1 /*
   2  * CDDL HEADER START
   3  *
   4  * The contents of this file are subject to the terms of the
   5  * Common Development and Distribution License (the "License").
   6  * You may not use this file except in compliance with the License.
   7  *
   8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
   9  * or http://www.opensolaris.org/os/licensing.
  10  * See the License for the specific language governing permissions
  11  * and limitations under the License.
  12  *
  13  * When distributing Covered Code, include this CDDL HEADER in each
  14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
  15  * If applicable, add the following below this CDDL HEADER, with the
  16  * fields enclosed by brackets "[]" replaced with your own identifying
  17  * information: Portions Copyright [yyyy] [name of copyright owner]
  18  *
  19  * CDDL HEADER END
  20  */
  21 
  22 /*
  23  * Copyright (c) 1991, 2010, Oracle and/or its affiliates. All rights reserved.
  24  * Copyright (c) 1990 Mentat Inc.
  25  * Copyright (c) 2011 Joyent, Inc. All rights reserved.
  26  */
  27 
  28 #include <sys/types.h>
  29 #include <sys/stream.h>
  30 #include <sys/dlpi.h>
  31 #include <sys/stropts.h>
  32 #include <sys/sysmacros.h>
  33 #include <sys/strsubr.h>
  34 #include <sys/strlog.h>
  35 #include <sys/strsun.h>
  36 #include <sys/zone.h>
  37 #define _SUN_TPI_VERSION 2
  38 #include <sys/tihdr.h>
  39 #include <sys/xti_inet.h>
  40 #include <sys/ddi.h>
  41 #include <sys/suntpi.h>
  42 #include <sys/cmn_err.h>
  43 #include <sys/debug.h>
  44 #include <sys/kobj.h>
  45 #include <sys/modctl.h>
  46 #include <sys/atomic.h>
  47 #include <sys/policy.h>
  48 #include <sys/priv.h>
  49 #include <sys/taskq.h>
  50 
  51 #include <sys/systm.h>
  52 #include <sys/param.h>
  53 #include <sys/kmem.h>
  54 #include <sys/sdt.h>
  55 #include <sys/socket.h>
  56 #include <sys/vtrace.h>
  57 #include <sys/isa_defs.h>
  58 #include <sys/mac.h>
  59 #include <net/if.h>
  60 #include <net/if_arp.h>
  61 #include <net/route.h>
  62 #include <sys/sockio.h>
  63 #include <netinet/in.h>
  64 #include <net/if_dl.h>
  65 
  66 #include <inet/common.h>
  67 #include <inet/mi.h>
  68 #include <inet/mib2.h>
  69 #include <inet/nd.h>
  70 #include <inet/arp.h>
  71 #include <inet/snmpcom.h>
  72 #include <inet/optcom.h>
  73 #include <inet/kstatcom.h>
  74 
  75 #include <netinet/igmp_var.h>
  76 #include <netinet/ip6.h>
  77 #include <netinet/icmp6.h>
  78 #include <netinet/sctp.h>
  79 
  80 #include <inet/ip.h>
  81 #include <inet/ip_impl.h>
  82 #include <inet/ip6.h>
  83 #include <inet/ip6_asp.h>
  84 #include <inet/tcp.h>
  85 #include <inet/tcp_impl.h>
  86 #include <inet/ip_multi.h>
  87 #include <inet/ip_if.h>
  88 #include <inet/ip_ire.h>
  89 #include <inet/ip_ftable.h>
  90 #include <inet/ip_rts.h>
  91 #include <inet/ip_ndp.h>
  92 #include <inet/ip_listutils.h>
  93 #include <netinet/igmp.h>
  94 #include <netinet/ip_mroute.h>
  95 #include <inet/ipp_common.h>
  96 
  97 #include <net/pfkeyv2.h>
  98 #include <inet/sadb.h>
  99 #include <inet/ipsec_impl.h>
 100 #include <inet/iptun/iptun_impl.h>
 101 #include <inet/ipdrop.h>
 102 #include <inet/ip_netinfo.h>
 103 #include <inet/ilb_ip.h>
 104 
 105 #include <sys/ethernet.h>
 106 #include <net/if_types.h>
 107 #include <sys/cpuvar.h>
 108 
 109 #include <ipp/ipp.h>
 110 #include <ipp/ipp_impl.h>
 111 #include <ipp/ipgpc/ipgpc.h>
 112 
 113 #include <sys/pattr.h>
 114 #include <inet/ipclassifier.h>
 115 #include <inet/sctp_ip.h>
 116 #include <inet/sctp/sctp_impl.h>
 117 #include <inet/udp_impl.h>
 118 #include <inet/rawip_impl.h>
 119 #include <inet/rts_impl.h>
 120 
 121 #include <sys/tsol/label.h>
 122 #include <sys/tsol/tnet.h>
 123 
 124 #include <sys/squeue_impl.h>
 125 #include <inet/ip_arp.h>
 126 
 127 #include <sys/clock_impl.h>       /* For LBOLT_FASTPATH{,64} */
 128 
 129 /*
 130  * Values for squeue switch:
 131  * IP_SQUEUE_ENTER_NODRAIN: SQ_NODRAIN
 132  * IP_SQUEUE_ENTER: SQ_PROCESS
 133  * IP_SQUEUE_FILL: SQ_FILL
 134  */
 135 int ip_squeue_enter = IP_SQUEUE_ENTER;  /* Setable in /etc/system */
 136 
 137 int ip_squeue_flag;
 138 
 139 /*
 140  * Setable in /etc/system
 141  */
 142 int ip_poll_normal_ms = 100;
 143 int ip_poll_normal_ticks = 0;
 144 int ip_modclose_ackwait_ms = 3000;
 145 
 146 /*
 147  * It would be nice to have these present only in DEBUG systems, but the
 148  * current design of the global symbol checking logic requires them to be
 149  * unconditionally present.
 150  */
 151 uint_t ip_thread_data;                  /* TSD key for debug support */
 152 krwlock_t ip_thread_rwlock;
 153 list_t  ip_thread_list;
 154 
 155 /*
 156  * Structure to represent a linked list of msgblks. Used by ip_snmp_ functions.
 157  */
 158 
 159 struct listptr_s {
 160         mblk_t  *lp_head;       /* pointer to the head of the list */
 161         mblk_t  *lp_tail;       /* pointer to the tail of the list */
 162 };
 163 
 164 typedef struct listptr_s listptr_t;
 165 
 166 /*
 167  * This is used by ip_snmp_get_mib2_ip_route_media and
 168  * ip_snmp_get_mib2_ip6_route_media to carry the lists of return data.
 169  */
 170 typedef struct iproutedata_s {
 171         uint_t          ird_idx;
 172         uint_t          ird_flags;      /* see below */
 173         listptr_t       ird_route;      /* ipRouteEntryTable */
 174         listptr_t       ird_netmedia;   /* ipNetToMediaEntryTable */
 175         listptr_t       ird_attrs;      /* ipRouteAttributeTable */
 176 } iproutedata_t;
 177 
 178 /* Include ire_testhidden and IRE_IF_CLONE routes */
 179 #define IRD_REPORT_ALL  0x01
 180 
 181 /*
 182  * Cluster specific hooks. These should be NULL when booted as a non-cluster
 183  */
 184 
 185 /*
 186  * Hook functions to enable cluster networking
 187  * On non-clustered systems these vectors must always be NULL.
 188  *
 189  * Hook function to Check ip specified ip address is a shared ip address
 190  * in the cluster
 191  *
 192  */
 193 int (*cl_inet_isclusterwide)(netstackid_t stack_id, uint8_t protocol,
 194     sa_family_t addr_family, uint8_t *laddrp, void *args) = NULL;
 195 
 196 /*
 197  * Hook function to generate cluster wide ip fragment identifier
 198  */
 199 uint32_t (*cl_inet_ipident)(netstackid_t stack_id, uint8_t protocol,
 200     sa_family_t addr_family, uint8_t *laddrp, uint8_t *faddrp,
 201     void *args) = NULL;
 202 
 203 /*
 204  * Hook function to generate cluster wide SPI.
 205  */
 206 void (*cl_inet_getspi)(netstackid_t, uint8_t, uint8_t *, size_t,
 207     void *) = NULL;
 208 
 209 /*
 210  * Hook function to verify if the SPI is already utlized.
 211  */
 212 
 213 int (*cl_inet_checkspi)(netstackid_t, uint8_t, uint32_t, void *) = NULL;
 214 
 215 /*
 216  * Hook function to delete the SPI from the cluster wide repository.
 217  */
 218 
 219 void (*cl_inet_deletespi)(netstackid_t, uint8_t, uint32_t, void *) = NULL;
 220 
 221 /*
 222  * Hook function to inform the cluster when packet received on an IDLE SA
 223  */
 224 
 225 void (*cl_inet_idlesa)(netstackid_t, uint8_t, uint32_t, sa_family_t,
 226     in6_addr_t, in6_addr_t, void *) = NULL;
 227 
 228 /*
 229  * Synchronization notes:
 230  *
 231  * IP is a fully D_MP STREAMS module/driver. Thus it does not depend on any
 232  * MT level protection given by STREAMS. IP uses a combination of its own
 233  * internal serialization mechanism and standard Solaris locking techniques.
 234  * The internal serialization is per phyint.  This is used to serialize
 235  * plumbing operations, IPMP operations, most set ioctls, etc.
 236  *
 237  * Plumbing is a long sequence of operations involving message
 238  * exchanges between IP, ARP and device drivers. Many set ioctls are typically
 239  * involved in plumbing operations. A natural model is to serialize these
 240  * ioctls one per ill. For example plumbing of hme0 and qfe0 can go on in
 241  * parallel without any interference. But various set ioctls on hme0 are best
 242  * serialized, along with IPMP operations and processing of DLPI control
 243  * messages received from drivers on a per phyint basis. This serialization is
 244  * provided by the ipsq_t and primitives operating on this. Details can
 245  * be found in ip_if.c above the core primitives operating on ipsq_t.
 246  *
 247  * Lookups of an ipif or ill by a thread return a refheld ipif / ill.
 248  * Simiarly lookup of an ire by a thread also returns a refheld ire.
 249  * In addition ipif's and ill's referenced by the ire are also indirectly
 250  * refheld. Thus no ipif or ill can vanish as long as an ipif is refheld
 251  * directly or indirectly. For example an SIOCSLIFADDR ioctl that changes the
 252  * address of an ipif has to go through the ipsq_t. This ensures that only
 253  * one such exclusive operation proceeds at any time on the ipif. It then
 254  * waits for all refcnts
 255  * associated with this ipif to come down to zero. The address is changed
 256  * only after the ipif has been quiesced. Then the ipif is brought up again.
 257  * More details are described above the comment in ip_sioctl_flags.
 258  *
 259  * Packet processing is based mostly on IREs and are fully multi-threaded
 260  * using standard Solaris MT techniques.
 261  *
 262  * There are explicit locks in IP to handle:
 263  * - The ip_g_head list maintained by mi_open_link() and friends.
 264  *
 265  * - The reassembly data structures (one lock per hash bucket)
 266  *
 267  * - conn_lock is meant to protect conn_t fields. The fields actually
 268  *   protected by conn_lock are documented in the conn_t definition.
 269  *
 270  * - ire_lock to protect some of the fields of the ire, IRE tables
 271  *   (one lock per hash bucket). Refer to ip_ire.c for details.
 272  *
 273  * - ndp_g_lock and ncec_lock for protecting NCEs.
 274  *
 275  * - ill_lock protects fields of the ill and ipif. Details in ip.h
 276  *
 277  * - ill_g_lock: This is a global reader/writer lock. Protects the following
 278  *      * The AVL tree based global multi list of all ills.
 279  *      * The linked list of all ipifs of an ill
 280  *      * The <ipsq-xop> mapping
 281  *      * <ill-phyint> association
 282  *   Insertion/deletion of an ill in the system, insertion/deletion of an ipif
 283  *   into an ill, changing the <ipsq-xop> mapping of an ill, changing the
 284  *   <ill-phyint> assoc of an ill will all have to hold the ill_g_lock as
 285  *   writer for the actual duration of the insertion/deletion/change.
 286  *
 287  * - ill_lock:  This is a per ill mutex.
 288  *   It protects some members of the ill_t struct; see ip.h for details.
 289  *   It also protects the <ill-phyint> assoc.
 290  *   It also protects the list of ipifs hanging off the ill.
 291  *
 292  * - ipsq_lock: This is a per ipsq_t mutex lock.
 293  *   This protects some members of the ipsq_t struct; see ip.h for details.
 294  *   It also protects the <ipsq-ipxop> mapping
 295  *
 296  * - ipx_lock: This is a per ipxop_t mutex lock.
 297  *   This protects some members of the ipxop_t struct; see ip.h for details.
 298  *
 299  * - phyint_lock: This is a per phyint mutex lock. Protects just the
 300  *   phyint_flags
 301  *
 302  * - ip_addr_avail_lock: This is used to ensure the uniqueness of IP addresses.
 303  *   This lock is held in ipif_up_done and the ipif is marked IPIF_UP and the
 304  *   uniqueness check also done atomically.
 305  *
 306  * - ill_g_usesrc_lock: This readers/writer lock protects the usesrc
 307  *   group list linked by ill_usesrc_grp_next. It also protects the
 308  *   ill_usesrc_ifindex field. It is taken as a writer when a member of the
 309  *   group is being added or deleted.  This lock is taken as a reader when
 310  *   walking the list/group(eg: to get the number of members in a usesrc group).
 311  *   Note, it is only necessary to take this lock if the ill_usesrc_grp_next
 312  *   field is changing state i.e from NULL to non-NULL or vice-versa. For
 313  *   example, it is not necessary to take this lock in the initial portion
 314  *   of ip_sioctl_slifusesrc or at all in ip_sioctl_flags since these
 315  *   operations are executed exclusively and that ensures that the "usesrc
 316  *   group state" cannot change. The "usesrc group state" change can happen
 317  *   only in the latter part of ip_sioctl_slifusesrc and in ill_delete.
 318  *
 319  * Changing <ill-phyint>, <ipsq-xop> assocications:
 320  *
 321  * To change the <ill-phyint> association, the ill_g_lock must be held
 322  * as writer, and the ill_locks of both the v4 and v6 instance of the ill
 323  * must be held.
 324  *
 325  * To change the <ipsq-xop> association, the ill_g_lock must be held as
 326  * writer, the ipsq_lock must be held, and one must be writer on the ipsq.
 327  * This is only done when ills are added or removed from IPMP groups.
 328  *
 329  * To add or delete an ipif from the list of ipifs hanging off the ill,
 330  * ill_g_lock (writer) and ill_lock must be held and the thread must be
 331  * a writer on the associated ipsq.
 332  *
 333  * To add or delete an ill to the system, the ill_g_lock must be held as
 334  * writer and the thread must be a writer on the associated ipsq.
 335  *
 336  * To add or delete an ilm to an ill, the ill_lock must be held and the thread
 337  * must be a writer on the associated ipsq.
 338  *
 339  * Lock hierarchy
 340  *
 341  * Some lock hierarchy scenarios are listed below.
 342  *
 343  * ill_g_lock -> conn_lock -> ill_lock -> ipsq_lock -> ipx_lock
 344  * ill_g_lock -> ill_lock(s) -> phyint_lock
 345  * ill_g_lock -> ndp_g_lock -> ill_lock -> ncec_lock
 346  * ill_g_lock -> ip_addr_avail_lock
 347  * conn_lock -> irb_lock -> ill_lock -> ire_lock
 348  * ill_g_lock -> ip_g_nd_lock
 349  * ill_g_lock -> ips_ipmp_lock -> ill_lock -> nce_lock
 350  * ill_g_lock -> ndp_g_lock -> ill_lock -> ncec_lock -> nce_lock
 351  * arl_lock -> ill_lock
 352  * ips_ire_dep_lock -> irb_lock
 353  *
 354  * When more than 1 ill lock is needed to be held, all ill lock addresses
 355  * are sorted on address and locked starting from highest addressed lock
 356  * downward.
 357  *
 358  * Multicast scenarios
 359  * ips_ill_g_lock -> ill_mcast_lock
 360  * conn_ilg_lock -> ips_ill_g_lock -> ill_lock
 361  * ill_mcast_serializer -> ill_mcast_lock -> ips_ipmp_lock -> ill_lock
 362  * ill_mcast_serializer -> ill_mcast_lock -> connf_lock -> conn_lock
 363  * ill_mcast_serializer -> ill_mcast_lock -> conn_ilg_lock
 364  * ill_mcast_serializer -> ill_mcast_lock -> ips_igmp_timer_lock
 365  *
 366  * IPsec scenarios
 367  *
 368  * ipsa_lock -> ill_g_lock -> ill_lock
 369  * ill_g_usesrc_lock -> ill_g_lock -> ill_lock
 370  *
 371  * Trusted Solaris scenarios
 372  *
 373  * igsa_lock -> gcgrp_rwlock -> gcgrp_lock
 374  * igsa_lock -> gcdb_lock
 375  * gcgrp_rwlock -> ire_lock
 376  * gcgrp_rwlock -> gcdb_lock
 377  *
 378  * squeue(sq_lock), flow related (ft_lock, fe_lock) locking
 379  *
 380  * cpu_lock --> ill_lock --> sqset_lock --> sq_lock
 381  * sq_lock -> conn_lock -> QLOCK(q)
 382  * ill_lock -> ft_lock -> fe_lock
 383  *
 384  * Routing/forwarding table locking notes:
 385  *
 386  * Lock acquisition order: Radix tree lock, irb_lock.
 387  * Requirements:
 388  * i.  Walker must not hold any locks during the walker callback.
 389  * ii  Walker must not see a truncated tree during the walk because of any node
 390  *     deletion.
 391  * iii Existing code assumes ire_bucket is valid if it is non-null and is used
 392  *     in many places in the code to walk the irb list. Thus even if all the
 393  *     ires in a bucket have been deleted, we still can't free the radix node
 394  *     until the ires have actually been inactive'd (freed).
 395  *
 396  * Tree traversal - Need to hold the global tree lock in read mode.
 397  * Before dropping the global tree lock, need to either increment the ire_refcnt
 398  * to ensure that the radix node can't be deleted.
 399  *
 400  * Tree add - Need to hold the global tree lock in write mode to add a
 401  * radix node. To prevent the node from being deleted, increment the
 402  * irb_refcnt, after the node is added to the tree. The ire itself is
 403  * added later while holding the irb_lock, but not the tree lock.
 404  *
 405  * Tree delete - Need to hold the global tree lock and irb_lock in write mode.
 406  * All associated ires must be inactive (i.e. freed), and irb_refcnt
 407  * must be zero.
 408  *
 409  * Walker - Increment irb_refcnt before calling the walker callback. Hold the
 410  * global tree lock (read mode) for traversal.
 411  *
 412  * IRE dependencies - In some cases we hold ips_ire_dep_lock across ire_refrele
 413  * hence we will acquire irb_lock while holding ips_ire_dep_lock.
 414  *
 415  * IPsec notes :
 416  *
 417  * IP interacts with the IPsec code (AH/ESP) by storing IPsec attributes
 418  * in the ip_xmit_attr_t ip_recv_attr_t. For outbound datagrams, the
 419  * ip_xmit_attr_t has the
 420  * information used by the IPsec code for applying the right level of
 421  * protection. The information initialized by IP in the ip_xmit_attr_t
 422  * is determined by the per-socket policy or global policy in the system.
 423  * For inbound datagrams, the ip_recv_attr_t
 424  * starts out with nothing in it. It gets filled
 425  * with the right information if it goes through the AH/ESP code, which
 426  * happens if the incoming packet is secure. The information initialized
 427  * by AH/ESP, is later used by IP (during fanouts to ULP) to see whether
 428  * the policy requirements needed by per-socket policy or global policy
 429  * is met or not.
 430  *
 431  * For fully connected sockets i.e dst, src [addr, port] is known,
 432  * conn_policy_cached is set indicating that policy has been cached.
 433  * conn_in_enforce_policy may or may not be set depending on whether
 434  * there is a global policy match or per-socket policy match.
 435  * Policy inheriting happpens in ip_policy_set once the destination is known.
 436  * Once the right policy is set on the conn_t, policy cannot change for
 437  * this socket. This makes life simpler for TCP (UDP ?) where
 438  * re-transmissions go out with the same policy. For symmetry, policy
 439  * is cached for fully connected UDP sockets also. Thus if policy is cached,
 440  * it also implies that policy is latched i.e policy cannot change
 441  * on these sockets. As we have the right policy on the conn, we don't
 442  * have to lookup global policy for every outbound and inbound datagram
 443  * and thus serving as an optimization. Note that a global policy change
 444  * does not affect fully connected sockets if they have policy. If fully
 445  * connected sockets did not have any policy associated with it, global
 446  * policy change may affect them.
 447  *
 448  * IP Flow control notes:
 449  * ---------------------
 450  * Non-TCP streams are flow controlled by IP. The way this is accomplished
 451  * differs when ILL_CAPAB_DLD_DIRECT is enabled for that IP instance. When
 452  * ILL_DIRECT_CAPABLE(ill) is TRUE, IP can do direct function calls into
 453  * GLDv3. Otherwise packets are sent down to lower layers using STREAMS
 454  * functions.
 455  *
 456  * Per Tx ring udp flow control:
 457  * This is applicable only when ILL_CAPAB_DLD_DIRECT capability is set in
 458  * the ill (i.e. ILL_DIRECT_CAPABLE(ill) is true).
 459  *
 460  * The underlying link can expose multiple Tx rings to the GLDv3 mac layer.
 461  * To achieve best performance, outgoing traffic need to be fanned out among
 462  * these Tx ring. mac_tx() is called (via str_mdata_fastpath_put()) to send
 463  * traffic out of the NIC and it takes a fanout hint. UDP connections pass
 464  * the address of connp as fanout hint to mac_tx(). Under flow controlled
 465  * condition, mac_tx() returns a non-NULL cookie (ip_mac_tx_cookie_t). This
 466  * cookie points to a specific Tx ring that is blocked. The cookie is used to
 467  * hash into an idl_tx_list[] entry in idl_tx_list[] array. Each idl_tx_list_t
 468  * point to drain_lists (idl_t's). These drain list will store the blocked UDP
 469  * connp's. The drain list is not a single list but a configurable number of
 470  * lists.
 471  *
 472  * The diagram below shows idl_tx_list_t's and their drain_lists. ip_stack_t
 473  * has an array of idl_tx_list_t. The size of the array is TX_FANOUT_SIZE
 474  * which is equal to 128. This array in turn contains a pointer to idl_t[],
 475  * the ip drain list. The idl_t[] array size is MIN(max_ncpus, 8). The drain
 476  * list will point to the list of connp's that are flow controlled.
 477  *
 478  *                      ---------------   -------   -------   -------
 479  *                   |->|drain_list[0]|-->|connp|-->|connp|-->|connp|-->
 480  *                   |  ---------------   -------   -------   -------
 481  *                   |  ---------------   -------   -------   -------
 482  *                   |->|drain_list[1]|-->|connp|-->|connp|-->|connp|-->
 483  * ----------------  |  ---------------   -------   -------   -------
 484  * |idl_tx_list[0]|->|  ---------------   -------   -------   -------
 485  * ----------------  |->|drain_list[2]|-->|connp|-->|connp|-->|connp|-->
 486  *                   |  ---------------   -------   -------   -------
 487  *                   .        .              .         .         .
 488  *                   |  ---------------   -------   -------   -------
 489  *                   |->|drain_list[n]|-->|connp|-->|connp|-->|connp|-->
 490  *                      ---------------   -------   -------   -------
 491  *                      ---------------   -------   -------   -------
 492  *                   |->|drain_list[0]|-->|connp|-->|connp|-->|connp|-->
 493  *                   |  ---------------   -------   -------   -------
 494  *                   |  ---------------   -------   -------   -------
 495  * ----------------  |->|drain_list[1]|-->|connp|-->|connp|-->|connp|-->
 496  * |idl_tx_list[1]|->|  ---------------   -------   -------   -------
 497  * ----------------  |        .              .         .         .
 498  *                   |  ---------------   -------   -------   -------
 499  *                   |->|drain_list[n]|-->|connp|-->|connp|-->|connp|-->
 500  *                      ---------------   -------   -------   -------
 501  *     .....
 502  * ----------------
 503  * |idl_tx_list[n]|-> ...
 504  * ----------------
 505  *
 506  * When mac_tx() returns a cookie, the cookie is hashed into an index into
 507  * ips_idl_tx_list[], and conn_drain_insert() is called with the idl_tx_list
 508  * to insert the conn onto.  conn_drain_insert() asserts flow control for the
 509  * sockets via su_txq_full() (non-STREAMS) or QFULL on conn_wq (STREAMS).
 510  * Further, conn_blocked is set to indicate that the conn is blocked.
 511  *
 512  * GLDv3 calls ill_flow_enable() when flow control is relieved.  The cookie
 513  * passed in the call to ill_flow_enable() identifies the blocked Tx ring and
 514  * is again hashed to locate the appropriate idl_tx_list, which is then
 515  * drained via conn_walk_drain().  conn_walk_drain() goes through each conn in
 516  * the drain list and calls conn_drain_remove() to clear flow control (via
 517  * calling su_txq_full() or clearing QFULL), and remove the conn from the
 518  * drain list.
 519  *
 520  * Note that the drain list is not a single list but a (configurable) array of
 521  * lists (8 elements by default).  Synchronization between drain insertion and
 522  * flow control wakeup is handled by using idl_txl->txl_lock, and only
 523  * conn_drain_insert() and conn_drain_remove() manipulate the drain list.
 524  *
 525  * Flow control via STREAMS is used when ILL_DIRECT_CAPABLE() returns FALSE.
 526  * On the send side, if the packet cannot be sent down to the driver by IP
 527  * (canput() fails), ip_xmit() drops the packet and returns EWOULDBLOCK to the
 528  * caller, who may then invoke ixa_check_drain_insert() to insert the conn on
 529  * the 0'th drain list.  When ip_wsrv() runs on the ill_wq because flow
 530  * control has been relieved, the blocked conns in the 0'th drain list are
 531  * drained as in the non-STREAMS case.
 532  *
 533  * In both the STREAMS and non-STREAMS cases, the sockfs upcall to set QFULL
 534  * is done when the conn is inserted into the drain list (conn_drain_insert())
 535  * and cleared when the conn is removed from the it (conn_drain_remove()).
 536  *
 537  * IPQOS notes:
 538  *
 539  * IPQoS Policies are applied to packets using IPPF (IP Policy framework)
 540  * and IPQoS modules. IPPF includes hooks in IP at different control points
 541  * (callout positions) which direct packets to IPQoS modules for policy
 542  * processing. Policies, if present, are global.
 543  *
 544  * The callout positions are located in the following paths:
 545  *              o local_in (packets destined for this host)
 546  *              o local_out (packets orginating from this host )
 547  *              o fwd_in  (packets forwarded by this m/c - inbound)
 548  *              o fwd_out (packets forwarded by this m/c - outbound)
 549  * Hooks at these callout points can be enabled/disabled using the ndd variable
 550  * ip_policy_mask (a bit mask with the 4 LSB indicating the callout positions).
 551  * By default all the callout positions are enabled.
 552  *
 553  * Outbound (local_out)
 554  * Hooks are placed in ire_send_wire_v4 and ire_send_wire_v6.
 555  *
 556  * Inbound (local_in)
 557  * Hooks are placed in ip_fanout_v4 and ip_fanout_v6.
 558  *
 559  * Forwarding (in and out)
 560  * Hooks are placed in ire_recv_forward_v4/v6.
 561  *
 562  * IP Policy Framework processing (IPPF processing)
 563  * Policy processing for a packet is initiated by ip_process, which ascertains
 564  * that the classifier (ipgpc) is loaded and configured, failing which the
 565  * packet resumes normal processing in IP. If the clasifier is present, the
 566  * packet is acted upon by one or more IPQoS modules (action instances), per
 567  * filters configured in ipgpc and resumes normal IP processing thereafter.
 568  * An action instance can drop a packet in course of its processing.
 569  *
 570  * Zones notes:
 571  *
 572  * The partitioning rules for networking are as follows:
 573  * 1) Packets coming from a zone must have a source address belonging to that
 574  * zone.
 575  * 2) Packets coming from a zone can only be sent on a physical interface on
 576  * which the zone has an IP address.
 577  * 3) Between two zones on the same machine, packet delivery is only allowed if
 578  * there's a matching route for the destination and zone in the forwarding
 579  * table.
 580  * 4) The TCP and UDP port spaces are per-zone; that is, two processes in
 581  * different zones can bind to the same port with the wildcard address
 582  * (INADDR_ANY).
 583  *
 584  * The granularity of interface partitioning is at the logical interface level.
 585  * Therefore, every zone has its own IP addresses, and incoming packets can be
 586  * attributed to a zone unambiguously. A logical interface is placed into a zone
 587  * using the SIOCSLIFZONE ioctl; this sets the ipif_zoneid field in the ipif_t
 588  * structure. Rule (1) is implemented by modifying the source address selection
 589  * algorithm so that the list of eligible addresses is filtered based on the
 590  * sending process zone.
 591  *
 592  * The Internet Routing Entries (IREs) are either exclusive to a zone or shared
 593  * across all zones, depending on their type. Here is the break-up:
 594  *
 595  * IRE type                             Shared/exclusive
 596  * --------                             ----------------
 597  * IRE_BROADCAST                        Exclusive
 598  * IRE_DEFAULT (default routes)         Shared (*)
 599  * IRE_LOCAL                            Exclusive (x)
 600  * IRE_LOOPBACK                         Exclusive
 601  * IRE_PREFIX (net routes)              Shared (*)
 602  * IRE_IF_NORESOLVER (interface routes) Exclusive
 603  * IRE_IF_RESOLVER (interface routes)   Exclusive
 604  * IRE_IF_CLONE (interface routes)      Exclusive
 605  * IRE_HOST (host routes)               Shared (*)
 606  *
 607  * (*) A zone can only use a default or off-subnet route if the gateway is
 608  * directly reachable from the zone, that is, if the gateway's address matches
 609  * one of the zone's logical interfaces.
 610  *
 611  * (x) IRE_LOCAL are handled a bit differently.
 612  * When ip_restrict_interzone_loopback is set (the default),
 613  * ire_route_recursive restricts loopback using an IRE_LOCAL
 614  * between zone to the case when L2 would have conceptually looped the packet
 615  * back, i.e. the loopback which is required since neither Ethernet drivers
 616  * nor Ethernet hardware loops them back. This is the case when the normal
 617  * routes (ignoring IREs with different zoneids) would send out the packet on
 618  * the same ill as the ill with which is IRE_LOCAL is associated.
 619  *
 620  * Multiple zones can share a common broadcast address; typically all zones
 621  * share the 255.255.255.255 address. Incoming as well as locally originated
 622  * broadcast packets must be dispatched to all the zones on the broadcast
 623  * network. For directed broadcasts (e.g. 10.16.72.255) this is not trivial
 624  * since some zones may not be on the 10.16.72/24 network. To handle this, each
 625  * zone has its own set of IRE_BROADCAST entries; then, broadcast packets are
 626  * sent to every zone that has an IRE_BROADCAST entry for the destination
 627  * address on the input ill, see ip_input_broadcast().
 628  *
 629  * Applications in different zones can join the same multicast group address.
 630  * The same logic applies for multicast as for broadcast. ip_input_multicast
 631  * dispatches packets to all zones that have members on the physical interface.
 632  */
 633 
 634 /*
 635  * Squeue Fanout flags:
 636  *      0: No fanout.
 637  *      1: Fanout across all squeues
 638  */
 639 boolean_t       ip_squeue_fanout = 0;
 640 
 641 /*
 642  * Maximum dups allowed per packet.
 643  */
 644 uint_t ip_max_frag_dups = 10;
 645 
 646 static int      ip_open(queue_t *q, dev_t *devp, int flag, int sflag,
 647                     cred_t *credp, boolean_t isv6);
 648 static mblk_t   *ip_xmit_attach_llhdr(mblk_t *, nce_t *);
 649 
 650 static boolean_t icmp_inbound_verify_v4(mblk_t *, icmph_t *, ip_recv_attr_t *);
 651 static void     icmp_inbound_too_big_v4(icmph_t *, ip_recv_attr_t *);
 652 static void     icmp_inbound_error_fanout_v4(mblk_t *, icmph_t *,
 653     ip_recv_attr_t *);
 654 static void     icmp_options_update(ipha_t *);
 655 static void     icmp_param_problem(mblk_t *, uint8_t,  ip_recv_attr_t *);
 656 static void     icmp_pkt(mblk_t *, void *, size_t, ip_recv_attr_t *);
 657 static mblk_t   *icmp_pkt_err_ok(mblk_t *, ip_recv_attr_t *);
 658 static void     icmp_redirect_v4(mblk_t *mp, ipha_t *, icmph_t *,
 659     ip_recv_attr_t *);
 660 static void     icmp_send_redirect(mblk_t *, ipaddr_t, ip_recv_attr_t *);
 661 static void     icmp_send_reply_v4(mblk_t *, ipha_t *, icmph_t *,
 662     ip_recv_attr_t *);
 663 
 664 mblk_t          *ip_dlpi_alloc(size_t, t_uscalar_t);
 665 char            *ip_dot_addr(ipaddr_t, char *);
 666 mblk_t          *ip_carve_mp(mblk_t **, ssize_t);
 667 int             ip_close(queue_t *, int);
 668 static char     *ip_dot_saddr(uchar_t *, char *);
 669 static void     ip_lrput(queue_t *, mblk_t *);
 670 ipaddr_t        ip_net_mask(ipaddr_t);
 671 char            *ip_nv_lookup(nv_t *, int);
 672 void    ip_rput(queue_t *, mblk_t *);
 673 static void     ip_rput_dlpi_writer(ipsq_t *dummy_sq, queue_t *q, mblk_t *mp,
 674                     void *dummy_arg);
 675 int             ip_snmp_get(queue_t *, mblk_t *, int, boolean_t);
 676 static mblk_t   *ip_snmp_get_mib2_ip(queue_t *, mblk_t *,
 677                     mib2_ipIfStatsEntry_t *, ip_stack_t *, boolean_t);
 678 static mblk_t   *ip_snmp_get_mib2_ip_traffic_stats(queue_t *, mblk_t *,
 679                     ip_stack_t *, boolean_t);
 680 static mblk_t   *ip_snmp_get_mib2_ip6(queue_t *, mblk_t *, ip_stack_t *,
 681                     boolean_t);
 682 static mblk_t   *ip_snmp_get_mib2_icmp(queue_t *, mblk_t *, ip_stack_t *ipst);
 683 static mblk_t   *ip_snmp_get_mib2_icmp6(queue_t *, mblk_t *, ip_stack_t *ipst);
 684 static mblk_t   *ip_snmp_get_mib2_igmp(queue_t *, mblk_t *, ip_stack_t *ipst);
 685 static mblk_t   *ip_snmp_get_mib2_multi(queue_t *, mblk_t *, ip_stack_t *ipst);
 686 static mblk_t   *ip_snmp_get_mib2_ip_addr(queue_t *, mblk_t *,
 687                     ip_stack_t *ipst, boolean_t);
 688 static mblk_t   *ip_snmp_get_mib2_ip6_addr(queue_t *, mblk_t *,
 689                     ip_stack_t *ipst, boolean_t);
 690 static mblk_t   *ip_snmp_get_mib2_ip_group_src(queue_t *, mblk_t *,
 691                     ip_stack_t *ipst);
 692 static mblk_t   *ip_snmp_get_mib2_ip6_group_src(queue_t *, mblk_t *,
 693                     ip_stack_t *ipst);
 694 static mblk_t   *ip_snmp_get_mib2_ip_group_mem(queue_t *, mblk_t *,
 695                     ip_stack_t *ipst);
 696 static mblk_t   *ip_snmp_get_mib2_ip6_group_mem(queue_t *, mblk_t *,
 697                     ip_stack_t *ipst);
 698 static mblk_t   *ip_snmp_get_mib2_virt_multi(queue_t *, mblk_t *,
 699                     ip_stack_t *ipst);
 700 static mblk_t   *ip_snmp_get_mib2_multi_rtable(queue_t *, mblk_t *,
 701                     ip_stack_t *ipst);
 702 static mblk_t   *ip_snmp_get_mib2_ip_route_media(queue_t *, mblk_t *, int,
 703                     ip_stack_t *ipst);
 704 static mblk_t   *ip_snmp_get_mib2_ip6_route_media(queue_t *, mblk_t *, int,
 705                     ip_stack_t *ipst);
 706 static void     ip_snmp_get2_v4(ire_t *, iproutedata_t *);
 707 static void     ip_snmp_get2_v6_route(ire_t *, iproutedata_t *);
 708 static int      ip_snmp_get2_v4_media(ncec_t *, iproutedata_t *);
 709 static int      ip_snmp_get2_v6_media(ncec_t *, iproutedata_t *);
 710 int             ip_snmp_set(queue_t *, int, int, uchar_t *, int);
 711 
 712 static mblk_t   *ip_fragment_copyhdr(uchar_t *, int, int, ip_stack_t *,
 713                     mblk_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(conn_t *connp, boolean_t closing);
 718 
 719 static void     conn_walk_drain(ip_stack_t *, idl_tx_list_t *);
 720 static void     conn_walk_sctp(pfv_t, void *, zoneid_t, netstack_t *);
 721 
 722 static void     *ip_stack_init(netstackid_t stackid, netstack_t *ns);
 723 static void     ip_stack_shutdown(netstackid_t stackid, void *arg);
 724 static void     ip_stack_fini(netstackid_t stackid, void *arg);
 725 
 726 static int      ip_multirt_apply_membership(int (*fn)(conn_t *, boolean_t,
 727     const in6_addr_t *, ipaddr_t, uint_t, mcast_record_t, const in6_addr_t *),
 728     ire_t *, conn_t *, boolean_t, const in6_addr_t *,  mcast_record_t,
 729     const in6_addr_t *);
 730 
 731 static int      ip_squeue_switch(int);
 732 
 733 static void     *ip_kstat_init(netstackid_t, ip_stack_t *);
 734 static void     ip_kstat_fini(netstackid_t, kstat_t *);
 735 static int      ip_kstat_update(kstat_t *kp, int rw);
 736 static void     *icmp_kstat_init(netstackid_t);
 737 static void     icmp_kstat_fini(netstackid_t, kstat_t *);
 738 static int      icmp_kstat_update(kstat_t *kp, int rw);
 739 static void     *ip_kstat2_init(netstackid_t, ip_stat_t *);
 740 static void     ip_kstat2_fini(netstackid_t, kstat_t *);
 741 
 742 static void     ipobs_init(ip_stack_t *);
 743 static void     ipobs_fini(ip_stack_t *);
 744 
 745 static int      ip_tp_cpu_update(cpu_setup_t, int, void *);
 746 
 747 ipaddr_t        ip_g_all_ones = IP_HOST_MASK;
 748 
 749 static long ip_rput_pullups;
 750 int     dohwcksum = 1;  /* use h/w cksum if supported by the hardware */
 751 
 752 vmem_t *ip_minor_arena_sa; /* for minor nos. from INET_MIN_DEV+2 thru 2^^18-1 */
 753 vmem_t *ip_minor_arena_la; /* for minor nos. from 2^^18 thru 2^^32-1 */
 754 
 755 int     ip_debug;
 756 
 757 /*
 758  * Multirouting/CGTP stuff
 759  */
 760 int     ip_cgtp_filter_rev = CGTP_FILTER_REV;   /* CGTP hooks version */
 761 
 762 /*
 763  * IP tunables related declarations. Definitions are in ip_tunables.c
 764  */
 765 extern mod_prop_info_t ip_propinfo_tbl[];
 766 extern int ip_propinfo_count;
 767 
 768 /*
 769  * Table of IP ioctls encoding the various properties of the ioctl and
 770  * indexed based on the last byte of the ioctl command. Occasionally there
 771  * is a clash, and there is more than 1 ioctl with the same last byte.
 772  * In such a case 1 ioctl is encoded in the ndx table and the remaining
 773  * ioctls are encoded in the misc table. An entry in the ndx table is
 774  * retrieved by indexing on the last byte of the ioctl command and comparing
 775  * the ioctl command with the value in the ndx table. In the event of a
 776  * mismatch the misc table is then searched sequentially for the desired
 777  * ioctl command.
