sizeof (struct arpreq), IPI_GET_CMD | IPI_REPL,
1009                         ARP_CMD, ip_sioctl_arp, NULL },
1010         /* 032 */ { SIOCDARP, sizeof (struct arpreq), IPI_PRIV,
1011                         ARP_CMD, ip_sioctl_arp, NULL },
1012 
1013         /* 033 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1014         /* 034 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1015         /* 035 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1016         /* 036 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1017         /* 037 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1018         /* 038 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1019         /* 039 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1020         /* 040 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1021         /* 041 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1022         /* 042 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1023         /* 043 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1024         /* 044 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1025         /* 045 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1026         /* 046 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1027         /* 047 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1028         /* 048 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1029         /* 049 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1030         /* 050 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1031         /* 051 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1032         /* 052 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1033         /* 053 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1034 
1035         /* 054 */ { IF_UNITSEL, sizeof (int), IPI_PRIV | IPI_WR | IPI_MODOK,
1036                         MISC_CMD, if_unitsel, if_unitsel_restart },
1037 
1038         /* 055 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1039         /* 056 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1040         /* 057 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1041         /* 058 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1042         /* 059 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1043         /* 060 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1044         /* 061 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1045         /* 062 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1046         /* 063 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1047         /* 064 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1048         /* 065 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1049         /* 066 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1050         /* 067 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1051         /* 068 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1052         /* 069 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1053         /* 070 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1054         /* 071 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1055         /* 072 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1056 
1057         /* 073 */ { SIOCSIFNAME, sizeof (struct ifreq),
1058                         IPI_PRIV | IPI_WR | IPI_MODOK,
1059                         IF_CMD, ip_sioctl_sifname, NULL },
1060 
1061         /* 074 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1062         /* 075 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1063         /* 076 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1064         /* 077 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1065         /* 078 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1066         /* 079 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1067         /* 080 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1068         /* 081 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1069         /* 082 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1070         /* 083 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1071         /* 084 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1072         /* 085 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1073         /* 086 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1074 
1075         /* 087 */ { SIOCGIFNUM, sizeof (int), IPI_GET_CMD | IPI_REPL,
1076                         MISC_CMD, ip_sioctl_get_ifnum, NULL },
1077         /* 088 */ { SIOCGIFMUXID, sizeof (struct ifreq), IPI_GET_CMD | IPI_REPL,
1078                         IF_CMD, ip_sioctl_get_muxid, NULL },
1079         /* 089 */ { SIOCSIFMUXID, sizeof (struct ifreq),
1080                         IPI_PRIV | IPI_WR | IPI_REPL,
1081                         IF_CMD, ip_sioctl_muxid, NULL },
1082 
1083         /* Both if and lif variants share same func */
1084         /* 090 */ { SIOCGIFINDEX, sizeof (struct ifreq), IPI_GET_CMD | IPI_REPL,
1085                         IF_CMD, ip_sioctl_get_lifindex, NULL },
1086         /* Both if and lif variants share same func */
1087         /* 091 */ { SIOCSIFINDEX, sizeof (struct ifreq),
1088                         IPI_PRIV | IPI_WR | IPI_REPL,
1089                         IF_CMD, ip_sioctl_slifindex, NULL },
1090 
1091         /* copyin size cannot be coded for SIOCGIFCONF */
1092         /* 092 */ { SIOCGIFCONF, 0, IPI_GET_CMD,
1093                         MISC_CMD, ip_sioctl_get_ifconf, NULL },
1094         /* 093 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1095         /* 094 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1096         /* 095 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1097         /* 096 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1098         /* 097 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1099         /* 098 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1100         /* 099 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1101         /* 100 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1102         /* 101 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1103         /* 102 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1104         /* 103 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1105         /* 104 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1106         /* 105 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1107         /* 106 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1108         /* 107 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1109         /* 108 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1110         /* 109 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1111 
1112         /* 110 */ { SIOCLIFREMOVEIF, sizeof (struct lifreq),
1113                         IPI_PRIV | IPI_WR | IPI_REPL,
1114                         LIF_CMD, ip_sioctl_removeif,
1115                         ip_sioctl_removeif_restart },
1116         /* 111 */ { SIOCLIFADDIF, sizeof (struct lifreq),
1117                         IPI_GET_CMD | IPI_PRIV | IPI_WR | IPI_REPL,
1118                         LIF_CMD, ip_sioctl_addif, NULL },
1119 #define SIOCLIFADDR_NDX 112
1120         /* 112 */ { SIOCSLIFADDR, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1121                         LIF_CMD, ip_sioctl_addr, ip_sioctl_addr_restart },
1122         /* 113 */ { SIOCGLIFADDR, sizeof (struct lifreq),
1123                         IPI_GET_CMD | IPI_REPL,
1124                         LIF_CMD, ip_sioctl_get_addr, NULL },
1125         /* 114 */ { SIOCSLIFDSTADDR, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1126                         LIF_CMD, ip_sioctl_dstaddr, ip_sioctl_dstaddr_restart },
1127         /* 115 */ { SIOCGLIFDSTADDR, sizeof (struct lifreq),
1128                         IPI_GET_CMD | IPI_REPL,
1129                         LIF_CMD, ip_sioctl_get_dstaddr, NULL },
1130         /* 116 */ { SIOCSLIFFLAGS, sizeof (struct lifreq),
1131                         IPI_PRIV | IPI_WR | IPI_REPL,
1132                         LIF_CMD, ip_sioctl_flags, ip_sioctl_flags_restart },
1133         /* 117 */ { SIOCGLIFFLAGS, sizeof (struct lifreq),
1134                         IPI_GET_CMD | IPI_MODOK | IPI_REPL,
1135                         LIF_CMD, ip_sioctl_get_flags, NULL },
1136 
1137         /* 118 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1138         /* 119 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1139 
1140         /* 120 */ { O_SIOCGLIFCONF, 0, IPI_GET_CMD, MISC_CMD,
1141                         ip_sioctl_get_lifconf, NULL },
1142         /* 121 */ { SIOCSLIFMTU, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1143                         LIF_CMD, ip_sioctl_mtu, NULL },
1144         /* 122 */ { SIOCGLIFMTU, sizeof (struct lifreq), IPI_GET_CMD | IPI_REPL,
1145                         LIF_CMD, ip_sioctl_get_mtu, NULL },
1146         /* 123 */ { SIOCGLIFBRDADDR, sizeof (struct lifreq),
1147                         IPI_GET_CMD | IPI_REPL,
1148                         LIF_CMD, ip_sioctl_get_brdaddr, NULL },
1149         /* 124 */ { SIOCSLIFBRDADDR, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1150                         LIF_CMD, ip_sioctl_brdaddr, NULL },
1151         /* 125 */ { SIOCGLIFNETMASK, sizeof (struct lifreq),
1152                         IPI_GET_CMD | IPI_REPL,
1153                         LIF_CMD, ip_sioctl_get_netmask, NULL },
1154         /* 126 */ { SIOCSLIFNETMASK, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1155                         LIF_CMD, ip_sioctl_netmask, ip_sioctl_netmask_restart },
1156         /* 127 */ { SIOCGLIFMETRIC, sizeof (struct lifreq),
1157                         IPI_GET_CMD | IPI_REPL,
1158                         LIF_CMD, ip_sioctl_get_metric, NULL },
1159         /* 128 */ { SIOCSLIFMETRIC, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1160                         LIF_CMD, ip_sioctl_metric, NULL },
1161         /* 129 */ { SIOCSLIFNAME, sizeof (struct lifreq),
1162                         IPI_PRIV | IPI_WR | IPI_MODOK | IPI_REPL,
1163                         LIF_CMD, ip_sioctl_slifname,
1164                         ip_sioctl_slifname_restart },
1165 
1166         /* 130 */ { SIOCGLIFNUM, sizeof (struct lifnum), IPI_GET_CMD | IPI_REPL,
1167                         MISC_CMD, ip_sioctl_get_lifnum, NULL },
1168         /* 131 */ { SIOCGLIFMUXID, sizeof (struct lifreq),
1169                         IPI_GET_CMD | IPI_REPL,
1170                         LIF_CMD, ip_sioctl_get_muxid, NULL },
1171         /* 132 */ { SIOCSLIFMUXID, sizeof (struct lifreq),
1172                         IPI_PRIV | IPI_WR | IPI_REPL,
1173                         LIF_CMD, ip_sioctl_muxid, NULL },
1174         /* 133 */ { SIOCGLIFINDEX, sizeof (struct lifreq),
1175                         IPI_GET_CMD | IPI_REPL,
1176                         LIF_CMD, ip_sioctl_get_lifindex, 0 },
1177         /* 134 */ { SIOCSLIFINDEX, sizeof (struct lifreq),
1178                         IPI_PRIV | IPI_WR | IPI_REPL,
1179                         LIF_CMD, ip_sioctl_slifindex, 0 },
1180         /* 135 */ { SIOCSLIFTOKEN, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1181                         LIF_CMD, ip_sioctl_token, NULL },
1182         /* 136 */ { SIOCGLIFTOKEN, sizeof (struct lifreq),
1183                         IPI_GET_CMD | IPI_REPL,
1184                         LIF_CMD, ip_sioctl_get_token, NULL },
1185         /* 137 */ { SIOCSLIFSUBNET, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1186                         LIF_CMD, ip_sioctl_subnet, ip_sioctl_subnet_restart },
1187         /* 138 */ { SIOCGLIFSUBNET, sizeof (struct lifreq),
1188                         IPI_GET_CMD | IPI_REPL,
1189                         LIF_CMD, ip_sioctl_get_subnet, NULL },
1190         /* 139 */ { SIOCSLIFLNKINFO, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1191                         LIF_CMD, ip_sioctl_lnkinfo, NULL },
1192 
1193         /* 140 */ { SIOCGLIFLNKINFO, sizeof (struct lifreq),
1194                         IPI_GET_CMD | IPI_REPL,
1195                         LIF_CMD, ip_sioctl_get_lnkinfo, NULL },
1196         /* 141 */ { SIOCLIFDELND, sizeof (struct lifreq), IPI_PRIV,
1197                         LIF_CMD, ip_siocdelndp_v6, NULL },
1198         /* 142 */ { SIOCLIFGETND, sizeof (struct lifreq), IPI_GET_CMD,
1199                         LIF_CMD, ip_siocqueryndp_v6, NULL },
1200         /* 143 */ { SIOCLIFSETND, sizeof (struct lifreq), IPI_PRIV,
1201                         LIF_CMD, ip_siocsetndp_v6, NULL },
1202         /* 144 */ { SIOCTMYADDR, sizeof (struct sioc_addrreq), IPI_GET_CMD,
1203                         MISC_CMD, ip_sioctl_tmyaddr, NULL },
1204         /* 145 */ { SIOCTONLINK, sizeof (struct sioc_addrreq), IPI_GET_CMD,
1205                         MISC_CMD, ip_sioctl_tonlink, NULL },
1206         /* 146 */ { SIOCTMYSITE, sizeof (struct sioc_addrreq), 0,
1207                         MISC_CMD, ip_sioctl_tmysite, NULL },
1208         /* 147 */ { SIOCGTUNPARAM, sizeof (struct iftun_req), IPI_REPL,
1209                         TUN_CMD, ip_sioctl_tunparam, NULL },
1210         /* 148 */ { SIOCSTUNPARAM, sizeof (struct iftun_req),
1211                         IPI_PRIV | IPI_WR,
1212                         TUN_CMD, ip_sioctl_tunparam, NULL },
1213 
1214         /* IPSECioctls handled in ip_sioctl_copyin_setup itself */
1215         /* 149 */ { SIOCFIPSECONFIG, 0, IPI_PRIV, MISC_CMD, NULL, NULL },
1216         /* 150 */ { SIOCSIPSECONFIG, 0, IPI_PRIV, MISC_CMD, NULL, NULL },
1217         /* 151 */ { SIOCDIPSECONFIG, 0, IPI_PRIV, MISC_CMD, NULL, NULL },
1218         /* 152 */ { SIOCLIPSECONFIG, 0, IPI_PRIV, MISC_CMD, NULL, NULL },
1219 
1220         /* 153 */ { SIOCLIFFAILOVER, sizeof (struct lifreq),
1221                         IPI_PRIV | IPI_WR | IPI_REPL,
1222                         LIF_CMD, ip_sioctl_move, ip_sioctl_move },
1223         /* 154 */ { SIOCLIFFAILBACK, sizeof (struct lifreq),
1224                         IPI_PRIV | IPI_WR | IPI_REPL,
1225                         LIF_CMD, ip_sioctl_move, ip_sioctl_move },
1226         /* 155 */ { SIOCSLIFGROUPNAME, sizeof (struct lifreq),
1227                         IPI_PRIV | IPI_WR | IPI_REPL,
1228                         LIF_CMD, ip_sioctl_groupname, ip_sioctl_groupname },
1229         /* 156 */ { SIOCGLIFGROUPNAME, sizeof (struct lifreq),
1230                         IPI_GET_CMD | IPI_REPL,
1231                         LIF_CMD, ip_sioctl_get_groupname, NULL },
1232         /* 157 */ { SIOCGLIFOINDEX, sizeof (struct lifreq),
1233                         IPI_GET_CMD | IPI_REPL,
1234                         LIF_CMD, ip_sioctl_get_oindex, NULL },
1235 
1236         /* Leave 158-160 unused; used to be SIOC*IFARP ioctls */
1237         /* 158 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1238         /* 159 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1239         /* 160 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1240 
1241         /* 161 */ { SIOCSLIFOINDEX, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1242                     LIF_CMD, ip_sioctl_slifoindex, NULL },
1243 
1244         /* These are handled in ip_sioctl_copyin_setup itself */
1245         /* 162 */ { SIOCGIP6ADDRPOLICY, 0, IPI_NULL_BCONT,
1246                         MISC_CMD, NULL, NULL },
1247         /* 163 */ { SIOCSIP6ADDRPOLICY, 0, IPI_PRIV | IPI_NULL_BCONT,
1248                         MISC_CMD, NULL, NULL },
1249         /* 164 */ { SIOCGDSTINFO, 0, IPI_GET_CMD, MISC_CMD, NULL, NULL },
1250 
1251         /* 165 */ { SIOCGLIFCONF, 0, IPI_GET_CMD, MISC_CMD,
1252                         ip_sioctl_get_lifconf, NULL },
1253 
1254         /* 166 */ { SIOCSXARP, sizeof (struct xarpreq), IPI_PRIV,
1255                         XARP_CMD, ip_sioctl_arp, NULL },
1256         /* 167 */ { SIOCGXARP, sizeof (struct xarpreq), IPI_GET_CMD | IPI_REPL,
1257                         XARP_CMD, ip_sioctl_arp, NULL },
1258         /* 168 */ { SIOCDXARP, sizeof (struct xarpreq), IPI_PRIV,
1259                         XARP_CMD, ip_sioctl_arp, NULL },
1260 
1261         /* SIOCPOPSOCKFS is not handled by IP */
1262         /* 169 */ { IPI_DONTCARE /* SIOCPOPSOCKFS */, 0, 0, 0, NULL, NULL },
1263 
1264         /* 170 */ { SIOCGLIFZONE, sizeof (struct lifreq),
1265                         IPI_GET_CMD | IPI_REPL,
1266                         LIF_CMD, ip_sioctl_get_lifzone, NULL },
1267         /* 171 */ { SIOCSLIFZONE, sizeof (struct lifreq),
1268                         IPI_PRIV | IPI_WR | IPI_REPL,
1269                         LIF_CMD, ip_sioctl_slifzone,
1270                         ip_sioctl_slifzone_restart },
1271         /* 172-174 are SCTP ioctls and not handled by IP */
1272         /* 172 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1273         /* 173 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1274         /* 174 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1275         /* 175 */ { SIOCGLIFUSESRC, sizeof (struct lifreq),
1276                         IPI_GET_CMD, LIF_CMD,
1277                         ip_sioctl_get_lifusesrc, 0 },
1278         /* 176 */ { SIOCSLIFUSESRC, sizeof (struct lifreq),
1279                         IPI_PRIV | IPI_WR,
1280                         LIF_CMD, ip_sioctl_slifusesrc,
1281                         NULL },
1282         /* 177 */ { SIOCGLIFSRCOF, 0, IPI_GET_CMD, MISC_CMD,
1283                         ip_sioctl_get_lifsrcof, NULL },
1284         /* 178 */ { SIOCGMSFILTER, sizeof (struct group_filter), IPI_GET_CMD,
1285                         MSFILT_CMD, ip_sioctl_msfilter, NULL },
1286         /* 179 */ { SIOCSMSFILTER, sizeof (struct group_filter), IPI_WR,
1287                         MSFILT_CMD, ip_sioctl_msfilter, NULL },
1288         /* 180 */ { SIOCGIPMSFILTER, sizeof (struct ip_msfilter), IPI_GET_CMD,
1289                         MSFILT_CMD, ip_sioctl_msfilter, NULL },
1290         /* 181 */ { SIOCSIPMSFILTER, sizeof (struct ip_msfilter), IPI_WR,
1291                         MSFILT_CMD, ip_sioctl_msfilter, NULL },
1292         /* 182 */ { SIOCSIPMPFAILBACK, sizeof (int), IPI_PRIV, MISC_CMD,
1293                         ip_sioctl_set_ipmpfailback, NULL },
1294         /* SIOCSENABLESDP is handled by SDP */
1295         /* 183 */ { IPI_DONTCARE /* SIOCSENABLESDP */, 0, 0, 0, NULL, NULL },
1296 };
1297 
1298 int ip_ndx_ioctl_count = sizeof (ip_ndx_ioctl_table) / sizeof (ip_ioctl_cmd_t);
1299 
1300 ip_ioctl_cmd_t ip_misc_ioctl_table[] = {
1301         { OSIOCGTUNPARAM, sizeof (struct old_iftun_req),
1302                 IPI_GET_CMD | IPI_REPL, TUN_CMD, ip_sioctl_tunparam, NULL },
1303         { OSIOCSTUNPARAM, sizeof (struct old_iftun_req), IPI_PRIV | IPI_WR,
1304                 TUN_CMD, ip_sioctl_tunparam, NULL },
1305         { I_LINK,       0, IPI_PRIV | IPI_WR | IPI_PASS_DOWN, 0, NULL, NULL },
1306         { I_UNLINK,     0, IPI_PRIV | IPI_WR | IPI_PASS_DOWN, 0, NULL, NULL },
1307         { I_PLINK,      0, IPI_PRIV | IPI_WR | IPI_PASS_DOWN, 0, NULL, NULL },
1308         { I_PUNLINK,    0, IPI_PRIV | IPI_WR | IPI_PASS_DOWN, 0, NULL, NULL },
1309         { ND_GET,       0, IPI_PASS_DOWN, 0, NULL, NULL },
1310         { ND_SET,       0, IPI_PRIV | IPI_WR | IPI_PASS_DOWN, 0, NULL, NULL },
1311         { IP_IOCTL,     0, 0, 0, NULL, NULL },
1312         { SIOCGETVIFCNT, sizeof (struct sioc_vif_req), IPI_REPL | IPI_GET_CMD,
1313                 MISC_CMD, mrt_ioctl},
1314         { SIOCGETSGCNT, sizeof (struct sioc_sg_req), IPI_REPL | IPI_GET_CMD,
1315                 MISC_CMD, mrt_ioctl},
1316         { SIOCGETLSGCNT, sizeof (struct sioc_lsg_req), IPI_REPL | IPI_GET_CMD,
1317                 MISC_CMD, mrt_ioctl}
1318 };
1319 
1320 int ip_misc_ioctl_count =
1321     sizeof (ip_misc_ioctl_table) / sizeof (ip_ioctl_cmd_t);
1322 
1323 int     conn_drain_nthreads;            /* Number of drainers reqd. */
1324                                         /* Settable in /etc/system */
1325 /* Defined in ip_ire.c */
1326 extern uint32_t ip_ire_max_bucket_cnt, ip6_ire_max_bucket_cnt;
1327 extern uint32_t ip_ire_min_bucket_cnt, ip6_ire_min_bucket_cnt;
1328 extern uint32_t ip_ire_mem_ratio, ip_ire_cpu_ratio;
1329 
1330 static nv_t     ire_nv_arr[] = {
1331         { IRE_BROADCAST, "BROADCAST" },
1332         { IRE_LOCAL, "LOCAL" },
1333         { IRE_LOOPBACK, "LOOPBACK" },
1334         { IRE_CACHE, "CACHE" },
1335         { IRE_DEFAULT, "DEFAULT" },
1336         { IRE_PREFIX, "PREFIX" },
1337         { IRE_IF_NORESOLVER, "IF_NORESOL" },
1338         { IRE_IF_RESOLVER, "IF_RESOLV" },
1339         { IRE_HOST, "HOST" },
1340         { 0 }
1341 };
1342 
1343 nv_t    *ire_nv_tbl = ire_nv_arr;
1344 
1345 /* Simple ICMP IP Header Template */
1346 static ipha_t icmp_ipha = {
1347         IP_SIMPLE_HDR_VERSION, 0, 0, 0, 0, 0, IPPROTO_ICMP
1348 };
1349 
1350 struct module_info ip_mod_info = {
1351         IP_MOD_ID, IP_MOD_NAME, 1, INFPSZ, 65536, 1024
1352 };
1353 
1354 /*
1355  * Duplicate static symbols within a module confuses mdb; so we avoid the
1356  * problem by making the symbols here distinct from those in udp.c.
1357  */
1358 
1359 /*
1360  * Entry points for IP as a device and as a module.
1361  * FIXME: down the road we might want a separate module and driver qinit.
1362  * We have separate open functions for the /dev/ip and /dev/ip6 devices.
1363  */
1364 static struct qinit iprinitv4 = {
1365         (pfi_t)ip_rput, NULL, ip_openv4, ip_close, NULL,
1366         &ip_mod_info
1367 };
1368 
1369 struct qinit iprinitv6 = {
1370         (pfi_t)ip_rput_v6, NULL, ip_openv6, ip_close, NULL,
1371         &ip_mod_info
1372 };
1373 
1374 static struct qinit ipwinitv4 = {
1375         (pfi_t)ip_wput, (pfi_t)ip_wsrv, NULL, NULL, NULL,
1376         &ip_mod_info
1377 };
1378 
1379 struct qinit ipwinitv6 = {
1380         (pfi_t)ip_wput_v6, (pfi_t)ip_wsrv, NULL, NULL, NULL,
1381         &ip_mod_info
1382 };
1383 
1384 static struct qinit iplrinit = {
1385         (pfi_t)ip_lrput, NULL, ip_openv4, ip_close, NULL,
1386         &ip_mod_info
1387 };
1388 
1389 static struct qinit iplwinit = {
1390         (pfi_t)ip_lwput, NULL, NULL, NULL, NULL,
1391         &ip_mod_info
1392 };
1393 
1394 /* For AF_INET aka /dev/ip */
1395 struct streamtab ipinfov4 = {
1396         &iprinitv4, &ipwinitv4, &iplrinit, &iplwinit
1397 };
1398 
1399 /* For AF_INET6 aka /dev/ip6 */
1400 struct streamtab ipinfov6 = {
1401         &iprinitv6, &ipwinitv6, &iplrinit, &iplwinit
1402 };
1403 
1404 #ifdef  DEBUG
1405 static boolean_t skip_sctp_cksum = B_FALSE;
1406 #endif
1407 
1408 /*
1409  * Prepend the zoneid using an ipsec_out_t for later use by functions like
1410  * ip_rput_v6(), ip_output(), etc.  If the message
1411  * block already has a M_CTL at the front of it, then simply set the zoneid
1412  * appropriately.
1413  */
1414 mblk_t *
1415 ip_prepend_zoneid(mblk_t *mp, zoneid_t zoneid, ip_stack_t *ipst)
1416 {
1417         mblk_t          *first_mp;
1418         ipsec_out_t     *io;
1419 
1420         ASSERT(zoneid != ALL_ZONES);
1421         if (mp->b_datap->db_type == M_CTL) {
1422                 io = (ipsec_out_t *)mp->b_rptr;
1423                 ASSERT(io->ipsec_out_type == IPSEC_OUT);
1424                 io->ipsec_out_zoneid = zoneid;
1425                 return (mp);
1426         }
1427 
1428         first_mp = ipsec_alloc_ipsec_out(ipst->ips_netstack);
1429         if (first_mp == NULL)
1430                 return (NULL);
1431         io = (ipsec_out_t *)first_mp->b_rptr;
1432         /* This is not a secure packet */
1433         io->ipsec_out_secure = B_FALSE;
1434         io->ipsec_out_zoneid = zoneid;
1435         first_mp->b_cont = mp;
1436         return (first_mp);
1437 }
1438 
1439 /*
1440  * Copy an M_CTL-tagged message, preserving reference counts appropriately.
1441  */
1442 mblk_t *
1443 ip_copymsg(mblk_t *mp)
1444 {
1445         mblk_t *nmp;
1446         ipsec_info_t *in;
1447 
1448         if (mp->b_datap->db_type != M_CTL)
1449                 return (copymsg(mp));
1450 
1451         in = (ipsec_info_t *)mp->b_rptr;
1452 
1453         /*
1454          * Note that M_CTL is also used for delivering ICMP error messages
1455          * upstream to transport layers.
1456          */
1457         if (in->ipsec_info_type != IPSEC_OUT &&
1458             in->ipsec_info_type != IPSEC_IN)
1459                 return (copymsg(mp));
1460 
1461         nmp = copymsg(mp->b_cont);
1462 
1463         if (in->ipsec_info_type == IPSEC_OUT) {
1464                 return (ipsec_out_tag(mp, nmp,
1465                     ((ipsec_out_t *)in)->ipsec_out_ns));
1466         } else {
1467                 return (ipsec_in_tag(mp, nmp,
1468                     ((ipsec_in_t *)in)->ipsec_in_ns));
1469         }
1470 }
1471 
1472 /* Generate an ICMP fragmentation needed message. */
1473 static void
1474 icmp_frag_needed(queue_t *q, mblk_t *mp, int mtu, zoneid_t zoneid,
1475     ip_stack_t *ipst)
1476 {
1477         icmph_t icmph;
1478         mblk_t *first_mp;
1479         boolean_t mctl_present;
1480 
1481         EXTRACT_PKT_MP(mp, first_mp, mctl_present);
1482 
1483         if (!(mp = icmp_pkt_err_ok(mp, ipst))) {
1484                 if (mctl_present)
1485                         freeb(first_mp);
1486                 return;
1487         }
1488 
1489         bzero(&icmph, sizeof (icmph_t));
1490         icmph.icmph_type = ICMP_DEST_UNREACHABLE;
1491         icmph.icmph_code = ICMP_FRAGMENTATION_NEEDED;
1492         icmph.icmph_du_mtu = htons((uint16_t)mtu);
1493         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutFragNeeded);
1494         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDestUnreachs);
1495         icmp_pkt(q, first_mp, &icmph, sizeof (icmph_t), mctl_present, zoneid,
1496             ipst);
1497 }
1498 
1499 /*
1500  * icmp_inbound deals with ICMP messages in the following ways.
1501  *
1502  * 1) It needs to send a reply back and possibly delivering it
1503  *    to the "interested" upper clients.
1504  * 2) It needs to send it to the upper clients only.
1505  * 3) It needs to change some values in IP only.
1506  * 4) It needs to change some values in IP and upper layers e.g TCP.
1507  *
1508  * We need to accomodate icmp messages coming in clear until we get
1509  * everything secure from the wire. If icmp_accept_clear_messages
1510  * is zero we check with the global policy and act accordingly. If
1511  * it is non-zero, we accept the message without any checks. But
1512  * *this does not mean* that this will be delivered to the upper
1513  * clients. By accepting we might send replies back, change our MTU
1514  * value etc. but delivery to the ULP/clients depends on their policy
1515  * dispositions.
1516  *
1517  * We handle the above 4 cases in the context of IPsec in the
1518  * following way :
1519  *
1520  * 1) Send the reply back in the same way as the request came in.
1521  *    If it came in encrypted, it goes out encrypted. If it came in
1522  *    clear, it goes out in clear. Thus, this will prevent chosen
1523  *    plain text attack.
1524  * 2) The client may or may not expect things to come in secure.
1525  *    If it comes in secure, the policy constraints are checked
1526  *    before delivering it to the upper layers. If it comes in
1527  *    clear, ipsec_inbound_accept_clear will decide whether to
1528  *    accept this in clear or not. In both the cases, if the returned
1529  *    message (IP header + 8 bytes) that caused the icmp message has
1530  *    AH/ESP headers, it is sent up to AH/ESP for validation before
1531  *    sending up. If there are only 8 bytes of returned message, then
1532  *    upper client will not be notified.
1533  * 3) Check with global policy to see whether it matches the constaints.
1534  *    But this will be done only if icmp_accept_messages_in_clear is
1535  *    zero.
1536  * 4) If we need to change both in IP and ULP, then the decision taken
1537  *    while affecting the values in IP and while delivering up to TCP
1538  *    should be the same.
1539  *
1540  *      There are two cases.
1541  *
1542  *      a) If we reject data at the IP layer (ipsec_check_global_policy()
1543  *         failed), we will not deliver it to the ULP, even though they
1544  *         are *willing* to accept in *clear*. This is fine as our global
1545  *         disposition to icmp messages asks us reject the datagram.
1546  *
1547  *      b) If we accept data at the IP layer (ipsec_check_global_policy()
1548  *         succeeded or icmp_accept_messages_in_clear is 1), and not able
1549  *         to deliver it to ULP (policy failed), it can lead to
1550  *         consistency problems. The cases known at this time are
1551  *         ICMP_DESTINATION_UNREACHABLE  messages with following code
1552  *         values :
1553  *
1554  *         - ICMP_FRAGMENTATION_NEEDED : IP adapts to the new value
1555  *           and Upper layer rejects. Then the communication will
1556  *           come to a stop. This is solved by making similar decisions
1557  *           at both levels. Currently, when we are unable to deliver
1558  *           to the Upper Layer (due to policy failures) while IP has
1559  *           adjusted ire_max_frag, the next outbound datagram would
1560  *           generate a local ICMP_FRAGMENTATION_NEEDED message - which
1561  *           will be with the right level of protection. Thus the right
1562  *           value will be communicated even if we are not able to
1563  *           communicate when we get from the wire initially. But this
1564  *           assumes there would be at least one outbound datagram after
1565  *           IP has adjusted its ire_max_frag value. To make things
1566  *           simpler, we accept in clear after the validation of
1567  *           AH/ESP headers.
1568  *
1569  *         - Other ICMP ERRORS : We may not be able to deliver it to the
1570  *           upper layer depending on the level of protection the upper
1571  *           layer expects and the disposition in ipsec_inbound_accept_clear().
1572  *           ipsec_inbound_accept_clear() decides whether a given ICMP error
1573  *           should be accepted in clear when the Upper layer expects secure.
1574  *           Thus the communication may get aborted by some bad ICMP
1575  *           packets.
1576  *
1577  * IPQoS Notes:
1578  * The only instance when a packet is sent for processing is when there
1579  * isn't an ICMP client and if we are interested in it.
1580  * If there is a client, IPPF processing will take place in the
1581  * ip_fanout_proto routine.
1582  *
1583  * Zones notes:
1584  * The packet is only processed in the context of the specified zone: typically
1585  * only this zone will reply to an echo request, and only interested clients in
1586  * this zone will receive a copy of the packet. This means that the caller must
1587  * call icmp_inbound() for each relevant zone.
1588  */
1589 static void
1590 icmp_inbound(queue_t *q, mblk_t *mp, boolean_t broadcast, ill_t *ill,
1591     int sum_valid, uint32_t sum, boolean_t mctl_present, boolean_t ip_policy,
1592     ill_t *recv_ill, zoneid_t zoneid)
1593 {
1594         icmph_t *icmph;
1595         ipha_t  *ipha;
1596         int     iph_hdr_length;
1597         int     hdr_length;
1598         boolean_t       interested;
1599         uint32_t        ts;
1600         uchar_t *wptr;
1601         ipif_t  *ipif;
1602         mblk_t *first_mp;
1603         ipsec_in_t *ii;
1604         ire_t *src_ire;
1605         boolean_t onlink;
1606         timestruc_t now;
1607         uint32_t ill_index;
1608         ip_stack_t *ipst;
1609 
1610         ASSERT(ill != NULL);
1611         ipst = ill->ill_ipst;
1612 
1613         first_mp = mp;
1614         if (mctl_present) {
1615                 mp = first_mp->b_cont;
1616                 ASSERT(mp != NULL);
1617         }
1618 
1619         ipha = (ipha_t *)mp->b_rptr;
1620         if (ipst->ips_icmp_accept_clear_messages == 0) {
1621                 first_mp = ipsec_check_global_policy(first_mp, NULL,
1622                     ipha, NULL, mctl_present, ipst->ips_netstack);
1623                 if (first_mp == NULL)
1624                         return;
1625         }
1626 
1627         /*
1628          * On a labeled system, we have to check whether the zone itself is
1629          * permitted to receive raw traffic.
1630          */
1631         if (is_system_labeled()) {
1632                 if (zoneid == ALL_ZONES)
1633                         zoneid = tsol_packet_to_zoneid(mp);
1634                 if (!tsol_can_accept_raw(mp, B_FALSE)) {
1635                         ip1dbg(("icmp_inbound: zone %d can't receive raw",
1636                             zoneid));
1637                         BUMP_MIB(&ipst->ips_icmp_mib, icmpInErrors);
1638                         freemsg(first_mp);
1639                         return;
1640                 }
1641         }
1642 
1643         /*
1644          * We have accepted the ICMP message. It means that we will
1645          * respond to the packet if needed. It may not be delivered
1646          * to the upper client depending on the policy constraints
1647          * and the disposition in ipsec_inbound_accept_clear.
1648          */
1649 
1650         ASSERT(ill != NULL);
1651 
1652         BUMP_MIB(&ipst->ips_icmp_mib, icmpInMsgs);
1653         iph_hdr_length = IPH_HDR_LENGTH(ipha);
1654         if ((mp->b_wptr - mp->b_rptr) < (iph_hdr_length + ICMPH_SIZE)) {
1655                 /* Last chance to get real. */
1656                 if (!pullupmsg(mp, iph_hdr_length + ICMPH_SIZE)) {
1657                         BUMP_MIB(&ipst->ips_icmp_mib, icmpInErrors);
1658                         freemsg(first_mp);
1659                         return;
1660                 }
1661                 /* Refresh iph following the pullup. */
1662                 ipha = (ipha_t *)mp->b_rptr;
1663         }
1664         /* ICMP header checksum, including checksum field, should be zero. */
1665         if (sum_valid ? (sum != 0 && sum != 0xFFFF) :
1666             IP_CSUM(mp, iph_hdr_length, 0)) {
1667                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInCksumErrs);
1668                 freemsg(first_mp);
1669                 return;
1670         }
1671         /* The IP header will always be a multiple of four bytes */
1672         icmph = (icmph_t *)&mp->b_rptr[iph_hdr_length];
1673         ip2dbg(("icmp_inbound: type %d code %d\n", icmph->icmph_type,
1674             icmph->icmph_code));
1675         wptr = (uchar_t *)icmph + ICMPH_SIZE;
1676         /* We will set "interested" to "true" if we want a copy */
1677         interested = B_FALSE;
1678         switch (icmph->icmph_type) {
1679         case ICMP_ECHO_REPLY:
1680                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInEchoReps);
1681                 break;
1682         case ICMP_DEST_UNREACHABLE:
1683                 if (icmph->icmph_code == ICMP_FRAGMENTATION_NEEDED)
1684                         BUMP_MIB(&ipst->ips_icmp_mib, icmpInFragNeeded);
1685                 interested = B_TRUE;    /* Pass up to transport */
1686                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInDestUnreachs);
1687                 break;
1688         case ICMP_SOURCE_QUENCH:
1689                 interested = B_TRUE;    /* Pass up to transport */
1690                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInSrcQuenchs);
1691                 break;
1692         case ICMP_REDIRECT:
1693                 if (!ipst->ips_ip_ignore_redirect)
1694                         interested = B_TRUE;
1695                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInRedirects);
1696                 break;
1697         case ICMP_ECHO_REQUEST:
1698                 /*
1699                  * Whether to respond to echo requests that come in as IP
1700                  * broadcasts or as IP multicast is subject to debate
1701                  * (what isn't?).  We aim to please, you pick it.
1702                  * Default is do it.
1703                  */
1704                 if (!broadcast && !CLASSD(ipha->ipha_dst)) {
1705                         /* unicast: always respond */
1706                         interested = B_TRUE;
1707                 } else if (CLASSD(ipha->ipha_dst)) {
1708                         /* multicast: respond based on tunable */
1709                         interested = ipst->ips_ip_g_resp_to_echo_mcast;
1710                 } else if (broadcast) {
1711                         /* broadcast: respond based on tunable */
1712                         interested = ipst->ips_ip_g_resp_to_echo_bcast;
1713                 }
1714                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInEchos);
1715                 break;
1716         case ICMP_ROUTER_ADVERTISEMENT:
1717         case ICMP_ROUTER_SOLICITATION:
1718                 break;
1719         case ICMP_TIME_EXCEEDED:
1720                 interested = B_TRUE;    /* Pass up to transport */
1721                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInTimeExcds);
1722                 break;
1723         case ICMP_PARAM_PROBLEM:
1724                 interested = B_TRUE;    /* Pass up to transport */
1725                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInParmProbs);
1726                 break;
1727         case ICMP_TIME_STAMP_REQUEST:
1728                 /* Response to Time Stamp Requests is local policy. */
1729                 if (ipst->ips_ip_g_resp_to_timestamp &&
1730                     /* So is whether to respond if it was an IP broadcast. */
1731                     (!broadcast || ipst->ips_ip_g_resp_to_timestamp_bcast)) {
1732                         int tstamp_len = 3 * sizeof (uint32_t);
1733 
1734                         if (wptr +  tstamp_len > mp->b_wptr) {
1735                                 if (!pullupmsg(mp, wptr + tstamp_len -
1736                                     mp->b_rptr)) {
1737                                         BUMP_MIB(ill->ill_ip_mib,
1738                                             ipIfStatsInDiscards);
1739                                         freemsg(first_mp);
1740                                         return;
1741                                 }
1742                                 /* Refresh ipha following the pullup. */
1743                                 ipha = (ipha_t *)mp->b_rptr;
1744                                 icmph = (icmph_t *)&mp->b_rptr[iph_hdr_length];
1745                                 wptr = (uchar_t *)icmph + ICMPH_SIZE;
1746                         }
1747                         interested = B_TRUE;
1748                 }
1749                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInTimestamps);
1750                 break;
1751         case ICMP_TIME_STAMP_REPLY:
1752                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInTimestampReps);
1753                 break;
1754         case ICMP_INFO_REQUEST:
1755                 /* Per RFC 1122 3.2.2.7, ignore this. */
1756         case ICMP_INFO_REPLY:
1757                 break;
1758         case ICMP_ADDRESS_MASK_REQUEST:
1759                 if ((ipst->ips_ip_respond_to_address_mask_broadcast ||
1760                     !broadcast) &&
1761                     /* TODO m_pullup of complete header? */
1762                     (mp->b_datap->db_lim - wptr) >= IP_ADDR_LEN) {
1763                         interested = B_TRUE;
1764                 }
1765                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInAddrMasks);
1766                 break;
1767         case ICMP_ADDRESS_MASK_REPLY:
1768                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInAddrMaskReps);
1769                 break;
1770         default:
1771                 interested = B_TRUE;    /* Pass up to transport */
1772                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInUnknowns);
1773                 break;
1774         }
1775         /* See if there is an ICMP client. */
1776         if (ipst->ips_ipcl_proto_fanout[IPPROTO_ICMP].connf_head != NULL) {
1777                 /* If there is an ICMP client and we want one too, copy it. */
1778                 mblk_t *first_mp1;
1779 
1780                 if (!interested) {
1781                         ip_fanout_proto(q, first_mp, ill, ipha, 0, mctl_present,
1782                             ip_policy, recv_ill, zoneid);
1783                         return;
1784                 }
1785                 first_mp1 = ip_copymsg(first_mp);
1786                 if (first_mp1 != NULL) {
1787                         ip_fanout_proto(q, first_mp1, ill, ipha,
1788                             0, mctl_present, ip_policy, recv_ill, zoneid);
1789                 }
1790         } else if (!interested) {
1791                 freemsg(first_mp);
1792                 return;
1793         } else {
1794                 /*
1795                  * Initiate policy processing for this packet if ip_policy
1796                  * is true.
1797                  */
1798                 if (IPP_ENABLED(IPP_LOCAL_IN, ipst) && ip_policy) {
1799                         ill_index = ill->ill_phyint->phyint_ifindex;
1800                         ip_process(IPP_LOCAL_IN, &mp, ill_index);
1801                         if (mp == NULL) {
1802                                 if (mctl_present) {
1803                                         freeb(first_mp);
1804                                 }
1805                                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInErrors);
1806                                 return;
1807                         }
1808                 }
1809         }
1810         /* We want to do something with it. */
1811         /* Check db_ref to make sure we can modify the packet. */
1812         if (mp->b_datap->db_ref > 1) {
1813                 mblk_t  *first_mp1;
1814 
1815                 first_mp1 = ip_copymsg(first_mp);
1816                 freemsg(first_mp);
1817                 if (!first_mp1) {
1818                         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
1819                         return;
1820                 }
1821                 first_mp = first_mp1;
1822                 if (mctl_present) {
1823                         mp = first_mp->b_cont;
1824                         ASSERT(mp != NULL);
1825                 } else {
1826                         mp = first_mp;
1827                 }
1828                 ipha = (ipha_t *)mp->b_rptr;
1829                 icmph = (icmph_t *)&mp->b_rptr[iph_hdr_length];
1830                 wptr = (uchar_t *)icmph + ICMPH_SIZE;
1831         }
1832         switch (icmph->icmph_type) {
1833         case ICMP_ADDRESS_MASK_REQUEST:
1834                 ipif = ipif_lookup_remote(ill, ipha->ipha_src, zoneid);
1835                 if (ipif == NULL) {
1836                         freemsg(first_mp);
1837                         return;
1838                 }
1839                 /*
1840                  * outging interface must be IPv4
1841                  */
1842                 ASSERT(ipif != NULL && !ipif->ipif_isv6);
1843                 icmph->icmph_type = ICMP_ADDRESS_MASK_REPLY;
1844                 bcopy(&ipif->ipif_net_mask, wptr, IP_ADDR_LEN);
1845                 ipif_refrele(ipif);
1846                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutAddrMaskReps);
1847                 break;
1848         case ICMP_ECHO_REQUEST:
1849                 icmph->icmph_type = ICMP_ECHO_REPLY;
1850                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutEchoReps);
1851                 break;
1852         case ICMP_TIME_STAMP_REQUEST: {
1853                 uint32_t *tsp;
1854 
1855                 icmph->icmph_type = ICMP_TIME_STAMP_REPLY;
1856                 tsp = (uint32_t *)wptr;
1857                 tsp++;          /* Skip past 'originate time' */
1858                 /* Compute # of milliseconds since midnight */
1859                 gethrestime(&now);
1860                 ts = (now.tv_sec % (24 * 60 * 60)) * 1000 +
1861                     now.tv_nsec / (NANOSEC / MILLISEC);
1862                 *tsp++ = htonl(ts);     /* Lay in 'receive time' */
1863                 *tsp++ = htonl(ts);     /* Lay in 'send time' */
1864                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutTimestampReps);
1865                 break;
1866         }
1867         default:
1868                 ipha = (ipha_t *)&icmph[1];
1869                 if ((uchar_t *)&ipha[1] > mp->b_wptr) {
1870                         if (!pullupmsg(mp, (uchar_t *)&ipha[1] - mp->b_rptr)) {
1871                                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1872                                 freemsg(first_mp);
1873                                 return;
1874                         }
1875                         icmph = (icmph_t *)&mp->b_rptr[iph_hdr_length];
1876                         ipha = (ipha_t *)&icmph[1];
1877                 }
1878                 if ((IPH_HDR_VERSION(ipha) != IPV4_VERSION)) {
1879                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1880                         freemsg(first_mp);
1881                         return;
1882                 }
1883                 hdr_length = IPH_HDR_LENGTH(ipha);
1884                 if (hdr_length < sizeof (ipha_t)) {
1885                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1886                         freemsg(first_mp);
1887                         return;
1888                 }
1889                 if ((uchar_t *)ipha + hdr_length > mp->b_wptr) {
1890                         if (!pullupmsg(mp,
1891                             (uchar_t *)ipha + hdr_length - mp->b_rptr)) {
1892                                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1893                                 freemsg(first_mp);
1894                                 return;
1895                         }
1896                         icmph = (icmph_t *)&mp->b_rptr[iph_hdr_length];
1897                         ipha = (ipha_t *)&icmph[1];
1898                 }
1899                 switch (icmph->icmph_type) {
1900                 case ICMP_REDIRECT:
1901                         /*
1902                          * As there is no upper client to deliver, we don't
1903                          * need the first_mp any more.
1904                          */
1905                         if (mctl_present) {
1906                                 freeb(first_mp);
1907                         }
1908                         icmp_redirect(ill, mp);
1909                         return;
1910                 case ICMP_DEST_UNREACHABLE:
1911                         if (icmph->icmph_code == ICMP_FRAGMENTATION_NEEDED) {
1912                                 if (!icmp_inbound_too_big(icmph, ipha, ill,
1913                                     zoneid, mp, iph_hdr_length, ipst)) {
1914                                         freemsg(first_mp);
1915                                         return;
1916                                 }
1917                                 /*
1918                                  * icmp_inbound_too_big() may alter mp.
1919                                  * Resynch ipha and icmph accordingly.
1920                                  */
1921                                 icmph = (icmph_t *)&mp->b_rptr[iph_hdr_length];
1922                                 ipha = (ipha_t *)&icmph[1];
1923                         }
1924                         /* FALLTHRU */
1925                 default :
1926                         /*
1927                          * IPQoS notes: Since we have already done IPQoS
1928                          * processing we don't want to do it again in
1929                          * the fanout routines called by
1930                          * icmp_inbound_error_fanout, hence the last
1931                          * argument, ip_policy, is B_FALSE.
1932                          */
1933                         icmp_inbound_error_fanout(q, ill, first_mp, icmph,
1934                             ipha, iph_hdr_length, hdr_length, mctl_present,
1935                             B_FALSE, recv_ill, zoneid);
1936                 }
1937                 return;
1938         }
1939         /* Send out an ICMP packet */
1940         icmph->icmph_checksum = 0;
1941         icmph->icmph_checksum = IP_CSUM(mp, iph_hdr_length, 0);
1942         if (broadcast || CLASSD(ipha->ipha_dst)) {
1943                 ipif_t  *ipif_chosen;
1944                 /*
1945                  * Make it look like it was directed to us, so we don't look
1946                  * like a fool with a broadcast or multicast source address.
1947                  */
1948                 ipif = ipif_lookup_remote(ill, ipha->ipha_src, zoneid);
1949                 /*
1950                  * Make sure that we haven't grabbed an interface that's DOWN.
1951                  */
1952                 if (ipif != NULL) {
1953                         ipif_chosen = ipif_select_source(ipif->ipif_ill,
1954                             ipha->ipha_src, zoneid);
1955                         if (ipif_chosen != NULL) {
1956                                 ipif_refrele(ipif);
1957                                 ipif = ipif_chosen;
1958                         }
1959                 }
1960                 if (ipif == NULL) {
1961                         ip0dbg(("icmp_inbound: "
1962                             "No source for broadcast/multicast:\n"
1963                             "\tsrc 0x%x dst 0x%x ill %p "
1964                             "ipif_lcl_addr 0x%x\n",
1965                             ntohl(ipha->ipha_src), ntohl(ipha->ipha_dst),
1966                             (void *)ill,
1967                             ill->ill_ipif->ipif_lcl_addr));
1968                         freemsg(first_mp);
1969                         return;
1970                 }
1971                 ASSERT(ipif != NULL && !ipif->ipif_isv6);
1972                 ipha->ipha_dst = ipif->ipif_src_addr;
1973                 ipif_refrele(ipif);
1974         }
1975         /* Reset time to live. */
1976         ipha->ipha_ttl = ipst->ips_ip_def_ttl;
1977         {
1978                 /* Swap source and destination addresses */
1979                 ipaddr_t tmp;
1980 
1981                 tmp = ipha->ipha_src;
1982                 ipha->ipha_src = ipha->ipha_dst;
1983                 ipha->ipha_dst = tmp;
1984         }
1985         ipha->ipha_ident = 0;
1986         if (!IS_SIMPLE_IPH(ipha))
1987                 icmp_options_update(ipha);
1988 
1989         /*
1990          * ICMP echo replies should go out on the same interface
1991          * the request came on as probes used by in.mpathd for detecting
1992          * NIC failures are ECHO packets. We turn-off load spreading
1993          * by setting ipsec_in_attach_if to B_TRUE, which is copied
1994          * to ipsec_out_attach_if by ipsec_in_to_out called later in this
1995          * function. This is in turn handled by ip_wput and ip_newroute
1996          * to make sure that the packet goes out on the interface it came
1997          * in on. If we don't turnoff load spreading, the packets might get
1998          * dropped if there are no non-FAILED/INACTIVE interfaces for it
1999          * to go out and in.mpathd would wrongly detect a failure or
2000          * mis-detect a NIC failure for link failure. As load spreading
2001          * can happen only if ill_group is not NULL, we do only for
2002          * that case and this does not affect the normal case.
2003          *
2004          * We turn off load spreading only on echo packets that came from
2005          * on-link hosts. If the interface route has been deleted, this will
2006          * not be enforced as we can't do much. For off-link hosts, as the
2007          * default routes in IPv4 does not typically have an ire_ipif
2008          * pointer, we can't force MATCH_IRE_ILL in ip_wput/ip_newroute.
2009          * Moreover, expecting a default route through this interface may
2010          * not be correct. We use ipha_dst because of the swap above.
2011          */
2012         onlink = B_FALSE;
2013         if (icmph->icmph_type == ICMP_ECHO_REPLY && ill->ill_group != NULL) {
2014                 /*
2015                  * First, we need to make sure that it is not one of our
2016                  * local addresses. If we set onlink when it is one of
2017                  * our local addresses, we will end up creating IRE_CACHES
2018                  * for one of our local addresses. Then, we will never
2019                  * accept packets for them afterwards.
2020                  */
2021                 src_ire = ire_ctable_lookup(ipha->ipha_dst, 0, IRE_LOCAL,
2022                     NULL, ALL_ZONES, NULL, MATCH_IRE_TYPE, ipst);
2023                 if (src_ire == NULL) {
2024                         ipif = ipif_get_next_ipif(NULL, ill);
2025                         if (ipif == NULL) {
2026                                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
2027                                 freemsg(mp);
2028                                 return;
2029                         }
2030                         src_ire = ire_ftable_lookup(ipha->ipha_dst, 0, 0,
2031                             IRE_INTERFACE, ipif, NULL, ALL_ZONES, 0,
2032                             NULL, MATCH_IRE_ILL | MATCH_IRE_TYPE, ipst);
2033                         ipif_refrele(ipif);
2034                         if (src_ire != NULL) {
2035                                 onlink = B_TRUE;
2036                                 ire_refrele(src_ire);
2037                         }
2038                 } else {
2039                         ire_refrele(src_ire);
2040                 }
2041         }
2042         if (!mctl_present) {
2043                 /*
2044                  * This packet should go out the same way as it
2045                  * came in i.e in clear. To make sure that global
2046                  * policy will not be applied to this in ip_wput_ire,
2047                  * we attach a IPSEC_IN mp and clear ipsec_in_secure.