 778  *
 779  * Entry: <command> <copyin_size> <flags> <cmd_type> <function> <restart_func>
 780  */
 781 ip_ioctl_cmd_t ip_ndx_ioctl_table[] = {
 782         /* 000 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 783         /* 001 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 784         /* 002 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 785         /* 003 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 786         /* 004 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 787         /* 005 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 788         /* 006 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 789         /* 007 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 790         /* 008 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 791         /* 009 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 792 
 793         /* 010 */ { SIOCADDRT,  sizeof (struct rtentry), IPI_PRIV,
 794                         MISC_CMD, ip_siocaddrt, NULL },
 795         /* 011 */ { SIOCDELRT,  sizeof (struct rtentry), IPI_PRIV,
 796                         MISC_CMD, ip_siocdelrt, NULL },
 797 
 798         /* 012 */ { SIOCSIFADDR, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
 799                         IF_CMD, ip_sioctl_addr, ip_sioctl_addr_restart },
 800         /* 013 */ { SIOCGIFADDR, sizeof (struct ifreq), IPI_GET_CMD,
 801                         IF_CMD, ip_sioctl_get_addr, NULL },
 802 
 803         /* 014 */ { SIOCSIFDSTADDR, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
 804                         IF_CMD, ip_sioctl_dstaddr, ip_sioctl_dstaddr_restart },
 805         /* 015 */ { SIOCGIFDSTADDR, sizeof (struct ifreq),
 806                         IPI_GET_CMD, IF_CMD, ip_sioctl_get_dstaddr, NULL },
 807 
 808         /* 016 */ { SIOCSIFFLAGS, sizeof (struct ifreq),
 809                         IPI_PRIV | IPI_WR,
 810                         IF_CMD, ip_sioctl_flags, ip_sioctl_flags_restart },
 811         /* 017 */ { SIOCGIFFLAGS, sizeof (struct ifreq),
 812                         IPI_MODOK | IPI_GET_CMD,
 813                         IF_CMD, ip_sioctl_get_flags, NULL },
 814 
 815         /* 018 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 816         /* 019 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 817 
 818         /* copyin size cannot be coded for SIOCGIFCONF */
 819         /* 020 */ { O_SIOCGIFCONF, 0, IPI_GET_CMD,
 820                         MISC_CMD, ip_sioctl_get_ifconf, NULL },
 821 
 822         /* 021 */ { SIOCSIFMTU, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
 823                         IF_CMD, ip_sioctl_mtu, NULL },
 824         /* 022 */ { SIOCGIFMTU, sizeof (struct ifreq), IPI_GET_CMD,
 825                         IF_CMD, ip_sioctl_get_mtu, NULL },
 826         /* 023 */ { SIOCGIFBRDADDR, sizeof (struct ifreq),
 827                         IPI_GET_CMD, IF_CMD, ip_sioctl_get_brdaddr, NULL },
 828         /* 024 */ { SIOCSIFBRDADDR, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
 829                         IF_CMD, ip_sioctl_brdaddr, NULL },
 830         /* 025 */ { SIOCGIFNETMASK, sizeof (struct ifreq),
 831                         IPI_GET_CMD, IF_CMD, ip_sioctl_get_netmask, NULL },
 832         /* 026 */ { SIOCSIFNETMASK, sizeof (struct ifreq), IPI_PRIV | IPI_WR,
 833                         IF_CMD, ip_sioctl_netmask, ip_sioctl_netmask_restart },
 834         /* 027 */ { SIOCGIFMETRIC, sizeof (struct ifreq),
 835                         IPI_GET_CMD, IF_CMD, ip_sioctl_get_metric, NULL },
 836         /* 028 */ { SIOCSIFMETRIC, sizeof (struct ifreq), IPI_PRIV,
 837                         IF_CMD, ip_sioctl_metric, NULL },
 838         /* 029 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 839 
 840         /* See 166-168 below for extended SIOC*XARP ioctls */
 841         /* 030 */ { SIOCSARP, sizeof (struct arpreq), IPI_PRIV | IPI_WR,
 842                         ARP_CMD, ip_sioctl_arp, NULL },
 843         /* 031 */ { SIOCGARP, sizeof (struct arpreq), IPI_GET_CMD,
 844                         ARP_CMD, ip_sioctl_arp, NULL },
 845         /* 032 */ { SIOCDARP, sizeof (struct arpreq), IPI_PRIV | IPI_WR,
 846                         ARP_CMD, ip_sioctl_arp, NULL },
 847 
 848         /* 033 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 849         /* 034 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 850         /* 035 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 851         /* 036 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 852         /* 037 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 853         /* 038 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 854         /* 039 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 855         /* 040 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 856         /* 041 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 857         /* 042 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 858         /* 043 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 859         /* 044 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 860         /* 045 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 861         /* 046 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 862         /* 047 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 863         /* 048 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 864         /* 049 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 865         /* 050 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 866         /* 051 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 867         /* 052 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 868         /* 053 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 869 
 870         /* 054 */ { IF_UNITSEL, sizeof (int), IPI_PRIV | IPI_WR | IPI_MODOK,
 871                         MISC_CMD, if_unitsel, if_unitsel_restart },
 872 
 873         /* 055 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 874         /* 056 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 875         /* 057 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 876         /* 058 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 877         /* 059 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 878         /* 060 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 879         /* 061 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 880         /* 062 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 881         /* 063 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 882         /* 064 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 883         /* 065 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 884         /* 066 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 885         /* 067 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 886         /* 068 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 887         /* 069 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 888         /* 070 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 889         /* 071 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 890         /* 072 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 891 
 892         /* 073 */ { SIOCSIFNAME, sizeof (struct ifreq),
 893                         IPI_PRIV | IPI_WR | IPI_MODOK,
 894                         IF_CMD, ip_sioctl_sifname, NULL },
 895 
 896         /* 074 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 897         /* 075 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 898         /* 076 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 899         /* 077 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 900         /* 078 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 901         /* 079 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 902         /* 080 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 903         /* 081 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 904         /* 082 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 905         /* 083 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 906         /* 084 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 907         /* 085 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 908         /* 086 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 909 
 910         /* 087 */ { SIOCGIFNUM, sizeof (int), IPI_GET_CMD,
 911                         MISC_CMD, ip_sioctl_get_ifnum, NULL },
 912         /* 088 */ { SIOCGIFMUXID, sizeof (struct ifreq), IPI_GET_CMD,
 913                         IF_CMD, ip_sioctl_get_muxid, NULL },
 914         /* 089 */ { SIOCSIFMUXID, sizeof (struct ifreq),
 915                         IPI_PRIV | IPI_WR, IF_CMD, ip_sioctl_muxid, NULL },
 916 
 917         /* Both if and lif variants share same func */
 918         /* 090 */ { SIOCGIFINDEX, sizeof (struct ifreq), IPI_GET_CMD,
 919                         IF_CMD, ip_sioctl_get_lifindex, NULL },
 920         /* Both if and lif variants share same func */
 921         /* 091 */ { SIOCSIFINDEX, sizeof (struct ifreq),
 922                         IPI_PRIV | IPI_WR, IF_CMD, ip_sioctl_slifindex, NULL },
 923 
 924         /* copyin size cannot be coded for SIOCGIFCONF */
 925         /* 092 */ { SIOCGIFCONF, 0, IPI_GET_CMD,
 926                         MISC_CMD, ip_sioctl_get_ifconf, NULL },
 927         /* 093 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 928         /* 094 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 929         /* 095 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 930         /* 096 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 931         /* 097 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 932         /* 098 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 933         /* 099 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 934         /* 100 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 935         /* 101 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 936         /* 102 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 937         /* 103 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 938         /* 104 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 939         /* 105 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 940         /* 106 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 941         /* 107 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 942         /* 108 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 943         /* 109 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 944 
 945         /* 110 */ { SIOCLIFREMOVEIF, sizeof (struct lifreq),
 946                         IPI_PRIV | IPI_WR, LIF_CMD, ip_sioctl_removeif,
 947                         ip_sioctl_removeif_restart },
 948         /* 111 */ { SIOCLIFADDIF, sizeof (struct lifreq),
 949                         IPI_GET_CMD | IPI_PRIV | IPI_WR,
 950                         LIF_CMD, ip_sioctl_addif, NULL },
 951 #define SIOCLIFADDR_NDX 112
 952         /* 112 */ { SIOCSLIFADDR, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
 953                         LIF_CMD, ip_sioctl_addr, ip_sioctl_addr_restart },
 954         /* 113 */ { SIOCGLIFADDR, sizeof (struct lifreq),
 955                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_addr, NULL },
 956         /* 114 */ { SIOCSLIFDSTADDR, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
 957                         LIF_CMD, ip_sioctl_dstaddr, ip_sioctl_dstaddr_restart },
 958         /* 115 */ { SIOCGLIFDSTADDR, sizeof (struct lifreq),
 959                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_dstaddr, NULL },
 960         /* 116 */ { SIOCSLIFFLAGS, sizeof (struct lifreq),
 961                         IPI_PRIV | IPI_WR,
 962                         LIF_CMD, ip_sioctl_flags, ip_sioctl_flags_restart },
 963         /* 117 */ { SIOCGLIFFLAGS, sizeof (struct lifreq),
 964                         IPI_GET_CMD | IPI_MODOK,
 965                         LIF_CMD, ip_sioctl_get_flags, NULL },
 966 
 967         /* 118 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 968         /* 119 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
 969 
 970         /* 120 */ { O_SIOCGLIFCONF, 0, IPI_GET_CMD, MISC_CMD,
 971                         ip_sioctl_get_lifconf, NULL },
 972         /* 121 */ { SIOCSLIFMTU, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
 973                         LIF_CMD, ip_sioctl_mtu, NULL },
 974         /* 122 */ { SIOCGLIFMTU, sizeof (struct lifreq), IPI_GET_CMD,
 975                         LIF_CMD, ip_sioctl_get_mtu, NULL },
 976         /* 123 */ { SIOCGLIFBRDADDR, sizeof (struct lifreq),
 977                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_brdaddr, NULL },
 978         /* 124 */ { SIOCSLIFBRDADDR, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
 979                         LIF_CMD, ip_sioctl_brdaddr, NULL },
 980         /* 125 */ { SIOCGLIFNETMASK, sizeof (struct lifreq),
 981                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_netmask, NULL },
 982         /* 126 */ { SIOCSLIFNETMASK, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
 983                         LIF_CMD, ip_sioctl_netmask, ip_sioctl_netmask_restart },
 984         /* 127 */ { SIOCGLIFMETRIC, sizeof (struct lifreq),
 985                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_metric, NULL },
 986         /* 128 */ { SIOCSLIFMETRIC, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
 987                         LIF_CMD, ip_sioctl_metric, NULL },
 988         /* 129 */ { SIOCSLIFNAME, sizeof (struct lifreq),
 989                         IPI_PRIV | IPI_WR | IPI_MODOK,
 990                         LIF_CMD, ip_sioctl_slifname,
 991                         ip_sioctl_slifname_restart },
 992 
 993         /* 130 */ { SIOCGLIFNUM, sizeof (struct lifnum), IPI_GET_CMD,
 994                         MISC_CMD, ip_sioctl_get_lifnum, NULL },
 995         /* 131 */ { SIOCGLIFMUXID, sizeof (struct lifreq),
 996                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_muxid, NULL },
 997         /* 132 */ { SIOCSLIFMUXID, sizeof (struct lifreq),
 998                         IPI_PRIV | IPI_WR, LIF_CMD, ip_sioctl_muxid, NULL },
 999         /* 133 */ { SIOCGLIFINDEX, sizeof (struct lifreq),
1000                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_lifindex, 0 },
1001         /* 134 */ { SIOCSLIFINDEX, sizeof (struct lifreq),
1002                         IPI_PRIV | IPI_WR, LIF_CMD, ip_sioctl_slifindex, 0 },
1003         /* 135 */ { SIOCSLIFTOKEN, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1004                         LIF_CMD, ip_sioctl_token, NULL },
1005         /* 136 */ { SIOCGLIFTOKEN, sizeof (struct lifreq),
1006                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_token, NULL },
1007         /* 137 */ { SIOCSLIFSUBNET, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1008                         LIF_CMD, ip_sioctl_subnet, ip_sioctl_subnet_restart },
1009         /* 138 */ { SIOCGLIFSUBNET, sizeof (struct lifreq),
1010                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_subnet, NULL },
1011         /* 139 */ { SIOCSLIFLNKINFO, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1012                         LIF_CMD, ip_sioctl_lnkinfo, NULL },
1013 
1014         /* 140 */ { SIOCGLIFLNKINFO, sizeof (struct lifreq),
1015                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_lnkinfo, NULL },
1016         /* 141 */ { SIOCLIFDELND, sizeof (struct lifreq), IPI_PRIV,
1017                         LIF_CMD, ip_siocdelndp_v6, NULL },
1018         /* 142 */ { SIOCLIFGETND, sizeof (struct lifreq), IPI_GET_CMD,
1019                         LIF_CMD, ip_siocqueryndp_v6, NULL },
1020         /* 143 */ { SIOCLIFSETND, sizeof (struct lifreq), IPI_PRIV,
1021                         LIF_CMD, ip_siocsetndp_v6, NULL },
1022         /* 144 */ { SIOCTMYADDR, sizeof (struct sioc_addrreq), IPI_GET_CMD,
1023                         MISC_CMD, ip_sioctl_tmyaddr, NULL },
1024         /* 145 */ { SIOCTONLINK, sizeof (struct sioc_addrreq), IPI_GET_CMD,
1025                         MISC_CMD, ip_sioctl_tonlink, NULL },
1026         /* 146 */ { SIOCTMYSITE, sizeof (struct sioc_addrreq), 0,
1027                         MISC_CMD, ip_sioctl_tmysite, NULL },
1028         /* 147 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1029         /* 148 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1030         /* IPSECioctls handled in ip_sioctl_copyin_setup itself */
1031         /* 149 */ { SIOCFIPSECONFIG, 0, IPI_PRIV, MISC_CMD, NULL, NULL },
1032         /* 150 */ { SIOCSIPSECONFIG, 0, IPI_PRIV, MISC_CMD, NULL, NULL },
1033         /* 151 */ { SIOCDIPSECONFIG, 0, IPI_PRIV, MISC_CMD, NULL, NULL },
1034         /* 152 */ { SIOCLIPSECONFIG, 0, IPI_PRIV, MISC_CMD, NULL, NULL },
1035 
1036         /* 153 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1037 
1038         /* 154 */ { SIOCGLIFBINDING, sizeof (struct lifreq), IPI_GET_CMD,
1039                         LIF_CMD, ip_sioctl_get_binding, NULL },
1040         /* 155 */ { SIOCSLIFGROUPNAME, sizeof (struct lifreq),
1041                         IPI_PRIV | IPI_WR,
1042                         LIF_CMD, ip_sioctl_groupname, ip_sioctl_groupname },
1043         /* 156 */ { SIOCGLIFGROUPNAME, sizeof (struct lifreq),
1044                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_groupname, NULL },
1045         /* 157 */ { SIOCGLIFGROUPINFO, sizeof (lifgroupinfo_t),
1046                         IPI_GET_CMD, MISC_CMD, ip_sioctl_groupinfo, NULL },
1047 
1048         /* Leave 158-160 unused; used to be SIOC*IFARP ioctls */
1049         /* 158 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1050         /* 159 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1051         /* 160 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1052 
1053         /* 161 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1054 
1055         /* These are handled in ip_sioctl_copyin_setup itself */
1056         /* 162 */ { SIOCGIP6ADDRPOLICY, 0, IPI_NULL_BCONT,
1057                         MISC_CMD, NULL, NULL },
1058         /* 163 */ { SIOCSIP6ADDRPOLICY, 0, IPI_PRIV | IPI_NULL_BCONT,
1059                         MISC_CMD, NULL, NULL },
1060         /* 164 */ { SIOCGDSTINFO, 0, IPI_GET_CMD, MISC_CMD, NULL, NULL },
1061 
1062         /* 165 */ { SIOCGLIFCONF, 0, IPI_GET_CMD, MISC_CMD,
1063                         ip_sioctl_get_lifconf, NULL },
1064 
1065         /* 166 */ { SIOCSXARP, sizeof (struct xarpreq), IPI_PRIV | IPI_WR,
1066                         XARP_CMD, ip_sioctl_arp, NULL },
1067         /* 167 */ { SIOCGXARP, sizeof (struct xarpreq), IPI_GET_CMD,
1068                         XARP_CMD, ip_sioctl_arp, NULL },
1069         /* 168 */ { SIOCDXARP, sizeof (struct xarpreq), IPI_PRIV | IPI_WR,
1070                         XARP_CMD, ip_sioctl_arp, NULL },
1071 
1072         /* SIOCPOPSOCKFS is not handled by IP */
1073         /* 169 */ { IPI_DONTCARE /* SIOCPOPSOCKFS */, 0, 0, 0, NULL, NULL },
1074 
1075         /* 170 */ { SIOCGLIFZONE, sizeof (struct lifreq),
1076                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_lifzone, NULL },
1077         /* 171 */ { SIOCSLIFZONE, sizeof (struct lifreq),
1078                         IPI_PRIV | IPI_WR, LIF_CMD, ip_sioctl_slifzone,
1079                         ip_sioctl_slifzone_restart },
1080         /* 172-174 are SCTP ioctls and not handled by IP */
1081         /* 172 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1082         /* 173 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1083         /* 174 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1084         /* 175 */ { SIOCGLIFUSESRC, sizeof (struct lifreq),
1085                         IPI_GET_CMD, LIF_CMD,
1086                         ip_sioctl_get_lifusesrc, 0 },
1087         /* 176 */ { SIOCSLIFUSESRC, sizeof (struct lifreq),
1088                         IPI_PRIV | IPI_WR,
1089                         LIF_CMD, ip_sioctl_slifusesrc,
1090                         NULL },
1091         /* 177 */ { SIOCGLIFSRCOF, 0, IPI_GET_CMD, MISC_CMD,
1092                         ip_sioctl_get_lifsrcof, NULL },
1093         /* 178 */ { SIOCGMSFILTER, sizeof (struct group_filter), IPI_GET_CMD,
1094                         MSFILT_CMD, ip_sioctl_msfilter, NULL },
1095         /* 179 */ { SIOCSMSFILTER, sizeof (struct group_filter), 0,
1096                         MSFILT_CMD, ip_sioctl_msfilter, NULL },
1097         /* 180 */ { SIOCGIPMSFILTER, sizeof (struct ip_msfilter), IPI_GET_CMD,
1098                         MSFILT_CMD, ip_sioctl_msfilter, NULL },
1099         /* 181 */ { SIOCSIPMSFILTER, sizeof (struct ip_msfilter), 0,
1100                         MSFILT_CMD, ip_sioctl_msfilter, NULL },
1101         /* 182 */ { IPI_DONTCARE, 0, 0, 0, NULL, NULL },
1102         /* SIOCSENABLESDP is handled by SDP */
1103         /* 183 */ { IPI_DONTCARE /* SIOCSENABLESDP */, 0, 0, 0, NULL, NULL },
1104         /* 184 */ { IPI_DONTCARE /* SIOCSQPTR */, 0, 0, 0, NULL, NULL },
1105         /* 185 */ { SIOCGIFHWADDR, sizeof (struct ifreq), IPI_GET_CMD,
1106                         IF_CMD, ip_sioctl_get_ifhwaddr, NULL },
1107         /* 186 */ { IPI_DONTCARE /* SIOCGSTAMP */, 0, 0, 0, NULL, NULL },
1108         /* 187 */ { SIOCILB, 0, IPI_PRIV | IPI_GET_CMD, MISC_CMD,
1109                         ip_sioctl_ilb_cmd, NULL },
1110         /* 188 */ { SIOCGETPROP, 0, IPI_GET_CMD, 0, NULL, NULL },
1111         /* 189 */ { SIOCSETPROP, 0, IPI_PRIV | IPI_WR, 0, NULL, NULL},
1112         /* 190 */ { SIOCGLIFDADSTATE, sizeof (struct lifreq),
1113                         IPI_GET_CMD, LIF_CMD, ip_sioctl_get_dadstate, NULL },
1114         /* 191 */ { SIOCSLIFPREFIX, sizeof (struct lifreq), IPI_PRIV | IPI_WR,
1115                         LIF_CMD, ip_sioctl_prefix, ip_sioctl_prefix_restart },
1116         /* 192 */ { SIOCGLIFHWADDR, sizeof (struct lifreq), IPI_GET_CMD,
1117                         LIF_CMD, ip_sioctl_get_lifhwaddr, NULL }
1118 };
1119 
1120 int ip_ndx_ioctl_count = sizeof (ip_ndx_ioctl_table) / sizeof (ip_ioctl_cmd_t);
1121 
1122 ip_ioctl_cmd_t ip_misc_ioctl_table[] = {
1123         { I_LINK,       0, IPI_PRIV | IPI_WR, 0, NULL, NULL },
1124         { I_UNLINK,     0, IPI_PRIV | IPI_WR, 0, NULL, NULL },
1125         { I_PLINK,      0, IPI_PRIV | IPI_WR, 0, NULL, NULL },
1126         { I_PUNLINK,    0, IPI_PRIV | IPI_WR, 0, NULL, NULL },
1127         { ND_GET,       0, 0, 0, NULL, NULL },
1128         { ND_SET,       0, IPI_PRIV | IPI_WR, 0, NULL, NULL },
1129         { IP_IOCTL,     0, 0, 0, NULL, NULL },
1130         { SIOCGETVIFCNT, sizeof (struct sioc_vif_req), IPI_GET_CMD,
1131                 MISC_CMD, mrt_ioctl},
1132         { SIOCGETSGCNT, sizeof (struct sioc_sg_req), IPI_GET_CMD,
1133                 MISC_CMD, mrt_ioctl},
1134         { SIOCGETLSGCNT, sizeof (struct sioc_lsg_req), IPI_GET_CMD,
1135                 MISC_CMD, mrt_ioctl}
1136 };
1137 
1138 int ip_misc_ioctl_count =
1139     sizeof (ip_misc_ioctl_table) / sizeof (ip_ioctl_cmd_t);
1140 
1141 int     conn_drain_nthreads;            /* Number of drainers reqd. */
1142                                         /* Settable in /etc/system */
1143 /* Defined in ip_ire.c */
1144 extern uint32_t ip_ire_max_bucket_cnt, ip6_ire_max_bucket_cnt;
1145 extern uint32_t ip_ire_min_bucket_cnt, ip6_ire_min_bucket_cnt;
1146 extern uint32_t ip_ire_mem_ratio, ip_ire_cpu_ratio;
1147 
1148 static nv_t     ire_nv_arr[] = {
1149         { IRE_BROADCAST, "BROADCAST" },
1150         { IRE_LOCAL, "LOCAL" },
1151         { IRE_LOOPBACK, "LOOPBACK" },
1152         { IRE_DEFAULT, "DEFAULT" },
1153         { IRE_PREFIX, "PREFIX" },
1154         { IRE_IF_NORESOLVER, "IF_NORESOL" },
1155         { IRE_IF_RESOLVER, "IF_RESOLV" },
1156         { IRE_IF_CLONE, "IF_CLONE" },
1157         { IRE_HOST, "HOST" },
1158         { IRE_MULTICAST, "MULTICAST" },
1159         { IRE_NOROUTE, "NOROUTE" },
1160         { 0 }
1161 };
1162 
1163 nv_t    *ire_nv_tbl = ire_nv_arr;
1164 
1165 /* Simple ICMP IP Header Template */
1166 static ipha_t icmp_ipha = {
1167         IP_SIMPLE_HDR_VERSION, 0, 0, 0, 0, 0, IPPROTO_ICMP
1168 };
1169 
1170 struct module_info ip_mod_info = {
1171         IP_MOD_ID, IP_MOD_NAME, IP_MOD_MINPSZ, IP_MOD_MAXPSZ, IP_MOD_HIWAT,
1172         IP_MOD_LOWAT
1173 };
1174 
1175 /*
1176  * Duplicate static symbols within a module confuses mdb; so we avoid the
1177  * problem by making the symbols here distinct from those in udp.c.
1178  */
1179 
1180 /*
1181  * Entry points for IP as a device and as a module.
1182  * We have separate open functions for the /dev/ip and /dev/ip6 devices.
1183  */
1184 static struct qinit iprinitv4 = {
1185         (pfi_t)ip_rput, NULL, ip_openv4, ip_close, NULL,
1186         &ip_mod_info
1187 };
1188 
1189 struct qinit iprinitv6 = {
1190         (pfi_t)ip_rput_v6, NULL, ip_openv6, ip_close, NULL,
1191         &ip_mod_info
1192 };
1193 
1194 static struct qinit ipwinit = {
1195         (pfi_t)ip_wput_nondata, (pfi_t)ip_wsrv, NULL, NULL, NULL,
1196         &ip_mod_info
1197 };
1198 
1199 static struct qinit iplrinit = {
1200         (pfi_t)ip_lrput, NULL, ip_openv4, ip_close, NULL,
1201         &ip_mod_info
1202 };
1203 
1204 static struct qinit iplwinit = {
1205         (pfi_t)ip_lwput, NULL, NULL, NULL, NULL,
1206         &ip_mod_info
1207 };
1208 
1209 /* For AF_INET aka /dev/ip */
1210 struct streamtab ipinfov4 = {
1211         &iprinitv4, &ipwinit, &iplrinit, &iplwinit
1212 };
1213 
1214 /* For AF_INET6 aka /dev/ip6 */
1215 struct streamtab ipinfov6 = {
1216         &iprinitv6, &ipwinit, &iplrinit, &iplwinit
1217 };
1218 
1219 #ifdef  DEBUG
1220 boolean_t skip_sctp_cksum = B_FALSE;
1221 #endif
1222 
1223 /*
1224  * Generate an ICMP fragmentation needed message.
1225  * When called from ip_output side a minimal ip_recv_attr_t needs to be
1226  * constructed by the caller.
1227  */
1228 void
1229 icmp_frag_needed(mblk_t *mp, int mtu, ip_recv_attr_t *ira)
1230 {
1231         icmph_t icmph;
1232         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
1233 
1234         mp = icmp_pkt_err_ok(mp, ira);
1235         if (mp == NULL)
1236                 return;
1237 
1238         bzero(&icmph, sizeof (icmph_t));
1239         icmph.icmph_type = ICMP_DEST_UNREACHABLE;
1240         icmph.icmph_code = ICMP_FRAGMENTATION_NEEDED;
1241         icmph.icmph_du_mtu = htons((uint16_t)mtu);
1242         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutFragNeeded);
1243         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDestUnreachs);
1244 
1245         icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
1246 }
1247 
1248 /*
1249  * icmp_inbound_v4 deals with ICMP messages that are handled by IP.
1250  * If the ICMP message is consumed by IP, i.e., it should not be delivered
1251  * to any IPPROTO_ICMP raw sockets, then it returns NULL.
1252  * Likewise, if the ICMP error is misformed (too short, etc), then it
1253  * returns NULL. The caller uses this to determine whether or not to send
1254  * to raw sockets.
1255  *
1256  * All error messages are passed to the matching transport stream.
1257  *
1258  * The following cases are handled by icmp_inbound:
1259  * 1) It needs to send a reply back and possibly delivering it
1260  *    to the "interested" upper clients.
1261  * 2) Return the mblk so that the caller can pass it to the RAW socket clients.
1262  * 3) It needs to change some values in IP only.
1263  * 4) It needs to change some values in IP and upper layers e.g TCP
1264  *    by delivering an error to the upper layers.
1265  *
1266  * We handle the above three cases in the context of IPsec in the
1267  * following way :
1268  *
1269  * 1) Send the reply back in the same way as the request came in.
1270  *    If it came in encrypted, it goes out encrypted. If it came in
1271  *    clear, it goes out in clear. Thus, this will prevent chosen
1272  *    plain text attack.
1273  * 2) The client may or may not expect things to come in secure.
1274  *    If it comes in secure, the policy constraints are checked
1275  *    before delivering it to the upper layers. If it comes in
1276  *    clear, ipsec_inbound_accept_clear will decide whether to
1277  *    accept this in clear or not. In both the cases, if the returned
1278  *    message (IP header + 8 bytes) that caused the icmp message has
1279  *    AH/ESP headers, it is sent up to AH/ESP for validation before
1280  *    sending up. If there are only 8 bytes of returned message, then
1281  *    upper client will not be notified.
1282  * 3) Check with global policy to see whether it matches the constaints.
1283  *    But this will be done only if icmp_accept_messages_in_clear is
1284  *    zero.
1285  * 4) If we need to change both in IP and ULP, then the decision taken
1286  *    while affecting the values in IP and while delivering up to TCP
1287  *    should be the same.
1288  *
1289  *      There are two cases.
1290  *
1291  *      a) If we reject data at the IP layer (ipsec_check_global_policy()
1292  *         failed), we will not deliver it to the ULP, even though they
1293  *         are *willing* to accept in *clear*. This is fine as our global
1294  *         disposition to icmp messages asks us reject the datagram.
1295  *
1296  *      b) If we accept data at the IP layer (ipsec_check_global_policy()
1297  *         succeeded or icmp_accept_messages_in_clear is 1), and not able
1298  *         to deliver it to ULP (policy failed), it can lead to
1299  *         consistency problems. The cases known at this time are
1300  *         ICMP_DESTINATION_UNREACHABLE  messages with following code
1301  *         values :
1302  *
1303  *         - ICMP_FRAGMENTATION_NEEDED : IP adapts to the new value
1304  *           and Upper layer rejects. Then the communication will
1305  *           come to a stop. This is solved by making similar decisions
1306  *           at both levels. Currently, when we are unable to deliver
1307  *           to the Upper Layer (due to policy failures) while IP has
1308  *           adjusted dce_pmtu, the next outbound datagram would
1309  *           generate a local ICMP_FRAGMENTATION_NEEDED message - which
1310  *           will be with the right level of protection. Thus the right
1311  *           value will be communicated even if we are not able to
1312  *           communicate when we get from the wire initially. But this
1313  *           assumes there would be at least one outbound datagram after
1314  *           IP has adjusted its dce_pmtu value. To make things
1315  *           simpler, we accept in clear after the validation of
1316  *           AH/ESP headers.
1317  *
1318  *         - Other ICMP ERRORS : We may not be able to deliver it to the
1319  *           upper layer depending on the level of protection the upper
1320  *           layer expects and the disposition in ipsec_inbound_accept_clear().
1321  *           ipsec_inbound_accept_clear() decides whether a given ICMP error
1322  *           should be accepted in clear when the Upper layer expects secure.
1323  *           Thus the communication may get aborted by some bad ICMP
1324  *           packets.
1325  */
1326 mblk_t *
1327 icmp_inbound_v4(mblk_t *mp, ip_recv_attr_t *ira)
1328 {
1329         icmph_t         *icmph;
1330         ipha_t          *ipha;          /* Outer header */
1331         int             ip_hdr_length;  /* Outer header length */
1332         boolean_t       interested;
1333         ipif_t          *ipif;
1334         uint32_t        ts;
1335         uint32_t        *tsp;
1336         timestruc_t     now;
1337         ill_t           *ill = ira->ira_ill;
1338         ip_stack_t      *ipst = ill->ill_ipst;
1339         zoneid_t        zoneid = ira->ira_zoneid;
1340         int             len_needed;
1341         mblk_t          *mp_ret = NULL;
1342 
1343         ipha = (ipha_t *)mp->b_rptr;
1344 
1345         BUMP_MIB(&ipst->ips_icmp_mib, icmpInMsgs);
1346 
1347         ip_hdr_length = ira->ira_ip_hdr_length;
1348         if ((mp->b_wptr - mp->b_rptr) < (ip_hdr_length + ICMPH_SIZE)) {
1349                 if (ira->ira_pktlen < (ip_hdr_length + ICMPH_SIZE)) {
1350                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
1351                         ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
1352                         freemsg(mp);
1353                         return (NULL);
1354                 }
1355                 /* Last chance to get real. */
1356                 ipha = ip_pullup(mp, ip_hdr_length + ICMPH_SIZE, ira);
1357                 if (ipha == NULL) {
1358                         BUMP_MIB(&ipst->ips_icmp_mib, icmpInErrors);
1359                         freemsg(mp);
1360                         return (NULL);
1361                 }
1362         }
1363 
1364         /* The IP header will always be a multiple of four bytes */
1365         icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1366         ip2dbg(("icmp_inbound_v4: type %d code %d\n", icmph->icmph_type,
1367             icmph->icmph_code));
1368 
1369         /*
1370          * We will set "interested" to "true" if we should pass a copy to
1371          * the transport or if we handle the packet locally.
1372          */
1373         interested = B_FALSE;
1374         switch (icmph->icmph_type) {
1375         case ICMP_ECHO_REPLY:
1376                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInEchoReps);
1377                 break;
1378         case ICMP_DEST_UNREACHABLE:
1379                 if (icmph->icmph_code == ICMP_FRAGMENTATION_NEEDED)
1380                         BUMP_MIB(&ipst->ips_icmp_mib, icmpInFragNeeded);
1381                 interested = B_TRUE;    /* Pass up to transport */
1382                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInDestUnreachs);
1383                 break;
1384         case ICMP_SOURCE_QUENCH:
1385                 interested = B_TRUE;    /* Pass up to transport */
1386                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInSrcQuenchs);
1387                 break;
1388         case ICMP_REDIRECT:
1389                 if (!ipst->ips_ip_ignore_redirect)
1390                         interested = B_TRUE;
1391                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInRedirects);
1392                 break;
1393         case ICMP_ECHO_REQUEST:
1394                 /*
1395                  * Whether to respond to echo requests that come in as IP
1396                  * broadcasts or as IP multicast is subject to debate
1397                  * (what isn't?).  We aim to please, you pick it.
1398                  * Default is do it.
1399                  */
1400                 if (ira->ira_flags & IRAF_MULTICAST) {
1401                         /* multicast: respond based on tunable */
1402                         interested = ipst->ips_ip_g_resp_to_echo_mcast;
1403                 } else if (ira->ira_flags & IRAF_BROADCAST) {
1404                         /* broadcast: respond based on tunable */
1405                         interested = ipst->ips_ip_g_resp_to_echo_bcast;
1406                 } else {
1407                         /* unicast: always respond */
1408                         interested = B_TRUE;
1409                 }
1410                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInEchos);
1411                 if (!interested) {
1412                         /* We never pass these to RAW sockets */
1413                         freemsg(mp);
1414                         return (NULL);
1415                 }
1416 
1417                 /* Check db_ref to make sure we can modify the packet. */
1418                 if (mp->b_datap->db_ref > 1) {
1419                         mblk_t  *mp1;
1420 
1421                         mp1 = copymsg(mp);
1422                         freemsg(mp);
1423                         if (!mp1) {
1424                                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
1425                                 return (NULL);
1426                         }
1427                         mp = mp1;
1428                         ipha = (ipha_t *)mp->b_rptr;
1429                         icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1430                 }
1431                 icmph->icmph_type = ICMP_ECHO_REPLY;
1432                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutEchoReps);
1433                 icmp_send_reply_v4(mp, ipha, icmph, ira);
1434                 return (NULL);
1435 
1436         case ICMP_ROUTER_ADVERTISEMENT:
1437         case ICMP_ROUTER_SOLICITATION:
1438                 break;
1439         case ICMP_TIME_EXCEEDED:
1440                 interested = B_TRUE;    /* Pass up to transport */
1441                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInTimeExcds);
1442                 break;
1443         case ICMP_PARAM_PROBLEM:
1444                 interested = B_TRUE;    /* Pass up to transport */
1445                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInParmProbs);
1446                 break;
1447         case ICMP_TIME_STAMP_REQUEST:
1448                 /* Response to Time Stamp Requests is local policy. */
1449                 if (ipst->ips_ip_g_resp_to_timestamp) {
1450                         if (ira->ira_flags & IRAF_MULTIBROADCAST)
1451                                 interested =
1452                                     ipst->ips_ip_g_resp_to_timestamp_bcast;
1453                         else
1454                                 interested = B_TRUE;
1455                 }
1456                 if (!interested) {
1457                         /* We never pass these to RAW sockets */
1458                         freemsg(mp);
1459                         return (NULL);
1460                 }
1461 
1462                 /* Make sure we have enough of the packet */
1463                 len_needed = ip_hdr_length + ICMPH_SIZE +
1464                     3 * sizeof (uint32_t);
1465 
1466                 if (mp->b_wptr - mp->b_rptr < len_needed) {
1467                         ipha = ip_pullup(mp, len_needed, ira);
1468                         if (ipha == NULL) {
1469                                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1470                                 ip_drop_input("ipIfStatsInDiscards - ip_pullup",
1471                                     mp, ill);
1472                                 freemsg(mp);
1473                                 return (NULL);
1474                         }
1475                         /* Refresh following the pullup. */
1476                         icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1477                 }
1478                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInTimestamps);
1479                 /* Check db_ref to make sure we can modify the packet. */
1480                 if (mp->b_datap->db_ref > 1) {
1481                         mblk_t  *mp1;
1482 
1483                         mp1 = copymsg(mp);
1484                         freemsg(mp);
1485                         if (!mp1) {
1486                                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
1487                                 return (NULL);
1488                         }
1489                         mp = mp1;
1490                         ipha = (ipha_t *)mp->b_rptr;
1491                         icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1492                 }
1493                 icmph->icmph_type = ICMP_TIME_STAMP_REPLY;
1494                 tsp = (uint32_t *)&icmph[1];
1495                 tsp++;          /* Skip past 'originate time' */
1496                 /* Compute # of milliseconds since midnight */
1497                 gethrestime(&now);
1498                 ts = (now.tv_sec % (24 * 60 * 60)) * 1000 +
1499                     now.tv_nsec / (NANOSEC / MILLISEC);
1500                 *tsp++ = htonl(ts);     /* Lay in 'receive time' */
1501                 *tsp++ = htonl(ts);     /* Lay in 'send time' */
1502                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutTimestampReps);
1503                 icmp_send_reply_v4(mp, ipha, icmph, ira);
1504                 return (NULL);
1505 
1506         case ICMP_TIME_STAMP_REPLY:
1507                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInTimestampReps);
1508                 break;
1509         case ICMP_INFO_REQUEST:
1510                 /* Per RFC 1122 3.2.2.7, ignore this. */
1511         case ICMP_INFO_REPLY:
1512                 break;
1513         case ICMP_ADDRESS_MASK_REQUEST:
1514                 if (ira->ira_flags & IRAF_MULTIBROADCAST) {
1515                         interested =
1516                             ipst->ips_ip_respond_to_address_mask_broadcast;
1517                 } else {
1518                         interested = B_TRUE;
1519                 }
1520                 if (!interested) {
1521                         /* We never pass these to RAW sockets */
1522                         freemsg(mp);
1523                         return (NULL);
1524                 }
1525                 len_needed = ip_hdr_length + ICMPH_SIZE + IP_ADDR_LEN;
1526                 if (mp->b_wptr - mp->b_rptr < len_needed) {
1527                         ipha = ip_pullup(mp, len_needed, ira);
1528                         if (ipha == NULL) {
1529                                 BUMP_MIB(ill->ill_ip_mib,
1530                                     ipIfStatsInTruncatedPkts);
1531                                 ip_drop_input("ipIfStatsInTruncatedPkts", mp,
1532                                     ill);
1533                                 freemsg(mp);
1534                                 return (NULL);
1535                         }
1536                         /* Refresh following the pullup. */
1537                         icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1538                 }
1539                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInAddrMasks);
1540                 /* Check db_ref to make sure we can modify the packet. */
1541                 if (mp->b_datap->db_ref > 1) {
1542                         mblk_t  *mp1;
1543 
1544                         mp1 = copymsg(mp);
1545                         freemsg(mp);
1546                         if (!mp1) {
1547                                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
1548                                 return (NULL);
1549                         }
1550                         mp = mp1;
1551                         ipha = (ipha_t *)mp->b_rptr;
1552                         icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1553                 }
1554                 /*
1555                  * Need the ipif with the mask be the same as the source
1556                  * address of the mask reply. For unicast we have a specific
1557                  * ipif. For multicast/broadcast we only handle onlink
1558                  * senders, and use the source address to pick an ipif.
1559                  */
1560                 ipif = ipif_lookup_addr(ipha->ipha_dst, ill, zoneid, ipst);
1561                 if (ipif == NULL) {
1562                         /* Broadcast or multicast */
1563                         ipif = ipif_lookup_remote(ill, ipha->ipha_src, zoneid);
1564                         if (ipif == NULL) {
1565                                 freemsg(mp);
1566                                 return (NULL);
1567                         }
1568                 }
1569                 icmph->icmph_type = ICMP_ADDRESS_MASK_REPLY;
1570                 bcopy(&ipif->ipif_net_mask, &icmph[1], IP_ADDR_LEN);
1571                 ipif_refrele(ipif);
1572                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutAddrMaskReps);
1573                 icmp_send_reply_v4(mp, ipha, icmph, ira);
1574                 return (NULL);
1575 
1576         case ICMP_ADDRESS_MASK_REPLY:
1577                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInAddrMaskReps);
1578                 break;
1579         default:
1580                 interested = B_TRUE;    /* Pass up to transport */
1581                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInUnknowns);
1582                 break;
1583         }
1584         /*
1585          * See if there is an ICMP client to avoid an extra copymsg/freemsg
1586          * if there isn't one.
1587          */
1588         if (ipst->ips_ipcl_proto_fanout_v4[IPPROTO_ICMP].connf_head != NULL) {
1589                 /* If there is an ICMP client and we want one too, copy it. */
1590 
1591                 if (!interested) {
1592                         /* Caller will deliver to RAW sockets */
1593                         return (mp);
1594                 }
1595                 mp_ret = copymsg(mp);
1596                 if (mp_ret == NULL) {
1597                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1598                         ip_drop_input("ipIfStatsInDiscards - copymsg", mp, ill);
1599                 }
1600         } else if (!interested) {
1601                 /* Neither we nor raw sockets are interested. Drop packet now */
1602                 freemsg(mp);
1603                 return (NULL);
1604         }
1605 
1606         /*
1607          * ICMP error or redirect packet. Make sure we have enough of
1608          * the header and that db_ref == 1 since we might end up modifying
1609          * the packet.
1610          */
1611         if (mp->b_cont != NULL) {
1612                 if (ip_pullup(mp, -1, ira) == NULL) {
1613                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1614                         ip_drop_input("ipIfStatsInDiscards - ip_pullup",
1615                             mp, ill);
1616                         freemsg(mp);
1617                         return (mp_ret);
1618                 }
1619         }
1620 
1621         if (mp->b_datap->db_ref > 1) {
1622                 mblk_t  *mp1;
1623 
1624                 mp1 = copymsg(mp);
1625                 if (mp1 == NULL) {
1626                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1627                         ip_drop_input("ipIfStatsInDiscards - copymsg", mp, ill);
1628                         freemsg(mp);
1629                         return (mp_ret);
1630                 }
1631                 freemsg(mp);
1632                 mp = mp1;
1633         }
1634 
1635         /*
1636          * In case mp has changed, verify the message before any further
1637          * processes.
1638          */
1639         ipha = (ipha_t *)mp->b_rptr;
1640         icmph = (icmph_t *)&mp->b_rptr[ip_hdr_length];
1641         if (!icmp_inbound_verify_v4(mp, icmph, ira)) {
1642                 freemsg(mp);
1643                 return (mp_ret);
1644         }
1645 
1646         switch (icmph->icmph_type) {
1647         case ICMP_REDIRECT:
1648                 icmp_redirect_v4(mp, ipha, icmph, ira);
1649                 break;
1650         case ICMP_DEST_UNREACHABLE:
1651                 if (icmph->icmph_code == ICMP_FRAGMENTATION_NEEDED) {
1652                         /* Update DCE and adjust MTU is icmp header if needed */
1653                         icmp_inbound_too_big_v4(icmph, ira);
1654                 }
1655                 /* FALLTHRU */
1656         default:
1657                 icmp_inbound_error_fanout_v4(mp, icmph, ira);
1658                 break;
1659         }
1660         return (mp_ret);
1661 }
1662 
1663 /*
1664  * Send an ICMP echo, timestamp or address mask reply.
1665  * The caller has already updated the payload part of the packet.
1666  * We handle the ICMP checksum, IP source address selection and feed
1667  * the packet into ip_output_simple.
1668  */
1669 static void
1670 icmp_send_reply_v4(mblk_t *mp, ipha_t *ipha, icmph_t *icmph,
1671     ip_recv_attr_t *ira)
1672 {
1673         uint_t          ip_hdr_length = ira->ira_ip_hdr_length;
1674         ill_t           *ill = ira->ira_ill;
1675         ip_stack_t      *ipst = ill->ill_ipst;
1676         ip_xmit_attr_t  ixas;
1677 
1678         /* Send out an ICMP packet */
1679         icmph->icmph_checksum = 0;
1680         icmph->icmph_checksum = IP_CSUM(mp, ip_hdr_length, 0);
1681         /* Reset time to live. */
1682         ipha->ipha_ttl = ipst->ips_ip_def_ttl;
1683         {
1684                 /* Swap source and destination addresses */
1685                 ipaddr_t tmp;
1686 
1687                 tmp = ipha->ipha_src;
1688                 ipha->ipha_src = ipha->ipha_dst;
1689                 ipha->ipha_dst = tmp;
1690         }
1691         ipha->ipha_ident = 0;
1692         if (!IS_SIMPLE_IPH(ipha))
1693                 icmp_options_update(ipha);
1694 
1695         bzero(&ixas, sizeof (ixas));
1696         ixas.ixa_flags = IXAF_BASIC_SIMPLE_V4;
1697         ixas.ixa_zoneid = ira->ira_zoneid;
1698         ixas.ixa_cred = kcred;
1699         ixas.ixa_cpid = NOPID;
1700         ixas.ixa_tsl = ira->ira_tsl; /* Behave as a multi-level responder */
1701         ixas.ixa_ifindex = 0;
1702         ixas.ixa_ipst = ipst;
1703         ixas.ixa_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
1704 
1705         if (!(ira->ira_flags & IRAF_IPSEC_SECURE)) {
1706                 /*
1707                  * This packet should go out the same way as it
1708                  * came in i.e in clear, independent of the IPsec policy
1709                  * for transmitting packets.
1710                  */
1711                 ixas.ixa_flags |= IXAF_NO_IPSEC;
1712         } else {
1713                 if (!ipsec_in_to_out(ira, &ixas, mp, ipha, NULL)) {
1714                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1715                         /* Note: mp already consumed and ip_drop_packet done */
1716                         return;
1717                 }
1718         }
1719         if (ira->ira_flags & IRAF_MULTIBROADCAST) {
1720                 /*
1721                  * Not one or our addresses (IRE_LOCALs), thus we let
1722                  * ip_output_simple pick the source.
1723                  */
1724                 ipha->ipha_src = INADDR_ANY;
1725                 ixas.ixa_flags |= IXAF_SET_SOURCE;
1726         }
1727         /* Should we send with DF and use dce_pmtu? */
1728         if (ipst->ips_ipv4_icmp_return_pmtu) {
1729                 ixas.ixa_flags |= IXAF_PMTU_DISCOVERY;
1730                 ipha->ipha_fragment_offset_and_flags |= IPH_DF_HTONS;
1731         }
1732 
1733         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutMsgs);
1734 
1735         (void) ip_output_simple(mp, &ixas);
1736         ixa_cleanup(&ixas);
1737 }
1738 
1739 /*
1740  * Verify the ICMP messages for either for ICMP error or redirect packet.
1741  * The caller should have fully pulled up the message. If it's a redirect
1742  * packet, only basic checks on IP header will be done; otherwise, verify
1743  * the packet by looking at the included ULP header.
1744  *
1745  * Called before icmp_inbound_error_fanout_v4 is called.
1746  */
1747 static boolean_t
1748 icmp_inbound_verify_v4(mblk_t *mp, icmph_t *icmph, ip_recv_attr_t *ira)
1749 {
1750         ill_t           *ill = ira->ira_ill;
1751         int             hdr_length;
1752         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
1753         conn_t          *connp;
1754         ipha_t          *ipha;  /* Inner IP header */
1755 
1756         ipha = (ipha_t *)&icmph[1];
1757         if ((uchar_t *)ipha + IP_SIMPLE_HDR_LENGTH > mp->b_wptr)
1758                 goto truncated;
1759 
1760         hdr_length = IPH_HDR_LENGTH(ipha);
1761 
1762         if ((IPH_HDR_VERSION(ipha) != IPV4_VERSION))
1763                 goto discard_pkt;
1764 
1765         if (hdr_length < sizeof (ipha_t))
1766                 goto truncated;
1767 
1768         if ((uchar_t *)ipha + hdr_length > mp->b_wptr)
1769                 goto truncated;
1770 
1771         /*
1772          * Stop here for ICMP_REDIRECT.
1773          */
1774         if (icmph->icmph_type == ICMP_REDIRECT)
1775                 return (B_TRUE);
1776 
1777         /*
1778          * ICMP errors only.
1779          */
1780         switch (ipha->ipha_protocol) {
1781         case IPPROTO_UDP:
1782                 /*
1783                  * Verify we have at least ICMP_MIN_TP_HDR_LEN bytes of
1784                  * transport header.
1785                  */
1786                 if ((uchar_t *)ipha + hdr_length + ICMP_MIN_TP_HDR_LEN >
1787                     mp->b_wptr)
1788                         goto truncated;
1789                 break;
1790         case IPPROTO_TCP: {
1791                 tcpha_t         *tcpha;
1792 
1793                 /*
1794                  * Verify we have at least ICMP_MIN_TP_HDR_LEN bytes of
1795                  * transport header.
1796                  */
1797                 if ((uchar_t *)ipha + hdr_length + ICMP_MIN_TP_HDR_LEN >
1798                     mp->b_wptr)
1799                         goto truncated;
1800 
1801                 tcpha = (tcpha_t *)((uchar_t *)ipha + hdr_length);
1802                 connp = ipcl_tcp_lookup_reversed_ipv4(ipha, tcpha, TCPS_LISTEN,
1803                     ipst);
1804                 if (connp == NULL)
1805                         goto discard_pkt;
1806 
1807                 if ((connp->conn_verifyicmp != NULL) &&
1808                     !connp->conn_verifyicmp(connp, tcpha, icmph, NULL, ira)) {
1809                         CONN_DEC_REF(connp);
1810                         goto discard_pkt;
1811                 }
1812                 CONN_DEC_REF(connp);
1813                 break;
1814         }
1815         case IPPROTO_SCTP:
1816                 /*
1817                  * Verify we have at least ICMP_MIN_TP_HDR_LEN bytes of
1818                  * transport header.
1819                  */
1820                 if ((uchar_t *)ipha + hdr_length + ICMP_MIN_TP_HDR_LEN >
1821                     mp->b_wptr)
1822                         goto truncated;
1823                 break;
1824         case IPPROTO_ESP:
1825         case IPPROTO_AH:
1826                 break;
1827         case IPPROTO_ENCAP:
1828                 if ((uchar_t *)ipha + hdr_length + sizeof (ipha_t) >
1829                     mp->b_wptr)
1830                         goto truncated;
1831                 break;
1832         default:
1833                 break;
1834         }
1835 
1836         return (B_TRUE);
1837 
1838 discard_pkt:
1839         /* Bogus ICMP error. */
1840         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
1841         return (B_FALSE);
1842 
1843 truncated:
1844         /* We pulled up everthing already. Must be truncated */
1845         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
1846         ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
1847         return (B_FALSE);
1848 }
1849 
1850 /* Table from RFC 1191 */
1851 static int icmp_frag_size_table[] =
1852 { 32000, 17914, 8166, 4352, 2002, 1496, 1006, 508, 296, 68 };
1853 
1854 /*
1855  * Process received ICMP Packet too big.
1856  * Just handles the DCE create/update, including using the above table of
1857  * PMTU guesses. The caller is responsible for validating the packet before
1858  * passing it in and also to fanout the ICMP error to any matching transport
1859  * conns. Assumes the message has been fully pulled up and verified.
1860  *
1861  * Before getting here, the caller has called icmp_inbound_verify_v4()
1862  * that should have verified with ULP to prevent undoing the changes we're
1863  * going to make to DCE. For example, TCP might have verified that the packet
1864  * which generated error is in the send window.
1865  *
1866  * In some cases modified this MTU in the ICMP header packet; the caller
1867  * should pass to the matching ULP after this returns.
1868  */
1869 static void
1870 icmp_inbound_too_big_v4(icmph_t *icmph, ip_recv_attr_t *ira)
1871 {
1872         dce_t           *dce;
1873         int             old_mtu;
1874         int             mtu, orig_mtu;
1875         ipaddr_t        dst;
1876         boolean_t       disable_pmtud;
1877         ill_t           *ill = ira->ira_ill;
1878         ip_stack_t      *ipst = ill->ill_ipst;
1879         uint_t          hdr_length;
1880         ipha_t          *ipha;
1881 
1882         /* Caller already pulled up everything. */
1883         ipha = (ipha_t *)&icmph[1];
1884         ASSERT(icmph->icmph_type == ICMP_DEST_UNREACHABLE &&
1885             icmph->icmph_code == ICMP_FRAGMENTATION_NEEDED);
1886         ASSERT(ill != NULL);
1887 
1888         hdr_length = IPH_HDR_LENGTH(ipha);
1889 
1890         /*
1891          * We handle path MTU for source routed packets since the DCE
1892          * is looked up using the final destination.
1893          */
1894         dst = ip_get_dst(ipha);
1895 
1896         dce = dce_lookup_and_add_v4(dst, ipst);
1897         if (dce == NULL) {
1898                 /* Couldn't add a unique one - ENOMEM */
1899                 ip1dbg(("icmp_inbound_too_big_v4: no dce for 0x%x\n",
1900                     ntohl(dst)));
1901                 return;
1902         }
1903 
1904         /* Check for MTU discovery advice as described in RFC 1191 */
1905         mtu = ntohs(icmph->icmph_du_mtu);
1906         orig_mtu = mtu;
1907         disable_pmtud = B_FALSE;
1908 
1909         mutex_enter(&dce->dce_lock);
1910         if (dce->dce_flags & DCEF_PMTU)
1911                 old_mtu = dce->dce_pmtu;
1912         else
1913                 old_mtu = ill->ill_mtu;
1914 
1915         if (icmph->icmph_du_zero != 0 || mtu < ipst->ips_ip_pmtu_min) {
1916                 uint32_t length;
1917                 int     i;
1918 
1919                 /*
1920                  * Use the table from RFC 1191 to figure out
1921                  * the next "plateau" based on the length in
1922                  * the original IP packet.
1923                  */
1924                 length = ntohs(ipha->ipha_length);
1925                 DTRACE_PROBE2(ip4__pmtu__guess, dce_t *, dce,
1926                     uint32_t, length);
1927                 if (old_mtu <= length &&
1928                     old_mtu >= length - hdr_length) {
1929                         /*
1930                          * Handle broken BSD 4.2 systems that
1931                          * return the wrong ipha_length in ICMP
1932                          * errors.
1933                          */
1934                         ip1dbg(("Wrong mtu: sent %d, dce %d\n",
1935                             length, old_mtu));
1936                         length -= hdr_length;
1937                 }
1938                 for (i = 0; i < A_CNT(icmp_frag_size_table); i++) {
1939                         if (length > icmp_frag_size_table[i])
1940                                 break;
1941                 }
1942                 if (i == A_CNT(icmp_frag_size_table)) {
1943                         /* Smaller than IP_MIN_MTU! */
1944                         ip1dbg(("Too big for packet size %d\n",
1945                             length));
1946                         disable_pmtud = B_TRUE;
1947                         mtu = ipst->ips_ip_pmtu_min;
1948                 } else {
1949                         mtu = icmp_frag_size_table[i];
1950                         ip1dbg(("Calculated mtu %d, packet size %d, "
1951                             "before %d\n", mtu, length, old_mtu));
1952                         if (mtu < ipst->ips_ip_pmtu_min) {
1953                                 mtu = ipst->ips_ip_pmtu_min;
1954                                 disable_pmtud = B_TRUE;
1955                         }
1956                 }
1957         }
1958         if (disable_pmtud)
1959                 dce->dce_flags |= DCEF_TOO_SMALL_PMTU;
1960         else
1961                 dce->dce_flags &= ~DCEF_TOO_SMALL_PMTU;
1962 
1963         dce->dce_pmtu = MIN(old_mtu, mtu);
1964         /* Prepare to send the new max frag size for the ULP. */
1965         icmph->icmph_du_zero = 0;
1966         icmph->icmph_du_mtu =  htons((uint16_t)dce->dce_pmtu);
1967         DTRACE_PROBE4(ip4__pmtu__change, icmph_t *, icmph, dce_t *,
1968             dce, int, orig_mtu, int, mtu);
1969 
1970         /* We now have a PMTU for sure */
1971         dce->dce_flags |= DCEF_PMTU;
1972         dce->dce_last_change_time = TICK_TO_SEC(ddi_get_lbolt64());
1973         mutex_exit(&dce->dce_lock);
1974         /*
1975          * After dropping the lock the new value is visible to everyone.
1976          * Then we bump the generation number so any cached values reinspect
1977          * the dce_t.
1978          */
1979         dce_increment_generation(dce);
1980         dce_refrele(dce);
1981 }
1982 
1983 /*
1984  * If the packet in error is Self-Encapsulated, icmp_inbound_error_fanout_v4
1985  * calls this function.
1986  */
1987 static mblk_t *
1988 icmp_inbound_self_encap_error_v4(mblk_t *mp, ipha_t *ipha, ipha_t *in_ipha)
1989 {
1990         int length;
1991 
1992         ASSERT(mp->b_datap->db_type == M_DATA);
1993 
1994         /* icmp_inbound_v4 has already pulled up the whole error packet */
1995         ASSERT(mp->b_cont == NULL);
1996 
1997         /*
1998          * The length that we want to overlay is the inner header
1999          * and what follows it.
2000          */
2001         length = msgdsize(mp) - ((uchar_t *)in_ipha - mp->b_rptr);
2002 
2003         /*
2004          * Overlay the inner header and whatever follows it over the
2005          * outer header.
2006          */
2007         bcopy((uchar_t *)in_ipha, (uchar_t *)ipha, length);
2008 
2009         /* Adjust for what we removed */
2010         mp->b_wptr -= (uchar_t *)in_ipha - (uchar_t *)ipha;
2011         return (mp);
2012 }
2013 
2014 /*
2015  * Try to pass the ICMP message upstream in case the ULP cares.
2016  *
2017  * If the packet that caused the ICMP error is secure, we send
2018  * it to AH/ESP to make sure that the attached packet has a
2019  * valid association. ipha in the code below points to the
2020  * IP header of the packet that caused the error.
2021  *
2022  * For IPsec cases, we let the next-layer-up (which has access to
2023  * cached policy on the conn_t, or can query the SPD directly)
2024  * subtract out any IPsec overhead if they must.  We therefore make no
2025  * adjustments here for IPsec overhead.
2026  *
2027  * IFN could have been generated locally or by some router.
2028  *
2029  * LOCAL : ire_send_wire (before calling ipsec_out_process) can call
2030  * icmp_frag_needed/icmp_pkt2big_v6 to generated a local IFN.
2031  *          This happens because IP adjusted its value of MTU on an
2032  *          earlier IFN message and could not tell the upper layer,
2033  *          the new adjusted value of MTU e.g. Packet was encrypted
2034  *          or there was not enough information to fanout to upper
2035  *          layers. Thus on the next outbound datagram, ire_send_wire
2036  *          generates the IFN, where IPsec processing has *not* been
2037  *          done.
2038  *
2039  *          Note that we retain ixa_fragsize across IPsec thus once
2040  *          we have picking ixa_fragsize and entered ipsec_out_process we do
2041  *          no change the fragsize even if the path MTU changes before
2042  *          we reach ip_output_post_ipsec.
2043  *
2044  *          In the local case, IRAF_LOOPBACK will be set indicating
2045  *          that IFN was generated locally.
2046  *
2047  * ROUTER : IFN could be secure or non-secure.
2048  *
2049  *          * SECURE : We use the IPSEC_IN to fanout to AH/ESP if the
2050  *            packet in error has AH/ESP headers to validate the AH/ESP
2051  *            headers. AH/ESP will verify whether there is a valid SA or
2052  *            not and send it back. We will fanout again if we have more
2053  *            data in the packet.
2054  *
2055  *            If the packet in error does not have AH/ESP, we handle it
2056  *            like any other case.
2057  *
2058  *          * NON_SECURE : If the packet in error has AH/ESP headers, we send it
2059  *            up to AH/ESP for validation. AH/ESP will verify whether there is a
2060  *            valid SA or not and send it back. We will fanout again if
2061  *            we have more data in the packet.
2062  *
2063  *            If the packet in error does not have AH/ESP, we handle it
2064  *            like any other case.
2065  *
2066  * The caller must have called icmp_inbound_verify_v4.
2067  */
2068 static void
2069 icmp_inbound_error_fanout_v4(mblk_t *mp, icmph_t *icmph, ip_recv_attr_t *ira)
2070 {
2071         uint16_t        *up;    /* Pointer to ports in ULP header */
2072         uint32_t        ports;  /* reversed ports for fanout */
2073         ipha_t          ripha;  /* With reversed addresses */
2074         ipha_t          *ipha;  /* Inner IP header */
2075         uint_t          hdr_length; /* Inner IP header length */
2076         tcpha_t         *tcpha;
2077         conn_t          *connp;
2078         ill_t           *ill = ira->ira_ill;
2079         ip_stack_t      *ipst = ill->ill_ipst;
2080         ipsec_stack_t   *ipss = ipst->ips_netstack->netstack_ipsec;
2081         ill_t           *rill = ira->ira_rill;
2082 
2083         /* Caller already pulled up everything. */
2084         ipha = (ipha_t *)&icmph[1];
2085         ASSERT((uchar_t *)&ipha[1] <= mp->b_wptr);
2086         ASSERT(mp->b_cont == NULL);
2087 
2088         hdr_length = IPH_HDR_LENGTH(ipha);
2089         ira->ira_protocol = ipha->ipha_protocol;
2090 
2091         /*
2092          * We need a separate IP header with the source and destination
2093          * addresses reversed to do fanout/classification because the ipha in
2094          * the ICMP error is in the form we sent it out.
2095          */
2096         ripha.ipha_src = ipha->ipha_dst;
2097         ripha.ipha_dst = ipha->ipha_src;
2098         ripha.ipha_protocol = ipha->ipha_protocol;
2099         ripha.ipha_version_and_hdr_length = ipha->ipha_version_and_hdr_length;
2100 
2101         ip2dbg(("icmp_inbound_error_v4: proto %d %x to %x: %d/%d\n",
2102             ripha.ipha_protocol, ntohl(ipha->ipha_src),
2103             ntohl(ipha->ipha_dst),
2104             icmph->icmph_type, icmph->icmph_code));
2105 
2106         switch (ipha->ipha_protocol) {
2107         case IPPROTO_UDP:
2108                 up = (uint16_t *)((uchar_t *)ipha + hdr_length);
2109 
2110                 /* Attempt to find a client stream based on port. */
2111                 ip2dbg(("icmp_inbound_error_v4: UDP ports %d to %d\n",
2112                     ntohs(up[0]), ntohs(up[1])));
2113 
2114                 /* Note that we send error to all matches. */
2115                 ira->ira_flags |= IRAF_ICMP_ERROR;
2116                 ip_fanout_udp_multi_v4(mp, &ripha, up[0], up[1], ira);
2117                 ira->ira_flags &= ~IRAF_ICMP_ERROR;
2118                 return;
2119 
2120         case IPPROTO_TCP:
2121                 /*
2122                  * Find a TCP client stream for this packet.
2123                  * Note that we do a reverse lookup since the header is
2124                  * in the form we sent it out.
2125                  */
2126                 tcpha = (tcpha_t *)((uchar_t *)ipha + hdr_length);
2127                 connp = ipcl_tcp_lookup_reversed_ipv4(ipha, tcpha, TCPS_LISTEN,
2128                     ipst);
2129                 if (connp == NULL)
2130                         goto discard_pkt;
2131 
2132                 if (CONN_INBOUND_POLICY_PRESENT(connp, ipss) ||
2133                     (ira->ira_flags & IRAF_IPSEC_SECURE)) {
2134                         mp = ipsec_check_inbound_policy(mp, connp,
2135                             ipha, NULL, ira);
2136                         if (mp == NULL) {
2137                                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
2138                                 /* Note that mp is NULL */
2139                                 ip_drop_input("ipIfStatsInDiscards", mp, ill);
2140                                 CONN_DEC_REF(connp);
2141                                 return;
2142                         }
2143                 }
2144 
2145                 ira->ira_flags |= IRAF_ICMP_ERROR;
2146                 ira->ira_ill = ira->ira_rill = NULL;
2147                 if (IPCL_IS_TCP(connp)) {
2148                         SQUEUE_ENTER_ONE(connp->conn_sqp, mp,
2149                             connp->conn_recvicmp, connp, ira, SQ_FILL,
2150                             SQTAG_TCP_INPUT_ICMP_ERR);
2151                 } else {
2152                         /* Not TCP; must be SOCK_RAW, IPPROTO_TCP */
2153                         (connp->conn_recv)(connp, mp, NULL, ira);
2154                         CONN_DEC_REF(connp);
2155                 }
2156                 ira->ira_ill = ill;
2157                 ira->ira_rill = rill;
2158                 ira->ira_flags &= ~IRAF_ICMP_ERROR;
2159                 return;
2160 
2161         case IPPROTO_SCTP:
2162                 up = (uint16_t *)((uchar_t *)ipha + hdr_length);
2163                 /* Find a SCTP client stream for this packet. */
2164                 ((uint16_t *)&ports)[0] = up[1];
2165                 ((uint16_t *)&ports)[1] = up[0];
2166 
2167                 ira->ira_flags |= IRAF_ICMP_ERROR;
2168                 ip_fanout_sctp(mp, &ripha, NULL, ports, ira);
2169                 ira->ira_flags &= ~IRAF_ICMP_ERROR;
2170                 return;
2171 
2172         case IPPROTO_ESP:
2173         case IPPROTO_AH:
2174                 if (!ipsec_loaded(ipss)) {
2175                         ip_proto_not_sup(mp, ira);
2176                         return;
2177                 }
2178 
2179                 if (ipha->ipha_protocol == IPPROTO_ESP)
2180                         mp = ipsecesp_icmp_error(mp, ira);
2181                 else
2182                         mp = ipsecah_icmp_error(mp, ira);
2183                 if (mp == NULL)
2184                         return;
2185 
2186                 /* Just in case ipsec didn't preserve the NULL b_cont */
2187                 if (mp->b_cont != NULL) {
2188                         if (!pullupmsg(mp, -1))
2189                                 goto discard_pkt;
2190                 }
2191 
2192                 /*
2193                  * Note that ira_pktlen and ira_ip_hdr_length are no longer
2194                  * correct, but we don't use them any more here.
2195                  *
2196                  * If succesful, the mp has been modified to not include
2197                  * the ESP/AH header so we can fanout to the ULP's icmp
2198                  * error handler.
2199                  */
2200                 if (mp->b_wptr - mp->b_rptr < IP_SIMPLE_HDR_LENGTH)
2201                         goto truncated;
2202 
2203                 /* Verify the modified message before any further processes. */
2204                 ipha = (ipha_t *)mp->b_rptr;
2205                 hdr_length = IPH_HDR_LENGTH(ipha);
2206                 icmph = (icmph_t *)&mp->b_rptr[hdr_length];
2207                 if (!icmp_inbound_verify_v4(mp, icmph, ira)) {
2208                         freemsg(mp);
2209                         return;
2210                 }
2211 
2212                 icmp_inbound_error_fanout_v4(mp, icmph, ira);
2213                 return;
2214 
2215         case IPPROTO_ENCAP: {
2216                 /* Look for self-encapsulated packets that caused an error */
2217                 ipha_t *in_ipha;
2218 
2219                 /*
2220                  * Caller has verified that length has to be
2221                  * at least the size of IP header.
2222                  */
2223                 ASSERT(hdr_length >= sizeof (ipha_t));
2224                 /*
2225                  * Check the sanity of the inner IP header like
2226                  * we did for the outer header.
2227                  */
2228                 in_ipha = (ipha_t *)((uchar_t *)ipha + hdr_length);
2229                 if ((IPH_HDR_VERSION(in_ipha) != IPV4_VERSION)) {
2230                         goto discard_pkt;
2231                 }
2232                 if (IPH_HDR_LENGTH(in_ipha) < sizeof (ipha_t)) {
2233                         goto discard_pkt;
2234                 }
2235                 /* Check for Self-encapsulated tunnels */
2236                 if (in_ipha->ipha_src == ipha->ipha_src &&
2237                     in_ipha->ipha_dst == ipha->ipha_dst) {
2238 
2239                         mp = icmp_inbound_self_encap_error_v4(mp, ipha,
2240                             in_ipha);
2241                         if (mp == NULL)
2242                                 goto discard_pkt;
2243 
2244                         /*
2245                          * Just in case self_encap didn't preserve the NULL
2246                          * b_cont
2247                          */
2248                         if (mp->b_cont != NULL) {
2249                                 if (!pullupmsg(mp, -1))
2250                                         goto discard_pkt;
2251                         }
2252                         /*
2253                          * Note that ira_pktlen and ira_ip_hdr_length are no
2254                          * longer correct, but we don't use them any more here.
2255                          */
2256                         if (mp->b_wptr - mp->b_rptr < IP_SIMPLE_HDR_LENGTH)
2257                                 goto truncated;
2258 
2259                         /*
2260                          * Verify the modified message before any further
2261                          * processes.
2262                          */
2263                         ipha = (ipha_t *)mp->b_rptr;
2264                         hdr_length = IPH_HDR_LENGTH(ipha);
2265                         icmph = (icmph_t *)&mp->b_rptr[hdr_length];
2266                         if (!icmp_inbound_verify_v4(mp, icmph, ira)) {
2267                                 freemsg(mp);
2268                                 return;
2269                         }
2270 
2271                         /*
2272                          * The packet in error is self-encapsualted.
2273                          * And we are finding it further encapsulated
2274                          * which we could not have possibly generated.
2275                          */
2276                         if (ipha->ipha_protocol == IPPROTO_ENCAP) {
2277                                 goto discard_pkt;
2278                         }
2279                         icmp_inbound_error_fanout_v4(mp, icmph, ira);
2280                         return;
2281                 }
2282                 /* No self-encapsulated */
2283                 /* FALLTHRU */
2284         }
2285         case IPPROTO_IPV6:
2286                 if ((connp = ipcl_iptun_classify_v4(&ripha.ipha_src,
2287                     &ripha.ipha_dst, ipst)) != NULL) {
2288                         ira->ira_flags |= IRAF_ICMP_ERROR;
2289                         connp->conn_recvicmp(connp, mp, NULL, ira);
2290                         CONN_DEC_REF(connp);
2291                         ira->ira_flags &= ~IRAF_ICMP_ERROR;
2292                         return;
2293                 }
2294                 /*
2295                  * No IP tunnel is interested, fallthrough and see
2296                  * if a raw socket will want it.
2297                  */
2298                 /* FALLTHRU */
2299         default:
2300                 ira->ira_flags |= IRAF_ICMP_ERROR;
2301                 ip_fanout_proto_v4(mp, &ripha, ira);
2302                 ira->ira_flags &= ~IRAF_ICMP_ERROR;
2303                 return;
2304         }
2305         /* NOTREACHED */
2306 discard_pkt:
2307         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
2308         ip1dbg(("icmp_inbound_error_fanout_v4: drop pkt\n"));
2309         ip_drop_input("ipIfStatsInDiscards", mp, ill);
2310         freemsg(mp);
2311         return;
2312 
2313 truncated:
2314         /* We pulled up everthing already. Must be truncated */
2315         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
2316         ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
2317         freemsg(mp);
2318 }
2319 
2320 /*
2321  * Common IP options parser.
2322  *
2323  * Setup routine: fill in *optp with options-parsing state, then
2324  * tail-call ipoptp_next to return the first option.
2325  */
2326 uint8_t
2327 ipoptp_first(ipoptp_t *optp, ipha_t *ipha)
2328 {
2329         uint32_t totallen; /* total length of all options */
2330 
2331         totallen = ipha->ipha_version_and_hdr_length -
2332             (uint8_t)((IP_VERSION << 4) + IP_SIMPLE_HDR_LENGTH_IN_WORDS);
2333         totallen <<= 2;
2334         optp->ipoptp_next = (uint8_t *)(&ipha[1]);
2335         optp->ipoptp_end = optp->ipoptp_next + totallen;
2336         optp->ipoptp_flags = 0;
2337         return (ipoptp_next(optp));
2338 }
2339 
2340 /* Like above but without an ipha_t */
2341 uint8_t
2342 ipoptp_first2(ipoptp_t *optp, uint32_t totallen, uint8_t *opt)
2343 {
2344         optp->ipoptp_next = opt;
2345         optp->ipoptp_end = optp->ipoptp_next + totallen;
2346         optp->ipoptp_flags = 0;
2347         return (ipoptp_next(optp));
2348 }
2349 
2350 /*
2351  * Common IP options parser: extract next option.
2352  */
2353 uint8_t
2354 ipoptp_next(ipoptp_t *optp)
2355 {
2356         uint8_t *end = optp->ipoptp_end;
2357         uint8_t *cur = optp->ipoptp_next;
2358         uint8_t opt, len, pointer;
2359 
2360         /*
2361          * If cur > end already, then the ipoptp_end or ipoptp_next pointer
2362          * has been corrupted.
2363          */
2364         ASSERT(cur <= end);
2365 
2366         if (cur == end)
2367                 return (IPOPT_EOL);
2368 
2369         opt = cur[IPOPT_OPTVAL];
2370 
2371         /*
2372          * Skip any NOP options.
2373          */
2374         while (opt == IPOPT_NOP) {
2375                 cur++;
2376                 if (cur == end)
2377                         return (IPOPT_EOL);
2378                 opt = cur[IPOPT_OPTVAL];
2379         }
2380 
2381         if (opt == IPOPT_EOL)
2382                 return (IPOPT_EOL);
2383 
2384         /*
2385          * Option requiring a length.
2386          */
2387         if ((cur + 1) >= end) {
2388                 optp->ipoptp_flags |= IPOPTP_ERROR;
2389                 return (IPOPT_EOL);
2390         }
2391         len = cur[IPOPT_OLEN];
2392         if (len < 2) {
2393                 optp->ipoptp_flags |= IPOPTP_ERROR;
2394                 return (IPOPT_EOL);
2395         }
2396         optp->ipoptp_cur = cur;
2397         optp->ipoptp_len = len;
2398         optp->ipoptp_next = cur + len;
2399         if (cur + len > end) {
2400                 optp->ipoptp_flags |= IPOPTP_ERROR;
2401                 return (IPOPT_EOL);
2402         }
2403 
2404         /*
2405          * For the options which require a pointer field, make sure
2406          * its there, and make sure it points to either something
2407          * inside this option, or the end of the option.
2408          */
2409         switch (opt) {
2410         case IPOPT_RR:
2411         case IPOPT_TS:
2412         case IPOPT_LSRR:
2413         case IPOPT_SSRR:
2414                 if (len <= IPOPT_OFFSET) {
2415                         optp->ipoptp_flags |= IPOPTP_ERROR;
2416                         return (opt);
2417                 }
2418                 pointer = cur[IPOPT_OFFSET];
2419                 if (pointer - 1 > len) {
2420                         optp->ipoptp_flags |= IPOPTP_ERROR;
2421                         return (opt);
2422                 }
2423                 break;
2424         }
2425 
2426         /*
2427          * Sanity check the pointer field based on the type of the
2428          * option.
2429          */
2430         switch (opt) {
2431         case IPOPT_RR:
2432         case IPOPT_SSRR:
2433         case IPOPT_LSRR:
2434                 if (pointer < IPOPT_MINOFF_SR)
2435                         optp->ipoptp_flags |= IPOPTP_ERROR;
2436                 break;
2437         case IPOPT_TS:
2438                 if (pointer < IPOPT_MINOFF_IT)
2439                         optp->ipoptp_flags |= IPOPTP_ERROR;
2440                 /*
2441                  * Note that the Internet Timestamp option also
2442                  * contains two four bit fields (the Overflow field,
2443                  * and the Flag field), which follow the pointer
2444                  * field.  We don't need to check that these fields
2445                  * fall within the length of the option because this
2446                  * was implicitely done above.  We've checked that the
2447                  * pointer value is at least IPOPT_MINOFF_IT, and that
2448                  * it falls within the option.  Since IPOPT_MINOFF_IT >
2449                  * IPOPT_POS_OV_FLG, we don't need the explicit check.
2450                  */
2451                 ASSERT(len > IPOPT_POS_OV_FLG);
2452                 break;
2453         }
2454 
2455         return (opt);
2456 }
2457 
2458 /*
2459  * Use the outgoing IP header to create an IP_OPTIONS option the way
2460  * it was passed down from the application.
2461  *
2462  * This is compatible with BSD in that it returns
2463  * the reverse source route with the final destination
2464  * as the last entry. The first 4 bytes of the option
2465  * will contain the final destination.
2466  */
2467 int
2468 ip_opt_get_user(conn_t *connp, uchar_t *buf)
2469 {
2470         ipoptp_t        opts;
2471         uchar_t         *opt;
2472         uint8_t         optval;
2473         uint8_t         optlen;
2474         uint32_t        len = 0;
2475         uchar_t         *buf1 = buf;
2476         uint32_t        totallen;
2477         ipaddr_t        dst;
2478         ip_pkt_t        *ipp = &connp->conn_xmit_ipp;
2479 
2480         if (!(ipp->ipp_fields & IPPF_IPV4_OPTIONS))
2481                 return (0);
2482 
2483         totallen = ipp->ipp_ipv4_options_len;
2484         if (totallen & 0x3)
2485                 return (0);
2486 
2487         buf += IP_ADDR_LEN;     /* Leave room for final destination */
2488         len += IP_ADDR_LEN;
2489         bzero(buf1, IP_ADDR_LEN);
2490 
2491         dst = connp->conn_faddr_v4;
2492 
2493         for (optval = ipoptp_first2(&opts, totallen, ipp->ipp_ipv4_options);
2494             optval != IPOPT_EOL;
2495             optval = ipoptp_next(&opts)) {
2496                 int     off;
2497 
2498                 opt = opts.ipoptp_cur;
2499                 if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
2500                         break;
2501                 }
2502                 optlen = opts.ipoptp_len;
2503 
2504                 switch (optval) {
2505                 case IPOPT_SSRR:
2506                 case IPOPT_LSRR:
2507 
2508                         /*
2509                          * Insert destination as the first entry in the source
2510                          * route and move down the entries on step.
2511                          * The last entry gets placed at buf1.
2512                          */
2513                         buf[IPOPT_OPTVAL] = optval;
2514                         buf[IPOPT_OLEN] = optlen;
2515                         buf[IPOPT_OFFSET] = optlen;
2516 
2517                         off = optlen - IP_ADDR_LEN;
2518                         if (off < 0) {
2519                                 /* No entries in source route */
2520                                 break;
2521                         }
2522                         /* Last entry in source route if not already set */
2523                         if (dst == INADDR_ANY)
2524                                 bcopy(opt + off, buf1, IP_ADDR_LEN);
2525                         off -= IP_ADDR_LEN;
2526 
2527                         while (off > 0) {
2528                                 bcopy(opt + off,
2529                                     buf + off + IP_ADDR_LEN,
2530                                     IP_ADDR_LEN);
2531                                 off -= IP_ADDR_LEN;
2532                         }
2533                         /* ipha_dst into first slot */
2534                         bcopy(&dst, buf + off + IP_ADDR_LEN,
2535                             IP_ADDR_LEN);
2536                         buf += optlen;
2537                         len += optlen;
2538                         break;
2539 
2540                 default:
2541                         bcopy(opt, buf, optlen);
2542                         buf += optlen;
2543                         len += optlen;
2544                         break;
2545                 }
2546         }
2547 done:
2548         /* Pad the resulting options */
2549         while (len & 0x3) {
2550                 *buf++ = IPOPT_EOL;
2551                 len++;
2552         }
2553         return (len);
2554 }
2555 
2556 /*
2557  * Update any record route or timestamp options to include this host.
2558  * Reverse any source route option.
2559  * This routine assumes that the options are well formed i.e. that they
2560  * have already been checked.
2561  */
2562 static void
2563 icmp_options_update(ipha_t *ipha)
2564 {
2565         ipoptp_t        opts;
2566         uchar_t         *opt;
2567         uint8_t         optval;
2568         ipaddr_t        src;            /* Our local address */
2569         ipaddr_t        dst;
2570 
2571         ip2dbg(("icmp_options_update\n"));
2572         src = ipha->ipha_src;
2573         dst = ipha->ipha_dst;
2574 
2575         for (optval = ipoptp_first(&opts, ipha);
2576             optval != IPOPT_EOL;
2577             optval = ipoptp_next(&opts)) {
2578                 ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
2579                 opt = opts.ipoptp_cur;
2580                 ip2dbg(("icmp_options_update: opt %d, len %d\n",
2581                     optval, opts.ipoptp_len));
2582                 switch (optval) {
2583                         int off1, off2;
2584                 case IPOPT_SSRR:
2585                 case IPOPT_LSRR:
2586                         /*
2587                          * Reverse the source route.  The first entry
2588                          * should be the next to last one in the current
2589                          * source route (the last entry is our address).
2590                          * The last entry should be the final destination.
2591                          */
2592                         off1 = IPOPT_MINOFF_SR - 1;
2593                         off2 = opt[IPOPT_OFFSET] - IP_ADDR_LEN - 1;
2594                         if (off2 < 0) {
2595                                 /* No entries in source route */
2596                                 ip1dbg((
2597                                     "icmp_options_update: bad src route\n"));
2598                                 break;
2599                         }
2600                         bcopy((char *)opt + off2, &dst, IP_ADDR_LEN);
2601                         bcopy(&ipha->ipha_dst, (char *)opt + off2, IP_ADDR_LEN);
2602                         bcopy(&dst, &ipha->ipha_dst, IP_ADDR_LEN);
2603                         off2 -= IP_ADDR_LEN;
2604 
2605                         while (off1 < off2) {
2606                                 bcopy((char *)opt + off1, &src, IP_ADDR_LEN);
2607                                 bcopy((char *)opt + off2, (char *)opt + off1,
2608                                     IP_ADDR_LEN);
2609                                 bcopy(&src, (char *)opt + off2, IP_ADDR_LEN);
2610                                 off1 += IP_ADDR_LEN;
2611                                 off2 -= IP_ADDR_LEN;
2612                         }
2613                         opt[IPOPT_OFFSET] = IPOPT_MINOFF_SR;
2614                         break;
2615                 }
2616         }
2617 }
2618 
2619 /*
2620  * Process received ICMP Redirect messages.