2048                  */
2049                 ASSERT(first_mp == mp);
2050                 first_mp = ipsec_in_alloc(B_TRUE, ipst->ips_netstack);
2051                 if (first_mp == NULL) {
2052                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
2053                         freemsg(mp);
2054                         return;
2055                 }
2056                 ii = (ipsec_in_t *)first_mp->b_rptr;
2057 
2058                 /* This is not a secure packet */
2059                 ii->ipsec_in_secure = B_FALSE;
2060                 if (onlink) {
2061                         ii->ipsec_in_attach_if = B_TRUE;
2062                         ii->ipsec_in_ill_index =
2063                             ill->ill_phyint->phyint_ifindex;
2064                         ii->ipsec_in_rill_index =
2065                             recv_ill->ill_phyint->phyint_ifindex;
2066                 }
2067                 first_mp->b_cont = mp;
2068         } else if (onlink) {
2069                 ii = (ipsec_in_t *)first_mp->b_rptr;
2070                 ii->ipsec_in_attach_if = B_TRUE;
2071                 ii->ipsec_in_ill_index = ill->ill_phyint->phyint_ifindex;
2072                 ii->ipsec_in_rill_index = recv_ill->ill_phyint->phyint_ifindex;
2073                 ii->ipsec_in_ns = ipst->ips_netstack;     /* No netstack_hold */
2074         } else {
2075                 ii = (ipsec_in_t *)first_mp->b_rptr;
2076                 ii->ipsec_in_ns = ipst->ips_netstack;     /* No netstack_hold */
2077         }
2078         ii->ipsec_in_zoneid = zoneid;
2079         ASSERT(zoneid != ALL_ZONES);
2080         if (!ipsec_in_to_out(first_mp, ipha, NULL)) {
2081                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
2082                 return;
2083         }
2084         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutMsgs);
2085         put(WR(q), first_mp);
2086 }
2087 
2088 static ipaddr_t
2089 icmp_get_nexthop_addr(ipha_t *ipha, ill_t *ill, zoneid_t zoneid, mblk_t *mp)
2090 {
2091         conn_t *connp;
2092         connf_t *connfp;
2093         ipaddr_t nexthop_addr = INADDR_ANY;
2094         int hdr_length = IPH_HDR_LENGTH(ipha);
2095         uint16_t *up;
2096         uint32_t ports;
2097         ip_stack_t *ipst = ill->ill_ipst;
2098 
2099         up = (uint16_t *)((uchar_t *)ipha + hdr_length);
2100         switch (ipha->ipha_protocol) {
2101                 case IPPROTO_TCP:
2102                 {
2103                         tcph_t *tcph;
2104 
2105                         /* do a reverse lookup */
2106                         tcph = (tcph_t *)((uchar_t *)ipha + hdr_length);
2107                         connp = ipcl_tcp_lookup_reversed_ipv4(ipha, tcph,
2108                             TCPS_LISTEN, ipst);
2109                         break;
2110                 }
2111                 case IPPROTO_UDP:
2112                 {
2113                         uint32_t dstport, srcport;
2114 
2115                         ((uint16_t *)&ports)[0] = up[1];
2116                         ((uint16_t *)&ports)[1] = up[0];
2117 
2118                         /* Extract ports in net byte order */
2119                         dstport = htons(ntohl(ports) & 0xFFFF);
2120                         srcport = htons(ntohl(ports) >> 16);
2121 
2122                         connfp = &ipst->ips_ipcl_udp_fanout[
2123                             IPCL_UDP_HASH(dstport, ipst)];
2124                         mutex_enter(&connfp->connf_lock);
2125                         connp = connfp->connf_head;
2126 
2127                         /* do a reverse lookup */
2128                         while ((connp != NULL) &&
2129                             (!IPCL_UDP_MATCH(connp, dstport,
2130                             ipha->ipha_src, srcport, ipha->ipha_dst) ||
2131                             !IPCL_ZONE_MATCH(connp, zoneid))) {
2132                                 connp = connp->conn_next;
2133                         }
2134                         if (connp != NULL)
2135                                 CONN_INC_REF(connp);
2136                         mutex_exit(&connfp->connf_lock);
2137                         break;
2138                 }
2139                 case IPPROTO_SCTP:
2140                 {
2141                         in6_addr_t map_src, map_dst;
2142 
2143                         IN6_IPADDR_TO_V4MAPPED(ipha->ipha_dst, &map_src);
2144                         IN6_IPADDR_TO_V4MAPPED(ipha->ipha_src, &map_dst);
2145                         ((uint16_t *)&ports)[0] = up[1];
2146                         ((uint16_t *)&ports)[1] = up[0];
2147 
2148                         connp = sctp_find_conn(&map_src, &map_dst, ports,
2149                             zoneid, ipst->ips_netstack->netstack_sctp);
2150                         if (connp == NULL) {
2151                                 connp = ipcl_classify_raw(mp, IPPROTO_SCTP,
2152                                     zoneid, ports, ipha, ipst);
2153                         } else {
2154                                 CONN_INC_REF(connp);
2155                                 SCTP_REFRELE(CONN2SCTP(connp));
2156                         }
2157                         break;
2158                 }
2159                 default:
2160                 {
2161                         ipha_t ripha;
2162 
2163                         ripha.ipha_src = ipha->ipha_dst;
2164                         ripha.ipha_dst = ipha->ipha_src;
2165                         ripha.ipha_protocol = ipha->ipha_protocol;
2166 
2167                         connfp = &ipst->ips_ipcl_proto_fanout[
2168                             ipha->ipha_protocol];
2169                         mutex_enter(&connfp->connf_lock);
2170                         connp = connfp->connf_head;
2171                         for (connp = connfp->connf_head; connp != NULL;
2172                             connp = connp->conn_next) {
2173                                 if (IPCL_PROTO_MATCH(connp,
2174                                     ipha->ipha_protocol, &ripha, ill,
2175                                     0, zoneid)) {
2176                                         CONN_INC_REF(connp);
2177                                         break;
2178                                 }
2179                         }
2180                         mutex_exit(&connfp->connf_lock);
2181                 }
2182         }
2183         if (connp != NULL) {
2184                 if (connp->conn_nexthop_set)
2185                         nexthop_addr = connp->conn_nexthop_v4;
2186                 CONN_DEC_REF(connp);
2187         }
2188         return (nexthop_addr);
2189 }
2190 
2191 /* Table from RFC 1191 */
2192 static int icmp_frag_size_table[] =
2193 { 32000, 17914, 8166, 4352, 2002, 1496, 1006, 508, 296, 68 };
2194 
2195 /*
2196  * Process received ICMP Packet too big.
2197  * After updating any IRE it does the fanout to any matching transport streams.
2198  * Assumes the message has been pulled up till the IP header that caused
2199  * the error.
2200  *
2201  * Returns B_FALSE on failure and B_TRUE on success.
2202  */
2203 static boolean_t
2204 icmp_inbound_too_big(icmph_t *icmph, ipha_t *ipha, ill_t *ill,
2205     zoneid_t zoneid, mblk_t *mp, int iph_hdr_length,
2206     ip_stack_t *ipst)
2207 {
2208         ire_t   *ire, *first_ire;
2209         int     mtu;
2210         int     hdr_length;
2211         ipaddr_t nexthop_addr;
2212 
2213         ASSERT(icmph->icmph_type == ICMP_DEST_UNREACHABLE &&
2214             icmph->icmph_code == ICMP_FRAGMENTATION_NEEDED);
2215         ASSERT(ill != NULL);
2216 
2217         hdr_length = IPH_HDR_LENGTH(ipha);
2218 
2219         /* Drop if the original packet contained a source route */
2220         if (ip_source_route_included(ipha)) {
2221                 return (B_FALSE);
2222         }
2223         /*
2224          * Verify we have atleast ICMP_MIN_TP_HDR_LENGTH bytes of transport
2225          * header.
2226          */
2227         if ((uchar_t *)ipha + hdr_length + ICMP_MIN_TP_HDR_LEN >
2228             mp->b_wptr) {
2229                 if (!pullupmsg(mp, (uchar_t *)ipha + hdr_length +
2230                     ICMP_MIN_TP_HDR_LEN - mp->b_rptr)) {
2231                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
2232                         ip1dbg(("icmp_inbound_too_big: insufficient hdr\n"));
2233                         return (B_FALSE);
2234                 }
2235                 icmph = (icmph_t *)&mp->b_rptr[iph_hdr_length];
2236                 ipha = (ipha_t *)&icmph[1];
2237         }
2238         nexthop_addr = icmp_get_nexthop_addr(ipha, ill, zoneid, mp);
2239         if (nexthop_addr != INADDR_ANY) {
2240                 /* nexthop set */
2241                 first_ire = ire_ctable_lookup(ipha->ipha_dst,
2242                     nexthop_addr, 0, NULL, ALL_ZONES, MBLK_GETLABEL(mp),
2243                     MATCH_IRE_MARK_PRIVATE_ADDR | MATCH_IRE_GW, ipst);
2244         } else {
2245                 /* nexthop not set */
2246                 first_ire = ire_ctable_lookup(ipha->ipha_dst, 0, IRE_CACHE,
2247                     NULL, ALL_ZONES, NULL, MATCH_IRE_TYPE, ipst);
2248         }
2249 
2250         if (!first_ire) {
2251                 ip1dbg(("icmp_inbound_too_big: no route for 0x%x\n",
2252                     ntohl(ipha->ipha_dst)));
2253                 return (B_FALSE);
2254         }
2255         /* Check for MTU discovery advice as described in RFC 1191 */
2256         mtu = ntohs(icmph->icmph_du_mtu);
2257         rw_enter(&first_ire->ire_bucket->irb_lock, RW_READER);
2258         for (ire = first_ire; ire != NULL && ire->ire_addr == ipha->ipha_dst;
2259             ire = ire->ire_next) {
2260                 /*
2261                  * Look for the connection to which this ICMP message is
2262                  * directed. If it has the IP_NEXTHOP option set, then the
2263                  * search is limited to IREs with the MATCH_IRE_PRIVATE
2264                  * option. Else the search is limited to regular IREs.
2265                  */
2266                 if (((ire->ire_marks & IRE_MARK_PRIVATE_ADDR) &&
2267                     (nexthop_addr != ire->ire_gateway_addr)) ||
2268                     (!(ire->ire_marks & IRE_MARK_PRIVATE_ADDR) &&
2269                     (nexthop_addr != INADDR_ANY)))
2270                         continue;
2271 
2272                 mutex_enter(&ire->ire_lock);
2273                 if (icmph->icmph_du_zero == 0 && mtu > 68) {
2274                         /* Reduce the IRE max frag value as advised. */
2275                         ip1dbg(("Received mtu from router: %d (was %d)\n",
2276                             mtu, ire->ire_max_frag));
2277                         ire->ire_max_frag = MIN(ire->ire_max_frag, mtu);
2278                 } else {
2279                         uint32_t length;
2280                         int     i;
2281 
2282                         /*
2283                          * Use the table from RFC 1191 to figure out
2284                          * the next "plateau" based on the length in
2285                          * the original IP packet.
2286                          */
2287                         length = ntohs(ipha->ipha_length);
2288                         if (ire->ire_max_frag <= length &&
2289                             ire->ire_max_frag >= length - hdr_length) {
2290                                 /*
2291                                  * Handle broken BSD 4.2 systems that
2292                                  * return the wrong iph_length in ICMP
2293                                  * errors.
2294                                  */
2295                                 ip1dbg(("Wrong mtu: sent %d, ire %d\n",
2296                                     length, ire->ire_max_frag));
2297                                 length -= hdr_length;
2298                         }
2299                         for (i = 0; i < A_CNT(icmp_frag_size_table); i++) {
2300                                 if (length > icmp_frag_size_table[i])
2301                                         break;
2302                         }
2303                         if (i == A_CNT(icmp_frag_size_table)) {
2304                                 /* Smaller than 68! */
2305                                 ip1dbg(("Too big for packet size %d\n",
2306                                     length));
2307                                 ire->ire_max_frag = MIN(ire->ire_max_frag, 576);
2308                                 ire->ire_frag_flag = 0;
2309                         } else {
2310                                 mtu = icmp_frag_size_table[i];
2311                                 ip1dbg(("Calculated mtu %d, packet size %d, "
2312                                     "before %d", mtu, length,
2313                                     ire->ire_max_frag));
2314                                 ire->ire_max_frag = MIN(ire->ire_max_frag, mtu);
2315                                 ip1dbg((", after %d\n", ire->ire_max_frag));
2316                         }
2317                         /* Record the new max frag size for the ULP. */
2318                         icmph->icmph_du_zero = 0;
2319                         icmph->icmph_du_mtu =
2320                             htons((uint16_t)ire->ire_max_frag);
2321                 }
2322                 mutex_exit(&ire->ire_lock);
2323         }
2324         rw_exit(&first_ire->ire_bucket->irb_lock);
2325         ire_refrele(first_ire);
2326         return (B_TRUE);
2327 }
2328 
2329 /*
2330  * If the packet in error is Self-Encapsulated, icmp_inbound_error_fanout
2331  * calls this function.
2332  */
2333 static mblk_t *
2334 icmp_inbound_self_encap_error(mblk_t *mp, int iph_hdr_length, int hdr_length)
2335 {
2336         ipha_t *ipha;
2337         icmph_t *icmph;
2338         ipha_t *in_ipha;
2339         int length;
2340 
2341         ASSERT(mp->b_datap->db_type == M_DATA);
2342 
2343         /*
2344          * For Self-encapsulated packets, we added an extra IP header
2345          * without the options. Inner IP header is the one from which
2346          * the outer IP header was formed. Thus, we need to remove the
2347          * outer IP header. To do this, we pullup the whole message
2348          * and overlay whatever follows the outer IP header over the
2349          * outer IP header.
2350          */
2351 
2352         if (!pullupmsg(mp, -1))
2353                 return (NULL);
2354 
2355         icmph = (icmph_t *)&mp->b_rptr[iph_hdr_length];
2356         ipha = (ipha_t *)&icmph[1];
2357         in_ipha = (ipha_t *)((uchar_t *)ipha + hdr_length);
2358 
2359         /*
2360          * The length that we want to overlay is following the inner
2361          * IP header. Subtracting the IP header + icmp header + outer
2362          * IP header's length should give us the length that we want to
2363          * overlay.
2364          */
2365         length = msgdsize(mp) - iph_hdr_length - sizeof (icmph_t) -
2366             hdr_length;
2367         /*
2368          * Overlay whatever follows the inner header over the
2369          * outer header.
2370          */
2371         bcopy((uchar_t *)in_ipha, (uchar_t *)ipha, length);
2372 
2373         /* Set the wptr to account for the outer header */
2374         mp->b_wptr -= hdr_length;
2375         return (mp);
2376 }
2377 
2378 /*
2379  * Try to pass the ICMP message upstream in case the ULP cares.
2380  *
2381  * If the packet that caused the ICMP error is secure, we send
2382  * it to AH/ESP to make sure that the attached packet has a
2383  * valid association. ipha in the code below points to the
2384  * IP header of the packet that caused the error.
2385  *
2386  * We handle ICMP_FRAGMENTATION_NEEDED(IFN) message differently
2387  * in the context of IPsec. Normally we tell the upper layer
2388  * whenever we send the ire (including ip_bind), the IPsec header
2389  * length in ire_ipsec_overhead. TCP can deduce the MSS as it
2390  * has both the MTU (ire_max_frag) and the ire_ipsec_overhead.
2391  * Similarly, we pass the new MTU icmph_du_mtu and TCP does the
2392  * same thing. As TCP has the IPsec options size that needs to be
2393  * adjusted, we just pass the MTU unchanged.
2394  *
2395  * IFN could have been generated locally or by some router.
2396  *
2397  * LOCAL : *ip_wput_ire -> icmp_frag_needed could have generated this.
2398  *          This happens because IP adjusted its value of MTU on an
2399  *          earlier IFN message and could not tell the upper layer,
2400  *          the new adjusted value of MTU e.g. Packet was encrypted
2401  *          or there was not enough information to fanout to upper
2402  *          layers. Thus on the next outbound datagram, ip_wput_ire
2403  *          generates the IFN, where IPsec processing has *not* been
2404  *          done.
2405  *
2406  *         *ip_wput_ire_fragmentit -> ip_wput_frag -> icmp_frag_needed
2407  *          could have generated this. This happens because ire_max_frag
2408  *          value in IP was set to a new value, while the IPsec processing
2409  *          was being done and after we made the fragmentation check in
2410  *          ip_wput_ire. Thus on return from IPsec processing,
2411  *          ip_wput_ipsec_out finds that the new length is > ire_max_frag
2412  *          and generates the IFN. As IPsec processing is over, we fanout
2413  *          to AH/ESP to remove the header.
2414  *
2415  *          In both these cases, ipsec_in_loopback will be set indicating
2416  *          that IFN was generated locally.
2417  *
2418  * ROUTER : IFN could be secure or non-secure.
2419  *
2420  *          * SECURE : We use the IPSEC_IN to fanout to AH/ESP if the
2421  *            packet in error has AH/ESP headers to validate the AH/ESP
2422  *            headers. AH/ESP will verify whether there is a valid SA or
2423  *            not and send it back. We will fanout again if we have more
2424  *            data in the packet.
2425  *
2426  *            If the packet in error does not have AH/ESP, we handle it
2427  *            like any other case.
2428  *
2429  *          * NON_SECURE : If the packet in error has AH/ESP headers,
2430  *            we attach a dummy ipsec_in and send it up to AH/ESP
2431  *            for validation. AH/ESP will verify whether there is a
2432  *            valid SA or not and send it back. We will fanout again if
2433  *            we have more data in the packet.
2434  *
2435  *            If the packet in error does not have AH/ESP, we handle it
2436  *            like any other case.
2437  */
2438 static void
2439 icmp_inbound_error_fanout(queue_t *q, ill_t *ill, mblk_t *mp,
2440     icmph_t *icmph, ipha_t *ipha, int iph_hdr_length, int hdr_length,
2441     boolean_t mctl_present, boolean_t ip_policy, ill_t *recv_ill,
2442     zoneid_t zoneid)
2443 {
2444         uint16_t *up;   /* Pointer to ports in ULP header */
2445         uint32_t ports; /* reversed ports for fanout */
2446         ipha_t ripha;   /* With reversed addresses */
2447         mblk_t *first_mp;
2448         ipsec_in_t *ii;
2449         tcph_t  *tcph;
2450         conn_t  *connp;
2451         ip_stack_t *ipst;
2452 
2453         ASSERT(ill != NULL);
2454 
2455         ASSERT(recv_ill != NULL);
2456         ipst = recv_ill->ill_ipst;
2457 
2458         first_mp = mp;
2459         if (mctl_present) {
2460                 mp = first_mp->b_cont;
2461                 ASSERT(mp != NULL);
2462 
2463                 ii = (ipsec_in_t *)first_mp->b_rptr;
2464                 ASSERT(ii->ipsec_in_type == IPSEC_IN);
2465         } else {
2466                 ii = NULL;
2467         }
2468 
2469         switch (ipha->ipha_protocol) {
2470         case IPPROTO_UDP:
2471                 /*
2472                  * Verify we have at least ICMP_MIN_TP_HDR_LEN bytes of
2473                  * transport header.
2474                  */
2475                 if ((uchar_t *)ipha + hdr_length + ICMP_MIN_TP_HDR_LEN >
2476                     mp->b_wptr) {
2477                         if (!pullupmsg(mp, (uchar_t *)ipha + hdr_length +
2478                             ICMP_MIN_TP_HDR_LEN - mp->b_rptr)) {
2479                                 goto discard_pkt;
2480                         }
2481                         icmph = (icmph_t *)&mp->b_rptr[iph_hdr_length];
2482                         ipha = (ipha_t *)&icmph[1];
2483                 }
2484                 up = (uint16_t *)((uchar_t *)ipha + hdr_length);
2485 
2486                 /*
2487                  * Attempt to find a client stream based on port.
2488                  * Note that we do a reverse lookup since the header is
2489                  * in the form we sent it out.
2490                  * The ripha header is only used for the IP_UDP_MATCH and we
2491                  * only set the src and dst addresses and protocol.
2492                  */
2493                 ripha.ipha_src = ipha->ipha_dst;
2494                 ripha.ipha_dst = ipha->ipha_src;
2495                 ripha.ipha_protocol = ipha->ipha_protocol;
2496                 ((uint16_t *)&ports)[0] = up[1];
2497                 ((uint16_t *)&ports)[1] = up[0];
2498                 ip2dbg(("icmp_inbound_error: UDP %x:%d to %x:%d: %d/%d\n",
2499                     ntohl(ipha->ipha_src), ntohs(up[0]),
2500                     ntohl(ipha->ipha_dst), ntohs(up[1]),
2501                     icmph->icmph_type, icmph->icmph_code));
2502 
2503                 /* Have to change db_type after any pullupmsg */
2504                 DB_TYPE(mp) = M_CTL;
2505 
2506                 ip_fanout_udp(q, first_mp, ill, &ripha, ports, B_FALSE, 0,
2507                     mctl_present, ip_policy, recv_ill, zoneid);
2508                 return;
2509 
2510         case IPPROTO_TCP:
2511                 /*
2512                  * Verify we have at least ICMP_MIN_TP_HDR_LEN bytes of
2513                  * transport header.
2514                  */
2515                 if ((uchar_t *)ipha + hdr_length + ICMP_MIN_TP_HDR_LEN >
2516                     mp->b_wptr) {
2517                         if (!pullupmsg(mp, (uchar_t *)ipha + hdr_length +
2518                             ICMP_MIN_TP_HDR_LEN - mp->b_rptr)) {
2519                                 goto discard_pkt;
2520                         }
2521                         icmph = (icmph_t *)&mp->b_rptr[iph_hdr_length];
2522                         ipha = (ipha_t *)&icmph[1];
2523                 }
2524                 /*
2525                  * Find a TCP client stream for this packet.
2526                  * Note that we do a reverse lookup since the header is
2527                  * in the form we sent it out.
2528                  */
2529                 tcph = (tcph_t *)((uchar_t *)ipha + hdr_length);
2530                 connp = ipcl_tcp_lookup_reversed_ipv4(ipha, tcph, TCPS_LISTEN,
2531                     ipst);
2532                 if (connp == NULL)
2533                         goto discard_pkt;
2534 
2535                 /* Have to change db_type after any pullupmsg */
2536                 DB_TYPE(mp) = M_CTL;
2537                 squeue_fill(connp->conn_sqp, first_mp, tcp_input,
2538                     connp, SQTAG_TCP_INPUT_ICMP_ERR);
2539                 return;
2540 
2541         case IPPROTO_SCTP:
2542                 /*
2543                  * Verify we have at least ICMP_MIN_TP_HDR_LEN bytes of
2544                  * transport header.
2545                  */
2546                 if ((uchar_t *)ipha + hdr_length + ICMP_MIN_TP_HDR_LEN >
2547                     mp->b_wptr) {
2548                         if (!pullupmsg(mp, (uchar_t *)ipha + hdr_length +
2549                             ICMP_MIN_TP_HDR_LEN - mp->b_rptr)) {
2550                                 goto discard_pkt;
2551                         }
2552                         icmph = (icmph_t *)&mp->b_rptr[iph_hdr_length];
2553                         ipha = (ipha_t *)&icmph[1];
2554                 }
2555                 up = (uint16_t *)((uchar_t *)ipha + hdr_length);
2556                 /*
2557                  * Find a SCTP client stream for this packet.
2558                  * Note that we do a reverse lookup since the header is
2559                  * in the form we sent it out.
2560                  * The ripha header is only used for the matching and we
2561                  * only set the src and dst addresses, protocol, and version.
2562                  */
2563                 ripha.ipha_src = ipha->ipha_dst;
2564                 ripha.ipha_dst = ipha->ipha_src;
2565                 ripha.ipha_protocol = ipha->ipha_protocol;
2566                 ripha.ipha_version_and_hdr_length =
2567                     ipha->ipha_version_and_hdr_length;
2568                 ((uint16_t *)&ports)[0] = up[1];
2569                 ((uint16_t *)&ports)[1] = up[0];
2570 
2571                 /* Have to change db_type after any pullupmsg */
2572                 DB_TYPE(mp) = M_CTL;
2573                 ip_fanout_sctp(first_mp, recv_ill, &ripha, ports, 0,
2574                     mctl_present, ip_policy, zoneid);
2575                 return;
2576 
2577         case IPPROTO_ESP:
2578         case IPPROTO_AH: {
2579                 int ipsec_rc;
2580                 ipsec_stack_t *ipss = ipst->ips_netstack->netstack_ipsec;
2581 
2582                 /*
2583                  * We need a IPSEC_IN in the front to fanout to AH/ESP.
2584                  * We will re-use the IPSEC_IN if it is already present as
2585                  * AH/ESP will not affect any fields in the IPSEC_IN for
2586                  * ICMP errors. If there is no IPSEC_IN, allocate a new
2587                  * one and attach it in the front.
2588                  */
2589                 if (ii != NULL) {
2590                         /*
2591                          * ip_fanout_proto_again converts the ICMP errors
2592                          * that come back from AH/ESP to M_DATA so that
2593                          * if it is non-AH/ESP and we do a pullupmsg in
2594                          * this function, it would work. Convert it back
2595                          * to M_CTL before we send up as this is a ICMP
2596                          * error. This could have been generated locally or
2597                          * by some router. Validate the inner IPsec
2598                          * headers.
2599                          *
2600                          * NOTE : ill_index is used by ip_fanout_proto_again
2601                          * to locate the ill.
2602                          */
2603                         ASSERT(ill != NULL);
2604                         ii->ipsec_in_ill_index =
2605                             ill->ill_phyint->phyint_ifindex;
2606                         ii->ipsec_in_rill_index =
2607                             recv_ill->ill_phyint->phyint_ifindex;
2608                         DB_TYPE(first_mp->b_cont) = M_CTL;
2609                 } else {
2610                         /*
2611                          * IPSEC_IN is not present. We attach a ipsec_in
2612                          * message and send up to IPsec for validating
2613                          * and removing the IPsec headers. Clear
2614                          * ipsec_in_secure so that when we return
2615                          * from IPsec, we don't mistakenly think that this
2616                          * is a secure packet came from the network.
2617                          *
2618                          * NOTE : ill_index is used by ip_fanout_proto_again
2619                          * to locate the ill.
2620                          */
2621                         ASSERT(first_mp == mp);
2622                         first_mp = ipsec_in_alloc(B_TRUE, ipst->ips_netstack);
2623                         if (first_mp == NULL) {
2624                                 freemsg(mp);
2625                                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
2626                                 return;
2627                         }
2628                         ii = (ipsec_in_t *)first_mp->b_rptr;
2629 
2630                         /* This is not a secure packet */
2631                         ii->ipsec_in_secure = B_FALSE;
2632                         first_mp->b_cont = mp;
2633                         DB_TYPE(mp) = M_CTL;
2634                         ASSERT(ill != NULL);
2635                         ii->ipsec_in_ill_index =
2636                             ill->ill_phyint->phyint_ifindex;
2637                         ii->ipsec_in_rill_index =
2638                             recv_ill->ill_phyint->phyint_ifindex;
2639                 }
2640                 ip2dbg(("icmp_inbound_error: ipsec\n"));
2641 
2642                 if (!ipsec_loaded(ipss)) {
2643                         ip_proto_not_sup(q, first_mp, 0, zoneid, ipst);
2644                         return;
2645                 }
2646 
2647                 if (ipha->ipha_protocol == IPPROTO_ESP)
2648                         ipsec_rc = ipsecesp_icmp_error(first_mp);
2649                 else
2650                         ipsec_rc = ipsecah_icmp_error(first_mp);
2651                 if (ipsec_rc == IPSEC_STATUS_FAILED)
2652                         return;
2653 
2654                 ip_fanout_proto_again(first_mp, ill, recv_ill, NULL);
2655                 return;
2656         }
2657         default:
2658                 /*
2659                  * The ripha header is only used for the lookup and we
2660                  * only set the src and dst addresses and protocol.
2661                  */
2662                 ripha.ipha_src = ipha->ipha_dst;
2663                 ripha.ipha_dst = ipha->ipha_src;
2664                 ripha.ipha_protocol = ipha->ipha_protocol;
2665                 ip2dbg(("icmp_inbound_error: proto %d %x to %x: %d/%d\n",
2666                     ripha.ipha_protocol, ntohl(ipha->ipha_src),
2667                     ntohl(ipha->ipha_dst),
2668                     icmph->icmph_type, icmph->icmph_code));
2669                 if (ipha->ipha_protocol == IPPROTO_ENCAP) {
2670                         ipha_t *in_ipha;
2671 
2672                         if ((uchar_t *)ipha + hdr_length + sizeof (ipha_t) >
2673                             mp->b_wptr) {
2674                                 if (!pullupmsg(mp, (uchar_t *)ipha +
2675                                     hdr_length + sizeof (ipha_t) -
2676                                     mp->b_rptr)) {
2677                                         goto discard_pkt;
2678                                 }
2679                                 icmph = (icmph_t *)&mp->b_rptr[iph_hdr_length];
2680                                 ipha = (ipha_t *)&icmph[1];
2681                         }
2682                         /*
2683                          * Caller has verified that length has to be
2684                          * at least the size of IP header.
2685                          */
2686                         ASSERT(hdr_length >= sizeof (ipha_t));
2687                         /*
2688                          * Check the sanity of the inner IP header like
2689                          * we did for the outer header.
2690                          */
2691                         in_ipha = (ipha_t *)((uchar_t *)ipha + hdr_length);
2692                         if ((IPH_HDR_VERSION(in_ipha) != IPV4_VERSION)) {
2693                                 goto discard_pkt;
2694                         }
2695                         if (IPH_HDR_LENGTH(in_ipha) < sizeof (ipha_t)) {
2696                                 goto discard_pkt;
2697                         }
2698                         /* Check for Self-encapsulated tunnels */
2699                         if (in_ipha->ipha_src == ipha->ipha_src &&
2700                             in_ipha->ipha_dst == ipha->ipha_dst) {
2701 
2702                                 mp = icmp_inbound_self_encap_error(mp,
2703                                     iph_hdr_length, hdr_length);
2704                                 if (mp == NULL)
2705                                         goto discard_pkt;
2706                                 icmph = (icmph_t *)&mp->b_rptr[iph_hdr_length];
2707                                 ipha = (ipha_t *)&icmph[1];
2708                                 hdr_length = IPH_HDR_LENGTH(ipha);
2709                                 /*
2710                                  * The packet in error is self-encapsualted.
2711                                  * And we are finding it further encapsulated
2712                                  * which we could not have possibly generated.
2713                                  */
2714                                 if (ipha->ipha_protocol == IPPROTO_ENCAP) {
2715                                         goto discard_pkt;
2716                                 }
2717                                 icmp_inbound_error_fanout(q, ill, first_mp,
2718                                     icmph, ipha, iph_hdr_length, hdr_length,
2719                                     mctl_present, ip_policy, recv_ill, zoneid);
2720                                 return;
2721                         }
2722                 }
2723                 if ((ipha->ipha_protocol == IPPROTO_ENCAP ||
2724                     ipha->ipha_protocol == IPPROTO_IPV6) &&
2725                     icmph->icmph_code == ICMP_FRAGMENTATION_NEEDED &&
2726                     ii != NULL &&
2727                     ii->ipsec_in_loopback &&
2728                     ii->ipsec_in_secure) {
2729                         /*
2730                          * For IP tunnels that get a looped-back
2731                          * ICMP_FRAGMENTATION_NEEDED message, adjust the
2732                          * reported new MTU to take into account the IPsec
2733                          * headers protecting this configured tunnel.
2734                          *
2735                          * This allows the tunnel module (tun.c) to blindly
2736                          * accept the MTU reported in an ICMP "too big"
2737                          * message.
2738                          *
2739                          * Non-looped back ICMP messages will just be
2740                          * handled by the security protocols (if needed),
2741                          * and the first subsequent packet will hit this
2742                          * path.
2743                          */
2744                         icmph->icmph_du_mtu = htons(ntohs(icmph->icmph_du_mtu) -
2745                             ipsec_in_extra_length(first_mp));
2746                 }
2747                 /* Have to change db_type after any pullupmsg */
2748                 DB_TYPE(mp) = M_CTL;
2749 
2750                 ip_fanout_proto(q, first_mp, ill, &ripha, 0, mctl_present,
2751                     ip_policy, recv_ill, zoneid);
2752                 return;
2753         }
2754         /* NOTREACHED */
2755 discard_pkt:
2756         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
2757 drop_pkt:;
2758         ip1dbg(("icmp_inbound_error_fanout: drop pkt\n"));
2759         freemsg(first_mp);
2760 }
2761 
2762 /*
2763  * Common IP options parser.
2764  *
2765  * Setup routine: fill in *optp with options-parsing state, then
2766  * tail-call ipoptp_next to return the first option.
2767  */
2768 uint8_t
2769 ipoptp_first(ipoptp_t *optp, ipha_t *ipha)
2770 {
2771         uint32_t totallen; /* total length of all options */
2772 
2773         totallen = ipha->ipha_version_and_hdr_length -
2774             (uint8_t)((IP_VERSION << 4) + IP_SIMPLE_HDR_LENGTH_IN_WORDS);
2775         totallen <<= 2;
2776         optp->ipoptp_next = (uint8_t *)(&ipha[1]);
2777         optp->ipoptp_end = optp->ipoptp_next + totallen;
2778         optp->ipoptp_flags = 0;
2779         return (ipoptp_next(optp));
2780 }
2781 
2782 /*
2783  * Common IP options parser: extract next option.
2784  */
2785 uint8_t
2786 ipoptp_next(ipoptp_t *optp)
2787 {
2788         uint8_t *end = optp->ipoptp_end;
2789         uint8_t *cur = optp->ipoptp_next;
2790         uint8_t opt, len, pointer;
2791 
2792         /*
2793          * If cur > end already, then the ipoptp_end or ipoptp_next pointer
2794          * has been corrupted.
2795          */
2796         ASSERT(cur <= end);
2797 
2798         if (cur == end)
2799                 return (IPOPT_EOL);
2800 
2801         opt = cur[IPOPT_OPTVAL];
2802 
2803         /*
2804          * Skip any NOP options.
2805          */
2806         while (opt == IPOPT_NOP) {
2807                 cur++;
2808                 if (cur == end)
2809                         return (IPOPT_EOL);
2810                 opt = cur[IPOPT_OPTVAL];
2811         }
2812 
2813         if (opt == IPOPT_EOL)
2814                 return (IPOPT_EOL);
2815 
2816         /*
2817          * Option requiring a length.
2818          */
2819         if ((cur + 1) >= end) {
2820                 optp->ipoptp_flags |= IPOPTP_ERROR;
2821                 return (IPOPT_EOL);
2822         }
2823         len = cur[IPOPT_OLEN];
2824         if (len < 2) {
2825                 optp->ipoptp_flags |= IPOPTP_ERROR;
2826                 return (IPOPT_EOL);
2827         }
2828         optp->ipoptp_cur = cur;
2829         optp->ipoptp_len = len;
2830         optp->ipoptp_next = cur + len;
2831         if (cur + len > end) {
2832                 optp->ipoptp_flags |= IPOPTP_ERROR;
2833                 return (IPOPT_EOL);
2834         }
2835 
2836         /*
2837          * For the options which require a pointer field, make sure
2838          * its there, and make sure it points to either something
2839          * inside this option, or the end of the option.
2840          */
2841         switch (opt) {
2842         case IPOPT_RR:
2843         case IPOPT_TS:
2844         case IPOPT_LSRR:
2845         case IPOPT_SSRR:
2846                 if (len <= IPOPT_OFFSET) {
2847                         optp->ipoptp_flags |= IPOPTP_ERROR;
2848                         return (opt);
2849                 }
2850                 pointer = cur[IPOPT_OFFSET];
2851                 if (pointer - 1 > len) {
2852                         optp->ipoptp_flags |= IPOPTP_ERROR;
2853                         return (opt);
2854                 }
2855                 break;
2856         }
2857 
2858         /*
2859          * Sanity check the pointer field based on the type of the
2860          * option.
2861          */
2862         switch (opt) {
2863         case IPOPT_RR:
2864         case IPOPT_SSRR:
2865         case IPOPT_LSRR:
2866                 if (pointer < IPOPT_MINOFF_SR)
2867                         optp->ipoptp_flags |= IPOPTP_ERROR;
2868                 break;
2869         case IPOPT_TS:
2870                 if (pointer < IPOPT_MINOFF_IT)
2871                         optp->ipoptp_flags |= IPOPTP_ERROR;
2872                 /*
2873                  * Note that the Internet Timestamp option also
2874                  * contains two four bit fields (the Overflow field,
2875                  * and the Flag field), which follow the pointer
2876                  * field.  We don't need to check that these fields
2877                  * fall within the length of the option because this
2878                  * was implicitely done above.  We've checked that the
2879                  * pointer value is at least IPOPT_MINOFF_IT, and that
2880                  * it falls within the option.  Since IPOPT_MINOFF_IT >
2881                  * IPOPT_POS_OV_FLG, we don't need the explicit check.
2882                  */
2883                 ASSERT(len > IPOPT_POS_OV_FLG);
2884                 break;
2885         }
2886 
2887         return (opt);
2888 }
2889 
2890 /*
2891  * Use the outgoing IP header to create an IP_OPTIONS option the way
2892  * it was passed down from the application.
2893  */
2894 int
2895 ip_opt_get_user(const ipha_t *ipha, uchar_t *buf)
2896 {
2897         ipoptp_t        opts;
2898         const uchar_t   *opt;
2899         uint8_t         optval;
2900         uint8_t         optlen;
2901         uint32_t        len = 0;
2902         uchar_t *buf1 = buf;
2903 
2904         buf += IP_ADDR_LEN;     /* Leave room for final destination */
2905         len += IP_ADDR_LEN;
2906         bzero(buf1, IP_ADDR_LEN);
2907 
2908         /*
2909          * OK to cast away const here, as we don't store through the returned
2910          * opts.ipoptp_cur pointer.
2911          */
2912         for (optval = ipoptp_first(&opts, (ipha_t *)ipha);
2913             optval != IPOPT_EOL;
2914             optval = ipoptp_next(&opts)) {
2915                 int     off;
2916 
2917                 opt = opts.ipoptp_cur;
2918                 optlen = opts.ipoptp_len;
2919                 switch (optval) {
2920                 case IPOPT_SSRR:
2921                 case IPOPT_LSRR:
2922 
2923                         /*
2924                          * Insert ipha_dst as the first entry in the source
2925                          * route and move down the entries on step.
2926                          * The last entry gets placed at buf1.
2927                          */
2928                         buf[IPOPT_OPTVAL] = optval;
2929                         buf[IPOPT_OLEN] = optlen;
2930                         buf[IPOPT_OFFSET] = optlen;
2931 
2932                         off = optlen - IP_ADDR_LEN;
2933                         if (off < 0) {
2934                                 /* No entries in source route */
2935                                 break;
2936                         }
2937                         /* Last entry in source route */
2938                         bcopy(opt + off, buf1, IP_ADDR_LEN);
2939                         off -= IP_ADDR_LEN;
2940 
2941                         while (off > 0) {
2942                                 bcopy(opt + off,
2943                                     buf + off + IP_ADDR_LEN,
2944                                     IP_ADDR_LEN);
2945                                 off -= IP_ADDR_LEN;
2946                         }
2947                         /* ipha_dst into first slot */
2948                         bcopy(&ipha->ipha_dst,
2949                             buf + off + IP_ADDR_LEN,
2950                             IP_ADDR_LEN);
2951                         buf += optlen;
2952                         len += optlen;
2953                         break;
2954 
2955                 case IPOPT_COMSEC:
2956                 case IPOPT_SECURITY:
2957                         /* if passing up a label is not ok, then remove */
2958                         if (is_system_labeled())
2959                                 break;
2960                         /* FALLTHROUGH */
2961                 default:
2962                         bcopy(opt, buf, optlen);
2963                         buf += optlen;
2964                         len += optlen;
2965                         break;
2966                 }
2967         }
2968 done:
2969         /* Pad the resulting options */
2970         while (len & 0x3) {
2971                 *buf++ = IPOPT_EOL;
2972                 len++;
2973         }
2974         return (len);
2975 }
2976 
2977 /*
2978  * Update any record route or timestamp options to include this host.
2979  * Reverse any source route option.
2980  * This routine assumes that the options are well formed i.e. that they
2981  * have already been checked.
2982  */
2983 static void
2984 icmp_options_update(ipha_t *ipha)
2985 {
2986         ipoptp_t        opts;
2987         uchar_t         *opt;
2988         uint8_t         optval;
2989         ipaddr_t        src;            /* Our local address */
2990         ipaddr_t        dst;
2991 
2992         ip2dbg(("icmp_options_update\n"));
2993         src = ipha->ipha_src;
2994         dst = ipha->ipha_dst;
2995 
2996         for (optval = ipoptp_first(&opts, ipha);
2997             optval != IPOPT_EOL;
2998             optval = ipoptp_next(&opts)) {
2999                 ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
3000                 opt = opts.ipoptp_cur;
3001                 ip2dbg(("icmp_options_update: opt %d, len %d\n",
3002                     optval, opts.ipoptp_len));
3003                 switch (optval) {
3004                         int off1, off2;
3005                 case IPOPT_SSRR:
3006                 case IPOPT_LSRR:
3007                         /*
3008                          * Reverse the source route.  The first entry
3009                          * should be the next to last one in the current
3010                          * source route (the last entry is our address).
3011                          * The last entry should be the final destination.
3012                          */
3013                         off1 = IPOPT_MINOFF_SR - 1;
3014                         off2 = opt[IPOPT_OFFSET] - IP_ADDR_LEN - 1;
3015                         if (off2 < 0) {
3016                                 /* No entries in source route */
3017                                 ip1dbg((
3018                                     "icmp_options_update: bad src route\n"));
3019                                 break;
3020                         }
3021                         bcopy((char *)opt + off2, &dst, IP_ADDR_LEN);
3022                         bcopy(&ipha->ipha_dst, (char *)opt + off2, IP_ADDR_LEN);
3023                         bcopy(&dst, &ipha->ipha_dst, IP_ADDR_LEN);
3024                         off2 -= IP_ADDR_LEN;
3025 
3026                         while (off1 < off2) {
3027                                 bcopy((char *)opt + off1, &src, IP_ADDR_LEN);
3028                                 bcopy((char *)opt + off2, (char *)opt + off1,
3029                                     IP_ADDR_LEN);
3030                                 bcopy(&src, (char *)opt + off2, IP_ADDR_LEN);
3031                                 off1 += IP_ADDR_LEN;
3032                                 off2 -= IP_ADDR_LEN;
3033                         }
3034                         opt[IPOPT_OFFSET] = IPOPT_MINOFF_SR;
3035                         break;
3036                 }
3037         }
3038 }
3039 
3040 /*
3041  * Process received ICMP Redirect messages.
3042  */
3043 static void
3044 icmp_redirect(ill_t *ill, mblk_t *mp)
3045 {
3046         ipha_t  *ipha;
3047         int     iph_hdr_length;
3048         icmph_t *icmph;
3049         ipha_t  *ipha_err;
3050         ire_t   *ire;
3051         ire_t   *prev_ire;
3052         ire_t   *save_ire;
3053         ipaddr_t  src, dst, gateway;
3054         iulp_t  ulp_info = { 0 };
3055         int     error;
3056         ip_stack_t *ipst;
3057 
3058         ASSERT(ill != NULL);
3059         ipst = ill->ill_ipst;
3060 
3061         ipha = (ipha_t *)mp->b_rptr;
3062         iph_hdr_length = IPH_HDR_LENGTH(ipha);
3063         if (((mp->b_wptr - mp->b_rptr) - iph_hdr_length) <
3064             sizeof (icmph_t) + IP_SIMPLE_HDR_LENGTH) {
3065                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInErrors);
3066                 freemsg(mp);
3067                 return;
3068         }
3069         icmph = (icmph_t *)&mp->b_rptr[iph_hdr_length];
3070         ipha_err = (ipha_t *)&icmph[1];
3071         src = ipha->ipha_src;
3072         dst = ipha_err->ipha_dst;
3073         gateway = icmph->icmph_rd_gateway;
3074         /* Make sure the new gateway is reachable somehow. */
3075         ire = ire_route_lookup(gateway, 0, 0, IRE_INTERFACE, NULL, NULL,
3076             ALL_ZONES, NULL, MATCH_IRE_TYPE, ipst);
3077         /*
3078          * Make sure we had a route for the dest in question and that
3079          * that route was pointing to the old gateway (the source of the
3080          * redirect packet.)
3081          */
3082         prev_ire = ire_route_lookup(dst, 0, src, 0, NULL, NULL, ALL_ZONES,
3083             NULL, MATCH_IRE_GW, ipst);
3084         /*
3085          * Check that
3086          *      the redirect was not from ourselves
3087          *      the new gateway and the old gateway are directly reachable
3088          */
3089         if (!prev_ire ||
3090             !ire ||
3091             ire->ire_type == IRE_LOCAL) {
3092                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInBadRedirects);
3093                 freemsg(mp);
3094                 if (ire != NULL)
3095                         ire_refrele(ire);
3096                 if (prev_ire != NULL)
3097                         ire_refrele(prev_ire);
3098                 return;
3099         }
3100 
3101         /*
3102          * Should we use the old ULP info to create the new gateway?  From
3103          * a user's perspective, we should inherit the info so that it
3104          * is a "smooth" transition.  If we do not do that, then new
3105          * connections going thru the new gateway will have no route metrics,
3106          * which is counter-intuitive to user.  From a network point of
3107          * view, this may or may not make sense even though the new gateway
3108          * is still directly connected to us so the route metrics should not
3109          * change much.
3110          *
3111          * But if the old ire_uinfo is not initialized, we do another
3112          * recursive lookup on the dest using the new gateway.  There may
3113          * be a route to that.  If so, use it to initialize the redirect
3114          * route.
3115          */
3116         if (prev_ire->ire_uinfo.iulp_set) {
3117                 bcopy(&prev_ire->ire_uinfo, &ulp_info, sizeof (iulp_t));
3118         } else {
3119                 ire_t *tmp_ire;
3120                 ire_t *sire;
3121 
3122                 tmp_ire = ire_ftable_lookup(dst, 0, gateway, 0, NULL, &sire,
3123                     ALL_ZONES, 0, NULL,
3124                     (MATCH_IRE_RECURSIVE | MATCH_IRE_GW | MATCH_IRE_DEFAULT),
3125                     ipst);
3126                 if (sire != NULL) {
3127                         bcopy(&sire->ire_uinfo, &ulp_info, sizeof (iulp_t));
3128                         /*
3129                          * If sire != NULL, ire_ftable_lookup() should not
3130                          * return a NULL value.
3131                          */
3132                         ASSERT(tmp_ire != NULL);
3133                         ire_refrele(tmp_ire);
3134                         ire_refrele(sire);
3135                 } else if (tmp_ire != NULL) {
3136                         bcopy(&tmp_ire->ire_uinfo, &ulp_info,
3137                             sizeof (iulp_t));
3138                         ire_refrele(tmp_ire);
3139                 }
3140         }
3141         if (prev_ire->ire_type == IRE_CACHE)
3142                 ire_delete(prev_ire);
3143         ire_refrele(prev_ire);
3144         /*
3145          * TODO: more precise handling for cases 0, 2, 3, the latter two
3146          * require TOS routing
3147          */
3148         switch (icmph->icmph_code) {
3149         case 0:
3150         case 1:
3151                 /* TODO: TOS specificity for cases 2 and 3 */
3152         case 2:
3153         case 3:
3154                 break;
3155         default:
3156                 freemsg(mp);
3157                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInBadRedirects);
3158                 ire_refrele(ire);
3159                 return;
3160         }
3161         /*
3162          * Create a Route Association.  This will allow us to remember that
3163          * someone we believe told us to use the particular gateway.
3164          */
3165         save_ire = ire;
3166         ire = ire_create(
3167             (uchar_t *)&dst,                        /* dest addr */
3168             (uchar_t *)&ip_g_all_ones,              /* mask */
3169             (uchar_t *)&save_ire->ire_src_addr,  /* source addr */
3170             (uchar_t *)&gateway,            /* gateway addr */
3171             &save_ire->ire_max_frag,             /* max frag */
3172             NULL,                               /* no src nce */
3173             NULL,                               /* no rfq */
3174             NULL,                               /* no stq */
3175             IRE_HOST,
3176             NULL,                               /* ipif */
3177             0,                                  /* cmask */
3178             0,                                  /* phandle */
3179             0,                                  /* ihandle */
3180             (RTF_DYNAMIC | RTF_GATEWAY | RTF_HOST),
3181             &ulp_info,
3182             NULL,                               /* tsol_gc_t */
3183             NULL,                               /* gcgrp */
3184             ipst);
3185 
3186         if (ire == NULL) {
3187                 freemsg(mp);
3188                 ire_refrele(save_ire);
3189                 return;
3190         }
3191         error = ire_add(&ire, NULL, NULL, NULL, B_FALSE);
3192         ire_refrele(save_ire);
3193         atomic_inc_32(&ipst->ips_ip_redirect_cnt);
3194 
3195         if (error == 0) {
3196                 ire_refrele(ire);               /* Held in ire_add_v4 */
3197                 /* tell routing sockets that we received a redirect */
3198                 ip_rts_change(RTM_REDIRECT, dst, gateway, IP_HOST_MASK, 0, src,
3199                     (RTF_DYNAMIC | RTF_GATEWAY | RTF_HOST), 0,
3200                     (RTA_DST | RTA_GATEWAY | RTA_NETMASK | RTA_AUTHOR), ipst);
3201         }
3202 
3203         /*
3204          * Delete any existing IRE_HOST type redirect ires for this destination.
3205          * This together with the added IRE has the effect of
3206          * modifying an existing redirect.
3207          */
3208         prev_ire = ire_ftable_lookup(dst, 0, src, IRE_HOST, NULL, NULL,
3209             ALL_ZONES, 0, NULL, (MATCH_IRE_GW | MATCH_IRE_TYPE), ipst);
3210         if (prev_ire != NULL) {
3211                 if (prev_ire ->ire_flags & RTF_DYNAMIC)
3212                         ire_delete(prev_ire);
3213                 ire_refrele(prev_ire);
3214         }
3215 
3216         freemsg(mp);
3217 }
3218 
3219 /*
3220  * Generate an ICMP parameter problem message.
3221  */
3222 static void
3223 icmp_param_problem(queue_t *q, mblk_t *mp, uint8_t ptr, zoneid_t zoneid,
3224         ip_stack_t *ipst)
3225 {
3226         icmph_t icmph;
3227         boolean_t mctl_present;
3228         mblk_t *first_mp;
3229 
3230         EXTRACT_PKT_MP(mp, first_mp, mctl_present);
3231 
3232         if (!(mp = icmp_pkt_err_ok(mp, ipst))) {
3233                 if (mctl_present)
3234                         freeb(first_mp);
3235                 return;
3236         }
3237 
3238         bzero(&icmph, sizeof (icmph_t));
3239         icmph.icmph_type = ICMP_PARAM_PROBLEM;
3240         icmph.icmph_pp_ptr = ptr;
3241         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutParmProbs);
3242         icmp_pkt(q, first_mp, &icmph, sizeof (icmph_t), mctl_present, zoneid,
3243             ipst);
3244 }
3245 
3246 /*
3247  * Build and ship an IPv4 ICMP message using the packet data in mp, and
3248  * the ICMP header pointed to by "stuff".  (May be called as writer.)
3249  * Note: assumes that icmp_pkt_err_ok has been called to verify that
3250  * an icmp error packet can be sent.
3251  * Assigns an appropriate source address to the packet. If ipha_dst is
3252  * one of our addresses use it for source. Otherwise pick a source based
3253  * on a route lookup back to ipha_src.
3254  * Note that ipha_src must be set here since the
3255  * packet is likely to arrive on an ill queue in ip_wput() which will
3256  * not set a source address.
3257  */
3258 static void
3259 icmp_pkt(queue_t *q, mblk_t *mp, void *stuff, size_t len,
3260     boolean_t mctl_present, zoneid_t zoneid, ip_stack_t *ipst)
3261 {
3262         ipaddr_t dst;
3263         icmph_t *icmph;
3264         ipha_t  *ipha;
3265         uint_t  len_needed;
3266         size_t  msg_len;
3267         mblk_t  *mp1;
3268         ipaddr_t src;
3269         ire_t   *ire;
3270         mblk_t *ipsec_mp;
3271         ipsec_out_t     *io = NULL;
3272 
3273         if (mctl_present) {
3274                 /*
3275                  * If it is :
3276                  *
3277                  * 1) a IPSEC_OUT, then this is caused by outbound
3278                  *    datagram originating on this host. IPsec processing
3279                  *    may or may not have been done. Refer to comments above
3280                  *    icmp_inbound_error_fanout for details.
3281                  *
3282                  * 2) a IPSEC_IN if we are generating a icmp_message
3283                  *    for an incoming datagram destined for us i.e called
3284                  *    from ip_fanout_send_icmp.
3285                  */
3286                 ipsec_info_t *in;
3287                 ipsec_mp = mp;
3288                 mp = ipsec_mp->b_cont;
3289 
3290                 in = (ipsec_info_t *)ipsec_mp->b_rptr;
3291                 ipha = (ipha_t *)mp->b_rptr;
3292 
3293                 ASSERT(in->ipsec_info_type == IPSEC_OUT ||
3294                     in->ipsec_info_type == IPSEC_IN);
3295 
3296                 if (in->ipsec_info_type == IPSEC_IN) {
3297                         /*
3298                          * Convert the IPSEC_IN to IPSEC_OUT.
3299                          */
3300                         if (!ipsec_in_to_out(ipsec_mp, ipha, NULL)) {
3301                                 BUMP_MIB(&ipst->ips_ip_mib,
3302                                     ipIfStatsOutDiscards);
3303                                 return;
3304                         }
3305                         io = (ipsec_out_t *)ipsec_mp->b_rptr;
3306                 } else {
3307                         ASSERT(in->ipsec_info_type == IPSEC_OUT);
3308                         io = (ipsec_out_t *)in;
3309                         /*
3310                          * Clear out ipsec_out_proc_begin, so we do a fresh
3311                          * ire lookup.
3312                          */
3313                         io->ipsec_out_proc_begin = B_FALSE;
3314                 }
3315                 ASSERT(zoneid == io->ipsec_out_zoneid);
3316                 ASSERT(zoneid != ALL_ZONES);
3317         } else {
3318                 /*
3319                  * This is in clear. The icmp message we are building
3320                  * here should go out in clear.
3321                  *
3322                  * Pardon the convolution of it all, but it's easier to
3323                  * allocate a "use cleartext" IPSEC_IN message and convert
3324                  * it than it is to allocate a new one.
3325                  */
3326                 ipsec_in_t *ii;
3327                 ASSERT(DB_TYPE(mp) == M_DATA);
3328                 ipsec_mp = ipsec_in_alloc(B_TRUE, ipst->ips_netstack);
3329                 if (ipsec_mp == NULL) {
3330                         freemsg(mp);
3331                         BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsOutDiscards);
3332                         return;
3333                 }
3334                 ii = (ipsec_in_t *)ipsec_mp->b_rptr;
3335 
3336                 /* This is not a secure packet */
3337                 ii->ipsec_in_secure = B_FALSE;
3338                 /*
3339                  * For trusted extensions using a shared IP address we can
3340                  * send using any zoneid.
3341                  */
3342                 if (zoneid == ALL_ZONES)
3343                         ii->ipsec_in_zoneid = GLOBAL_ZONEID;
3344                 else
3345                         ii->ipsec_in_zoneid = zoneid;
3346                 ipsec_mp->b_cont = mp;
3347                 ipha = (ipha_t *)mp->b_rptr;
3348                 /*
3349                  * Convert the IPSEC_IN to IPSEC_OUT.
3350                  */
3351                 if (!ipsec_in_to_out(ipsec_mp, ipha, NULL)) {
3352                         BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsOutDiscards);
3353                         return;
3354                 }
3355                 io = (ipsec_out_t *)ipsec_mp->b_rptr;
3356         }
3357 
3358         /* Remember our eventual destination */
3359         dst = ipha->ipha_src;
3360 
3361         ire = ire_route_lookup(ipha->ipha_dst, 0, 0, (IRE_LOCAL|IRE_LOOPBACK),
3362             NULL, NULL, zoneid, NULL, MATCH_IRE_TYPE, ipst);
3363         if (ire != NULL &&
3364             (ire->ire_zoneid == zoneid || ire->ire_zoneid == ALL_ZONES)) {
3365                 src = ipha->ipha_dst;
3366         } else {
3367                 if (ire != NULL)
3368                         ire_refrele(ire);
3369                 ire = ire_route_lookup(dst, 0, 0, 0, NULL, NULL, zoneid, NULL,
3370                     (MATCH_IRE_DEFAULT|MATCH_IRE_RECURSIVE|MATCH_IRE_ZONEONLY),
3371                     ipst);
3372                 if (ire == NULL) {
3373                         BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsOutNoRoutes);
3374                         freemsg(ipsec_mp);
3375                         return;
3376                 }
3377                 src = ire->ire_src_addr;
3378         }
3379 
3380         if (ire != NULL)
3381                 ire_refrele(ire);
3382 
3383         /*
3384          * Check if we can send back more then 8 bytes in addition to
3385          * the IP header.  We try to send 64 bytes of data and the internal
3386          * header in the special cases of ipv4 encapsulated ipv4 or ipv6.
3387          */
3388         len_needed = IPH_HDR_LENGTH(ipha);
3389         if (ipha->ipha_protocol == IPPROTO_ENCAP ||
3390             ipha->ipha_protocol == IPPROTO_IPV6) {
3391 
3392                 if (!pullupmsg(mp, -1)) {
3393                         BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsOutDiscards);
3394                         freemsg(ipsec_mp);
3395                         return;
3396                 }
3397                 ipha = (ipha_t *)mp->b_rptr;
3398 
3399                 if (ipha->ipha_protocol == IPPROTO_ENCAP) {
3400                         len_needed += IPH_HDR_LENGTH(((uchar_t *)ipha +
3401                             len_needed));
3402                 } else {
3403                         ip6_t *ip6h = (ip6_t *)((uchar_t *)ipha + len_needed);
3404 
3405                         ASSERT(ipha->ipha_protocol == IPPROTO_IPV6);
3406                         len_needed += ip_hdr_length_v6(mp, ip6h);
3407                 }
3408         }
3409         len_needed += ipst->ips_ip_icmp_return;
3410         msg_len = msgdsize(mp);
3411         if (msg_len > len_needed) {
3412                 (void) adjmsg(mp, len_needed - msg_len);
3413                 msg_len = len_needed;
3414         }
3415         mp1 = allocb_tmpl(sizeof (icmp_ipha) + len, mp);
3416         if (mp1 == NULL) {
3417                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutErrors);
3418                 freemsg(ipsec_mp);
3419                 return;
3420         }
3421         mp1->b_cont = mp;
3422         mp = mp1;
3423         ASSERT(ipsec_mp->b_datap->db_type == M_CTL &&
3424             ipsec_mp->b_rptr == (uint8_t *)io &&
3425             io->ipsec_out_type == IPSEC_OUT);
3426         ipsec_mp->b_cont = mp;
3427 
3428         /*
3429          * Set ipsec_out_icmp_loopback so we can let the ICMP messages this
3430          * node generates be accepted in peace by all on-host destinations.
3431          * If we do NOT assume that all on-host destinations trust
3432          * self-generated ICMP messages, then rework here, ip6.c, and spd.c.
3433          * (Look for ipsec_out_icmp_loopback).
3434          */
3435         io->ipsec_out_icmp_loopback = B_TRUE;
3436 
3437         ipha = (ipha_t *)mp->b_rptr;
3438         mp1->b_wptr = (uchar_t *)ipha + (sizeof (icmp_ipha) + len);
3439         *ipha = icmp_ipha;
3440         ipha->ipha_src = src;
3441         ipha->ipha_dst = dst;
3442         ipha->ipha_ttl = ipst->ips_ip_def_ttl;
3443         msg_len += sizeof (icmp_ipha) + len;
3444         if (msg_len > IP_MAXPACKET) {
3445                 (void) adjmsg(mp, IP_MAXPACKET - msg_len);
3446                 msg_len = IP_MAXPACKET;
3447         }
3448         ipha->ipha_length = htons((uint16_t)msg_len);
3449         icmph = (icmph_t *)&ipha[1];
3450         bcopy(stuff, icmph, len);
3451         icmph->icmph_checksum = 0;
3452         icmph->icmph_checksum = IP_CSUM(mp, (int32_t)sizeof (ipha_t), 0);
3453         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutMsgs);
3454         put(q, ipsec_mp);
3455 }
3456 
3457 /*
3458  * Determine if an ICMP error packet can be sent given the rate limit.
3459  * The limit consists of an average frequency (icmp_pkt_err_interval measured
3460  * in milliseconds) and a burst size. Burst size number of packets can
3461  * be sent arbitrarely closely spaced.
3462  * The state is tracked using two variables to implement an approximate
3463  * token bucket filter:
3464  *      icmp_pkt_err_last - lbolt value when the last burst started
3465  *      icmp_pkt_err_sent - number of packets sent in current burst
3466  */
3467 boolean_t
3468 icmp_err_rate_limit(ip_stack_t *ipst)
3469 {
3470         clock_t now = TICK_TO_MSEC(lbolt);
3471         uint_t refilled; /* Number of packets refilled in tbf since last */
3472         /* Guard against changes by loading into local variable */
3473         uint_t err_interval = ipst->ips_ip_icmp_err_interval;
3474 
3475         if (err_interval == 0)
3476                 return (B_FALSE);
3477 
3478         if (ipst->ips_icmp_pkt_err_last > now) {
3479                 /* 100HZ lbolt in ms for 32bit arch wraps every 49.7 days */
3480                 ipst->ips_icmp_pkt_err_last = 0;
3481                 ipst->ips_icmp_pkt_err_sent = 0;
3482         }
3483         /*
3484          * If we are in a burst update the token bucket filter.
3485          * Update the "last" time to be close to "now" but make sure
3486          * we don't loose precision.
3487          */
3488         if (ipst->ips_icmp_pkt_err_sent != 0) {
3489                 refilled = (now - ipst->ips_icmp_pkt_err_last)/err_interval;
3490                 if (refilled > ipst->ips_icmp_pkt_err_sent) {
3491                         ipst->ips_icmp_pkt_err_sent = 0;
3492                 } else {
3493                         ipst->ips_icmp_pkt_err_sent -= refilled;
3494                         ipst->ips_icmp_pkt_err_last += refilled * err_interval;
3495                 }
3496         }
3497         if (ipst->ips_icmp_pkt_err_sent == 0) {
3498                 /* Start of new burst */
3499                 ipst->ips_icmp_pkt_err_last = now;
3500         }
3501         if (ipst->ips_icmp_pkt_err_sent < ipst->ips_ip_icmp_err_burst) {
3502                 ipst->ips_icmp_pkt_err_sent++;
3503                 ip1dbg(("icmp_err_rate_limit: %d sent in burst\n",
3504                     ipst->ips_icmp_pkt_err_sent));
3505                 return (B_FALSE);
3506         }
3507         ip1dbg(("icmp_err_rate_limit: dropped\n"));
3508         return (B_TRUE);
3509 }
3510 
3511 /*
3512  * Check if it is ok to send an IPv4 ICMP error packet in
3513  * response to the IPv4 packet in mp.
3514  * Free the message and return null if no
3515  * ICMP error packet should be sent.
3516  */
3517 static mblk_t *
3518 icmp_pkt_err_ok(mblk_t *mp, ip_stack_t *ipst)
3519 {
3520         icmph_t *icmph;
3521         ipha_t  *ipha;
3522         uint_t  len_needed;
3523         ire_t   *src_ire;
3524         ire_t   *dst_ire;
3525 
3526         if (!mp)
3527                 return (NULL);
3528         ipha = (ipha_t *)mp->b_rptr;
3529         if (ip_csum_hdr(ipha)) {
3530                 BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsInCksumErrs);
3531                 freemsg(mp);
3532                 return (NULL);
3533         }
3534         src_ire = ire_ctable_lookup(ipha->ipha_dst, 0, IRE_BROADCAST,
3535             NULL, ALL_ZONES, NULL, MATCH_IRE_TYPE, ipst);
3536         dst_ire = ire_ctable_lookup(ipha->ipha_src, 0, IRE_BROADCAST,
3537             NULL, ALL_ZONES, NULL, MATCH_IRE_TYPE, ipst);
3538         if (src_ire != NULL || dst_ire != NULL ||
3539             CLASSD(ipha->ipha_dst) ||
3540             CLASSD(ipha->ipha_src) ||
3541             (ntohs(ipha->ipha_fragment_offset_and_flags) & IPH_OFFSET)) {
3542                 /* Note: only errors to the fragment with offset 0 */
3543                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
3544                 freemsg(mp);
3545                 if (src_ire != NULL)
3546                         ire_refrele(src_ire);
3547                 if (dst_ire != NULL)
3548                         ire_refrele(dst_ire);
3549                 return (NULL);
3550         }
3551         if (ipha->ipha_protocol == IPPROTO_ICMP) {
3552                 /*
3553                  * Check the ICMP type.  RFC 1122 sez:  don't send ICMP
3554                  * errors in response to any ICMP errors.
3555                  */
3556                 len_needed = IPH_HDR_LENGTH(ipha) + ICMPH_SIZE;
3557                 if (mp->b_wptr - mp->b_rptr < len_needed) {
3558                         if (!pullupmsg(mp, len_needed)) {
3559                                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInErrors);
3560                                 freemsg(mp);
3561                                 return (NULL);
3562                         }
3563                         ipha = (ipha_t *)mp->b_rptr;
3564                 }
3565                 icmph = (icmph_t *)
3566                     (&((char *)ipha)[IPH_HDR_LENGTH(ipha)]);
3567                 switch (icmph->icmph_type) {
3568                 case ICMP_DEST_UNREACHABLE:
3569                 case ICMP_SOURCE_QUENCH:
3570                 case ICMP_TIME_EXCEEDED:
3571                 case ICMP_PARAM_PROBLEM:
3572                 case ICMP_REDIRECT:
3573                         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
3574                         freemsg(mp);
3575                         return (NULL);
3576                 default:
3577                         break;
3578                 }
3579         }
3580         /*
3581          * If this is a labeled system, then check to see if we're allowed to
3582          * send a response to this particular sender.  If not, then just drop.
3583          */
3584         if (is_system_labeled() && !tsol_can_reply_error(mp)) {
3585                 ip2dbg(("icmp_pkt_err_ok: can't respond to packet\n"));
3586                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
3587                 freemsg(mp);
3588                 return (NULL);
3589         }
3590         if (icmp_err_rate_limit(ipst)) {
3591                 /*
3592                  * Only send ICMP error packets every so often.
3593                  * This should be done on a per port/source basis,
3594                  * but for now this will suffice.
3595                  */
3596                 freemsg(mp);
3597                 return (NULL);
3598         }
3599         return (mp);
3600 }
3601 
3602 /*
3603  * Generate an ICMP redirect message.
3604  */
3605 static void
3606 icmp_send_redirect(queue_t *q, mblk_t *mp, ipaddr_t gateway, ip_stack_t *ipst)
3607 {
3608         icmph_t icmph;
3609 
3610         /*
3611          * We are called from ip_rput where we could
3612          * not have attached an IPSEC_IN.
3613          */
3614         ASSERT(mp->b_datap->db_type == M_DATA);
3615 
3616         if (!(mp = icmp_pkt_err_ok(mp, ipst))) {
3617                 return;
3618         }
3619 
3620         bzero(&icmph, sizeof (icmph_t));
3621         icmph.icmph_type = ICMP_REDIRECT;
3622         icmph.icmph_code = 1;
3623         icmph.icmph_rd_gateway = gateway;
3624         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutRedirects);
3625         /* Redirects sent by router, and router is global zone */
3626         icmp_pkt(q, mp, &icmph, sizeof (icmph_t), B_FALSE, GLOBAL_ZONEID, ipst);
3627 }
3628 
3629 /*
3630  * Generate an ICMP time exceeded message.
3631  */
3632 void
3633 icmp_time_exceeded(queue_t *q, mblk_t *mp, uint8_t code, zoneid_t zoneid,
3634     ip_stack_t *ipst)
3635 {
3636         icmph_t icmph;
3637         boolean_t mctl_present;
3638         mblk_t *first_mp;
3639 
3640         EXTRACT_PKT_MP(mp, first_mp, mctl_present);
3641 
3642         if (!(mp = icmp_pkt_err_ok(mp, ipst))) {
3643                 if (mctl_present)
3644                         freeb(first_mp);
3645                 return;
3646         }
3647 
3648         bzero(&icmph, sizeof (icmph_t));
3649         icmph.icmph_type = ICMP_TIME_EXCEEDED;
3650         icmph.icmph_code = code;
3651         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutTimeExcds);
3652         icmp_pkt(q, first_mp, &icmph, sizeof (icmph_t), mctl_present, zoneid,
3653             ipst);
3654 }
3655 
3656 /*
3657  * Generate an ICMP unreachable message.
3658  */
3659 void
3660 icmp_unreachable(queue_t *q, mblk_t *mp, uint8_t code, zoneid_t zoneid,
3661     ip_stack_t *ipst)
3662 {
3663         icmph_t icmph;
3664         mblk_t *first_mp;
3665         boolean_t mctl_present;
3666 
3667         EXTRACT_PKT_MP(mp, first_mp, mctl_present);
3668 
3669         if (!(mp = icmp_pkt_err_ok(mp, ipst))) {
3670                 if (mctl_present)
3671                         freeb(first_mp);
3672                 return;
3673         }
3674 
3675         bzero(&icmph, sizeof (icmph_t));
3676         icmph.icmph_type = ICMP_DEST_UNREACHABLE;
3677         icmph.icmph_code = code;
3678         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDestUnreachs);
3679         ip2dbg(("send icmp destination unreachable code %d\n", code));
3680         icmp_pkt(q, first_mp, (char *)&icmph, sizeof (icmph_t), mctl_present,
3681             zoneid, ipst);
3682 }
3683 
3684 /*
3685  * Attempt to start recovery of an IPv4 interface that's been shut down as a
3686  * duplicate.  As long as someone else holds the address, the interface will
3687  * stay down.  When that conflict goes away, the interface is brought back up.
3688  * This is done so that accidental shutdowns of addresses aren't made
3689  * permanent.  Your server will recover from a failure.
3690  *
3691  * For DHCP, recovery is not done in the kernel.  Instead, it's handled by a
3692  * user space process (dhcpagent).
3693  *
3694  * Recovery completes if ARP reports that the address is now ours (via
3695  * AR_CN_READY).  In that case, we go to ip_arp_excl to finish the operation.
3696  *
3697  * This function is entered on a timer expiry; the ID is in ipif_recovery_id.
3698  */
3699 static void
3700 ipif_dup_recovery(void *arg)
3701 {
3702         ipif_t *ipif = arg;
3703         ill_t *ill = ipif->ipif_ill;
3704         mblk_t *arp_add_mp;
3705         mblk_t *arp_del_mp;
3706         area_t *area;
3707         ip_stack_t *ipst = ill->ill_ipst;
3708 
3709         ipif->ipif_recovery_id = 0;
3710 
3711         /*
3712          * No lock needed for moving or condemned check, as this is just an
3713          * optimization.
3714          */
3715         if (ill->ill_arp_closing || !(ipif->ipif_flags & IPIF_DUPLICATE) ||
3716             (ipif->ipif_flags & IPIF_POINTOPOINT) ||
3717             (ipif->ipif_state_flags & (IPIF_MOVING | IPIF_CONDEMNED))) {
3718                 /* No reason to try to bring this address back. */
3719                 return;
3720         }
3721 
3722         if ((arp_add_mp = ipif_area_alloc(ipif)) == NULL)
3723                 goto alloc_fail;
3724 
3725         if (ipif->ipif_arp_del_mp == NULL) {
3726                 if ((arp_del_mp = ipif_ared_alloc(ipif)) == NULL)
3727                         goto alloc_fail;
3728                 ipif->ipif_arp_del_mp = arp_del_mp;
3729         }
3730 
3731         /* Setting the 'unverified' flag restarts DAD */
3732         area = (area_t *)arp_add_mp->b_rptr;
3733         area->area_flags = ACE_F_PERMANENT | ACE_F_PUBLISH | ACE_F_MYADDR |
3734             ACE_F_UNVERIFIED;
3735         putnext(ill->ill_rq, arp_add_mp);
3736         return;
3737 
3738 alloc_fail:
3739         /*
3740          * On allocation failure, just restart the timer.  Note that the ipif
3741          * is down here, so no other thread could be trying to start a recovery
3742          * timer.  The ill_lock protects the condemned flag and the recovery
3743          * timer ID.
3744          */
3745         freemsg(arp_add_mp);
3746         mutex_enter(&ill->ill_lock);
3747         if (ipst->ips_ip_dup_recovery > 0 && ipif->ipif_recovery_id == 0 &&
3748             !(ipif->ipif_state_flags & IPIF_CONDEMNED)) {
3749                 ipif->ipif_recovery_id = timeout(ipif_dup_recovery, ipif,
3750                     MSEC_TO_TICK(ipst->ips_ip_dup_recovery));
3751         }
3752         mutex_exit(&ill->ill_lock);
3753 }
3754 
3755 /*
3756  * This is for exclusive changes due to ARP.  Either tear down an interface due
3757  * to AR_CN_FAILED and AR_CN_BOGON, or bring one up for successful recovery.
3758  */
3759 /* ARGSUSED */
3760 static void
3761 ip_arp_excl(ipsq_t *ipsq, queue_t *rq, mblk_t *mp, void *dummy_arg)
3762 {
3763         ill_t   *ill = rq->q_ptr;
3764         arh_t *arh;
3765         ipaddr_t src;
3766         ipif_t  *ipif;
3767         char ibuf[LIFNAMSIZ + 10];      /* 10 digits for logical i/f number */
3768         char hbuf[MAC_STR_LEN];
3769         char sbuf[INET_ADDRSTRLEN];
3770         const char *failtype;
3771         boolean_t bring_up;
3772         ip_stack_t *ipst = ill->ill_ipst;
3773 
3774         switch (((arcn_t *)mp->b_rptr)->arcn_code) {
3775         case AR_CN_READY:
3776                 failtype = NULL;
3777                 bring_up = B_TRUE;
3778                 break;
3779         case AR_CN_FAILED:
3780                 failtype = "in use";
3781                 bring_up = B_FALSE;
3782                 break;
3783         default:
3784                 failtype = "claimed";
3785                 bring_up = B_FALSE;
3786                 break;
3787         }
3788 
3789         arh = (arh_t *)mp->b_cont->b_rptr;
3790         bcopy((char *)&arh[1] + arh->arh_hlen, &src, IP_ADDR_LEN);
3791 
3792         (void) mac_colon_addr((uint8_t *)(arh + 1), arh->arh_hlen, hbuf,
3793             sizeof (hbuf));
3794         (void) ip_dot_addr(src, sbuf);
3795         for (ipif = ill->ill_ipif; ipif != NULL; ipif = ipif->ipif_next) {
3796 
3797                 if ((ipif->ipif_flags & IPIF_POINTOPOINT) ||
3798                     ipif->ipif_lcl_addr != src) {
3799                         continue;
3800                 }
3801 
3802                 /*
3803                  * If we failed on a recovery probe, then restart the timer to
3804                  * try again later.
3805                  */
3806                 if (!bring_up && (ipif->ipif_flags & IPIF_DUPLICATE) &&
3807                     !(ipif->ipif_flags & (IPIF_DHCPRUNNING|IPIF_TEMPORARY)) &&
3808                     ill->ill_net_type == IRE_IF_RESOLVER &&
3809                     !(ipif->ipif_state_flags & IPIF_CONDEMNED) &&
3810                     ipst->ips_ip_dup_recovery > 0 &&
3811                     ipif->ipif_recovery_id == 0) {
3812                         ipif->ipif_recovery_id = timeout(ipif_dup_recovery,
3813                             ipif, MSEC_TO_TICK(ipst->ips_ip_dup_recovery));
3814                         continue;
3815                 }
3816 
3817                 /*
3818                  * If what we're trying to do has already been done, then do
3819                  * nothing.
3820                  */
3821                 if (bring_up == ((ipif->ipif_flags & IPIF_UP) != 0))
3822                         continue;
3823 
3824                 ipif_get_name(ipif, ibuf, sizeof (ibuf));
3825 
3826                 if (failtype == NULL) {
3827                         cmn_err(CE_NOTE, "recovered address %s on %s", sbuf,
3828                             ibuf);
3829                 } else {
3830                         cmn_err(CE_WARN, "%s has duplicate address %s (%s "
3831                             "by %s); disabled", ibuf, sbuf, failtype, hbuf);
3832                 }
3833 
3834                 if (bring_up) {
3835                         ASSERT(ill->ill_dl_up);
3836                         /*
3837                          * Free up the ARP delete message so we can allocate
3838                          * a fresh one through the normal path.
3839                          */
3840                         freemsg(ipif->ipif_arp_del_mp);
3841                         ipif->ipif_arp_del_mp = NULL;
3842                         if (ipif_resolver_up(ipif, Res_act_initial) !=
3843                             EINPROGRESS) {
3844                                 ipif->ipif_addr_ready = 1;
3845                                 (void) ipif_up_done(ipif);
3846                         }
3847                         continue;
3848                 }
3849 
3850                 mutex_enter(&ill->ill_lock);
3851                 ASSERT(!(ipif->ipif_flags & IPIF_DUPLICATE));
3852                 ipif->ipif_flags |= IPIF_DUPLICATE;
3853                 ill->ill_ipif_dup_count++;
3854                 mutex_exit(&ill->ill_lock);
3855                 /*
3856                  * Already exclusive on the ill; no need to handle deferred
3857                  * processing here.
3858                  */
3859                 (void) ipif_down(ipif, NULL, NULL);
3860                 ipif_down_tail(ipif);
3861                 mutex_enter(&ill->ill_lock);
3862                 if (!(ipif->ipif_flags & (IPIF_DHCPRUNNING|IPIF_TEMPORARY)) &&
3863                     ill->ill_net_type == IRE_IF_RESOLVER &&
3864                     !(ipif->ipif_state_flags & IPIF_CONDEMNED) &&
3865                     ipst->ips_ip_dup_recovery > 0) {
3866                         ipif->ipif_recovery_id = timeout(ipif_dup_recovery,
3867                             ipif, MSEC_TO_TICK(ipst->ips_ip_dup_recovery));
3868                 }
3869                 mutex_exit(&ill->ill_lock);
3870         }
3871         freemsg(mp);
3872 }
3873 
3874 /* ARGSUSED */
3875 static void
3876 ip_arp_defend(ipsq_t *ipsq, queue_t *rq, mblk_t *mp, void *dummy_arg)
3877 {
3878         ill_t   *ill = rq->q_ptr;
3879         arh_t *arh;
3880         ipaddr_t src;
3881         ipif_t  *ipif;
3882 
3883         arh = (arh_t *)mp->b_cont->b_rptr;
3884         bcopy((char *)&arh[1] + arh->arh_hlen, &src, IP_ADDR_LEN);
3885         for (ipif = ill->ill_ipif; ipif != NULL; ipif = ipif->ipif_next) {
3886                 if ((ipif->ipif_flags & IPIF_UP) && ipif->ipif_lcl_addr == src)
3887                         (void) ipif_resolver_up(ipif, Res_act_defend);
3888         }
3889         freemsg(mp);
3890 }
3891 
3892 /*
3893  * News from ARP.  ARP sends notification of interesting events down
3894  * to its clients using M_CTL messages with the interesting ARP packet
3895  * attached via b_cont.
3896  * The interesting event from a device comes up the corresponding ARP-IP-DEV
3897  * queue as opposed to ARP sending the message to all the clients, i.e. all
3898  * its ARP-IP-DEV instances. Thus, for AR_CN_ANNOUNCE, we must walk the cache
3899  * table if a cache IRE is found to delete all the entries for the address in
3900  * the packet.
3901  */
3902 static void
3903 ip_arp_news(queue_t *q, mblk_t *mp)
3904 {
3905         arcn_t          *arcn;
3906         arh_t           *arh;
3907         ire_t           *ire = NULL;
3908         char            hbuf[MAC_STR_LEN];
3909         char            sbuf[INET_ADDRSTRLEN];
3910         ipaddr_t        src;
3911         in6_addr_t      v6src;
3912         boolean_t       isv6 = B_FALSE;
3913         ipif_t          *ipif;
3914         ill_t           *ill;
3915         ip_stack_t      *ipst;
3916 
3917         if (CONN_Q(q)) {
3918                 conn_t *connp = Q_TO_CONN(q);
3919 
3920                 ipst = connp->conn_netstack->netstack_ip;
3921         } else {
3922                 ill_t *ill = (ill_t *)q->q_ptr;
3923 
3924                 ipst = ill->ill_ipst;
3925         }
3926 
3927         if ((mp->b_wptr - mp->b_rptr) < sizeof (arcn_t)        || !mp->b_cont) {
3928                 if (q->q_next) {
3929                         putnext(q, mp);
3930                 } else
3931                         freemsg(mp);
3932                 return;
3933         }
3934         arh = (arh_t *)mp->b_cont->b_rptr;
3935         /* Is it one we are interested in? */
3936         if (BE16_TO_U16(arh->arh_proto) == IP6_DL_SAP) {
3937                 isv6 = B_TRUE;
3938                 bcopy((char *)&arh[1] + (arh->arh_hlen & 0xFF), &v6src,
3939                     IPV6_ADDR_LEN);
3940         } else if (BE16_TO_U16(arh->arh_proto) == IP_ARP_PROTO_TYPE) {
3941                 bcopy((char *)&arh[1] + (arh->arh_hlen & 0xFF), &src,
3942                     IP_ADDR_LEN);
3943         } else {
3944                 freemsg(mp);
3945                 return;
3946         }
3947 
3948         ill = q->q_ptr;
3949 
3950         arcn = (arcn_t *)mp->b_rptr;
3951         switch (arcn->arcn_code) {
3952         case AR_CN_BOGON:
3953                 /*
3954                  * Someone is sending ARP packets with a source protocol
3955                  * address that we have published and for which we believe our
3956                  * entry is authoritative and (when ill_arp_extend is set)
3957                  * verified to be unique on the network.
3958                  *
3959                  * The ARP module internally handles the cases where the sender
3960                  * is just probing (for DAD) and where the hardware address of
3961                  * a non-authoritative entry has changed.  Thus, these are the
3962                  * real conflicts, and we have to do resolution.
3963                  *
3964                  * We back away quickly from the address if it's from DHCP or
3965                  * otherwise temporary and hasn't been used recently (or at
3966                  * all).  We'd like to include "deprecated" addresses here as
3967                  * well (as there's no real reason to defend something we're
3968                  * discarding), but IPMP "reuses" this flag to mean something
3969                  * other than the standard meaning.
3970                  *
3971                  * If the ARP module above is not extended (meaning that it
3972                  * doesn't know how to defend the address), then we just log
3973                  * the problem as we always did and continue on.  It's not
3974                  * right, but there's little else we can do, and those old ATM
3975                  * users are going away anyway.
3976                  */
3977                 (void) mac_colon_addr((uint8_t *)(arh + 1), arh->arh_hlen,
3978                     hbuf, sizeof (hbuf));
3979                 (void) ip_dot_addr(src, sbuf);
3980                 if (isv6) {
3981                         ire = ire_cache_lookup_v6(&v6src, ALL_ZONES, NULL,
3982                             ipst);
3983                 } else {
3984                         ire = ire_cache_lookup(src, ALL_ZONES, NULL, ipst);
3985                 }
3986                 if (ire != NULL && IRE_IS_LOCAL(ire)) {
3987                         uint32_t now;
3988                         uint32_t maxage;
3989                         clock_t lused;
3990                         uint_t maxdefense;
3991                         uint_t defs;
3992 
3993                         /*
3994                          * First, figure out if this address hasn't been used
3995                          * in a while.  If it hasn't, then it's a better
3996                          * candidate for abandoning.
3997                          */
3998                         ipif = ire->ire_ipif;
3999                         ASSERT(ipif != NULL);
4000                         now = gethrestime_sec();
4001                         maxage = now - ire->ire_create_time;
4002                         if (maxage > ipst->ips_ip_max_temp_idle)
4003                                 maxage = ipst->ips_ip_max_temp_idle;
4004                         lused = drv_hztousec(ddi_get_lbolt() -
4005                             ire->ire_last_used_time) / MICROSEC + 1;
4006                         if (lused >= maxage && (ipif->ipif_flags &
4007                             (IPIF_DHCPRUNNING | IPIF_TEMPORARY)))
4008                                 maxdefense = ipst->ips_ip_max_temp_defend;
4009                         else
4010                                 maxdefense = ipst->ips_ip_max_defend;
4011 
4012                         /*
4013                          * Now figure out how many times we've defended
4014                          * ourselves.  Ignore defenses that happened long in
4015                          * the past.
4016                          */
4017                         mutex_enter(&ire->ire_lock);
4018                         if ((defs = ire->ire_defense_count) > 0 &&
4019                             now - ire->ire_defense_time >
4020                             ipst->ips_ip_defend_interval) {
4021                                 ire->ire_defense_count = defs = 0;
4022                         }
4023                         ire->ire_defense_count++;
4024                         ire->ire_defense_time = now;
4025                         mutex_exit(&ire->ire_lock);
4026                         ill_refhold(ill);
4027                         ire_refrele(ire);
4028 
4029                         /*
4030                          * If we've defended ourselves too many times already,
4031                          * then give up and tear down the interface(s) using
4032                          * this address.  Otherwise, defend by sending out a
4033                          * gratuitous ARP.
4034                          */
4035                         if (defs >= maxdefense && ill->ill_arp_extend) {
4036                                 qwriter_ip(ill, q, mp, ip_arp_excl, NEW_OP,
4037                                     B_FALSE);
4038                         } else {
4039                                 cmn_err(CE_WARN,
4040                                     "node %s is using our IP address %s on %s",
4041                                     hbuf, sbuf, ill->ill_name);
4042                                 /*
4043                                  * If this is an old (ATM) ARP module, then
4044                                  * don't try to defend the address.  Remain
4045                                  * compatible with the old behavior.  Defend
4046                                  * only with new ARP.
4047                                  */
4048                                 if (ill->ill_arp_extend) {
4049                                         qwriter_ip(ill, q, mp, ip_arp_defend,
4050                                             NEW_OP, B_FALSE);
4051                                 } else {
4052                                         ill_refrele(ill);
4053                                 }
4054                         }
4055                         return;
4056                 }
4057                 cmn_err(CE_WARN,
4058                     "proxy ARP problem?  Node '%s' is using %s on %s",
4059                     hbuf, sbuf, ill->ill_name);
4060                 if (ire != NULL)
4061                         ire_refrele(ire);
4062                 break;
4063         case AR_CN_ANNOUNCE:
4064                 if (isv6) {
4065                         /*
4066                          * For XRESOLV interfaces.
4067                          * Delete the IRE cache entry and NCE for this
4068                          * v6 address
4069                          */
4070                         ip_ire_clookup_and_delete_v6(&v6src, ipst);
4071                         /*
4072                          * If v6src is a non-zero, it's a router address
4073                          * as below. Do the same sort of thing to clean
4074                          * out off-net IRE_CACHE entries that go through
4075                          * the router.
4076                          */
4077                         if (!IN6_IS_ADDR_UNSPECIFIED(&v6src)) {
4078                                 ire_walk_v6(ire_delete_cache_gw_v6,
4079                                     (char *)&v6src, ALL_ZONES, ipst);
4080                         }
4081                 } else {
4082                         nce_hw_map_t hwm;
4083 
4084                         /*
4085                          * ARP gives us a copy of any packet where it thinks
4086                          * the address has changed, so that we can update our
4087                          * caches.  We're responsible for caching known answers
4088                          * in the current design.  We check whether the
4089                          * hardware address really has changed in all of our
4090                          * entries that have cached this mapping, and if so, we
4091                          * blow them away.  This way we will immediately pick
4092                          * up the rare case of a host changing hardware
4093                          * address.
4094                          */
4095                         if (src == 0)
4096                                 break;
4097                         hwm.hwm_addr = src;
4098                         hwm.hwm_hwlen = arh->arh_hlen;
4099                         hwm.hwm_hwaddr = (uchar_t *)(arh + 1);
4100                         NDP_HW_CHANGE_INCR(ipst->ips_ndp4);
4101                         ndp_walk_common(ipst->ips_ndp4, NULL,
4102                             (pfi_t)nce_delete_hw_changed, &hwm, ALL_ZONES);
4103                         NDP_HW_CHANGE_DECR(ipst->ips_ndp4);
4104                 }
4105                 break;
4106         case AR_CN_READY:
4107                 /* No external v6 resolver has a contract to use this */
4108                 if (isv6)
4109                         break;
4110                 /* If the link is down, we'll retry this later */
4111                 if (!(ill->ill_phyint->phyint_flags & PHYI_RUNNING))
4112                         break;
4113                 ipif = ipif_lookup_addr(src, ill, ALL_ZONES, NULL, NULL,
4114                     NULL, NULL, ipst);
4115                 if (ipif != NULL) {
4116                         /*
4117                          * If this is a duplicate recovery, then we now need to
4118                          * go exclusive to bring this thing back up.
4119                          */
4120                         if ((ipif->ipif_flags & (IPIF_UP|IPIF_DUPLICATE)) ==
4121                             IPIF_DUPLICATE) {
4122                                 ipif_refrele(ipif);
4123                                 ill_refhold(ill);
4124                                 qwriter_ip(ill, q, mp, ip_arp_excl, NEW_OP,
4125                                     B_FALSE);
4126                                 return;
4127                         }
4128                         /*
4129                          * If this is the first notice that this address is
4130                          * ready, then let the user know now.
4131                          */
4132                         if ((ipif->ipif_flags & IPIF_UP) &&
4133                             !ipif->ipif_addr_ready) {
4134                                 ipif_mask_reply(ipif);
4135                                 ip_rts_ifmsg(ipif);
4136                                 ip_rts_newaddrmsg(RTM_ADD, 0, ipif);
4137                                 sctp_update_ipif(ipif, SCTP_IPIF_UP);
4138                         }
4139                         ipif->ipif_addr_ready = 1;
4140                         ipif_refrele(ipif);
4141                 }
4142                 ire = ire_cache_lookup(src, ALL_ZONES, MBLK_GETLABEL(mp), ipst);
4143                 if (ire != NULL) {
4144                         ire->ire_defense_count = 0;
4145                         ire_refrele(ire);
4146                 }
4147                 break;
4148         case AR_CN_FAILED:
4149                 /* No external v6 resolver has a contract to use this */
4150                 if (isv6)
4151                         break;
4152                 ill_refhold(ill);
4153                 qwriter_ip(ill, q, mp, ip_arp_excl, NEW_OP, B_FALSE);
4154                 return;
4155         }
4156         freemsg(mp);
4157 }
4158 
4159 /*
4160  * Create a mblk suitable for carrying the interface index and/or source link
4161  * address. This mblk is tagged as an M_CTL and is sent to ULP. This is used
4162  * when the IP_RECVIF and/or IP_RECVSLLA socket option is set by the user
4163  * application.
4164  */
4165 mblk_t *
4166 ip_add_info(mblk_t *data_mp, ill_t *ill, uint_t flags, zoneid_t zoneid,
4167     ip_stack_t *ipst)
4168 {
4169         mblk_t          *mp;
4170         ip_pktinfo_t    *pinfo;
4171         ipha_t *ipha;
4172         struct ether_header *pether;
4173 
4174         mp = allocb(sizeof (ip_pktinfo_t), BPRI_MED);
4175         if (mp == NULL) {
4176                 ip1dbg(("ip_add_info: allocation failure.\n"));
4177                 return (data_mp);
4178         }
4179 
4180         ipha    = (ipha_t *)data_mp->b_rptr;
4181         pinfo = (ip_pktinfo_t *)mp->b_rptr;
4182         bzero(pinfo, sizeof (ip_pktinfo_t));
4183         pinfo->ip_pkt_flags = (uchar_t)flags;
4184         pinfo->ip_pkt_ulp_type = IN_PKTINFO; /* Tell ULP what type of info */
4185 
4186         if (flags & (IPF_RECVIF | IPF_RECVADDR))
4187                 pinfo->ip_pkt_ifindex = ill->ill_phyint->phyint_ifindex;
4188         if (flags & IPF_RECVADDR) {
4189                 ipif_t  *ipif;
4190                 ire_t   *ire;
4191 
4192                 /*
4193                  * Only valid for V4
4194                  */
4195                 ASSERT((ipha->ipha_version_and_hdr_length & 0xf0) ==
4196                     (IPV4_VERSION << 4));
4197 
4198                 ipif = ipif_get_next_ipif(NULL, ill);
4199                 if (ipif != NULL) {
4200                         /*
4201                          * Since a decision has already been made to deliver the
4202                          * packet, there is no need to test for SECATTR and
4203                          * ZONEONLY.
4204                          * When a multicast packet is transmitted
4205                          * a cache entry is created for the multicast address.
4206                          * When delivering a copy of the packet or when new
4207                          * packets are received we do not want to match on the
4208                          * cached entry so explicitly match on
4209                          * IRE_LOCAL and IRE_LOOPBACK
4210                          */
4211                         ire = ire_ctable_lookup(ipha->ipha_dst, 0,
4212                             IRE_LOCAL | IRE_LOOPBACK,
4213                             ipif, zoneid, NULL,
4214                             MATCH_IRE_TYPE | MATCH_IRE_ILL_GROUP, ipst);
4215                         if (ire == NULL) {
4216                                 /*
4217                                  * packet must have come on a different
4218                                  * interface.
4219                                  * Since a decision has already been made to
4220                                  * deliver the packet, there is no need to test
4221                                  * for SECATTR and ZONEONLY.
4222                                  * Only match on local and broadcast ire's.
4223                                  * See detailed comment above.
4224                                  */
4225                                 ire = ire_ctable_lookup(ipha->ipha_dst, 0,
4226                                     IRE_LOCAL | IRE_LOOPBACK, ipif, zoneid,
4227                                     NULL, MATCH_IRE_TYPE, ipst);
4228                         }
4229 
4230                         if (ire == NULL) {
4231                                 /*
4232                                  * This is either a multicast packet or
4233                                  * the address has been removed since
4234                                  * the packet was received.
4235                                  * Return INADDR_ANY so that normal source
4236                                  * selection occurs for the response.
4237                                  */
4238 
4239                                 pinfo->ip_pkt_match_addr.s_addr = INADDR_ANY;
4240                         } else {
4241                                 pinfo->ip_pkt_match_addr.s_addr =
4242                                     ire->ire_src_addr;
4243                                 ire_refrele(ire);
4244                         }
4245                         ipif_refrele(ipif);
4246                 } else {
4247                         pinfo->ip_pkt_match_addr.s_addr = INADDR_ANY;
4248                 }
4249         }
4250 
4251         pether = (struct ether_header *)((char *)ipha
4252             - sizeof (struct ether_header));
4253         /*
4254          * Make sure the interface is an ethernet type, since this option
4255          * is currently supported only on this type of interface. Also make
4256          * sure we are pointing correctly above db_base.
4257          */
4258 
4259         if ((flags & IPF_RECVSLLA) &&
4260             ((uchar_t *)pether >= data_mp->b_datap->db_base) &&
4261             (ill->ill_type == IFT_ETHER) &&
4262             (ill->ill_net_type == IRE_IF_RESOLVER)) {
4263 
4264                 pinfo->ip_pkt_slla.sdl_type = IFT_ETHER;
4265                 bcopy((uchar_t *)pether->ether_shost.ether_addr_octet,
4266                     (uchar_t *)pinfo->ip_pkt_slla.sdl_data, ETHERADDRL);
4267         } else {
4268                 /*
4269                  * Clear the bit. Indicate to upper layer that IP is not
4270                  * sending this ancillary info.
4271                  */
4272                 pinfo->ip_pkt_flags = pinfo->ip_pkt_flags & ~IPF_RECVSLLA;
4273         }
4274 
4275         mp->b_datap->db_type = M_CTL;
4276         mp->b_wptr += sizeof (ip_pktinfo_t);
4277         mp->b_cont = data_mp;
4278 
4279         return (mp);
4280 }
4281 
4282 /*
4283  * Latch in the IPsec state for a stream based on the ipsec_in_t passed in as
4284  * part of the bind request.
4285  */
4286 
4287 boolean_t
4288 ip_bind_ipsec_policy_set(conn_t *connp, mblk_t *policy_mp)
4289 {
4290         ipsec_in_t *ii;
4291 
4292         ASSERT(policy_mp != NULL);
4293         ASSERT(policy_mp->b_datap->db_type == IPSEC_POLICY_SET);
4294 
4295         ii = (ipsec_in_t *)policy_mp->b_rptr;
4296         ASSERT(ii->ipsec_in_type == IPSEC_IN);
4297 
4298         connp->conn_policy = ii->ipsec_in_policy;
4299         ii->ipsec_in_policy = NULL;
4300 
4301         if (ii->ipsec_in_action != NULL) {
4302                 if (connp->conn_latch == NULL) {
4303                         connp->conn_latch = iplatch_create();
4304                         if (connp->conn_latch == NULL)
4305                                 return (B_FALSE);
4306                 }
4307                 ipsec_latch_inbound(connp->conn_latch, ii);
4308         }
4309         return (B_TRUE);
4310 }
4311 
4312 /*
4313  * Upper level protocols (ULP) pass through bind requests to IP for inspection
4314  * and to arrange for power-fanout assist.  The ULP is identified by
4315  * adding a single byte at the end of the original bind message.
4316  * A ULP other than UDP or TCP that wishes to be recognized passes
4317  * down a bind with a zero length address.
4318  *
4319  * The binding works as follows:
4320  * - A zero byte address means just bind to the protocol.
4321  * - A four byte address is treated as a request to validate
4322  *   that the address is a valid local address, appropriate for
4323  *   an application to bind to. This does not affect any fanout
4324  *   information in IP.
4325  * - A sizeof sin_t byte address is used to bind to only the local address
4326  *   and port.
4327  * - A sizeof ipa_conn_t byte address contains complete fanout information
4328  *   consisting of local and remote addresses and ports.  In
4329  *   this case, the addresses are both validated as appropriate
4330  *   for this operation, and, if so, the information is retained
4331  *   for use in the inbound fanout.
4332  *
4333  * The ULP (except in the zero-length bind) can append an
4334  * additional mblk of db_type IRE_DB_REQ_TYPE or IPSEC_POLICY_SET to the
4335  * T_BIND_REQ/O_T_BIND_REQ. IRE_DB_REQ_TYPE indicates that the ULP wants
4336  * a copy of the source or destination IRE (source for local bind;
4337  * destination for complete bind). IPSEC_POLICY_SET indicates that the
4338  * policy information contained should be copied on to the conn.
4339  *
4340  * NOTE : Only one of IRE_DB_REQ_TYPE or IPSEC_POLICY_SET can be present.
4341  */
4342 mblk_t *
4343 ip_bind_v4(queue_t *q, mblk_t *mp, conn_t *connp)
4344 {
4345         ssize_t         len;
4346         struct T_bind_req       *tbr;
4347         sin_t           *sin;
4348         ipa_conn_t      *ac;
4349         uchar_t         *ucp;
4350         mblk_t          *mp1;
4351         boolean_t       ire_requested;
4352         boolean_t       ipsec_policy_set = B_FALSE;
4353         int             error = 0;
4354         int             protocol;
4355         ipa_conn_x_t    *acx;
4356 
4357         ASSERT(!connp->conn_af_isv6);
4358         connp->conn_pkt_isv6 = B_FALSE;
4359 
4360         len = MBLKL(mp);
4361         if (len < (sizeof (*tbr) + 1)) {
4362                 (void) mi_strlog(q, 1, SL_ERROR|SL_TRACE,
4363                     "ip_bind: bogus msg, len %ld", len);
4364                 /* XXX: Need to return something better */
4365                 goto bad_addr;
4366         }
4367         /* Back up and extract the protocol identifier. */
4368         mp->b_wptr--;
4369         protocol = *mp->b_wptr & 0xFF;
4370         tbr = (struct T_bind_req *)mp->b_rptr;
4371         /* Reset the message type in preparation for shipping it back. */
4372         DB_TYPE(mp) = M_PCPROTO;
4373 
4374         connp->conn_ulp = (uint8_t)protocol;
4375 
4376         /*
4377          * Check for a zero length address.  This is from a protocol that
4378          * wants to register to receive all packets of its type.
4379          */
4380         if (tbr->ADDR_length == 0) {
4381                 /*
4382                  * These protocols are now intercepted in ip_bind_v6().
4383                  * Reject protocol-level binds here for now.
4384                  *
4385                  * For SCTP raw socket, ICMP sends down a bind with sin_t
4386                  * so that the protocol type cannot be SCTP.
4387                  */
4388                 if (protocol == IPPROTO_TCP || protocol == IPPROTO_AH ||
4389                     protocol == IPPROTO_ESP || protocol == IPPROTO_SCTP) {
4390                         goto bad_addr;
4391                 }
4392 
4393                 /*
4394                  *
4395                  * The udp module never sends down a zero-length address,
4396                  * and allowing this on a labeled system will break MLP
4397                  * functionality.
4398                  */
4399                 if (is_system_labeled() && protocol == IPPROTO_UDP)
4400                         goto bad_addr;
4401 
4402                 if (connp->conn_mac_exempt)
4403                         goto bad_addr;
4404 
4405                 /* No hash here really.  The table is big enough. */
4406                 connp->conn_srcv6 = ipv6_all_zeros;
4407 
4408                 ipcl_proto_insert(connp, protocol);
4409 
4410                 tbr->PRIM_type = T_BIND_ACK;
4411                 return (mp);
4412         }
4413 
4414         /* Extract the address pointer from the message. */
4415         ucp = (uchar_t *)mi_offset_param(mp, tbr->ADDR_offset,
4416             tbr->ADDR_length);
4417         if (ucp == NULL) {
4418                 ip1dbg(("ip_bind: no address\n"));
4419                 goto bad_addr;
4420         }
4421         if (!OK_32PTR(ucp)) {
4422                 ip1dbg(("ip_bind: unaligned address\n"));
4423                 goto bad_addr;
4424         }
4425         /*
4426          * Check for trailing mps.
4427          */
4428 
4429         mp1 = mp->b_cont;
4430         ire_requested = (mp1 != NULL && DB_TYPE(mp1) == IRE_DB_REQ_TYPE);
4431         ipsec_policy_set = (mp1 != NULL && DB_TYPE(mp1) == IPSEC_POLICY_SET);
4432 
4433         switch (tbr->ADDR_length) {
4434         default:
4435                 ip1dbg(("ip_bind: bad address length %d\n",
4436                     (int)tbr->ADDR_length));
4437                 goto bad_addr;
4438 
4439         case IP_ADDR_LEN:
4440                 /* Verification of local address only */
4441                 error = ip_bind_laddr(connp, mp, *(ipaddr_t *)ucp, 0,
4442                     ire_requested, ipsec_policy_set, B_FALSE);
4443                 break;
4444 
4445         case sizeof (sin_t):
4446                 sin = (sin_t *)ucp;
4447                 error = ip_bind_laddr(connp, mp, sin->sin_addr.s_addr,
4448                     sin->sin_port, ire_requested, ipsec_policy_set, B_TRUE);
4449                 break;
4450 
4451         case sizeof (ipa_conn_t):
4452                 ac = (ipa_conn_t *)ucp;
4453                 /* For raw socket, the local port is not set. */
4454                 if (ac->ac_lport == 0)
4455                         ac->ac_lport = connp->conn_lport;
4456                 /* Always verify destination reachability. */
4457                 error = ip_bind_connected(connp, mp, &ac->ac_laddr,
4458                     ac->ac_lport, ac->ac_faddr, ac->ac_fport, ire_requested,
4459                     ipsec_policy_set, B_TRUE, B_TRUE);
4460                 break;
4461 
4462         case sizeof (ipa_conn_x_t):
4463                 acx = (ipa_conn_x_t *)ucp;
4464                 /*
4465                  * Whether or not to verify destination reachability depends
4466                  * on the setting of the ACX_VERIFY_DST flag in acx->acx_flags.
4467                  */
4468                 error = ip_bind_connected(connp, mp, &acx->acx_conn.ac_laddr,
4469                     acx->acx_conn.ac_lport, acx->acx_conn.ac_faddr,
4470                     acx->acx_conn.ac_fport, ire_requested, ipsec_policy_set,
4471                     B_TRUE, (acx->acx_flags & ACX_VERIFY_DST) != 0);
4472                 break;
4473         }
4474         if (error == EINPROGRESS)
4475                 return (NULL);
4476         else if (error != 0)
4477                 goto bad_addr;
4478         /*
4479          * Pass the IPsec headers size in ire_ipsec_overhead.
4480          * We can't do this in ip_bind_insert_ire because the policy
4481          * may not have been inherited at that point in time and hence
4482          * conn_out_enforce_policy may not be set.
4483          */
4484         mp1 = mp->b_cont;
4485         if (ire_requested && connp->conn_out_enforce_policy &&
4486             mp1 != NULL && DB_TYPE(mp1) == IRE_DB_REQ_TYPE) {
4487                 ire_t *ire = (ire_t *)mp1->b_rptr;
4488                 ASSERT(MBLKL(mp1) >= sizeof (ire_t));
4489                 ire->ire_ipsec_overhead = conn_ipsec_length(connp);
4490         }
4491 
4492         /* Send it home. */
4493         mp->b_datap->db_type = M_PCPROTO;
4494         tbr->PRIM_type = T_BIND_ACK;
4495         return (mp);
4496 
4497 bad_addr:
4498         /*
4499          * If error = -1 then we generate a TBADADDR - otherwise error is
4500          * a unix errno.
4501          */
4502         if (error > 0)
4503                 mp = mi_tpi_err_ack_alloc(mp, TSYSERR, error);
4504         else
4505                 mp = mi_tpi_err_ack_alloc(mp, TBADADDR, 0);
4506         return (mp);
4507 }
4508 
4509 /*
4510  * Here address is verified to be a valid local address.
4511  * If the IRE_DB_REQ_TYPE mp is present, a broadcast/multicast
4512  * address is also considered a valid local address.