2621  * Assumes the caller has verified that the headers are in the pulled up mblk.
2622  * Consumes mp.
2623  */
2624 static void
2625 icmp_redirect_v4(mblk_t *mp, ipha_t *ipha, icmph_t *icmph, ip_recv_attr_t *ira)
2626 {
2627         ire_t           *ire, *nire;
2628         ire_t           *prev_ire;
2629         ipaddr_t        src, dst, gateway;
2630         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
2631         ipha_t          *inner_ipha;    /* Inner IP header */
2632 
2633         /* Caller already pulled up everything. */
2634         inner_ipha = (ipha_t *)&icmph[1];
2635         src = ipha->ipha_src;
2636         dst = inner_ipha->ipha_dst;
2637         gateway = icmph->icmph_rd_gateway;
2638         /* Make sure the new gateway is reachable somehow. */
2639         ire = ire_ftable_lookup_v4(gateway, 0, 0, IRE_ONLINK, NULL,
2640             ALL_ZONES, NULL, MATCH_IRE_TYPE, 0, ipst, NULL);
2641         /*
2642          * Make sure we had a route for the dest in question and that
2643          * that route was pointing to the old gateway (the source of the
2644          * redirect packet.)
2645          * We do longest match and then compare ire_gateway_addr below.
2646          */
2647         prev_ire = ire_ftable_lookup_v4(dst, 0, 0, 0, NULL, ALL_ZONES,
2648             NULL, MATCH_IRE_DSTONLY, 0, ipst, NULL);
2649         /*
2650          * Check that
2651          *      the redirect was not from ourselves
2652          *      the new gateway and the old gateway are directly reachable
2653          */
2654         if (prev_ire == NULL || ire == NULL ||
2655             (prev_ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK)) ||
2656             (prev_ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) ||
2657             !(ire->ire_type & IRE_IF_ALL) ||
2658             prev_ire->ire_gateway_addr != src) {
2659                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInBadRedirects);
2660                 ip_drop_input("icmpInBadRedirects - ire", mp, ira->ira_ill);
2661                 freemsg(mp);
2662                 if (ire != NULL)
2663                         ire_refrele(ire);
2664                 if (prev_ire != NULL)
2665                         ire_refrele(prev_ire);
2666                 return;
2667         }
2668 
2669         ire_refrele(prev_ire);
2670         ire_refrele(ire);
2671 
2672         /*
2673          * TODO: more precise handling for cases 0, 2, 3, the latter two
2674          * require TOS routing
2675          */
2676         switch (icmph->icmph_code) {
2677         case 0:
2678         case 1:
2679                 /* TODO: TOS specificity for cases 2 and 3 */
2680         case 2:
2681         case 3:
2682                 break;
2683         default:
2684                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInBadRedirects);
2685                 ip_drop_input("icmpInBadRedirects - code", mp, ira->ira_ill);
2686                 freemsg(mp);
2687                 return;
2688         }
2689         /*
2690          * Create a Route Association.  This will allow us to remember that
2691          * someone we believe told us to use the particular gateway.
2692          */
2693         ire = ire_create(
2694             (uchar_t *)&dst,                        /* dest addr */
2695             (uchar_t *)&ip_g_all_ones,              /* mask */
2696             (uchar_t *)&gateway,            /* gateway addr */
2697             IRE_HOST,
2698             NULL,                               /* ill */
2699             ALL_ZONES,
2700             (RTF_DYNAMIC | RTF_GATEWAY | RTF_HOST),
2701             NULL,                               /* tsol_gc_t */
2702             ipst);
2703 
2704         if (ire == NULL) {
2705                 freemsg(mp);
2706                 return;
2707         }
2708         nire = ire_add(ire);
2709         /* Check if it was a duplicate entry */
2710         if (nire != NULL && nire != ire) {
2711                 ASSERT(nire->ire_identical_ref > 1);
2712                 ire_delete(nire);
2713                 ire_refrele(nire);
2714                 nire = NULL;
2715         }
2716         ire = nire;
2717         if (ire != NULL) {
2718                 ire_refrele(ire);               /* Held in ire_add */
2719 
2720                 /* tell routing sockets that we received a redirect */
2721                 ip_rts_change(RTM_REDIRECT, dst, gateway, IP_HOST_MASK, 0, src,
2722                     (RTF_DYNAMIC | RTF_GATEWAY | RTF_HOST), 0,
2723                     (RTA_DST | RTA_GATEWAY | RTA_NETMASK | RTA_AUTHOR), ipst);
2724         }
2725 
2726         /*
2727          * Delete any existing IRE_HOST type redirect ires for this destination.
2728          * This together with the added IRE has the effect of
2729          * modifying an existing redirect.
2730          */
2731         prev_ire = ire_ftable_lookup_v4(dst, 0, src, IRE_HOST, NULL,
2732             ALL_ZONES, NULL, (MATCH_IRE_GW | MATCH_IRE_TYPE), 0, ipst, NULL);
2733         if (prev_ire != NULL) {
2734                 if (prev_ire ->ire_flags & RTF_DYNAMIC)
2735                         ire_delete(prev_ire);
2736                 ire_refrele(prev_ire);
2737         }
2738 
2739         freemsg(mp);
2740 }
2741 
2742 /*
2743  * Generate an ICMP parameter problem message.
2744  * When called from ip_output side a minimal ip_recv_attr_t needs to be
2745  * constructed by the caller.
2746  */
2747 static void
2748 icmp_param_problem(mblk_t *mp, uint8_t ptr, ip_recv_attr_t *ira)
2749 {
2750         icmph_t icmph;
2751         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
2752 
2753         mp = icmp_pkt_err_ok(mp, ira);
2754         if (mp == NULL)
2755                 return;
2756 
2757         bzero(&icmph, sizeof (icmph_t));
2758         icmph.icmph_type = ICMP_PARAM_PROBLEM;
2759         icmph.icmph_pp_ptr = ptr;
2760         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutParmProbs);
2761         icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
2762 }
2763 
2764 /*
2765  * Build and ship an IPv4 ICMP message using the packet data in mp, and
2766  * the ICMP header pointed to by "stuff".  (May be called as writer.)
2767  * Note: assumes that icmp_pkt_err_ok has been called to verify that
2768  * an icmp error packet can be sent.
2769  * Assigns an appropriate source address to the packet. If ipha_dst is
2770  * one of our addresses use it for source. Otherwise let ip_output_simple
2771  * pick the source address.
2772  */
2773 static void
2774 icmp_pkt(mblk_t *mp, void *stuff, size_t len, ip_recv_attr_t *ira)
2775 {
2776         ipaddr_t dst;
2777         icmph_t *icmph;
2778         ipha_t  *ipha;
2779         uint_t  len_needed;
2780         size_t  msg_len;
2781         mblk_t  *mp1;
2782         ipaddr_t src;
2783         ire_t   *ire;
2784         ip_xmit_attr_t ixas;
2785         ip_stack_t *ipst = ira->ira_ill->ill_ipst;
2786 
2787         ipha = (ipha_t *)mp->b_rptr;
2788 
2789         bzero(&ixas, sizeof (ixas));
2790         ixas.ixa_flags = IXAF_BASIC_SIMPLE_V4;
2791         ixas.ixa_zoneid = ira->ira_zoneid;
2792         ixas.ixa_ifindex = 0;
2793         ixas.ixa_ipst = ipst;
2794         ixas.ixa_cred = kcred;
2795         ixas.ixa_cpid = NOPID;
2796         ixas.ixa_tsl = ira->ira_tsl; /* Behave as a multi-level responder */
2797         ixas.ixa_multicast_ttl = IP_DEFAULT_MULTICAST_TTL;
2798 
2799         if (ira->ira_flags & IRAF_IPSEC_SECURE) {
2800                 /*
2801                  * Apply IPsec based on how IPsec was applied to
2802                  * the packet that had the error.
2803                  *
2804                  * If it was an outbound packet that caused the ICMP
2805                  * error, then the caller will have setup the IRA
2806                  * appropriately.
2807                  */
2808                 if (!ipsec_in_to_out(ira, &ixas, mp, ipha, NULL)) {
2809                         BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsOutDiscards);
2810                         /* Note: mp already consumed and ip_drop_packet done */
2811                         return;
2812                 }
2813         } else {
2814                 /*
2815                  * This is in clear. The icmp message we are building
2816                  * here should go out in clear, independent of our policy.
2817                  */
2818                 ixas.ixa_flags |= IXAF_NO_IPSEC;
2819         }
2820 
2821         /* Remember our eventual destination */
2822         dst = ipha->ipha_src;
2823 
2824         /*
2825          * If the packet was for one of our unicast addresses, make
2826          * sure we respond with that as the source. Otherwise
2827          * have ip_output_simple pick the source address.
2828          */
2829         ire = ire_ftable_lookup_v4(ipha->ipha_dst, 0, 0,
2830             (IRE_LOCAL|IRE_LOOPBACK), NULL, ira->ira_zoneid, NULL,
2831             MATCH_IRE_TYPE|MATCH_IRE_ZONEONLY, 0, ipst, NULL);
2832         if (ire != NULL) {
2833                 ire_refrele(ire);
2834                 src = ipha->ipha_dst;
2835         } else {
2836                 src = INADDR_ANY;
2837                 ixas.ixa_flags |= IXAF_SET_SOURCE;
2838         }
2839 
2840         /*
2841          * Check if we can send back more then 8 bytes in addition to
2842          * the IP header.  We try to send 64 bytes of data and the internal
2843          * header in the special cases of ipv4 encapsulated ipv4 or ipv6.
2844          */
2845         len_needed = IPH_HDR_LENGTH(ipha);
2846         if (ipha->ipha_protocol == IPPROTO_ENCAP ||
2847             ipha->ipha_protocol == IPPROTO_IPV6) {
2848                 if (!pullupmsg(mp, -1)) {
2849                         BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsOutDiscards);
2850                         ip_drop_output("ipIfStatsOutDiscards", mp, NULL);
2851                         freemsg(mp);
2852                         return;
2853                 }
2854                 ipha = (ipha_t *)mp->b_rptr;
2855 
2856                 if (ipha->ipha_protocol == IPPROTO_ENCAP) {
2857                         len_needed += IPH_HDR_LENGTH(((uchar_t *)ipha +
2858                             len_needed));
2859                 } else {
2860                         ip6_t *ip6h = (ip6_t *)((uchar_t *)ipha + len_needed);
2861 
2862                         ASSERT(ipha->ipha_protocol == IPPROTO_IPV6);
2863                         len_needed += ip_hdr_length_v6(mp, ip6h);
2864                 }
2865         }
2866         len_needed += ipst->ips_ip_icmp_return;
2867         msg_len = msgdsize(mp);
2868         if (msg_len > len_needed) {
2869                 (void) adjmsg(mp, len_needed - msg_len);
2870                 msg_len = len_needed;
2871         }
2872         mp1 = allocb(sizeof (icmp_ipha) + len, BPRI_MED);
2873         if (mp1 == NULL) {
2874                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutErrors);
2875                 freemsg(mp);
2876                 return;
2877         }
2878         mp1->b_cont = mp;
2879         mp = mp1;
2880 
2881         /*
2882          * Set IXAF_TRUSTED_ICMP so we can let the ICMP messages this
2883          * node generates be accepted in peace by all on-host destinations.
2884          * If we do NOT assume that all on-host destinations trust
2885          * self-generated ICMP messages, then rework here, ip6.c, and spd.c.
2886          * (Look for IXAF_TRUSTED_ICMP).
2887          */
2888         ixas.ixa_flags |= IXAF_TRUSTED_ICMP;
2889 
2890         ipha = (ipha_t *)mp->b_rptr;
2891         mp1->b_wptr = (uchar_t *)ipha + (sizeof (icmp_ipha) + len);
2892         *ipha = icmp_ipha;
2893         ipha->ipha_src = src;
2894         ipha->ipha_dst = dst;
2895         ipha->ipha_ttl = ipst->ips_ip_def_ttl;
2896         msg_len += sizeof (icmp_ipha) + len;
2897         if (msg_len > IP_MAXPACKET) {
2898                 (void) adjmsg(mp, IP_MAXPACKET - msg_len);
2899                 msg_len = IP_MAXPACKET;
2900         }
2901         ipha->ipha_length = htons((uint16_t)msg_len);
2902         icmph = (icmph_t *)&ipha[1];
2903         bcopy(stuff, icmph, len);
2904         icmph->icmph_checksum = 0;
2905         icmph->icmph_checksum = IP_CSUM(mp, (int32_t)sizeof (ipha_t), 0);
2906         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutMsgs);
2907 
2908         (void) ip_output_simple(mp, &ixas);
2909         ixa_cleanup(&ixas);
2910 }
2911 
2912 /*
2913  * Determine if an ICMP error packet can be sent given the rate limit.
2914  * The limit consists of an average frequency (icmp_pkt_err_interval measured
2915  * in milliseconds) and a burst size. Burst size number of packets can
2916  * be sent arbitrarely closely spaced.
2917  * The state is tracked using two variables to implement an approximate
2918  * token bucket filter:
2919  *      icmp_pkt_err_last - lbolt value when the last burst started
2920  *      icmp_pkt_err_sent - number of packets sent in current burst
2921  */
2922 boolean_t
2923 icmp_err_rate_limit(ip_stack_t *ipst)
2924 {
2925         clock_t now = TICK_TO_MSEC(ddi_get_lbolt());
2926         uint_t refilled; /* Number of packets refilled in tbf since last */
2927         /* Guard against changes by loading into local variable */
2928         uint_t err_interval = ipst->ips_ip_icmp_err_interval;
2929 
2930         if (err_interval == 0)
2931                 return (B_FALSE);
2932 
2933         if (ipst->ips_icmp_pkt_err_last > now) {
2934                 /* 100HZ lbolt in ms for 32bit arch wraps every 49.7 days */
2935                 ipst->ips_icmp_pkt_err_last = 0;
2936                 ipst->ips_icmp_pkt_err_sent = 0;
2937         }
2938         /*
2939          * If we are in a burst update the token bucket filter.
2940          * Update the "last" time to be close to "now" but make sure
2941          * we don't loose precision.
2942          */
2943         if (ipst->ips_icmp_pkt_err_sent != 0) {
2944                 refilled = (now - ipst->ips_icmp_pkt_err_last)/err_interval;
2945                 if (refilled > ipst->ips_icmp_pkt_err_sent) {
2946                         ipst->ips_icmp_pkt_err_sent = 0;
2947                 } else {
2948                         ipst->ips_icmp_pkt_err_sent -= refilled;
2949                         ipst->ips_icmp_pkt_err_last += refilled * err_interval;
2950                 }
2951         }
2952         if (ipst->ips_icmp_pkt_err_sent == 0) {
2953                 /* Start of new burst */
2954                 ipst->ips_icmp_pkt_err_last = now;
2955         }
2956         if (ipst->ips_icmp_pkt_err_sent < ipst->ips_ip_icmp_err_burst) {
2957                 ipst->ips_icmp_pkt_err_sent++;
2958                 ip1dbg(("icmp_err_rate_limit: %d sent in burst\n",
2959                     ipst->ips_icmp_pkt_err_sent));
2960                 return (B_FALSE);
2961         }
2962         ip1dbg(("icmp_err_rate_limit: dropped\n"));
2963         return (B_TRUE);
2964 }
2965 
2966 /*
2967  * Check if it is ok to send an IPv4 ICMP error packet in
2968  * response to the IPv4 packet in mp.
2969  * Free the message and return null if no
2970  * ICMP error packet should be sent.
2971  */
2972 static mblk_t *
2973 icmp_pkt_err_ok(mblk_t *mp, ip_recv_attr_t *ira)
2974 {
2975         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
2976         icmph_t *icmph;
2977         ipha_t  *ipha;
2978         uint_t  len_needed;
2979 
2980         if (!mp)
2981                 return (NULL);
2982         ipha = (ipha_t *)mp->b_rptr;
2983         if (ip_csum_hdr(ipha)) {
2984                 BUMP_MIB(&ipst->ips_ip_mib, ipIfStatsInCksumErrs);
2985                 ip_drop_input("ipIfStatsInCksumErrs", mp, NULL);
2986                 freemsg(mp);
2987                 return (NULL);
2988         }
2989         if (ip_type_v4(ipha->ipha_dst, ipst) == IRE_BROADCAST ||
2990             ip_type_v4(ipha->ipha_src, ipst) == IRE_BROADCAST ||
2991             CLASSD(ipha->ipha_dst) ||
2992             CLASSD(ipha->ipha_src) ||
2993             (ntohs(ipha->ipha_fragment_offset_and_flags) & IPH_OFFSET)) {
2994                 /* Note: only errors to the fragment with offset 0 */
2995                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
2996                 freemsg(mp);
2997                 return (NULL);
2998         }
2999         if (ipha->ipha_protocol == IPPROTO_ICMP) {
3000                 /*
3001                  * Check the ICMP type.  RFC 1122 sez:  don't send ICMP
3002                  * errors in response to any ICMP errors.
3003                  */
3004                 len_needed = IPH_HDR_LENGTH(ipha) + ICMPH_SIZE;
3005                 if (mp->b_wptr - mp->b_rptr < len_needed) {
3006                         if (!pullupmsg(mp, len_needed)) {
3007                                 BUMP_MIB(&ipst->ips_icmp_mib, icmpInErrors);
3008                                 freemsg(mp);
3009                                 return (NULL);
3010                         }
3011                         ipha = (ipha_t *)mp->b_rptr;
3012                 }
3013                 icmph = (icmph_t *)
3014                     (&((char *)ipha)[IPH_HDR_LENGTH(ipha)]);
3015                 switch (icmph->icmph_type) {
3016                 case ICMP_DEST_UNREACHABLE:
3017                 case ICMP_SOURCE_QUENCH:
3018                 case ICMP_TIME_EXCEEDED:
3019                 case ICMP_PARAM_PROBLEM:
3020                 case ICMP_REDIRECT:
3021                         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
3022                         freemsg(mp);
3023                         return (NULL);
3024                 default:
3025                         break;
3026                 }
3027         }
3028         /*
3029          * If this is a labeled system, then check to see if we're allowed to
3030          * send a response to this particular sender.  If not, then just drop.
3031          */
3032         if (is_system_labeled() && !tsol_can_reply_error(mp, ira)) {
3033                 ip2dbg(("icmp_pkt_err_ok: can't respond to packet\n"));
3034                 BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDrops);
3035                 freemsg(mp);
3036                 return (NULL);
3037         }
3038         if (icmp_err_rate_limit(ipst)) {
3039                 /*
3040                  * Only send ICMP error packets every so often.
3041                  * This should be done on a per port/source basis,
3042                  * but for now this will suffice.
3043                  */
3044                 freemsg(mp);
3045                 return (NULL);
3046         }
3047         return (mp);
3048 }
3049 
3050 /*
3051  * Called when a packet was sent out the same link that it arrived on.
3052  * Check if it is ok to send a redirect and then send it.
3053  */
3054 void
3055 ip_send_potential_redirect_v4(mblk_t *mp, ipha_t *ipha, ire_t *ire,
3056     ip_recv_attr_t *ira)
3057 {
3058         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
3059         ipaddr_t        src, nhop;
3060         mblk_t          *mp1;
3061         ire_t           *nhop_ire;
3062 
3063         /*
3064          * Check the source address to see if it originated
3065          * on the same logical subnet it is going back out on.
3066          * If so, we should be able to send it a redirect.
3067          * Avoid sending a redirect if the destination
3068          * is directly connected (i.e., we matched an IRE_ONLINK),
3069          * or if the packet was source routed out this interface.
3070          *
3071          * We avoid sending a redirect if the
3072          * destination is directly connected
3073          * because it is possible that multiple
3074          * IP subnets may have been configured on
3075          * the link, and the source may not
3076          * be on the same subnet as ip destination,
3077          * even though they are on the same
3078          * physical link.
3079          */
3080         if ((ire->ire_type & IRE_ONLINK) ||
3081             ip_source_routed(ipha, ipst))
3082                 return;
3083 
3084         nhop_ire = ire_nexthop(ire);
3085         if (nhop_ire == NULL)
3086                 return;
3087 
3088         nhop = nhop_ire->ire_addr;
3089 
3090         if (nhop_ire->ire_type & IRE_IF_CLONE) {
3091                 ire_t   *ire2;
3092 
3093                 /* Follow ire_dep_parent to find non-clone IRE_INTERFACE */
3094                 mutex_enter(&nhop_ire->ire_lock);
3095                 ire2 = nhop_ire->ire_dep_parent;
3096                 if (ire2 != NULL)
3097                         ire_refhold(ire2);
3098                 mutex_exit(&nhop_ire->ire_lock);
3099                 ire_refrele(nhop_ire);
3100                 nhop_ire = ire2;
3101         }
3102         if (nhop_ire == NULL)
3103                 return;
3104 
3105         ASSERT(!(nhop_ire->ire_type & IRE_IF_CLONE));
3106 
3107         src = ipha->ipha_src;
3108 
3109         /*
3110          * We look at the interface ire for the nexthop,
3111          * to see if ipha_src is in the same subnet
3112          * as the nexthop.
3113          */
3114         if ((src & nhop_ire->ire_mask) == (nhop & nhop_ire->ire_mask)) {
3115                 /*
3116                  * The source is directly connected.
3117                  */
3118                 mp1 = copymsg(mp);
3119                 if (mp1 != NULL) {
3120                         icmp_send_redirect(mp1, nhop, ira);
3121                 }
3122         }
3123         ire_refrele(nhop_ire);
3124 }
3125 
3126 /*
3127  * Generate an ICMP redirect message.
3128  */
3129 static void
3130 icmp_send_redirect(mblk_t *mp, ipaddr_t gateway, ip_recv_attr_t *ira)
3131 {
3132         icmph_t icmph;
3133         ip_stack_t *ipst = ira->ira_ill->ill_ipst;
3134 
3135         mp = icmp_pkt_err_ok(mp, ira);
3136         if (mp == NULL)
3137                 return;
3138 
3139         bzero(&icmph, sizeof (icmph_t));
3140         icmph.icmph_type = ICMP_REDIRECT;
3141         icmph.icmph_code = 1;
3142         icmph.icmph_rd_gateway = gateway;
3143         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutRedirects);
3144         icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
3145 }
3146 
3147 /*
3148  * Generate an ICMP time exceeded message.
3149  */
3150 void
3151 icmp_time_exceeded(mblk_t *mp, uint8_t code, ip_recv_attr_t *ira)
3152 {
3153         icmph_t icmph;
3154         ip_stack_t *ipst = ira->ira_ill->ill_ipst;
3155 
3156         mp = icmp_pkt_err_ok(mp, ira);
3157         if (mp == NULL)
3158                 return;
3159 
3160         bzero(&icmph, sizeof (icmph_t));
3161         icmph.icmph_type = ICMP_TIME_EXCEEDED;
3162         icmph.icmph_code = code;
3163         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutTimeExcds);
3164         icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
3165 }
3166 
3167 /*
3168  * Generate an ICMP unreachable message.
3169  * When called from ip_output side a minimal ip_recv_attr_t needs to be
3170  * constructed by the caller.
3171  */
3172 void
3173 icmp_unreachable(mblk_t *mp, uint8_t code, ip_recv_attr_t *ira)
3174 {
3175         icmph_t icmph;
3176         ip_stack_t *ipst = ira->ira_ill->ill_ipst;
3177 
3178         mp = icmp_pkt_err_ok(mp, ira);
3179         if (mp == NULL)
3180                 return;
3181 
3182         bzero(&icmph, sizeof (icmph_t));
3183         icmph.icmph_type = ICMP_DEST_UNREACHABLE;
3184         icmph.icmph_code = code;
3185         BUMP_MIB(&ipst->ips_icmp_mib, icmpOutDestUnreachs);
3186         icmp_pkt(mp, &icmph, sizeof (icmph_t), ira);
3187 }
3188 
3189 /*
3190  * Latch in the IPsec state for a stream based the policy in the listener
3191  * and the actions in the ip_recv_attr_t.
3192  * Called directly from TCP and SCTP.
3193  */
3194 boolean_t
3195 ip_ipsec_policy_inherit(conn_t *connp, conn_t *lconnp, ip_recv_attr_t *ira)
3196 {
3197         ASSERT(lconnp->conn_policy != NULL);
3198         ASSERT(connp->conn_policy == NULL);
3199 
3200         IPPH_REFHOLD(lconnp->conn_policy);
3201         connp->conn_policy = lconnp->conn_policy;
3202 
3203         if (ira->ira_ipsec_action != NULL) {
3204                 if (connp->conn_latch == NULL) {
3205                         connp->conn_latch = iplatch_create();
3206                         if (connp->conn_latch == NULL)
3207                                 return (B_FALSE);
3208                 }
3209                 ipsec_latch_inbound(connp, ira);
3210         }
3211         return (B_TRUE);
3212 }
3213 
3214 /*
3215  * Verify whether or not the IP address is a valid local address.
3216  * Could be a unicast, including one for a down interface.
3217  * If allow_mcbc then a multicast or broadcast address is also
3218  * acceptable.
3219  *
3220  * In the case of a broadcast/multicast address, however, the
3221  * upper protocol is expected to reset the src address
3222  * to zero when we return IPVL_MCAST/IPVL_BCAST so that
3223  * no packets are emitted with broadcast/multicast address as
3224  * source address (that violates hosts requirements RFC 1122)
3225  * The addresses valid for bind are:
3226  *      (1) - INADDR_ANY (0)
3227  *      (2) - IP address of an UP interface
3228  *      (3) - IP address of a DOWN interface
3229  *      (4) - valid local IP broadcast addresses. In this case
3230  *      the conn will only receive packets destined to
3231  *      the specified broadcast address.
3232  *      (5) - a multicast address. In this case
3233  *      the conn will only receive packets destined to
3234  *      the specified multicast address. Note: the
3235  *      application still has to issue an
3236  *      IP_ADD_MEMBERSHIP socket option.
3237  *
3238  * In all the above cases, the bound address must be valid in the current zone.
3239  * When the address is loopback, multicast or broadcast, there might be many
3240  * matching IREs so bind has to look up based on the zone.
3241  */
3242 ip_laddr_t
3243 ip_laddr_verify_v4(ipaddr_t src_addr, zoneid_t zoneid,
3244     ip_stack_t *ipst, boolean_t allow_mcbc)
3245 {
3246         ire_t *src_ire;
3247 
3248         ASSERT(src_addr != INADDR_ANY);
3249 
3250         src_ire = ire_ftable_lookup_v4(src_addr, 0, 0, 0,
3251             NULL, zoneid, NULL, MATCH_IRE_ZONEONLY, 0, ipst, NULL);
3252 
3253         /*
3254          * If an address other than in6addr_any is requested,
3255          * we verify that it is a valid address for bind
3256          * Note: Following code is in if-else-if form for
3257          * readability compared to a condition check.
3258          */
3259         if (src_ire != NULL && (src_ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK))) {
3260                 /*
3261                  * (2) Bind to address of local UP interface
3262                  */
3263                 ire_refrele(src_ire);
3264                 return (IPVL_UNICAST_UP);
3265         } else if (src_ire != NULL && src_ire->ire_type & IRE_BROADCAST) {
3266                 /*
3267                  * (4) Bind to broadcast address
3268                  */
3269                 ire_refrele(src_ire);
3270                 if (allow_mcbc)
3271                         return (IPVL_BCAST);
3272                 else
3273                         return (IPVL_BAD);
3274         } else if (CLASSD(src_addr)) {
3275                 /* (5) bind to multicast address. */
3276                 if (src_ire != NULL)
3277                         ire_refrele(src_ire);
3278 
3279                 if (allow_mcbc)
3280                         return (IPVL_MCAST);
3281                 else
3282                         return (IPVL_BAD);
3283         } else {
3284                 ipif_t *ipif;
3285 
3286                 /*
3287                  * (3) Bind to address of local DOWN interface?
3288                  * (ipif_lookup_addr() looks up all interfaces
3289                  * but we do not get here for UP interfaces
3290                  * - case (2) above)
3291                  */
3292                 if (src_ire != NULL)
3293                         ire_refrele(src_ire);
3294 
3295                 ipif = ipif_lookup_addr(src_addr, NULL, zoneid, ipst);
3296                 if (ipif == NULL)
3297                         return (IPVL_BAD);
3298 
3299                 /* Not a useful source? */
3300                 if (ipif->ipif_flags & (IPIF_NOLOCAL | IPIF_ANYCAST)) {
3301                         ipif_refrele(ipif);
3302                         return (IPVL_BAD);
3303                 }
3304                 ipif_refrele(ipif);
3305                 return (IPVL_UNICAST_DOWN);
3306         }
3307 }
3308 
3309 /*
3310  * Insert in the bind fanout for IPv4 and IPv6.
3311  * The caller should already have used ip_laddr_verify_v*() before calling
3312  * this.
3313  */
3314 int
3315 ip_laddr_fanout_insert(conn_t *connp)
3316 {
3317         int             error;
3318 
3319         /*
3320          * Allow setting new policies. For example, disconnects result
3321          * in us being called. As we would have set conn_policy_cached
3322          * to B_TRUE before, we should set it to B_FALSE, so that policy
3323          * can change after the disconnect.
3324          */
3325         connp->conn_policy_cached = B_FALSE;
3326 
3327         error = ipcl_bind_insert(connp);
3328         if (error != 0) {
3329                 if (connp->conn_anon_port) {
3330                         (void) tsol_mlp_anon(crgetzone(connp->conn_cred),
3331                             connp->conn_mlp_type, connp->conn_proto,
3332                             ntohs(connp->conn_lport), B_FALSE);
3333                 }
3334                 connp->conn_mlp_type = mlptSingle;
3335         }
3336         return (error);
3337 }
3338 
3339 /*
3340  * Verify that both the source and destination addresses are valid. If
3341  * IPDF_VERIFY_DST is not set, then the destination address may be unreachable,
3342  * i.e. have no route to it.  Protocols like TCP want to verify destination
3343  * reachability, while tunnels do not.
3344  *
3345  * Determine the route, the interface, and (optionally) the source address
3346  * to use to reach a given destination.
3347  * Note that we allow connect to broadcast and multicast addresses when
3348  * IPDF_ALLOW_MCBC is set.
3349  * first_hop and dst_addr are normally the same, but if source routing
3350  * they will differ; in that case the first_hop is what we'll use for the
3351  * routing lookup but the dce and label checks will be done on dst_addr,
3352  *
3353  * If uinfo is set, then we fill in the best available information
3354  * we have for the destination. This is based on (in priority order) any
3355  * metrics and path MTU stored in a dce_t, route metrics, and finally the
3356  * ill_mtu/ill_mc_mtu.
3357  *
3358  * Tsol note: If we have a source route then dst_addr != firsthop. But we
3359  * always do the label check on dst_addr.
3360  */
3361 int
3362 ip_set_destination_v4(ipaddr_t *src_addrp, ipaddr_t dst_addr, ipaddr_t firsthop,
3363     ip_xmit_attr_t *ixa, iulp_t *uinfo, uint32_t flags, uint_t mac_mode)
3364 {
3365         ire_t           *ire = NULL;
3366         int             error = 0;
3367         ipaddr_t        setsrc;                         /* RTF_SETSRC */
3368         zoneid_t        zoneid = ixa->ixa_zoneid;    /* Honors SO_ALLZONES */
3369         ip_stack_t      *ipst = ixa->ixa_ipst;
3370         dce_t           *dce;
3371         uint_t          pmtu;
3372         uint_t          generation;
3373         nce_t           *nce;
3374         ill_t           *ill = NULL;
3375         boolean_t       multirt = B_FALSE;
3376 
3377         ASSERT(ixa->ixa_flags & IXAF_IS_IPV4);
3378 
3379         /*
3380          * We never send to zero; the ULPs map it to the loopback address.
3381          * We can't allow it since we use zero to mean unitialized in some
3382          * places.
3383          */
3384         ASSERT(dst_addr != INADDR_ANY);
3385 
3386         if (is_system_labeled()) {
3387                 ts_label_t *tsl = NULL;
3388 
3389                 error = tsol_check_dest(ixa->ixa_tsl, &dst_addr, IPV4_VERSION,
3390                     mac_mode, (flags & IPDF_ZONE_IS_GLOBAL) != 0, &tsl);
3391                 if (error != 0)
3392                         return (error);
3393                 if (tsl != NULL) {
3394                         /* Update the label */
3395                         ip_xmit_attr_replace_tsl(ixa, tsl);
3396                 }
3397         }
3398 
3399         setsrc = INADDR_ANY;
3400         /*
3401          * Select a route; For IPMP interfaces, we would only select
3402          * a "hidden" route (i.e., going through a specific under_ill)
3403          * if ixa_ifindex has been specified.
3404          */
3405         ire = ip_select_route_v4(firsthop, *src_addrp, ixa,
3406             &generation, &setsrc, &error, &multirt);
3407         ASSERT(ire != NULL);    /* IRE_NOROUTE if none found */
3408         if (error != 0)
3409                 goto bad_addr;
3410 
3411         /*
3412          * ire can't be a broadcast or multicast unless IPDF_ALLOW_MCBC is set.
3413          * If IPDF_VERIFY_DST is set, the destination must be reachable;
3414          * Otherwise the destination needn't be reachable.
3415          *
3416          * If we match on a reject or black hole, then we've got a
3417          * local failure.  May as well fail out the connect() attempt,
3418          * since it's never going to succeed.
3419          */
3420         if (ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
3421                 /*
3422                  * If we're verifying destination reachability, we always want
3423                  * to complain here.
3424                  *
3425                  * If we're not verifying destination reachability but the
3426                  * destination has a route, we still want to fail on the
3427                  * temporary address and broadcast address tests.
3428                  *
3429                  * In both cases do we let the code continue so some reasonable
3430                  * information is returned to the caller. That enables the
3431                  * caller to use (and even cache) the IRE. conn_ip_ouput will
3432                  * use the generation mismatch path to check for the unreachable
3433                  * case thereby avoiding any specific check in the main path.
3434                  */
3435                 ASSERT(generation == IRE_GENERATION_VERIFY);
3436                 if (flags & IPDF_VERIFY_DST) {
3437                         /*
3438                          * Set errno but continue to set up ixa_ire to be
3439                          * the RTF_REJECT|RTF_BLACKHOLE IRE.
3440                          * That allows callers to use ip_output to get an
3441                          * ICMP error back.
3442                          */
3443                         if (!(ire->ire_type & IRE_HOST))
3444                                 error = ENETUNREACH;
3445                         else
3446                                 error = EHOSTUNREACH;
3447                 }
3448         }
3449 
3450         if ((ire->ire_type & (IRE_BROADCAST|IRE_MULTICAST)) &&
3451             !(flags & IPDF_ALLOW_MCBC)) {
3452                 ire_refrele(ire);
3453                 ire = ire_reject(ipst, B_FALSE);
3454                 generation = IRE_GENERATION_VERIFY;
3455                 error = ENETUNREACH;
3456         }
3457 
3458         /* Cache things */
3459         if (ixa->ixa_ire != NULL)
3460                 ire_refrele_notr(ixa->ixa_ire);
3461 #ifdef DEBUG
3462         ire_refhold_notr(ire);
3463         ire_refrele(ire);
3464 #endif
3465         ixa->ixa_ire = ire;
3466         ixa->ixa_ire_generation = generation;
3467 
3468         /*
3469          * Ensure that ixa_dce is always set any time that ixa_ire is set,
3470          * since some callers will send a packet to conn_ip_output() even if
3471          * there's an error.
3472          */
3473         if (flags & IPDF_UNIQUE_DCE) {
3474                 /* Fallback to the default dce if allocation fails */
3475                 dce = dce_lookup_and_add_v4(dst_addr, ipst);
3476                 if (dce != NULL)
3477                         generation = dce->dce_generation;
3478                 else
3479                         dce = dce_lookup_v4(dst_addr, ipst, &generation);
3480         } else {
3481                 dce = dce_lookup_v4(dst_addr, ipst, &generation);
3482         }
3483         ASSERT(dce != NULL);
3484         if (ixa->ixa_dce != NULL)
3485                 dce_refrele_notr(ixa->ixa_dce);
3486 #ifdef DEBUG
3487         dce_refhold_notr(dce);
3488         dce_refrele(dce);
3489 #endif
3490         ixa->ixa_dce = dce;
3491         ixa->ixa_dce_generation = generation;
3492 
3493         /*
3494          * For multicast with multirt we have a flag passed back from
3495          * ire_lookup_multi_ill_v4 since we don't have an IRE for each
3496          * possible multicast address.
3497          * We also need a flag for multicast since we can't check
3498          * whether RTF_MULTIRT is set in ixa_ire for multicast.
3499          */
3500         if (multirt) {
3501                 ixa->ixa_postfragfn = ip_postfrag_multirt_v4;
3502                 ixa->ixa_flags |= IXAF_MULTIRT_MULTICAST;
3503         } else {
3504                 ixa->ixa_postfragfn = ire->ire_postfragfn;
3505                 ixa->ixa_flags &= ~IXAF_MULTIRT_MULTICAST;
3506         }
3507         if (!(ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE))) {
3508                 /* Get an nce to cache. */
3509                 nce = ire_to_nce(ire, firsthop, NULL);
3510                 if (nce == NULL) {
3511                         /* Allocation failure? */
3512                         ixa->ixa_ire_generation = IRE_GENERATION_VERIFY;
3513                 } else {
3514                         if (ixa->ixa_nce != NULL)
3515                                 nce_refrele(ixa->ixa_nce);
3516                         ixa->ixa_nce = nce;
3517                 }
3518         }
3519 
3520         /*
3521          * If the source address is a loopback address, the
3522          * destination had best be local or multicast.
3523          * If we are sending to an IRE_LOCAL using a loopback source then
3524          * it had better be the same zoneid.
3525          */
3526         if (*src_addrp == htonl(INADDR_LOOPBACK)) {
3527                 if ((ire->ire_type & IRE_LOCAL) && ire->ire_zoneid != zoneid) {
3528                         ire = NULL;     /* Stored in ixa_ire */
3529                         error = EADDRNOTAVAIL;
3530                         goto bad_addr;
3531                 }
3532                 if (!(ire->ire_type & (IRE_LOOPBACK|IRE_LOCAL|IRE_MULTICAST))) {
3533                         ire = NULL;     /* Stored in ixa_ire */
3534                         error = EADDRNOTAVAIL;
3535                         goto bad_addr;
3536                 }
3537         }
3538         if (ire->ire_type & IRE_BROADCAST) {
3539                 /*
3540                  * If the ULP didn't have a specified source, then we
3541                  * make sure we reselect the source when sending
3542                  * broadcasts out different interfaces.
3543                  */
3544                 if (flags & IPDF_SELECT_SRC)
3545                         ixa->ixa_flags |= IXAF_SET_SOURCE;
3546                 else
3547                         ixa->ixa_flags &= ~IXAF_SET_SOURCE;
3548         }
3549 
3550         /*
3551          * Does the caller want us to pick a source address?
3552          */
3553         if (flags & IPDF_SELECT_SRC) {
3554                 ipaddr_t        src_addr;
3555 
3556                 /*
3557                  * We use use ire_nexthop_ill to avoid the under ipmp
3558                  * interface for source address selection. Note that for ipmp
3559                  * probe packets, ixa_ifindex would have been specified, and
3560                  * the ip_select_route() invocation would have picked an ire
3561                  * will ire_ill pointing at an under interface.
3562                  */
3563                 ill = ire_nexthop_ill(ire);
3564 
3565                 /* If unreachable we have no ill but need some source */
3566                 if (ill == NULL) {
3567                         src_addr = htonl(INADDR_LOOPBACK);
3568                         /* Make sure we look for a better source address */
3569                         generation = SRC_GENERATION_VERIFY;
3570                 } else {
3571                         error = ip_select_source_v4(ill, setsrc, dst_addr,
3572                             ixa->ixa_multicast_ifaddr, zoneid,
3573                             ipst, &src_addr, &generation, NULL);
3574                         if (error != 0) {
3575                                 ire = NULL;     /* Stored in ixa_ire */
3576                                 goto bad_addr;
3577                         }
3578                 }
3579 
3580                 /*
3581                  * We allow the source address to to down.
3582                  * However, we check that we don't use the loopback address
3583                  * as a source when sending out on the wire.
3584                  */
3585                 if ((src_addr == htonl(INADDR_LOOPBACK)) &&
3586                     !(ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK|IRE_MULTICAST)) &&
3587                     !(ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE))) {
3588                         ire = NULL;     /* Stored in ixa_ire */
3589                         error = EADDRNOTAVAIL;
3590                         goto bad_addr;
3591                 }
3592 
3593                 *src_addrp = src_addr;
3594                 ixa->ixa_src_generation = generation;
3595         }
3596 
3597         /*
3598          * Make sure we don't leave an unreachable ixa_nce in place
3599          * since ip_select_route is used when we unplumb i.e., remove
3600          * references on ixa_ire, ixa_nce, and ixa_dce.