4513  * In the case of a broadcast/multicast address, however, the
4514  * upper protocol is expected to reset the src address
4515  * to 0 if it sees a IRE_BROADCAST type returned so that
4516  * no packets are emitted with broadcast/multicast address as
4517  * source address (that violates hosts requirements RFC1122)
4518  * The addresses valid for bind are:
4519  *      (1) - INADDR_ANY (0)
4520  *      (2) - IP address of an UP interface
4521  *      (3) - IP address of a DOWN interface
4522  *      (4) - valid local IP broadcast addresses. In this case
4523  *      the conn will only receive packets destined to
4524  *      the specified broadcast address.
4525  *      (5) - a multicast address. In this case
4526  *      the conn will only receive packets destined to
4527  *      the specified multicast address. Note: the
4528  *      application still has to issue an
4529  *      IP_ADD_MEMBERSHIP socket option.
4530  *
4531  * On error, return -1 for TBADADDR otherwise pass the
4532  * errno with TSYSERR reply.
4533  *
4534  * In all the above cases, the bound address must be valid in the current zone.
4535  * When the address is loopback, multicast or broadcast, there might be many
4536  * matching IREs so bind has to look up based on the zone.
4537  *
4538  * Note: lport is in network byte order.
4539  */
4540 int
4541 ip_bind_laddr(conn_t *connp, mblk_t *mp, ipaddr_t src_addr, uint16_t lport,
4542     boolean_t ire_requested, boolean_t ipsec_policy_set,
4543     boolean_t fanout_insert)
4544 {
4545         int             error = 0;
4546         ire_t           *src_ire;
4547         mblk_t          *policy_mp;
4548         ipif_t          *ipif;
4549         zoneid_t        zoneid;
4550         ip_stack_t      *ipst = connp->conn_netstack->netstack_ip;
4551 
4552         if (ipsec_policy_set) {
4553                 policy_mp = mp->b_cont;
4554         }
4555 
4556         /*
4557          * If it was previously connected, conn_fully_bound would have
4558          * been set.
4559          */
4560         connp->conn_fully_bound = B_FALSE;
4561 
4562         src_ire = NULL;
4563         ipif = NULL;
4564 
4565         zoneid = IPCL_ZONEID(connp);
4566 
4567         if (src_addr) {
4568                 src_ire = ire_route_lookup(src_addr, 0, 0, 0,
4569                     NULL, NULL, zoneid, NULL, MATCH_IRE_ZONEONLY, ipst);
4570                 /*
4571                  * If an address other than 0.0.0.0 is requested,
4572                  * we verify that it is a valid address for bind
4573                  * Note: Following code is in if-else-if form for
4574                  * readability compared to a condition check.
4575                  */
4576                 /* LINTED - statement has no consequent */
4577                 if (IRE_IS_LOCAL(src_ire)) {
4578                         /*
4579                          * (2) Bind to address of local UP interface
4580                          */
4581                 } else if (src_ire && src_ire->ire_type == IRE_BROADCAST) {
4582                         /*
4583                          * (4) Bind to broadcast address
4584                          * Note: permitted only from transports that
4585                          * request IRE
4586                          */
4587                         if (!ire_requested)
4588                                 error = EADDRNOTAVAIL;
4589                 } else {
4590                         /*
4591                          * (3) Bind to address of local DOWN interface
4592                          * (ipif_lookup_addr() looks up all interfaces
4593                          * but we do not get here for UP interfaces
4594                          * - case (2) above)
4595                          * We put the protocol byte back into the mblk
4596                          * since we may come back via ip_wput_nondata()
4597                          * later with this mblk if ipif_lookup_addr chooses
4598                          * to defer processing.
4599                          */
4600                         *mp->b_wptr++ = (char)connp->conn_ulp;
4601                         if ((ipif = ipif_lookup_addr(src_addr, NULL, zoneid,
4602                             CONNP_TO_WQ(connp), mp, ip_wput_nondata,
4603                             &error, ipst)) != NULL) {
4604                                 ipif_refrele(ipif);
4605                         } else if (error == EINPROGRESS) {
4606                                 if (src_ire != NULL)
4607                                         ire_refrele(src_ire);
4608                                 return (EINPROGRESS);
4609                         } else if (CLASSD(src_addr)) {
4610                                 error = 0;
4611                                 if (src_ire != NULL)
4612                                         ire_refrele(src_ire);
4613                                 /*
4614                                  * (5) bind to multicast address.
4615                                  * Fake out the IRE returned to upper
4616                                  * layer to be a broadcast IRE.
4617                                  */
4618                                 src_ire = ire_ctable_lookup(
4619                                     INADDR_BROADCAST, INADDR_ANY,
4620                                     IRE_BROADCAST, NULL, zoneid, NULL,
4621                                     (MATCH_IRE_TYPE | MATCH_IRE_ZONEONLY),
4622                                     ipst);
4623                                 if (src_ire == NULL || !ire_requested)
4624                                         error = EADDRNOTAVAIL;
4625                         } else {
4626                                 /*
4627                                  * Not a valid address for bind
4628                                  */
4629                                 error = EADDRNOTAVAIL;
4630                         }
4631                         /*
4632                          * Just to keep it consistent with the processing in
4633                          * ip_bind_v4()
4634                          */
4635                         mp->b_wptr--;
4636                 }
4637                 if (error) {
4638                         /* Red Alert!  Attempting to be a bogon! */
4639                         ip1dbg(("ip_bind: bad src address 0x%x\n",
4640                             ntohl(src_addr)));
4641                         goto bad_addr;
4642                 }
4643         }
4644 
4645         /*
4646          * Allow setting new policies. For example, disconnects come
4647          * down as ipa_t bind. As we would have set conn_policy_cached
4648          * to B_TRUE before, we should set it to B_FALSE, so that policy
4649          * can change after the disconnect.
4650          */
4651         connp->conn_policy_cached = B_FALSE;
4652 
4653         /*
4654          * If not fanout_insert this was just an address verification
4655          */
4656         if (fanout_insert) {
4657                 /*
4658                  * The addresses have been verified. Time to insert in
4659                  * the correct fanout list.
4660                  */
4661                 IN6_IPADDR_TO_V4MAPPED(src_addr, &connp->conn_srcv6);
4662                 IN6_IPADDR_TO_V4MAPPED(INADDR_ANY, &connp->conn_remv6);
4663                 connp->conn_lport = lport;
4664                 connp->conn_fport = 0;
4665                 /*
4666                  * Do we need to add a check to reject Multicast packets
4667                  */
4668                 error = ipcl_bind_insert(connp, *mp->b_wptr, src_addr, lport);
4669         }
4670 
4671         if (error == 0) {
4672                 if (ire_requested) {
4673                         if (!ip_bind_insert_ire(mp, src_ire, NULL, ipst)) {
4674                                 error = -1;
4675                                 /* Falls through to bad_addr */
4676                         }
4677                 } else if (ipsec_policy_set) {
4678                         if (!ip_bind_ipsec_policy_set(connp, policy_mp)) {
4679                                 error = -1;
4680                                 /* Falls through to bad_addr */
4681                         }
4682                 }
4683         }
4684 bad_addr:
4685         if (error != 0) {
4686                 if (connp->conn_anon_port) {
4687                         (void) tsol_mlp_anon(crgetzone(connp->conn_cred),
4688                             connp->conn_mlp_type, connp->conn_ulp, ntohs(lport),
4689                             B_FALSE);
4690                 }
4691                 connp->conn_mlp_type = mlptSingle;
4692         }
4693         if (src_ire != NULL)
4694                 IRE_REFRELE(src_ire);
4695         if (ipsec_policy_set) {
4696                 ASSERT(policy_mp == mp->b_cont);
4697                 ASSERT(policy_mp != NULL);
4698                 freeb(policy_mp);
4699                 /*
4700                  * As of now assume that nothing else accompanies
4701                  * IPSEC_POLICY_SET.
4702                  */
4703                 mp->b_cont = NULL;
4704         }
4705         return (error);
4706 }
4707 
4708 /*
4709  * Verify that both the source and destination addresses
4710  * are valid.  If verify_dst is false, then the destination address may be
4711  * unreachable, i.e. have no route to it.  Protocols like TCP want to verify
4712  * destination reachability, while tunnels do not.
4713  * Note that we allow connect to broadcast and multicast
4714  * addresses when ire_requested is set. Thus the ULP
4715  * has to check for IRE_BROADCAST and multicast.
4716  *
4717  * Returns zero if ok.
4718  * On error: returns -1 to mean TBADADDR otherwise returns an errno
4719  * (for use with TSYSERR reply).
4720  *
4721  * Note: lport and fport are in network byte order.
4722  */
4723 int
4724 ip_bind_connected(conn_t *connp, mblk_t *mp, ipaddr_t *src_addrp,
4725     uint16_t lport, ipaddr_t dst_addr, uint16_t fport,
4726     boolean_t ire_requested, boolean_t ipsec_policy_set,
4727     boolean_t fanout_insert, boolean_t verify_dst)
4728 {
4729         ire_t           *src_ire;
4730         ire_t           *dst_ire;
4731         int             error = 0;
4732         int             protocol;
4733         mblk_t          *policy_mp;
4734         ire_t           *sire = NULL;
4735         ire_t           *md_dst_ire = NULL;
4736         ire_t           *lso_dst_ire = NULL;
4737         ill_t           *ill = NULL;
4738         zoneid_t        zoneid;
4739         ipaddr_t        src_addr = *src_addrp;
4740         ip_stack_t      *ipst = connp->conn_netstack->netstack_ip;
4741 
4742         src_ire = dst_ire = NULL;
4743         protocol = *mp->b_wptr & 0xFF;
4744 
4745         /*
4746          * If we never got a disconnect before, clear it now.
4747          */
4748         connp->conn_fully_bound = B_FALSE;
4749 
4750         if (ipsec_policy_set) {
4751                 policy_mp = mp->b_cont;
4752         }
4753 
4754         zoneid = IPCL_ZONEID(connp);
4755 
4756         if (CLASSD(dst_addr)) {
4757                 /* Pick up an IRE_BROADCAST */
4758                 dst_ire = ire_route_lookup(ip_g_all_ones, 0, 0, 0, NULL,
4759                     NULL, zoneid, MBLK_GETLABEL(mp),
4760                     (MATCH_IRE_RECURSIVE |
4761                     MATCH_IRE_DEFAULT | MATCH_IRE_RJ_BHOLE |
4762                     MATCH_IRE_SECATTR), ipst);
4763         } else {
4764                 /*
4765                  * If conn_dontroute is set or if conn_nexthop_set is set,
4766                  * and onlink ipif is not found set ENETUNREACH error.
4767                  */
4768                 if (connp->conn_dontroute || connp->conn_nexthop_set) {
4769                         ipif_t *ipif;
4770 
4771                         ipif = ipif_lookup_onlink_addr(connp->conn_dontroute ?
4772                             dst_addr : connp->conn_nexthop_v4, zoneid, ipst);
4773                         if (ipif == NULL) {
4774                                 error = ENETUNREACH;
4775                                 goto bad_addr;
4776                         }
4777                         ipif_refrele(ipif);
4778                 }
4779 
4780                 if (connp->conn_nexthop_set) {
4781                         dst_ire = ire_route_lookup(connp->conn_nexthop_v4, 0,
4782                             0, 0, NULL, NULL, zoneid, MBLK_GETLABEL(mp),
4783                             MATCH_IRE_SECATTR, ipst);
4784                 } else {
4785                         dst_ire = ire_route_lookup(dst_addr, 0, 0, 0, NULL,
4786                             &sire, zoneid, MBLK_GETLABEL(mp),
4787                             (MATCH_IRE_RECURSIVE | MATCH_IRE_DEFAULT |
4788                             MATCH_IRE_PARENT | MATCH_IRE_RJ_BHOLE |
4789                             MATCH_IRE_SECATTR), ipst);
4790                 }
4791         }
4792         /*
4793          * dst_ire can't be a broadcast when not ire_requested.
4794          * We also prevent ire's with src address INADDR_ANY to
4795          * be used, which are created temporarily for
4796          * sending out packets from endpoints that have
4797          * conn_unspec_src set.  If verify_dst is true, the destination must be
4798          * reachable.  If verify_dst is false, the destination needn't be
4799          * reachable.
4800          *
4801          * If we match on a reject or black hole, then we've got a
4802          * local failure.  May as well fail out the connect() attempt,
4803          * since it's never going to succeed.
4804          */
4805         if (dst_ire == NULL || dst_ire->ire_src_addr == INADDR_ANY ||
4806             (dst_ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) ||
4807             ((dst_ire->ire_type & IRE_BROADCAST) && !ire_requested)) {
4808                 /*
4809                  * If we're verifying destination reachability, we always want
4810                  * to complain here.
4811                  *
4812                  * If we're not verifying destination reachability but the
4813                  * destination has a route, we still want to fail on the
4814                  * temporary address and broadcast address tests.
4815                  */
4816                 if (verify_dst || (dst_ire != NULL)) {
4817                         if (ip_debug > 2) {
4818                                 pr_addr_dbg("ip_bind_connected: bad connected "
4819                                     "dst %s\n", AF_INET, &dst_addr);
4820                         }
4821                         if (dst_ire == NULL || !(dst_ire->ire_type & IRE_HOST))
4822                                 error = ENETUNREACH;
4823                         else
4824                                 error = EHOSTUNREACH;
4825                         goto bad_addr;
4826                 }
4827         }
4828 
4829         /*
4830          * We now know that routing will allow us to reach the destination.
4831          * Check whether Trusted Solaris policy allows communication with this
4832          * host, and pretend that the destination is unreachable if not.
4833          *
4834          * This is never a problem for TCP, since that transport is known to
4835          * compute the label properly as part of the tcp_rput_other T_BIND_ACK
4836          * handling.  If the remote is unreachable, it will be detected at that
4837          * point, so there's no reason to check it here.
4838          *
4839          * Note that for sendto (and other datagram-oriented friends), this
4840          * check is done as part of the data path label computation instead.
4841          * The check here is just to make non-TCP connect() report the right
4842          * error.
4843          */
4844         if (dst_ire != NULL && is_system_labeled() &&
4845             !IPCL_IS_TCP(connp) &&
4846             tsol_compute_label(DB_CREDDEF(mp, connp->conn_cred), dst_addr, NULL,
4847             connp->conn_mac_exempt, ipst) != 0) {
4848                 error = EHOSTUNREACH;
4849                 if (ip_debug > 2) {
4850                         pr_addr_dbg("ip_bind_connected: no label for dst %s\n",
4851                             AF_INET, &dst_addr);
4852                 }
4853                 goto bad_addr;
4854         }
4855 
4856         /*
4857          * If the app does a connect(), it means that it will most likely
4858          * send more than 1 packet to the destination.  It makes sense
4859          * to clear the temporary flag.
4860          */
4861         if (dst_ire != NULL && dst_ire->ire_type == IRE_CACHE &&
4862             (dst_ire->ire_marks & IRE_MARK_TEMPORARY)) {
4863                 irb_t *irb = dst_ire->ire_bucket;
4864 
4865                 rw_enter(&irb->irb_lock, RW_WRITER);
4866                 /*
4867                  * We need to recheck for IRE_MARK_TEMPORARY after acquiring
4868                  * the lock to guarantee irb_tmp_ire_cnt.
4869                  */
4870                 if (dst_ire->ire_marks & IRE_MARK_TEMPORARY) {
4871                         dst_ire->ire_marks &= ~IRE_MARK_TEMPORARY;
4872                         irb->irb_tmp_ire_cnt--;
4873                 }
4874                 rw_exit(&irb->irb_lock);
4875         }
4876 
4877         /*
4878          * See if we should notify ULP about LSO/MDT; we do this whether or not
4879          * ire_requested is TRUE, in order to handle active connects; LSO/MDT
4880          * eligibility tests for passive connects are handled separately
4881          * through tcp_adapt_ire().  We do this before the source address
4882          * selection, because dst_ire may change after a call to
4883          * ipif_select_source().  This is a best-effort check, as the
4884          * packet for this connection may not actually go through
4885          * dst_ire->ire_stq, and the exact IRE can only be known after
4886          * calling ip_newroute().  This is why we further check on the
4887          * IRE during LSO/Multidata packet transmission in
4888          * tcp_lsosend()/tcp_multisend().
4889          */
4890         if (!ipsec_policy_set && dst_ire != NULL &&
4891             !(dst_ire->ire_type & (IRE_LOCAL | IRE_LOOPBACK | IRE_BROADCAST)) &&
4892             (ill = ire_to_ill(dst_ire), ill != NULL)) {
4893                 if (ipst->ips_ip_lso_outbound && ILL_LSO_CAPABLE(ill)) {
4894                         lso_dst_ire = dst_ire;
4895                         IRE_REFHOLD(lso_dst_ire);
4896                 } else if (ipst->ips_ip_multidata_outbound &&
4897                     ILL_MDT_CAPABLE(ill)) {
4898                         md_dst_ire = dst_ire;
4899                         IRE_REFHOLD(md_dst_ire);
4900                 }
4901         }
4902 
4903         if (dst_ire != NULL &&
4904             dst_ire->ire_type == IRE_LOCAL &&
4905             dst_ire->ire_zoneid != zoneid && dst_ire->ire_zoneid != ALL_ZONES) {
4906                 /*
4907                  * If the IRE belongs to a different zone, look for a matching
4908                  * route in the forwarding table and use the source address from
4909                  * that route.
4910                  */
4911                 src_ire = ire_ftable_lookup(dst_addr, 0, 0, 0, NULL, NULL,
4912                     zoneid, 0, NULL,
4913                     MATCH_IRE_RECURSIVE | MATCH_IRE_DEFAULT |
4914                     MATCH_IRE_RJ_BHOLE, ipst);
4915                 if (src_ire == NULL) {
4916                         error = EHOSTUNREACH;
4917                         goto bad_addr;
4918                 } else if (src_ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
4919                         if (!(src_ire->ire_type & IRE_HOST))
4920                                 error = ENETUNREACH;
4921                         else
4922                                 error = EHOSTUNREACH;
4923                         goto bad_addr;
4924                 }
4925                 if (src_addr == INADDR_ANY)
4926                         src_addr = src_ire->ire_src_addr;
4927                 ire_refrele(src_ire);
4928                 src_ire = NULL;
4929         } else if ((src_addr == INADDR_ANY) && (dst_ire != NULL)) {
4930                 if ((sire != NULL) && (sire->ire_flags & RTF_SETSRC)) {
4931                         src_addr = sire->ire_src_addr;
4932                         ire_refrele(dst_ire);
4933                         dst_ire = sire;
4934                         sire = NULL;
4935                 } else {
4936                         /*
4937                          * Pick a source address so that a proper inbound
4938                          * load spreading would happen.
4939                          */
4940                         ill_t *dst_ill = dst_ire->ire_ipif->ipif_ill;
4941                         ipif_t *src_ipif = NULL;
4942                         ire_t *ipif_ire;
4943 
4944                         /*
4945                          * Supply a local source address such that inbound
4946                          * load spreading happens.
4947                          *
4948                          * Determine the best source address on this ill for
4949                          * the destination.
4950                          *
4951                          * 1) For broadcast, we should return a broadcast ire
4952                          *    found above so that upper layers know that the
4953                          *    destination address is a broadcast address.
4954                          *
4955                          * 2) If this is part of a group, select a better
4956                          *    source address so that better inbound load
4957                          *    balancing happens. Do the same if the ipif
4958                          *    is DEPRECATED.
4959                          *
4960                          * 3) If the outgoing interface is part of a usesrc
4961                          *    group, then try selecting a source address from
4962                          *    the usesrc ILL.
4963                          */
4964                         if ((dst_ire->ire_zoneid != zoneid &&
4965                             dst_ire->ire_zoneid != ALL_ZONES) ||
4966                             (!(dst_ire->ire_flags & RTF_SETSRC)) &&
4967                             (!(dst_ire->ire_type & IRE_BROADCAST) &&
4968                             ((dst_ill->ill_group != NULL) ||
4969                             (dst_ire->ire_ipif->ipif_flags & IPIF_DEPRECATED) ||
4970                             (dst_ill->ill_usesrc_ifindex != 0)))) {
4971                                 /*
4972                                  * If the destination is reachable via a
4973                                  * given gateway, the selected source address
4974                                  * should be in the same subnet as the gateway.
4975                                  * Otherwise, the destination is not reachable.
4976                                  *
4977                                  * If there are no interfaces on the same subnet
4978                                  * as the destination, ipif_select_source gives
4979                                  * first non-deprecated interface which might be
4980                                  * on a different subnet than the gateway.
4981                                  * This is not desirable. Hence pass the dst_ire
4982                                  * source address to ipif_select_source.
4983                                  * It is sure that the destination is reachable
4984                                  * with the dst_ire source address subnet.
4985                                  * So passing dst_ire source address to
4986                                  * ipif_select_source will make sure that the
4987                                  * selected source will be on the same subnet
4988                                  * as dst_ire source address.
4989                                  */
4990                                 ipaddr_t saddr =
4991                                     dst_ire->ire_ipif->ipif_src_addr;
4992                                 src_ipif = ipif_select_source(dst_ill,
4993                                     saddr, zoneid);
4994                                 if (src_ipif != NULL) {
4995                                         if (IS_VNI(src_ipif->ipif_ill)) {
4996                                                 /*
4997                                                  * For VNI there is no
4998                                                  * interface route
4999                                                  */
5000                                                 src_addr =
5001                                                     src_ipif->ipif_src_addr;
5002                                         } else {
5003                                                 ipif_ire =
5004                                                     ipif_to_ire(src_ipif);
5005                                                 if (ipif_ire != NULL) {
5006                                                         IRE_REFRELE(dst_ire);
5007                                                         dst_ire = ipif_ire;
5008                                                 }
5009                                                 src_addr =
5010                                                     dst_ire->ire_src_addr;
5011                                         }
5012                                         ipif_refrele(src_ipif);
5013                                 } else {
5014                                         src_addr = dst_ire->ire_src_addr;
5015                                 }
5016                         } else {
5017                                 src_addr = dst_ire->ire_src_addr;
5018                         }
5019                 }
5020         }
5021 
5022         /*
5023          * We do ire_route_lookup() here (and not
5024          * interface lookup as we assert that
5025          * src_addr should only come from an
5026          * UP interface for hard binding.
5027          */
5028         ASSERT(src_ire == NULL);
5029         src_ire = ire_route_lookup(src_addr, 0, 0, 0, NULL,
5030             NULL, zoneid, NULL, MATCH_IRE_ZONEONLY, ipst);
5031         /* src_ire must be a local|loopback */
5032         if (!IRE_IS_LOCAL(src_ire)) {
5033                 if (ip_debug > 2) {
5034                         pr_addr_dbg("ip_bind_connected: bad connected "
5035                             "src %s\n", AF_INET, &src_addr);
5036                 }
5037                 error = EADDRNOTAVAIL;
5038                 goto bad_addr;
5039         }
5040 
5041         /*
5042          * If the source address is a loopback address, the
5043          * destination had best be local or multicast.
5044          * The transports that can't handle multicast will reject
5045          * those addresses.
5046          */
5047         if (src_ire->ire_type == IRE_LOOPBACK &&
5048             !(IRE_IS_LOCAL(dst_ire) || CLASSD(dst_addr))) {
5049                 ip1dbg(("ip_bind_connected: bad connected loopback\n"));
5050                 error = -1;
5051                 goto bad_addr;
5052         }
5053 
5054         /*
5055          * Allow setting new policies. For example, disconnects come
5056          * down as ipa_t bind. As we would have set conn_policy_cached
5057          * to B_TRUE before, we should set it to B_FALSE, so that policy
5058          * can change after the disconnect.
5059          */
5060         connp->conn_policy_cached = B_FALSE;
5061 
5062         /*
5063          * Set the conn addresses/ports immediately, so the IPsec policy calls
5064          * can handle their passed-in conn's.
5065          */
5066 
5067         IN6_IPADDR_TO_V4MAPPED(src_addr, &connp->conn_srcv6);
5068         IN6_IPADDR_TO_V4MAPPED(dst_addr, &connp->conn_remv6);
5069         connp->conn_lport = lport;
5070         connp->conn_fport = fport;
5071         *src_addrp = src_addr;
5072 
5073         ASSERT(!(ipsec_policy_set && ire_requested));
5074         if (ire_requested) {
5075                 iulp_t *ulp_info = NULL;
5076 
5077                 /*
5078                  * Note that sire will not be NULL if this is an off-link
5079                  * connection and there is not cache for that dest yet.
5080                  *
5081                  * XXX Because of an existing bug, if there are multiple
5082                  * default routes, the IRE returned now may not be the actual
5083                  * default route used (default routes are chosen in a
5084                  * round robin fashion).  So if the metrics for different
5085                  * default routes are different, we may return the wrong
5086                  * metrics.  This will not be a problem if the existing
5087                  * bug is fixed.
5088                  */
5089                 if (sire != NULL) {
5090                         ulp_info = &(sire->ire_uinfo);
5091                 }
5092                 if (!ip_bind_insert_ire(mp, dst_ire, ulp_info, ipst)) {
5093                         error = -1;
5094                         goto bad_addr;
5095                 }
5096         } else if (ipsec_policy_set) {
5097                 if (!ip_bind_ipsec_policy_set(connp, policy_mp)) {
5098                         error = -1;
5099                         goto bad_addr;
5100                 }
5101         }
5102 
5103         /*
5104          * Cache IPsec policy in this conn.  If we have per-socket policy,
5105          * we'll cache that.  If we don't, we'll inherit global policy.
5106          *
5107          * We can't insert until the conn reflects the policy. Note that
5108          * conn_policy_cached is set by ipsec_conn_cache_policy() even for
5109          * connections where we don't have a policy. This is to prevent
5110          * global policy lookups in the inbound path.
5111          *
5112          * If we insert before we set conn_policy_cached,
5113          * CONN_INBOUND_POLICY_PRESENT() check can still evaluate true
5114          * because global policy cound be non-empty. We normally call
5115          * ipsec_check_policy() for conn_policy_cached connections only if
5116          * ipc_in_enforce_policy is set. But in this case,
5117          * conn_policy_cached can get set anytime since we made the
5118          * CONN_INBOUND_POLICY_PRESENT() check and ipsec_check_policy() is
5119          * called, which will make the above assumption false.  Thus, we
5120          * need to insert after we set conn_policy_cached.
5121          */
5122         if ((error = ipsec_conn_cache_policy(connp, B_TRUE)) != 0)
5123                 goto bad_addr;
5124 
5125         if (fanout_insert) {
5126                 /*
5127                  * The addresses have been verified. Time to insert in
5128                  * the correct fanout list.
5129                  */
5130                 error = ipcl_conn_insert(connp, protocol, src_addr,
5131                     dst_addr, connp->conn_ports);
5132         }
5133 
5134         if (error == 0) {
5135                 connp->conn_fully_bound = B_TRUE;
5136                 /*
5137                  * Our initial checks for LSO/MDT have passed; the IRE is not
5138                  * LOCAL/LOOPBACK/BROADCAST, and the link layer seems to
5139                  * be supporting LSO/MDT.  Pass the IRE, IPC and ILL into
5140                  * ip_xxinfo_return(), which performs further checks
5141                  * against them and upon success, returns the LSO/MDT info
5142                  * mblk which we will attach to the bind acknowledgment.
5143                  */
5144                 if (lso_dst_ire != NULL) {
5145                         mblk_t *lsoinfo_mp;
5146 
5147                         ASSERT(ill->ill_lso_capab != NULL);
5148                         if ((lsoinfo_mp = ip_lsoinfo_return(lso_dst_ire, connp,
5149                             ill->ill_name, ill->ill_lso_capab)) != NULL)
5150                                 linkb(mp, lsoinfo_mp);
5151                 } else if (md_dst_ire != NULL) {
5152                         mblk_t *mdinfo_mp;
5153 
5154                         ASSERT(ill->ill_mdt_capab != NULL);
5155                         if ((mdinfo_mp = ip_mdinfo_return(md_dst_ire, connp,
5156                             ill->ill_name, ill->ill_mdt_capab)) != NULL)
5157                                 linkb(mp, mdinfo_mp);
5158                 }
5159         }
5160 bad_addr:
5161         if (ipsec_policy_set) {
5162                 ASSERT(policy_mp == mp->b_cont);
5163                 ASSERT(policy_mp != NULL);
5164                 freeb(policy_mp);
5165                 /*
5166                  * As of now assume that nothing else accompanies
5167                  * IPSEC_POLICY_SET.
5168                  */
5169                 mp->b_cont = NULL;
5170         }
5171         if (src_ire != NULL)
5172                 IRE_REFRELE(src_ire);
5173         if (dst_ire != NULL)
5174                 IRE_REFRELE(dst_ire);
5175         if (sire != NULL)
5176                 IRE_REFRELE(sire);
5177         if (md_dst_ire != NULL)
5178                 IRE_REFRELE(md_dst_ire);
5179         if (lso_dst_ire != NULL)
5180                 IRE_REFRELE(lso_dst_ire);
5181         return (error);
5182 }
5183 
5184 /*
5185  * Insert the ire in b_cont. Returns false if it fails (due to lack of space).
5186  * Prefers dst_ire over src_ire.
5187  */
5188 static boolean_t
5189 ip_bind_insert_ire(mblk_t *mp, ire_t *ire, iulp_t *ulp_info, ip_stack_t *ipst)
5190 {
5191         mblk_t  *mp1;
5192         ire_t *ret_ire = NULL;
5193 
5194         mp1 = mp->b_cont;
5195         ASSERT(mp1 != NULL);
5196 
5197         if (ire != NULL) {
5198                 /*
5199                  * mp1 initialized above to IRE_DB_REQ_TYPE
5200                  * appended mblk. Its <upper protocol>'s
5201                  * job to make sure there is room.
5202                  */
5203                 if ((mp1->b_datap->db_lim - mp1->b_rptr) < sizeof (ire_t))
5204                         return (0);
5205 
5206                 mp1->b_datap->db_type = IRE_DB_TYPE;
5207                 mp1->b_wptr = mp1->b_rptr + sizeof (ire_t);
5208                 bcopy(ire, mp1->b_rptr, sizeof (ire_t));
5209                 ret_ire = (ire_t *)mp1->b_rptr;
5210                 /*
5211                  * Pass the latest setting of the ip_path_mtu_discovery and
5212                  * copy the ulp info if any.
5213                  */
5214                 ret_ire->ire_frag_flag |= (ipst->ips_ip_path_mtu_discovery) ?
5215                     IPH_DF : 0;
5216                 if (ulp_info != NULL) {
5217                         bcopy(ulp_info, &(ret_ire->ire_uinfo),
5218                             sizeof (iulp_t));
5219                 }
5220                 ret_ire->ire_mp = mp1;
5221         } else {
5222                 /*
5223                  * No IRE was found. Remove IRE mblk.
5224                  */
5225                 mp->b_cont = mp1->b_cont;
5226                 freeb(mp1);
5227         }
5228 
5229         return (1);
5230 }
5231 
5232 /*
5233  * Carve "len" bytes out of an mblk chain, consuming any we empty, and duping
5234  * the final piece where we don't.  Return a pointer to the first mblk in the
5235  * result, and update the pointer to the next mblk to chew on.  If anything
5236  * goes wrong (i.e., dupb fails), we waste everything in sight and return a
5237  * NULL pointer.
5238  */
5239 mblk_t *
5240 ip_carve_mp(mblk_t **mpp, ssize_t len)
5241 {
5242         mblk_t  *mp0;
5243         mblk_t  *mp1;
5244         mblk_t  *mp2;
5245 
5246         if (!len || !mpp || !(mp0 = *mpp))
5247                 return (NULL);
5248         /* If we aren't going to consume the first mblk, we need a dup. */
5249         if (mp0->b_wptr - mp0->b_rptr > len) {
5250                 mp1 = dupb(mp0);
5251                 if (mp1) {
5252                         /* Partition the data between the two mblks. */
5253                         mp1->b_wptr = mp1->b_rptr + len;
5254                         mp0->b_rptr = mp1->b_wptr;
5255                         /*
5256                          * after adjustments if mblk not consumed is now
5257                          * unaligned, try to align it. If this fails free
5258                          * all messages and let upper layer recover.
5259                          */
5260                         if (!OK_32PTR(mp0->b_rptr)) {
5261                                 if (!pullupmsg(mp0, -1)) {
5262                                         freemsg(mp0);
5263                                         freemsg(mp1);
5264                                         *mpp = NULL;
5265                                         return (NULL);
5266                                 }
5267                         }
5268                 }
5269                 return (mp1);
5270         }
5271         /* Eat through as many mblks as we need to get len bytes. */
5272         len -= mp0->b_wptr - mp0->b_rptr;
5273         for (mp2 = mp1 = mp0; (mp2 = mp2->b_cont) != 0 && len; mp1 = mp2) {
5274                 if (mp2->b_wptr - mp2->b_rptr > len) {
5275                         /*
5276                          * We won't consume the entire last mblk.  Like
5277                          * above, dup and partition it.
5278                          */
5279                         mp1->b_cont = dupb(mp2);
5280                         mp1 = mp1->b_cont;
5281                         if (!mp1) {
5282                                 /*
5283                                  * Trouble.  Rather than go to a lot of
5284                                  * trouble to clean up, we free the messages.
5285                                  * This won't be any worse than losing it on
5286                                  * the wire.
5287                                  */
5288                                 freemsg(mp0);
5289                                 freemsg(mp2);
5290                                 *mpp = NULL;
5291                                 return (NULL);
5292                         }
5293                         mp1->b_wptr = mp1->b_rptr + len;
5294                         mp2->b_rptr = mp1->b_wptr;
5295                         /*
5296                          * after adjustments if mblk not consumed is now
5297                          * unaligned, try to align it. If this fails free
5298                          * all messages and let upper layer recover.
5299                          */
5300                         if (!OK_32PTR(mp2->b_rptr)) {
5301                                 if (!pullupmsg(mp2, -1)) {
5302                                         freemsg(mp0);
5303                                         freemsg(mp2);
5304                                         *mpp = NULL;
5305                                         return (NULL);
5306                                 }
5307                         }
5308                         *mpp = mp2;
5309                         return (mp0);
5310                 }
5311                 /* Decrement len by the amount we just got. */
5312                 len -= mp2->b_wptr - mp2->b_rptr;
5313         }
5314         /*
5315          * len should be reduced to zero now.  If not our caller has
5316          * screwed up.
5317          */
5318         if (len) {
5319                 /* Shouldn't happen! */
5320                 freemsg(mp0);
5321                 *mpp = NULL;
5322                 return (NULL);
5323         }
5324         /*
5325          * We consumed up to exactly the end of an mblk.  Detach the part
5326          * we are returning from the rest of the chain.
5327          */
5328         mp1->b_cont = NULL;
5329         *mpp = mp2;
5330         return (mp0);
5331 }
5332 
5333 /* The ill stream is being unplumbed. Called from ip_close */
5334 int
5335 ip_modclose(ill_t *ill)
5336 {
5337         boolean_t success;
5338         ipsq_t  *ipsq;
5339         ipif_t  *ipif;
5340         queue_t *q = ill->ill_rq;
5341         ip_stack_t      *ipst = ill->ill_ipst;
5342         clock_t timeout;
5343 
5344         /*
5345          * Wait for the ACKs of all deferred control messages to be processed.
5346          * In particular, we wait for a potential capability reset initiated
5347          * in ip_sioctl_plink() to complete before proceeding.
5348          *
5349          * Note: we wait for at most ip_modclose_ackwait_ms (by default 3000 ms)
5350          * in case the driver never replies.
5351          */
5352         timeout = lbolt + MSEC_TO_TICK(ip_modclose_ackwait_ms);
5353         mutex_enter(&ill->ill_lock);
5354         while (ill->ill_dlpi_pending != DL_PRIM_INVAL) {
5355                 if (cv_timedwait(&ill->ill_cv, &ill->ill_lock, timeout) < 0) {
5356                         /* Timeout */
5357                         break;
5358                 }
5359         }
5360         mutex_exit(&ill->ill_lock);
5361 
5362         /*
5363          * Forcibly enter the ipsq after some delay. This is to take
5364          * care of the case when some ioctl does not complete because
5365          * we sent a control message to the driver and it did not
5366          * send us a reply. We want to be able to at least unplumb
5367          * and replumb rather than force the user to reboot the system.
5368          */
5369         success = ipsq_enter(ill, B_FALSE);
5370 
5371         /*
5372          * Open/close/push/pop is guaranteed to be single threaded
5373          * per stream by STREAMS. FS guarantees that all references
5374          * from top are gone before close is called. So there can't
5375          * be another close thread that has set CONDEMNED on this ill.
5376          * and cause ipsq_enter to return failure.
5377          */
5378         ASSERT(success);
5379         ipsq = ill->ill_phyint->phyint_ipsq;
5380 
5381         /*
5382          * Mark it condemned. No new reference will be made to this ill.
5383          * Lookup functions will return an error. Threads that try to
5384          * increment the refcnt must check for ILL_CAN_LOOKUP. This ensures
5385          * that the refcnt will drop down to zero.
5386          */
5387         mutex_enter(&ill->ill_lock);
5388         ill->ill_state_flags |= ILL_CONDEMNED;
5389         for (ipif = ill->ill_ipif; ipif != NULL;
5390             ipif = ipif->ipif_next) {
5391                 ipif->ipif_state_flags |= IPIF_CONDEMNED;
5392         }
5393         /*
5394          * Wake up anybody waiting to enter the ipsq. ipsq_enter
5395          * returns  error if ILL_CONDEMNED is set
5396          */
5397         cv_broadcast(&ill->ill_cv);
5398         mutex_exit(&ill->ill_lock);
5399 
5400         /*
5401          * Send all the deferred DLPI messages downstream which came in
5402          * during the small window right before ipsq_enter(). We do this
5403          * without waiting for the ACKs because all the ACKs for M_PROTO
5404          * messages are ignored in ip_rput() when ILL_CONDEMNED is set.
5405          */
5406         ill_dlpi_send_deferred(ill);
5407 
5408         /*
5409          * Shut down fragmentation reassembly.
5410          * ill_frag_timer won't start a timer again.
5411          * Now cancel any existing timer
5412          */
5413         (void) untimeout(ill->ill_frag_timer_id);
5414         (void) ill_frag_timeout(ill, 0);
5415 
5416         /*
5417          * If MOVE was in progress, clear the
5418          * move_in_progress fields also.
5419          */
5420         if (ill->ill_move_in_progress) {
5421                 ILL_CLEAR_MOVE(ill);
5422         }
5423 
5424         /*
5425          * Call ill_delete to bring down the ipifs, ilms and ill on
5426          * this ill. Then wait for the refcnts to drop to zero.
5427          * ill_is_freeable checks whether the ill is really quiescent.
5428          * Then make sure that threads that are waiting to enter the
5429          * ipsq have seen the error returned by ipsq_enter and have
5430          * gone away. Then we call ill_delete_tail which does the
5431          * DL_UNBIND_REQ with the driver and then qprocsoff.
5432          */
5433         ill_delete(ill);
5434         mutex_enter(&ill->ill_lock);
5435         while (!ill_is_freeable(ill))
5436                 cv_wait(&ill->ill_cv, &ill->ill_lock);
5437         while (ill->ill_waiters)
5438                 cv_wait(&ill->ill_cv, &ill->ill_lock);
5439 
5440         mutex_exit(&ill->ill_lock);
5441 
5442         /*
5443          * ill_delete_tail drops reference on ill_ipst, but we need to keep
5444          * it held until the end of the function since the cleanup
5445          * below needs to be able to use the ip_stack_t.
5446          */
5447         netstack_hold(ipst->ips_netstack);
5448 
5449         /* qprocsoff is called in ill_delete_tail */
5450         ill_delete_tail(ill);
5451         ASSERT(ill->ill_ipst == NULL);
5452 
5453         /*
5454          * Walk through all upper (conn) streams and qenable
5455          * those that have queued data.
5456          * close synchronization needs this to
5457          * be done to ensure that all upper layers blocked
5458          * due to flow control to the closing device
5459          * get unblocked.
5460          */
5461         ip1dbg(("ip_wsrv: walking\n"));
5462         conn_walk_drain(ipst);
5463 
5464         mutex_enter(&ipst->ips_ip_mi_lock);
5465         mi_close_unlink(&ipst->ips_ip_g_head, (IDP)ill);
5466         mutex_exit(&ipst->ips_ip_mi_lock);
5467 
5468         /*
5469          * credp could be null if the open didn't succeed and ip_modopen
5470          * itself calls ip_close.
5471          */
5472         if (ill->ill_credp != NULL)
5473                 crfree(ill->ill_credp);
5474 
5475         mutex_enter(&ill->ill_lock);
5476         ill_nic_info_dispatch(ill);
5477         mutex_exit(&ill->ill_lock);
5478 
5479         /*
5480          * Now we are done with the module close pieces that
5481          * need the netstack_t.
5482          */
5483         netstack_rele(ipst->ips_netstack);
5484 
5485         mi_close_free((IDP)ill);
5486         q->q_ptr = WR(q)->q_ptr = NULL;
5487 
5488         ipsq_exit(ipsq);
5489 
5490         return (0);
5491 }
5492 
5493 /*
5494  * This is called as part of close() for IP, UDP, ICMP, and RTS
5495  * in order to quiesce the conn.
5496  */
5497 void
5498 ip_quiesce_conn(conn_t *connp)
5499 {
5500         boolean_t       drain_cleanup_reqd = B_FALSE;
5501         boolean_t       conn_ioctl_cleanup_reqd = B_FALSE;
5502         boolean_t       ilg_cleanup_reqd = B_FALSE;
5503         ip_stack_t      *ipst;
5504 
5505         ASSERT(!IPCL_IS_TCP(connp));
5506         ipst = connp->conn_netstack->netstack_ip;
5507 
5508         /*
5509          * Mark the conn as closing, and this conn must not be
5510          * inserted in future into any list. Eg. conn_drain_insert(),
5511          * won't insert this conn into the conn_drain_list.
5512          * Similarly ill_pending_mp_add() will not add any mp to
5513          * the pending mp list, after this conn has started closing.
5514          *
5515          * conn_idl, conn_pending_ill, conn_down_pending_ill, conn_ilg
5516          * cannot get set henceforth.
5517          */
5518         mutex_enter(&connp->conn_lock);
5519         ASSERT(!(connp->conn_state_flags & CONN_QUIESCED));
5520         connp->conn_state_flags |= CONN_CLOSING;
5521         if (connp->conn_idl != NULL)
5522                 drain_cleanup_reqd = B_TRUE;
5523         if (connp->conn_oper_pending_ill != NULL)
5524                 conn_ioctl_cleanup_reqd = B_TRUE;
5525         if (connp->conn_dhcpinit_ill != NULL) {
5526                 ASSERT(connp->conn_dhcpinit_ill->ill_dhcpinit != 0);
5527                 atomic_dec_32(&connp->conn_dhcpinit_ill->ill_dhcpinit);
5528                 connp->conn_dhcpinit_ill = NULL;
5529         }
5530         if (connp->conn_ilg_inuse != 0)
5531                 ilg_cleanup_reqd = B_TRUE;
5532         mutex_exit(&connp->conn_lock);
5533 
5534         if (conn_ioctl_cleanup_reqd)
5535                 conn_ioctl_cleanup(connp);
5536 
5537         if (is_system_labeled() && connp->conn_anon_port) {
5538                 (void) tsol_mlp_anon(crgetzone(connp->conn_cred),
5539                     connp->conn_mlp_type, connp->conn_ulp,
5540                     ntohs(connp->conn_lport), B_FALSE);
5541                 connp->conn_anon_port = 0;
5542         }
5543         connp->conn_mlp_type = mlptSingle;
5544 
5545         /*
5546          * Remove this conn from any fanout list it is on.
5547          * and then wait for any threads currently operating
5548          * on this endpoint to finish
5549          */
5550         ipcl_hash_remove(connp);
5551 
5552         /*
5553          * Remove this conn from the drain list, and do
5554          * any other cleanup that may be required.
5555          * (Only non-tcp streams may have a non-null conn_idl.
5556          * TCP streams are never flow controlled, and
5557          * conn_idl will be null)
5558          */
5559         if (drain_cleanup_reqd)
5560                 conn_drain_tail(connp, B_TRUE);
5561 
5562         if (connp == ipst->ips_ip_g_mrouter)
5563                 (void) ip_mrouter_done(NULL, ipst);
5564 
5565         if (ilg_cleanup_reqd)
5566                 ilg_delete_all(connp);
5567 
5568         conn_delete_ire(connp, NULL);
5569 
5570         /*
5571          * Now conn refcnt can increase only thru CONN_INC_REF_LOCKED.
5572          * callers from write side can't be there now because close
5573          * is in progress. The only other caller is ipcl_walk
5574          * which checks for the condemned flag.
5575          */
5576         mutex_enter(&connp->conn_lock);
5577         connp->conn_state_flags |= CONN_CONDEMNED;
5578         while (connp->conn_ref != 1)
5579                 cv_wait(&connp->conn_cv, &connp->conn_lock);
5580         connp->conn_state_flags |= CONN_QUIESCED;
5581         mutex_exit(&connp->conn_lock);
5582 }
5583 
5584 /* ARGSUSED */
5585 int
5586 ip_close(queue_t *q, int flags)
5587 {
5588         conn_t          *connp;
5589 
5590         TRACE_1(TR_FAC_IP, TR_IP_CLOSE, "ip_close: q %p", q);
5591 
5592         /*
5593          * Call the appropriate delete routine depending on whether this is
5594          * a module or device.
5595          */
5596         if (WR(q)->q_next != NULL) {
5597                 /* This is a module close */
5598                 return (ip_modclose((ill_t *)q->q_ptr));
5599         }
5600 
5601         connp = q->q_ptr;
5602         ip_quiesce_conn(connp);
5603 
5604         qprocsoff(q);
5605 
5606         /*
5607          * Now we are truly single threaded on this stream, and can
5608          * delete the things hanging off the connp, and finally the connp.
5609          * We removed this connp from the fanout list, it cannot be
5610          * accessed thru the fanouts, and we already waited for the
5611          * conn_ref to drop to 0. We are already in close, so
5612          * there cannot be any other thread from the top. qprocsoff
5613          * has completed, and service has completed or won't run in
5614          * future.
5615          */
5616         ASSERT(connp->conn_ref == 1);
5617 
5618         inet_minor_free(connp->conn_minor_arena, connp->conn_dev);
5619 
5620         connp->conn_ref--;
5621         ipcl_conn_destroy(connp);
5622 
5623         q->q_ptr = WR(q)->q_ptr = NULL;
5624         return (0);
5625 }
5626 
5627 /*
5628  * Wapper around putnext() so that ip_rts_request can merely use
5629  * conn_recv.
5630  */
5631 /*ARGSUSED2*/
5632 static void
5633 ip_conn_input(void *arg1, mblk_t *mp, void *arg2)
5634 {
5635         conn_t *connp = (conn_t *)arg1;
5636 
5637         putnext(connp->conn_rq, mp);
5638 }
5639 
5640 /* Return the IP checksum for the IP header at "iph". */
5641 uint16_t
5642 ip_csum_hdr(ipha_t *ipha)
5643 {
5644         uint16_t        *uph;
5645         uint32_t        sum;
5646         int             opt_len;
5647 
5648         opt_len = (ipha->ipha_version_and_hdr_length & 0xF) -
5649             IP_SIMPLE_HDR_LENGTH_IN_WORDS;
5650         uph = (uint16_t *)ipha;
5651         sum = uph[0] + uph[1] + uph[2] + uph[3] + uph[4] +
5652             uph[5] + uph[6] + uph[7] + uph[8] + uph[9];
5653         if (opt_len > 0) {
5654                 do {
5655                         sum += uph[10];
5656                         sum += uph[11];
5657                         uph += 2;
5658                 } while (--opt_len);
5659         }
5660         sum = (sum & 0xFFFF) + (sum >> 16);
5661         sum = ~(sum + (sum >> 16)) & 0xFFFF;
5662         if (sum == 0xffff)
5663                 sum = 0;
5664         return ((uint16_t)sum);
5665 }
5666 
5667 /*
5668  * Called when the module is about to be unloaded
5669  */
5670 void
5671 ip_ddi_destroy(void)
5672 {
5673         tnet_fini();
5674 
5675         icmp_ddi_destroy();
5676         rts_ddi_destroy();
5677         udp_ddi_destroy();
5678         sctp_ddi_g_destroy();
5679         tcp_ddi_g_destroy();
5680         ipsec_policy_g_destroy();
5681         ipcl_g_destroy();
5682         ip_net_g_destroy();
5683         ip_ire_g_fini();
5684         inet_minor_destroy(ip_minor_arena_sa);
5685 #if defined(_LP64)
5686         inet_minor_destroy(ip_minor_arena_la);
5687 #endif
5688 
5689 #ifdef DEBUG
5690         list_destroy(&ip_thread_list);
5691         rw_destroy(&ip_thread_rwlock);
5692         tsd_destroy(&ip_thread_data);
5693 #endif
5694 
5695         netstack_unregister(NS_IP);
5696 }
5697 
5698 /*
5699  * First step in cleanup.
5700  */
5701 /* ARGSUSED */
5702 static void
5703 ip_stack_shutdown(netstackid_t stackid, void *arg)
5704 {
5705         ip_stack_t *ipst = (ip_stack_t *)arg;
5706 
5707 #ifdef NS_DEBUG
5708         printf("ip_stack_shutdown(%p, stack %d)\n", (void *)ipst, stackid);
5709 #endif
5710 
5711         /* Get rid of loopback interfaces and their IREs */
5712         ip_loopback_cleanup(ipst);
5713 }
5714 
5715 /*
5716  * Free the IP stack instance.
5717  */
5718 static void
5719 ip_stack_fini(netstackid_t stackid, void *arg)
5720 {
5721         ip_stack_t *ipst = (ip_stack_t *)arg;
5722         int ret;
5723 
5724 #ifdef NS_DEBUG
5725         printf("ip_stack_fini(%p, stack %d)\n", (void *)ipst, stackid);
5726 #endif
5727         ipv4_hook_destroy(ipst);
5728         ipv6_hook_destroy(ipst);
5729         ip_net_destroy(ipst);
5730 
5731         rw_destroy(&ipst->ips_srcid_lock);
5732 
5733         ip_kstat_fini(stackid, ipst->ips_ip_mibkp);
5734         ipst->ips_ip_mibkp = NULL;
5735         icmp_kstat_fini(stackid, ipst->ips_icmp_mibkp);
5736         ipst->ips_icmp_mibkp = NULL;
5737         ip_kstat2_fini(stackid, ipst->ips_ip_kstat);
5738         ipst->ips_ip_kstat = NULL;
5739         bzero(&ipst->ips_ip_statistics, sizeof (ipst->ips_ip_statistics));
5740         ip6_kstat_fini(stackid, ipst->ips_ip6_kstat);
5741         ipst->ips_ip6_kstat = NULL;
5742         bzero(&ipst->ips_ip6_statistics, sizeof (ipst->ips_ip6_statistics));
5743 
5744         nd_free(&ipst->ips_ip_g_nd);
5745         kmem_free(ipst->ips_param_arr, sizeof (lcl_param_arr));
5746         ipst->ips_param_arr = NULL;
5747         kmem_free(ipst->ips_ndp_arr, sizeof (lcl_ndp_arr));
5748         ipst->ips_ndp_arr = NULL;
5749 
5750         ip_mrouter_stack_destroy(ipst);
5751 
5752         mutex_destroy(&ipst->ips_ip_mi_lock);
5753         rw_destroy(&ipst->ips_ipsec_capab_ills_lock);
5754         rw_destroy(&ipst->ips_ill_g_usesrc_lock);
5755         rw_destroy(&ipst->ips_ip_g_nd_lock);
5756 
5757         ret = untimeout(ipst->ips_igmp_timeout_id);
5758         if (ret == -1) {
5759                 ASSERT(ipst->ips_igmp_timeout_id == 0);
5760         } else {
5761                 ASSERT(ipst->ips_igmp_timeout_id != 0);
5762                 ipst->ips_igmp_timeout_id = 0;
5763         }
5764         ret = untimeout(ipst->ips_igmp_slowtimeout_id);
5765         if (ret == -1) {
5766                 ASSERT(ipst->ips_igmp_slowtimeout_id == 0);
5767         } else {
5768                 ASSERT(ipst->ips_igmp_slowtimeout_id != 0);
5769                 ipst->ips_igmp_slowtimeout_id = 0;
5770         }
5771         ret = untimeout(ipst->ips_mld_timeout_id);
5772         if (ret == -1) {
5773                 ASSERT(ipst->ips_mld_timeout_id == 0);
5774         } else {
5775                 ASSERT(ipst->ips_mld_timeout_id != 0);
5776                 ipst->ips_mld_timeout_id = 0;
5777         }
5778         ret = untimeout(ipst->ips_mld_slowtimeout_id);
5779         if (ret == -1) {
5780                 ASSERT(ipst->ips_mld_slowtimeout_id == 0);
5781         } else {
5782                 ASSERT(ipst->ips_mld_slowtimeout_id != 0);
5783                 ipst->ips_mld_slowtimeout_id = 0;
5784         }
5785         ret = untimeout(ipst->ips_ip_ire_expire_id);
5786         if (ret == -1) {
5787                 ASSERT(ipst->ips_ip_ire_expire_id == 0);
5788         } else {
5789                 ASSERT(ipst->ips_ip_ire_expire_id != 0);
5790                 ipst->ips_ip_ire_expire_id = 0;
5791         }
5792 
5793         mutex_destroy(&ipst->ips_igmp_timer_lock);
5794         mutex_destroy(&ipst->ips_mld_timer_lock);
5795         mutex_destroy(&ipst->ips_igmp_slowtimeout_lock);
5796         mutex_destroy(&ipst->ips_mld_slowtimeout_lock);
5797         mutex_destroy(&ipst->ips_ip_addr_avail_lock);
5798         rw_destroy(&ipst->ips_ill_g_lock);
5799 
5800         ip_ire_fini(ipst);
5801         ip6_asp_free(ipst);
5802         conn_drain_fini(ipst);
5803         ipcl_destroy(ipst);
5804 
5805         mutex_destroy(&ipst->ips_ndp4->ndp_g_lock);
5806         mutex_destroy(&ipst->ips_ndp6->ndp_g_lock);
5807         kmem_free(ipst->ips_ndp4, sizeof (ndp_g_t));
5808         ipst->ips_ndp4 = NULL;
5809         kmem_free(ipst->ips_ndp6, sizeof (ndp_g_t));
5810         ipst->ips_ndp6 = NULL;
5811 
5812         if (ipst->ips_loopback_ksp != NULL) {
5813                 kstat_delete_netstack(ipst->ips_loopback_ksp, stackid);
5814                 ipst->ips_loopback_ksp = NULL;
5815         }
5816 
5817         kmem_free(ipst->ips_phyint_g_list, sizeof (phyint_list_t));
5818         ipst->ips_phyint_g_list = NULL;
5819         kmem_free(ipst->ips_ill_g_heads, sizeof (ill_g_head_t) * MAX_G_HEADS);
5820         ipst->ips_ill_g_heads = NULL;
5821 
5822         kmem_free(ipst, sizeof (*ipst));
5823 }
5824 
5825 /*
5826  * This function is called from the TSD destructor, and is used to debug
5827  * reference count issues in IP. See block comment in <inet/ip_if.h> for
5828  * details.
5829  */
5830 static void
5831 ip_thread_exit(void *phash)
5832 {
5833         th_hash_t *thh = phash;
5834 
5835         rw_enter(&ip_thread_rwlock, RW_WRITER);
5836         list_remove(&ip_thread_list, thh);
5837         rw_exit(&ip_thread_rwlock);
5838         mod_hash_destroy_hash(thh->thh_hash);
5839         kmem_free(thh, sizeof (*thh));
5840 }
5841 
5842 /*
5843  * Called when the IP kernel module is loaded into the kernel
5844  */
5845 void
5846 ip_ddi_init(void)
5847 {
5848         ip_input_proc = ip_squeue_switch(ip_squeue_enter);
5849 
5850         /*
5851          * For IP and TCP the minor numbers should start from 2 since we have 4
5852          * initial devices: ip, ip6, tcp, tcp6.
5853          */
5854         /*
5855          * If this is a 64-bit kernel, then create two separate arenas -
5856          * one for TLIs in the range of INET_MIN_DEV+2 through 2^^18-1, and the
5857          * other for socket apps in the range 2^^18 through 2^^32-1.
5858          */
5859         ip_minor_arena_la = NULL;
5860         ip_minor_arena_sa = NULL;
5861 #if defined(_LP64)
5862         if ((ip_minor_arena_sa = inet_minor_create("ip_minor_arena_sa",
5863             INET_MIN_DEV + 2, MAXMIN32, KM_SLEEP)) == NULL) {
5864                 cmn_err(CE_PANIC,
5865                     "ip_ddi_init: ip_minor_arena_sa creation failed\n");
5866         }
5867         if ((ip_minor_arena_la = inet_minor_create("ip_minor_arena_la",
5868             MAXMIN32 + 1, MAXMIN64, KM_SLEEP)) == NULL) {
5869                 cmn_err(CE_PANIC,
5870                     "ip_ddi_init: ip_minor_arena_la creation failed\n");
5871         }
5872 #else
5873         if ((ip_minor_arena_sa = inet_minor_create("ip_minor_arena_sa",
5874             INET_MIN_DEV + 2, MAXMIN, KM_SLEEP)) == NULL) {
5875                 cmn_err(CE_PANIC,
5876                     "ip_ddi_init: ip_minor_arena_sa creation failed\n");
5877         }
5878 #endif
5879         ip_poll_normal_ticks = MSEC_TO_TICK_ROUNDUP(ip_poll_normal_ms);
5880 
5881         ipcl_g_init();
5882         ip_ire_g_init();
5883         ip_net_g_init();
5884 
5885 #ifdef DEBUG
5886         tsd_create(&ip_thread_data, ip_thread_exit);
5887         rw_init(&ip_thread_rwlock, NULL, RW_DEFAULT, NULL);
5888         list_create(&ip_thread_list, sizeof (th_hash_t),
5889             offsetof(th_hash_t, thh_link));
5890 #endif
5891 
5892         /*
5893          * We want to be informed each time a stack is created or
5894          * destroyed in the kernel, so we can maintain the
5895          * set of udp_stack_t's.
5896          */
5897         netstack_register(NS_IP, ip_stack_init, ip_stack_shutdown,
5898             ip_stack_fini);
5899 
5900         ipsec_policy_g_init();
5901         tcp_ddi_g_init();
5902         sctp_ddi_g_init();
5903 
5904         tnet_init();
5905 
5906         udp_ddi_init();
5907         rts_ddi_init();
5908         icmp_ddi_init();
5909 }
5910 
5911 /*
5912  * Initialize the IP stack instance.
5913  */
5914 static void *
5915 ip_stack_init(netstackid_t stackid, netstack_t *ns)
5916 {
5917         ip_stack_t      *ipst;
5918         ipparam_t       *pa;
5919         ipndp_t         *na;
5920 
5921 #ifdef NS_DEBUG
5922         printf("ip_stack_init(stack %d)\n", stackid);
5923 #endif
5924 
5925         ipst = (ip_stack_t *)kmem_zalloc(sizeof (*ipst), KM_SLEEP);
5926         ipst->ips_netstack = ns;
5927 
5928         ipst->ips_ill_g_heads = kmem_zalloc(sizeof (ill_g_head_t) * MAX_G_HEADS,
5929             KM_SLEEP);
5930         ipst->ips_phyint_g_list = kmem_zalloc(sizeof (phyint_list_t),
5931             KM_SLEEP);
5932         ipst->ips_ndp4 = kmem_zalloc(sizeof (ndp_g_t), KM_SLEEP);
5933         ipst->ips_ndp6 = kmem_zalloc(sizeof (ndp_g_t), KM_SLEEP);
5934         mutex_init(&ipst->ips_ndp4->ndp_g_lock, NULL, MUTEX_DEFAULT, NULL);
5935         mutex_init(&ipst->ips_ndp6->ndp_g_lock, NULL, MUTEX_DEFAULT, NULL);
5936 
5937         rw_init(&ipst->ips_ip_g_nd_lock, NULL, RW_DEFAULT, NULL);
5938         mutex_init(&ipst->ips_igmp_timer_lock, NULL, MUTEX_DEFAULT, NULL);
5939         ipst->ips_igmp_deferred_next = INFINITY;
5940         mutex_init(&ipst->ips_mld_timer_lock, NULL, MUTEX_DEFAULT, NULL);
5941         ipst->ips_mld_deferred_next = INFINITY;
5942         mutex_init(&ipst->ips_igmp_slowtimeout_lock, NULL, MUTEX_DEFAULT, NULL);
5943         mutex_init(&ipst->ips_mld_slowtimeout_lock, NULL, MUTEX_DEFAULT, NULL);
5944         mutex_init(&ipst->ips_ip_mi_lock, NULL, MUTEX_DEFAULT, NULL);
5945         mutex_init(&ipst->ips_ip_addr_avail_lock, NULL, MUTEX_DEFAULT, NULL);
5946         rw_init(&ipst->ips_ill_g_lock, NULL, RW_DEFAULT, NULL);
5947         rw_init(&ipst->ips_ipsec_capab_ills_lock, NULL, RW_DEFAULT, NULL);
5948         rw_init(&ipst->ips_ill_g_usesrc_lock, NULL, RW_DEFAULT, NULL);
5949 
5950         ipcl_init(ipst);
5951         ip_ire_init(ipst);
5952         ip6_asp_init(ipst);
5953         ipif_init(ipst);
5954         conn_drain_init(ipst);
5955         ip_mrouter_stack_init(ipst);
5956 
5957         ipst->ips_ip_g_frag_timeout = IP_FRAG_TIMEOUT;
5958         ipst->ips_ip_g_frag_timo_ms = IP_FRAG_TIMEOUT * 1000;
5959 
5960         ipst->ips_ip_multirt_log_interval = 1000;
5961 
5962         ipst->ips_ip_g_forward = IP_FORWARD_DEFAULT;
5963         ipst->ips_ipv6_forward = IP_FORWARD_DEFAULT;
5964         ipst->ips_ill_index = 1;
5965 
5966         ipst->ips_saved_ip_g_forward = -1;
5967         ipst->ips_reg_vif_num = ALL_VIFS;    /* Index to Register vif */
5968 
5969         pa = (ipparam_t *)kmem_alloc(sizeof (lcl_param_arr), KM_SLEEP);
5970         ipst->ips_param_arr = pa;
5971         bcopy(lcl_param_arr, ipst->ips_param_arr, sizeof (lcl_param_arr));
5972 
5973         na = (ipndp_t *)kmem_alloc(sizeof (lcl_ndp_arr), KM_SLEEP);
5974         ipst->ips_ndp_arr = na;
5975         bcopy(lcl_ndp_arr, ipst->ips_ndp_arr, sizeof (lcl_ndp_arr));
5976         ipst->ips_ndp_arr[IPNDP_IP_FORWARDING_OFFSET].ip_ndp_data =
5977             (caddr_t)&ipst->ips_ip_g_forward;
5978         ipst->ips_ndp_arr[IPNDP_IP6_FORWARDING_OFFSET].ip_ndp_data =
5979             (caddr_t)&ipst->ips_ipv6_forward;
5980         ASSERT(strcmp(ipst->ips_ndp_arr[IPNDP_CGTP_FILTER_OFFSET].ip_ndp_name,
5981             "ip_cgtp_filter") == 0);
5982         ipst->ips_ndp_arr[IPNDP_CGTP_FILTER_OFFSET].ip_ndp_data =
5983             (caddr_t)&ipst->ips_ip_cgtp_filter;
5984         ASSERT(strcmp(ipst->ips_ndp_arr[IPNDP_IPMP_HOOK_OFFSET].ip_ndp_name,
5985             "ipmp_hook_emulation") == 0);
5986         ipst->ips_ndp_arr[IPNDP_IPMP_HOOK_OFFSET].ip_ndp_data =
5987             (caddr_t)&ipst->ips_ipmp_hook_emulation;
5988 
5989         (void) ip_param_register(&ipst->ips_ip_g_nd,
5990             ipst->ips_param_arr, A_CNT(lcl_param_arr),
5991             ipst->ips_ndp_arr, A_CNT(lcl_ndp_arr));
5992 
5993         ipst->ips_ip_mibkp = ip_kstat_init(stackid, ipst);
5994         ipst->ips_icmp_mibkp = icmp_kstat_init(stackid);
5995         ipst->ips_ip_kstat = ip_kstat2_init(stackid, &ipst->ips_ip_statistics);
5996         ipst->ips_ip6_kstat =
5997             ip6_kstat_init(stackid, &ipst->ips_ip6_statistics);
5998 
5999         ipst->ips_ipmp_enable_failback = B_TRUE;
6000 
6001         ipst->ips_ip_src_id = 1;
6002         rw_init(&ipst->ips_srcid_lock, NULL, RW_DEFAULT, NULL);
6003 
6004         ip_net_init(ipst, ns);
6005         ipv4_hook_init(ipst);
6006         ipv6_hook_init(ipst);
6007 
6008         return (ipst);
6009 }
6010 
6011 /*
6012  * Allocate and initialize a DLPI template of the specified length.  (May be
6013  * called as writer.)
6014  */
6015 mblk_t *
6016 ip_dlpi_alloc(size_t len, t_uscalar_t prim)
6017 {
6018         mblk_t  *mp;
6019 
6020         mp = allocb(len, BPRI_MED);
6021         if (!mp)
6022                 return (NULL);
6023 
6024         /*
6025          * DLPIv2 says that DL_INFO_REQ and DL_TOKEN_REQ (the latter
6026          * of which we don't seem to use) are sent with M_PCPROTO, and
6027          * that other DLPI are M_PROTO.
6028          */
6029         if (prim == DL_INFO_REQ) {
6030                 mp->b_datap->db_type = M_PCPROTO;
6031         } else {
6032                 mp->b_datap->db_type = M_PROTO;
6033         }
6034 
6035         mp->b_wptr = mp->b_rptr + len;
6036         bzero(mp->b_rptr, len);
6037         ((dl_unitdata_req_t *)mp->b_rptr)->dl_primitive = prim;
6038         return (mp);
6039 }
6040 
6041 /*
6042  * Debug formatting routine.  Returns a character string representation of the
6043  * addr in buf, of the form xxx.xxx.xxx.xxx.  This routine takes the address
6044  * in the form of a ipaddr_t and calls ip_dot_saddr with a pointer.
6045  *
6046  * Once the ndd table-printing interfaces are removed, this can be changed to
6047  * standard dotted-decimal form.
6048  */
6049 char *
6050 ip_dot_addr(ipaddr_t addr, char *buf)
6051 {
6052         uint8_t *ap = (uint8_t *)&addr;
6053 
6054         (void) mi_sprintf(buf, "%03d.%03d.%03d.%03d",
6055             ap[0] & 0xFF, ap[1] & 0xFF, ap[2] & 0xFF, ap[3] & 0xFF);
6056         return (buf);
6057 }
6058 
6059 /*
6060  * Write the given MAC address as a printable string in the usual colon-
6061  * separated format.
6062  */
6063 const char *
6064 mac_colon_addr(const uint8_t *addr, size_t alen, char *buf, size_t buflen)
6065 {
6066         char *bp;
6067 
6068         if (alen == 0 || buflen < 4)
6069                 return ("?");
6070         bp = buf;
6071         for (;;) {
6072                 /*
6073                  * If there are more MAC address bytes available, but we won't
6074                  * have any room to print them, then add "..." to the string
6075                  * instead.  See below for the 'magic number' explanation.
6076                  */
6077                 if ((alen == 2 && buflen < 6) || (alen > 2 && buflen < 7)) {
6078                         (void) strcpy(bp, "...");
6079                         break;
6080                 }
6081                 (void) sprintf(bp, "%02x", *addr++);
6082                 bp += 2;
6083                 if (--alen == 0)
6084                         break;
6085                 *bp++ = ':';
6086                 buflen -= 3;
6087                 /*
6088                  * At this point, based on the first 'if' statement above,
6089                  * either alen == 1 and buflen >= 3, or alen > 1 and
6090                  * buflen >= 4.  The first case leaves room for the final "xx"
6091                  * number and trailing NUL byte.  The second leaves room for at
6092                  * least "...".  Thus the apparently 'magic' numbers chosen for
6093                  * that statement.
6094                  */
6095         }
6096         return (buf);
6097 }
6098 
6099 /*
6100  * Send an ICMP error after patching up the packet appropriately.  Returns
6101  * non-zero if the appropriate MIB should be bumped; zero otherwise.
6102  */
6103 static boolean_t
6104 ip_fanout_send_icmp(queue_t *q, mblk_t *mp, uint_t flags,
6105     uint_t icmp_type, uint_t icmp_code, boolean_t mctl_present,
6106     zoneid_t zoneid, ip_stack_t *ipst)
6107 {
6108         ipha_t *ipha;
6109         mblk_t *first_mp;
6110         boolean_t secure;
6111         unsigned char db_type;
6112         ipsec_stack_t   *ipss = ipst->ips_netstack->netstack_ipsec;
6113 
6114         first_mp = mp;
6115         if (mctl_present) {
6116                 mp = mp->b_cont;
6117                 secure = ipsec_in_is_secure(first_mp);
6118                 ASSERT(mp != NULL);
6119         } else {
6120                 /*
6121                  * If this is an ICMP error being reported - which goes
6122                  * up as M_CTLs, we need to convert them to M_DATA till
6123                  * we finish checking with global policy because
6124                  * ipsec_check_global_policy() assumes M_DATA as clear
6125                  * and M_CTL as secure.
6126                  */
6127                 db_type = DB_TYPE(mp);
6128                 DB_TYPE(mp) = M_DATA;
6129                 secure = B_FALSE;
6130         }
6131         /*
6132          * We are generating an icmp error for some inbound packet.
6133          * Called from all ip_fanout_(udp, tcp, proto) functions.
6134          * Before we generate an error, check with global policy
6135          * to see whether this is allowed to enter the system. As
6136          * there is no "conn", we are checking with global policy.
6137          */
6138         ipha = (ipha_t *)mp->b_rptr;
6139         if (secure || ipss->ipsec_inbound_v4_policy_present) {
6140                 first_mp = ipsec_check_global_policy(first_mp, NULL,
6141                     ipha, NULL, mctl_present, ipst->ips_netstack);
6142                 if (first_mp == NULL)
6143                         return (B_FALSE);
6144         }
6145 
6146         if (!mctl_present)
6147                 DB_TYPE(mp) = db_type;
6148 
6149         if (flags & IP_FF_SEND_ICMP) {
6150                 if (flags & IP_FF_HDR_COMPLETE) {
6151                         if (ip_hdr_complete(ipha, zoneid, ipst)) {
6152                                 freemsg(first_mp);
6153                                 return (B_TRUE);
6154                         }
6155                 }
6156                 if (flags & IP_FF_CKSUM) {
6157                         /*
6158                          * Have to correct checksum since
6159                          * the packet might have been
6160                          * fragmented and the reassembly code in ip_rput
6161                          * does not restore the IP checksum.
6162                          */
6163                         ipha->ipha_hdr_checksum = 0;
6164                         ipha->ipha_hdr_checksum = ip_csum_hdr(ipha);
6165                 }
6166                 switch (icmp_type) {
6167                 case ICMP_DEST_UNREACHABLE:
6168                         icmp_unreachable(WR(q), first_mp, icmp_code, zoneid,
6169                             ipst);
6170                         break;
6171                 default:
6172                         freemsg(first_mp);
6173                         break;
6174                 }
6175         } else {
6176                 freemsg(first_mp);
6177                 return (B_FALSE);
6178         }
6179 
6180         return (B_TRUE);
6181 }
6182 
6183 /*
6184  * Used to send an ICMP error message when a packet is received for
6185  * a protocol that is not supported. The mblk passed as argument
6186  * is consumed by this function.
6187  */
6188 void
6189 ip_proto_not_sup(queue_t *q, mblk_t *ipsec_mp, uint_t flags, zoneid_t zoneid,
6190     ip_stack_t *ipst)
6191 {
6192         mblk_t *mp;
6193         ipha_t *ipha;
6194         ill_t *ill;
6195         ipsec_in_t *ii;
6196 
6197         ii = (ipsec_in_t *)ipsec_mp->b_rptr;
6198         ASSERT(ii->ipsec_in_type == IPSEC_IN);
6199 
6200         mp = ipsec_mp->b_cont;
6201         ipsec_mp->b_cont = NULL;
6202         ipha = (ipha_t *)mp->b_rptr;
6203         /* Get ill from index in ipsec_in_t. */
6204         ill = ill_lookup_on_ifindex(ii->ipsec_in_ill_index,
6205             (IPH_HDR_VERSION(ipha) == IPV6_VERSION), NULL, NULL, NULL, NULL,
6206             ipst);
6207         if (ill != NULL) {
6208                 if (IPH_HDR_VERSION(ipha) == IP_VERSION) {
6209                         if (ip_fanout_send_icmp(q, mp, flags,
6210                             ICMP_DEST_UNREACHABLE,
6211                             ICMP_PROTOCOL_UNREACHABLE, B_FALSE, zoneid, ipst)) {
6212                                 BUMP_MIB(ill->ill_ip_mib,
6213                                     ipIfStatsInUnknownProtos);
6214                         }
6215                 } else {
6216                         if (ip_fanout_send_icmp_v6(q, mp, flags,
6217                             ICMP6_PARAM_PROB, ICMP6_PARAMPROB_NEXTHEADER,
6218                             0, B_FALSE, zoneid, ipst)) {
6219                                 BUMP_MIB(ill->ill_ip_mib,
6220                                     ipIfStatsInUnknownProtos);
6221                         }
6222                 }
6223                 ill_refrele(ill);
6224         } else { /* re-link for the freemsg() below. */
6225                 ipsec_mp->b_cont = mp;
6226         }
6227 
6228         /* If ICMP delivered, ipsec_mp will be a singleton (b_cont == NULL). */
6229         freemsg(ipsec_mp);
6230 }
6231 
6232 /*
6233  * See if the inbound datagram has had IPsec processing applied to it.
6234  */
6235 boolean_t
6236 ipsec_in_is_secure(mblk_t *ipsec_mp)
6237 {
6238         ipsec_in_t *ii;
6239 
6240         ii = (ipsec_in_t *)ipsec_mp->b_rptr;
6241         ASSERT(ii->ipsec_in_type == IPSEC_IN);
6242 
6243         if (ii->ipsec_in_loopback) {
6244                 return (ii->ipsec_in_secure);
6245         } else {
6246                 return (ii->ipsec_in_ah_sa != NULL ||
6247                     ii->ipsec_in_esp_sa != NULL ||
6248                     ii->ipsec_in_decaps);
6249         }
6250 }
6251 
6252 /*
6253  * Handle protocols with which IP is less intimate.  There
6254  * can be more than one stream bound to a particular
6255  * protocol.  When this is the case, normally each one gets a copy
6256  * of any incoming packets.
6257  *
6258  * IPsec NOTE :
6259  *
6260  * Don't allow a secure packet going up a non-secure connection.
6261  * We don't allow this because
6262  *
6263  * 1) Reply might go out in clear which will be dropped at
6264  *    the sending side.
6265  * 2) If the reply goes out in clear it will give the
6266  *    adversary enough information for getting the key in
6267  *    most of the cases.
6268  *
6269  * Moreover getting a secure packet when we expect clear
6270  * implies that SA's were added without checking for
6271  * policy on both ends. This should not happen once ISAKMP
6272  * is used to negotiate SAs as SAs will be added only after
6273  * verifying the policy.
6274  *
6275  * NOTE : If the packet was tunneled and not multicast we only send
6276  * to it the first match. Unlike TCP and UDP fanouts this doesn't fall
6277  * back to delivering packets to AF_INET6 raw sockets.
6278  *
6279  * IPQoS Notes:
6280  * Once we have determined the client, invoke IPPF processing.
6281  * Policy processing takes place only if the callout_position, IPP_LOCAL_IN,
6282  * is enabled. If we get here from icmp_inbound_error_fanout or ip_wput_local
6283  * ip_policy will be false.
6284  *
6285  * Zones notes:
6286  * Currently only applications in the global zone can create raw sockets for
6287  * protocols other than ICMP. So unlike the broadcast / multicast case of
6288  * ip_fanout_udp(), we only send a copy of the packet to streams in the
6289  * specified zone. For ICMP, this is handled by the callers of icmp_inbound().
6290  */
6291 static void
6292 ip_fanout_proto(queue_t *q, mblk_t *mp, ill_t *ill, ipha_t *ipha, uint_t flags,
6293     boolean_t mctl_present, boolean_t ip_policy, ill_t *recv_ill,
6294     zoneid_t zoneid)
6295 {
6296         queue_t *rq;
6297         mblk_t  *mp1, *first_mp1;
6298         uint_t  protocol = ipha->ipha_protocol;
6299         ipaddr_t dst;
6300         boolean_t one_only;
6301         mblk_t *first_mp = mp;
6302         boolean_t secure;
6303         uint32_t ill_index;
6304         conn_t  *connp, *first_connp, *next_connp;
6305         connf_t *connfp;
6306         boolean_t shared_addr;
6307         mib2_ipIfStatsEntry_t *mibptr;
6308         ip_stack_t *ipst = recv_ill->ill_ipst;
6309         ipsec_stack_t   *ipss = ipst->ips_netstack->netstack_ipsec;
6310 
6311         mibptr = (ill != NULL) ? ill->ill_ip_mib : &ipst->ips_ip_mib;
6312         if (mctl_present) {
6313                 mp = first_mp->b_cont;
6314                 secure = ipsec_in_is_secure(first_mp);
6315                 ASSERT(mp != NULL);
6316         } else {
6317                 secure = B_FALSE;
6318         }
6319         dst = ipha->ipha_dst;
6320         /*
6321          * If the packet was tunneled and not multicast we only send to it
6322          * the first match.
6323          */
6324         one_only = ((protocol == IPPROTO_ENCAP || protocol == IPPROTO_IPV6) &&
6325             !CLASSD(dst));
6326 
6327         shared_addr = (zoneid == ALL_ZONES);
6328         if (shared_addr) {
6329                 /*
6330                  * We don't allow multilevel ports for raw IP, so no need to
6331                  * check for that here.
6332                  */
6333                 zoneid = tsol_packet_to_zoneid(mp);
6334         }
6335 
6336         connfp = &ipst->ips_ipcl_proto_fanout[protocol];
6337         mutex_enter(&connfp->connf_lock);
6338         connp = connfp->connf_head;
6339         for (connp = connfp->connf_head; connp != NULL;
6340             connp = connp->conn_next) {
6341                 if (IPCL_PROTO_MATCH(connp, protocol, ipha, ill, flags,
6342                     zoneid) &&
6343                     (!is_system_labeled() ||
6344                     tsol_receive_local(mp, &dst, IPV4_VERSION, shared_addr,
6345                     connp))) {
6346                         break;
6347                 }
6348         }
6349 
6350         if (connp == NULL || connp->conn_upq == NULL) {
6351                 /*
6352                  * No one bound to these addresses.  Is
6353                  * there a client that wants all
6354                  * unclaimed datagrams?
6355                  */
6356                 mutex_exit(&connfp->connf_lock);
6357                 /*
6358                  * Check for IPPROTO_ENCAP...
6359                  */
6360                 if (protocol == IPPROTO_ENCAP && ipst->ips_ip_g_mrouter) {
6361                         /*
6362                          * If an IPsec mblk is here on a multicast
6363                          * tunnel (using ip_mroute stuff), check policy here,
6364                          * THEN ship off to ip_mroute_decap().
6365                          *
6366                          * BTW,  If I match a configured IP-in-IP
6367                          * tunnel, this path will not be reached, and
6368                          * ip_mroute_decap will never be called.
6369                          */
6370                         first_mp = ipsec_check_global_policy(first_mp, connp,
6371                             ipha, NULL, mctl_present, ipst->ips_netstack);
6372                         if (first_mp != NULL) {
6373                                 if (mctl_present)
6374                                         freeb(first_mp);
6375                                 ip_mroute_decap(q, mp, ill);
6376                         } /* Else we already freed everything! */
6377                 } else {
6378                         /*
6379                          * Otherwise send an ICMP protocol unreachable.
6380                          */
6381                         if (ip_fanout_send_icmp(q, first_mp, flags,
6382                             ICMP_DEST_UNREACHABLE, ICMP_PROTOCOL_UNREACHABLE,
6383                             mctl_present, zoneid, ipst)) {
6384                                 BUMP_MIB(mibptr, ipIfStatsInUnknownProtos);
6385                         }
6386                 }
6387                 return;
6388         }
6389         CONN_INC_REF(connp);
6390         first_connp = connp;
6391 
6392         /*
6393          * Only send message to one tunnel driver by immediately
6394          * terminating the loop.
6395          */
6396         connp = one_only ? NULL : connp->conn_next;
6397 
6398         for (;;) {
6399                 while (connp != NULL) {
6400                         if (IPCL_PROTO_MATCH(connp, protocol, ipha, ill,
6401                             flags, zoneid) &&
6402                             (!is_system_labeled() ||
6403                             tsol_receive_local(mp, &dst, IPV4_VERSION,
6404                             shared_addr, connp)))
6405                                 break;
6406                         connp = connp->conn_next;
6407                 }
6408 
6409                 /*
6410                  * Copy the packet.
6411                  */
6412                 if (connp == NULL || connp->conn_upq == NULL ||
6413                     (((first_mp1 = dupmsg(first_mp)) == NULL) &&
6414                     ((first_mp1 = ip_copymsg(first_mp)) == NULL))) {
6415                         /*
6416                          * No more interested clients or memory
6417                          * allocation failed
6418                          */
6419                         connp = first_connp;
6420                         break;
6421                 }
6422                 mp1 = mctl_present ? first_mp1->b_cont : first_mp1;
6423                 CONN_INC_REF(connp);
6424                 mutex_exit(&connfp->connf_lock);
6425                 rq = connp->conn_rq;
6426                 if (!canputnext(rq)) {
6427                         if (flags & IP_FF_RAWIP) {
6428                                 BUMP_MIB(mibptr, rawipIfStatsInOverflows);
6429                         } else {
6430                                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInOverflows);
6431                         }
6432 
6433                         freemsg(first_mp1);
6434                 } else {
6435                         /*
6436                          * Don't enforce here if we're an actual tunnel -
6437                          * let "tun" do it instead.
6438                          */
6439                         if (!IPCL_IS_IPTUN(connp) &&
6440                             (CONN_INBOUND_POLICY_PRESENT(connp, ipss) ||
6441                             secure)) {
6442                                 first_mp1 = ipsec_check_inbound_policy
6443                                     (first_mp1, connp, ipha, NULL,
6444                                     mctl_present);
6445                         }
6446                         if (first_mp1 != NULL) {
6447                                 int in_flags = 0;
6448                                 /*
6449                                  * ip_fanout_proto also gets called from
6450                                  * icmp_inbound_error_fanout, in which case
6451                                  * the msg type is M_CTL.  Don't add info
6452                                  * in this case for the time being. In future
6453                                  * when there is a need for knowing the
6454                                  * inbound iface index for ICMP error msgs,
6455                                  * then this can be changed.
6456                                  */
6457                                 if (connp->conn_recvif)
6458                                         in_flags = IPF_RECVIF;
6459                                 /*
6460                                  * The ULP may support IP_RECVPKTINFO for both
6461                                  * IP v4 and v6 so pass the appropriate argument
6462                                  * based on conn IP version.
6463                                  */
6464                                 if (connp->conn_ip_recvpktinfo) {
6465                                         if (connp->conn_af_isv6) {
6466                                                 /*
6467                                                  * V6 only needs index
6468                                                  */
6469                                                 in_flags |= IPF_RECVIF;
6470                                         } else {
6471                                                 /*
6472                                                  * V4 needs index +
6473                                                  * matching address.
6474                                                  */
6475                                                 in_flags |= IPF_RECVADDR;
6476                                         }
6477                                 }
6478                                 if ((in_flags != 0) &&
6479                                     (mp->b_datap->db_type != M_CTL)) {
6480                                         /*
6481                                          * the actual data will be
6482                                          * contained in b_cont upon
6483                                          * successful return of the
6484                                          * following call else
6485                                          * original mblk is returned
6486                                          */
6487                                         ASSERT(recv_ill != NULL);
6488                                         mp1 = ip_add_info(mp1, recv_ill,
6489                                             in_flags, IPCL_ZONEID(connp), ipst);
6490                                 }
6491                                 BUMP_MIB(mibptr, ipIfStatsHCInDelivers);
6492                                 if (mctl_present)
6493                                         freeb(first_mp1);
6494                                 (connp->conn_recv)(connp, mp1, NULL);
6495                         }
6496                 }
6497                 mutex_enter(&connfp->connf_lock);
6498                 /* Follow the next pointer before releasing the conn. */
6499                 next_connp = connp->conn_next;
6500                 CONN_DEC_REF(connp);
6501                 connp = next_connp;
6502         }
6503 
6504         /* Last one.  Send it upstream. */
6505         mutex_exit(&connfp->connf_lock);
6506 
6507         /*
6508          * If this packet is coming from icmp_inbound_error_fanout ip_policy
6509          * will be set to false.
6510          */
6511         if (IPP_ENABLED(IPP_LOCAL_IN, ipst) && ip_policy) {
6512                 ill_index = ill->ill_phyint->phyint_ifindex;
6513                 ip_process(IPP_LOCAL_IN, &mp, ill_index);
6514                 if (mp == NULL) {
6515                         CONN_DEC_REF(connp);
6516                         if (mctl_present) {
6517                                 freeb(first_mp);
6518                         }
6519                         return;
6520                 }
6521         }
6522 
6523         rq = connp->conn_rq;
6524         if (!canputnext(rq)) {
6525                 if (flags & IP_FF_RAWIP) {
6526                         BUMP_MIB(mibptr, rawipIfStatsInOverflows);
6527                 } else {
6528                         BUMP_MIB(&ipst->ips_icmp_mib, icmpInOverflows);
6529                 }
6530 
6531                 freemsg(first_mp);
6532         } else {
6533                 if (IPCL_IS_IPTUN(connp)) {
6534                         /*
6535                          * Tunneled packet.  We enforce policy in the tunnel
6536                          * module itself.
6537                          *
6538                          * Send the WHOLE packet up (incl. IPSEC_IN) without
6539                          * a policy check.
6540                          * FIXME to use conn_recv for tun later.
6541                          */
6542                         putnext(rq, first_mp);
6543                         CONN_DEC_REF(connp);
6544                         return;
6545                 }
6546 
6547                 if ((CONN_INBOUND_POLICY_PRESENT(connp, ipss) || secure)) {
6548                         first_mp = ipsec_check_inbound_policy(first_mp, connp,
6549                             ipha, NULL, mctl_present);
6550                 }
6551 
6552                 if (first_mp != NULL) {
6553                         int in_flags = 0;
6554 
6555                         /*
6556                          * ip_fanout_proto also gets called
6557                          * from icmp_inbound_error_fanout, in
6558                          * which case the msg type is M_CTL.
6559                          * Don't add info in this case for time
6560                          * being. In future when there is a
6561                          * need for knowing the inbound iface
6562                          * index for ICMP error msgs, then this
6563                          * can be changed
6564                          */
6565                         if (connp->conn_recvif)
6566                                 in_flags = IPF_RECVIF;
6567                         if (connp->conn_ip_recvpktinfo) {
6568                                 if (connp->conn_af_isv6) {
6569                                         /*
6570                                          * V6 only needs index
6571                                          */
6572                                         in_flags |= IPF_RECVIF;
6573                                 } else {
6574                                         /*
6575                                          * V4 needs index +
6576                                          * matching address.
6577                                          */
6578                                         in_flags |= IPF_RECVADDR;
6579                                 }
6580                         }
6581                         if ((in_flags != 0) &&
6582                             (mp->b_datap->db_type != M_CTL)) {
6583 
6584                                 /*
6585                                  * the actual data will be contained in
6586                                  * b_cont upon successful return
6587                                  * of the following call else original
6588                                  * mblk is returned
6589                                  */
6590                                 ASSERT(recv_ill != NULL);
6591                                 mp = ip_add_info(mp, recv_ill,
6592                                     in_flags, IPCL_ZONEID(connp), ipst);
6593                         }
6594                         BUMP_MIB(mibptr, ipIfStatsHCInDelivers);
6595                         (connp->conn_recv)(connp, mp, NULL);
6596                         if (mctl_present)
6597                                 freeb(first_mp);
6598                 }
6599         }
6600         CONN_DEC_REF(connp);
6601 }
6602 
6603 /*
6604  * Fanout for TCP packets
6605  * The caller puts <fport, lport> in the ports parameter.
6606  *
6607  * IPQoS Notes
6608  * Before sending it to the client, invoke IPPF processing.
6609  * Policy processing takes place only if the callout_position, IPP_LOCAL_IN,
6610  * is enabled. If we get here from icmp_inbound_error_fanout or ip_wput_local
6611  * ip_policy is false.
6612  */
6613 static void
6614 ip_fanout_tcp(queue_t *q, mblk_t *mp, ill_t *recv_ill, ipha_t *ipha,
6615     uint_t flags, boolean_t mctl_present, boolean_t ip_policy, zoneid_t zoneid)
6616 {
6617         mblk_t  *first_mp;
6618         boolean_t secure;
6619         uint32_t ill_index;
6620         int     ip_hdr_len;
6621         tcph_t  *tcph;
6622         boolean_t syn_present = B_FALSE;
6623         conn_t  *connp;
6624         ip_stack_t      *ipst = recv_ill->ill_ipst;
6625         ipsec_stack_t   *ipss = ipst->ips_netstack->netstack_ipsec;
6626 
6627         ASSERT(recv_ill != NULL);
6628 
6629         first_mp = mp;
6630         if (mctl_present) {
6631                 ASSERT(first_mp->b_datap->db_type == M_CTL);
6632                 mp = first_mp->b_cont;
6633                 secure = ipsec_in_is_secure(first_mp);
6634                 ASSERT(mp != NULL);
6635         } else {
6636                 secure = B_FALSE;
6637         }
6638 
6639         ip_hdr_len = IPH_HDR_LENGTH(mp->b_rptr);
6640 
6641         if ((connp = ipcl_classify_v4(mp, IPPROTO_TCP, ip_hdr_len,
6642             zoneid, ipst)) == NULL) {
6643                 /*
6644                  * No connected connection or listener. Send a
6645                  * TH_RST via tcp_xmit_listeners_reset.
6646                  */
6647 
6648                 /* Initiate IPPf processing, if needed. */
6649                 if (IPP_ENABLED(IPP_LOCAL_IN, ipst)) {
6650                         uint32_t ill_index;
6651                         ill_index = recv_ill->ill_phyint->phyint_ifindex;
6652                         ip_process(IPP_LOCAL_IN, &first_mp, ill_index);
6653                         if (first_mp == NULL)
6654                                 return;
6655                 }
6656                 BUMP_MIB(recv_ill->ill_ip_mib, ipIfStatsHCInDelivers);
6657                 ip2dbg(("ip_fanout_tcp: no listener; send reset to zone %d\n",
6658                     zoneid));
6659                 tcp_xmit_listeners_reset(first_mp, ip_hdr_len, zoneid,
6660                     ipst->ips_netstack->netstack_tcp, NULL);
6661                 return;
6662         }
6663 
6664         /*
6665          * Allocate the SYN for the TCP connection here itself
6666          */
6667         tcph = (tcph_t *)&mp->b_rptr[ip_hdr_len];
6668         if ((tcph->th_flags[0] & (TH_SYN|TH_ACK|TH_RST|TH_URG)) == TH_SYN) {
6669                 if (IPCL_IS_TCP(connp)) {
6670                         squeue_t *sqp;
6671 
6672                         /*
6673                          * For fused tcp loopback, assign the eager's
6674                          * squeue to be that of the active connect's.
6675                          * Note that we don't check for IP_FF_LOOPBACK
6676                          * here since this routine gets called only
6677                          * for loopback (unlike the IPv6 counterpart).
6678                          */
6679                         ASSERT(Q_TO_CONN(q) != NULL);
6680                         if (do_tcp_fusion &&
6681                             !CONN_INBOUND_POLICY_PRESENT(connp, ipss) &&
6682                             !secure &&
6683                             !IPP_ENABLED(IPP_LOCAL_IN, ipst) && !ip_policy &&
6684                             IPCL_IS_TCP(Q_TO_CONN(q))) {
6685                                 ASSERT(Q_TO_CONN(q)->conn_sqp != NULL);
6686                                 sqp = Q_TO_CONN(q)->conn_sqp;
6687                         } else {
6688                                 sqp = IP_SQUEUE_GET(lbolt);
6689                         }
6690 
6691                         mp->b_datap->db_struioflag |= STRUIO_EAGER;
6692                         DB_CKSUMSTART(mp) = (intptr_t)sqp;
6693                         syn_present = B_TRUE;
6694                 }
6695         }
6696 
6697         if (IPCL_IS_TCP(connp) && IPCL_IS_BOUND(connp) && !syn_present) {
6698                 uint_t  flags = (unsigned int)tcph->th_flags[0] & 0xFF;
6699                 BUMP_MIB(recv_ill->ill_ip_mib, ipIfStatsHCInDelivers);
6700                 if ((flags & TH_RST) || (flags & TH_URG)) {
6701                         DTRACE_TCP5(receive, mblk_t *, NULL, conn_t *, NULL,
6702                             void_ip_t *, ipha, tcp_t *, NULL, tcph_t *, tcph);
6703                         CONN_DEC_REF(connp);
6704                         freemsg(first_mp);
6705                         return;
6706                 }
6707                 if (flags & TH_ACK) {
6708                         tcp_xmit_listeners_reset(first_mp, ip_hdr_len, zoneid,
6709                             ipst->ips_netstack->netstack_tcp, connp);
6710                         CONN_DEC_REF(connp);
6711                         return;
6712                 }
6713 
6714                 DTRACE_TCP5(receive, mblk_t *, NULL, conn_t *, NULL,
6715                     void_ip_t *, ipha, tcp_t *, NULL, tcph_t *, tcph);
6716                 CONN_DEC_REF(connp);
6717                 freemsg(first_mp);
6718                 return;
6719         }
6720 
6721         if (CONN_INBOUND_POLICY_PRESENT(connp, ipss) || secure) {
6722                 first_mp = ipsec_check_inbound_policy(first_mp, connp, ipha,
6723                     NULL, mctl_present);
6724                 if (first_mp == NULL) {
6725                         BUMP_MIB(recv_ill->ill_ip_mib, ipIfStatsInDiscards);
6726                         CONN_DEC_REF(connp);
6727                         return;
6728                 }
6729                 if (IPCL_IS_TCP(connp) && IPCL_IS_BOUND(connp)) {
6730                         ASSERT(syn_present);
6731                         if (mctl_present) {
6732                                 ASSERT(first_mp != mp);
6733                                 first_mp->b_datap->db_struioflag |=
6734                                     STRUIO_POLICY;
6735                         } else {
6736                                 ASSERT(first_mp == mp);
6737                                 mp->b_datap->db_struioflag &=
6738                                     ~STRUIO_EAGER;
6739                                 mp->b_datap->db_struioflag |=
6740                                     STRUIO_POLICY;
6741                         }
6742                 } else {
6743                         /*
6744                          * Discard first_mp early since we're dealing with a
6745                          * fully-connected conn_t and tcp doesn't do policy in
6746                          * this case.
6747                          */
6748                         if (mctl_present) {
6749                                 freeb(first_mp);
6750                                 mctl_present = B_FALSE;
6751                         }
6752                         first_mp = mp;
6753                 }
6754         }
6755 
6756         /*
6757          * Initiate policy processing here if needed. If we get here from
6758          * icmp_inbound_error_fanout, ip_policy is false.
6759          */
6760         if (IPP_ENABLED(IPP_LOCAL_IN, ipst) && ip_policy) {
6761                 ill_index = recv_ill->ill_phyint->phyint_ifindex;
6762                 ip_process(IPP_LOCAL_IN, &mp, ill_index);
6763                 if (mp == NULL) {
6764                         CONN_DEC_REF(connp);
6765                         if (mctl_present)
6766                                 freeb(first_mp);
6767                         return;
6768                 } else if (mctl_present) {
6769                         ASSERT(first_mp != mp);
6770                         first_mp->b_cont = mp;
6771                 } else {
6772                         first_mp = mp;
6773                 }
6774         }
6775 
6776 
6777 
6778         /* Handle socket options. */
6779         if (!syn_present &&
6780             connp->conn_ip_recvpktinfo && (flags & IP_FF_IPINFO)) {
6781                 /* Add header */
6782                 ASSERT(recv_ill != NULL);
6783                 /*
6784                  * Since tcp does not support IP_RECVPKTINFO for V4, only pass
6785                  * IPF_RECVIF.
6786                  */
6787                 mp = ip_add_info(mp, recv_ill, IPF_RECVIF, IPCL_ZONEID(connp),
6788                     ipst);
6789                 if (mp == NULL) {
6790                         BUMP_MIB(recv_ill->ill_ip_mib, ipIfStatsInDiscards);
6791                         CONN_DEC_REF(connp);
6792                         if (mctl_present)
6793                                 freeb(first_mp);
6794                         return;
6795                 } else if (mctl_present) {
6796                         /*
6797                          * ip_add_info might return a new mp.
6798                          */
6799                         ASSERT(first_mp != mp);
6800                         first_mp->b_cont = mp;
6801                 } else {
6802                         first_mp = mp;
6803                 }
6804         }
6805         BUMP_MIB(recv_ill->ill_ip_mib, ipIfStatsHCInDelivers);
6806         if (IPCL_IS_TCP(connp)) {
6807                 /* do not drain, certain use cases can blow the stack */
6808                 squeue_enter_nodrain(connp->conn_sqp, first_mp,
6809                     connp->conn_recv, connp, SQTAG_IP_FANOUT_TCP);
6810         } else {
6811                 /* Not TCP; must be SOCK_RAW, IPPROTO_TCP */
6812                 (connp->conn_recv)(connp, first_mp, NULL);
6813                 CONN_DEC_REF(connp);
6814         }
6815 }
6816 
6817 /*
6818  * If we have a IPsec NAT-Traversal packet, strip the zero-SPI or
6819  * pass it along to ESP if the SPI is non-zero.  Returns TRUE if the mblk
6820  * is not consumed.
6821  *
6822  * One of four things can happen, all of which affect the passed-in mblk:
6823  *
6824  * 1.) ICMP messages that go through here just get returned TRUE.
6825  *
6826  * 2.) The packet is stock UDP and gets its zero-SPI stripped.  Return TRUE.
6827  *
6828  * 3.) The packet is ESP-in-UDP, gets transformed into an equivalent
6829  *     ESP packet, and is passed along to ESP for consumption.  Return FALSE.
6830  *
6831  * 4.) The packet is an ESP-in-UDP Keepalive.  Drop it and return FALSE.
6832  */
6833 static boolean_t
6834 zero_spi_check(queue_t *q, mblk_t *mp, ire_t *ire, ill_t *recv_ill,
6835     ipsec_stack_t *ipss)
6836 {
6837         int shift, plen, iph_len;
6838         ipha_t *ipha;
6839         udpha_t *udpha;
6840         uint32_t *spi;
6841         uint32_t esp_ports;
6842         uint8_t *orptr;
6843         boolean_t free_ire;
6844 
6845         if (DB_TYPE(mp) == M_CTL) {
6846                 /*
6847                  * ICMP message with UDP inside.  Don't bother stripping, just
6848                  * send it up.
6849                  *
6850                  * NOTE: Any app with UDP_NAT_T_ENDPOINT set is probably going
6851                  * to ignore errors set by ICMP anyway ('cause they might be
6852                  * forged), but that's the app's decision, not ours.
6853                  */
6854 
6855                 /* Bunch of reality checks for DEBUG kernels... */
6856                 ASSERT(IPH_HDR_VERSION(mp->b_rptr) == IPV4_VERSION);
6857                 ASSERT(((ipha_t *)mp->b_rptr)->ipha_protocol == IPPROTO_ICMP);
6858 
6859                 return (B_TRUE);
6860         }
6861 
6862         ipha = (ipha_t *)mp->b_rptr;
6863         iph_len = IPH_HDR_LENGTH(ipha);
6864         plen = ntohs(ipha->ipha_length);
6865 
6866         if (plen - iph_len - sizeof (udpha_t) < sizeof (uint32_t)) {
6867                 /*
6868                  * Most likely a keepalive for the benefit of an intervening
6869                  * NAT.  These aren't for us, per se, so drop it.
6870                  *
6871                  * RFC 3947/8 doesn't say for sure what to do for 2-3
6872                  * byte packets (keepalives are 1-byte), but we'll drop them
6873                  * also.
6874                  */
6875                 ip_drop_packet(mp, B_TRUE, recv_ill, NULL,
6876                     DROPPER(ipss, ipds_esp_nat_t_ka), &ipss->ipsec_dropper);
6877                 return (B_FALSE);
6878         }
6879 
6880         if (MBLKL(mp) < iph_len + sizeof (udpha_t) + sizeof (*spi)) {
6881                 /* might as well pull it all up - it might be ESP. */
6882                 if (!pullupmsg(mp, -1)) {
6883                         ip_drop_packet(mp, B_TRUE, recv_ill, NULL,
6884                             DROPPER(ipss, ipds_esp_nomem),
6885                             &ipss->ipsec_dropper);
6886                         return (B_FALSE);
6887                 }
6888 
6889                 ipha = (ipha_t *)mp->b_rptr;
6890         }
6891         spi = (uint32_t *)(mp->b_rptr + iph_len + sizeof (udpha_t));
6892         if (*spi == 0) {
6893                 /* UDP packet - remove 0-spi. */
6894                 shift = sizeof (uint32_t);
6895         } else {
6896                 /* ESP-in-UDP packet - reduce to ESP. */
6897                 ipha->ipha_protocol = IPPROTO_ESP;
6898                 shift = sizeof (udpha_t);
6899         }
6900 
6901         /* Fix IP header */
6902         ipha->ipha_length = htons(plen - shift);
6903         ipha->ipha_hdr_checksum = 0;
6904 
6905         orptr = mp->b_rptr;
6906         mp->b_rptr += shift;
6907 
6908         udpha = (udpha_t *)(orptr + iph_len);
6909         if (*spi == 0) {
6910                 ASSERT((uint8_t *)ipha == orptr);
6911                 udpha->uha_length = htons(plen - shift - iph_len);
6912                 iph_len += sizeof (udpha_t);    /* For the call to ovbcopy(). */
6913                 esp_ports = 0;
6914         } else {
6915                 esp_ports = *((uint32_t *)udpha);
6916                 ASSERT(esp_ports != 0);
6917         }
6918         ovbcopy(orptr, orptr + shift, iph_len);
6919         if (esp_ports != 0) /* Punt up for ESP processing. */ {
6920                 ipha = (ipha_t *)(orptr + shift);
6921 
6922                 free_ire = (ire == NULL);
6923                 if (free_ire) {
6924                         /* Re-acquire ire. */
6925                         ire = ire_cache_lookup(ipha->ipha_dst, ALL_ZONES, NULL,
6926                             ipss->ipsec_netstack->netstack_ip);
6927                         if (ire == NULL || !(ire->ire_type & IRE_LOCAL)) {
6928                                 if (ire != NULL)
6929                                         ire_refrele(ire);
6930                                 /*
6931                                  * Do a regular freemsg(), as this is an IP
6932                                  * error (no local route) not an IPsec one.
6933                                  */
6934                                 freemsg(mp);
6935                         }
6936                 }
6937 
6938                 ip_proto_input(q, mp, ipha, ire, recv_ill, esp_ports);
6939                 if (free_ire)
6940                         ire_refrele(ire);
6941         }
6942 
6943         return (esp_ports == 0);
6944 }
6945 
6946 /*
6947  * Deliver a udp packet to the given conn, possibly applying ipsec policy.
6948  * We are responsible for disposing of mp, such as by freemsg() or putnext()
6949  * Caller is responsible for dropping references to the conn, and freeing
6950  * first_mp.
6951  *
6952  * IPQoS Notes
6953  * Before sending it to the client, invoke IPPF processing. Policy processing
6954  * takes place only if the callout_position, IPP_LOCAL_IN, is enabled and
6955  * ip_policy is true. If we get here from icmp_inbound_error_fanout or
6956  * ip_wput_local, ip_policy is false.