3601          */
3602         nce = ixa->ixa_nce;
3603         if (nce != NULL && nce->nce_is_condemned) {
3604                 nce_refrele(nce);
3605                 ixa->ixa_nce = NULL;
3606                 ixa->ixa_ire_generation = IRE_GENERATION_VERIFY;
3607         }
3608 
3609         /*
3610          * The caller has set IXAF_PMTU_DISCOVERY if path MTU is desired.
3611          * However, we can't do it for IPv4 multicast or broadcast.
3612          */
3613         if (ire->ire_type & (IRE_BROADCAST|IRE_MULTICAST))
3614                 ixa->ixa_flags &= ~IXAF_PMTU_DISCOVERY;
3615 
3616         /*
3617          * Set initial value for fragmentation limit. Either conn_ip_output
3618          * or ULP might updates it when there are routing changes.
3619          * Handles a NULL ixa_ire->ire_ill or a NULL ixa_nce for RTF_REJECT.
3620          */
3621         pmtu = ip_get_pmtu(ixa);
3622         ixa->ixa_fragsize = pmtu;
3623         /* Make sure ixa_fragsize and ixa_pmtu remain identical */
3624         if (ixa->ixa_flags & IXAF_VERIFY_PMTU)
3625                 ixa->ixa_pmtu = pmtu;
3626 
3627         /*
3628          * Extract information useful for some transports.
3629          * First we look for DCE metrics. Then we take what we have in
3630          * the metrics in the route, where the offlink is used if we have
3631          * one.
3632          */
3633         if (uinfo != NULL) {
3634                 bzero(uinfo, sizeof (*uinfo));
3635 
3636                 if (dce->dce_flags & DCEF_UINFO)
3637                         *uinfo = dce->dce_uinfo;
3638 
3639                 rts_merge_metrics(uinfo, &ire->ire_metrics);
3640 
3641                 /* Allow ire_metrics to decrease the path MTU from above */
3642                 if (uinfo->iulp_mtu == 0 || uinfo->iulp_mtu > pmtu)
3643                         uinfo->iulp_mtu = pmtu;
3644 
3645                 uinfo->iulp_localnet = (ire->ire_type & IRE_ONLINK) != 0;
3646                 uinfo->iulp_loopback = (ire->ire_type & IRE_LOOPBACK) != 0;
3647                 uinfo->iulp_local = (ire->ire_type & IRE_LOCAL) != 0;
3648         }
3649 
3650         if (ill != NULL)
3651                 ill_refrele(ill);
3652 
3653         return (error);
3654 
3655 bad_addr:
3656         if (ire != NULL)
3657                 ire_refrele(ire);
3658 
3659         if (ill != NULL)
3660                 ill_refrele(ill);
3661 
3662         /*
3663          * Make sure we don't leave an unreachable ixa_nce in place
3664          * since ip_select_route is used when we unplumb i.e., remove
3665          * references on ixa_ire, ixa_nce, and ixa_dce.
3666          */
3667         nce = ixa->ixa_nce;
3668         if (nce != NULL && nce->nce_is_condemned) {
3669                 nce_refrele(nce);
3670                 ixa->ixa_nce = NULL;
3671                 ixa->ixa_ire_generation = IRE_GENERATION_VERIFY;
3672         }
3673 
3674         return (error);
3675 }
3676 
3677 
3678 /*
3679  * Get the base MTU for the case when path MTU discovery is not used.
3680  * Takes the MTU of the IRE into account.
3681  */
3682 uint_t
3683 ip_get_base_mtu(ill_t *ill, ire_t *ire)
3684 {
3685         uint_t mtu;
3686         uint_t iremtu = ire->ire_metrics.iulp_mtu;
3687 
3688         if (ire->ire_type & (IRE_MULTICAST|IRE_BROADCAST))
3689                 mtu = ill->ill_mc_mtu;
3690         else
3691                 mtu = ill->ill_mtu;
3692 
3693         if (iremtu != 0 && iremtu < mtu)
3694                 mtu = iremtu;
3695 
3696         return (mtu);
3697 }
3698 
3699 /*
3700  * Get the PMTU for the attributes. Handles both IPv4 and IPv6.
3701  * Assumes that ixa_ire, dce, and nce have already been set up.
3702  *
3703  * The caller has set IXAF_PMTU_DISCOVERY if path MTU discovery is desired.
3704  * We avoid path MTU discovery if it is disabled with ndd.
3705  * Furtermore, if the path MTU is too small, then we don't set DF for IPv4.
3706  *
3707  * NOTE: We also used to turn it off for source routed packets. That
3708  * is no longer required since the dce is per final destination.
3709  */
3710 uint_t
3711 ip_get_pmtu(ip_xmit_attr_t *ixa)
3712 {
3713         ip_stack_t      *ipst = ixa->ixa_ipst;
3714         dce_t           *dce;
3715         nce_t           *nce;
3716         ire_t           *ire;
3717         uint_t          pmtu;
3718 
3719         ire = ixa->ixa_ire;
3720         dce = ixa->ixa_dce;
3721         nce = ixa->ixa_nce;
3722 
3723         /*
3724          * If path MTU discovery has been turned off by ndd, then we ignore
3725          * any dce_pmtu and for IPv4 we will not set DF.
3726          */
3727         if (!ipst->ips_ip_path_mtu_discovery)
3728                 ixa->ixa_flags &= ~IXAF_PMTU_DISCOVERY;
3729 
3730         pmtu = IP_MAXPACKET;
3731         /*
3732          * Decide whether whether IPv4 sets DF
3733          * For IPv6 "no DF" means to use the 1280 mtu
3734          */
3735         if (ixa->ixa_flags & IXAF_PMTU_DISCOVERY) {
3736                 ixa->ixa_flags |= IXAF_PMTU_IPV4_DF;
3737         } else {
3738                 ixa->ixa_flags &= ~IXAF_PMTU_IPV4_DF;
3739                 if (!(ixa->ixa_flags & IXAF_IS_IPV4))
3740                         pmtu = IPV6_MIN_MTU;
3741         }
3742 
3743         /* Check if the PMTU is to old before we use it */
3744         if ((dce->dce_flags & DCEF_PMTU) &&
3745             TICK_TO_SEC(ddi_get_lbolt64()) - dce->dce_last_change_time >
3746             ipst->ips_ip_pathmtu_interval) {
3747                 /*
3748                  * Older than 20 minutes. Drop the path MTU information.
3749                  */
3750                 mutex_enter(&dce->dce_lock);
3751                 dce->dce_flags &= ~(DCEF_PMTU|DCEF_TOO_SMALL_PMTU);
3752                 dce->dce_last_change_time = TICK_TO_SEC(ddi_get_lbolt64());
3753                 mutex_exit(&dce->dce_lock);
3754                 dce_increment_generation(dce);
3755         }
3756 
3757         /* The metrics on the route can lower the path MTU */
3758         if (ire->ire_metrics.iulp_mtu != 0 &&
3759             ire->ire_metrics.iulp_mtu < pmtu)
3760                 pmtu = ire->ire_metrics.iulp_mtu;
3761 
3762         /*
3763          * If the path MTU is smaller than some minimum, we still use dce_pmtu
3764          * above (would be 576 for IPv4 and 1280 for IPv6), but we clear
3765          * IXAF_PMTU_IPV4_DF so that we avoid setting DF for IPv4.
3766          */
3767         if (ixa->ixa_flags & IXAF_PMTU_DISCOVERY) {
3768                 if (dce->dce_flags & DCEF_PMTU) {
3769                         if (dce->dce_pmtu < pmtu)
3770                                 pmtu = dce->dce_pmtu;
3771 
3772                         if (dce->dce_flags & DCEF_TOO_SMALL_PMTU) {
3773                                 ixa->ixa_flags |= IXAF_PMTU_TOO_SMALL;
3774                                 ixa->ixa_flags &= ~IXAF_PMTU_IPV4_DF;
3775                         } else {
3776                                 ixa->ixa_flags &= ~IXAF_PMTU_TOO_SMALL;
3777                                 ixa->ixa_flags |= IXAF_PMTU_IPV4_DF;
3778                         }
3779                 } else {
3780                         ixa->ixa_flags &= ~IXAF_PMTU_TOO_SMALL;
3781                         ixa->ixa_flags |= IXAF_PMTU_IPV4_DF;
3782                 }
3783         }
3784 
3785         /*
3786          * If we have an IRE_LOCAL we use the loopback mtu instead of
3787          * the ill for going out the wire i.e., IRE_LOCAL gets the same
3788          * mtu as IRE_LOOPBACK.
3789          */
3790         if (ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK)) {
3791                 uint_t loopback_mtu;
3792 
3793                 loopback_mtu = (ire->ire_ipversion == IPV6_VERSION) ?
3794                     ip_loopback_mtu_v6plus : ip_loopback_mtuplus;
3795 
3796                 if (loopback_mtu < pmtu)
3797                         pmtu = loopback_mtu;
3798         } else if (nce != NULL) {
3799                 /*
3800                  * Make sure we don't exceed the interface MTU.
3801                  * In the case of RTF_REJECT or RTF_BLACKHOLE we might not have
3802                  * an ill. We'd use the above IP_MAXPACKET in that case just
3803                  * to tell the transport something larger than zero.
3804                  */
3805                 if (ire->ire_type & (IRE_MULTICAST|IRE_BROADCAST)) {
3806                         if (nce->nce_common->ncec_ill->ill_mc_mtu < pmtu)
3807                                 pmtu = nce->nce_common->ncec_ill->ill_mc_mtu;
3808                         if (nce->nce_common->ncec_ill != nce->nce_ill &&
3809                             nce->nce_ill->ill_mc_mtu < pmtu) {
3810                                 /*
3811                                  * for interfaces in an IPMP group, the mtu of
3812                                  * the nce_ill (under_ill) could be different
3813                                  * from the mtu of the ncec_ill, so we take the
3814                                  * min of the two.
3815                                  */
3816                                 pmtu = nce->nce_ill->ill_mc_mtu;
3817                         }
3818                 } else {
3819                         if (nce->nce_common->ncec_ill->ill_mtu < pmtu)
3820                                 pmtu = nce->nce_common->ncec_ill->ill_mtu;
3821                         if (nce->nce_common->ncec_ill != nce->nce_ill &&
3822                             nce->nce_ill->ill_mtu < pmtu) {
3823                                 /*
3824                                  * for interfaces in an IPMP group, the mtu of
3825                                  * the nce_ill (under_ill) could be different
3826                                  * from the mtu of the ncec_ill, so we take the
3827                                  * min of the two.
3828                                  */
3829                                 pmtu = nce->nce_ill->ill_mtu;
3830                         }
3831                 }
3832         }
3833 
3834         /*
3835          * Handle the IPV6_USE_MIN_MTU socket option or ancillary data.
3836          * Only applies to IPv6.
3837          */
3838         if (!(ixa->ixa_flags & IXAF_IS_IPV4)) {
3839                 if (ixa->ixa_flags & IXAF_USE_MIN_MTU) {
3840                         switch (ixa->ixa_use_min_mtu) {
3841                         case IPV6_USE_MIN_MTU_MULTICAST:
3842                                 if (ire->ire_type & IRE_MULTICAST)
3843                                         pmtu = IPV6_MIN_MTU;
3844                                 break;
3845                         case IPV6_USE_MIN_MTU_ALWAYS:
3846                                 pmtu = IPV6_MIN_MTU;
3847                                 break;
3848                         case IPV6_USE_MIN_MTU_NEVER:
3849                                 break;
3850                         }
3851                 } else {
3852                         /* Default is IPV6_USE_MIN_MTU_MULTICAST */
3853                         if (ire->ire_type & IRE_MULTICAST)
3854                                 pmtu = IPV6_MIN_MTU;
3855                 }
3856         }
3857 
3858         /*
3859          * After receiving an ICMPv6 "packet too big" message with a
3860          * MTU < 1280, and for multirouted IPv6 packets, the IP layer
3861          * will insert a 8-byte fragment header in every packet. We compensate
3862          * for those cases by returning a smaller path MTU to the ULP.
3863          *
3864          * In the case of CGTP then ip_output will add a fragment header.
3865          * Make sure there is room for it by telling a smaller number
3866          * to the transport.
3867          *
3868          * When IXAF_IPV6_ADDR_FRAGHDR we subtract the frag hdr here
3869          * so the ULPs consistently see a iulp_pmtu and ip_get_pmtu()
3870          * which is the size of the packets it can send.
3871          */
3872         if (!(ixa->ixa_flags & IXAF_IS_IPV4)) {
3873                 if ((dce->dce_flags & DCEF_TOO_SMALL_PMTU) ||
3874                     (ire->ire_flags & RTF_MULTIRT) ||
3875                     (ixa->ixa_flags & IXAF_MULTIRT_MULTICAST)) {
3876                         pmtu -= sizeof (ip6_frag_t);
3877                         ixa->ixa_flags |= IXAF_IPV6_ADD_FRAGHDR;
3878                 }
3879         }
3880 
3881         return (pmtu);
3882 }
3883 
3884 /*
3885  * Carve "len" bytes out of an mblk chain, consuming any we empty, and duping
3886  * the final piece where we don't.  Return a pointer to the first mblk in the
3887  * result, and update the pointer to the next mblk to chew on.  If anything
3888  * goes wrong (i.e., dupb fails), we waste everything in sight and return a
3889  * NULL pointer.
3890  */
3891 mblk_t *
3892 ip_carve_mp(mblk_t **mpp, ssize_t len)
3893 {
3894         mblk_t  *mp0;
3895         mblk_t  *mp1;
3896         mblk_t  *mp2;
3897 
3898         if (!len || !mpp || !(mp0 = *mpp))
3899                 return (NULL);
3900         /* If we aren't going to consume the first mblk, we need a dup. */
3901         if (mp0->b_wptr - mp0->b_rptr > len) {
3902                 mp1 = dupb(mp0);
3903                 if (mp1) {
3904                         /* Partition the data between the two mblks. */
3905                         mp1->b_wptr = mp1->b_rptr + len;
3906                         mp0->b_rptr = mp1->b_wptr;
3907                         /*
3908                          * after adjustments if mblk not consumed is now
3909                          * unaligned, try to align it. If this fails free
3910                          * all messages and let upper layer recover.
3911                          */
3912                         if (!OK_32PTR(mp0->b_rptr)) {
3913                                 if (!pullupmsg(mp0, -1)) {
3914                                         freemsg(mp0);
3915                                         freemsg(mp1);
3916                                         *mpp = NULL;
3917                                         return (NULL);
3918                                 }
3919                         }
3920                 }
3921                 return (mp1);
3922         }
3923         /* Eat through as many mblks as we need to get len bytes. */
3924         len -= mp0->b_wptr - mp0->b_rptr;
3925         for (mp2 = mp1 = mp0; (mp2 = mp2->b_cont) != 0 && len; mp1 = mp2) {
3926                 if (mp2->b_wptr - mp2->b_rptr > len) {
3927                         /*
3928                          * We won't consume the entire last mblk.  Like
3929                          * above, dup and partition it.
3930                          */
3931                         mp1->b_cont = dupb(mp2);
3932                         mp1 = mp1->b_cont;
3933                         if (!mp1) {
3934                                 /*
3935                                  * Trouble.  Rather than go to a lot of
3936                                  * trouble to clean up, we free the messages.
3937                                  * This won't be any worse than losing it on
3938                                  * the wire.
3939                                  */
3940                                 freemsg(mp0);
3941                                 freemsg(mp2);
3942                                 *mpp = NULL;
3943                                 return (NULL);
3944                         }
3945                         mp1->b_wptr = mp1->b_rptr + len;
3946                         mp2->b_rptr = mp1->b_wptr;
3947                         /*
3948                          * after adjustments if mblk not consumed is now
3949                          * unaligned, try to align it. If this fails free
3950                          * all messages and let upper layer recover.
3951                          */
3952                         if (!OK_32PTR(mp2->b_rptr)) {
3953                                 if (!pullupmsg(mp2, -1)) {
3954                                         freemsg(mp0);
3955                                         freemsg(mp2);
3956                                         *mpp = NULL;
3957                                         return (NULL);
3958                                 }
3959                         }
3960                         *mpp = mp2;
3961                         return (mp0);
3962                 }
3963                 /* Decrement len by the amount we just got. */
3964                 len -= mp2->b_wptr - mp2->b_rptr;
3965         }
3966         /*
3967          * len should be reduced to zero now.  If not our caller has
3968          * screwed up.
3969          */
3970         if (len) {
3971                 /* Shouldn't happen! */
3972                 freemsg(mp0);
3973                 *mpp = NULL;
3974                 return (NULL);
3975         }
3976         /*
3977          * We consumed up to exactly the end of an mblk.  Detach the part
3978          * we are returning from the rest of the chain.
3979          */
3980         mp1->b_cont = NULL;
3981         *mpp = mp2;
3982         return (mp0);
3983 }
3984 
3985 /* The ill stream is being unplumbed. Called from ip_close */
3986 int
3987 ip_modclose(ill_t *ill)
3988 {
3989         boolean_t success;
3990         ipsq_t  *ipsq;
3991         ipif_t  *ipif;
3992         queue_t *q = ill->ill_rq;
3993         ip_stack_t      *ipst = ill->ill_ipst;
3994         int     i;
3995         arl_ill_common_t *ai = ill->ill_common;
3996 
3997         /*
3998          * The punlink prior to this may have initiated a capability
3999          * negotiation. But ipsq_enter will block until that finishes or
4000          * times out.
4001          */
4002         success = ipsq_enter(ill, B_FALSE, NEW_OP);
4003 
4004         /*
4005          * Open/close/push/pop is guaranteed to be single threaded
4006          * per stream by STREAMS. FS guarantees that all references
4007          * from top are gone before close is called. So there can't
4008          * be another close thread that has set CONDEMNED on this ill.
4009          * and cause ipsq_enter to return failure.
4010          */
4011         ASSERT(success);
4012         ipsq = ill->ill_phyint->phyint_ipsq;
4013 
4014         /*
4015          * Mark it condemned. No new reference will be made to this ill.
4016          * Lookup functions will return an error. Threads that try to
4017          * increment the refcnt must check for ILL_CAN_LOOKUP. This ensures
4018          * that the refcnt will drop down to zero.
4019          */
4020         mutex_enter(&ill->ill_lock);
4021         ill->ill_state_flags |= ILL_CONDEMNED;
4022         for (ipif = ill->ill_ipif; ipif != NULL;
4023             ipif = ipif->ipif_next) {
4024                 ipif->ipif_state_flags |= IPIF_CONDEMNED;
4025         }
4026         /*
4027          * Wake up anybody waiting to enter the ipsq. ipsq_enter
4028          * returns  error if ILL_CONDEMNED is set
4029          */
4030         cv_broadcast(&ill->ill_cv);
4031         mutex_exit(&ill->ill_lock);
4032 
4033         /*
4034          * Send all the deferred DLPI messages downstream which came in
4035          * during the small window right before ipsq_enter(). We do this
4036          * without waiting for the ACKs because all the ACKs for M_PROTO
4037          * messages are ignored in ip_rput() when ILL_CONDEMNED is set.
4038          */
4039         ill_dlpi_send_deferred(ill);
4040 
4041         /*
4042          * Shut down fragmentation reassembly.
4043          * ill_frag_timer won't start a timer again.
4044          * Now cancel any existing timer
4045          */
4046         (void) untimeout(ill->ill_frag_timer_id);
4047         (void) ill_frag_timeout(ill, 0);
4048 
4049         /*
4050          * Call ill_delete to bring down the ipifs, ilms and ill on
4051          * this ill. Then wait for the refcnts to drop to zero.
4052          * ill_is_freeable checks whether the ill is really quiescent.
4053          * Then make sure that threads that are waiting to enter the
4054          * ipsq have seen the error returned by ipsq_enter and have
4055          * gone away. Then we call ill_delete_tail which does the
4056          * DL_UNBIND_REQ with the driver and then qprocsoff.
4057          */
4058         ill_delete(ill);
4059         mutex_enter(&ill->ill_lock);
4060         while (!ill_is_freeable(ill))
4061                 cv_wait(&ill->ill_cv, &ill->ill_lock);
4062 
4063         while (ill->ill_waiters)
4064                 cv_wait(&ill->ill_cv, &ill->ill_lock);
4065 
4066         mutex_exit(&ill->ill_lock);
4067 
4068         /*
4069          * ill_delete_tail drops reference on ill_ipst, but we need to keep
4070          * it held until the end of the function since the cleanup
4071          * below needs to be able to use the ip_stack_t.
4072          */
4073         netstack_hold(ipst->ips_netstack);
4074 
4075         /* qprocsoff is done via ill_delete_tail */
4076         ill_delete_tail(ill);
4077         /*
4078          * synchronously wait for arp stream to unbind. After this, we
4079          * cannot get any data packets up from the driver.
4080          */
4081         arp_unbind_complete(ill);
4082         ASSERT(ill->ill_ipst == NULL);
4083 
4084         /*
4085          * Walk through all conns and qenable those that have queued data.
4086          * Close synchronization needs this to
4087          * be done to ensure that all upper layers blocked
4088          * due to flow control to the closing device
4089          * get unblocked.
4090          */
4091         ip1dbg(("ip_wsrv: walking\n"));
4092         for (i = 0; i < TX_FANOUT_SIZE; i++) {
4093                 conn_walk_drain(ipst, &ipst->ips_idl_tx_list[i]);
4094         }
4095 
4096         /*
4097          * ai can be null if this is an IPv6 ill, or if the IPv4
4098          * stream is being torn down before ARP was plumbed (e.g.,
4099          * /sbin/ifconfig plumbing a stream twice, and encountering
4100          * an error
4101          */
4102         if (ai != NULL) {
4103                 ASSERT(!ill->ill_isv6);
4104                 mutex_enter(&ai->ai_lock);
4105                 ai->ai_ill = NULL;
4106                 if (ai->ai_arl == NULL) {
4107                         mutex_destroy(&ai->ai_lock);
4108                         kmem_free(ai, sizeof (*ai));
4109                 } else {
4110                         cv_signal(&ai->ai_ill_unplumb_done);
4111                         mutex_exit(&ai->ai_lock);
4112                 }
4113         }
4114 
4115         mutex_enter(&ipst->ips_ip_mi_lock);
4116         mi_close_unlink(&ipst->ips_ip_g_head, (IDP)ill);
4117         mutex_exit(&ipst->ips_ip_mi_lock);
4118 
4119         /*
4120          * credp could be null if the open didn't succeed and ip_modopen
4121          * itself calls ip_close.
4122          */
4123         if (ill->ill_credp != NULL)
4124                 crfree(ill->ill_credp);
4125 
4126         mutex_destroy(&ill->ill_saved_ire_lock);
4127         mutex_destroy(&ill->ill_lock);
4128         rw_destroy(&ill->ill_mcast_lock);
4129         mutex_destroy(&ill->ill_mcast_serializer);
4130         list_destroy(&ill->ill_nce);
4131 
4132         /*
4133          * Now we are done with the module close pieces that
4134          * need the netstack_t.
4135          */
4136         netstack_rele(ipst->ips_netstack);
4137 
4138         mi_close_free((IDP)ill);
4139         q->q_ptr = WR(q)->q_ptr = NULL;
4140 
4141         ipsq_exit(ipsq);
4142 
4143         return (0);
4144 }
4145 
4146 /*
4147  * This is called as part of close() for IP, UDP, ICMP, and RTS
4148  * in order to quiesce the conn.
4149  */
4150 void
4151 ip_quiesce_conn(conn_t *connp)
4152 {
4153         boolean_t       drain_cleanup_reqd = B_FALSE;
4154         boolean_t       conn_ioctl_cleanup_reqd = B_FALSE;
4155         boolean_t       ilg_cleanup_reqd = B_FALSE;
4156         ip_stack_t      *ipst;
4157 
4158         ASSERT(!IPCL_IS_TCP(connp));
4159         ipst = connp->conn_netstack->netstack_ip;
4160 
4161         /*
4162          * Mark the conn as closing, and this conn must not be
4163          * inserted in future into any list. Eg. conn_drain_insert(),
4164          * won't insert this conn into the conn_drain_list.
4165          *
4166          * conn_idl, and conn_ilg cannot get set henceforth.
4167          */
4168         mutex_enter(&connp->conn_lock);
4169         ASSERT(!(connp->conn_state_flags & CONN_QUIESCED));
4170         connp->conn_state_flags |= CONN_CLOSING;
4171         if (connp->conn_idl != NULL)
4172                 drain_cleanup_reqd = B_TRUE;
4173         if (connp->conn_oper_pending_ill != NULL)
4174                 conn_ioctl_cleanup_reqd = B_TRUE;
4175         if (connp->conn_dhcpinit_ill != NULL) {
4176                 ASSERT(connp->conn_dhcpinit_ill->ill_dhcpinit != 0);
4177                 atomic_dec_32(&connp->conn_dhcpinit_ill->ill_dhcpinit);
4178                 ill_set_inputfn(connp->conn_dhcpinit_ill);
4179                 connp->conn_dhcpinit_ill = NULL;
4180         }
4181         if (connp->conn_ilg != NULL)
4182                 ilg_cleanup_reqd = B_TRUE;
4183         mutex_exit(&connp->conn_lock);
4184 
4185         if (conn_ioctl_cleanup_reqd)
4186                 conn_ioctl_cleanup(connp);
4187 
4188         if (is_system_labeled() && connp->conn_anon_port) {
4189                 (void) tsol_mlp_anon(crgetzone(connp->conn_cred),
4190                     connp->conn_mlp_type, connp->conn_proto,
4191                     ntohs(connp->conn_lport), B_FALSE);
4192                 connp->conn_anon_port = 0;
4193         }
4194         connp->conn_mlp_type = mlptSingle;
4195 
4196         /*
4197          * Remove this conn from any fanout list it is on.
4198          * and then wait for any threads currently operating
4199          * on this endpoint to finish
4200          */
4201         ipcl_hash_remove(connp);
4202 
4203         /*
4204          * Remove this conn from the drain list, and do any other cleanup that
4205          * may be required.  (TCP conns are never flow controlled, and
4206          * conn_idl will be NULL.)
4207          */
4208         if (drain_cleanup_reqd && connp->conn_idl != NULL) {
4209                 idl_t *idl = connp->conn_idl;
4210 
4211                 mutex_enter(&idl->idl_lock);
4212                 conn_drain(connp, B_TRUE);
4213                 mutex_exit(&idl->idl_lock);
4214         }
4215 
4216         if (connp == ipst->ips_ip_g_mrouter)
4217                 (void) ip_mrouter_done(ipst);
4218 
4219         if (ilg_cleanup_reqd)
4220                 ilg_delete_all(connp);
4221 
4222         /*
4223          * Now conn refcnt can increase only thru CONN_INC_REF_LOCKED.
4224          * callers from write side can't be there now because close
4225          * is in progress. The only other caller is ipcl_walk
4226          * which checks for the condemned flag.
4227          */
4228         mutex_enter(&connp->conn_lock);
4229         connp->conn_state_flags |= CONN_CONDEMNED;
4230         while (connp->conn_ref != 1)
4231                 cv_wait(&connp->conn_cv, &connp->conn_lock);
4232         connp->conn_state_flags |= CONN_QUIESCED;
4233         mutex_exit(&connp->conn_lock);
4234 }
4235 
4236 /* ARGSUSED */
4237 int
4238 ip_close(queue_t *q, int flags)
4239 {
4240         conn_t          *connp;
4241 
4242         /*
4243          * Call the appropriate delete routine depending on whether this is
4244          * a module or device.
4245          */
4246         if (WR(q)->q_next != NULL) {
4247                 /* This is a module close */
4248                 return (ip_modclose((ill_t *)q->q_ptr));
4249         }
4250 
4251         connp = q->q_ptr;
4252         ip_quiesce_conn(connp);
4253 
4254         qprocsoff(q);
4255 
4256         /*
4257          * Now we are truly single threaded on this stream, and can
4258          * delete the things hanging off the connp, and finally the connp.
4259          * We removed this connp from the fanout list, it cannot be
4260          * accessed thru the fanouts, and we already waited for the
4261          * conn_ref to drop to 0. We are already in close, so
4262          * there cannot be any other thread from the top. qprocsoff
4263          * has completed, and service has completed or won't run in
4264          * future.
4265          */
4266         ASSERT(connp->conn_ref == 1);
4267 
4268         inet_minor_free(connp->conn_minor_arena, connp->conn_dev);
4269 
4270         connp->conn_ref--;
4271         ipcl_conn_destroy(connp);
4272 
4273         q->q_ptr = WR(q)->q_ptr = NULL;
4274         return (0);
4275 }
4276 
4277 /*
4278  * Wapper around putnext() so that ip_rts_request can merely use
4279  * conn_recv.
4280  */
4281 /*ARGSUSED2*/
4282 static void
4283 ip_conn_input(void *arg1, mblk_t *mp, void *arg2, ip_recv_attr_t *ira)
4284 {
4285         conn_t *connp = (conn_t *)arg1;
4286 
4287         putnext(connp->conn_rq, mp);
4288 }
4289 
4290 /* Dummy in case ICMP error delivery is attempted to a /dev/ip instance */
4291 /* ARGSUSED */
4292 static void
4293 ip_conn_input_icmp(void *arg1, mblk_t *mp, void *arg2, ip_recv_attr_t *ira)
4294 {
4295         freemsg(mp);
4296 }
4297 
4298 /*
4299  * Called when the module is about to be unloaded
4300  */
4301 void
4302 ip_ddi_destroy(void)
4303 {
4304         /* This needs to be called before destroying any transports. */
4305         mutex_enter(&cpu_lock);
4306         unregister_cpu_setup_func(ip_tp_cpu_update, NULL);
4307         mutex_exit(&cpu_lock);
4308 
4309         tnet_fini();
4310 
4311         icmp_ddi_g_destroy();
4312         rts_ddi_g_destroy();
4313         udp_ddi_g_destroy();
4314         sctp_ddi_g_destroy();
4315         tcp_ddi_g_destroy();
4316         ilb_ddi_g_destroy();
4317         dce_g_destroy();
4318         ipsec_policy_g_destroy();
4319         ipcl_g_destroy();
4320         ip_net_g_destroy();
4321         ip_ire_g_fini();
4322         inet_minor_destroy(ip_minor_arena_sa);
4323 #if defined(_LP64)
4324         inet_minor_destroy(ip_minor_arena_la);
4325 #endif
4326 
4327 #ifdef DEBUG
4328         list_destroy(&ip_thread_list);
4329         rw_destroy(&ip_thread_rwlock);
4330         tsd_destroy(&ip_thread_data);
4331 #endif
4332 
4333         netstack_unregister(NS_IP);
4334 }
4335 
4336 /*
4337  * First step in cleanup.
4338  */
4339 /* ARGSUSED */
4340 static void
4341 ip_stack_shutdown(netstackid_t stackid, void *arg)
4342 {
4343         ip_stack_t *ipst = (ip_stack_t *)arg;
4344 
4345 #ifdef NS_DEBUG
4346         printf("ip_stack_shutdown(%p, stack %d)\n", (void *)ipst, stackid);
4347 #endif
4348 
4349         /*
4350          * Perform cleanup for special interfaces (loopback and IPMP).
4351          */
4352         ip_interface_cleanup(ipst);
4353 
4354         /*
4355          * The *_hook_shutdown()s start the process of notifying any
4356          * consumers that things are going away.... nothing is destroyed.
4357          */
4358         ipv4_hook_shutdown(ipst);
4359         ipv6_hook_shutdown(ipst);
4360         arp_hook_shutdown(ipst);
4361 
4362         mutex_enter(&ipst->ips_capab_taskq_lock);
4363         ipst->ips_capab_taskq_quit = B_TRUE;
4364         cv_signal(&ipst->ips_capab_taskq_cv);
4365         mutex_exit(&ipst->ips_capab_taskq_lock);
4366 }
4367 
4368 /*
4369  * Free the IP stack instance.
4370  */
4371 static void
4372 ip_stack_fini(netstackid_t stackid, void *arg)
4373 {
4374         ip_stack_t *ipst = (ip_stack_t *)arg;
4375         int ret;
4376 
4377 #ifdef NS_DEBUG
4378         printf("ip_stack_fini(%p, stack %d)\n", (void *)ipst, stackid);
4379 #endif
4380         /*
4381          * At this point, all of the notifications that the events and
4382          * protocols are going away have been run, meaning that we can
4383          * now set about starting to clean things up.
4384          */
4385         ipobs_fini(ipst);
4386         ipv4_hook_destroy(ipst);
4387         ipv6_hook_destroy(ipst);
4388         arp_hook_destroy(ipst);
4389         ip_net_destroy(ipst);
4390 
4391         ipmp_destroy(ipst);
4392 
4393         ip_kstat_fini(stackid, ipst->ips_ip_mibkp);
4394         ipst->ips_ip_mibkp = NULL;
4395         icmp_kstat_fini(stackid, ipst->ips_icmp_mibkp);
4396         ipst->ips_icmp_mibkp = NULL;
4397         ip_kstat2_fini(stackid, ipst->ips_ip_kstat);
4398         ipst->ips_ip_kstat = NULL;
4399         bzero(&ipst->ips_ip_statistics, sizeof (ipst->ips_ip_statistics));
4400         ip6_kstat_fini(stackid, ipst->ips_ip6_kstat);
4401         ipst->ips_ip6_kstat = NULL;
4402         bzero(&ipst->ips_ip6_statistics, sizeof (ipst->ips_ip6_statistics));
4403 
4404         kmem_free(ipst->ips_propinfo_tbl,
4405             ip_propinfo_count * sizeof (mod_prop_info_t));
4406         ipst->ips_propinfo_tbl = NULL;
4407 
4408         dce_stack_destroy(ipst);
4409         ip_mrouter_stack_destroy(ipst);
4410 
4411         ret = untimeout(ipst->ips_igmp_timeout_id);
4412         if (ret == -1) {
4413                 ASSERT(ipst->ips_igmp_timeout_id == 0);
4414         } else {
4415                 ASSERT(ipst->ips_igmp_timeout_id != 0);
4416                 ipst->ips_igmp_timeout_id = 0;
4417         }
4418         ret = untimeout(ipst->ips_igmp_slowtimeout_id);
4419         if (ret == -1) {
4420                 ASSERT(ipst->ips_igmp_slowtimeout_id == 0);
4421         } else {
4422                 ASSERT(ipst->ips_igmp_slowtimeout_id != 0);
4423                 ipst->ips_igmp_slowtimeout_id = 0;
4424         }
4425         ret = untimeout(ipst->ips_mld_timeout_id);
4426         if (ret == -1) {
4427                 ASSERT(ipst->ips_mld_timeout_id == 0);
4428         } else {
4429                 ASSERT(ipst->ips_mld_timeout_id != 0);
4430                 ipst->ips_mld_timeout_id = 0;
4431         }
4432         ret = untimeout(ipst->ips_mld_slowtimeout_id);
4433         if (ret == -1) {
4434                 ASSERT(ipst->ips_mld_slowtimeout_id == 0);
4435         } else {
4436                 ASSERT(ipst->ips_mld_slowtimeout_id != 0);
4437                 ipst->ips_mld_slowtimeout_id = 0;
4438         }
4439 
4440         ip_ire_fini(ipst);
4441         ip6_asp_free(ipst);
4442         conn_drain_fini(ipst);
4443         ipcl_destroy(ipst);
4444 
4445         mutex_destroy(&ipst->ips_ndp4->ndp_g_lock);
4446         mutex_destroy(&ipst->ips_ndp6->ndp_g_lock);
4447         kmem_free(ipst->ips_ndp4, sizeof (ndp_g_t));
4448         ipst->ips_ndp4 = NULL;
4449         kmem_free(ipst->ips_ndp6, sizeof (ndp_g_t));
4450         ipst->ips_ndp6 = NULL;
4451 
4452         if (ipst->ips_loopback_ksp != NULL) {
4453                 kstat_delete_netstack(ipst->ips_loopback_ksp, stackid);
4454                 ipst->ips_loopback_ksp = NULL;
4455         }
4456 
4457         mutex_destroy(&ipst->ips_capab_taskq_lock);
4458         cv_destroy(&ipst->ips_capab_taskq_cv);
4459 
4460         rw_destroy(&ipst->ips_srcid_lock);
4461 
4462         mutex_destroy(&ipst->ips_ip_mi_lock);
4463         rw_destroy(&ipst->ips_ill_g_usesrc_lock);
4464 
4465         mutex_destroy(&ipst->ips_igmp_timer_lock);
4466         mutex_destroy(&ipst->ips_mld_timer_lock);
4467         mutex_destroy(&ipst->ips_igmp_slowtimeout_lock);
4468         mutex_destroy(&ipst->ips_mld_slowtimeout_lock);
4469         mutex_destroy(&ipst->ips_ip_addr_avail_lock);
4470         rw_destroy(&ipst->ips_ill_g_lock);
4471 
4472         kmem_free(ipst->ips_phyint_g_list, sizeof (phyint_list_t));
4473         ipst->ips_phyint_g_list = NULL;
4474         kmem_free(ipst->ips_ill_g_heads, sizeof (ill_g_head_t) * MAX_G_HEADS);
4475         ipst->ips_ill_g_heads = NULL;
4476 
4477         ldi_ident_release(ipst->ips_ldi_ident);
4478         kmem_free(ipst, sizeof (*ipst));
4479 }
4480 
4481 /*
4482  * This function is called from the TSD destructor, and is used to debug
4483  * reference count issues in IP. See block comment in <inet/ip_if.h> for
4484  * details.
4485  */
4486 static void
4487 ip_thread_exit(void *phash)
4488 {
4489         th_hash_t *thh = phash;
4490 
4491         rw_enter(&ip_thread_rwlock, RW_WRITER);
4492         list_remove(&ip_thread_list, thh);
4493         rw_exit(&ip_thread_rwlock);
4494         mod_hash_destroy_hash(thh->thh_hash);
4495         kmem_free(thh, sizeof (*thh));
4496 }
4497 
4498 /*
4499  * Called when the IP kernel module is loaded into the kernel
4500  */
4501 void
4502 ip_ddi_init(void)
4503 {
4504         ip_squeue_flag = ip_squeue_switch(ip_squeue_enter);
4505 
4506         /*
4507          * For IP and TCP the minor numbers should start from 2 since we have 4
4508          * initial devices: ip, ip6, tcp, tcp6.
4509          */
4510         /*
4511          * If this is a 64-bit kernel, then create two separate arenas -
4512          * one for TLIs in the range of INET_MIN_DEV+2 through 2^^18-1, and the
4513          * other for socket apps in the range 2^^18 through 2^^32-1.
4514          */
4515         ip_minor_arena_la = NULL;
4516         ip_minor_arena_sa = NULL;
4517 #if defined(_LP64)
4518         if ((ip_minor_arena_sa = inet_minor_create("ip_minor_arena_sa",
4519             INET_MIN_DEV + 2, MAXMIN32, KM_SLEEP)) == NULL) {
4520                 cmn_err(CE_PANIC,
4521                     "ip_ddi_init: ip_minor_arena_sa creation failed\n");
4522         }
4523         if ((ip_minor_arena_la = inet_minor_create("ip_minor_arena_la",
4524             MAXMIN32 + 1, MAXMIN64, KM_SLEEP)) == NULL) {
4525                 cmn_err(CE_PANIC,
4526                     "ip_ddi_init: ip_minor_arena_la creation failed\n");
4527         }
4528 #else
4529         if ((ip_minor_arena_sa = inet_minor_create("ip_minor_arena_sa",
4530             INET_MIN_DEV + 2, MAXMIN, KM_SLEEP)) == NULL) {
4531                 cmn_err(CE_PANIC,
4532                     "ip_ddi_init: ip_minor_arena_sa creation failed\n");
4533         }
4534 #endif
4535         ip_poll_normal_ticks = MSEC_TO_TICK_ROUNDUP(ip_poll_normal_ms);
4536 
4537         ipcl_g_init();
4538         ip_ire_g_init();
4539         ip_net_g_init();
4540 
4541 #ifdef DEBUG
4542         tsd_create(&ip_thread_data, ip_thread_exit);
4543         rw_init(&ip_thread_rwlock, NULL, RW_DEFAULT, NULL);
4544         list_create(&ip_thread_list, sizeof (th_hash_t),
4545             offsetof(th_hash_t, thh_link));
4546 #endif
4547         ipsec_policy_g_init();
4548         tcp_ddi_g_init();
4549         sctp_ddi_g_init();
4550         dce_g_init();
4551 
4552         /*
4553          * We want to be informed each time a stack is created or
4554          * destroyed in the kernel, so we can maintain the
4555          * set of udp_stack_t's.