6957  */
6958 static void
6959 ip_fanout_udp_conn(conn_t *connp, mblk_t *first_mp, mblk_t *mp,
6960     boolean_t secure, ill_t *ill, ipha_t *ipha, uint_t flags, ill_t *recv_ill,
6961     boolean_t ip_policy)
6962 {
6963         boolean_t       mctl_present = (first_mp != NULL);
6964         uint32_t        in_flags = 0; /* set to IP_RECVSLLA and/or IP_RECVIF */
6965         uint32_t        ill_index;
6966         ip_stack_t      *ipst = recv_ill->ill_ipst;
6967         ipsec_stack_t   *ipss = ipst->ips_netstack->netstack_ipsec;
6968 
6969         ASSERT(ill != NULL);
6970 
6971         if (mctl_present)
6972                 first_mp->b_cont = mp;
6973         else
6974                 first_mp = mp;
6975 
6976         if (CONN_UDP_FLOWCTLD(connp)) {
6977                 BUMP_MIB(ill->ill_ip_mib, udpIfStatsInOverflows);
6978                 freemsg(first_mp);
6979                 return;
6980         }
6981 
6982         if (CONN_INBOUND_POLICY_PRESENT(connp, ipss) || secure) {
6983                 first_mp = ipsec_check_inbound_policy(first_mp, connp, ipha,
6984                     NULL, mctl_present);
6985                 if (first_mp == NULL) {
6986                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
6987                         return; /* Freed by ipsec_check_inbound_policy(). */
6988                 }
6989         }
6990         if (mctl_present)
6991                 freeb(first_mp);
6992 
6993         /* Let's hope the compilers utter "branch, predict-not-taken..." ;) */
6994         if (connp->conn_udp->udp_nat_t_endpoint) {
6995                 if (mctl_present) {
6996                         /* mctl_present *shouldn't* happen. */
6997                         ip_drop_packet(mp, B_TRUE, NULL, NULL,
6998                             DROPPER(ipss, ipds_esp_nat_t_ipsec),
6999                             &ipss->ipsec_dropper);
7000                         return;
7001                 }
7002 
7003                 if (!zero_spi_check(ill->ill_rq, mp, NULL, recv_ill, ipss))
7004                         return;
7005         }
7006 
7007         /* Handle options. */
7008         if (connp->conn_recvif)
7009                 in_flags = IPF_RECVIF;
7010         /*
7011          * UDP supports IP_RECVPKTINFO option for both v4 and v6 so the flag
7012          * passed to ip_add_info is based on IP version of connp.
7013          */
7014         if (connp->conn_ip_recvpktinfo && (flags & IP_FF_IPINFO)) {
7015                 if (connp->conn_af_isv6) {
7016                         /*
7017                          * V6 only needs index
7018                          */
7019                         in_flags |= IPF_RECVIF;
7020                 } else {
7021                         /*
7022                          * V4 needs index + matching address.
7023                          */
7024                         in_flags |= IPF_RECVADDR;
7025                 }
7026         }
7027 
7028         if (connp->conn_recvslla && !(flags & IP_FF_SEND_SLLA))
7029                 in_flags |= IPF_RECVSLLA;
7030 
7031         /*
7032          * Initiate IPPF processing here, if needed. Note first_mp won't be
7033          * freed if the packet is dropped. The caller will do so.
7034          */
7035         if (IPP_ENABLED(IPP_LOCAL_IN, ipst) && ip_policy) {
7036                 ill_index = recv_ill->ill_phyint->phyint_ifindex;
7037                 ip_process(IPP_LOCAL_IN, &mp, ill_index);
7038                 if (mp == NULL) {
7039                         return;
7040                 }
7041         }
7042         if ((in_flags != 0) &&
7043             (mp->b_datap->db_type != M_CTL)) {
7044                 /*
7045                  * The actual data will be contained in b_cont
7046                  * upon successful return of the following call
7047                  * else original mblk is returned
7048                  */
7049                 ASSERT(recv_ill != NULL);
7050                 mp = ip_add_info(mp, recv_ill, in_flags, IPCL_ZONEID(connp),
7051                     ipst);
7052         }
7053         BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCInDelivers);
7054         /* Send it upstream */
7055         (connp->conn_recv)(connp, mp, NULL);
7056 }
7057 
7058 /*
7059  * Fanout for UDP packets.
7060  * The caller puts <fport, lport> in the ports parameter.
7061  *
7062  * If SO_REUSEADDR is set all multicast and broadcast packets
7063  * will be delivered to all streams bound to the same port.
7064  *
7065  * Zones notes:
7066  * Multicast and broadcast packets will be distributed to streams in all zones.
7067  * In the special case where an AF_INET socket binds to 0.0.0.0/<port> and an
7068  * AF_INET6 socket binds to ::/<port>, only the AF_INET socket receives the IPv4
7069  * packets. To maintain this behavior with multiple zones, the conns are grouped
7070  * by zone and the SO_REUSEADDR flag is checked for the first matching conn in
7071  * each zone. If unset, all the following conns in the same zone are skipped.
7072  */
7073 static void
7074 ip_fanout_udp(queue_t *q, mblk_t *mp, ill_t *ill, ipha_t *ipha,
7075     uint32_t ports, boolean_t broadcast, uint_t flags, boolean_t mctl_present,
7076     boolean_t ip_policy, ill_t *recv_ill, zoneid_t zoneid)
7077 {
7078         uint32_t        dstport, srcport;
7079         ipaddr_t        dst;
7080         mblk_t          *first_mp;
7081         boolean_t       secure;
7082         in6_addr_t      v6src;
7083         conn_t          *connp;
7084         connf_t         *connfp;
7085         conn_t          *first_connp;
7086         conn_t          *next_connp;
7087         mblk_t          *mp1, *first_mp1;
7088         ipaddr_t        src;
7089         zoneid_t        last_zoneid;
7090         boolean_t       reuseaddr;
7091         boolean_t       shared_addr;
7092         boolean_t       unlabeled;
7093         ip_stack_t      *ipst;
7094 
7095         ASSERT(recv_ill != NULL);
7096         ipst = recv_ill->ill_ipst;
7097 
7098         first_mp = mp;
7099         if (mctl_present) {
7100                 mp = first_mp->b_cont;
7101                 first_mp->b_cont = NULL;
7102                 secure = ipsec_in_is_secure(first_mp);
7103                 ASSERT(mp != NULL);
7104         } else {
7105                 first_mp = NULL;
7106                 secure = B_FALSE;
7107         }
7108 
7109         /* Extract ports in net byte order */
7110         dstport = htons(ntohl(ports) & 0xFFFF);
7111         srcport = htons(ntohl(ports) >> 16);
7112         dst = ipha->ipha_dst;
7113         src = ipha->ipha_src;
7114 
7115         unlabeled = B_FALSE;
7116         if (is_system_labeled())
7117                 /* Cred cannot be null on IPv4 */
7118                 unlabeled = (crgetlabel(DB_CRED(mp))->tsl_flags &
7119                     TSLF_UNLABELED) != 0;
7120         shared_addr = (zoneid == ALL_ZONES);
7121         if (shared_addr) {
7122                 /*
7123                  * No need to handle exclusive-stack zones since ALL_ZONES
7124                  * only applies to the shared stack.
7125                  */
7126                 zoneid = tsol_mlp_findzone(IPPROTO_UDP, dstport);
7127                 /*
7128                  * If no shared MLP is found, tsol_mlp_findzone returns
7129                  * ALL_ZONES.  In that case, we assume it's SLP, and
7130                  * search for the zone based on the packet label.
7131                  *
7132                  * If there is such a zone, we prefer to find a
7133                  * connection in it.  Otherwise, we look for a
7134                  * MAC-exempt connection in any zone whose label
7135                  * dominates the default label on the packet.
7136                  */
7137                 if (zoneid == ALL_ZONES)
7138                         zoneid = tsol_packet_to_zoneid(mp);
7139                 else
7140                         unlabeled = B_FALSE;
7141         }
7142 
7143         connfp = &ipst->ips_ipcl_udp_fanout[IPCL_UDP_HASH(dstport, ipst)];
7144         mutex_enter(&connfp->connf_lock);
7145         connp = connfp->connf_head;
7146         if (!broadcast && !CLASSD(dst)) {
7147                 /*
7148                  * Not broadcast or multicast. Send to the one (first)
7149                  * client we find. No need to check conn_wantpacket()
7150                  * since IP_BOUND_IF/conn_incoming_ill does not apply to
7151                  * IPv4 unicast packets.
7152                  */
7153                 while ((connp != NULL) &&
7154                     (!IPCL_UDP_MATCH(connp, dstport, dst, srcport, src) ||
7155                     (!IPCL_ZONE_MATCH(connp, zoneid) &&
7156                     !(unlabeled && connp->conn_mac_exempt)))) {
7157                         /*
7158                          * We keep searching since the conn did not match,
7159                          * or its zone did not match and it is not either
7160                          * an allzones conn or a mac exempt conn (if the
7161                          * sender is unlabeled.)
7162                          */
7163                         connp = connp->conn_next;
7164                 }
7165 
7166                 if (connp == NULL || connp->conn_upq == NULL)
7167                         goto notfound;
7168 
7169                 if (is_system_labeled() &&
7170                     !tsol_receive_local(mp, &dst, IPV4_VERSION, shared_addr,
7171                     connp))
7172                         goto notfound;
7173 
7174                 CONN_INC_REF(connp);
7175                 mutex_exit(&connfp->connf_lock);
7176                 ip_fanout_udp_conn(connp, first_mp, mp, secure, ill, ipha,
7177                     flags, recv_ill, ip_policy);
7178                 IP_STAT(ipst, ip_udp_fannorm);
7179                 CONN_DEC_REF(connp);
7180                 return;
7181         }
7182 
7183         /*
7184          * Broadcast and multicast case
7185          *
7186          * Need to check conn_wantpacket().
7187          * If SO_REUSEADDR has been set on the first we send the
7188          * packet to all clients that have joined the group and
7189          * match the port.
7190          */
7191 
7192         while (connp != NULL) {
7193                 if ((IPCL_UDP_MATCH(connp, dstport, dst, srcport, src)) &&
7194                     conn_wantpacket(connp, ill, ipha, flags, zoneid) &&
7195                     (!is_system_labeled() ||
7196                     tsol_receive_local(mp, &dst, IPV4_VERSION, shared_addr,
7197                     connp)))
7198                         break;
7199                 connp = connp->conn_next;
7200         }
7201 
7202         if (connp == NULL || connp->conn_upq == NULL)
7203                 goto notfound;
7204 
7205         first_connp = connp;
7206         /*
7207          * When SO_REUSEADDR is not set, send the packet only to the first
7208          * matching connection in its zone by keeping track of the zoneid.
7209          */
7210         reuseaddr = first_connp->conn_reuseaddr;
7211         last_zoneid = first_connp->conn_zoneid;
7212 
7213         CONN_INC_REF(connp);
7214         connp = connp->conn_next;
7215         for (;;) {
7216                 while (connp != NULL) {
7217                         if (IPCL_UDP_MATCH(connp, dstport, dst, srcport, src) &&
7218                             (reuseaddr || connp->conn_zoneid != last_zoneid) &&
7219                             conn_wantpacket(connp, ill, ipha, flags, zoneid) &&
7220                             (!is_system_labeled() ||
7221                             tsol_receive_local(mp, &dst, IPV4_VERSION,
7222                             shared_addr, connp)))
7223                                 break;
7224                         connp = connp->conn_next;
7225                 }
7226                 /*
7227                  * Just copy the data part alone. The mctl part is
7228                  * needed just for verifying policy and it is never
7229                  * sent up.
7230                  */
7231                 if (connp == NULL || (((mp1 = dupmsg(mp)) == NULL) &&
7232                     ((mp1 = copymsg(mp)) == NULL))) {
7233                         /*
7234                          * No more interested clients or memory
7235                          * allocation failed
7236                          */
7237                         connp = first_connp;
7238                         break;
7239                 }
7240                 if (connp->conn_zoneid != last_zoneid) {
7241                         /*
7242                          * Update the zoneid so that the packet isn't sent to
7243                          * any more conns in the same zone unless SO_REUSEADDR
7244                          * is set.
7245                          */
7246                         reuseaddr = connp->conn_reuseaddr;
7247                         last_zoneid = connp->conn_zoneid;
7248                 }
7249                 if (first_mp != NULL) {
7250                         ASSERT(((ipsec_info_t *)first_mp->b_rptr)->
7251                             ipsec_info_type == IPSEC_IN);
7252                         first_mp1 = ipsec_in_tag(first_mp, NULL,
7253                             ipst->ips_netstack);
7254                         if (first_mp1 == NULL) {
7255                                 freemsg(mp1);
7256                                 connp = first_connp;
7257                                 break;
7258                         }
7259                 } else {
7260                         first_mp1 = NULL;
7261                 }
7262                 CONN_INC_REF(connp);
7263                 mutex_exit(&connfp->connf_lock);
7264                 /*
7265                  * IPQoS notes: We don't send the packet for policy
7266                  * processing here, will do it for the last one (below).
7267                  * i.e. we do it per-packet now, but if we do policy
7268                  * processing per-conn, then we would need to do it
7269                  * here too.
7270                  */
7271                 ip_fanout_udp_conn(connp, first_mp1, mp1, secure, ill,
7272                     ipha, flags, recv_ill, B_FALSE);
7273                 mutex_enter(&connfp->connf_lock);
7274                 /* Follow the next pointer before releasing the conn. */
7275                 next_connp = connp->conn_next;
7276                 IP_STAT(ipst, ip_udp_fanmb);
7277                 CONN_DEC_REF(connp);
7278                 connp = next_connp;
7279         }
7280 
7281         /* Last one.  Send it upstream. */
7282         mutex_exit(&connfp->connf_lock);
7283         ip_fanout_udp_conn(connp, first_mp, mp, secure, ill, ipha, flags,
7284             recv_ill, ip_policy);
7285         IP_STAT(ipst, ip_udp_fanmb);
7286         CONN_DEC_REF(connp);
7287         return;
7288 
7289 notfound:
7290 
7291         mutex_exit(&connfp->connf_lock);
7292         IP_STAT(ipst, ip_udp_fanothers);
7293         /*
7294          * IPv6 endpoints bound to unicast or multicast IPv4-mapped addresses
7295          * have already been matched above, since they live in the IPv4
7296          * fanout tables. This implies we only need to
7297          * check for IPv6 in6addr_any endpoints here.
7298          * Thus we compare using ipv6_all_zeros instead of the destination
7299          * address, except for the multicast group membership lookup which
7300          * uses the IPv4 destination.
7301          */
7302         IN6_IPADDR_TO_V4MAPPED(ipha->ipha_src, &v6src);
7303         connfp = &ipst->ips_ipcl_udp_fanout[IPCL_UDP_HASH(dstport, ipst)];
7304         mutex_enter(&connfp->connf_lock);
7305         connp = connfp->connf_head;
7306         if (!broadcast && !CLASSD(dst)) {
7307                 while (connp != NULL) {
7308                         if (IPCL_UDP_MATCH_V6(connp, dstport, ipv6_all_zeros,
7309                             srcport, v6src) && IPCL_ZONE_MATCH(connp, zoneid) &&
7310                             conn_wantpacket(connp, ill, ipha, flags, zoneid) &&
7311                             !connp->conn_ipv6_v6only)
7312                                 break;
7313                         connp = connp->conn_next;
7314                 }
7315 
7316                 if (connp != NULL && is_system_labeled() &&
7317                     !tsol_receive_local(mp, &dst, IPV4_VERSION, shared_addr,
7318                     connp))
7319                         connp = NULL;
7320 
7321                 if (connp == NULL || connp->conn_upq == NULL) {
7322                         /*
7323                          * No one bound to this port.  Is
7324                          * there a client that wants all
7325                          * unclaimed datagrams?
7326                          */
7327                         mutex_exit(&connfp->connf_lock);
7328 
7329                         if (mctl_present)
7330                                 first_mp->b_cont = mp;
7331                         else
7332                                 first_mp = mp;
7333                         if (ipst->ips_ipcl_proto_fanout[IPPROTO_UDP].
7334                             connf_head != NULL) {
7335                                 ip_fanout_proto(q, first_mp, ill, ipha,
7336                                     flags | IP_FF_RAWIP, mctl_present,
7337                                     ip_policy, recv_ill, zoneid);
7338                         } else {
7339                                 if (ip_fanout_send_icmp(q, first_mp, flags,
7340                                     ICMP_DEST_UNREACHABLE,
7341                                     ICMP_PORT_UNREACHABLE,
7342                                     mctl_present, zoneid, ipst)) {
7343                                         BUMP_MIB(ill->ill_ip_mib,
7344                                             udpIfStatsNoPorts);
7345                                 }
7346                         }
7347                         return;
7348                 }
7349 
7350                 CONN_INC_REF(connp);
7351                 mutex_exit(&connfp->connf_lock);
7352                 ip_fanout_udp_conn(connp, first_mp, mp, secure, ill, ipha,
7353                     flags, recv_ill, ip_policy);
7354                 CONN_DEC_REF(connp);
7355                 return;
7356         }
7357         /*
7358          * IPv4 multicast packet being delivered to an AF_INET6
7359          * in6addr_any endpoint.
7360          * Need to check conn_wantpacket(). Note that we use conn_wantpacket()
7361          * and not conn_wantpacket_v6() since any multicast membership is
7362          * for an IPv4-mapped multicast address.
7363          * The packet is sent to all clients in all zones that have joined the
7364          * group and match the port.
7365          */
7366         while (connp != NULL) {
7367                 if (IPCL_UDP_MATCH_V6(connp, dstport, ipv6_all_zeros,
7368                     srcport, v6src) &&
7369                     conn_wantpacket(connp, ill, ipha, flags, zoneid) &&
7370                     (!is_system_labeled() ||
7371                     tsol_receive_local(mp, &dst, IPV4_VERSION, shared_addr,
7372                     connp)))
7373                         break;
7374                 connp = connp->conn_next;
7375         }
7376 
7377         if (connp == NULL || connp->conn_upq == NULL) {
7378                 /*
7379                  * No one bound to this port.  Is
7380                  * there a client that wants all
7381                  * unclaimed datagrams?
7382                  */
7383                 mutex_exit(&connfp->connf_lock);
7384 
7385                 if (mctl_present)
7386                         first_mp->b_cont = mp;
7387                 else
7388                         first_mp = mp;
7389                 if (ipst->ips_ipcl_proto_fanout[IPPROTO_UDP].connf_head !=
7390                     NULL) {
7391                         ip_fanout_proto(q, first_mp, ill, ipha,
7392                             flags | IP_FF_RAWIP, mctl_present, ip_policy,
7393                             recv_ill, zoneid);
7394                 } else {
7395                         /*
7396                          * We used to attempt to send an icmp error here, but
7397                          * since this is known to be a multicast packet
7398                          * and we don't send icmp errors in response to
7399                          * multicast, just drop the packet and give up sooner.
7400                          */
7401                         BUMP_MIB(ill->ill_ip_mib, udpIfStatsNoPorts);
7402                         freemsg(first_mp);
7403                 }
7404                 return;
7405         }
7406 
7407         first_connp = connp;
7408 
7409         CONN_INC_REF(connp);
7410         connp = connp->conn_next;
7411         for (;;) {
7412                 while (connp != NULL) {
7413                         if (IPCL_UDP_MATCH_V6(connp, dstport,
7414                             ipv6_all_zeros, srcport, v6src) &&
7415                             conn_wantpacket(connp, ill, ipha, flags, zoneid) &&
7416                             (!is_system_labeled() ||
7417                             tsol_receive_local(mp, &dst, IPV4_VERSION,
7418                             shared_addr, connp)))
7419                                 break;
7420                         connp = connp->conn_next;
7421                 }
7422                 /*
7423                  * Just copy the data part alone. The mctl part is
7424                  * needed just for verifying policy and it is never
7425                  * sent up.
7426                  */
7427                 if (connp == NULL || (((mp1 = dupmsg(mp)) == NULL) &&
7428                     ((mp1 = copymsg(mp)) == NULL))) {
7429                         /*
7430                          * No more intested clients or memory
7431                          * allocation failed
7432                          */
7433                         connp = first_connp;
7434                         break;
7435                 }
7436                 if (first_mp != NULL) {
7437                         ASSERT(((ipsec_info_t *)first_mp->b_rptr)->
7438                             ipsec_info_type == IPSEC_IN);
7439                         first_mp1 = ipsec_in_tag(first_mp, NULL,
7440                             ipst->ips_netstack);
7441                         if (first_mp1 == NULL) {
7442                                 freemsg(mp1);
7443                                 connp = first_connp;
7444                                 break;
7445                         }
7446                 } else {
7447                         first_mp1 = NULL;
7448                 }
7449                 CONN_INC_REF(connp);
7450                 mutex_exit(&connfp->connf_lock);
7451                 /*
7452                  * IPQoS notes: We don't send the packet for policy
7453                  * processing here, will do it for the last one (below).
7454                  * i.e. we do it per-packet now, but if we do policy
7455                  * processing per-conn, then we would need to do it
7456                  * here too.
7457                  */
7458                 ip_fanout_udp_conn(connp, first_mp1, mp1, secure, ill,
7459                     ipha, flags, recv_ill, B_FALSE);
7460                 mutex_enter(&connfp->connf_lock);
7461                 /* Follow the next pointer before releasing the conn. */
7462                 next_connp = connp->conn_next;
7463                 CONN_DEC_REF(connp);
7464                 connp = next_connp;
7465         }
7466 
7467         /* Last one.  Send it upstream. */
7468         mutex_exit(&connfp->connf_lock);
7469         ip_fanout_udp_conn(connp, first_mp, mp, secure, ill, ipha, flags,
7470             recv_ill, ip_policy);
7471         CONN_DEC_REF(connp);
7472 }
7473 
7474 /*
7475  * Complete the ip_wput header so that it
7476  * is possible to generate ICMP
7477  * errors.
7478  */
7479 int
7480 ip_hdr_complete(ipha_t *ipha, zoneid_t zoneid, ip_stack_t *ipst)
7481 {
7482         ire_t *ire;
7483 
7484         if (ipha->ipha_src == INADDR_ANY) {
7485                 ire = ire_lookup_local(zoneid, ipst);
7486                 if (ire == NULL) {
7487                         ip1dbg(("ip_hdr_complete: no source IRE\n"));
7488                         return (1);
7489                 }
7490                 ipha->ipha_src = ire->ire_addr;
7491                 ire_refrele(ire);
7492         }
7493         ipha->ipha_ttl = ipst->ips_ip_def_ttl;
7494         ipha->ipha_hdr_checksum = 0;
7495         ipha->ipha_hdr_checksum = ip_csum_hdr(ipha);
7496         return (0);
7497 }
7498 
7499 /*
7500  * Nobody should be sending
7501  * packets up this stream
7502  */
7503 static void
7504 ip_lrput(queue_t *q, mblk_t *mp)
7505 {
7506         mblk_t *mp1;
7507 
7508         switch (mp->b_datap->db_type) {
7509         case M_FLUSH:
7510                 /* Turn around */
7511                 if (*mp->b_rptr & FLUSHW) {
7512                         *mp->b_rptr &= ~FLUSHR;
7513                         qreply(q, mp);
7514                         return;
7515                 }
7516                 break;
7517         }
7518         /* Could receive messages that passed through ar_rput */
7519         for (mp1 = mp; mp1; mp1 = mp1->b_cont)
7520                 mp1->b_prev = mp1->b_next = NULL;
7521         freemsg(mp);
7522 }
7523 
7524 /* Nobody should be sending packets down this stream */
7525 /* ARGSUSED */
7526 void
7527 ip_lwput(queue_t *q, mblk_t *mp)
7528 {
7529         freemsg(mp);
7530 }
7531 
7532 /*
7533  * Move the first hop in any source route to ipha_dst and remove that part of
7534  * the source route.  Called by other protocols.  Errors in option formatting
7535  * are ignored - will be handled by ip_wput_options Return the final
7536  * destination (either ipha_dst or the last entry in a source route.)
7537  */
7538 ipaddr_t
7539 ip_massage_options(ipha_t *ipha, netstack_t *ns)
7540 {
7541         ipoptp_t        opts;
7542         uchar_t         *opt;
7543         uint8_t         optval;
7544         uint8_t         optlen;
7545         ipaddr_t        dst;
7546         int             i;
7547         ire_t           *ire;
7548         ip_stack_t      *ipst = ns->netstack_ip;
7549 
7550         ip2dbg(("ip_massage_options\n"));
7551         dst = ipha->ipha_dst;
7552         for (optval = ipoptp_first(&opts, ipha);
7553             optval != IPOPT_EOL;
7554             optval = ipoptp_next(&opts)) {
7555                 opt = opts.ipoptp_cur;
7556                 switch (optval) {
7557                         uint8_t off;
7558                 case IPOPT_SSRR:
7559                 case IPOPT_LSRR:
7560                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
7561                                 ip1dbg(("ip_massage_options: bad src route\n"));
7562                                 break;
7563                         }
7564                         optlen = opts.ipoptp_len;
7565                         off = opt[IPOPT_OFFSET];
7566                         off--;
7567                 redo_srr:
7568                         if (optlen < IP_ADDR_LEN ||
7569                             off > optlen - IP_ADDR_LEN) {
7570                                 /* End of source route */
7571                                 ip1dbg(("ip_massage_options: end of SR\n"));
7572                                 break;
7573                         }
7574                         bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
7575                         ip1dbg(("ip_massage_options: next hop 0x%x\n",
7576                             ntohl(dst)));
7577                         /*
7578                          * Check if our address is present more than
7579                          * once as consecutive hops in source route.
7580                          * XXX verify per-interface ip_forwarding
7581                          * for source route?
7582                          */
7583                         ire = ire_ctable_lookup(dst, 0, IRE_LOCAL, NULL,
7584                             ALL_ZONES, NULL, MATCH_IRE_TYPE, ipst);
7585                         if (ire != NULL) {
7586                                 ire_refrele(ire);
7587                                 off += IP_ADDR_LEN;
7588                                 goto redo_srr;
7589                         }
7590                         if (dst == htonl(INADDR_LOOPBACK)) {
7591                                 ip1dbg(("ip_massage_options: loopback addr in "
7592                                     "source route!\n"));
7593                                 break;
7594                         }
7595                         /*
7596                          * Update ipha_dst to be the first hop and remove the
7597                          * first hop from the source route (by overwriting
7598                          * part of the option with NOP options).
7599                          */
7600                         ipha->ipha_dst = dst;
7601                         /* Put the last entry in dst */
7602                         off = ((optlen - IP_ADDR_LEN - 3) & ~(IP_ADDR_LEN-1)) +
7603                             3;
7604                         bcopy(&opt[off], &dst, IP_ADDR_LEN);
7605 
7606                         ip1dbg(("ip_massage_options: last hop 0x%x\n",
7607                             ntohl(dst)));
7608                         /* Move down and overwrite */
7609                         opt[IP_ADDR_LEN] = opt[0];
7610                         opt[IP_ADDR_LEN+1] = opt[IPOPT_OLEN] - IP_ADDR_LEN;
7611                         opt[IP_ADDR_LEN+2] = opt[IPOPT_OFFSET];
7612                         for (i = 0; i < IP_ADDR_LEN; i++)
7613                                 opt[i] = IPOPT_NOP;
7614                         break;
7615                 }
7616         }
7617         return (dst);
7618 }
7619 
7620 /*
7621  * Return the network mask
7622  * associated with the specified address.
7623  */
7624 ipaddr_t
7625 ip_net_mask(ipaddr_t addr)
7626 {
7627         uchar_t *up = (uchar_t *)&addr;
7628         ipaddr_t mask = 0;
7629         uchar_t *maskp = (uchar_t *)&mask;
7630 
7631 #if defined(__i386) || defined(__amd64)
7632 #define TOTALLY_BRAIN_DAMAGED_C_COMPILER
7633 #endif
7634 #ifdef  TOTALLY_BRAIN_DAMAGED_C_COMPILER
7635         maskp[0] = maskp[1] = maskp[2] = maskp[3] = 0;
7636 #endif
7637         if (CLASSD(addr)) {
7638                 maskp[0] = 0xF0;
7639                 return (mask);
7640         }
7641 
7642         /* We assume Class E default netmask to be 32 */
7643         if (CLASSE(addr))
7644                 return (0xffffffffU);
7645 
7646         if (addr == 0)
7647                 return (0);
7648         maskp[0] = 0xFF;
7649         if ((up[0] & 0x80) == 0)
7650                 return (mask);
7651 
7652         maskp[1] = 0xFF;
7653         if ((up[0] & 0xC0) == 0x80)
7654                 return (mask);
7655 
7656         maskp[2] = 0xFF;
7657         if ((up[0] & 0xE0) == 0xC0)
7658                 return (mask);
7659 
7660         /* Otherwise return no mask */
7661         return ((ipaddr_t)0);
7662 }
7663 
7664 /*
7665  * Select an ill for the packet by considering load spreading across
7666  * a different ill in the group if dst_ill is part of some group.
7667  */
7668 ill_t *
7669 ip_newroute_get_dst_ill(ill_t *dst_ill)
7670 {
7671         ill_t *ill;
7672 
7673         /*
7674          * We schedule irrespective of whether the source address is
7675          * INADDR_ANY or not. illgrp_scheduler returns a held ill.
7676          */
7677         ill = illgrp_scheduler(dst_ill);
7678         if (ill == NULL)
7679                 return (NULL);
7680 
7681         /*
7682          * For groups with names ip_sioctl_groupname ensures that all
7683          * ills are of same type. For groups without names, ifgrp_insert
7684          * ensures this.
7685          */
7686         ASSERT(dst_ill->ill_type == ill->ill_type);
7687 
7688         return (ill);
7689 }
7690 
7691 /*
7692  * Helper function for the IPIF_NOFAILOVER/ATTACH_IF interface attachment case.
7693  */
7694 ill_t *
7695 ip_grab_attach_ill(ill_t *ill, mblk_t *first_mp, int ifindex, boolean_t isv6,
7696     ip_stack_t *ipst)
7697 {
7698         ill_t *ret_ill;
7699 
7700         ASSERT(ifindex != 0);
7701         ret_ill = ill_lookup_on_ifindex(ifindex, isv6, NULL, NULL, NULL, NULL,
7702             ipst);
7703         if (ret_ill == NULL ||
7704             (ret_ill->ill_phyint->phyint_flags & PHYI_OFFLINE)) {
7705                 if (isv6) {
7706                         if (ill != NULL) {
7707                                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
7708                         } else {
7709                                 BUMP_MIB(&ipst->ips_ip6_mib,
7710                                     ipIfStatsOutDiscards);
7711                         }
7712                         ip1dbg(("ip_grab_attach_ill (IPv6): "
7713                             "bad ifindex %d.\n", ifindex));
7714                 } else {
7715                         if (ill != NULL) {
7716                                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
7717                         } else {
7718                                 BUMP_MIB(&ipst->ips_ip_mib,
7719                                     ipIfStatsOutDiscards);
7720                         }
7721                         ip1dbg(("ip_grab_attach_ill (IPv4): "
7722                             "bad ifindex %d.\n", ifindex));
7723                 }
7724                 if (ret_ill != NULL)
7725                         ill_refrele(ret_ill);
7726                 freemsg(first_mp);
7727                 return (NULL);
7728         }
7729 
7730         return (ret_ill);
7731 }
7732 
7733 /*
7734  * IPv4 -
7735  * ip_newroute is called by ip_rput or ip_wput whenever we need to send
7736  * out a packet to a destination address for which we do not have specific
7737  * (or sufficient) routing information.
7738  *
7739  * NOTE : These are the scopes of some of the variables that point at IRE,
7740  *        which needs to be followed while making any future modifications
7741  *        to avoid memory leaks.
7742  *
7743  *      - ire and sire are the entries looked up initially by
7744  *        ire_ftable_lookup.
7745  *      - ipif_ire is used to hold the interface ire associated with
7746  *        the new cache ire. But it's scope is limited, so we always REFRELE
7747  *        it before branching out to error paths.
7748  *      - save_ire is initialized before ire_create, so that ire returned
7749  *        by ire_create will not over-write the ire. We REFRELE save_ire
7750  *        before breaking out of the switch.
7751  *
7752  *      Thus on failures, we have to REFRELE only ire and sire, if they
7753  *      are not NULL.
7754  */
7755 void
7756 ip_newroute(queue_t *q, mblk_t *mp, ipaddr_t dst, conn_t *connp,
7757     zoneid_t zoneid, ip_stack_t *ipst)
7758 {
7759         areq_t  *areq;
7760         ipaddr_t gw = 0;
7761         ire_t   *ire = NULL;
7762         mblk_t  *res_mp;
7763         ipaddr_t *addrp;
7764         ipaddr_t nexthop_addr;
7765         ipif_t  *src_ipif = NULL;
7766         ill_t   *dst_ill = NULL;
7767         ipha_t  *ipha;
7768         ire_t   *sire = NULL;
7769         mblk_t  *first_mp;
7770         ire_t   *save_ire;
7771         ill_t   *attach_ill = NULL;     /* Bind to IPIF_NOFAILOVER address */
7772         ushort_t ire_marks = 0;
7773         boolean_t mctl_present;
7774         ipsec_out_t *io;
7775         mblk_t  *saved_mp;
7776         ire_t   *first_sire = NULL;
7777         mblk_t  *copy_mp = NULL;
7778         mblk_t  *xmit_mp = NULL;
7779         ipaddr_t save_dst;
7780         uint32_t multirt_flags =
7781             MULTIRT_CACHEGW | MULTIRT_USESTAMP | MULTIRT_SETSTAMP;
7782         boolean_t multirt_is_resolvable;
7783         boolean_t multirt_resolve_next;
7784         boolean_t unspec_src;
7785         boolean_t do_attach_ill = B_FALSE;
7786         boolean_t ip_nexthop = B_FALSE;
7787         tsol_ire_gw_secattr_t *attrp = NULL;
7788         tsol_gcgrp_t *gcgrp = NULL;
7789         tsol_gcgrp_addr_t ga;
7790 
7791         if (ip_debug > 2) {
7792                 /* ip1dbg */
7793                 pr_addr_dbg("ip_newroute: dst %s\n", AF_INET, &dst);
7794         }
7795 
7796         EXTRACT_PKT_MP(mp, first_mp, mctl_present);
7797         if (mctl_present) {
7798                 io = (ipsec_out_t *)first_mp->b_rptr;
7799                 ASSERT(io->ipsec_out_type == IPSEC_OUT);
7800                 ASSERT(zoneid == io->ipsec_out_zoneid);
7801                 ASSERT(zoneid != ALL_ZONES);
7802         }
7803 
7804         ipha = (ipha_t *)mp->b_rptr;
7805 
7806         /* All multicast lookups come through ip_newroute_ipif() */
7807         if (CLASSD(dst)) {
7808                 ip0dbg(("ip_newroute: CLASSD 0x%x (b_prev %p, b_next %p)\n",
7809                     ntohl(dst), (void *)mp->b_prev, (void *)mp->b_next));
7810                 freemsg(first_mp);
7811                 return;
7812         }
7813 
7814         if (mctl_present && io->ipsec_out_attach_if) {
7815                 /* ip_grab_attach_ill returns a held ill */
7816                 attach_ill = ip_grab_attach_ill(NULL, first_mp,
7817                     io->ipsec_out_ill_index, B_FALSE, ipst);
7818 
7819                 /* Failure case frees things for us. */
7820                 if (attach_ill == NULL)
7821                         return;
7822 
7823                 /*
7824                  * Check if we need an ire that will not be
7825                  * looked up by anybody else i.e. HIDDEN.
7826                  */
7827                 if (ill_is_probeonly(attach_ill))
7828                         ire_marks = IRE_MARK_HIDDEN;
7829         }
7830         if (mctl_present && io->ipsec_out_ip_nexthop) {
7831                 ip_nexthop = B_TRUE;
7832                 nexthop_addr = io->ipsec_out_nexthop_addr;
7833         }
7834         /*
7835          * If this IRE is created for forwarding or it is not for
7836          * traffic for congestion controlled protocols, mark it as temporary.
7837          */
7838         if (mp->b_prev != NULL || !IP_FLOW_CONTROLLED_ULP(ipha->ipha_protocol))
7839                 ire_marks |= IRE_MARK_TEMPORARY;
7840 
7841         /*
7842          * Get what we can from ire_ftable_lookup which will follow an IRE
7843          * chain until it gets the most specific information available.
7844          * For example, we know that there is no IRE_CACHE for this dest,
7845          * but there may be an IRE_OFFSUBNET which specifies a gateway.
7846          * ire_ftable_lookup will look up the gateway, etc.
7847          * Otherwise, given ire_ftable_lookup algorithm, only one among routes
7848          * to the destination, of equal netmask length in the forward table,
7849          * will be recursively explored. If no information is available
7850          * for the final gateway of that route, we force the returned ire
7851          * to be equal to sire using MATCH_IRE_PARENT.
7852          * At least, in this case we have a starting point (in the buckets)
7853          * to look for other routes to the destination in the forward table.
7854          * This is actually used only for multirouting, where a list
7855          * of routes has to be processed in sequence.
7856          *
7857          * In the process of coming up with the most specific information,
7858          * ire_ftable_lookup may end up with an incomplete IRE_CACHE entry
7859          * for the gateway (i.e., one for which the ire_nce->nce_state is
7860          * not yet ND_REACHABLE, and is in the middle of arp resolution).
7861          * Two caveats when handling incomplete ire's in ip_newroute:
7862          * - we should be careful when accessing its ire_nce (specifically
7863          *   the nce_res_mp) ast it might change underneath our feet, and,
7864          * - not all legacy code path callers are prepared to handle
7865          *   incomplete ire's, so we should not create/add incomplete
7866          *   ire_cache entries here. (See discussion about temporary solution
7867          *   further below).
7868          *
7869          * In order to minimize packet dropping, and to preserve existing
7870          * behavior, we treat this case as if there were no IRE_CACHE for the
7871          * gateway, and instead use the IF_RESOLVER ire to send out
7872          * another request to ARP (this is achieved by passing the
7873          * MATCH_IRE_COMPLETE flag to ire_ftable_lookup). When the
7874          * arp response comes back in ip_wput_nondata, we will create
7875          * a per-dst ire_cache that has an ND_COMPLETE ire.
7876          *
7877          * Note that this is a temporary solution; the correct solution is
7878          * to create an incomplete  per-dst ire_cache entry, and send the
7879          * packet out when the gw's nce is resolved. In order to achieve this,
7880          * all packet processing must have been completed prior to calling
7881          * ire_add_then_send. Some legacy code paths (e.g. cgtp) would need
7882          * to be modified to accomodate this solution.
7883          */
7884         if (ip_nexthop) {
7885                 /*
7886                  * The first time we come here, we look for an IRE_INTERFACE
7887                  * entry for the specified nexthop, set the dst to be the
7888                  * nexthop address and create an IRE_CACHE entry for the
7889                  * nexthop. The next time around, we are able to find an
7890                  * IRE_CACHE entry for the nexthop, set the gateway to be the
7891                  * nexthop address and create an IRE_CACHE entry for the
7892                  * destination address via the specified nexthop.
7893                  */
7894                 ire = ire_cache_lookup(nexthop_addr, zoneid,
7895                     MBLK_GETLABEL(mp), ipst);
7896                 if (ire != NULL) {
7897                         gw = nexthop_addr;
7898                         ire_marks |= IRE_MARK_PRIVATE_ADDR;
7899                 } else {
7900                         ire = ire_ftable_lookup(nexthop_addr, 0, 0,
7901                             IRE_INTERFACE, NULL, NULL, zoneid, 0,
7902                             MBLK_GETLABEL(mp),
7903                             MATCH_IRE_TYPE | MATCH_IRE_SECATTR,
7904                             ipst);
7905                         if (ire != NULL) {
7906                                 dst = nexthop_addr;
7907                         }
7908                 }
7909         } else if (attach_ill == NULL) {
7910                 ire = ire_ftable_lookup(dst, 0, 0, 0,
7911                     NULL, &sire, zoneid, 0, MBLK_GETLABEL(mp),
7912                     MATCH_IRE_RECURSIVE | MATCH_IRE_DEFAULT |
7913                     MATCH_IRE_RJ_BHOLE | MATCH_IRE_PARENT |
7914                     MATCH_IRE_SECATTR | MATCH_IRE_COMPLETE,
7915                     ipst);
7916         } else {
7917                 /*
7918                  * attach_ill is set only for communicating with
7919                  * on-link hosts. So, don't look for DEFAULT.
7920                  */
7921                 ipif_t  *attach_ipif;
7922 
7923                 attach_ipif = ipif_get_next_ipif(NULL, attach_ill);
7924                 if (attach_ipif == NULL) {
7925                         ill_refrele(attach_ill);
7926                         goto icmp_err_ret;
7927                 }
7928                 ire = ire_ftable_lookup(dst, 0, 0, 0, attach_ipif,
7929                     &sire, zoneid, 0, MBLK_GETLABEL(mp),
7930                     MATCH_IRE_RJ_BHOLE | MATCH_IRE_ILL |
7931                     MATCH_IRE_SECATTR, ipst);
7932                 ipif_refrele(attach_ipif);
7933         }
7934         ip3dbg(("ip_newroute: ire_ftable_lookup() "
7935             "returned ire %p, sire %p\n", (void *)ire, (void *)sire));
7936 
7937         /*
7938          * This loop is run only once in most cases.
7939          * We loop to resolve further routes only when the destination
7940          * can be reached through multiple RTF_MULTIRT-flagged ires.
7941          */
7942         do {
7943                 /* Clear the previous iteration's values */
7944                 if (src_ipif != NULL) {
7945                         ipif_refrele(src_ipif);
7946                         src_ipif = NULL;
7947                 }
7948                 if (dst_ill != NULL) {
7949                         ill_refrele(dst_ill);
7950                         dst_ill = NULL;
7951                 }
7952 
7953                 multirt_resolve_next = B_FALSE;
7954                 /*
7955                  * We check if packets have to be multirouted.
7956                  * In this case, given the current <ire, sire> couple,
7957                  * we look for the next suitable <ire, sire>.
7958                  * This check is done in ire_multirt_lookup(),
7959                  * which applies various criteria to find the next route
7960                  * to resolve. ire_multirt_lookup() leaves <ire, sire>
7961                  * unchanged if it detects it has not been tried yet.
7962                  */
7963                 if ((sire != NULL) && (sire->ire_flags & RTF_MULTIRT)) {
7964                         ip3dbg(("ip_newroute: starting next_resolution "
7965                             "with first_mp %p, tag %d\n",
7966                             (void *)first_mp,
7967                             MULTIRT_DEBUG_TAGGED(first_mp)));
7968 
7969                         ASSERT(sire != NULL);
7970                         multirt_is_resolvable =
7971                             ire_multirt_lookup(&ire, &sire, multirt_flags,
7972                             MBLK_GETLABEL(mp), ipst);
7973 
7974                         ip3dbg(("ip_newroute: multirt_is_resolvable %d, "
7975                             "ire %p, sire %p\n",
7976                             multirt_is_resolvable,
7977                             (void *)ire, (void *)sire));
7978 
7979                         if (!multirt_is_resolvable) {
7980                                 /*
7981                                  * No more multirt route to resolve; give up
7982                                  * (all routes resolved or no more
7983                                  * resolvable routes).
7984                                  */
7985                                 if (ire != NULL) {
7986                                         ire_refrele(ire);
7987                                         ire = NULL;
7988                                 }
7989                         } else {
7990                                 ASSERT(sire != NULL);
7991                                 ASSERT(ire != NULL);
7992                                 /*
7993                                  * We simply use first_sire as a flag that
7994                                  * indicates if a resolvable multirt route
7995                                  * has already been found.
7996                                  * If it is not the case, we may have to send
7997                                  * an ICMP error to report that the
7998                                  * destination is unreachable.
7999                                  * We do not IRE_REFHOLD first_sire.
8000                                  */
8001                                 if (first_sire == NULL) {
8002                                         first_sire = sire;
8003                                 }
8004                         }
8005                 }
8006                 if (ire == NULL) {
8007                         if (ip_debug > 3) {
8008                                 /* ip2dbg */
8009                                 pr_addr_dbg("ip_newroute: "
8010                                     "can't resolve %s\n", AF_INET, &dst);
8011                         }
8012                         ip3dbg(("ip_newroute: "
8013                             "ire %p, sire %p, first_sire %p\n",
8014                             (void *)ire, (void *)sire, (void *)first_sire));
8015 
8016                         if (sire != NULL) {
8017                                 ire_refrele(sire);
8018                                 sire = NULL;
8019                         }
8020 
8021                         if (first_sire != NULL) {
8022                                 /*
8023                                  * At least one multirt route has been found
8024                                  * in the same call to ip_newroute();
8025                                  * there is no need to report an ICMP error.
8026                                  * first_sire was not IRE_REFHOLDed.
8027                                  */
8028                                 MULTIRT_DEBUG_UNTAG(first_mp);
8029                                 freemsg(first_mp);
8030                                 return;
8031                         }
8032                         ip_rts_change(RTM_MISS, dst, 0, 0, 0, 0, 0, 0,
8033                             RTA_DST, ipst);
8034                         if (attach_ill != NULL)
8035                                 ill_refrele(attach_ill);
8036                         goto icmp_err_ret;
8037                 }
8038 
8039                 /*
8040                  * Verify that the returned IRE does not have either
8041                  * the RTF_REJECT or RTF_BLACKHOLE flags set and that the IRE is
8042                  * either an IRE_CACHE, IRE_IF_NORESOLVER or IRE_IF_RESOLVER.
8043                  */
8044                 if ((ire->ire_flags & (RTF_REJECT | RTF_BLACKHOLE)) ||
8045                     (ire->ire_type & (IRE_CACHE | IRE_INTERFACE)) == 0) {
8046                         if (attach_ill != NULL)
8047                                 ill_refrele(attach_ill);
8048                         goto icmp_err_ret;
8049                 }
8050                 /*
8051                  * Increment the ire_ob_pkt_count field for ire if it is an
8052                  * INTERFACE (IF_RESOLVER or IF_NORESOLVER) IRE type, and
8053                  * increment the same for the parent IRE, sire, if it is some
8054                  * sort of prefix IRE (which includes DEFAULT, PREFIX, and HOST)
8055                  */
8056                 if ((ire->ire_type & IRE_INTERFACE) != 0) {
8057                         UPDATE_OB_PKT_COUNT(ire);
8058                         ire->ire_last_used_time = lbolt;
8059                 }
8060 
8061                 if (sire != NULL) {
8062                         gw = sire->ire_gateway_addr;
8063                         ASSERT((sire->ire_type & (IRE_CACHETABLE |
8064                             IRE_INTERFACE)) == 0);
8065                         UPDATE_OB_PKT_COUNT(sire);
8066                         sire->ire_last_used_time = lbolt;
8067                 }
8068                 /*
8069                  * We have a route to reach the destination.
8070                  *
8071                  * 1) If the interface is part of ill group, try to get a new
8072                  *    ill taking load spreading into account.
8073                  *
8074                  * 2) After selecting the ill, get a source address that
8075                  *    might create good inbound load spreading.
8076                  *    ipif_select_source does this for us.
8077                  *
8078                  * If the application specified the ill (ifindex), we still
8079                  * load spread. Only if the packets needs to go out
8080                  * specifically on a given ill e.g. binding to
8081                  * IPIF_NOFAILOVER address, then we don't try to use a
8082                  * different ill for load spreading.
8083                  */
8084                 if (attach_ill == NULL) {
8085                         /*
8086                          * Don't perform outbound load spreading in the
8087                          * case of an RTF_MULTIRT route, as we actually
8088                          * typically want to replicate outgoing packets
8089                          * through particular interfaces.
8090                          */
8091                         if ((sire != NULL) && (sire->ire_flags & RTF_MULTIRT)) {
8092                                 dst_ill = ire->ire_ipif->ipif_ill;
8093                                 /* for uniformity */
8094                                 ill_refhold(dst_ill);
8095                         } else {
8096                                 /*
8097                                  * If we are here trying to create an IRE_CACHE
8098                                  * for an offlink destination and have the
8099                                  * IRE_CACHE for the next hop and the latter is
8100                                  * using virtual IP source address selection i.e
8101                                  * it's ire->ire_ipif is pointing to a virtual
8102                                  * network interface (vni) then
8103                                  * ip_newroute_get_dst_ll() will return the vni
8104                                  * interface as the dst_ill. Since the vni is
8105                                  * virtual i.e not associated with any physical
8106                                  * interface, it cannot be the dst_ill, hence
8107                                  * in such a case call ip_newroute_get_dst_ll()
8108                                  * with the stq_ill instead of the ire_ipif ILL.
8109                                  * The function returns a refheld ill.
8110                                  */
8111                                 if ((ire->ire_type == IRE_CACHE) &&
8112                                     IS_VNI(ire->ire_ipif->ipif_ill))
8113                                         dst_ill = ip_newroute_get_dst_ill(
8114                                             ire->ire_stq->q_ptr);
8115                                 else
8116                                         dst_ill = ip_newroute_get_dst_ill(
8117                                             ire->ire_ipif->ipif_ill);
8118                         }
8119                         if (dst_ill == NULL) {
8120                                 if (ip_debug > 2) {
8121                                         pr_addr_dbg("ip_newroute: "
8122                                             "no dst ill for dst"
8123                                             " %s\n", AF_INET, &dst);
8124                                 }
8125                                 goto icmp_err_ret;
8126                         }
8127                 } else {
8128                         dst_ill = ire->ire_ipif->ipif_ill;
8129                         /* for uniformity */
8130                         ill_refhold(dst_ill);
8131                         /*
8132                          * We should have found a route matching ill as we
8133                          * called ire_ftable_lookup with MATCH_IRE_ILL.
8134                          * Rather than asserting, when there is a mismatch,
8135                          * we just drop the packet.
8136                          */
8137                         if (dst_ill != attach_ill) {
8138                                 ip0dbg(("ip_newroute: Packet dropped as "
8139                                     "IPIF_NOFAILOVER ill is %s, "
8140                                     "ire->ire_ipif->ipif_ill is %s\n",
8141                                     attach_ill->ill_name,
8142                                     dst_ill->ill_name));
8143                                 ill_refrele(attach_ill);
8144                                 goto icmp_err_ret;
8145                         }
8146                 }
8147                 /* attach_ill can't go in loop. IPMP and CGTP are disjoint */
8148                 if (attach_ill != NULL) {
8149                         ill_refrele(attach_ill);
8150                         attach_ill = NULL;
8151                         do_attach_ill = B_TRUE;
8152                 }
8153                 ASSERT(dst_ill != NULL);
8154                 ip2dbg(("ip_newroute: dst_ill %s\n", dst_ill->ill_name));
8155 
8156                 /*
8157                  * Pick the best source address from dst_ill.
8158                  *
8159                  * 1) If it is part of a multipathing group, we would
8160                  *    like to spread the inbound packets across different
8161                  *    interfaces. ipif_select_source picks a random source
8162                  *    across the different ills in the group.
8163                  *
8164                  * 2) If it is not part of a multipathing group, we try
8165                  *    to pick the source address from the destination
8166                  *    route. Clustering assumes that when we have multiple
8167                  *    prefixes hosted on an interface, the prefix of the
8168                  *    source address matches the prefix of the destination
8169                  *    route. We do this only if the address is not
8170                  *    DEPRECATED.
8171                  *
8172                  * 3) If the conn is in a different zone than the ire, we
8173                  *    need to pick a source address from the right zone.
8174                  *
8175                  * NOTE : If we hit case (1) above, the prefix of the source
8176                  *        address picked may not match the prefix of the
8177                  *        destination routes prefix as ipif_select_source
8178                  *        does not look at "dst" while picking a source
8179                  *        address.
8180                  *        If we want the same behavior as (2), we will need
8181                  *        to change the behavior of ipif_select_source.
8182                  */
8183                 ASSERT(src_ipif == NULL);
8184                 if ((sire != NULL) && (sire->ire_flags & RTF_SETSRC)) {
8185                         /*
8186                          * The RTF_SETSRC flag is set in the parent ire (sire).
8187                          * Check that the ipif matching the requested source
8188                          * address still exists.
8189                          */
8190                         src_ipif = ipif_lookup_addr(sire->ire_src_addr, NULL,
8191                             zoneid, NULL, NULL, NULL, NULL, ipst);
8192                 }
8193 
8194                 unspec_src = (connp != NULL && connp->conn_unspec_src);
8195 
8196                 if (src_ipif == NULL &&
8197                     (!unspec_src || ipha->ipha_src != INADDR_ANY)) {
8198                         ire_marks |= IRE_MARK_USESRC_CHECK;
8199                         if ((dst_ill->ill_group != NULL) ||
8200                             (ire->ire_ipif->ipif_flags & IPIF_DEPRECATED) ||
8201                             (connp != NULL && ire->ire_zoneid != zoneid &&
8202                             ire->ire_zoneid != ALL_ZONES) ||
8203                             (dst_ill->ill_usesrc_ifindex != 0)) {
8204                                 /*
8205                                  * If the destination is reachable via a
8206                                  * given gateway, the selected source address
8207                                  * should be in the same subnet as the gateway.
8208                                  * Otherwise, the destination is not reachable.
8209                                  *
8210                                  * If there are no interfaces on the same subnet
8211                                  * as the destination, ipif_select_source gives
8212                                  * first non-deprecated interface which might be
8213                                  * on a different subnet than the gateway.
8214                                  * This is not desirable. Hence pass the dst_ire
8215                                  * source address to ipif_select_source.
8216                                  * It is sure that the destination is reachable
8217                                  * with the dst_ire source address subnet.
8218                                  * So passing dst_ire source address to
8219                                  * ipif_select_source will make sure that the
8220                                  * selected source will be on the same subnet
8221                                  * as dst_ire source address.