4556          */
4557         netstack_register(NS_IP, ip_stack_init, ip_stack_shutdown,
4558             ip_stack_fini);
4559 
4560         tnet_init();
4561 
4562         udp_ddi_g_init();
4563         rts_ddi_g_init();
4564         icmp_ddi_g_init();
4565         ilb_ddi_g_init();
4566 
4567         /* This needs to be called after all transports are initialized. */
4568         mutex_enter(&cpu_lock);
4569         register_cpu_setup_func(ip_tp_cpu_update, NULL);
4570         mutex_exit(&cpu_lock);
4571 }
4572 
4573 /*
4574  * Initialize the IP stack instance.
4575  */
4576 static void *
4577 ip_stack_init(netstackid_t stackid, netstack_t *ns)
4578 {
4579         ip_stack_t      *ipst;
4580         size_t          arrsz;
4581         major_t         major;
4582 
4583 #ifdef NS_DEBUG
4584         printf("ip_stack_init(stack %d)\n", stackid);
4585 #endif
4586 
4587         ipst = (ip_stack_t *)kmem_zalloc(sizeof (*ipst), KM_SLEEP);
4588         ipst->ips_netstack = ns;
4589 
4590         ipst->ips_ill_g_heads = kmem_zalloc(sizeof (ill_g_head_t) * MAX_G_HEADS,
4591             KM_SLEEP);
4592         ipst->ips_phyint_g_list = kmem_zalloc(sizeof (phyint_list_t),
4593             KM_SLEEP);
4594         ipst->ips_ndp4 = kmem_zalloc(sizeof (ndp_g_t), KM_SLEEP);
4595         ipst->ips_ndp6 = kmem_zalloc(sizeof (ndp_g_t), KM_SLEEP);
4596         mutex_init(&ipst->ips_ndp4->ndp_g_lock, NULL, MUTEX_DEFAULT, NULL);
4597         mutex_init(&ipst->ips_ndp6->ndp_g_lock, NULL, MUTEX_DEFAULT, NULL);
4598 
4599         mutex_init(&ipst->ips_igmp_timer_lock, NULL, MUTEX_DEFAULT, NULL);
4600         ipst->ips_igmp_deferred_next = INFINITY;
4601         mutex_init(&ipst->ips_mld_timer_lock, NULL, MUTEX_DEFAULT, NULL);
4602         ipst->ips_mld_deferred_next = INFINITY;
4603         mutex_init(&ipst->ips_igmp_slowtimeout_lock, NULL, MUTEX_DEFAULT, NULL);
4604         mutex_init(&ipst->ips_mld_slowtimeout_lock, NULL, MUTEX_DEFAULT, NULL);
4605         mutex_init(&ipst->ips_ip_mi_lock, NULL, MUTEX_DEFAULT, NULL);
4606         mutex_init(&ipst->ips_ip_addr_avail_lock, NULL, MUTEX_DEFAULT, NULL);
4607         rw_init(&ipst->ips_ill_g_lock, NULL, RW_DEFAULT, NULL);
4608         rw_init(&ipst->ips_ill_g_usesrc_lock, NULL, RW_DEFAULT, NULL);
4609 
4610         ipcl_init(ipst);
4611         ip_ire_init(ipst);
4612         ip6_asp_init(ipst);
4613         ipif_init(ipst);
4614         conn_drain_init(ipst);
4615         ip_mrouter_stack_init(ipst);
4616         dce_stack_init(ipst);
4617 
4618         ipst->ips_ip_multirt_log_interval = 1000;
4619 
4620         ipst->ips_ill_index = 1;
4621 
4622         ipst->ips_saved_ip_forwarding = -1;
4623         ipst->ips_reg_vif_num = ALL_VIFS;    /* Index to Register vif */
4624 
4625         arrsz = ip_propinfo_count * sizeof (mod_prop_info_t);
4626         ipst->ips_propinfo_tbl = (mod_prop_info_t *)kmem_alloc(arrsz, KM_SLEEP);
4627         bcopy(ip_propinfo_tbl, ipst->ips_propinfo_tbl, arrsz);
4628 
4629         ipst->ips_ip_mibkp = ip_kstat_init(stackid, ipst);
4630         ipst->ips_icmp_mibkp = icmp_kstat_init(stackid);
4631         ipst->ips_ip_kstat = ip_kstat2_init(stackid, &ipst->ips_ip_statistics);
4632         ipst->ips_ip6_kstat =
4633             ip6_kstat_init(stackid, &ipst->ips_ip6_statistics);
4634 
4635         ipst->ips_ip_src_id = 1;
4636         rw_init(&ipst->ips_srcid_lock, NULL, RW_DEFAULT, NULL);
4637 
4638         ipst->ips_src_generation = SRC_GENERATION_INITIAL;
4639 
4640         ip_net_init(ipst, ns);
4641         ipv4_hook_init(ipst);
4642         ipv6_hook_init(ipst);
4643         arp_hook_init(ipst);
4644         ipmp_init(ipst);
4645         ipobs_init(ipst);
4646 
4647         /*
4648          * Create the taskq dispatcher thread and initialize related stuff.
4649          */
4650         mutex_init(&ipst->ips_capab_taskq_lock, NULL, MUTEX_DEFAULT, NULL);
4651         cv_init(&ipst->ips_capab_taskq_cv, NULL, CV_DEFAULT, NULL);
4652         ipst->ips_capab_taskq_thread = thread_create(NULL, 0,
4653             ill_taskq_dispatch, ipst, 0, &p0, TS_RUN, minclsyspri);
4654 
4655         major = mod_name_to_major(INET_NAME);
4656         (void) ldi_ident_from_major(major, &ipst->ips_ldi_ident);
4657         return (ipst);
4658 }
4659 
4660 /*
4661  * Allocate and initialize a DLPI template of the specified length.  (May be
4662  * called as writer.)
4663  */
4664 mblk_t *
4665 ip_dlpi_alloc(size_t len, t_uscalar_t prim)
4666 {
4667         mblk_t  *mp;
4668 
4669         mp = allocb(len, BPRI_MED);
4670         if (!mp)
4671                 return (NULL);
4672 
4673         /*
4674          * DLPIv2 says that DL_INFO_REQ and DL_TOKEN_REQ (the latter
4675          * of which we don't seem to use) are sent with M_PCPROTO, and
4676          * that other DLPI are M_PROTO.
4677          */
4678         if (prim == DL_INFO_REQ) {
4679                 mp->b_datap->db_type = M_PCPROTO;
4680         } else {
4681                 mp->b_datap->db_type = M_PROTO;
4682         }
4683 
4684         mp->b_wptr = mp->b_rptr + len;
4685         bzero(mp->b_rptr, len);
4686         ((dl_unitdata_req_t *)mp->b_rptr)->dl_primitive = prim;
4687         return (mp);
4688 }
4689 
4690 /*
4691  * Allocate and initialize a DLPI notification.  (May be called as writer.)
4692  */
4693 mblk_t *
4694 ip_dlnotify_alloc(uint_t notification, uint_t data)
4695 {
4696         dl_notify_ind_t *notifyp;
4697         mblk_t          *mp;
4698 
4699         if ((mp = ip_dlpi_alloc(DL_NOTIFY_IND_SIZE, DL_NOTIFY_IND)) == NULL)
4700                 return (NULL);
4701 
4702         notifyp = (dl_notify_ind_t *)mp->b_rptr;
4703         notifyp->dl_notification = notification;
4704         notifyp->dl_data = data;
4705         return (mp);
4706 }
4707 
4708 mblk_t *
4709 ip_dlnotify_alloc2(uint_t notification, uint_t data1, uint_t data2)
4710 {
4711         dl_notify_ind_t *notifyp;
4712         mblk_t          *mp;
4713 
4714         if ((mp = ip_dlpi_alloc(DL_NOTIFY_IND_SIZE, DL_NOTIFY_IND)) == NULL)
4715                 return (NULL);
4716 
4717         notifyp = (dl_notify_ind_t *)mp->b_rptr;
4718         notifyp->dl_notification = notification;
4719         notifyp->dl_data1 = data1;
4720         notifyp->dl_data2 = data2;
4721         return (mp);
4722 }
4723 
4724 /*
4725  * Debug formatting routine.  Returns a character string representation of the
4726  * addr in buf, of the form xxx.xxx.xxx.xxx.  This routine takes the address
4727  * in the form of a ipaddr_t and calls ip_dot_saddr with a pointer.
4728  *
4729  * Once the ndd table-printing interfaces are removed, this can be changed to
4730  * standard dotted-decimal form.
4731  */
4732 char *
4733 ip_dot_addr(ipaddr_t addr, char *buf)
4734 {
4735         uint8_t *ap = (uint8_t *)&addr;
4736 
4737         (void) mi_sprintf(buf, "%03d.%03d.%03d.%03d",
4738             ap[0] & 0xFF, ap[1] & 0xFF, ap[2] & 0xFF, ap[3] & 0xFF);
4739         return (buf);
4740 }
4741 
4742 /*
4743  * Write the given MAC address as a printable string in the usual colon-
4744  * separated format.
4745  */
4746 const char *
4747 mac_colon_addr(const uint8_t *addr, size_t alen, char *buf, size_t buflen)
4748 {
4749         char *bp;
4750 
4751         if (alen == 0 || buflen < 4)
4752                 return ("?");
4753         bp = buf;
4754         for (;;) {
4755                 /*
4756                  * If there are more MAC address bytes available, but we won't
4757                  * have any room to print them, then add "..." to the string
4758                  * instead.  See below for the 'magic number' explanation.
4759                  */
4760                 if ((alen == 2 && buflen < 6) || (alen > 2 && buflen < 7)) {
4761                         (void) strcpy(bp, "...");
4762                         break;
4763                 }
4764                 (void) sprintf(bp, "%02x", *addr++);
4765                 bp += 2;
4766                 if (--alen == 0)
4767                         break;
4768                 *bp++ = ':';
4769                 buflen -= 3;
4770                 /*
4771                  * At this point, based on the first 'if' statement above,
4772                  * either alen == 1 and buflen >= 3, or alen > 1 and
4773                  * buflen >= 4.  The first case leaves room for the final "xx"
4774                  * number and trailing NUL byte.  The second leaves room for at
4775                  * least "...".  Thus the apparently 'magic' numbers chosen for
4776                  * that statement.
4777                  */
4778         }
4779         return (buf);
4780 }
4781 
4782 /*
4783  * Called when it is conceptually a ULP that would sent the packet
4784  * e.g., port unreachable and protocol unreachable. Check that the packet
4785  * would have passed the IPsec global policy before sending the error.
4786  *
4787  * Send an ICMP error after patching up the packet appropriately.
4788  * Uses ip_drop_input and bumps the appropriate MIB.
4789  */
4790 void
4791 ip_fanout_send_icmp_v4(mblk_t *mp, uint_t icmp_type, uint_t icmp_code,
4792     ip_recv_attr_t *ira)
4793 {
4794         ipha_t          *ipha;
4795         boolean_t       secure;
4796         ill_t           *ill = ira->ira_ill;
4797         ip_stack_t      *ipst = ill->ill_ipst;
4798         netstack_t      *ns = ipst->ips_netstack;
4799         ipsec_stack_t   *ipss = ns->netstack_ipsec;
4800 
4801         secure = ira->ira_flags & IRAF_IPSEC_SECURE;
4802 
4803         /*
4804          * We are generating an icmp error for some inbound packet.
4805          * Called from all ip_fanout_(udp, tcp, proto) functions.
4806          * Before we generate an error, check with global policy
4807          * to see whether this is allowed to enter the system. As
4808          * there is no "conn", we are checking with global policy.
4809          */
4810         ipha = (ipha_t *)mp->b_rptr;
4811         if (secure || ipss->ipsec_inbound_v4_policy_present) {
4812                 mp = ipsec_check_global_policy(mp, NULL, ipha, NULL, ira, ns);
4813                 if (mp == NULL)
4814                         return;
4815         }
4816 
4817         /* We never send errors for protocols that we do implement */
4818         if (ira->ira_protocol == IPPROTO_ICMP ||
4819             ira->ira_protocol == IPPROTO_IGMP) {
4820                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
4821                 ip_drop_input("ip_fanout_send_icmp_v4", mp, ill);
4822                 freemsg(mp);
4823                 return;
4824         }
4825         /*
4826          * Have to correct checksum since
4827          * the packet might have been
4828          * fragmented and the reassembly code in ip_rput
4829          * does not restore the IP checksum.
4830          */
4831         ipha->ipha_hdr_checksum = 0;
4832         ipha->ipha_hdr_checksum = ip_csum_hdr(ipha);
4833 
4834         switch (icmp_type) {
4835         case ICMP_DEST_UNREACHABLE:
4836                 switch (icmp_code) {
4837                 case ICMP_PROTOCOL_UNREACHABLE:
4838                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInUnknownProtos);
4839                         ip_drop_input("ipIfStatsInUnknownProtos", mp, ill);
4840                         break;
4841                 case ICMP_PORT_UNREACHABLE:
4842                         BUMP_MIB(ill->ill_ip_mib, udpIfStatsNoPorts);
4843                         ip_drop_input("ipIfStatsNoPorts", mp, ill);
4844                         break;
4845                 }
4846 
4847                 icmp_unreachable(mp, icmp_code, ira);
4848                 break;
4849         default:
4850 #ifdef DEBUG
4851                 panic("ip_fanout_send_icmp_v4: wrong type");
4852                 /*NOTREACHED*/
4853 #else
4854                 freemsg(mp);
4855                 break;
4856 #endif
4857         }
4858 }
4859 
4860 /*
4861  * Used to send an ICMP error message when a packet is received for
4862  * a protocol that is not supported. The mblk passed as argument
4863  * is consumed by this function.
4864  */
4865 void
4866 ip_proto_not_sup(mblk_t *mp, ip_recv_attr_t *ira)
4867 {
4868         ipha_t          *ipha;
4869 
4870         ipha = (ipha_t *)mp->b_rptr;
4871         if (ira->ira_flags & IRAF_IS_IPV4) {
4872                 ASSERT(IPH_HDR_VERSION(ipha) == IP_VERSION);
4873                 ip_fanout_send_icmp_v4(mp, ICMP_DEST_UNREACHABLE,
4874                     ICMP_PROTOCOL_UNREACHABLE, ira);
4875         } else {
4876                 ASSERT(IPH_HDR_VERSION(ipha) == IPV6_VERSION);
4877                 ip_fanout_send_icmp_v6(mp, ICMP6_PARAM_PROB,
4878                     ICMP6_PARAMPROB_NEXTHEADER, ira);
4879         }
4880 }
4881 
4882 /*
4883  * Deliver a rawip packet to the given conn, possibly applying ipsec policy.
4884  * Handles IPv4 and IPv6.
4885  * We are responsible for disposing of mp, such as by freemsg() or putnext()
4886  * Caller is responsible for dropping references to the conn.
4887  */
4888 void
4889 ip_fanout_proto_conn(conn_t *connp, mblk_t *mp, ipha_t *ipha, ip6_t *ip6h,
4890     ip_recv_attr_t *ira)
4891 {
4892         ill_t           *ill = ira->ira_ill;
4893         ip_stack_t      *ipst = ill->ill_ipst;
4894         ipsec_stack_t   *ipss = ipst->ips_netstack->netstack_ipsec;
4895         boolean_t       secure;
4896         uint_t          protocol = ira->ira_protocol;
4897         iaflags_t       iraflags = ira->ira_flags;
4898         queue_t         *rq;
4899 
4900         secure = iraflags & IRAF_IPSEC_SECURE;
4901 
4902         rq = connp->conn_rq;
4903         if (IPCL_IS_NONSTR(connp) ? connp->conn_flow_cntrld : !canputnext(rq)) {
4904                 switch (protocol) {
4905                 case IPPROTO_ICMPV6:
4906                         BUMP_MIB(ill->ill_icmp6_mib, ipv6IfIcmpInOverflows);
4907                         break;
4908                 case IPPROTO_ICMP:
4909                         BUMP_MIB(&ipst->ips_icmp_mib, icmpInOverflows);
4910                         break;
4911                 default:
4912                         BUMP_MIB(ill->ill_ip_mib, rawipIfStatsInOverflows);
4913                         break;
4914                 }
4915                 freemsg(mp);
4916                 return;
4917         }
4918 
4919         ASSERT(!(IPCL_IS_IPTUN(connp)));
4920 
4921         if (((iraflags & IRAF_IS_IPV4) ?
4922             CONN_INBOUND_POLICY_PRESENT(connp, ipss) :
4923             CONN_INBOUND_POLICY_PRESENT_V6(connp, ipss)) ||
4924             secure) {
4925                 mp = ipsec_check_inbound_policy(mp, connp, ipha,
4926                     ip6h, ira);
4927                 if (mp == NULL) {
4928                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
4929                         /* Note that mp is NULL */
4930                         ip_drop_input("ipIfStatsInDiscards", mp, ill);
4931                         return;
4932                 }
4933         }
4934 
4935         if (iraflags & IRAF_ICMP_ERROR) {
4936                 (connp->conn_recvicmp)(connp, mp, NULL, ira);
4937         } else {
4938                 ill_t *rill = ira->ira_rill;
4939 
4940                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCInDelivers);
4941                 ira->ira_ill = ira->ira_rill = NULL;
4942                 /* Send it upstream */
4943                 (connp->conn_recv)(connp, mp, NULL, ira);
4944                 ira->ira_ill = ill;
4945                 ira->ira_rill = rill;
4946         }
4947 }
4948 
4949 /*
4950  * Handle protocols with which IP is less intimate.  There
4951  * can be more than one stream bound to a particular
4952  * protocol.  When this is the case, normally each one gets a copy
4953  * of any incoming packets.
4954  *
4955  * IPsec NOTE :
4956  *
4957  * Don't allow a secure packet going up a non-secure connection.
4958  * We don't allow this because
4959  *
4960  * 1) Reply might go out in clear which will be dropped at
4961  *    the sending side.
4962  * 2) If the reply goes out in clear it will give the
4963  *    adversary enough information for getting the key in
4964  *    most of the cases.
4965  *
4966  * Moreover getting a secure packet when we expect clear
4967  * implies that SA's were added without checking for
4968  * policy on both ends. This should not happen once ISAKMP
4969  * is used to negotiate SAs as SAs will be added only after
4970  * verifying the policy.
4971  *
4972  * Zones notes:
4973  * Earlier in ip_input on a system with multiple shared-IP zones we
4974  * duplicate the multicast and broadcast packets and send them up
4975  * with each explicit zoneid that exists on that ill.
4976  * This means that here we can match the zoneid with SO_ALLZONES being special.
4977  */
4978 void
4979 ip_fanout_proto_v4(mblk_t *mp, ipha_t *ipha, ip_recv_attr_t *ira)
4980 {
4981         mblk_t          *mp1;
4982         ipaddr_t        laddr;
4983         conn_t          *connp, *first_connp, *next_connp;
4984         connf_t         *connfp;
4985         ill_t           *ill = ira->ira_ill;
4986         ip_stack_t      *ipst = ill->ill_ipst;
4987 
4988         laddr = ipha->ipha_dst;
4989 
4990         connfp = &ipst->ips_ipcl_proto_fanout_v4[ira->ira_protocol];
4991         mutex_enter(&connfp->connf_lock);
4992         connp = connfp->connf_head;
4993         for (connp = connfp->connf_head; connp != NULL;
4994             connp = connp->conn_next) {
4995                 /* Note: IPCL_PROTO_MATCH includes conn_wantpacket */
4996                 if (IPCL_PROTO_MATCH(connp, ira, ipha) &&
4997                     (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
4998                     tsol_receive_local(mp, &laddr, IPV4_VERSION, ira, connp))) {
4999                         break;
5000                 }
5001         }
5002 
5003         if (connp == NULL) {
5004                 /*
5005                  * No one bound to these addresses.  Is
5006                  * there a client that wants all
5007                  * unclaimed datagrams?
5008                  */
5009                 mutex_exit(&connfp->connf_lock);
5010                 ip_fanout_send_icmp_v4(mp, ICMP_DEST_UNREACHABLE,
5011                     ICMP_PROTOCOL_UNREACHABLE, ira);
5012                 return;
5013         }
5014 
5015         ASSERT(IPCL_IS_NONSTR(connp) || connp->conn_rq != NULL);
5016 
5017         CONN_INC_REF(connp);
5018         first_connp = connp;
5019         connp = connp->conn_next;
5020 
5021         for (;;) {
5022                 while (connp != NULL) {
5023                         /* Note: IPCL_PROTO_MATCH includes conn_wantpacket */
5024                         if (IPCL_PROTO_MATCH(connp, ira, ipha) &&
5025                             (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5026                             tsol_receive_local(mp, &laddr, IPV4_VERSION,
5027                             ira, connp)))
5028                                 break;
5029                         connp = connp->conn_next;
5030                 }
5031 
5032                 if (connp == NULL) {
5033                         /* No more interested clients */
5034                         connp = first_connp;
5035                         break;
5036                 }
5037                 if (((mp1 = dupmsg(mp)) == NULL) &&
5038                     ((mp1 = copymsg(mp)) == NULL)) {
5039                         /* Memory allocation failed */
5040                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
5041                         ip_drop_input("ipIfStatsInDiscards", mp, ill);
5042                         connp = first_connp;
5043                         break;
5044                 }
5045 
5046                 CONN_INC_REF(connp);
5047                 mutex_exit(&connfp->connf_lock);
5048 
5049                 ip_fanout_proto_conn(connp, mp1, (ipha_t *)mp1->b_rptr, NULL,
5050                     ira);
5051 
5052                 mutex_enter(&connfp->connf_lock);
5053                 /* Follow the next pointer before releasing the conn. */
5054                 next_connp = connp->conn_next;
5055                 CONN_DEC_REF(connp);
5056                 connp = next_connp;
5057         }
5058 
5059         /* Last one.  Send it upstream. */
5060         mutex_exit(&connfp->connf_lock);
5061 
5062         ip_fanout_proto_conn(connp, mp, ipha, NULL, ira);
5063 
5064         CONN_DEC_REF(connp);
5065 }
5066 
5067 /*
5068  * If we have a IPsec NAT-Traversal packet, strip the zero-SPI or
5069  * pass it along to ESP if the SPI is non-zero.  Returns the mblk if the mblk
5070  * is not consumed.
5071  *
5072  * One of three things can happen, all of which affect the passed-in mblk:
5073  *
5074  * 1.) The packet is stock UDP and gets its zero-SPI stripped.  Return mblk..
5075  *
5076  * 2.) The packet is ESP-in-UDP, gets transformed into an equivalent
5077  *     ESP packet, and is passed along to ESP for consumption.  Return NULL.
5078  *
5079  * 3.) The packet is an ESP-in-UDP Keepalive.  Drop it and return NULL.
5080  */
5081 mblk_t *
5082 zero_spi_check(mblk_t *mp, ip_recv_attr_t *ira)
5083 {
5084         int shift, plen, iph_len;
5085         ipha_t *ipha;
5086         udpha_t *udpha;
5087         uint32_t *spi;
5088         uint32_t esp_ports;
5089         uint8_t *orptr;
5090         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
5091         ipsec_stack_t   *ipss = ipst->ips_netstack->netstack_ipsec;
5092 
5093         ipha = (ipha_t *)mp->b_rptr;
5094         iph_len = ira->ira_ip_hdr_length;
5095         plen = ira->ira_pktlen;
5096 
5097         if (plen - iph_len - sizeof (udpha_t) < sizeof (uint32_t)) {
5098                 /*
5099                  * Most likely a keepalive for the benefit of an intervening
5100                  * NAT.  These aren't for us, per se, so drop it.
5101                  *
5102                  * RFC 3947/8 doesn't say for sure what to do for 2-3
5103                  * byte packets (keepalives are 1-byte), but we'll drop them
5104                  * also.
5105                  */
5106                 ip_drop_packet(mp, B_TRUE, ira->ira_ill,
5107                     DROPPER(ipss, ipds_esp_nat_t_ka), &ipss->ipsec_dropper);
5108                 return (NULL);
5109         }
5110 
5111         if (MBLKL(mp) < iph_len + sizeof (udpha_t) + sizeof (*spi)) {
5112                 /* might as well pull it all up - it might be ESP. */
5113                 if (!pullupmsg(mp, -1)) {
5114                         ip_drop_packet(mp, B_TRUE, ira->ira_ill,
5115                             DROPPER(ipss, ipds_esp_nomem),
5116                             &ipss->ipsec_dropper);
5117                         return (NULL);
5118                 }
5119 
5120                 ipha = (ipha_t *)mp->b_rptr;
5121         }
5122         spi = (uint32_t *)(mp->b_rptr + iph_len + sizeof (udpha_t));
5123         if (*spi == 0) {
5124                 /* UDP packet - remove 0-spi. */
5125                 shift = sizeof (uint32_t);
5126         } else {
5127                 /* ESP-in-UDP packet - reduce to ESP. */
5128                 ipha->ipha_protocol = IPPROTO_ESP;
5129                 shift = sizeof (udpha_t);
5130         }
5131 
5132         /* Fix IP header */
5133         ira->ira_pktlen = (plen - shift);
5134         ipha->ipha_length = htons(ira->ira_pktlen);
5135         ipha->ipha_hdr_checksum = 0;
5136 
5137         orptr = mp->b_rptr;
5138         mp->b_rptr += shift;
5139 
5140         udpha = (udpha_t *)(orptr + iph_len);
5141         if (*spi == 0) {
5142                 ASSERT((uint8_t *)ipha == orptr);
5143                 udpha->uha_length = htons(plen - shift - iph_len);
5144                 iph_len += sizeof (udpha_t);    /* For the call to ovbcopy(). */
5145                 esp_ports = 0;
5146         } else {
5147                 esp_ports = *((uint32_t *)udpha);
5148                 ASSERT(esp_ports != 0);
5149         }
5150         ovbcopy(orptr, orptr + shift, iph_len);
5151         if (esp_ports != 0) /* Punt up for ESP processing. */ {
5152                 ipha = (ipha_t *)(orptr + shift);
5153 
5154                 ira->ira_flags |= IRAF_ESP_UDP_PORTS;
5155                 ira->ira_esp_udp_ports = esp_ports;
5156                 ip_fanout_v4(mp, ipha, ira);
5157                 return (NULL);
5158         }
5159         return (mp);
5160 }
5161 
5162 /*
5163  * Deliver a udp packet to the given conn, possibly applying ipsec policy.
5164  * Handles IPv4 and IPv6.
5165  * We are responsible for disposing of mp, such as by freemsg() or putnext()
5166  * Caller is responsible for dropping references to the conn.
5167  */
5168 void
5169 ip_fanout_udp_conn(conn_t *connp, mblk_t *mp, ipha_t *ipha, ip6_t *ip6h,
5170     ip_recv_attr_t *ira)
5171 {
5172         ill_t           *ill = ira->ira_ill;
5173         ip_stack_t      *ipst = ill->ill_ipst;
5174         ipsec_stack_t   *ipss = ipst->ips_netstack->netstack_ipsec;
5175         boolean_t       secure;
5176         iaflags_t       iraflags = ira->ira_flags;
5177 
5178         secure = iraflags & IRAF_IPSEC_SECURE;
5179 
5180         if (IPCL_IS_NONSTR(connp) ? connp->conn_flow_cntrld :
5181             !canputnext(connp->conn_rq)) {
5182                 BUMP_MIB(ill->ill_ip_mib, udpIfStatsInOverflows);
5183                 freemsg(mp);
5184                 return;
5185         }
5186 
5187         if (((iraflags & IRAF_IS_IPV4) ?
5188             CONN_INBOUND_POLICY_PRESENT(connp, ipss) :
5189             CONN_INBOUND_POLICY_PRESENT_V6(connp, ipss)) ||
5190             secure) {
5191                 mp = ipsec_check_inbound_policy(mp, connp, ipha,
5192                     ip6h, ira);
5193                 if (mp == NULL) {
5194                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
5195                         /* Note that mp is NULL */
5196                         ip_drop_input("ipIfStatsInDiscards", mp, ill);
5197                         return;
5198                 }
5199         }
5200 
5201         /*
5202          * Since this code is not used for UDP unicast we don't need a NAT_T
5203          * check. Only ip_fanout_v4 has that check.
5204          */
5205         if (ira->ira_flags & IRAF_ICMP_ERROR) {
5206                 (connp->conn_recvicmp)(connp, mp, NULL, ira);
5207         } else {
5208                 ill_t *rill = ira->ira_rill;
5209 
5210                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCInDelivers);
5211                 ira->ira_ill = ira->ira_rill = NULL;
5212                 /* Send it upstream */
5213                 (connp->conn_recv)(connp, mp, NULL, ira);
5214                 ira->ira_ill = ill;
5215                 ira->ira_rill = rill;
5216         }
5217 }
5218 
5219 /*
5220  * Fanout for UDP packets that are multicast or broadcast, and ICMP errors.
5221  * (Unicast fanout is handled in ip_input_v4.)
5222  *
5223  * If SO_REUSEADDR is set all multicast and broadcast packets
5224  * will be delivered to all conns bound to the same port.
5225  *
5226  * If there is at least one matching AF_INET receiver, then we will
5227  * ignore any AF_INET6 receivers.
5228  * In the special case where an AF_INET socket binds to 0.0.0.0/<port> and an
5229  * AF_INET6 socket binds to ::/<port>, only the AF_INET socket receives the IPv4
5230  * packets.
5231  *
5232  * Zones notes:
5233  * Earlier in ip_input on a system with multiple shared-IP zones we
5234  * duplicate the multicast and broadcast packets and send them up
5235  * with each explicit zoneid that exists on that ill.
5236  * This means that here we can match the zoneid with SO_ALLZONES being special.
5237  */
5238 void
5239 ip_fanout_udp_multi_v4(mblk_t *mp, ipha_t *ipha, uint16_t lport, uint16_t fport,
5240     ip_recv_attr_t *ira)
5241 {
5242         ipaddr_t        laddr;
5243         in6_addr_t      v6faddr;
5244         conn_t          *connp;
5245         connf_t         *connfp;
5246         ipaddr_t        faddr;
5247         ill_t           *ill = ira->ira_ill;
5248         ip_stack_t      *ipst = ill->ill_ipst;
5249 
5250         ASSERT(ira->ira_flags & (IRAF_MULTIBROADCAST|IRAF_ICMP_ERROR));
5251 
5252         laddr = ipha->ipha_dst;
5253         faddr = ipha->ipha_src;
5254 
5255         connfp = &ipst->ips_ipcl_udp_fanout[IPCL_UDP_HASH(lport, ipst)];
5256         mutex_enter(&connfp->connf_lock);
5257         connp = connfp->connf_head;
5258 
5259         /*
5260          * If SO_REUSEADDR has been set on the first we send the
5261          * packet to all clients that have joined the group and
5262          * match the port.
5263          */
5264         while (connp != NULL) {
5265                 if ((IPCL_UDP_MATCH(connp, lport, laddr, fport, faddr)) &&
5266                     conn_wantpacket(connp, ira, ipha) &&
5267                     (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5268                     tsol_receive_local(mp, &laddr, IPV4_VERSION, ira, connp)))
5269                         break;
5270                 connp = connp->conn_next;
5271         }
5272 
5273         if (connp == NULL)
5274                 goto notfound;
5275 
5276         CONN_INC_REF(connp);
5277 
5278         if (connp->conn_reuseaddr) {
5279                 conn_t          *first_connp = connp;
5280                 conn_t          *next_connp;
5281                 mblk_t          *mp1;
5282 
5283                 connp = connp->conn_next;
5284                 for (;;) {
5285                         while (connp != NULL) {
5286                                 if (IPCL_UDP_MATCH(connp, lport, laddr,
5287                                     fport, faddr) &&
5288                                     conn_wantpacket(connp, ira, ipha) &&
5289                                     (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5290                                     tsol_receive_local(mp, &laddr, IPV4_VERSION,
5291                                     ira, connp)))
5292                                         break;
5293                                 connp = connp->conn_next;
5294                         }
5295                         if (connp == NULL) {
5296                                 /* No more interested clients */
5297                                 connp = first_connp;
5298                                 break;
5299                         }
5300                         if (((mp1 = dupmsg(mp)) == NULL) &&
5301                             ((mp1 = copymsg(mp)) == NULL)) {
5302                                 /* Memory allocation failed */
5303                                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
5304                                 ip_drop_input("ipIfStatsInDiscards", mp, ill);
5305                                 connp = first_connp;
5306                                 break;
5307                         }
5308                         CONN_INC_REF(connp);
5309                         mutex_exit(&connfp->connf_lock);
5310 
5311                         IP_STAT(ipst, ip_udp_fanmb);
5312                         ip_fanout_udp_conn(connp, mp1, (ipha_t *)mp1->b_rptr,
5313                             NULL, ira);
5314                         mutex_enter(&connfp->connf_lock);
5315                         /* Follow the next pointer before releasing the conn */
5316                         next_connp = connp->conn_next;
5317                         CONN_DEC_REF(connp);
5318                         connp = next_connp;
5319                 }
5320         }
5321 
5322         /* Last one.  Send it upstream. */
5323         mutex_exit(&connfp->connf_lock);
5324         IP_STAT(ipst, ip_udp_fanmb);
5325         ip_fanout_udp_conn(connp, mp, ipha, NULL, ira);
5326         CONN_DEC_REF(connp);
5327         return;
5328 
5329 notfound:
5330         mutex_exit(&connfp->connf_lock);
5331         /*
5332          * IPv6 endpoints bound to multicast IPv4-mapped addresses
5333          * have already been matched above, since they live in the IPv4
5334          * fanout tables. This implies we only need to
5335          * check for IPv6 in6addr_any endpoints here.
5336          * Thus we compare using ipv6_all_zeros instead of the destination
5337          * address, except for the multicast group membership lookup which
5338          * uses the IPv4 destination.
5339          */
5340         IN6_IPADDR_TO_V4MAPPED(ipha->ipha_src, &v6faddr);
5341         connfp = &ipst->ips_ipcl_udp_fanout[IPCL_UDP_HASH(lport, ipst)];
5342         mutex_enter(&connfp->connf_lock);
5343         connp = connfp->connf_head;
5344         /*
5345          * IPv4 multicast packet being delivered to an AF_INET6
5346          * in6addr_any endpoint.
5347          * Need to check conn_wantpacket(). Note that we use conn_wantpacket()
5348          * and not conn_wantpacket_v6() since any multicast membership is
5349          * for an IPv4-mapped multicast address.
5350          */
5351         while (connp != NULL) {
5352                 if (IPCL_UDP_MATCH_V6(connp, lport, ipv6_all_zeros,
5353                     fport, v6faddr) &&
5354                     conn_wantpacket(connp, ira, ipha) &&
5355                     (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5356                     tsol_receive_local(mp, &laddr, IPV4_VERSION, ira, connp)))
5357                         break;
5358                 connp = connp->conn_next;
5359         }
5360 
5361         if (connp == NULL) {
5362                 /*
5363                  * No one bound to this port.  Is
5364                  * there a client that wants all
5365                  * unclaimed datagrams?
5366                  */
5367                 mutex_exit(&connfp->connf_lock);
5368 
5369                 if (ipst->ips_ipcl_proto_fanout_v4[IPPROTO_UDP].connf_head !=
5370                     NULL) {
5371                         ASSERT(ira->ira_protocol == IPPROTO_UDP);
5372                         ip_fanout_proto_v4(mp, ipha, ira);
5373                 } else {
5374                         /*
5375                          * We used to attempt to send an icmp error here, but
5376                          * since this is known to be a multicast packet
5377                          * and we don't send icmp errors in response to
5378                          * multicast, just drop the packet and give up sooner.
5379                          */
5380                         BUMP_MIB(ill->ill_ip_mib, udpIfStatsNoPorts);
5381                         freemsg(mp);
5382                 }
5383                 return;
5384         }
5385         ASSERT(IPCL_IS_NONSTR(connp) || connp->conn_rq != NULL);
5386 
5387         /*
5388          * If SO_REUSEADDR has been set on the first we send the
5389          * packet to all clients that have joined the group and
5390          * match the port.
5391          */
5392         if (connp->conn_reuseaddr) {
5393                 conn_t          *first_connp = connp;
5394                 conn_t          *next_connp;
5395                 mblk_t          *mp1;
5396 
5397                 CONN_INC_REF(connp);
5398                 connp = connp->conn_next;
5399                 for (;;) {
5400                         while (connp != NULL) {
5401                                 if (IPCL_UDP_MATCH_V6(connp, lport,
5402                                     ipv6_all_zeros, fport, v6faddr) &&
5403                                     conn_wantpacket(connp, ira, ipha) &&
5404                                     (!(ira->ira_flags & IRAF_SYSTEM_LABELED) ||
5405                                     tsol_receive_local(mp, &laddr, IPV4_VERSION,
5406                                     ira, connp)))
5407                                         break;
5408                                 connp = connp->conn_next;
5409                         }
5410                         if (connp == NULL) {
5411                                 /* No more interested clients */
5412                                 connp = first_connp;
5413                                 break;
5414                         }
5415                         if (((mp1 = dupmsg(mp)) == NULL) &&
5416                             ((mp1 = copymsg(mp)) == NULL)) {
5417                                 /* Memory allocation failed */
5418                                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
5419                                 ip_drop_input("ipIfStatsInDiscards", mp, ill);
5420                                 connp = first_connp;
5421                                 break;
5422                         }
5423                         CONN_INC_REF(connp);
5424                         mutex_exit(&connfp->connf_lock);
5425 
5426                         IP_STAT(ipst, ip_udp_fanmb);
5427                         ip_fanout_udp_conn(connp, mp1, (ipha_t *)mp1->b_rptr,
5428                             NULL, ira);
5429                         mutex_enter(&connfp->connf_lock);
5430                         /* Follow the next pointer before releasing the conn */
5431                         next_connp = connp->conn_next;
5432                         CONN_DEC_REF(connp);
5433                         connp = next_connp;
5434                 }
5435         }
5436 
5437         /* Last one.  Send it upstream. */
5438         mutex_exit(&connfp->connf_lock);
5439         IP_STAT(ipst, ip_udp_fanmb);
5440         ip_fanout_udp_conn(connp, mp, ipha, NULL, ira);
5441         CONN_DEC_REF(connp);
5442 }
5443 
5444 /*
5445  * Split an incoming packet's IPv4 options into the label and the other options.
5446  * If 'allocate' is set it does memory allocation for the ip_pkt_t, including
5447  * clearing out any leftover label or options.
5448  * Otherwise it just makes ipp point into the packet.
5449  *
5450  * Returns zero if ok; ENOMEM if the buffer couldn't be allocated.
5451  */
5452 int
5453 ip_find_hdr_v4(ipha_t *ipha, ip_pkt_t *ipp, boolean_t allocate)
5454 {
5455         uchar_t         *opt;
5456         uint32_t        totallen;
5457         uint32_t        optval;
5458         uint32_t        optlen;
5459 
5460         ipp->ipp_fields |= IPPF_HOPLIMIT | IPPF_TCLASS | IPPF_ADDR;
5461         ipp->ipp_hoplimit = ipha->ipha_ttl;
5462         ipp->ipp_type_of_service = ipha->ipha_type_of_service;
5463         IN6_IPADDR_TO_V4MAPPED(ipha->ipha_dst, &ipp->ipp_addr);
5464 
5465         /*
5466          * Get length (in 4 byte octets) of IP header options.
5467          */
5468         totallen = ipha->ipha_version_and_hdr_length -
5469             (uint8_t)((IP_VERSION << 4) + IP_SIMPLE_HDR_LENGTH_IN_WORDS);
5470 
5471         if (totallen == 0) {
5472                 if (!allocate)
5473                         return (0);
5474 
5475                 /* Clear out anything from a previous packet */
5476                 if (ipp->ipp_fields & IPPF_IPV4_OPTIONS) {
5477                         kmem_free(ipp->ipp_ipv4_options,
5478                             ipp->ipp_ipv4_options_len);
5479                         ipp->ipp_ipv4_options = NULL;
5480                         ipp->ipp_ipv4_options_len = 0;
5481                         ipp->ipp_fields &= ~IPPF_IPV4_OPTIONS;
5482                 }
5483                 if (ipp->ipp_fields & IPPF_LABEL_V4) {
5484                         kmem_free(ipp->ipp_label_v4, ipp->ipp_label_len_v4);
5485                         ipp->ipp_label_v4 = NULL;
5486                         ipp->ipp_label_len_v4 = 0;
5487                         ipp->ipp_fields &= ~IPPF_LABEL_V4;
5488                 }
5489                 return (0);
5490         }
5491 
5492         totallen <<= 2;
5493         opt = (uchar_t *)&ipha[1];
5494         if (!is_system_labeled()) {
5495 
5496         copyall:
5497                 if (!allocate) {
5498                         if (totallen != 0) {
5499                                 ipp->ipp_ipv4_options = opt;
5500                                 ipp->ipp_ipv4_options_len = totallen;
5501                                 ipp->ipp_fields |= IPPF_IPV4_OPTIONS;
5502                         }
5503                         return (0);
5504                 }
5505                 /* Just copy all of options */
5506                 if (ipp->ipp_fields & IPPF_IPV4_OPTIONS) {
5507                         if (totallen == ipp->ipp_ipv4_options_len) {
5508                                 bcopy(opt, ipp->ipp_ipv4_options, totallen);
5509                                 return (0);
5510                         }
5511                         kmem_free(ipp->ipp_ipv4_options,
5512                             ipp->ipp_ipv4_options_len);
5513                         ipp->ipp_ipv4_options = NULL;
5514                         ipp->ipp_ipv4_options_len = 0;
5515                         ipp->ipp_fields &= ~IPPF_IPV4_OPTIONS;
5516                 }
5517                 if (totallen == 0)
5518                         return (0);
5519 
5520                 ipp->ipp_ipv4_options = kmem_alloc(totallen, KM_NOSLEEP);
5521                 if (ipp->ipp_ipv4_options == NULL)
5522                         return (ENOMEM);
5523                 ipp->ipp_ipv4_options_len = totallen;
5524                 ipp->ipp_fields |= IPPF_IPV4_OPTIONS;
5525                 bcopy(opt, ipp->ipp_ipv4_options, totallen);
5526                 return (0);
5527         }
5528 
5529         if (allocate && (ipp->ipp_fields & IPPF_LABEL_V4)) {
5530                 kmem_free(ipp->ipp_label_v4, ipp->ipp_label_len_v4);
5531                 ipp->ipp_label_v4 = NULL;
5532                 ipp->ipp_label_len_v4 = 0;
5533                 ipp->ipp_fields &= ~IPPF_LABEL_V4;
5534         }
5535 
5536         /*
5537          * Search for CIPSO option.
5538          * We assume CIPSO is first in options if it is present.
5539          * If it isn't, then ipp_opt_ipv4_options will not include the options
5540          * prior to the CIPSO option.
5541          */
5542         while (totallen != 0) {
5543                 switch (optval = opt[IPOPT_OPTVAL]) {
5544                 case IPOPT_EOL:
5545                         return (0);
5546                 case IPOPT_NOP:
5547                         optlen = 1;
5548                         break;
5549                 default:
5550                         if (totallen <= IPOPT_OLEN)
5551                                 return (EINVAL);
5552                         optlen = opt[IPOPT_OLEN];
5553                         if (optlen < 2)
5554                                 return (EINVAL);
5555                 }
5556                 if (optlen > totallen)
5557                         return (EINVAL);
5558 
5559                 switch (optval) {
5560                 case IPOPT_COMSEC:
5561                         if (!allocate) {
5562                                 ipp->ipp_label_v4 = opt;
5563                                 ipp->ipp_label_len_v4 = optlen;
5564                                 ipp->ipp_fields |= IPPF_LABEL_V4;
5565                         } else {
5566                                 ipp->ipp_label_v4 = kmem_alloc(optlen,
5567                                     KM_NOSLEEP);
5568                                 if (ipp->ipp_label_v4 == NULL)
5569                                         return (ENOMEM);
5570                                 ipp->ipp_label_len_v4 = optlen;
5571                                 ipp->ipp_fields |= IPPF_LABEL_V4;
5572                                 bcopy(opt, ipp->ipp_label_v4, optlen);
5573                         }
5574                         totallen -= optlen;
5575                         opt += optlen;
5576 
5577                         /* Skip padding bytes until we get to a multiple of 4 */
5578                         while ((totallen & 3) != 0 && opt[0] == IPOPT_NOP) {
5579                                 totallen--;
5580                                 opt++;
5581                         }
5582                         /* Remaining as ipp_ipv4_options */
5583                         goto copyall;
5584                 }
5585                 totallen -= optlen;
5586                 opt += optlen;
5587         }
5588         /* No CIPSO found; return everything as ipp_ipv4_options */
5589         totallen = ipha->ipha_version_and_hdr_length -
5590             (uint8_t)((IP_VERSION << 4) + IP_SIMPLE_HDR_LENGTH_IN_WORDS);
5591         totallen <<= 2;
5592         opt = (uchar_t *)&ipha[1];
5593         goto copyall;
5594 }
5595 
5596 /*
5597  * Efficient versions of lookup for an IRE when we only
5598  * match the address.
5599  * For RTF_REJECT or BLACKHOLE we return IRE_NOROUTE.
5600  * Does not handle multicast addresses.
5601  */
5602 uint_t
5603 ip_type_v4(ipaddr_t addr, ip_stack_t *ipst)
5604 {
5605         ire_t *ire;
5606         uint_t result;
5607 
5608         ire = ire_ftable_lookup_simple_v4(addr, 0, ipst, NULL);
5609         ASSERT(ire != NULL);
5610         if (ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE))
5611                 result = IRE_NOROUTE;
5612         else
5613                 result = ire->ire_type;
5614         ire_refrele(ire);
5615         return (result);
5616 }
5617 
5618 /*
5619  * Efficient versions of lookup for an IRE when we only
5620  * match the address.
5621  * For RTF_REJECT or BLACKHOLE we return IRE_NOROUTE.
5622  * Does not handle multicast addresses.
5623  */
5624 uint_t
5625 ip_type_v6(const in6_addr_t *addr, ip_stack_t *ipst)
5626 {
5627         ire_t *ire;
5628         uint_t result;
5629 
5630         ire = ire_ftable_lookup_simple_v6(addr, 0, ipst, NULL);
5631         ASSERT(ire != NULL);
5632         if (ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE))
5633                 result = IRE_NOROUTE;
5634         else
5635                 result = ire->ire_type;
5636         ire_refrele(ire);
5637         return (result);
5638 }
5639 
5640 /*
5641  * Nobody should be sending
5642  * packets up this stream
5643  */
5644 static void
5645 ip_lrput(queue_t *q, mblk_t *mp)
5646 {
5647         switch (mp->b_datap->db_type) {
5648         case M_FLUSH:
5649                 /* Turn around */
5650                 if (*mp->b_rptr & FLUSHW) {
5651                         *mp->b_rptr &= ~FLUSHR;
5652                         qreply(q, mp);
5653                         return;
5654                 }
5655                 break;
5656         }
5657         freemsg(mp);
5658 }
5659 
5660 /* Nobody should be sending packets down this stream */
5661 /* ARGSUSED */
5662 void
5663 ip_lwput(queue_t *q, mblk_t *mp)
5664 {
5665         freemsg(mp);
5666 }
5667 
5668 /*
5669  * Move the first hop in any source route to ipha_dst and remove that part of
5670  * the source route.  Called by other protocols.  Errors in option formatting
5671  * are ignored - will be handled by ip_output_options. Return the final
5672  * destination (either ipha_dst or the last entry in a source route.)
5673  */
5674 ipaddr_t
5675 ip_massage_options(ipha_t *ipha, netstack_t *ns)
5676 {
5677         ipoptp_t        opts;
5678         uchar_t         *opt;
5679         uint8_t         optval;
5680         uint8_t         optlen;
5681         ipaddr_t        dst;
5682         int             i;
5683         ip_stack_t      *ipst = ns->netstack_ip;
5684 
5685         ip2dbg(("ip_massage_options\n"));
5686         dst = ipha->ipha_dst;
5687         for (optval = ipoptp_first(&opts, ipha);
5688             optval != IPOPT_EOL;
5689             optval = ipoptp_next(&opts)) {
5690                 opt = opts.ipoptp_cur;
5691                 switch (optval) {
5692                         uint8_t off;
5693                 case IPOPT_SSRR:
5694                 case IPOPT_LSRR:
5695                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
5696                                 ip1dbg(("ip_massage_options: bad src route\n"));
5697                                 break;
5698                         }
5699                         optlen = opts.ipoptp_len;
5700                         off = opt[IPOPT_OFFSET];
5701                         off--;
5702                 redo_srr:
5703                         if (optlen < IP_ADDR_LEN ||
5704                             off > optlen - IP_ADDR_LEN) {
5705                                 /* End of source route */
5706                                 ip1dbg(("ip_massage_options: end of SR\n"));
5707                                 break;
5708                         }
5709                         bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
5710                         ip1dbg(("ip_massage_options: next hop 0x%x\n",
5711                             ntohl(dst)));
5712                         /*
5713                          * Check if our address is present more than
5714                          * once as consecutive hops in source route.
5715                          * XXX verify per-interface ip_forwarding
5716                          * for source route?
5717                          */
5718                         if (ip_type_v4(dst, ipst) == IRE_LOCAL) {
5719                                 off += IP_ADDR_LEN;
5720                                 goto redo_srr;
5721                         }
5722                         if (dst == htonl(INADDR_LOOPBACK)) {
5723                                 ip1dbg(("ip_massage_options: loopback addr in "
5724                                     "source route!\n"));
5725                                 break;
5726                         }
5727                         /*
5728                          * Update ipha_dst to be the first hop and remove the
5729                          * first hop from the source route (by overwriting
5730                          * part of the option with NOP options).
5731                          */
5732                         ipha->ipha_dst = dst;
5733                         /* Put the last entry in dst */
5734                         off = ((optlen - IP_ADDR_LEN - 3) & ~(IP_ADDR_LEN-1)) +
5735                             3;
5736                         bcopy(&opt[off], &dst, IP_ADDR_LEN);
5737 
5738                         ip1dbg(("ip_massage_options: last hop 0x%x\n",
5739                             ntohl(dst)));
5740                         /* Move down and overwrite */
5741                         opt[IP_ADDR_LEN] = opt[0];
5742                         opt[IP_ADDR_LEN+1] = opt[IPOPT_OLEN] - IP_ADDR_LEN;
5743                         opt[IP_ADDR_LEN+2] = opt[IPOPT_OFFSET];
5744                         for (i = 0; i < IP_ADDR_LEN; i++)
5745                                 opt[i] = IPOPT_NOP;
5746                         break;
5747                 }
5748         }
5749         return (dst);
5750 }
5751 
5752 /*
5753  * Return the network mask
5754  * associated with the specified address.
5755  */
5756 ipaddr_t
5757 ip_net_mask(ipaddr_t addr)
5758 {
5759         uchar_t *up = (uchar_t *)&addr;
5760         ipaddr_t mask = 0;
5761         uchar_t *maskp = (uchar_t *)&mask;
5762 
5763 #if defined(__i386) || defined(__amd64)
5764 #define TOTALLY_BRAIN_DAMAGED_C_COMPILER
5765 #endif
5766 #ifdef  TOTALLY_BRAIN_DAMAGED_C_COMPILER
5767         maskp[0] = maskp[1] = maskp[2] = maskp[3] = 0;
5768 #endif
5769         if (CLASSD(addr)) {
5770                 maskp[0] = 0xF0;
5771                 return (mask);
5772         }
5773 
5774         /* We assume Class E default netmask to be 32 */
5775         if (CLASSE(addr))
5776                 return (0xffffffffU);
5777 
5778         if (addr == 0)
5779                 return (0);
5780         maskp[0] = 0xFF;
5781         if ((up[0] & 0x80) == 0)
5782                 return (mask);
5783 
5784         maskp[1] = 0xFF;
5785         if ((up[0] & 0xC0) == 0x80)
5786                 return (mask);
5787 
5788         maskp[2] = 0xFF;
5789         if ((up[0] & 0xE0) == 0xC0)
5790                 return (mask);
5791 
5792         /* Otherwise return no mask */
5793         return ((ipaddr_t)0);
5794 }
5795 
5796 /* Name/Value Table Lookup Routine */
5797 char *
5798 ip_nv_lookup(nv_t *nv, int value)
5799 {
5800         if (!nv)
5801                 return (NULL);
5802         for (; nv->nv_name; nv++) {
5803                 if (nv->nv_value == value)
5804                         return (nv->nv_name);
5805         }
5806         return ("unknown");
5807 }
5808 
5809 static int
5810 ip_wait_for_info_ack(ill_t *ill)
5811 {
5812         int err;
5813 
5814         mutex_enter(&ill->ill_lock);
5815         while (ill->ill_state_flags & ILL_LL_SUBNET_PENDING) {
5816                 /*
5817                  * Return value of 0 indicates a pending signal.
5818                  */
5819                 err = cv_wait_sig(&ill->ill_cv, &ill->ill_lock);
5820                 if (err == 0) {
5821                         mutex_exit(&ill->ill_lock);
5822                         return (EINTR);
5823                 }
5824         }
5825         mutex_exit(&ill->ill_lock);
5826         /*
5827          * ip_rput_other could have set an error  in ill_error on
5828          * receipt of M_ERROR.
5829          */
5830         return (ill->ill_error);
5831 }
5832 
5833 /*
5834  * This is a module open, i.e. this is a control stream for access
5835  * to a DLPI device.  We allocate an ill_t as the instance data in
5836  * this case.
5837  */
5838 static int
5839 ip_modopen(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
5840 {
5841         ill_t   *ill;
5842         int     err;
5843         zoneid_t zoneid;
5844         netstack_t *ns;
5845         ip_stack_t *ipst;
5846 
5847         /*
5848          * Prevent unprivileged processes from pushing IP so that
5849          * they can't send raw IP.
5850          */
5851         if (secpolicy_net_rawaccess(credp) != 0)
5852                 return (EPERM);
5853 
5854         ns = netstack_find_by_cred(credp);
5855         ASSERT(ns != NULL);
5856         ipst = ns->netstack_ip;
5857         ASSERT(ipst != NULL);
5858 
5859         /*
5860          * For exclusive stacks we set the zoneid to zero
5861          * to make IP operate as if in the global zone.
5862          */
5863         if (ipst->ips_netstack->netstack_stackid != GLOBAL_NETSTACKID)
5864                 zoneid = GLOBAL_ZONEID;
5865         else
5866                 zoneid = crgetzoneid(credp);
5867 
5868         ill = (ill_t *)mi_open_alloc_sleep(sizeof (ill_t));
5869         q->q_ptr = WR(q)->q_ptr = ill;
5870         ill->ill_ipst = ipst;
5871         ill->ill_zoneid = zoneid;
5872 
5873         /*
5874          * ill_init initializes the ill fields and then sends down
5875          * down a DL_INFO_REQ after calling qprocson.
5876          */
5877         err = ill_init(q, ill);
5878 
5879         if (err != 0) {
5880                 mi_free(ill);
5881                 netstack_rele(ipst->ips_netstack);
5882                 q->q_ptr = NULL;
5883                 WR(q)->q_ptr = NULL;
5884                 return (err);
5885         }
5886 
5887         /*
5888          * Wait for the DL_INFO_ACK if a DL_INFO_REQ was sent.
5889          *
5890          * ill_init initializes the ipsq marking this thread as
5891          * writer
5892          */
5893         ipsq_exit(ill->ill_phyint->phyint_ipsq);
5894         err = ip_wait_for_info_ack(ill);
5895         if (err == 0)
5896                 ill->ill_credp = credp;
5897         else
5898                 goto fail;
5899 
5900         crhold(credp);
5901 
5902         mutex_enter(&ipst->ips_ip_mi_lock);
5903         err = mi_open_link(&ipst->ips_ip_g_head, (IDP)q->q_ptr, devp, flag,
5904             sflag, credp);
5905         mutex_exit(&ipst->ips_ip_mi_lock);
5906 fail:
5907         if (err) {
5908                 (void) ip_close(q, 0);
5909                 return (err);
5910         }
5911         return (0);
5912 }
5913 
5914 /* For /dev/ip aka AF_INET open */
5915 int
5916 ip_openv4(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
5917 {
5918         return (ip_open(q, devp, flag, sflag, credp, B_FALSE));
5919 }
5920 
5921 /* For /dev/ip6 aka AF_INET6 open */
5922 int
5923 ip_openv6(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp)
5924 {
5925         return (ip_open(q, devp, flag, sflag, credp, B_TRUE));
5926 }
5927 
5928 /* IP open routine. */
5929 int
5930 ip_open(queue_t *q, dev_t *devp, int flag, int sflag, cred_t *credp,
5931     boolean_t isv6)
5932 {
5933         conn_t          *connp;
5934         major_t         maj;
5935         zoneid_t        zoneid;
5936         netstack_t      *ns;
5937         ip_stack_t      *ipst;
5938 
5939         /* Allow reopen. */
5940         if (q->q_ptr != NULL)
5941                 return (0);
5942 
5943         if (sflag & MODOPEN) {
5944                 /* This is a module open */
5945                 return (ip_modopen(q, devp, flag, sflag, credp));
5946         }
5947 
5948         if ((flag & ~(FKLYR)) == IP_HELPER_STR) {
5949                 /*
5950                  * Non streams based socket looking for a stream
5951                  * to access IP
5952                  */
5953                 return (ip_helper_stream_setup(q, devp, flag, sflag,
5954                     credp, isv6));
5955         }
5956 
5957         ns = netstack_find_by_cred(credp);
5958         ASSERT(ns != NULL);
5959         ipst = ns->netstack_ip;
5960         ASSERT(ipst != NULL);
5961 
5962         /*
5963          * For exclusive stacks we set the zoneid to zero
5964          * to make IP operate as if in the global zone.
5965          */
5966         if (ipst->ips_netstack->netstack_stackid != GLOBAL_NETSTACKID)
5967                 zoneid = GLOBAL_ZONEID;
5968         else
5969                 zoneid = crgetzoneid(credp);
5970 
5971         /*
5972          * We are opening as a device. This is an IP client stream, and we
5973          * allocate an conn_t as the instance data.
5974          */
5975         connp = ipcl_conn_create(IPCL_IPCCONN, KM_SLEEP, ipst->ips_netstack);
5976 
5977         /*
5978          * ipcl_conn_create did a netstack_hold. Undo the hold that was
5979          * done by netstack_find_by_cred()
5980          */
5981         netstack_rele(ipst->ips_netstack);
5982 
5983         connp->conn_ixa->ixa_flags |= IXAF_MULTICAST_LOOP | IXAF_SET_ULP_CKSUM;
5984         /* conn_allzones can not be set this early, hence no IPCL_ZONEID */
5985         connp->conn_ixa->ixa_zoneid = zoneid;
5986         connp->conn_zoneid = zoneid;
5987 
5988         connp->conn_rq = q;
5989         q->q_ptr = WR(q)->q_ptr = connp;
5990 
5991         /* Minor tells us which /dev entry was opened */
5992         if (isv6) {
5993                 connp->conn_family = AF_INET6;
5994                 connp->conn_ipversion = IPV6_VERSION;
5995                 connp->conn_ixa->ixa_flags &= ~IXAF_IS_IPV4;
5996                 connp->conn_ixa->ixa_src_preferences = IPV6_PREFER_SRC_DEFAULT;
5997         } else {
5998                 connp->conn_family = AF_INET;
5999                 connp->conn_ipversion = IPV4_VERSION;
6000                 connp->conn_ixa->ixa_flags |= IXAF_IS_IPV4;
6001         }
6002 
6003         if ((ip_minor_arena_la != NULL) && (flag & SO_SOCKSTR) &&
6004             ((connp->conn_dev = inet_minor_alloc(ip_minor_arena_la)) != 0)) {
6005                 connp->conn_minor_arena = ip_minor_arena_la;
6006         } else {
6007                 /*
6008                  * Either minor numbers in the large arena were exhausted
6009                  * or a non socket application is doing the open.
6010                  * Try to allocate from the small arena.
6011                  */
6012                 if ((connp->conn_dev =
6013                     inet_minor_alloc(ip_minor_arena_sa)) == 0) {
6014                         /* CONN_DEC_REF takes care of netstack_rele() */
6015                         q->q_ptr = WR(q)->q_ptr = NULL;
6016                         CONN_DEC_REF(connp);
6017                         return (EBUSY);
6018                 }
6019                 connp->conn_minor_arena = ip_minor_arena_sa;
6020         }
6021 
6022         maj = getemajor(*devp);
6023         *devp = makedevice(maj, (minor_t)connp->conn_dev);
6024 
6025         /*
6026          * connp->conn_cred is crfree()ed in ipcl_conn_destroy()
6027          */
6028         connp->conn_cred = credp;
6029         connp->conn_cpid = curproc->p_pid;
6030         /* Cache things in ixa without an extra refhold */
6031         ASSERT(!(connp->conn_ixa->ixa_free_flags & IXA_FREE_CRED));
6032         connp->conn_ixa->ixa_cred = connp->conn_cred;
6033         connp->conn_ixa->ixa_cpid = connp->conn_cpid;
6034         if (is_system_labeled())
6035                 connp->conn_ixa->ixa_tsl = crgetlabel(connp->conn_cred);
6036 
6037         /*
6038          * Handle IP_IOC_RTS_REQUEST and other ioctls which use conn_recv
6039          */
6040         connp->conn_recv = ip_conn_input;
6041         connp->conn_recvicmp = ip_conn_input_icmp;
6042 
6043         crhold(connp->conn_cred);
6044 
6045         /*
6046          * If the caller has the process-wide flag set, then default to MAC
6047          * exempt mode.  This allows read-down to unlabeled hosts.
6048          */
6049         if (getpflags(NET_MAC_AWARE, credp) != 0)
6050                 connp->conn_mac_mode = CONN_MAC_AWARE;
6051 
6052         connp->conn_zone_is_global = (crgetzoneid(credp) == GLOBAL_ZONEID);
6053 
6054         connp->conn_rq = q;
6055         connp->conn_wq = WR(q);
6056 
6057         /* Non-zero default values */
6058         connp->conn_ixa->ixa_flags |= IXAF_MULTICAST_LOOP;
6059 
6060         /*
6061          * Make the conn globally visible to walkers
6062          */
6063         ASSERT(connp->conn_ref == 1);
6064         mutex_enter(&connp->conn_lock);
6065         connp->conn_state_flags &= ~CONN_INCIPIENT;
6066         mutex_exit(&connp->conn_lock);
6067 
6068         qprocson(q);
6069 
6070         return (0);
6071 }
6072 
6073 /*
6074  * Set IPsec policy from an ipsec_req_t. If the req is not "zero" and valid,
6075  * all of them are copied to the conn_t. If the req is "zero", the policy is
6076  * zeroed out. A "zero" policy has zero ipsr_{ah,req,self_encap}_req
6077  * fields.
6078  * We keep only the latest setting of the policy and thus policy setting
6079  * is not incremental/cumulative.
6080  *
6081  * Requests to set policies with multiple alternative actions will
6082  * go through a different API.
6083  */
6084 int
6085 ipsec_set_req(cred_t *cr, conn_t *connp, ipsec_req_t *req)
6086 {
6087         uint_t ah_req = 0;
6088         uint_t esp_req = 0;
6089         uint_t se_req = 0;
6090         ipsec_act_t *actp = NULL;
6091         uint_t nact;
6092         ipsec_policy_head_t *ph;
6093         boolean_t is_pol_reset, is_pol_inserted = B_FALSE;
6094         int error = 0;
6095         netstack_t      *ns = connp->conn_netstack;
6096         ip_stack_t      *ipst = ns->netstack_ip;
6097         ipsec_stack_t   *ipss = ns->netstack_ipsec;
6098 
6099 #define REQ_MASK (IPSEC_PREF_REQUIRED|IPSEC_PREF_NEVER)
6100 
6101         /*
6102          * The IP_SEC_OPT option does not allow variable length parameters,
6103          * hence a request cannot be NULL.
6104          */
6105         if (req == NULL)
6106                 return (EINVAL);
6107 
6108         ah_req = req->ipsr_ah_req;
6109         esp_req = req->ipsr_esp_req;
6110         se_req = req->ipsr_self_encap_req;
6111 
6112         /* Don't allow setting self-encap without one or more of AH/ESP. */
6113         if (se_req != 0 && esp_req == 0 && ah_req == 0)
6114                 return (EINVAL);
6115 
6116         /*
6117          * Are we dealing with a request to reset the policy (i.e.
6118          * zero requests).
6119          */
6120         is_pol_reset = ((ah_req & REQ_MASK) == 0 &&
6121             (esp_req & REQ_MASK) == 0 &&
6122             (se_req & REQ_MASK) == 0);
6123 
6124         if (!is_pol_reset) {
6125                 /*
6126                  * If we couldn't load IPsec, fail with "protocol
6127                  * not supported".
6128                  * IPsec may not have been loaded for a request with zero
6129                  * policies, so we don't fail in this case.
6130                  */
6131                 mutex_enter(&ipss->ipsec_loader_lock);
6132                 if (ipss->ipsec_loader_state != IPSEC_LOADER_SUCCEEDED) {
6133                         mutex_exit(&ipss->ipsec_loader_lock);
6134                         return (EPROTONOSUPPORT);
6135                 }
6136                 mutex_exit(&ipss->ipsec_loader_lock);
6137 
6138                 /*
6139                  * Test for valid requests. Invalid algorithms
6140                  * need to be tested by IPsec code because new
6141                  * algorithms can be added dynamically.
6142                  */
6143                 if ((ah_req & ~(REQ_MASK|IPSEC_PREF_UNIQUE)) != 0 ||
6144                     (esp_req & ~(REQ_MASK|IPSEC_PREF_UNIQUE)) != 0 ||
6145                     (se_req & ~(REQ_MASK|IPSEC_PREF_UNIQUE)) != 0) {
6146                         return (EINVAL);
6147                 }
6148 
6149                 /*
6150                  * Only privileged users can issue these
6151                  * requests.
6152                  */
6153                 if (((ah_req & IPSEC_PREF_NEVER) ||
6154                     (esp_req & IPSEC_PREF_NEVER) ||
6155                     (se_req & IPSEC_PREF_NEVER)) &&
6156                     secpolicy_ip_config(cr, B_FALSE) != 0) {
6157                         return (EPERM);
6158                 }
6159 
6160                 /*
6161                  * The IPSEC_PREF_REQUIRED and IPSEC_PREF_NEVER
6162                  * are mutually exclusive.
6163                  */
6164                 if (((ah_req & REQ_MASK) == REQ_MASK) ||
6165                     ((esp_req & REQ_MASK) == REQ_MASK) ||
6166                     ((se_req & REQ_MASK) == REQ_MASK)) {
6167                         /* Both of them are set */
6168                         return (EINVAL);
6169                 }
6170         }
6171 
6172         ASSERT(MUTEX_HELD(&connp->conn_lock));
6173 
6174         /*
6175          * If we have already cached policies in conn_connect(), don't
6176          * let them change now. We cache policies for connections
6177          * whose src,dst [addr, port] is known.
6178          */
6179         if (connp->conn_policy_cached) {
6180                 return (EINVAL);
6181         }
6182 
6183         /*
6184          * We have a zero policies, reset the connection policy if already
6185          * set. This will cause the connection to inherit the
6186          * global policy, if any.
6187          */
6188         if (is_pol_reset) {
6189                 if (connp->conn_policy != NULL) {
6190                         IPPH_REFRELE(connp->conn_policy, ipst->ips_netstack);
6191                         connp->conn_policy = NULL;
6192                 }
6193                 connp->conn_in_enforce_policy = B_FALSE;
6194                 connp->conn_out_enforce_policy = B_FALSE;
6195                 return (0);
6196         }
6197 
6198         ph = connp->conn_policy = ipsec_polhead_split(connp->conn_policy,
6199             ipst->ips_netstack);
6200         if (ph == NULL)
6201                 goto enomem;
6202 
6203         ipsec_actvec_from_req(req, &actp, &nact, ipst->ips_netstack);
6204         if (actp == NULL)
6205                 goto enomem;
6206 
6207         /*
6208          * Always insert IPv4 policy entries, since they can also apply to
6209          * ipv6 sockets being used in ipv4-compat mode.
6210          */
6211         if (!ipsec_polhead_insert(ph, actp, nact, IPSEC_AF_V4,
6212             IPSEC_TYPE_INBOUND, ns))
6213                 goto enomem;
6214         is_pol_inserted = B_TRUE;
6215         if (!ipsec_polhead_insert(ph, actp, nact, IPSEC_AF_V4,
6216             IPSEC_TYPE_OUTBOUND, ns))
6217                 goto enomem;
6218 
6219         /*
6220          * We're looking at a v6 socket, also insert the v6-specific
6221          * entries.
6222          */
6223         if (connp->conn_family == AF_INET6) {
6224                 if (!ipsec_polhead_insert(ph, actp, nact, IPSEC_AF_V6,
6225                     IPSEC_TYPE_INBOUND, ns))
6226                         goto enomem;
6227                 if (!ipsec_polhead_insert(ph, actp, nact, IPSEC_AF_V6,
6228                     IPSEC_TYPE_OUTBOUND, ns))
6229                         goto enomem;
6230         }
6231 
6232         ipsec_actvec_free(actp, nact);
6233 
6234         /*
6235          * If the requests need security, set enforce_policy.
6236          * If the requests are IPSEC_PREF_NEVER, one should
6237          * still set conn_out_enforce_policy so that ip_set_destination
6238          * marks the ip_xmit_attr_t appropriatly. This is needed so that
6239          * for connections that we don't cache policy in at connect time,
6240          * if global policy matches in ip_output_attach_policy, we
6241          * don't wrongly inherit global policy. Similarly, we need
6242          * to set conn_in_enforce_policy also so that we don't verify
6243          * policy wrongly.
6244          */
6245         if ((ah_req & REQ_MASK) != 0 ||
6246             (esp_req & REQ_MASK) != 0 ||
6247             (se_req & REQ_MASK) != 0) {
6248                 connp->conn_in_enforce_policy = B_TRUE;
6249                 connp->conn_out_enforce_policy = B_TRUE;
6250         }
6251 
6252         return (error);
6253 #undef REQ_MASK
6254 
6255         /*
6256          * Common memory-allocation-failure exit path.
6257          */
6258 enomem:
6259         if (actp != NULL)
6260                 ipsec_actvec_free(actp, nact);
6261         if (is_pol_inserted)
6262                 ipsec_polhead_flush(ph, ns);
6263         return (ENOMEM);
6264 }
6265 
6266 /*
6267  * Set socket options for joining and leaving multicast groups.
6268  * Common to IPv4 and IPv6; inet6 indicates the type of socket.
6269  * The caller has already check that the option name is consistent with
6270  * the address family of the socket.
6271  */
6272 int
6273 ip_opt_set_multicast_group(conn_t *connp, t_scalar_t name,
6274     uchar_t *invalp, boolean_t inet6, boolean_t checkonly)
6275 {
6276         int             *i1 = (int *)invalp;
6277         int             error = 0;
6278         ip_stack_t      *ipst = connp->conn_netstack->netstack_ip;
6279         struct ip_mreq  *v4_mreqp;
6280         struct ipv6_mreq *v6_mreqp;
6281         struct group_req *greqp;
6282         ire_t *ire;
6283         boolean_t done = B_FALSE;
6284         ipaddr_t ifaddr;
6285         in6_addr_t v6group;
6286         uint_t ifindex;
6287         boolean_t mcast_opt = B_TRUE;
6288         mcast_record_t fmode;
6289         int (*optfn)(conn_t *, boolean_t, const in6_addr_t *,
6290             ipaddr_t, uint_t, mcast_record_t, const in6_addr_t *);
6291 
6292         switch (name) {
6293         case IP_ADD_MEMBERSHIP:
6294         case IPV6_JOIN_GROUP:
6295                 mcast_opt = B_FALSE;
6296                 /* FALLTHRU */
6297         case MCAST_JOIN_GROUP:
6298                 fmode = MODE_IS_EXCLUDE;
6299                 optfn = ip_opt_add_group;
6300                 break;
6301 
6302         case IP_DROP_MEMBERSHIP:
6303         case IPV6_LEAVE_GROUP:
6304                 mcast_opt = B_FALSE;
6305                 /* FALLTHRU */
6306         case MCAST_LEAVE_GROUP:
6307                 fmode = MODE_IS_INCLUDE;
6308                 optfn = ip_opt_delete_group;
6309                 break;
6310         default:
6311                 ASSERT(0);
6312         }
6313 
6314         if (mcast_opt) {
6315                 struct sockaddr_in *sin;
6316                 struct sockaddr_in6 *sin6;
6317 
6318                 greqp = (struct group_req *)i1;
6319                 if (greqp->gr_group.ss_family == AF_INET) {
6320                         sin = (struct sockaddr_in *)&(greqp->gr_group);
6321                         IN6_INADDR_TO_V4MAPPED(&sin->sin_addr, &v6group);
6322                 } else {
6323                         if (!inet6)
6324                                 return (EINVAL);        /* Not on INET socket */
6325 
6326                         sin6 = (struct sockaddr_in6 *)&(greqp->gr_group);
6327                         v6group = sin6->sin6_addr;
6328                 }
6329                 ifaddr = INADDR_ANY;
6330                 ifindex = greqp->gr_interface;
6331         } else if (inet6) {
6332                 v6_mreqp = (struct ipv6_mreq *)i1;
6333                 v6group = v6_mreqp->ipv6mr_multiaddr;
6334                 ifaddr = INADDR_ANY;
6335                 ifindex = v6_mreqp->ipv6mr_interface;
6336         } else {
6337                 v4_mreqp = (struct ip_mreq *)i1;
6338                 IN6_INADDR_TO_V4MAPPED(&v4_mreqp->imr_multiaddr, &v6group);
6339                 ifaddr = (ipaddr_t)v4_mreqp->imr_interface.s_addr;
6340                 ifindex = 0;
6341         }
6342 
6343         /*
6344          * In the multirouting case, we need to replicate
6345          * the request on all interfaces that will take part
6346          * in replication.  We do so because multirouting is
6347          * reflective, thus we will probably receive multi-
6348          * casts on those interfaces.
6349          * The ip_multirt_apply_membership() succeeds if
6350          * the operation succeeds on at least one interface.
6351          */
6352         if (IN6_IS_ADDR_V4MAPPED(&v6group)) {
6353                 ipaddr_t group;
6354 
6355                 IN6_V4MAPPED_TO_IPADDR(&v6group, group);
6356 
6357                 ire = ire_ftable_lookup_v4(group, IP_HOST_MASK, 0,
6358                     IRE_HOST | IRE_INTERFACE, NULL, ALL_ZONES, NULL,
6359                     MATCH_IRE_MASK | MATCH_IRE_TYPE, 0, ipst, NULL);
6360         } else {
6361                 ire = ire_ftable_lookup_v6(&v6group, &ipv6_all_ones, 0,
6362                     IRE_HOST | IRE_INTERFACE, NULL, ALL_ZONES, NULL,
6363                     MATCH_IRE_MASK | MATCH_IRE_TYPE, 0, ipst, NULL);
6364         }
6365         if (ire != NULL) {
6366                 if (ire->ire_flags & RTF_MULTIRT) {
6367                         error = ip_multirt_apply_membership(optfn, ire, connp,
6368                             checkonly, &v6group, fmode, &ipv6_all_zeros);
6369                         done = B_TRUE;
6370                 }
6371                 ire_refrele(ire);
6372         }
6373 
6374         if (!done) {
6375                 error = optfn(connp, checkonly, &v6group, ifaddr, ifindex,
6376                     fmode, &ipv6_all_zeros);
6377         }
6378         return (error);
6379 }
6380 
6381 /*
6382  * Set socket options for joining and leaving multicast groups
6383  * for specific sources.
6384  * Common to IPv4 and IPv6; inet6 indicates the type of socket.
6385  * The caller has already check that the option name is consistent with
6386  * the address family of the socket.
6387  */
6388 int
6389 ip_opt_set_multicast_sources(conn_t *connp, t_scalar_t name,
6390     uchar_t *invalp, boolean_t inet6, boolean_t checkonly)
6391 {
6392         int             *i1 = (int *)invalp;
6393         int             error = 0;
6394         ip_stack_t      *ipst = connp->conn_netstack->netstack_ip;
6395         struct ip_mreq_source *imreqp;
6396         struct group_source_req *gsreqp;
6397         in6_addr_t v6group, v6src;
6398         uint32_t ifindex;
6399         ipaddr_t ifaddr;
6400         boolean_t mcast_opt = B_TRUE;
6401         mcast_record_t fmode;
6402         ire_t *ire;
6403         boolean_t done = B_FALSE;
6404         int (*optfn)(conn_t *, boolean_t, const in6_addr_t *,
6405             ipaddr_t, uint_t, mcast_record_t, const in6_addr_t *);
6406 
6407         switch (name) {
6408         case IP_BLOCK_SOURCE:
6409                 mcast_opt = B_FALSE;
6410                 /* FALLTHRU */
6411         case MCAST_BLOCK_SOURCE:
6412                 fmode = MODE_IS_EXCLUDE;
6413                 optfn = ip_opt_add_group;
6414                 break;
6415 
6416         case IP_UNBLOCK_SOURCE:
6417                 mcast_opt = B_FALSE;
6418                 /* FALLTHRU */
6419         case MCAST_UNBLOCK_SOURCE:
6420                 fmode = MODE_IS_EXCLUDE;
6421                 optfn = ip_opt_delete_group;
6422                 break;
6423 
6424         case IP_ADD_SOURCE_MEMBERSHIP:
6425                 mcast_opt = B_FALSE;
6426                 /* FALLTHRU */
6427         case MCAST_JOIN_SOURCE_GROUP:
6428                 fmode = MODE_IS_INCLUDE;
6429                 optfn = ip_opt_add_group;
6430                 break;
6431 
6432         case IP_DROP_SOURCE_MEMBERSHIP:
6433                 mcast_opt = B_FALSE;
6434                 /* FALLTHRU */
6435         case MCAST_LEAVE_SOURCE_GROUP:
6436                 fmode = MODE_IS_INCLUDE;
6437                 optfn = ip_opt_delete_group;
6438                 break;
6439         default:
6440                 ASSERT(0);
6441         }
6442 
6443         if (mcast_opt) {
6444                 gsreqp = (struct group_source_req *)i1;
6445                 ifindex = gsreqp->gsr_interface;
6446                 if (gsreqp->gsr_group.ss_family == AF_INET) {
6447                         struct sockaddr_in *s;
6448                         s = (struct sockaddr_in *)&gsreqp->gsr_group;
6449                         IN6_INADDR_TO_V4MAPPED(&s->sin_addr, &v6group);
6450                         s = (struct sockaddr_in *)&gsreqp->gsr_source;
6451                         IN6_INADDR_TO_V4MAPPED(&s->sin_addr, &v6src);
6452                 } else {
6453                         struct sockaddr_in6 *s6;
6454 
6455                         if (!inet6)
6456                                 return (EINVAL);        /* Not on INET socket */
6457 
6458                         s6 = (struct sockaddr_in6 *)&gsreqp->gsr_group;
6459                         v6group = s6->sin6_addr;
6460                         s6 = (struct sockaddr_in6 *)&gsreqp->gsr_source;
6461                         v6src = s6->sin6_addr;
6462                 }
6463                 ifaddr = INADDR_ANY;
6464         } else {
6465                 imreqp = (struct ip_mreq_source *)i1;
6466                 IN6_INADDR_TO_V4MAPPED(&imreqp->imr_multiaddr, &v6group);
6467                 IN6_INADDR_TO_V4MAPPED(&imreqp->imr_sourceaddr, &v6src);
6468                 ifaddr = (ipaddr_t)imreqp->imr_interface.s_addr;
6469                 ifindex = 0;
6470         }
6471 
6472         /*
6473          * Handle src being mapped INADDR_ANY by changing it to unspecified.
6474          */
6475         if (IN6_IS_ADDR_V4MAPPED_ANY(&v6src))
6476                 v6src = ipv6_all_zeros;
6477 
6478         /*
6479          * In the multirouting case, we need to replicate
6480          * the request as noted in the mcast cases above.
6481          */
6482         if (IN6_IS_ADDR_V4MAPPED(&v6group)) {
6483                 ipaddr_t group;
6484 
6485                 IN6_V4MAPPED_TO_IPADDR(&v6group, group);
6486 
6487                 ire = ire_ftable_lookup_v4(group, IP_HOST_MASK, 0,
6488                     IRE_HOST | IRE_INTERFACE, NULL, ALL_ZONES, NULL,
6489                     MATCH_IRE_MASK | MATCH_IRE_TYPE, 0, ipst, NULL);
6490         } else {
6491                 ire = ire_ftable_lookup_v6(&v6group, &ipv6_all_ones, 0,
6492                     IRE_HOST | IRE_INTERFACE, NULL, ALL_ZONES, NULL,
6493                     MATCH_IRE_MASK | MATCH_IRE_TYPE, 0, ipst, NULL);
6494         }
6495         if (ire != NULL) {
6496                 if (ire->ire_flags & RTF_MULTIRT) {
6497                         error = ip_multirt_apply_membership(optfn, ire, connp,
6498                             checkonly, &v6group, fmode, &v6src);
6499                         done = B_TRUE;
6500                 }
6501                 ire_refrele(ire);
6502         }
6503         if (!done) {
6504                 error = optfn(connp, checkonly, &v6group, ifaddr, ifindex,
6505                     fmode, &v6src);
6506         }
6507         return (error);
6508 }
6509 
6510 /*
6511  * Given a destination address and a pointer to where to put the information
6512  * this routine fills in the mtuinfo.
6513  * The socket must be connected.
6514  * For sctp conn_faddr is the primary address.