8222                                  */
8223                                 ipaddr_t saddr = ire->ire_ipif->ipif_src_addr;
8224                                 src_ipif = ipif_select_source(dst_ill, saddr,
8225                                     zoneid);
8226                                 if (src_ipif == NULL) {
8227                                         if (ip_debug > 2) {
8228                                                 pr_addr_dbg("ip_newroute: "
8229                                                     "no src for dst %s ",
8230                                                     AF_INET, &dst);
8231                                                 printf("through interface %s\n",
8232                                                     dst_ill->ill_name);
8233                                         }
8234                                         goto icmp_err_ret;
8235                                 }
8236                         } else {
8237                                 src_ipif = ire->ire_ipif;
8238                                 ASSERT(src_ipif != NULL);
8239                                 /* hold src_ipif for uniformity */
8240                                 ipif_refhold(src_ipif);
8241                         }
8242                 }
8243 
8244                 /*
8245                  * Assign a source address while we have the conn.
8246                  * We can't have ip_wput_ire pick a source address when the
8247                  * packet returns from arp since we need to look at
8248                  * conn_unspec_src and conn_zoneid, and we lose the conn when
8249                  * going through arp.
8250                  *
8251                  * NOTE : ip_newroute_v6 does not have this piece of code as
8252                  *        it uses ip6i to store this information.
8253                  */
8254                 if (ipha->ipha_src == INADDR_ANY && !unspec_src)
8255                         ipha->ipha_src = src_ipif->ipif_src_addr;
8256 
8257                 if (ip_debug > 3) {
8258                         /* ip2dbg */
8259                         pr_addr_dbg("ip_newroute: first hop %s\n",
8260                             AF_INET, &gw);
8261                 }
8262                 ip2dbg(("\tire type %s (%d)\n",
8263                     ip_nv_lookup(ire_nv_tbl, ire->ire_type), ire->ire_type));
8264 
8265                 /*
8266                  * The TTL of multirouted packets is bounded by the
8267                  * ip_multirt_ttl ndd variable.
8268                  */
8269                 if ((sire != NULL) && (sire->ire_flags & RTF_MULTIRT)) {
8270                         /* Force TTL of multirouted packets */
8271                         if ((ipst->ips_ip_multirt_ttl > 0) &&
8272                             (ipha->ipha_ttl > ipst->ips_ip_multirt_ttl)) {
8273                                 ip2dbg(("ip_newroute: forcing multirt TTL "
8274                                     "to %d (was %d), dst 0x%08x\n",
8275                                     ipst->ips_ip_multirt_ttl, ipha->ipha_ttl,
8276                                     ntohl(sire->ire_addr)));
8277                                 ipha->ipha_ttl = ipst->ips_ip_multirt_ttl;
8278                         }
8279                 }
8280                 /*
8281                  * At this point in ip_newroute(), ire is either the
8282                  * IRE_CACHE of the next-hop gateway for an off-subnet
8283                  * destination or an IRE_INTERFACE type that should be used
8284                  * to resolve an on-subnet destination or an on-subnet
8285                  * next-hop gateway.
8286                  *
8287                  * In the IRE_CACHE case, we have the following :
8288                  *
8289                  * 1) src_ipif - used for getting a source address.
8290                  *
8291                  * 2) dst_ill - from which we derive ire_stq/ire_rfq. This
8292                  *    means packets using this IRE_CACHE will go out on
8293                  *    dst_ill.
8294                  *
8295                  * 3) The IRE sire will point to the prefix that is the
8296                  *    longest  matching route for the destination. These
8297                  *    prefix types include IRE_DEFAULT, IRE_PREFIX, IRE_HOST.
8298                  *
8299                  *    The newly created IRE_CACHE entry for the off-subnet
8300                  *    destination is tied to both the prefix route and the
8301                  *    interface route used to resolve the next-hop gateway
8302                  *    via the ire_phandle and ire_ihandle fields,
8303                  *    respectively.
8304                  *
8305                  * In the IRE_INTERFACE case, we have the following :
8306                  *
8307                  * 1) src_ipif - used for getting a source address.
8308                  *
8309                  * 2) dst_ill - from which we derive ire_stq/ire_rfq. This
8310                  *    means packets using the IRE_CACHE that we will build
8311                  *    here will go out on dst_ill.
8312                  *
8313                  * 3) sire may or may not be NULL. But, the IRE_CACHE that is
8314                  *    to be created will only be tied to the IRE_INTERFACE
8315                  *    that was derived from the ire_ihandle field.
8316                  *
8317                  *    If sire is non-NULL, it means the destination is
8318                  *    off-link and we will first create the IRE_CACHE for the
8319                  *    gateway. Next time through ip_newroute, we will create
8320                  *    the IRE_CACHE for the final destination as described
8321                  *    above.
8322                  *
8323                  * In both cases, after the current resolution has been
8324                  * completed (or possibly initialised, in the IRE_INTERFACE
8325                  * case), the loop may be re-entered to attempt the resolution
8326                  * of another RTF_MULTIRT route.
8327                  *
8328                  * When an IRE_CACHE entry for the off-subnet destination is
8329                  * created, RTF_SETSRC and RTF_MULTIRT are inherited from sire,
8330                  * for further processing in emission loops.
8331                  */
8332                 save_ire = ire;
8333                 switch (ire->ire_type) {
8334                 case IRE_CACHE: {
8335                         ire_t   *ipif_ire;
8336 
8337                         ASSERT(save_ire->ire_nce->nce_state == ND_REACHABLE);
8338                         if (gw == 0)
8339                                 gw = ire->ire_gateway_addr;
8340                         /*
8341                          * We need 3 ire's to create a new cache ire for an
8342                          * off-link destination from the cache ire of the
8343                          * gateway.
8344                          *
8345                          *      1. The prefix ire 'sire' (Note that this does
8346                          *         not apply to the conn_nexthop_set case)
8347                          *      2. The cache ire of the gateway 'ire'
8348                          *      3. The interface ire 'ipif_ire'
8349                          *
8350                          * We have (1) and (2). We lookup (3) below.
8351                          *
8352                          * If there is no interface route to the gateway,
8353                          * it is a race condition, where we found the cache
8354                          * but the interface route has been deleted.
8355                          */
8356                         if (ip_nexthop) {
8357                                 ipif_ire = ire_ihandle_lookup_onlink(ire);
8358                         } else {
8359                                 ipif_ire =
8360                                     ire_ihandle_lookup_offlink(ire, sire);
8361                         }
8362                         if (ipif_ire == NULL) {
8363                                 ip1dbg(("ip_newroute: "
8364                                     "ire_ihandle_lookup_offlink failed\n"));
8365                                 goto icmp_err_ret;
8366                         }
8367 
8368                         /*
8369                          * Check cached gateway IRE for any security
8370                          * attributes; if found, associate the gateway
8371                          * credentials group to the destination IRE.
8372                          */
8373                         if ((attrp = save_ire->ire_gw_secattr) != NULL) {
8374                                 mutex_enter(&attrp->igsa_lock);
8375                                 if ((gcgrp = attrp->igsa_gcgrp) != NULL)
8376                                         GCGRP_REFHOLD(gcgrp);
8377                                 mutex_exit(&attrp->igsa_lock);
8378                         }
8379 
8380                         /*
8381                          * XXX For the source of the resolver mp,
8382                          * we are using the same DL_UNITDATA_REQ
8383                          * (from save_ire->ire_nce->nce_res_mp)
8384                          * though the save_ire is not pointing at the same ill.
8385                          * This is incorrect. We need to send it up to the
8386                          * resolver to get the right res_mp. For ethernets
8387                          * this may be okay (ill_type == DL_ETHER).
8388                          */
8389 
8390                         ire = ire_create(
8391                             (uchar_t *)&dst,                /* dest address */
8392                             (uchar_t *)&ip_g_all_ones,      /* mask */
8393                             (uchar_t *)&src_ipif->ipif_src_addr, /* src addr */
8394                             (uchar_t *)&gw,         /* gateway address */
8395                             &save_ire->ire_max_frag,
8396                             save_ire->ire_nce,               /* src nce */
8397                             dst_ill->ill_rq,         /* recv-from queue */
8398                             dst_ill->ill_wq,         /* send-to queue */
8399                             IRE_CACHE,                  /* IRE type */
8400                             src_ipif,
8401                             (sire != NULL) ?
8402                             sire->ire_mask : 0,      /* Parent mask */
8403                             (sire != NULL) ?
8404                             sire->ire_phandle : 0,   /* Parent handle */
8405                             ipif_ire->ire_ihandle,   /* Interface handle */
8406                             (sire != NULL) ? (sire->ire_flags &
8407                             (RTF_SETSRC | RTF_MULTIRT)) : 0, /* flags */
8408                             (sire != NULL) ?
8409                             &(sire->ire_uinfo) : &(save_ire->ire_uinfo),
8410                             NULL,
8411                             gcgrp,
8412                             ipst);
8413 
8414                         if (ire == NULL) {
8415                                 if (gcgrp != NULL) {
8416                                         GCGRP_REFRELE(gcgrp);
8417                                         gcgrp = NULL;
8418                                 }
8419                                 ire_refrele(ipif_ire);
8420                                 ire_refrele(save_ire);
8421                                 break;
8422                         }
8423 
8424                         /* reference now held by IRE */
8425                         gcgrp = NULL;
8426 
8427                         ire->ire_marks |= ire_marks;
8428 
8429                         /*
8430                          * Prevent sire and ipif_ire from getting deleted.
8431                          * The newly created ire is tied to both of them via
8432                          * the phandle and ihandle respectively.
8433                          */
8434                         if (sire != NULL) {
8435                                 IRB_REFHOLD(sire->ire_bucket);
8436                                 /* Has it been removed already ? */
8437                                 if (sire->ire_marks & IRE_MARK_CONDEMNED) {
8438                                         IRB_REFRELE(sire->ire_bucket);
8439                                         ire_refrele(ipif_ire);
8440                                         ire_refrele(save_ire);
8441                                         break;
8442                                 }
8443                         }
8444 
8445                         IRB_REFHOLD(ipif_ire->ire_bucket);
8446                         /* Has it been removed already ? */
8447                         if (ipif_ire->ire_marks & IRE_MARK_CONDEMNED) {
8448                                 IRB_REFRELE(ipif_ire->ire_bucket);
8449                                 if (sire != NULL)
8450                                         IRB_REFRELE(sire->ire_bucket);
8451                                 ire_refrele(ipif_ire);
8452                                 ire_refrele(save_ire);
8453                                 break;
8454                         }
8455 
8456                         xmit_mp = first_mp;
8457                         /*
8458                          * In the case of multirouting, a copy
8459                          * of the packet is done before its sending.
8460                          * The copy is used to attempt another
8461                          * route resolution, in a next loop.
8462                          */
8463                         if (ire->ire_flags & RTF_MULTIRT) {
8464                                 copy_mp = copymsg(first_mp);
8465                                 if (copy_mp != NULL) {
8466                                         xmit_mp = copy_mp;
8467                                         MULTIRT_DEBUG_TAG(first_mp);
8468                                 }
8469                         }
8470                         ire_add_then_send(q, ire, xmit_mp);
8471                         ire_refrele(save_ire);
8472 
8473                         /* Assert that sire is not deleted yet. */
8474                         if (sire != NULL) {
8475                                 ASSERT(sire->ire_ptpn != NULL);
8476                                 IRB_REFRELE(sire->ire_bucket);
8477                         }
8478 
8479                         /* Assert that ipif_ire is not deleted yet. */
8480                         ASSERT(ipif_ire->ire_ptpn != NULL);
8481                         IRB_REFRELE(ipif_ire->ire_bucket);
8482                         ire_refrele(ipif_ire);
8483 
8484                         /*
8485                          * If copy_mp is not NULL, multirouting was
8486                          * requested. We loop to initiate a next
8487                          * route resolution attempt, starting from sire.
8488                          */
8489                         if (copy_mp != NULL) {
8490                                 /*
8491                                  * Search for the next unresolved
8492                                  * multirt route.
8493                                  */
8494                                 copy_mp = NULL;
8495                                 ipif_ire = NULL;
8496                                 ire = NULL;
8497                                 multirt_resolve_next = B_TRUE;
8498                                 continue;
8499                         }
8500                         if (sire != NULL)
8501                                 ire_refrele(sire);
8502                         ipif_refrele(src_ipif);
8503                         ill_refrele(dst_ill);
8504                         return;
8505                 }
8506                 case IRE_IF_NORESOLVER: {
8507 
8508                         if (dst_ill->ill_phys_addr_length != IP_ADDR_LEN &&
8509                             dst_ill->ill_resolver_mp == NULL) {
8510                                 ip1dbg(("ip_newroute: dst_ill %p "
8511                                     "for IRE_IF_NORESOLVER ire %p has "
8512                                     "no ill_resolver_mp\n",
8513                                     (void *)dst_ill, (void *)ire));
8514                                 break;
8515                         }
8516 
8517                         /*
8518                          * TSol note: We are creating the ire cache for the
8519                          * destination 'dst'. If 'dst' is offlink, going
8520                          * through the first hop 'gw', the security attributes
8521                          * of 'dst' must be set to point to the gateway
8522                          * credentials of gateway 'gw'. If 'dst' is onlink, it
8523                          * is possible that 'dst' is a potential gateway that is
8524                          * referenced by some route that has some security
8525                          * attributes. Thus in the former case, we need to do a
8526                          * gcgrp_lookup of 'gw' while in the latter case we
8527                          * need to do gcgrp_lookup of 'dst' itself.
8528                          */
8529                         ga.ga_af = AF_INET;
8530                         IN6_IPADDR_TO_V4MAPPED(gw != INADDR_ANY ? gw : dst,
8531                             &ga.ga_addr);
8532                         gcgrp = gcgrp_lookup(&ga, B_FALSE);
8533 
8534                         ire = ire_create(
8535                             (uchar_t *)&dst,                /* dest address */
8536                             (uchar_t *)&ip_g_all_ones,      /* mask */
8537                             (uchar_t *)&src_ipif->ipif_src_addr, /* src addr */
8538                             (uchar_t *)&gw,         /* gateway address */
8539                             &save_ire->ire_max_frag,
8540                             NULL,                       /* no src nce */
8541                             dst_ill->ill_rq,         /* recv-from queue */
8542                             dst_ill->ill_wq,         /* send-to queue */
8543                             IRE_CACHE,
8544                             src_ipif,
8545                             save_ire->ire_mask,              /* Parent mask */
8546                             (sire != NULL) ?            /* Parent handle */
8547                             sire->ire_phandle : 0,
8548                             save_ire->ire_ihandle,   /* Interface handle */
8549                             (sire != NULL) ? sire->ire_flags &
8550                             (RTF_SETSRC | RTF_MULTIRT) : 0, /* flags */
8551                             &(save_ire->ire_uinfo),
8552                             NULL,
8553                             gcgrp,
8554                             ipst);
8555 
8556                         if (ire == NULL) {
8557                                 if (gcgrp != NULL) {
8558                                         GCGRP_REFRELE(gcgrp);
8559                                         gcgrp = NULL;
8560                                 }
8561                                 ire_refrele(save_ire);
8562                                 break;
8563                         }
8564 
8565                         /* reference now held by IRE */
8566                         gcgrp = NULL;
8567 
8568                         ire->ire_marks |= ire_marks;
8569 
8570                         /* Prevent save_ire from getting deleted */
8571                         IRB_REFHOLD(save_ire->ire_bucket);
8572                         /* Has it been removed already ? */
8573                         if (save_ire->ire_marks & IRE_MARK_CONDEMNED) {
8574                                 IRB_REFRELE(save_ire->ire_bucket);
8575                                 ire_refrele(save_ire);
8576                                 break;
8577                         }
8578 
8579                         /*
8580                          * In the case of multirouting, a copy
8581                          * of the packet is made before it is sent.
8582                          * The copy is used in the next
8583                          * loop to attempt another resolution.
8584                          */
8585                         xmit_mp = first_mp;
8586                         if ((sire != NULL) &&
8587                             (sire->ire_flags & RTF_MULTIRT)) {
8588                                 copy_mp = copymsg(first_mp);
8589                                 if (copy_mp != NULL) {
8590                                         xmit_mp = copy_mp;
8591                                         MULTIRT_DEBUG_TAG(first_mp);
8592                                 }
8593                         }
8594                         ire_add_then_send(q, ire, xmit_mp);
8595 
8596                         /* Assert that it is not deleted yet. */
8597                         ASSERT(save_ire->ire_ptpn != NULL);
8598                         IRB_REFRELE(save_ire->ire_bucket);
8599                         ire_refrele(save_ire);
8600 
8601                         if (copy_mp != NULL) {
8602                                 /*
8603                                  * If we found a (no)resolver, we ignore any
8604                                  * trailing top priority IRE_CACHE in further
8605                                  * loops. This ensures that we do not omit any
8606                                  * (no)resolver.
8607                                  * This IRE_CACHE, if any, will be processed
8608                                  * by another thread entering ip_newroute().
8609                                  * IRE_CACHE entries, if any, will be processed
8610                                  * by another thread entering ip_newroute(),
8611                                  * (upon resolver response, for instance).
8612                                  * This aims to force parallel multirt
8613                                  * resolutions as soon as a packet must be sent.
8614                                  * In the best case, after the tx of only one
8615                                  * packet, all reachable routes are resolved.
8616                                  * Otherwise, the resolution of all RTF_MULTIRT
8617                                  * routes would require several emissions.
8618                                  */
8619                                 multirt_flags &= ~MULTIRT_CACHEGW;
8620 
8621                                 /*
8622                                  * Search for the next unresolved multirt
8623                                  * route.
8624                                  */
8625                                 copy_mp = NULL;
8626                                 save_ire = NULL;
8627                                 ire = NULL;
8628                                 multirt_resolve_next = B_TRUE;
8629                                 continue;
8630                         }
8631 
8632                         /*
8633                          * Don't need sire anymore
8634                          */
8635                         if (sire != NULL)
8636                                 ire_refrele(sire);
8637 
8638                         ipif_refrele(src_ipif);
8639                         ill_refrele(dst_ill);
8640                         return;
8641                 }
8642                 case IRE_IF_RESOLVER:
8643                         /*
8644                          * We can't build an IRE_CACHE yet, but at least we
8645                          * found a resolver that can help.
8646                          */
8647                         res_mp = dst_ill->ill_resolver_mp;
8648                         if (!OK_RESOLVER_MP(res_mp))
8649                                 break;
8650 
8651                         /*
8652                          * To be at this point in the code with a non-zero gw
8653                          * means that dst is reachable through a gateway that
8654                          * we have never resolved.  By changing dst to the gw
8655                          * addr we resolve the gateway first.
8656                          * When ire_add_then_send() tries to put the IP dg
8657                          * to dst, it will reenter ip_newroute() at which
8658                          * time we will find the IRE_CACHE for the gw and
8659                          * create another IRE_CACHE in case IRE_CACHE above.
8660                          */
8661                         if (gw != INADDR_ANY) {
8662                                 /*
8663                                  * The source ipif that was determined above was
8664                                  * relative to the destination address, not the
8665                                  * gateway's. If src_ipif was not taken out of
8666                                  * the IRE_IF_RESOLVER entry, we'll need to call
8667                                  * ipif_select_source() again.
8668                                  */
8669                                 if (src_ipif != ire->ire_ipif) {
8670                                         ipif_refrele(src_ipif);
8671                                         src_ipif = ipif_select_source(dst_ill,
8672                                             gw, zoneid);
8673                                         if (src_ipif == NULL) {
8674                                                 if (ip_debug > 2) {
8675                                                         pr_addr_dbg(
8676                                                             "ip_newroute: no "
8677                                                             "src for gw %s ",
8678                                                             AF_INET, &gw);
8679                                                         printf("through "
8680                                                             "interface %s\n",
8681                                                             dst_ill->ill_name);
8682                                                 }
8683                                                 goto icmp_err_ret;
8684                                         }
8685                                 }
8686                                 save_dst = dst;
8687                                 dst = gw;
8688                                 gw = INADDR_ANY;
8689                         }
8690 
8691                         /*
8692                          * We obtain a partial IRE_CACHE which we will pass
8693                          * along with the resolver query.  When the response
8694                          * comes back it will be there ready for us to add.
8695                          * The ire_max_frag is atomically set under the
8696                          * irebucket lock in ire_add_v[46].
8697                          */
8698 
8699                         ire = ire_create_mp(
8700                             (uchar_t *)&dst,                /* dest address */
8701                             (uchar_t *)&ip_g_all_ones,      /* mask */
8702                             (uchar_t *)&src_ipif->ipif_src_addr, /* src addr */
8703                             (uchar_t *)&gw,         /* gateway address */
8704                             NULL,                       /* ire_max_frag */
8705                             NULL,                       /* no src nce */
8706                             dst_ill->ill_rq,         /* recv-from queue */
8707                             dst_ill->ill_wq,         /* send-to queue */
8708                             IRE_CACHE,
8709                             src_ipif,                   /* Interface ipif */
8710                             save_ire->ire_mask,              /* Parent mask */
8711                             0,
8712                             save_ire->ire_ihandle,   /* Interface handle */
8713                             0,                          /* flags if any */
8714                             &(save_ire->ire_uinfo),
8715                             NULL,
8716                             NULL,
8717                             ipst);
8718 
8719                         if (ire == NULL) {
8720                                 ire_refrele(save_ire);
8721                                 break;
8722                         }
8723 
8724                         if ((sire != NULL) &&
8725                             (sire->ire_flags & RTF_MULTIRT)) {
8726                                 copy_mp = copymsg(first_mp);
8727                                 if (copy_mp != NULL)
8728                                         MULTIRT_DEBUG_TAG(copy_mp);
8729                         }
8730 
8731                         ire->ire_marks |= ire_marks;
8732 
8733                         /*
8734                          * Construct message chain for the resolver
8735                          * of the form:
8736                          *      ARP_REQ_MBLK-->IRE_MBLK-->Packet
8737                          * Packet could contain a IPSEC_OUT mp.
8738                          *
8739                          * NOTE : ire will be added later when the response
8740                          * comes back from ARP. If the response does not
8741                          * come back, ARP frees the packet. For this reason,
8742                          * we can't REFHOLD the bucket of save_ire to prevent
8743                          * deletions. We may not be able to REFRELE the bucket
8744                          * if the response never comes back. Thus, before
8745                          * adding the ire, ire_add_v4 will make sure that the
8746                          * interface route does not get deleted. This is the
8747                          * only case unlike ip_newroute_v6, ip_newroute_ipif_v6
8748                          * where we can always prevent deletions because of
8749                          * the synchronous nature of adding IRES i.e
8750                          * ire_add_then_send is called after creating the IRE.
8751                          */
8752                         ASSERT(ire->ire_mp != NULL);
8753                         ire->ire_mp->b_cont = first_mp;
8754                         /* Have saved_mp handy, for cleanup if canput fails */
8755                         saved_mp = mp;
8756                         mp = copyb(res_mp);
8757                         if (mp == NULL) {
8758                                 /* Prepare for cleanup */
8759                                 mp = saved_mp; /* pkt */
8760                                 ire_delete(ire); /* ire_mp */
8761                                 ire = NULL;
8762                                 ire_refrele(save_ire);
8763                                 if (copy_mp != NULL) {
8764                                         MULTIRT_DEBUG_UNTAG(copy_mp);
8765                                         freemsg(copy_mp);
8766                                         copy_mp = NULL;
8767                                 }
8768                                 break;
8769                         }
8770                         linkb(mp, ire->ire_mp);
8771 
8772                         /*
8773                          * Fill in the source and dest addrs for the resolver.
8774                          * NOTE: this depends on memory layouts imposed by
8775                          * ill_init().
8776                          */
8777                         areq = (areq_t *)mp->b_rptr;
8778                         addrp = (ipaddr_t *)((char *)areq +
8779                             areq->areq_sender_addr_offset);
8780                         if (do_attach_ill) {
8781                                 /*
8782                                  * This is bind to no failover case.
8783                                  * arp packet also must go out on attach_ill.
8784                                  */
8785                                 ASSERT(ipha->ipha_src != NULL);
8786                                 *addrp = ipha->ipha_src;
8787                         } else {
8788                                 *addrp = save_ire->ire_src_addr;
8789                         }
8790 
8791                         ire_refrele(save_ire);
8792                         addrp = (ipaddr_t *)((char *)areq +
8793                             areq->areq_target_addr_offset);
8794                         *addrp = dst;
8795                         /* Up to the resolver. */
8796                         if (canputnext(dst_ill->ill_rq) &&
8797                             !(dst_ill->ill_arp_closing)) {
8798                                 putnext(dst_ill->ill_rq, mp);
8799                                 ire = NULL;
8800                                 if (copy_mp != NULL) {
8801                                         /*
8802                                          * If we found a resolver, we ignore
8803                                          * any trailing top priority IRE_CACHE
8804                                          * in the further loops. This ensures
8805                                          * that we do not omit any resolver.
8806                                          * IRE_CACHE entries, if any, will be
8807                                          * processed next time we enter
8808                                          * ip_newroute().
8809                                          */
8810                                         multirt_flags &= ~MULTIRT_CACHEGW;
8811                                         /*
8812                                          * Search for the next unresolved
8813                                          * multirt route.
8814                                          */
8815                                         first_mp = copy_mp;
8816                                         copy_mp = NULL;
8817                                         /* Prepare the next resolution loop. */
8818                                         mp = first_mp;
8819                                         EXTRACT_PKT_MP(mp, first_mp,
8820                                             mctl_present);
8821                                         if (mctl_present)
8822                                                 io = (ipsec_out_t *)
8823                                                     first_mp->b_rptr;
8824                                         ipha = (ipha_t *)mp->b_rptr;
8825 
8826                                         ASSERT(sire != NULL);
8827 
8828                                         dst = save_dst;
8829                                         multirt_resolve_next = B_TRUE;
8830                                         continue;
8831                                 }
8832 
8833                                 if (sire != NULL)
8834                                         ire_refrele(sire);
8835 
8836                                 /*
8837                                  * The response will come back in ip_wput
8838                                  * with db_type IRE_DB_TYPE.
8839                                  */
8840                                 ipif_refrele(src_ipif);
8841                                 ill_refrele(dst_ill);
8842                                 return;
8843                         } else {
8844                                 /* Prepare for cleanup */
8845                                 DTRACE_PROBE1(ip__newroute__drop, mblk_t *,
8846                                     mp);
8847                                 mp->b_cont = NULL;
8848                                 freeb(mp); /* areq */
8849                                 /*
8850                                  * this is an ire that is not added to the
8851                                  * cache. ire_freemblk will handle the release
8852                                  * of any resources associated with the ire.
8853                                  */
8854                                 ire_delete(ire); /* ire_mp */
8855                                 mp = saved_mp; /* pkt */
8856                                 ire = NULL;
8857                                 if (copy_mp != NULL) {
8858                                         MULTIRT_DEBUG_UNTAG(copy_mp);
8859                                         freemsg(copy_mp);
8860                                         copy_mp = NULL;
8861                                 }
8862                                 break;
8863                         }
8864                 default:
8865                         break;
8866                 }
8867         } while (multirt_resolve_next);
8868 
8869         ip1dbg(("ip_newroute: dropped\n"));
8870         /* Did this packet originate externally? */
8871         if (mp->b_prev) {
8872                 mp->b_next = NULL;
8873                 mp->b_prev = NULL;
8874                 BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsInDiscards);
8875         } else {
8876                 if (dst_ill != NULL) {
8877                         BUMP_MIB(dst_ill->ill_ip_mib, ipIfStatsOutDiscards);
8878                 } else {
8879                         BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsOutDiscards);
8880                 }
8881         }
8882         ASSERT(copy_mp == NULL);
8883         MULTIRT_DEBUG_UNTAG(first_mp);
8884         freemsg(first_mp);
8885         if (ire != NULL)
8886                 ire_refrele(ire);
8887         if (sire != NULL)
8888                 ire_refrele(sire);
8889         if (src_ipif != NULL)
8890                 ipif_refrele(src_ipif);
8891         if (dst_ill != NULL)
8892                 ill_refrele(dst_ill);
8893         return;
8894 
8895 icmp_err_ret:
8896         ip1dbg(("ip_newroute: no route\n"));
8897         if (src_ipif != NULL)
8898                 ipif_refrele(src_ipif);
8899         if (dst_ill != NULL)
8900                 ill_refrele(dst_ill);
8901         if (sire != NULL)
8902                 ire_refrele(sire);
8903         /* Did this packet originate externally? */
8904         if (mp->b_prev) {
8905                 mp->b_next = NULL;
8906                 mp->b_prev = NULL;
8907                 BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsInNoRoutes);
8908                 q = WR(q);
8909         } else {
8910                 /*
8911                  * There is no outgoing ill, so just increment the
8912                  * system MIB.
8913                  */
8914                 BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsOutNoRoutes);
8915                 /*
8916                  * Since ip_wput() isn't close to finished, we fill
8917                  * in enough of the header for credible error reporting.
8918                  */
8919                 if (ip_hdr_complete(ipha, zoneid, ipst)) {
8920                         /* Failed */
8921                         MULTIRT_DEBUG_UNTAG(first_mp);
8922                         freemsg(first_mp);
8923                         if (ire != NULL)
8924                                 ire_refrele(ire);
8925                         return;
8926                 }
8927         }
8928 
8929         /*
8930          * At this point we will have ire only if RTF_BLACKHOLE
8931          * or RTF_REJECT flags are set on the IRE. It will not
8932          * generate ICMP_HOST_UNREACHABLE if RTF_BLACKHOLE is set.
8933          */
8934         if (ire != NULL) {
8935                 if (ire->ire_flags & RTF_BLACKHOLE) {
8936                         ire_refrele(ire);
8937                         MULTIRT_DEBUG_UNTAG(first_mp);
8938                         freemsg(first_mp);
8939                         return;
8940                 }
8941                 ire_refrele(ire);
8942         }
8943         if (ip_source_routed(ipha, ipst)) {
8944                 icmp_unreachable(q, first_mp, ICMP_SOURCE_ROUTE_FAILED,
8945                     zoneid, ipst);
8946                 return;
8947         }
8948         icmp_unreachable(q, first_mp, ICMP_HOST_UNREACHABLE, zoneid, ipst);
8949 }
8950 
8951 ip_opt_info_t zero_info;
8952 
8953 /*
8954  * IPv4 -
8955  * ip_newroute_ipif is called by ip_wput_multicast and
8956  * ip_rput_forward_multicast whenever we need to send
8957  * out a packet to a destination address for which we do not have specific
8958  * routing information. It is used when the packet will be sent out
8959  * on a specific interface. It is also called by ip_wput() when IP_BOUND_IF
8960  * socket option is set or icmp error message wants to go out on a particular
8961  * interface for a unicast packet.
8962  *
8963  * In most cases, the destination address is resolved thanks to the ipif
8964  * intrinsic resolver. However, there are some cases where the call to
8965  * ip_newroute_ipif must take into account the potential presence of
8966  * RTF_SETSRC and/or RTF_MULITRT flags in an IRE_OFFSUBNET ire
8967  * that uses the interface. This is specified through flags,
8968  * which can be a combination of:
8969  * - RTF_SETSRC: if an IRE_OFFSUBNET ire exists that has the RTF_SETSRC
8970  *   flag, the resulting ire will inherit the IRE_OFFSUBNET source address
8971  *   and flags. Additionally, the packet source address has to be set to
8972  *   the specified address. The caller is thus expected to set this flag
8973  *   if the packet has no specific source address yet.
8974  * - RTF_MULTIRT: if an IRE_OFFSUBNET ire exists that has the RTF_MULTIRT
8975  *   flag, the resulting ire will inherit the flag. All unresolved routes
8976  *   to the destination must be explored in the same call to
8977  *   ip_newroute_ipif().
8978  */
8979 static void
8980 ip_newroute_ipif(queue_t *q, mblk_t *mp, ipif_t *ipif, ipaddr_t dst,
8981     conn_t *connp, uint32_t flags, zoneid_t zoneid, ip_opt_info_t *infop)
8982 {
8983         areq_t  *areq;
8984         ire_t   *ire = NULL;
8985         mblk_t  *res_mp;
8986         ipaddr_t *addrp;
8987         mblk_t *first_mp;
8988         ire_t   *save_ire = NULL;
8989         ill_t   *attach_ill = NULL;             /* Bind to IPIF_NOFAILOVER */
8990         ipif_t  *src_ipif = NULL;
8991         ushort_t ire_marks = 0;
8992         ill_t   *dst_ill = NULL;
8993         boolean_t mctl_present;
8994         ipsec_out_t *io;
8995         ipha_t *ipha;
8996         int     ihandle = 0;
8997         mblk_t  *saved_mp;
8998         ire_t   *fire = NULL;
8999         mblk_t  *copy_mp = NULL;
9000         boolean_t multirt_resolve_next;
9001         boolean_t unspec_src;
9002         ipaddr_t ipha_dst;
9003         ip_stack_t *ipst = ipif->ipif_ill->ill_ipst;
9004 
9005         /*
9006          * CGTP goes in a loop which looks up a new ipif, do an ipif_refhold
9007          * here for uniformity
9008          */
9009         ipif_refhold(ipif);
9010 
9011         /*
9012          * This loop is run only once in most cases.
9013          * We loop to resolve further routes only when the destination
9014          * can be reached through multiple RTF_MULTIRT-flagged ires.
9015          */
9016         do {
9017                 if (dst_ill != NULL) {
9018                         ill_refrele(dst_ill);
9019                         dst_ill = NULL;
9020                 }
9021                 if (src_ipif != NULL) {
9022                         ipif_refrele(src_ipif);
9023                         src_ipif = NULL;
9024                 }
9025                 multirt_resolve_next = B_FALSE;
9026 
9027                 ip1dbg(("ip_newroute_ipif: dst 0x%x, if %s\n", ntohl(dst),
9028                     ipif->ipif_ill->ill_name));
9029 
9030                 EXTRACT_PKT_MP(mp, first_mp, mctl_present);
9031                 if (mctl_present)
9032                         io = (ipsec_out_t *)first_mp->b_rptr;
9033 
9034                 ipha = (ipha_t *)mp->b_rptr;
9035 
9036                 /*
9037                  * Save the packet destination address, we may need it after
9038                  * the packet has been consumed.
9039                  */
9040                 ipha_dst = ipha->ipha_dst;
9041 
9042                 /*
9043                  * If the interface is a pt-pt interface we look for an
9044                  * IRE_IF_RESOLVER or IRE_IF_NORESOLVER that matches both the
9045                  * local_address and the pt-pt destination address. Otherwise
9046                  * we just match the local address.
9047                  * NOTE: dst could be different than ipha->ipha_dst in case
9048                  * of sending igmp multicast packets over a point-to-point
9049                  * connection.
9050                  * Thus we must be careful enough to check ipha_dst to be a
9051                  * multicast address, otherwise it will take xmit_if path for
9052                  * multicast packets resulting into kernel stack overflow by
9053                  * repeated calls to ip_newroute_ipif from ire_send().
9054                  */
9055                 if (CLASSD(ipha_dst) &&
9056                     !(ipif->ipif_ill->ill_flags & ILLF_MULTICAST)) {
9057                         goto err_ret;
9058                 }
9059 
9060                 /*
9061                  * We check if an IRE_OFFSUBNET for the addr that goes through
9062                  * ipif exists. We need it to determine if the RTF_SETSRC and/or
9063                  * RTF_MULTIRT flags must be honored. This IRE_OFFSUBNET ire may
9064                  * propagate its flags to the new ire.
9065                  */
9066                 if (CLASSD(ipha_dst) && (flags & (RTF_MULTIRT | RTF_SETSRC))) {
9067                         fire = ipif_lookup_multi_ire(ipif, ipha_dst);
9068                         ip2dbg(("ip_newroute_ipif: "
9069                             "ipif_lookup_multi_ire("
9070                             "ipif %p, dst %08x) = fire %p\n",
9071                             (void *)ipif, ntohl(dst), (void *)fire));
9072                 }
9073 
9074                 if (mctl_present && io->ipsec_out_attach_if) {
9075                         attach_ill = ip_grab_attach_ill(NULL, first_mp,
9076                             io->ipsec_out_ill_index, B_FALSE, ipst);
9077 
9078                         /* Failure case frees things for us. */
9079                         if (attach_ill == NULL) {
9080                                 ipif_refrele(ipif);
9081                                 if (fire != NULL)
9082                                         ire_refrele(fire);
9083                                 return;
9084                         }
9085 
9086                         /*
9087                          * Check if we need an ire that will not be
9088                          * looked up by anybody else i.e. HIDDEN.
9089                          */
9090                         if (ill_is_probeonly(attach_ill)) {
9091                                 ire_marks = IRE_MARK_HIDDEN;
9092                         }
9093                         /*
9094                          * ip_wput passes the right ipif for IPIF_NOFAILOVER
9095                          * case.
9096                          */
9097                         dst_ill = ipif->ipif_ill;
9098                         /* attach_ill has been refheld by ip_grab_attach_ill */
9099                         ASSERT(dst_ill == attach_ill);
9100                 } else {
9101                         /*
9102                          * If the interface belongs to an interface group,
9103                          * make sure the next possible interface in the group
9104                          * is used.  This encourages load spreading among
9105                          * peers in an interface group.
9106                          * Note: load spreading is disabled for RTF_MULTIRT
9107                          * routes.
9108                          */
9109                         if ((flags & RTF_MULTIRT) && (fire != NULL) &&
9110                             (fire->ire_flags & RTF_MULTIRT)) {
9111                                 /*
9112                                  * Don't perform outbound load spreading
9113                                  * in the case of an RTF_MULTIRT issued route,
9114                                  * we actually typically want to replicate
9115                                  * outgoing packets through particular
9116                                  * interfaces.
9117                                  */
9118                                 dst_ill = ipif->ipif_ill;
9119                                 ill_refhold(dst_ill);
9120                         } else {
9121                                 dst_ill = ip_newroute_get_dst_ill(
9122                                     ipif->ipif_ill);
9123                         }
9124                         if (dst_ill == NULL) {
9125                                 if (ip_debug > 2) {
9126                                         pr_addr_dbg("ip_newroute_ipif: "
9127                                             "no dst ill for dst %s\n",
9128                                             AF_INET, &dst);
9129                                 }
9130                                 goto err_ret;
9131                         }
9132                 }
9133 
9134                 /*
9135                  * Pick a source address preferring non-deprecated ones.
9136                  * Unlike ip_newroute, we don't do any source address
9137                  * selection here since for multicast it really does not help
9138                  * in inbound load spreading as in the unicast case.
9139                  */
9140                 if ((flags & RTF_SETSRC) && (fire != NULL) &&
9141                     (fire->ire_flags & RTF_SETSRC)) {
9142                         /*
9143                          * As requested by flags, an IRE_OFFSUBNET was looked up
9144                          * on that interface. This ire has RTF_SETSRC flag, so
9145                          * the source address of the packet must be changed.
9146                          * Check that the ipif matching the requested source
9147                          * address still exists.
9148                          */
9149                         src_ipif = ipif_lookup_addr(fire->ire_src_addr, NULL,
9150                             zoneid, NULL, NULL, NULL, NULL, ipst);
9151                 }
9152 
9153                 unspec_src = (connp != NULL && connp->conn_unspec_src);
9154 
9155                 if (((!ipif->ipif_isv6 && ipif->ipif_lcl_addr == INADDR_ANY) ||
9156                     (ipif->ipif_flags & (IPIF_DEPRECATED|IPIF_UP)) != IPIF_UP ||
9157                     (connp != NULL && ipif->ipif_zoneid != zoneid &&
9158                     ipif->ipif_zoneid != ALL_ZONES)) &&
9159                     (src_ipif == NULL) &&
9160                     (!unspec_src || ipha->ipha_src != INADDR_ANY)) {
9161                         src_ipif = ipif_select_source(dst_ill, dst, zoneid);
9162                         if (src_ipif == NULL) {
9163                                 if (ip_debug > 2) {
9164                                         /* ip1dbg */
9165                                         pr_addr_dbg("ip_newroute_ipif: "
9166                                             "no src for dst %s",
9167                                             AF_INET, &dst);
9168                                 }
9169                                 ip1dbg((" through interface %s\n",
9170                                     dst_ill->ill_name));
9171                                 goto err_ret;
9172                         }
9173                         ipif_refrele(ipif);
9174                         ipif = src_ipif;
9175                         ipif_refhold(ipif);
9176                 }
9177                 if (src_ipif == NULL) {
9178                         src_ipif = ipif;
9179                         ipif_refhold(src_ipif);
9180                 }
9181 
9182                 /*
9183                  * Assign a source address while we have the conn.
9184                  * We can't have ip_wput_ire pick a source address when the
9185                  * packet returns from arp since conn_unspec_src might be set
9186                  * and we lose the conn when going through arp.
9187                  */
9188                 if (ipha->ipha_src == INADDR_ANY && !unspec_src)
9189                         ipha->ipha_src = src_ipif->ipif_src_addr;
9190 
9191                 /*
9192                  * In the case of IP_BOUND_IF and IP_PKTINFO, it is possible
9193                  * that the outgoing interface does not have an interface ire.
9194                  */
9195                 if (CLASSD(ipha_dst) && (connp == NULL ||
9196                     connp->conn_outgoing_ill == NULL) &&
9197                     infop->ip_opt_ill_index == 0) {
9198                         /* ipif_to_ire returns an held ire */
9199                         ire = ipif_to_ire(ipif);
9200                         if (ire == NULL)
9201                                 goto err_ret;
9202                         if (ire->ire_flags & (RTF_REJECT | RTF_BLACKHOLE))
9203                                 goto err_ret;
9204                         /*
9205                          * ihandle is needed when the ire is added to
9206                          * cache table.
9207                          */
9208                         save_ire = ire;
9209                         ihandle = save_ire->ire_ihandle;
9210 
9211                         ip2dbg(("ip_newroute_ipif: ire %p, ipif %p, "
9212                             "flags %04x\n",
9213                             (void *)ire, (void *)ipif, flags));
9214                         if ((flags & RTF_MULTIRT) && (fire != NULL) &&
9215                             (fire->ire_flags & RTF_MULTIRT)) {
9216                                 /*
9217                                  * As requested by flags, an IRE_OFFSUBNET was
9218                                  * looked up on that interface. This ire has
9219                                  * RTF_MULTIRT flag, so the resolution loop will
9220                                  * be re-entered to resolve additional routes on
9221                                  * other interfaces. For that purpose, a copy of
9222                                  * the packet is performed at this point.
9223                                  */
9224                                 fire->ire_last_used_time = lbolt;
9225                                 copy_mp = copymsg(first_mp);
9226                                 if (copy_mp) {
9227                                         MULTIRT_DEBUG_TAG(copy_mp);
9228                                 }
9229                         }
9230                         if ((flags & RTF_SETSRC) && (fire != NULL) &&
9231                             (fire->ire_flags & RTF_SETSRC)) {
9232                                 /*
9233                                  * As requested by flags, an IRE_OFFSUBET was
9234                                  * looked up on that interface. This ire has
9235                                  * RTF_SETSRC flag, so the source address of the
9236                                  * packet must be changed.
9237                                  */
9238                                 ipha->ipha_src = fire->ire_src_addr;
9239                         }
9240                 } else {
9241                         ASSERT((connp == NULL) ||
9242                             (connp->conn_outgoing_ill != NULL) ||
9243                             (connp->conn_dontroute) ||
9244                             infop->ip_opt_ill_index != 0);
9245                         /*
9246                          * The only ways we can come here are:
9247                          * 1) IP_BOUND_IF socket option is set
9248                          * 2) SO_DONTROUTE socket option is set
9249                          * 3) IP_PKTINFO option is passed in as ancillary data.
9250                          * In all cases, the new ire will not be added
9251                          * into cache table.
9252                          */
9253                         ire_marks |= IRE_MARK_NOADD;
9254                 }
9255 
9256                 switch (ipif->ipif_net_type) {
9257                 case IRE_IF_NORESOLVER: {
9258                         /* We have what we need to build an IRE_CACHE. */
9259 
9260                         if ((dst_ill->ill_phys_addr_length != IP_ADDR_LEN) &&
9261                             (dst_ill->ill_resolver_mp == NULL)) {
9262                                 ip1dbg(("ip_newroute_ipif: dst_ill %p "
9263                                     "for IRE_IF_NORESOLVER ire %p has "
9264                                     "no ill_resolver_mp\n",
9265                                     (void *)dst_ill, (void *)ire));
9266                                 break;
9267                         }
9268 
9269                         /*
9270                          * The new ire inherits the IRE_OFFSUBNET flags
9271                          * and source address, if this was requested.
9272                          */
9273                         ire = ire_create(
9274                             (uchar_t *)&dst,                /* dest address */
9275                             (uchar_t *)&ip_g_all_ones,      /* mask */
9276                             (uchar_t *)&src_ipif->ipif_src_addr, /* src addr */
9277                             NULL,                       /* gateway address */
9278                             &ipif->ipif_mtu,
9279                             NULL,                       /* no src nce */
9280                             dst_ill->ill_rq,         /* recv-from queue */
9281                             dst_ill->ill_wq,         /* send-to queue */
9282                             IRE_CACHE,
9283                             src_ipif,
9284                             (save_ire != NULL ? save_ire->ire_mask : 0),
9285                             (fire != NULL) ?            /* Parent handle */
9286                             fire->ire_phandle : 0,
9287                             ihandle,                    /* Interface handle */
9288                             (fire != NULL) ?
9289                             (fire->ire_flags &
9290                             (RTF_SETSRC | RTF_MULTIRT)) : 0,
9291                             (save_ire == NULL ? &ire_uinfo_null :
9292                             &save_ire->ire_uinfo),
9293                             NULL,
9294                             NULL,
9295                             ipst);
9296 
9297                         if (ire == NULL) {
9298                                 if (save_ire != NULL)
9299                                         ire_refrele(save_ire);
9300                                 break;
9301                         }
9302 
9303                         ire->ire_marks |= ire_marks;
9304 
9305                         /*
9306                          * If IRE_MARK_NOADD is set then we need to convert
9307                          * the max_fragp to a useable value now. This is
9308                          * normally done in ire_add_v[46]. We also need to
9309                          * associate the ire with an nce (normally would be
9310                          * done in ip_wput_nondata()).
9311                          *
9312                          * Note that IRE_MARK_NOADD packets created here
9313                          * do not have a non-null ire_mp pointer. The null
9314                          * value of ire_bucket indicates that they were
9315                          * never added.
9316                          */
9317                         if (ire->ire_marks & IRE_MARK_NOADD) {
9318                                 uint_t  max_frag;
9319 
9320                                 max_frag = *ire->ire_max_fragp;
9321                                 ire->ire_max_fragp = NULL;
9322                                 ire->ire_max_frag = max_frag;
9323 
9324                                 if ((ire->ire_nce = ndp_lookup_v4(
9325                                     ire_to_ill(ire),
9326                                     (ire->ire_gateway_addr != INADDR_ANY ?
9327                                     &ire->ire_gateway_addr : &ire->ire_addr),
9328                                     B_FALSE)) == NULL) {
9329                                         if (save_ire != NULL)
9330                                                 ire_refrele(save_ire);
9331                                         break;
9332                                 }
9333                                 ASSERT(ire->ire_nce->nce_state ==
9334                                     ND_REACHABLE);
9335                                 NCE_REFHOLD_TO_REFHOLD_NOTR(ire->ire_nce);
9336                         }
9337 
9338                         /* Prevent save_ire from getting deleted */
9339                         if (save_ire != NULL) {
9340                                 IRB_REFHOLD(save_ire->ire_bucket);
9341                                 /* Has it been removed already ? */
9342                                 if (save_ire->ire_marks & IRE_MARK_CONDEMNED) {
9343                                         IRB_REFRELE(save_ire->ire_bucket);
9344                                         ire_refrele(save_ire);
9345                                         break;
9346                                 }
9347                         }
9348 
9349                         ire_add_then_send(q, ire, first_mp);
9350 
9351                         /* Assert that save_ire is not deleted yet. */
9352                         if (save_ire != NULL) {
9353                                 ASSERT(save_ire->ire_ptpn != NULL);
9354                                 IRB_REFRELE(save_ire->ire_bucket);
9355                                 ire_refrele(save_ire);
9356                                 save_ire = NULL;
9357                         }
9358                         if (fire != NULL) {
9359                                 ire_refrele(fire);
9360                                 fire = NULL;
9361                         }
9362 
9363                         /*
9364                          * the resolution loop is re-entered if this
9365                          * was requested through flags and if we
9366                          * actually are in a multirouting case.
9367                          */
9368                         if ((flags & RTF_MULTIRT) && (copy_mp != NULL)) {
9369                                 boolean_t need_resolve =
9370                                     ire_multirt_need_resolve(ipha_dst,
9371                                     MBLK_GETLABEL(copy_mp), ipst);
9372                                 if (!need_resolve) {
9373                                         MULTIRT_DEBUG_UNTAG(copy_mp);
9374                                         freemsg(copy_mp);
9375                                         copy_mp = NULL;
9376                                 } else {
9377                                         /*
9378                                          * ipif_lookup_group() calls
9379                                          * ire_lookup_multi() that uses
9380                                          * ire_ftable_lookup() to find
9381                                          * an IRE_INTERFACE for the group.
9382                                          * In the multirt case,
9383                                          * ire_lookup_multi() then invokes
9384                                          * ire_multirt_lookup() to find
9385                                          * the next resolvable ire.
9386                                          * As a result, we obtain an new
9387                                          * interface, derived from the
9388                                          * next ire.
9389                                          */
9390                                         ipif_refrele(ipif);
9391                                         ipif = ipif_lookup_group(ipha_dst,
9392                                             zoneid, ipst);
9393                                         ip2dbg(("ip_newroute_ipif: "
9394                                             "multirt dst %08x, ipif %p\n",
9395                                             htonl(dst), (void *)ipif));
9396                                         if (ipif != NULL) {
9397                                                 mp = copy_mp;
9398                                                 copy_mp = NULL;
9399                                                 multirt_resolve_next = B_TRUE;
9400                                                 continue;
9401                                         } else {
9402                                                 freemsg(copy_mp);
9403                                         }
9404                                 }
9405                         }
9406                         if (ipif != NULL)
9407                                 ipif_refrele(ipif);
9408                         ill_refrele(dst_ill);
9409                         ipif_refrele(src_ipif);
9410                         return;
9411                 }
9412                 case IRE_IF_RESOLVER:
9413                         /*
9414                          * We can't build an IRE_CACHE yet, but at least
9415                          * we found a resolver that can help.
9416                          */
9417                         res_mp = dst_ill->ill_resolver_mp;
9418                         if (!OK_RESOLVER_MP(res_mp))
9419                                 break;
9420 
9421                         /*
9422                          * We obtain a partial IRE_CACHE which we will pass
9423                          * along with the resolver query.  When the response
9424                          * comes back it will be there ready for us to add.
9425                          * The new ire inherits the IRE_OFFSUBNET flags
9426                          * and source address, if this was requested.
9427                          * The ire_max_frag is atomically set under the
9428                          * irebucket lock in ire_add_v[46]. Only in the
9429                          * case of IRE_MARK_NOADD, we set it here itself.
9430                          */
9431                         ire = ire_create_mp(
9432                             (uchar_t *)&dst,                /* dest address */
9433                             (uchar_t *)&ip_g_all_ones,      /* mask */
9434                             (uchar_t *)&src_ipif->ipif_src_addr, /* src addr */
9435                             NULL,                       /* gateway address */
9436                             (ire_marks & IRE_MARK_NOADD) ?
9437                             ipif->ipif_mtu : 0,              /* max_frag */
9438                             NULL,                       /* no src nce */
9439                             dst_ill->ill_rq,         /* recv-from queue */
9440                             dst_ill->ill_wq,         /* send-to queue */
9441                             IRE_CACHE,
9442                             src_ipif,
9443                             (save_ire != NULL ? save_ire->ire_mask : 0),
9444                             (fire != NULL) ?            /* Parent handle */
9445                             fire->ire_phandle : 0,
9446                             ihandle,                    /* Interface handle */
9447                             (fire != NULL) ?            /* flags if any */
9448                             (fire->ire_flags &
9449                             (RTF_SETSRC | RTF_MULTIRT)) : 0,
9450                             (save_ire == NULL ? &ire_uinfo_null :
9451                             &save_ire->ire_uinfo),
9452                             NULL,
9453                             NULL,
9454                             ipst);
9455 
9456                         if (save_ire != NULL) {
9457                                 ire_refrele(save_ire);
9458                                 save_ire = NULL;
9459                         }
9460                         if (ire == NULL)
9461                                 break;
9462 
9463                         ire->ire_marks |= ire_marks;
9464                         /*
9465                          * Construct message chain for the resolver of the
9466                          * form:
9467                          *      ARP_REQ_MBLK-->IRE_MBLK-->Packet
9468                          *
9469                          * NOTE : ire will be added later when the response
9470                          * comes back from ARP. If the response does not
9471                          * come back, ARP frees the packet. For this reason,
9472                          * we can't REFHOLD the bucket of save_ire to prevent
9473                          * deletions. We may not be able to REFRELE the
9474                          * bucket if the response never comes back.
9475                          * Thus, before adding the ire, ire_add_v4 will make
9476                          * sure that the interface route does not get deleted.
9477                          * This is the only case unlike ip_newroute_v6,
9478                          * ip_newroute_ipif_v6 where we can always prevent
9479                          * deletions because ire_add_then_send is called after
9480                          * creating the IRE.
9481                          * If IRE_MARK_NOADD is set, then ire_add_then_send
9482                          * does not add this IRE into the IRE CACHE.
9483                          */
9484                         ASSERT(ire->ire_mp != NULL);
9485                         ire->ire_mp->b_cont = first_mp;
9486                         /* Have saved_mp handy, for cleanup if canput fails */
9487                         saved_mp = mp;
9488                         mp = copyb(res_mp);
9489                         if (mp == NULL) {
9490                                 /* Prepare for cleanup */
9491                                 mp = saved_mp; /* pkt */
9492                                 ire_delete(ire); /* ire_mp */
9493                                 ire = NULL;
9494                                 if (copy_mp != NULL) {
9495                                         MULTIRT_DEBUG_UNTAG(copy_mp);
9496                                         freemsg(copy_mp);
9497                                         copy_mp = NULL;
9498                                 }
9499                                 break;
9500                         }
9501                         linkb(mp, ire->ire_mp);
9502 
9503                         /*
9504                          * Fill in the source and dest addrs for the resolver.
9505                          * NOTE: this depends on memory layouts imposed by
9506                          * ill_init().
9507                          */
9508                         areq = (areq_t *)mp->b_rptr;
9509                         addrp = (ipaddr_t *)((char *)areq +
9510                             areq->areq_sender_addr_offset);
9511                         *addrp = ire->ire_src_addr;
9512                         addrp = (ipaddr_t *)((char *)areq +
9513                             areq->areq_target_addr_offset);
9514                         *addrp = dst;
9515                         /* Up to the resolver. */
9516                         if (canputnext(dst_ill->ill_rq) &&
9517                             !(dst_ill->ill_arp_closing)) {
9518                                 putnext(dst_ill->ill_rq, mp);
9519                                 /*
9520                                  * The response will come back in ip_wput
9521                                  * with db_type IRE_DB_TYPE.
9522                                  */
9523                         } else {
9524                                 mp->b_cont = NULL;
9525                                 freeb(mp); /* areq */
9526                                 ire_delete(ire); /* ire_mp */
9527                                 saved_mp->b_next = NULL;
9528                                 saved_mp->b_prev = NULL;
9529                                 freemsg(first_mp); /* pkt */
9530                                 ip2dbg(("ip_newroute_ipif: dropped\n"));
9531                         }
9532 
9533                         if (fire != NULL) {
9534                                 ire_refrele(fire);
9535                                 fire = NULL;
9536                         }
9537 
9538 
9539                         /*
9540                          * The resolution loop is re-entered if this was
9541                          * requested through flags and we actually are
9542                          * in a multirouting case.
9543                          */
9544                         if ((flags & RTF_MULTIRT) && (copy_mp != NULL)) {
9545                                 boolean_t need_resolve =
9546                                     ire_multirt_need_resolve(ipha_dst,
9547                                     MBLK_GETLABEL(copy_mp), ipst);
9548                                 if (!need_resolve) {
9549                                         MULTIRT_DEBUG_UNTAG(copy_mp);
9550                                         freemsg(copy_mp);
9551                                         copy_mp = NULL;
9552                                 } else {
9553                                         /*
9554                                          * ipif_lookup_group() calls
9555                                          * ire_lookup_multi() that uses
9556                                          * ire_ftable_lookup() to find
9557                                          * an IRE_INTERFACE for the group.
9558                                          * In the multirt case,
9559                                          * ire_lookup_multi() then invokes
9560                                          * ire_multirt_lookup() to find
9561                                          * the next resolvable ire.
9562                                          * As a result, we obtain an new
9563                                          * interface, derived from the
9564                                          * next ire.
9565                                          */
9566                                         ipif_refrele(ipif);
9567                                         ipif = ipif_lookup_group(ipha_dst,
9568                                             zoneid, ipst);
9569                                         if (ipif != NULL) {
9570                                                 mp = copy_mp;
9571                                                 copy_mp = NULL;
9572                                                 multirt_resolve_next = B_TRUE;
9573                                                 continue;
9574                                         } else {
9575                                                 freemsg(copy_mp);
9576                                         }
9577                                 }
9578                         }
9579                         if (ipif != NULL)
9580                                 ipif_refrele(ipif);
9581                         ill_refrele(dst_ill);
9582                         ipif_refrele(src_ipif);
9583                         return;
9584                 default:
9585                         break;
9586                 }
9587         } while (multirt_resolve_next);
9588 
9589 err_ret:
9590         ip2dbg(("ip_newroute_ipif: dropped\n"));
9591         if (fire != NULL)
9592                 ire_refrele(fire);
9593         ipif_refrele(ipif);
9594         /* Did this packet originate externally? */
9595         if (dst_ill != NULL)
9596                 ill_refrele(dst_ill);
9597         if (src_ipif != NULL)
9598                 ipif_refrele(src_ipif);
9599         if (mp->b_prev || mp->b_next) {
9600                 mp->b_next = NULL;
9601                 mp->b_prev = NULL;
9602         } else {
9603                 /*
9604                  * Since ip_wput() isn't close to finished, we fill
9605                  * in enough of the header for credible error reporting.
9606                  */
9607                 if (ip_hdr_complete((ipha_t *)mp->b_rptr, zoneid, ipst)) {
9608                         /* Failed */
9609                         freemsg(first_mp);
9610                         if (ire != NULL)
9611                                 ire_refrele(ire);
9612                         return;
9613                 }
9614         }
9615         /*
9616          * At this point we will have ire only if RTF_BLACKHOLE
9617          * or RTF_REJECT flags are set on the IRE. It will not
9618          * generate ICMP_HOST_UNREACHABLE if RTF_BLACKHOLE is set.
9619          */
9620         if (ire != NULL) {
9621                 if (ire->ire_flags & RTF_BLACKHOLE) {
9622                         ire_refrele(ire);
9623                         freemsg(first_mp);
9624                         return;
9625                 }
9626                 ire_refrele(ire);
9627         }
9628         icmp_unreachable(q, first_mp, ICMP_HOST_UNREACHABLE, zoneid, ipst);
9629 }
9630 
9631 /* Name/Value Table Lookup Routine */
9632 char *
9633 ip_nv_lookup(nv_t *nv, int value)
9634 {
9635         if (!nv)
9636                 return (NULL);
9637         for (; nv->nv_name; nv++) {
9638                 if (nv->nv_value == value)
9639                         return (nv->nv_name);
9640         }
9641         return ("unknown");
9642 }
9643 
9644 /*
9645  * This is a module open, i.e. this is a control stream for access
9646  * to a DLPI device.  We allocate an ill_t as the instance data in
9647  * this case.
9648  */
9649 int
9650 ip_modopen(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
9651 {
9652         ill_t   *ill;
9653         int     err;
9654         zoneid_t zoneid;
9655         netstack_t *ns;
9656         ip_stack_t *ipst;
9657 
9658         /*
9659          * Prevent unprivileged processes from pushing IP so that
9660          * they can't send raw IP.
9661          */
9662         if (secpolicy_net_rawaccess(credp) != 0)
9663                 return (EPERM);
9664 
9665         ns = netstack_find_by_cred(credp);
9666         ASSERT(ns != NULL);
9667         ipst = ns->netstack_ip;
9668         ASSERT(ipst != NULL);
9669 
9670         /*
9671          * For exclusive stacks we set the zoneid to zero
9672          * to make IP operate as if in the global zone.
9673          */
9674         if (ipst->ips_netstack->netstack_stackid != GLOBAL_NETSTACKID)
9675                 zoneid = GLOBAL_ZONEID;
9676         else
9677                 zoneid = crgetzoneid(credp);
9678 
9679         ill = (ill_t *)mi_open_alloc_sleep(sizeof (ill_t));
9680         q->q_ptr = WR(q)->q_ptr = ill;
9681         ill->ill_ipst = ipst;
9682         ill->ill_zoneid = zoneid;
9683 
9684         /*
9685          * ill_init initializes the ill fields and then sends down
9686          * down a DL_INFO_REQ after calling qprocson.
9687          */
9688         err = ill_init(q, ill);
9689         if (err != 0) {
9690                 mi_free(ill);
9691                 netstack_rele(ipst->ips_netstack);
9692                 q->q_ptr = NULL;
9693                 WR(q)->q_ptr = NULL;
9694                 return (err);
9695         }
9696 
9697         /* ill_init initializes the ipsq marking this thread as writer */
9698         ipsq_exit(ill->ill_phyint->phyint_ipsq);
9699         /* Wait for the DL_INFO_ACK */
9700         mutex_enter(&ill->ill_lock);
9701         while (ill->ill_state_flags & ILL_LL_SUBNET_PENDING) {
9702                 /*
9703                  * Return value of 0 indicates a pending signal.
9704                  */
9705                 err = cv_wait_sig(&ill->ill_cv, &ill->ill_lock);
9706                 if (err == 0) {
9707                         mutex_exit(&ill->ill_lock);
9708                         (void) ip_close(q, 0);
9709                         return (EINTR);
9710                 }
9711         }
9712         mutex_exit(&ill->ill_lock);
9713 
9714         /*
9715          * ip_rput_other could have set an error  in ill_error on
9716          * receipt of M_ERROR.
9717          */
9718 
9719         err = ill->ill_error;
9720         if (err != 0) {
9721                 (void) ip_close(q, 0);
9722                 return (err);
9723         }
9724 
9725         ill->ill_credp = credp;
9726         crhold(credp);
9727 
9728         mutex_enter(&ipst->ips_ip_mi_lock);
9729         err = mi_open_link(&ipst->ips_ip_g_head, (IDP)ill, devp, flag, sflag,
9730             credp);
9731         mutex_exit(&ipst->ips_ip_mi_lock);
9732         if (err) {
9733                 (void) ip_close(q, 0);
9734                 return (err);
9735         }
9736         return (0);
9737 }
9738 
9739 /* For /dev/ip aka AF_INET open */
9740 int
9741 ip_openv4(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
9742 {
9743         return (ip_open(q, devp, flag, sflag, credp, B_FALSE));
9744 }
9745 
9746 /* For /dev/ip6 aka AF_INET6 open */
9747 int
9748 ip_openv6(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
9749 {
9750         return (ip_open(q, devp, flag, sflag, credp, B_TRUE));
9751 }
9752 
9753 /* IP open routine. */
9754 int
9755 ip_open(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp,
9756     boolean_t isv6)
9757 {
9758         conn_t          *connp;
9759         major_t         maj;
9760         zoneid_t        zoneid;
9761         netstack_t      *ns;
9762         ip_stack_t      *ipst;
9763 
9764         TRACE_1(TR_FAC_IP, TR_IP_OPEN, "ip_open: q %p", q);
9765 
9766         /* Allow reopen. */
9767         if (q->q_ptr != NULL)
9768                 return (0);
9769 
9770         if (sflag & MODOPEN) {
9771                 /* This is a module open */
9772                 return (ip_modopen(q, devp, flag, sflag, credp));
9773         }
9774 
9775         ns = netstack_find_by_cred(credp);
9776         ASSERT(ns != NULL);
9777         ipst = ns->netstack_ip;
9778         ASSERT(ipst != NULL);
9779 
9780         /*
9781          * For exclusive stacks we set the zoneid to zero
9782          * to make IP operate as if in the global zone.
9783          */
9784         if (ipst->ips_netstack->netstack_stackid != GLOBAL_NETSTACKID)
9785                 zoneid = GLOBAL_ZONEID;
9786         else
9787                 zoneid = crgetzoneid(credp);
9788 
9789         /*
9790          * We are opening as a device. This is an IP client stream, and we
9791          * allocate an conn_t as the instance data.
9792          */
9793         connp = ipcl_conn_create(IPCL_IPCCONN, KM_SLEEP, ipst->ips_netstack);
9794 
9795         /*
9796          * ipcl_conn_create did a netstack_hold. Undo the hold that was
9797          * done by netstack_find_by_cred()
9798          */
9799         netstack_rele(ipst->ips_netstack);
9800 
9801         connp->conn_zoneid = zoneid;
9802 
9803         connp->conn_upq = q;
9804         q->q_ptr = WR(q)->q_ptr = connp;
9805 
9806         if (flag & SO_SOCKSTR)
9807                 connp->conn_flags |= IPCL_SOCKET;
9808 
9809         /* Minor tells us which /dev entry was opened */
9810         if (isv6) {
9811                 connp->conn_flags |= IPCL_ISV6;
9812                 connp->conn_af_isv6 = B_TRUE;
9813                 ip_setpktversion(connp, isv6, B_FALSE, ipst);
9814                 connp->conn_src_preferences = IPV6_PREFER_SRC_DEFAULT;
9815         } else {
9816                 connp->conn_af_isv6 = B_FALSE;
9817                 connp->conn_pkt_isv6 = B_FALSE;
9818         }
9819 
9820         if ((ip_minor_arena_la != NULL) && (flag & SO_SOCKSTR) &&
9821             ((connp->conn_dev = inet_minor_alloc(ip_minor_arena_la)) != 0)) {
9822                 connp->conn_minor_arena = ip_minor_arena_la;
9823         } else {
9824                 /*
9825                  * Either minor numbers in the large arena were exhausted
9826                  * or a non socket application is doing the open.
9827                  * Try to allocate from the small arena.
9828                  */
9829                 if ((connp->conn_dev =
9830                     inet_minor_alloc(ip_minor_arena_sa)) == 0) {
9831                         /* CONN_DEC_REF takes care of netstack_rele() */
9832                         q->q_ptr = WR(q)->q_ptr = NULL;
9833                         CONN_DEC_REF(connp);
9834                         return (EBUSY);
9835                 }
9836                 connp->conn_minor_arena = ip_minor_arena_sa;
9837         }
9838 
9839         maj = getemajor(*devp);
9840         *devp = makedevice(maj, (minor_t)connp->conn_dev);
9841 
9842         /*
9843          * connp->conn_cred is crfree()ed in ipcl_conn_destroy()
9844          */
9845         connp->conn_cred = credp;
9846 
9847         /*
9848          * Handle IP_RTS_REQUEST and other ioctls which use conn_recv
9849          */
9850         connp->conn_recv = ip_conn_input;
9851 
9852         crhold(connp->conn_cred);
9853 
9854         /*
9855          * If the caller has the process-wide flag set, then default to MAC
9856          * exempt mode.  This allows read-down to unlabeled hosts.
9857          */
9858         if (getpflags(NET_MAC_AWARE, credp) != 0)
9859                 connp->conn_mac_exempt = B_TRUE;
9860 
9861         connp->conn_rq = q;
9862         connp->conn_wq = WR(q);
9863 
9864         /* Non-zero default values */
9865         connp->conn_multicast_loop = IP_DEFAULT_MULTICAST_LOOP;
9866 
9867         /*
9868          * Make the conn globally visible to walkers
9869          */
9870         ASSERT(connp->conn_ref == 1);
9871         mutex_enter(&connp->conn_lock);
9872         connp->conn_state_flags &= ~CONN_INCIPIENT;
9873         mutex_exit(&connp->conn_lock);
9874 
9875         qprocson(q);
9876 
9877         return (0);
9878 }
9879 
9880 /*
9881  * Change the output format (IPv4 vs. IPv6) for a conn_t.
9882  * Note that there is no race since either ip_output function works - it
9883  * is just an optimization to enter the best ip_output routine directly.
9884  */
9885 void
9886 ip_setpktversion(conn_t *connp, boolean_t isv6, boolean_t bump_mib,
9887     ip_stack_t *ipst)
9888 {
9889         if (isv6)  {
9890                 if (bump_mib) {
9891                         BUMP_MIB(&ipst->ips_ip6_mib,
9892                             ipIfStatsOutSwitchIPVersion);
9893                 }
9894                 connp->conn_send = ip_output_v6;
9895                 connp->conn_pkt_isv6 = B_TRUE;
9896         } else {
9897                 if (bump_mib) {
9898                         BUMP_MIB(&ipst->ips_ip_mib,
9899                             ipIfStatsOutSwitchIPVersion);
9900                 }
9901                 connp->conn_send = ip_output;
9902                 connp->conn_pkt_isv6 = B_FALSE;
9903         }
9904 
9905 }
9906 
9907 /*
9908  * See if IPsec needs loading because of the options in mp.
9909  */
9910 static boolean_t
9911 ipsec_opt_present(mblk_t *mp)
9912 {
9913         uint8_t *optcp, *next_optcp, *opt_endcp;
9914         struct opthdr *opt;
9915         struct T_opthdr *topt;
9916         int opthdr_len;
9917         t_uscalar_t optname, optlevel;
9918         struct T_optmgmt_req *tor = (struct T_optmgmt_req *)mp->b_rptr;
9919         ipsec_req_t *ipsr;
9920 
9921         /*
9922          * Walk through the mess, and find IP_SEC_OPT.  If it's there,
9923          * return TRUE.
9924          */
9925 
9926         optcp = mi_offset_param(mp, tor->OPT_offset, tor->OPT_length);
9927         opt_endcp = optcp + tor->OPT_length;
9928         if (tor->PRIM_type == T_OPTMGMT_REQ) {
9929                 opthdr_len = sizeof (struct T_opthdr);
9930         } else {                /* O_OPTMGMT_REQ */
9931                 ASSERT(tor->PRIM_type == T_SVR4_OPTMGMT_REQ);
9932                 opthdr_len = sizeof (struct opthdr);
9933         }
9934         for (; optcp < opt_endcp; optcp = next_optcp) {
9935                 if (optcp + opthdr_len > opt_endcp)
9936                         return (B_FALSE);       /* Not enough option header. */
9937                 if (tor->PRIM_type == T_OPTMGMT_REQ) {
9938                         topt = (struct T_opthdr *)optcp;
9939                         optlevel = topt->level;
9940                         optname = topt->name;
9941                         next_optcp = optcp + _TPI_ALIGN_TOPT(topt->len);
9942                 } else {
9943                         opt = (struct opthdr *)optcp;
9944                         optlevel = opt->level;
9945                         optname = opt->name;
9946                         next_optcp = optcp + opthdr_len +
9947                             _TPI_ALIGN_OPT(opt->len);
9948                 }
9949                 if ((next_optcp < optcp) || /* wraparound pointer space */
9950                     ((next_optcp >= opt_endcp) && /* last option bad len */
9951                     ((next_optcp - opt_endcp) >= __TPI_ALIGN_SIZE)))
9952                         return (B_FALSE); /* bad option buffer */
9953                 if ((optlevel == IPPROTO_IP && optname == IP_SEC_OPT) ||
9954                     (optlevel == IPPROTO_IPV6 && optname == IPV6_SEC_OPT)) {
9955                         /*
9956                          * Check to see if it's an all-bypass or all-zeroes
9957                          * IPsec request.  Don't bother loading IPsec if
9958                          * the socket doesn't want to use it.  (A good example
9959                          * is a bypass request.)
9960                          *
9961                          * Basically, if any of the non-NEVER bits are set,
9962                          * load IPsec.
9963                          */
9964                         ipsr = (ipsec_req_t *)(optcp + opthdr_len);
9965                         if ((ipsr->ipsr_ah_req & ~IPSEC_PREF_NEVER) != 0 ||
9966                             (ipsr->ipsr_esp_req & ~IPSEC_PREF_NEVER) != 0 ||
9967                             (ipsr->ipsr_self_encap_req & ~IPSEC_PREF_NEVER)
9968                             != 0)
9969                                 return (B_TRUE);
9970                 }
9971         }
9972         return (B_FALSE);
9973 }
9974 
9975 /*
9976  * If conn is is waiting for ipsec to finish loading, kick it.
9977  */
9978 /* ARGSUSED */
9979 static void
9980 conn_restart_ipsec_waiter(conn_t *connp, void *arg)
9981 {
9982         t_scalar_t      optreq_prim;
9983         mblk_t          *mp;
9984         cred_t          *cr;
9985         int             err = 0;
9986 
9987         /*
9988          * This function is called, after ipsec loading is complete.
9989          * Since IP checks exclusively and atomically (i.e it prevents
9990          * ipsec load from completing until ip_optcom_req completes)
9991          * whether ipsec load is complete, there cannot be a race with IP
9992          * trying to set the CONN_IPSEC_LOAD_WAIT flag on any conn now.
9993          */
9994         mutex_enter(&connp->conn_lock);
9995         if (connp->conn_state_flags & CONN_IPSEC_LOAD_WAIT) {
9996                 ASSERT(connp->conn_ipsec_opt_mp != NULL);
9997                 mp = connp->conn_ipsec_opt_mp;
9998                 connp->conn_ipsec_opt_mp = NULL;
9999                 connp->conn_state_flags  &= ~CONN_IPSEC_LOAD_WAIT;
10000                 cr = DB_CREDDEF(mp, GET_QUEUE_CRED(CONNP_TO_WQ(connp)));
10001                 mutex_exit(&connp->conn_lock);
10002 
10003                 ASSERT(DB_TYPE(mp) == M_PROTO || DB_TYPE(mp) == M_PCPROTO);
10004 
10005                 optreq_prim = ((union T_primitives *)mp->b_rptr)->type;
10006                 if (optreq_prim == T_OPTMGMT_REQ) {
10007                         err = tpi_optcom_req(CONNP_TO_WQ(connp), mp, cr,
10008                             &ip_opt_obj, B_FALSE);
10009                 } else {
10010                         ASSERT(optreq_prim == T_SVR4_OPTMGMT_REQ);
10011                         err = svr4_optcom_req(CONNP_TO_WQ(connp), mp, cr,
10012                             &ip_opt_obj, B_FALSE);
10013                 }
10014                 if (err != EINPROGRESS)
10015                         CONN_OPER_PENDING_DONE(connp);
10016                 return;
10017         }
10018         mutex_exit(&connp->conn_lock);
10019 }
10020 
10021 /*
10022  * Called from the ipsec_loader thread, outside any perimeter, to tell
10023  * ip qenable any of the queues waiting for the ipsec loader to
10024  * complete.
10025  */
10026 void
10027 ip_ipsec_load_complete(ipsec_stack_t *ipss)
10028 {
10029         netstack_t *ns = ipss->ipsec_netstack;
10030 
10031         ipcl_walk(conn_restart_ipsec_waiter, NULL, ns->netstack_ip);
10032 }
10033 
10034 /*
10035  * Can't be used. Need to call svr4* -> optset directly. the leaf routine
10036  * determines the grp on which it has to become exclusive, queues the mp
10037  * and sq draining restarts the optmgmt
10038  */
10039 static boolean_t
10040 ip_check_for_ipsec_opt(queue_t *q, mblk_t *mp)
10041 {
10042         conn_t *connp = Q_TO_CONN(q);
10043         ipsec_stack_t *ipss = connp->conn_netstack->netstack_ipsec;
10044 
10045         /*
10046          * Take IPsec requests and treat them special.
10047          */
10048         if (ipsec_opt_present(mp)) {
10049                 /* First check if IPsec is loaded. */
10050                 mutex_enter(&ipss->ipsec_loader_lock);
10051                 if (ipss->ipsec_loader_state != IPSEC_LOADER_WAIT) {
10052                         mutex_exit(&ipss->ipsec_loader_lock);
10053                         return (B_FALSE);
10054                 }
10055                 mutex_enter(&connp->conn_lock);
10056                 connp->conn_state_flags |= CONN_IPSEC_LOAD_WAIT;
10057 
10058                 ASSERT(connp->conn_ipsec_opt_mp == NULL);
10059                 connp->conn_ipsec_opt_mp = mp;
10060                 mutex_exit(&connp->conn_lock);
10061                 mutex_exit(&ipss->ipsec_loader_lock);
10062 
10063                 ipsec_loader_loadnow(ipss);
10064                 return (B_TRUE);
10065         }
10066         return (B_FALSE);
10067 }
10068 
10069 /*
10070  * Set IPsec policy from an ipsec_req_t. If the req is not "zero" and valid,
10071  * all of them are copied to the conn_t. If the req is "zero", the policy is
10072  * zeroed out. A "zero" policy has zero ipsr_{ah,req,self_encap}_req
10073  * fields.
10074  * We keep only the latest setting of the policy and thus policy setting
10075  * is not incremental/cumulative.
10076  *
10077  * Requests to set policies with multiple alternative actions will
10078  * go through a different API.
10079  */
10080 int
10081 ipsec_set_req(cred_t *cr, conn_t *connp, ipsec_req_t *req)
10082 {
10083         uint_t ah_req = 0;
10084         uint_t esp_req = 0;
10085         uint_t se_req = 0;
10086         ipsec_selkey_t sel;
10087         ipsec_act_t *actp = NULL;
10088         uint_t nact;
10089         ipsec_policy_t *pin4 = NULL, *pout4 = NULL;
10090         ipsec_policy_t *pin6 = NULL, *pout6 = NULL;
10091         ipsec_policy_root_t *pr;
10092         ipsec_policy_head_t *ph;
10093         int fam;
10094         boolean_t is_pol_reset;
10095         int error = 0;
10096         netstack_t      *ns = connp->conn_netstack;
10097         ip_stack_t      *ipst = ns->netstack_ip;
10098         ipsec_stack_t   *ipss = ns->netstack_ipsec;
10099 
10100 #define REQ_MASK (IPSEC_PREF_REQUIRED|IPSEC_PREF_NEVER)
10101 
10102         /*
10103          * The IP_SEC_OPT option does not allow variable length parameters,
10104          * hence a request cannot be NULL.
10105          */
10106         if (req == NULL)
10107                 return (EINVAL);
10108 
10109         ah_req = req->ipsr_ah_req;
10110         esp_req = req->ipsr_esp_req;
10111         se_req = req->ipsr_self_encap_req;
10112 
10113         /* Don't allow setting self-encap without one or more of AH/ESP. */
10114         if (se_req != 0 && esp_req == 0 && ah_req == 0)
10115                 return (EINVAL);
10116 
10117         /*
10118          * Are we dealing with a request to reset the policy (i.e.
10119          * zero requests).
10120          */
10121         is_pol_reset = ((ah_req & REQ_MASK) == 0 &&
10122             (esp_req & REQ_MASK) == 0 &&
10123             (se_req & REQ_MASK) == 0);
10124 
10125         if (!is_pol_reset) {
10126                 /*
10127                  * If we couldn't load IPsec, fail with "protocol
10128                  * not supported".
10129                  * IPsec may not have been loaded for a request with zero
10130                  * policies, so we don't fail in this case.
10131                  */
10132                 mutex_enter(&ipss->ipsec_loader_lock);
10133                 if (ipss->ipsec_loader_state != IPSEC_LOADER_SUCCEEDED) {
10134                         mutex_exit(&ipss->ipsec_loader_lock);
10135                         return (EPROTONOSUPPORT);
10136                 }
10137                 mutex_exit(&ipss->ipsec_loader_lock);
10138 
10139                 /*
10140                  * Test for valid requests. Invalid algorithms
10141                  * need to be tested by IPsec code because new
10142                  * algorithms can be added dynamically.
10143                  */
10144                 if ((ah_req & ~(REQ_MASK|IPSEC_PREF_UNIQUE)) != 0 ||
10145                     (esp_req & ~(REQ_MASK|IPSEC_PREF_UNIQUE)) != 0 ||
10146                     (se_req & ~(REQ_MASK|IPSEC_PREF_UNIQUE)) != 0) {
10147                         return (EINVAL);
10148                 }
10149 
10150                 /*
10151                  * Only privileged users can issue these
10152                  * requests.
10153                  */
10154                 if (((ah_req & IPSEC_PREF_NEVER) ||
10155                     (esp_req & IPSEC_PREF_NEVER) ||
10156                     (se_req & IPSEC_PREF_NEVER)) &&
10157                     secpolicy_ip_config(cr, B_FALSE) != 0) {
10158                         return (EPERM);
10159                 }
10160 
10161                 /*
10162                  * The IPSEC_PREF_REQUIRED and IPSEC_PREF_NEVER
10163                  * are mutually exclusive.
10164                  */
10165                 if (((ah_req & REQ_MASK) == REQ_MASK) ||
10166                     ((esp_req & REQ_MASK) == REQ_MASK) ||
10167                     ((se_req & REQ_MASK) == REQ_MASK)) {
10168                         /* Both of them are set */
10169                         return (EINVAL);
10170                 }
10171         }
10172 
10173         mutex_enter(&connp->conn_lock);
10174 
10175         /*
10176          * If we have already cached policies in ip_bind_connected*(), don't
10177          * let them change now. We cache policies for connections
10178          * whose src,dst [addr, port] is known.
10179          */
10180         if (connp->conn_policy_cached) {
10181                 mutex_exit(&connp->conn_lock);
10182                 return (EINVAL);
10183         }
10184 
10185         /*
10186          * We have a zero policies, reset the connection policy if already
10187          * set. This will cause the connection to inherit the
10188          * global policy, if any.
10189          */
10190         if (is_pol_reset) {
10191                 if (connp->conn_policy != NULL) {
10192                         IPPH_REFRELE(connp->conn_policy, ipst->ips_netstack);
10193                         connp->conn_policy = NULL;
10194                 }
10195                 connp->conn_flags &= ~IPCL_CHECK_POLICY;
10196                 connp->conn_in_enforce_policy = B_FALSE;
10197                 connp->conn_out_enforce_policy = B_FALSE;
10198                 mutex_exit(&connp->conn_lock);
10199                 return (0);
10200         }
10201 
10202         ph = connp->conn_policy = ipsec_polhead_split(connp->conn_policy,
10203             ipst->ips_netstack);
10204         if (ph == NULL)
10205                 goto enomem;
10206 
10207         ipsec_actvec_from_req(req, &actp, &nact, ipst->ips_netstack);
10208         if (actp == NULL)
10209                 goto enomem;
10210 
10211         /*
10212          * Always allocate IPv4 policy entries, since they can also
10213          * apply to ipv6 sockets being used in ipv4-compat mode.
10214          */
10215         bzero(&sel, sizeof (sel));
10216         sel.ipsl_valid = IPSL_IPV4;
10217 
10218         pin4 = ipsec_policy_create(&sel, actp, nact, IPSEC_PRIO_SOCKET, NULL,
10219             ipst->ips_netstack);
10220         if (pin4 == NULL)
10221                 goto enomem;
10222 
10223         pout4 = ipsec_policy_create(&sel, actp, nact, IPSEC_PRIO_SOCKET, NULL,
10224             ipst->ips_netstack);
10225         if (pout4 == NULL)
10226                 goto enomem;
10227 
10228         if (connp->conn_af_isv6) {
10229                 /*
10230                  * We're looking at a v6 socket, also allocate the
10231                  * v6-specific entries...
10232                  */
10233                 sel.ipsl_valid = IPSL_IPV6;
10234                 pin6 = ipsec_policy_create(&sel, actp, nact,
10235                     IPSEC_PRIO_SOCKET, NULL, ipst->ips_netstack);
10236                 if (pin6 == NULL)
10237                         goto enomem;
10238 
10239                 pout6 = ipsec_policy_create(&sel, actp, nact,
10240                     IPSEC_PRIO_SOCKET, NULL, ipst->ips_netstack);
10241                 if (pout6 == NULL)
10242                         goto enomem;
10243 
10244                 /*
10245                  * .. and file them away in the right place.
10246                  */
10247                 fam = IPSEC_AF_V6;
10248                 pr = &ph->iph_root[IPSEC_TYPE_INBOUND];
10249                 HASHLIST_INSERT(pin6, ipsp_hash, pr->ipr_nonhash[fam]);
10250                 ipsec_insert_always(&ph->iph_rulebyid, pin6);
10251                 pr = &ph->iph_root[IPSEC_TYPE_OUTBOUND];
10252                 HASHLIST_INSERT(pout6, ipsp_hash, pr->ipr_nonhash[fam]);
10253                 ipsec_insert_always(&ph->iph_rulebyid, pout6);
10254         }
10255 
10256         ipsec_actvec_free(actp, nact);
10257 
10258         /*
10259          * File the v4 policies.
10260          */
10261         fam = IPSEC_AF_V4;
10262         pr = &ph->iph_root[IPSEC_TYPE_INBOUND];
10263         HASHLIST_INSERT(pin4, ipsp_hash, pr->ipr_nonhash[fam]);
10264         ipsec_insert_always(&ph->iph_rulebyid, pin4);
10265 
10266         pr = &ph->iph_root[IPSEC_TYPE_OUTBOUND];
10267         HASHLIST_INSERT(pout4, ipsp_hash, pr->ipr_nonhash[fam]);
10268         ipsec_insert_always(&ph->iph_rulebyid, pout4);
10269 
10270         /*
10271          * If the requests need security, set enforce_policy.
10272          * If the requests are IPSEC_PREF_NEVER, one should
10273          * still set conn_out_enforce_policy so that an ipsec_out
10274          * gets attached in ip_wput. This is needed so that
10275          * for connections that we don't cache policy in ip_bind,
10276          * if global policy matches in ip_wput_attach_policy, we
10277          * don't wrongly inherit global policy. Similarly, we need
10278          * to set conn_in_enforce_policy also so that we don't verify
10279          * policy wrongly.
10280          */
10281         if ((ah_req & REQ_MASK) != 0 ||
10282             (esp_req & REQ_MASK) != 0 ||
10283             (se_req & REQ_MASK) != 0) {
10284                 connp->conn_in_enforce_policy = B_TRUE;
10285                 connp->conn_out_enforce_policy = B_TRUE;
10286                 connp->conn_flags |= IPCL_CHECK_POLICY;
10287         }
10288 
10289         mutex_exit(&connp->conn_lock);
10290         return (error);
10291 #undef REQ_MASK
10292 
10293         /*
10294          * Common memory-allocation-failure exit path.
10295          */
10296 enomem:
10297         mutex_exit(&connp->conn_lock);
10298         if (actp != NULL)
10299                 ipsec_actvec_free(actp, nact);
10300         if (pin4 != NULL)
10301                 IPPOL_REFRELE(pin4, ipst->ips_netstack);
10302         if (pout4 != NULL)
10303                 IPPOL_REFRELE(pout4, ipst->ips_netstack);
10304         if (pin6 != NULL)
10305                 IPPOL_REFRELE(pin6, ipst->ips_netstack);
10306         if (pout6 != NULL)
10307                 IPPOL_REFRELE(pout6, ipst->ips_netstack);
10308         return (ENOMEM);
10309 }
10310 
10311 /*
10312  * Only for options that pass in an IP addr. Currently only V4 options
10313  * pass in an ipif. V6 options always pass an ifindex specifying the ill.
10314  * So this function assumes level is IPPROTO_IP
10315  */
10316 int
10317 ip_opt_set_ipif(conn_t *connp, ipaddr_t addr, boolean_t checkonly, int option,
10318     mblk_t *first_mp)
10319 {
10320         ipif_t *ipif = NULL;
10321         int error;
10322         ill_t *ill;
10323         int zoneid;
10324         ip_stack_t *ipst = connp->conn_netstack->netstack_ip;
10325 
10326         ip2dbg(("ip_opt_set_ipif: ipaddr %X\n", addr));
10327 
10328         if (addr != INADDR_ANY || checkonly) {
10329                 ASSERT(connp != NULL);
10330                 zoneid = IPCL_ZONEID(connp);
10331                 if (option == IP_NEXTHOP) {
10332                         ipif = ipif_lookup_onlink_addr(addr,
10333                             connp->conn_zoneid, ipst);
10334                 } else {
10335                         ipif = ipif_lookup_addr(addr, NULL, zoneid,
10336                             CONNP_TO_WQ(connp), first_mp, ip_restart_optmgmt,
10337                             &error, ipst);
10338                 }
10339                 if (ipif == NULL) {
10340                         if (error == EINPROGRESS)
10341                                 return (error);
10342                         else if ((option == IP_MULTICAST_IF) ||
10343                             (option == IP_NEXTHOP))
10344                                 return (EHOSTUNREACH);
10345                         else
10346                                 return (EINVAL);
10347                 } else if (checkonly) {
10348                         if (option == IP_MULTICAST_IF) {
10349                                 ill = ipif->ipif_ill;
10350                                 /* not supported by the virtual network iface */
10351                                 if (IS_VNI(ill)) {
10352                                         ipif_refrele(ipif);
10353                                         return (EINVAL);
10354                                 }
10355                         }
10356                         ipif_refrele(ipif);
10357                         return (0);
10358                 }
10359                 ill = ipif->ipif_ill;
10360                 mutex_enter(&connp->conn_lock);
10361                 mutex_enter(&ill->ill_lock);
10362                 if ((ill->ill_state_flags & ILL_CONDEMNED) ||
10363                     (ipif->ipif_state_flags & IPIF_CONDEMNED)) {
10364                         mutex_exit(&ill->ill_lock);
10365                         mutex_exit(&connp->conn_lock);
10366                         ipif_refrele(ipif);
10367                         return (option == IP_MULTICAST_IF ?
10368                             EHOSTUNREACH : EINVAL);
10369                 }
10370         } else {
10371                 mutex_enter(&connp->conn_lock);
10372         }
10373 
10374         /* None of the options below are supported on the VNI */
10375         if (ipif != NULL && IS_VNI(ipif->ipif_ill)) {
10376                 mutex_exit(&ill->ill_lock);
10377                 mutex_exit(&connp->conn_lock);
10378                 ipif_refrele(ipif);
10379                 return (EINVAL);
10380         }
10381 
10382         switch (option) {
10383         case IP_DONTFAILOVER_IF:
10384                 /*
10385                  * This option is used by in.mpathd to ensure
10386                  * that IPMP probe packets only go out on the
10387                  * test interfaces. in.mpathd sets this option
10388                  * on the non-failover interfaces.
10389                  * For backward compatibility, this option
10390                  * implicitly sets IP_MULTICAST_IF, as used
10391                  * be done in bind(), so that ip_wput gets
10392                  * this ipif to send mcast packets.
10393                  */
10394                 if (ipif != NULL) {
10395                         ASSERT(addr != INADDR_ANY);
10396                         connp->conn_nofailover_ill = ipif->ipif_ill;
10397                         connp->conn_multicast_ipif = ipif;
10398                 } else {
10399                         ASSERT(addr == INADDR_ANY);
10400                         connp->conn_nofailover_ill = NULL;
10401                         connp->conn_multicast_ipif = NULL;
10402                 }
10403                 break;
10404 
10405         case IP_MULTICAST_IF:
10406                 connp->conn_multicast_ipif = ipif;
10407                 break;
10408         case IP_NEXTHOP:
10409                 connp->conn_nexthop_v4 = addr;
10410                 connp->conn_nexthop_set = B_TRUE;
10411                 break;
10412         }
10413 
10414         if (ipif != NULL) {
10415                 mutex_exit(&ill->ill_lock);
10416                 mutex_exit(&connp->conn_lock);
10417                 ipif_refrele(ipif);
10418                 return (0);
10419         }
10420         mutex_exit(&connp->conn_lock);
10421         /* We succeded in cleared the option */
10422         return (0);
10423 }
10424 
10425 /*
10426  * For options that pass in an ifindex specifying the ill. V6 options always
10427  * pass in an ill. Some v4 options also pass in ifindex specifying the ill.
10428  */
10429 int
10430 ip_opt_set_ill(conn_t *connp, int ifindex, boolean_t isv6, boolean_t checkonly,
10431     int level, int option, mblk_t *first_mp)
10432 {
10433         ill_t *ill = NULL;
10434         int error = 0;
10435         ip_stack_t      *ipst = connp->conn_netstack->netstack_ip;
10436 
10437         ip2dbg(("ip_opt_set_ill: ifindex %d\n", ifindex));
10438         if (ifindex != 0) {
10439                 ASSERT(connp != NULL);
10440                 ill = ill_lookup_on_ifindex(ifindex, isv6, CONNP_TO_WQ(connp),
10441                     first_mp, ip_restart_optmgmt, &error, ipst);
10442                 if (ill != NULL) {
10443                         if (checkonly) {
10444                                 /* not supported by the virtual network iface */
10445                                 if (IS_VNI(ill)) {
10446                                         ill_refrele(ill);
10447                                         return (EINVAL);
10448                                 }
10449                                 ill_refrele(ill);
10450                                 return (0);
10451                         }
10452                         if (!ipif_lookup_zoneid_group(ill, connp->conn_zoneid,
10453                             0, NULL)) {
10454                                 ill_refrele(ill);
10455                                 ill = NULL;
10456                                 mutex_enter(&connp->conn_lock);
10457                                 goto setit;
10458                         }
10459                         mutex_enter(&connp->conn_lock);
10460                         mutex_enter(&ill->ill_lock);
10461                         if (ill->ill_state_flags & ILL_CONDEMNED) {
10462                                 mutex_exit(&ill->ill_lock);
10463                                 mutex_exit(&connp->conn_lock);
10464                                 ill_refrele(ill);
10465                                 ill = NULL;
10466                                 mutex_enter(&connp->conn_lock);
10467                         }
10468                         goto setit;
10469                 } else if (error == EINPROGRESS) {
10470                         return (error);
10471                 } else {
10472                         error = 0;
10473                 }
10474         }
10475         mutex_enter(&connp->conn_lock);
10476 setit:
10477         ASSERT((level == IPPROTO_IP || level == IPPROTO_IPV6));
10478 
10479         /*
10480          * The options below assume that the ILL (if any) transmits and/or
10481          * receives traffic. Neither of which is true for the virtual network
10482          * interface, so fail setting these on a VNI.
10483          */
10484         if (IS_VNI(ill)) {
10485                 ASSERT(ill != NULL);
10486                 mutex_exit(&ill->ill_lock);
10487                 mutex_exit(&connp->conn_lock);
10488                 ill_refrele(ill);
10489                 return (EINVAL);
10490         }
10491 
10492         if (level == IPPROTO_IP) {
10493                 switch (option) {
10494                 case IP_BOUND_IF:
10495                         connp->conn_incoming_ill = ill;
10496                         connp->conn_outgoing_ill = ill;
10497                         connp->conn_orig_bound_ifindex = (ill == NULL) ?
10498                             0 : ifindex;
10499                         break;
10500 
10501                 case IP_MULTICAST_IF:
10502                         /*
10503                          * This option is an internal special. The socket
10504                          * level IP_MULTICAST_IF specifies an 'ipaddr' and
10505                          * is handled in ip_opt_set_ipif. IPV6_MULTICAST_IF
10506                          * specifies an ifindex and we try first on V6 ill's.
10507                          * If we don't find one, we they try using on v4 ill's
10508                          * intenally and we come here.
10509                          */
10510                         if (!checkonly && ill != NULL) {
10511                                 ipif_t  *ipif;
10512                                 ipif = ill->ill_ipif;
10513 
10514                                 if (ipif->ipif_state_flags & IPIF_CONDEMNED) {
10515                                         mutex_exit(&ill->ill_lock);
10516                                         mutex_exit(&connp->conn_lock);
10517                                         ill_refrele(ill);
10518                                         ill = NULL;
10519                                         mutex_enter(&connp->conn_lock);
10520                                 } else {
10521                                         connp->conn_multicast_ipif = ipif;
10522                                 }
10523                         }
10524                         break;
10525 
10526                 case IP_DHCPINIT_IF:
10527                         if (connp->conn_dhcpinit_ill != NULL) {
10528                                 /*
10529                                  * We've locked the conn so conn_cleanup_ill()
10530                                  * cannot clear conn_dhcpinit_ill -- so it's
10531                                  * safe to access the ill.
10532                                  */
10533                                 ill_t *oill = connp->conn_dhcpinit_ill;
10534 
10535                                 ASSERT(oill->ill_dhcpinit != 0);
10536                                 atomic_dec_32(&oill->ill_dhcpinit);
10537                                 connp->conn_dhcpinit_ill = NULL;
10538                         }
10539 
10540                         if (ill != NULL) {
10541                                 connp->conn_dhcpinit_ill = ill;
10542                                 atomic_inc_32(&ill->ill_dhcpinit);
10543                         }
10544                         break;
10545                 }
10546         } else {
10547                 switch (option) {
10548                 case IPV6_BOUND_IF:
10549                         connp->conn_incoming_ill = ill;
10550                         connp->conn_outgoing_ill = ill;
10551                         connp->conn_orig_bound_ifindex = (ill == NULL) ?
10552                             0 : ifindex;
10553                         break;
10554 
10555                 case IPV6_BOUND_PIF:
10556                         /*
10557                          * Limit all transmit to this ill.
10558                          * Unlike IPV6_BOUND_IF, using this option
10559                          * prevents load spreading and failover from
10560                          * happening when the interface is part of the
10561                          * group. That's why we don't need to remember
10562                          * the ifindex in orig_bound_ifindex as in
10563                          * IPV6_BOUND_IF.
10564                          */
10565                         connp->conn_outgoing_pill = ill;
10566                         break;
10567 
10568                 case IPV6_DONTFAILOVER_IF:
10569                         /*
10570                          * This option is used by in.mpathd to ensure
10571                          * that IPMP probe packets only go out on the
10572                          * test interfaces. in.mpathd sets this option
10573                          * on the non-failover interfaces.
10574                          */
10575                         connp->conn_nofailover_ill = ill;
10576                         /*
10577                          * For backward compatibility, this option
10578                          * implicitly sets ip_multicast_ill as used in
10579                          * IPV6_MULTICAST_IF so that ip_wput gets
10580                          * this ill to send mcast packets.
10581                          */
10582                         connp->conn_multicast_ill = ill;
10583                         connp->conn_orig_multicast_ifindex = (ill == NULL) ?
10584                             0 : ifindex;
10585                         break;
10586 
10587                 case IPV6_MULTICAST_IF:
10588                         /*
10589                          * Set conn_multicast_ill to be the IPv6 ill.
10590                          * Set conn_multicast_ipif to be an IPv4 ipif
10591                          * for ifindex to make IPv4 mapped addresses
10592                          * on PF_INET6 sockets honor IPV6_MULTICAST_IF.
10593                          * Even if no IPv6 ill exists for the ifindex
10594                          * we need to check for an IPv4 ifindex in order
10595                          * for this to work with mapped addresses. In that
10596                          * case only set conn_multicast_ipif.
10597                          */
10598                         if (!checkonly) {
10599                                 if (ifindex == 0) {
10600                                         connp->conn_multicast_ill = NULL;
10601                                         connp->conn_orig_multicast_ifindex = 0;
10602                                         connp->conn_multicast_ipif = NULL;
10603                                 } else if (ill != NULL) {
10604                                         connp->conn_multicast_ill = ill;
10605                                         connp->conn_orig_multicast_ifindex =
10606                                             ifindex;
10607                                 }
10608                         }
10609                         break;
10610                 }
10611         }
10612 
10613         if (ill != NULL) {
10614                 mutex_exit(&ill->ill_lock);
10615                 mutex_exit(&connp->conn_lock);
10616                 ill_refrele(ill);
10617                 return (0);
10618         }
10619         mutex_exit(&connp->conn_lock);
10620         /*
10621          * We succeeded in clearing the option (ifindex == 0) or failed to
10622          * locate the ill and could not set the option (ifindex != 0)
10623          */
10624         return (ifindex == 0 ? 0 : EINVAL);
10625 }
10626 
10627 /* This routine sets socket options. */
10628 /* ARGSUSED */
10629 int
10630 ip_opt_set(queue_t *q, uint_t optset_context, int level, int name,
10631     uint_t inlen, uchar_t *invalp, uint_t *outlenp, uchar_t *outvalp,
10632     void *dummy, cred_t *cr, mblk_t *first_mp)
10633 {
10634         int             *i1 = (int *)invalp;
10635         conn_t          *connp = Q_TO_CONN(q);
10636         int             error = 0;
10637         boolean_t       checkonly;
10638         ire_t           *ire;
10639         boolean_t       found;
10640         ip_stack_t      *ipst = connp->conn_netstack->netstack_ip;
10641 
10642         switch (optset_context) {
10643 
10644         case SETFN_OPTCOM_CHECKONLY:
10645                 checkonly = B_TRUE;
10646                 /*
10647                  * Note: Implies T_CHECK semantics for T_OPTCOM_REQ
10648                  * inlen != 0 implies value supplied and
10649                  *      we have to "pretend" to set it.
10650                  * inlen == 0 implies that there is no
10651                  *      value part in T_CHECK request and just validation
10652                  * done elsewhere should be enough, we just return here.
10653                  */
10654                 if (inlen == 0) {
10655                         *outlenp = 0;
10656                         return (0);
10657                 }
10658                 break;
10659         case SETFN_OPTCOM_NEGOTIATE:
10660         case SETFN_UD_NEGOTIATE:
10661         case SETFN_CONN_NEGOTIATE:
10662                 checkonly = B_FALSE;
10663                 break;
10664         default:
10665                 /*
10666                  * We should never get here
10667                  */
10668                 *outlenp = 0;
10669                 return (EINVAL);
10670         }
10671 
10672         ASSERT((optset_context != SETFN_OPTCOM_CHECKONLY) ||
10673             (optset_context == SETFN_OPTCOM_CHECKONLY && inlen != 0));
10674 
10675         /*
10676          * For fixed length options, no sanity check
10677          * of passed in length is done. It is assumed *_optcom_req()
10678          * routines do the right thing.
10679          */
10680 
10681         switch (level) {
10682         case SOL_SOCKET:
10683                 /*
10684                  * conn_lock protects the bitfields, and is used to
10685                  * set the fields atomically.
10686                  */
10687                 switch (name) {
10688                 case SO_BROADCAST:
10689                         if (!checkonly) {
10690                                 /* TODO: use value someplace? */
10691                                 mutex_enter(&connp->conn_lock);
10692                                 connp->conn_broadcast = *i1 ? 1 : 0;
10693                                 mutex_exit(&connp->conn_lock);
10694                         }
10695                         break;  /* goto sizeof (int) option return */
10696                 case SO_USELOOPBACK:
10697                         if (!checkonly) {
10698                                 /* TODO: use value someplace? */
10699                                 mutex_enter(&connp->conn_lock);
10700                                 connp->conn_loopback = *i1 ? 1 : 0;
10701                                 mutex_exit(&connp->conn_lock);
10702                         }
10703                         break;  /* goto sizeof (int) option return */
10704                 case SO_DONTROUTE:
10705                         if (!checkonly) {
10706                                 mutex_enter(&connp->conn_lock);
10707                                 connp->conn_dontroute = *i1 ? 1 : 0;
10708                                 mutex_exit(&connp->conn_lock);
10709                         }
10710                         break;  /* goto sizeof (int) option return */
10711                 case SO_REUSEADDR:
10712                         if (!checkonly) {
10713                                 mutex_enter(&connp->conn_lock);
10714                                 connp->conn_reuseaddr = *i1 ? 1 : 0;
10715                                 mutex_exit(&connp->conn_lock);
10716                         }
10717                         break;  /* goto sizeof (int) option return */
10718                 case SO_PROTOTYPE:
10719                         if (!checkonly) {
10720                                 mutex_enter(&connp->conn_lock);
10721                                 connp->conn_proto = *i1;
10722                                 mutex_exit(&connp->conn_lock);
10723                         }
10724                         break;  /* goto sizeof (int) option return */
10725                 case SO_ALLZONES:
10726                         if (!checkonly) {
10727                                 mutex_enter(&connp->conn_lock);
10728                                 if (IPCL_IS_BOUND(connp)) {
10729                                         mutex_exit(&connp->conn_lock);
10730                                         return (EINVAL);
10731                                 }
10732                                 connp->conn_allzones = *i1 != 0 ? 1 : 0;
10733                                 mutex_exit(&connp->conn_lock);
10734                         }
10735                         break;  /* goto sizeof (int) option return */
10736                 case SO_ANON_MLP:
10737                         if (!checkonly) {
10738                                 mutex_enter(&connp->conn_lock);
10739                                 connp->conn_anon_mlp = *i1 != 0 ? 1 : 0;
10740                                 mutex_exit(&connp->conn_lock);
10741                         }
10742                         break;  /* goto sizeof (int) option return */
10743                 case SO_MAC_EXEMPT:
10744                         if (secpolicy_net_mac_aware(cr) != 0 ||
10745                             IPCL_IS_BOUND(connp))
10746                                 return (EACCES);
10747                         if (!checkonly) {
10748                                 mutex_enter(&connp->conn_lock);
10749                                 connp->conn_mac_exempt = *i1 != 0 ? 1 : 0;
10750                                 mutex_exit(&connp->conn_lock);
10751                         }
10752