6515  */
6516 int
6517 ip_fill_mtuinfo(conn_t *connp, ip_xmit_attr_t *ixa, struct ip6_mtuinfo *mtuinfo)
6518 {
6519         uint32_t        pmtu = IP_MAXPACKET;
6520         uint_t          scopeid;
6521 
6522         if (IN6_IS_ADDR_UNSPECIFIED(&connp->conn_faddr_v6))
6523                 return (-1);
6524 
6525         /* In case we never sent or called ip_set_destination_v4/v6 */
6526         if (ixa->ixa_ire != NULL)
6527                 pmtu = ip_get_pmtu(ixa);
6528 
6529         if (ixa->ixa_flags & IXAF_SCOPEID_SET)
6530                 scopeid = ixa->ixa_scopeid;
6531         else
6532                 scopeid = 0;
6533 
6534         bzero(mtuinfo, sizeof (*mtuinfo));
6535         mtuinfo->ip6m_addr.sin6_family = AF_INET6;
6536         mtuinfo->ip6m_addr.sin6_port = connp->conn_fport;
6537         mtuinfo->ip6m_addr.sin6_addr = connp->conn_faddr_v6;
6538         mtuinfo->ip6m_addr.sin6_scope_id = scopeid;
6539         mtuinfo->ip6m_mtu = pmtu;
6540 
6541         return (sizeof (struct ip6_mtuinfo));
6542 }
6543 
6544 /*
6545  * When the src multihoming is changed from weak to [strong, preferred]
6546  * ip_ire_rebind_walker is called to walk the list of all ire_t entries
6547  * and identify routes that were created by user-applications in the
6548  * unbound state (i.e., without RTA_IFP), and for which an ire_ill is not
6549  * currently defined. These routes are then 'rebound', i.e., their ire_ill
6550  * is selected by finding an interface route for the gateway.
6551  */
6552 /* ARGSUSED */
6553 void
6554 ip_ire_rebind_walker(ire_t *ire, void *notused)
6555 {
6556         if (!ire->ire_unbound || ire->ire_ill != NULL)
6557                 return;
6558         ire_rebind(ire);
6559         ire_delete(ire);
6560 }
6561 
6562 /*
6563  * When the src multihoming is changed from  [strong, preferred] to weak,
6564  * ip_ire_unbind_walker is called to walk the list of all ire_t entries, and
6565  * set any entries that were created by user-applications in the unbound state
6566  * (i.e., without RTA_IFP) back to having a NULL ire_ill.
6567  */
6568 /* ARGSUSED */
6569 void
6570 ip_ire_unbind_walker(ire_t *ire, void *notused)
6571 {
6572         ire_t *new_ire;
6573 
6574         if (!ire->ire_unbound || ire->ire_ill == NULL)
6575                 return;
6576         if (ire->ire_ipversion == IPV6_VERSION) {
6577                 new_ire = ire_create_v6(&ire->ire_addr_v6, &ire->ire_mask_v6,
6578                     &ire->ire_gateway_addr_v6, ire->ire_type, NULL,
6579                     ire->ire_zoneid, ire->ire_flags, NULL, ire->ire_ipst);
6580         } else {
6581                 new_ire = ire_create((uchar_t *)&ire->ire_addr,
6582                     (uchar_t *)&ire->ire_mask,
6583                     (uchar_t *)&ire->ire_gateway_addr, ire->ire_type, NULL,
6584                     ire->ire_zoneid, ire->ire_flags, NULL, ire->ire_ipst);
6585         }
6586         if (new_ire == NULL)
6587                 return;
6588         new_ire->ire_unbound = B_TRUE;
6589         /*
6590          * The bound ire must first be deleted so that we don't return
6591          * the existing one on the attempt to add the unbound new_ire.
6592          */
6593         ire_delete(ire);
6594         new_ire = ire_add(new_ire);
6595         if (new_ire != NULL)
6596                 ire_refrele(new_ire);
6597 }
6598 
6599 /*
6600  * When the settings of ip*_strict_src_multihoming tunables are changed,
6601  * all cached routes need to be recomputed. This recomputation needs to be
6602  * done when going from weaker to stronger modes so that the cached ire
6603  * for the connection does not violate the current ip*_strict_src_multihoming
6604  * setting. It also needs to be done when going from stronger to weaker modes,
6605  * so that we fall back to matching on the longest-matching-route (as opposed
6606  * to a shorter match that may have been selected in the strong mode
6607  * to satisfy src_multihoming settings).
6608  *
6609  * The cached ixa_ire entires for all conn_t entries are marked as
6610  * "verify" so that they will be recomputed for the next packet.
6611  */
6612 void
6613 conn_ire_revalidate(conn_t *connp, void *arg)
6614 {
6615         boolean_t isv6 = (boolean_t)arg;
6616 
6617         if ((isv6 && connp->conn_ipversion != IPV6_VERSION) ||
6618             (!isv6 && connp->conn_ipversion != IPV4_VERSION))
6619                 return;
6620         connp->conn_ixa->ixa_ire_generation = IRE_GENERATION_VERIFY;
6621 }
6622 
6623 /*
6624  * Handles both IPv4 and IPv6 reassembly - doing the out-of-order cases,
6625  * When an ipf is passed here for the first time, if
6626  * we already have in-order fragments on the queue, we convert from the fast-
6627  * path reassembly scheme to the hard-case scheme.  From then on, additional
6628  * fragments are reassembled here.  We keep track of the start and end offsets
6629  * of each piece, and the number of holes in the chain.  When the hole count
6630  * goes to zero, we are done!
6631  *
6632  * The ipf_count will be updated to account for any mblk(s) added (pointed to
6633  * by mp) or subtracted (freeb()ed dups), upon return the caller must update
6634  * ipfb_count and ill_frag_count by the difference of ipf_count before and
6635  * after the call to ip_reassemble().
6636  */
6637 int
6638 ip_reassemble(mblk_t *mp, ipf_t *ipf, uint_t start, boolean_t more, ill_t *ill,
6639     size_t msg_len)
6640 {
6641         uint_t  end;
6642         mblk_t  *next_mp;
6643         mblk_t  *mp1;
6644         uint_t  offset;
6645         boolean_t incr_dups = B_TRUE;
6646         boolean_t offset_zero_seen = B_FALSE;
6647         boolean_t pkt_boundary_checked = B_FALSE;
6648 
6649         /* If start == 0 then ipf_nf_hdr_len has to be set. */
6650         ASSERT(start != 0 || ipf->ipf_nf_hdr_len != 0);
6651 
6652         /* Add in byte count */
6653         ipf->ipf_count += msg_len;
6654         if (ipf->ipf_end) {
6655                 /*
6656                  * We were part way through in-order reassembly, but now there
6657                  * is a hole.  We walk through messages already queued, and
6658                  * mark them for hard case reassembly.  We know that up till
6659                  * now they were in order starting from offset zero.
6660                  */
6661                 offset = 0;
6662                 for (mp1 = ipf->ipf_mp->b_cont; mp1; mp1 = mp1->b_cont) {
6663                         IP_REASS_SET_START(mp1, offset);
6664                         if (offset == 0) {
6665                                 ASSERT(ipf->ipf_nf_hdr_len != 0);
6666                                 offset = -ipf->ipf_nf_hdr_len;
6667                         }
6668                         offset += mp1->b_wptr - mp1->b_rptr;
6669                         IP_REASS_SET_END(mp1, offset);
6670                 }
6671                 /* One hole at the end. */
6672                 ipf->ipf_hole_cnt = 1;
6673                 /* Brand it as a hard case, forever. */
6674                 ipf->ipf_end = 0;
6675         }
6676         /* Walk through all the new pieces. */
6677         do {
6678                 end = start + (mp->b_wptr - mp->b_rptr);
6679                 /*
6680                  * If start is 0, decrease 'end' only for the first mblk of
6681                  * the fragment. Otherwise 'end' can get wrong value in the
6682                  * second pass of the loop if first mblk is exactly the
6683                  * size of ipf_nf_hdr_len.
6684                  */
6685                 if (start == 0 && !offset_zero_seen) {
6686                         /* First segment */
6687                         ASSERT(ipf->ipf_nf_hdr_len != 0);
6688                         end -= ipf->ipf_nf_hdr_len;
6689                         offset_zero_seen = B_TRUE;
6690                 }
6691                 next_mp = mp->b_cont;
6692                 /*
6693                  * We are checking to see if there is any interesing data
6694                  * to process.  If there isn't and the mblk isn't the
6695                  * one which carries the unfragmentable header then we
6696                  * drop it.  It's possible to have just the unfragmentable
6697                  * header come through without any data.  That needs to be
6698                  * saved.
6699                  *
6700                  * If the assert at the top of this function holds then the
6701                  * term "ipf->ipf_nf_hdr_len != 0" isn't needed.  This code
6702                  * is infrequently traveled enough that the test is left in
6703                  * to protect against future code changes which break that
6704                  * invariant.
6705                  */
6706                 if (start == end && start != 0 && ipf->ipf_nf_hdr_len != 0) {
6707                         /* Empty.  Blast it. */
6708                         IP_REASS_SET_START(mp, 0);
6709                         IP_REASS_SET_END(mp, 0);
6710                         /*
6711                          * If the ipf points to the mblk we are about to free,
6712                          * update ipf to point to the next mblk (or NULL
6713                          * if none).
6714                          */
6715                         if (ipf->ipf_mp->b_cont == mp)
6716                                 ipf->ipf_mp->b_cont = next_mp;
6717                         freeb(mp);
6718                         continue;
6719                 }
6720                 mp->b_cont = NULL;
6721                 IP_REASS_SET_START(mp, start);
6722                 IP_REASS_SET_END(mp, end);
6723                 if (!ipf->ipf_tail_mp) {
6724                         ipf->ipf_tail_mp = mp;
6725                         ipf->ipf_mp->b_cont = mp;
6726                         if (start == 0 || !more) {
6727                                 ipf->ipf_hole_cnt = 1;
6728                                 /*
6729                                  * if the first fragment comes in more than one
6730                                  * mblk, this loop will be executed for each
6731                                  * mblk. Need to adjust hole count so exiting
6732                                  * this routine will leave hole count at 1.
6733                                  */
6734                                 if (next_mp)
6735                                         ipf->ipf_hole_cnt++;
6736                         } else
6737                                 ipf->ipf_hole_cnt = 2;
6738                         continue;
6739                 } else if (ipf->ipf_last_frag_seen && !more &&
6740                     !pkt_boundary_checked) {
6741                         /*
6742                          * We check datagram boundary only if this fragment
6743                          * claims to be the last fragment and we have seen a
6744                          * last fragment in the past too. We do this only
6745                          * once for a given fragment.
6746                          *
6747                          * start cannot be 0 here as fragments with start=0
6748                          * and MF=0 gets handled as a complete packet. These
6749                          * fragments should not reach here.
6750                          */
6751 
6752                         if (start + msgdsize(mp) !=
6753                             IP_REASS_END(ipf->ipf_tail_mp)) {
6754                                 /*
6755                                  * We have two fragments both of which claim
6756                                  * to be the last fragment but gives conflicting
6757                                  * information about the whole datagram size.
6758                                  * Something fishy is going on. Drop the
6759                                  * fragment and free up the reassembly list.
6760                                  */
6761                                 return (IP_REASS_FAILED);
6762                         }
6763 
6764                         /*
6765                          * We shouldn't come to this code block again for this
6766                          * particular fragment.
6767                          */
6768                         pkt_boundary_checked = B_TRUE;
6769                 }
6770 
6771                 /* New stuff at or beyond tail? */
6772                 offset = IP_REASS_END(ipf->ipf_tail_mp);
6773                 if (start >= offset) {
6774                         if (ipf->ipf_last_frag_seen) {
6775                                 /* current fragment is beyond last fragment */
6776                                 return (IP_REASS_FAILED);
6777                         }
6778                         /* Link it on end. */
6779                         ipf->ipf_tail_mp->b_cont = mp;
6780                         ipf->ipf_tail_mp = mp;
6781                         if (more) {
6782                                 if (start != offset)
6783                                         ipf->ipf_hole_cnt++;
6784                         } else if (start == offset && next_mp == NULL)
6785                                         ipf->ipf_hole_cnt--;
6786                         continue;
6787                 }
6788                 mp1 = ipf->ipf_mp->b_cont;
6789                 offset = IP_REASS_START(mp1);
6790                 /* New stuff at the front? */
6791                 if (start < offset) {
6792                         if (start == 0) {
6793                                 if (end >= offset) {
6794                                         /* Nailed the hole at the begining. */
6795                                         ipf->ipf_hole_cnt--;
6796                                 }
6797                         } else if (end < offset) {
6798                                 /*
6799                                  * A hole, stuff, and a hole where there used
6800                                  * to be just a hole.
6801                                  */
6802                                 ipf->ipf_hole_cnt++;
6803                         }
6804                         mp->b_cont = mp1;
6805                         /* Check for overlap. */
6806                         while (end > offset) {
6807                                 if (end < IP_REASS_END(mp1)) {
6808                                         mp->b_wptr -= end - offset;
6809                                         IP_REASS_SET_END(mp, offset);
6810                                         BUMP_MIB(ill->ill_ip_mib,
6811                                             ipIfStatsReasmPartDups);
6812                                         break;
6813                                 }
6814                                 /* Did we cover another hole? */
6815                                 if ((mp1->b_cont &&
6816                                     IP_REASS_END(mp1) !=
6817                                     IP_REASS_START(mp1->b_cont) &&
6818                                     end >= IP_REASS_START(mp1->b_cont)) ||
6819                                     (!ipf->ipf_last_frag_seen && !more)) {
6820                                         ipf->ipf_hole_cnt--;
6821                                 }
6822                                 /* Clip out mp1. */
6823                                 if ((mp->b_cont = mp1->b_cont) == NULL) {
6824                                         /*
6825                                          * After clipping out mp1, this guy
6826                                          * is now hanging off the end.
6827                                          */
6828                                         ipf->ipf_tail_mp = mp;
6829                                 }
6830                                 IP_REASS_SET_START(mp1, 0);
6831                                 IP_REASS_SET_END(mp1, 0);
6832                                 /* Subtract byte count */
6833                                 ipf->ipf_count -= mp1->b_datap->db_lim -
6834                                     mp1->b_datap->db_base;
6835                                 freeb(mp1);
6836                                 BUMP_MIB(ill->ill_ip_mib,
6837                                     ipIfStatsReasmPartDups);
6838                                 mp1 = mp->b_cont;
6839                                 if (!mp1)
6840                                         break;
6841                                 offset = IP_REASS_START(mp1);
6842                         }
6843                         ipf->ipf_mp->b_cont = mp;
6844                         continue;
6845                 }
6846                 /*
6847                  * The new piece starts somewhere between the start of the head
6848                  * and before the end of the tail.
6849                  */
6850                 for (; mp1; mp1 = mp1->b_cont) {
6851                         offset = IP_REASS_END(mp1);
6852                         if (start < offset) {
6853                                 if (end <= offset) {
6854                                         /* Nothing new. */
6855                                         IP_REASS_SET_START(mp, 0);
6856                                         IP_REASS_SET_END(mp, 0);
6857                                         /* Subtract byte count */
6858                                         ipf->ipf_count -= mp->b_datap->db_lim -
6859                                             mp->b_datap->db_base;
6860                                         if (incr_dups) {
6861                                                 ipf->ipf_num_dups++;
6862                                                 incr_dups = B_FALSE;
6863                                         }
6864                                         freeb(mp);
6865                                         BUMP_MIB(ill->ill_ip_mib,
6866                                             ipIfStatsReasmDuplicates);
6867                                         break;
6868                                 }
6869                                 /*
6870                                  * Trim redundant stuff off beginning of new
6871                                  * piece.
6872                                  */
6873                                 IP_REASS_SET_START(mp, offset);
6874                                 mp->b_rptr += offset - start;
6875                                 BUMP_MIB(ill->ill_ip_mib,
6876                                     ipIfStatsReasmPartDups);
6877                                 start = offset;
6878                                 if (!mp1->b_cont) {
6879                                         /*
6880                                          * After trimming, this guy is now
6881                                          * hanging off the end.
6882                                          */
6883                                         mp1->b_cont = mp;
6884                                         ipf->ipf_tail_mp = mp;
6885                                         if (!more) {
6886                                                 ipf->ipf_hole_cnt--;
6887                                         }
6888                                         break;
6889                                 }
6890                         }
6891                         if (start >= IP_REASS_START(mp1->b_cont))
6892                                 continue;
6893                         /* Fill a hole */
6894                         if (start > offset)
6895                                 ipf->ipf_hole_cnt++;
6896                         mp->b_cont = mp1->b_cont;
6897                         mp1->b_cont = mp;
6898                         mp1 = mp->b_cont;
6899                         offset = IP_REASS_START(mp1);
6900                         if (end >= offset) {
6901                                 ipf->ipf_hole_cnt--;
6902                                 /* Check for overlap. */
6903                                 while (end > offset) {
6904                                         if (end < IP_REASS_END(mp1)) {
6905                                                 mp->b_wptr -= end - offset;
6906                                                 IP_REASS_SET_END(mp, offset);
6907                                                 /*
6908                                                  * TODO we might bump
6909                                                  * this up twice if there is
6910                                                  * overlap at both ends.
6911                                                  */
6912                                                 BUMP_MIB(ill->ill_ip_mib,
6913                                                     ipIfStatsReasmPartDups);
6914                                                 break;
6915                                         }
6916                                         /* Did we cover another hole? */
6917                                         if ((mp1->b_cont &&
6918                                             IP_REASS_END(mp1)
6919                                             != IP_REASS_START(mp1->b_cont) &&
6920                                             end >=
6921                                             IP_REASS_START(mp1->b_cont)) ||
6922                                             (!ipf->ipf_last_frag_seen &&
6923                                             !more)) {
6924                                                 ipf->ipf_hole_cnt--;
6925                                         }
6926                                         /* Clip out mp1. */
6927                                         if ((mp->b_cont = mp1->b_cont) ==
6928                                             NULL) {
6929                                                 /*
6930                                                  * After clipping out mp1,
6931                                                  * this guy is now hanging
6932                                                  * off the end.
6933                                                  */
6934                                                 ipf->ipf_tail_mp = mp;
6935                                         }
6936                                         IP_REASS_SET_START(mp1, 0);
6937                                         IP_REASS_SET_END(mp1, 0);
6938                                         /* Subtract byte count */
6939                                         ipf->ipf_count -=
6940                                             mp1->b_datap->db_lim -
6941                                             mp1->b_datap->db_base;
6942                                         freeb(mp1);
6943                                         BUMP_MIB(ill->ill_ip_mib,
6944                                             ipIfStatsReasmPartDups);
6945                                         mp1 = mp->b_cont;
6946                                         if (!mp1)
6947                                                 break;
6948                                         offset = IP_REASS_START(mp1);
6949                                 }
6950                         }
6951                         break;
6952                 }
6953         } while (start = end, mp = next_mp);
6954 
6955         /* Fragment just processed could be the last one. Remember this fact */
6956         if (!more)
6957                 ipf->ipf_last_frag_seen = B_TRUE;
6958 
6959         /* Still got holes? */
6960         if (ipf->ipf_hole_cnt)
6961                 return (IP_REASS_PARTIAL);
6962         /* Clean up overloaded fields to avoid upstream disasters. */
6963         for (mp1 = ipf->ipf_mp->b_cont; mp1; mp1 = mp1->b_cont) {
6964                 IP_REASS_SET_START(mp1, 0);
6965                 IP_REASS_SET_END(mp1, 0);
6966         }
6967         return (IP_REASS_COMPLETE);
6968 }
6969 
6970 /*
6971  * Fragmentation reassembly.  Each ILL has a hash table for
6972  * queuing packets undergoing reassembly for all IPIFs
6973  * associated with the ILL.  The hash is based on the packet
6974  * IP ident field.  The ILL frag hash table was allocated
6975  * as a timer block at the time the ILL was created.  Whenever
6976  * there is anything on the reassembly queue, the timer will
6977  * be running.  Returns the reassembled packet if reassembly completes.
6978  */
6979 mblk_t *
6980 ip_input_fragment(mblk_t *mp, ipha_t *ipha, ip_recv_attr_t *ira)
6981 {
6982         uint32_t        frag_offset_flags;
6983         mblk_t          *t_mp;
6984         ipaddr_t        dst;
6985         uint8_t         proto = ipha->ipha_protocol;
6986         uint32_t        sum_val;
6987         uint16_t        sum_flags;
6988         ipf_t           *ipf;
6989         ipf_t           **ipfp;
6990         ipfb_t          *ipfb;
6991         uint16_t        ident;
6992         uint32_t        offset;
6993         ipaddr_t        src;
6994         uint_t          hdr_length;
6995         uint32_t        end;
6996         mblk_t          *mp1;
6997         mblk_t          *tail_mp;
6998         size_t          count;
6999         size_t          msg_len;
7000         uint8_t         ecn_info = 0;
7001         uint32_t        packet_size;
7002         boolean_t       pruned = B_FALSE;
7003         ill_t           *ill = ira->ira_ill;
7004         ip_stack_t      *ipst = ill->ill_ipst;
7005 
7006         /*
7007          * Drop the fragmented as early as possible, if
7008          * we don't have resource(s) to re-assemble.
7009          */
7010         if (ipst->ips_ip_reass_queue_bytes == 0) {
7011                 freemsg(mp);
7012                 return (NULL);
7013         }
7014 
7015         /* Check for fragmentation offset; return if there's none */
7016         if ((frag_offset_flags = ntohs(ipha->ipha_fragment_offset_and_flags) &
7017             (IPH_MF | IPH_OFFSET)) == 0)
7018                 return (mp);
7019 
7020         /*
7021          * We utilize hardware computed checksum info only for UDP since
7022          * IP fragmentation is a normal occurrence for the protocol.  In
7023          * addition, checksum offload support for IP fragments carrying
7024          * UDP payload is commonly implemented across network adapters.
7025          */
7026         ASSERT(ira->ira_rill != NULL);
7027         if (proto == IPPROTO_UDP && dohwcksum &&
7028             ILL_HCKSUM_CAPABLE(ira->ira_rill) &&
7029             (DB_CKSUMFLAGS(mp) & (HCK_FULLCKSUM | HCK_PARTIALCKSUM))) {
7030                 mblk_t *mp1 = mp->b_cont;
7031                 int32_t len;
7032 
7033                 /* Record checksum information from the packet */
7034                 sum_val = (uint32_t)DB_CKSUM16(mp);
7035                 sum_flags = DB_CKSUMFLAGS(mp);
7036 
7037                 /* IP payload offset from beginning of mblk */
7038                 offset = ((uchar_t *)ipha + IPH_HDR_LENGTH(ipha)) - mp->b_rptr;
7039 
7040                 if ((sum_flags & HCK_PARTIALCKSUM) &&
7041                     (mp1 == NULL || mp1->b_cont == NULL) &&
7042                     offset >= DB_CKSUMSTART(mp) &&
7043                     ((len = offset - DB_CKSUMSTART(mp)) & 1) == 0) {
7044                         uint32_t adj;
7045                         /*
7046                          * Partial checksum has been calculated by hardware
7047                          * and attached to the packet; in addition, any
7048                          * prepended extraneous data is even byte aligned.
7049                          * If any such data exists, we adjust the checksum;
7050                          * this would also handle any postpended data.
7051                          */
7052                         IP_ADJCKSUM_PARTIAL(mp->b_rptr + DB_CKSUMSTART(mp),
7053                             mp, mp1, len, adj);
7054 
7055                         /* One's complement subtract extraneous checksum */
7056                         if (adj >= sum_val)
7057                                 sum_val = ~(adj - sum_val) & 0xFFFF;
7058                         else
7059                                 sum_val -= adj;
7060                 }
7061         } else {
7062                 sum_val = 0;
7063                 sum_flags = 0;
7064         }
7065 
7066         /* Clear hardware checksumming flag */
7067         DB_CKSUMFLAGS(mp) = 0;
7068 
7069         ident = ipha->ipha_ident;
7070         offset = (frag_offset_flags << 3) & 0xFFFF;
7071         src = ipha->ipha_src;
7072         dst = ipha->ipha_dst;
7073         hdr_length = IPH_HDR_LENGTH(ipha);
7074         end = ntohs(ipha->ipha_length) - hdr_length;
7075 
7076         /* If end == 0 then we have a packet with no data, so just free it */
7077         if (end == 0) {
7078                 freemsg(mp);
7079                 return (NULL);
7080         }
7081 
7082         /* Record the ECN field info. */
7083         ecn_info = (ipha->ipha_type_of_service & 0x3);
7084         if (offset != 0) {
7085                 /*
7086                  * If this isn't the first piece, strip the header, and
7087                  * add the offset to the end value.
7088                  */
7089                 mp->b_rptr += hdr_length;
7090                 end += offset;
7091         }
7092 
7093         /* Handle vnic loopback of fragments */
7094         if (mp->b_datap->db_ref > 2)
7095                 msg_len = 0;
7096         else
7097                 msg_len = MBLKSIZE(mp);
7098 
7099         tail_mp = mp;
7100         while (tail_mp->b_cont != NULL) {
7101                 tail_mp = tail_mp->b_cont;
7102                 if (tail_mp->b_datap->db_ref <= 2)
7103                         msg_len += MBLKSIZE(tail_mp);
7104         }
7105 
7106         /* If the reassembly list for this ILL will get too big, prune it */
7107         if ((msg_len + sizeof (*ipf) + ill->ill_frag_count) >=
7108             ipst->ips_ip_reass_queue_bytes) {
7109                 DTRACE_PROBE3(ip_reass_queue_bytes, uint_t, msg_len,
7110                     uint_t, ill->ill_frag_count,
7111                     uint_t, ipst->ips_ip_reass_queue_bytes);
7112                 ill_frag_prune(ill,
7113                     (ipst->ips_ip_reass_queue_bytes < msg_len) ? 0 :
7114                     (ipst->ips_ip_reass_queue_bytes - msg_len));
7115                 pruned = B_TRUE;
7116         }
7117 
7118         ipfb = &ill->ill_frag_hash_tbl[ILL_FRAG_HASH(src, ident)];
7119         mutex_enter(&ipfb->ipfb_lock);
7120 
7121         ipfp = &ipfb->ipfb_ipf;
7122         /* Try to find an existing fragment queue for this packet. */
7123         for (;;) {
7124                 ipf = ipfp[0];
7125                 if (ipf != NULL) {
7126                         /*
7127                          * It has to match on ident and src/dst address.
7128                          */
7129                         if (ipf->ipf_ident == ident &&
7130                             ipf->ipf_src == src &&
7131                             ipf->ipf_dst == dst &&
7132                             ipf->ipf_protocol == proto) {
7133                                 /*
7134                                  * If we have received too many
7135                                  * duplicate fragments for this packet
7136                                  * free it.
7137                                  */
7138                                 if (ipf->ipf_num_dups > ip_max_frag_dups) {
7139                                         ill_frag_free_pkts(ill, ipfb, ipf, 1);
7140                                         freemsg(mp);
7141                                         mutex_exit(&ipfb->ipfb_lock);
7142                                         return (NULL);
7143                                 }
7144                                 /* Found it. */
7145                                 break;
7146                         }
7147                         ipfp = &ipf->ipf_hash_next;
7148                         continue;
7149                 }
7150 
7151                 /*
7152                  * If we pruned the list, do we want to store this new
7153                  * fragment?. We apply an optimization here based on the
7154                  * fact that most fragments will be received in order.
7155                  * So if the offset of this incoming fragment is zero,
7156                  * it is the first fragment of a new packet. We will
7157                  * keep it.  Otherwise drop the fragment, as we have
7158                  * probably pruned the packet already (since the
7159                  * packet cannot be found).
7160                  */
7161                 if (pruned && offset != 0) {
7162                         mutex_exit(&ipfb->ipfb_lock);
7163                         freemsg(mp);
7164                         return (NULL);
7165                 }
7166 
7167                 if (ipfb->ipfb_frag_pkts >= MAX_FRAG_PKTS(ipst))  {
7168                         /*
7169                          * Too many fragmented packets in this hash
7170                          * bucket. Free the oldest.
7171                          */
7172                         ill_frag_free_pkts(ill, ipfb, ipfb->ipfb_ipf, 1);
7173                 }
7174 
7175                 /* New guy.  Allocate a frag message. */
7176                 mp1 = allocb(sizeof (*ipf), BPRI_MED);
7177                 if (mp1 == NULL) {
7178                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
7179                         ip_drop_input("ipIfStatsInDiscards", mp, ill);
7180                         freemsg(mp);
7181 reass_done:
7182                         mutex_exit(&ipfb->ipfb_lock);
7183                         return (NULL);
7184                 }
7185 
7186                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsReasmReqds);
7187                 mp1->b_cont = mp;
7188 
7189                 /* Initialize the fragment header. */
7190                 ipf = (ipf_t *)mp1->b_rptr;
7191                 ipf->ipf_mp = mp1;
7192                 ipf->ipf_ptphn = ipfp;
7193                 ipfp[0] = ipf;
7194                 ipf->ipf_hash_next = NULL;
7195                 ipf->ipf_ident = ident;
7196                 ipf->ipf_protocol = proto;
7197                 ipf->ipf_src = src;
7198                 ipf->ipf_dst = dst;
7199                 ipf->ipf_nf_hdr_len = 0;
7200                 /* Record reassembly start time. */
7201                 ipf->ipf_timestamp = gethrestime_sec();
7202                 /* Record ipf generation and account for frag header */
7203                 ipf->ipf_gen = ill->ill_ipf_gen++;
7204                 ipf->ipf_count = MBLKSIZE(mp1);
7205                 ipf->ipf_last_frag_seen = B_FALSE;
7206                 ipf->ipf_ecn = ecn_info;
7207                 ipf->ipf_num_dups = 0;
7208                 ipfb->ipfb_frag_pkts++;
7209                 ipf->ipf_checksum = 0;
7210                 ipf->ipf_checksum_flags = 0;
7211 
7212                 /* Store checksum value in fragment header */
7213                 if (sum_flags != 0) {
7214                         sum_val = (sum_val & 0xFFFF) + (sum_val >> 16);
7215                         sum_val = (sum_val & 0xFFFF) + (sum_val >> 16);
7216                         ipf->ipf_checksum = sum_val;
7217                         ipf->ipf_checksum_flags = sum_flags;
7218                 }
7219 
7220                 /*
7221                  * We handle reassembly two ways.  In the easy case,
7222                  * where all the fragments show up in order, we do
7223                  * minimal bookkeeping, and just clip new pieces on
7224                  * the end.  If we ever see a hole, then we go off
7225                  * to ip_reassemble which has to mark the pieces and
7226                  * keep track of the number of holes, etc.  Obviously,
7227                  * the point of having both mechanisms is so we can
7228                  * handle the easy case as efficiently as possible.
7229                  */
7230                 if (offset == 0) {
7231                         /* Easy case, in-order reassembly so far. */
7232                         ipf->ipf_count += msg_len;
7233                         ipf->ipf_tail_mp = tail_mp;
7234                         /*
7235                          * Keep track of next expected offset in
7236                          * ipf_end.
7237                          */
7238                         ipf->ipf_end = end;
7239                         ipf->ipf_nf_hdr_len = hdr_length;
7240                 } else {
7241                         /* Hard case, hole at the beginning. */
7242                         ipf->ipf_tail_mp = NULL;
7243                         /*
7244                          * ipf_end == 0 means that we have given up
7245                          * on easy reassembly.
7246                          */
7247                         ipf->ipf_end = 0;
7248 
7249                         /* Forget checksum offload from now on */
7250                         ipf->ipf_checksum_flags = 0;
7251 
7252                         /*
7253                          * ipf_hole_cnt is set by ip_reassemble.
7254                          * ipf_count is updated by ip_reassemble.
7255                          * No need to check for return value here
7256                          * as we don't expect reassembly to complete
7257                          * or fail for the first fragment itself.
7258                          */
7259                         (void) ip_reassemble(mp, ipf,
7260                             (frag_offset_flags & IPH_OFFSET) << 3,
7261                             (frag_offset_flags & IPH_MF), ill, msg_len);
7262                 }
7263                 /* Update per ipfb and ill byte counts */
7264                 ipfb->ipfb_count += ipf->ipf_count;
7265                 ASSERT(ipfb->ipfb_count > 0);     /* Wraparound */
7266                 atomic_add_32(&ill->ill_frag_count, ipf->ipf_count);
7267                 /* If the frag timer wasn't already going, start it. */
7268                 mutex_enter(&ill->ill_lock);
7269                 ill_frag_timer_start(ill);
7270                 mutex_exit(&ill->ill_lock);
7271                 goto reass_done;
7272         }
7273 
7274         /*
7275          * If the packet's flag has changed (it could be coming up
7276          * from an interface different than the previous, therefore
7277          * possibly different checksum capability), then forget about
7278          * any stored checksum states.  Otherwise add the value to
7279          * the existing one stored in the fragment header.
7280          */
7281         if (sum_flags != 0 && sum_flags == ipf->ipf_checksum_flags) {
7282                 sum_val += ipf->ipf_checksum;
7283                 sum_val = (sum_val & 0xFFFF) + (sum_val >> 16);
7284                 sum_val = (sum_val & 0xFFFF) + (sum_val >> 16);
7285                 ipf->ipf_checksum = sum_val;
7286         } else if (ipf->ipf_checksum_flags != 0) {
7287                 /* Forget checksum offload from now on */
7288                 ipf->ipf_checksum_flags = 0;
7289         }
7290 
7291         /*
7292          * We have a new piece of a datagram which is already being
7293          * reassembled.  Update the ECN info if all IP fragments
7294          * are ECN capable.  If there is one which is not, clear
7295          * all the info.  If there is at least one which has CE
7296          * code point, IP needs to report that up to transport.
7297          */
7298         if (ecn_info != IPH_ECN_NECT && ipf->ipf_ecn != IPH_ECN_NECT) {
7299                 if (ecn_info == IPH_ECN_CE)
7300                         ipf->ipf_ecn = IPH_ECN_CE;
7301         } else {
7302                 ipf->ipf_ecn = IPH_ECN_NECT;
7303         }
7304         if (offset && ipf->ipf_end == offset) {
7305                 /* The new fragment fits at the end */
7306                 ipf->ipf_tail_mp->b_cont = mp;
7307                 /* Update the byte count */
7308                 ipf->ipf_count += msg_len;
7309                 /* Update per ipfb and ill byte counts */
7310                 ipfb->ipfb_count += msg_len;
7311                 ASSERT(ipfb->ipfb_count > 0);     /* Wraparound */
7312                 atomic_add_32(&ill->ill_frag_count, msg_len);
7313                 if (frag_offset_flags & IPH_MF) {
7314                         /* More to come. */
7315                         ipf->ipf_end = end;
7316                         ipf->ipf_tail_mp = tail_mp;
7317                         goto reass_done;
7318                 }
7319         } else {
7320                 /* Go do the hard cases. */
7321                 int ret;
7322 
7323                 if (offset == 0)
7324                         ipf->ipf_nf_hdr_len = hdr_length;
7325 
7326                 /* Save current byte count */
7327                 count = ipf->ipf_count;
7328                 ret = ip_reassemble(mp, ipf,
7329                     (frag_offset_flags & IPH_OFFSET) << 3,
7330                     (frag_offset_flags & IPH_MF), ill, msg_len);
7331                 /* Count of bytes added and subtracted (freeb()ed) */
7332                 count = ipf->ipf_count - count;
7333                 if (count) {
7334                         /* Update per ipfb and ill byte counts */
7335                         ipfb->ipfb_count += count;
7336                         ASSERT(ipfb->ipfb_count > 0); /* Wraparound */
7337                         atomic_add_32(&ill->ill_frag_count, count);
7338                 }
7339                 if (ret == IP_REASS_PARTIAL) {
7340                         goto reass_done;
7341                 } else if (ret == IP_REASS_FAILED) {
7342                         /* Reassembly failed. Free up all resources */
7343                         ill_frag_free_pkts(ill, ipfb, ipf, 1);
7344                         for (t_mp = mp; t_mp != NULL; t_mp = t_mp->b_cont) {
7345                                 IP_REASS_SET_START(t_mp, 0);
7346                                 IP_REASS_SET_END(t_mp, 0);
7347                         }
7348                         freemsg(mp);
7349                         goto reass_done;
7350                 }
7351                 /* We will reach here iff 'ret' is IP_REASS_COMPLETE */
7352         }
7353         /*
7354          * We have completed reassembly.  Unhook the frag header from
7355          * the reassembly list.
7356          *
7357          * Before we free the frag header, record the ECN info
7358          * to report back to the transport.
7359          */
7360         ecn_info = ipf->ipf_ecn;
7361         BUMP_MIB(ill->ill_ip_mib, ipIfStatsReasmOKs);
7362         ipfp = ipf->ipf_ptphn;
7363 
7364         /* We need to supply these to caller */
7365         if ((sum_flags = ipf->ipf_checksum_flags) != 0)
7366                 sum_val = ipf->ipf_checksum;
7367         else
7368                 sum_val = 0;
7369 
7370         mp1 = ipf->ipf_mp;
7371         count = ipf->ipf_count;
7372         ipf = ipf->ipf_hash_next;
7373         if (ipf != NULL)
7374                 ipf->ipf_ptphn = ipfp;
7375         ipfp[0] = ipf;
7376         atomic_add_32(&ill->ill_frag_count, -count);
7377         ASSERT(ipfb->ipfb_count >= count);
7378         ipfb->ipfb_count -= count;
7379         ipfb->ipfb_frag_pkts--;
7380         mutex_exit(&ipfb->ipfb_lock);
7381         /* Ditch the frag header. */
7382         mp = mp1->b_cont;
7383 
7384         freeb(mp1);
7385 
7386         /* Restore original IP length in header. */
7387         packet_size = (uint32_t)msgdsize(mp);
7388         if (packet_size > IP_MAXPACKET) {
7389                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7390                 ip_drop_input("Reassembled packet too large", mp, ill);
7391                 freemsg(mp);
7392                 return (NULL);
7393         }
7394 
7395         if (DB_REF(mp) > 1) {
7396                 mblk_t *mp2 = copymsg(mp);
7397 
7398                 if (mp2 == NULL) {
7399                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
7400                         ip_drop_input("ipIfStatsInDiscards", mp, ill);
7401                         freemsg(mp);
7402                         return (NULL);
7403                 }
7404                 freemsg(mp);
7405                 mp = mp2;
7406         }
7407         ipha = (ipha_t *)mp->b_rptr;
7408 
7409         ipha->ipha_length = htons((uint16_t)packet_size);
7410         /* We're now complete, zip the frag state */
7411         ipha->ipha_fragment_offset_and_flags = 0;
7412         /* Record the ECN info. */
7413         ipha->ipha_type_of_service &= 0xFC;
7414         ipha->ipha_type_of_service |= ecn_info;
7415 
7416         /* Update the receive attributes */
7417         ira->ira_pktlen = packet_size;
7418         ira->ira_ip_hdr_length = IPH_HDR_LENGTH(ipha);
7419 
7420         /* Reassembly is successful; set checksum information in packet */
7421         DB_CKSUM16(mp) = (uint16_t)sum_val;
7422         DB_CKSUMFLAGS(mp) = sum_flags;
7423         DB_CKSUMSTART(mp) = ira->ira_ip_hdr_length;
7424 
7425         return (mp);
7426 }
7427 
7428 /*
7429  * Pullup function that should be used for IP input in order to
7430  * ensure we do not loose the L2 source address; we need the l2 source
7431  * address for IP_RECVSLLA and for ndp_input.
7432  *
7433  * We return either NULL or b_rptr.
7434  */
7435 void *
7436 ip_pullup(mblk_t *mp, ssize_t len, ip_recv_attr_t *ira)
7437 {
7438         ill_t           *ill = ira->ira_ill;
7439 
7440         if (ip_rput_pullups++ == 0) {
7441                 (void) mi_strlog(ill->ill_rq, 1, SL_ERROR|SL_TRACE,
7442                     "ip_pullup: %s forced us to "
7443                     " pullup pkt, hdr len %ld, hdr addr %p",
7444                     ill->ill_name, len, (void *)mp->b_rptr);
7445         }
7446         if (!(ira->ira_flags & IRAF_L2SRC_SET))
7447                 ip_setl2src(mp, ira, ira->ira_rill);
7448         ASSERT(ira->ira_flags & IRAF_L2SRC_SET);
7449         if (!pullupmsg(mp, len))
7450                 return (NULL);
7451         else
7452                 return (mp->b_rptr);
7453 }
7454 
7455 /*
7456  * Make sure ira_l2src has an address. If we don't have one fill with zeros.
7457  * When called from the ULP ira_rill will be NULL hence the caller has to
7458  * pass in the ill.
7459  */
7460 /* ARGSUSED */
7461 void
7462 ip_setl2src(mblk_t *mp, ip_recv_attr_t *ira, ill_t *ill)
7463 {
7464         const uchar_t *addr;
7465         int alen;
7466 
7467         if (ira->ira_flags & IRAF_L2SRC_SET)
7468                 return;
7469 
7470         ASSERT(ill != NULL);
7471         alen = ill->ill_phys_addr_length;
7472         ASSERT(alen <= sizeof (ira->ira_l2src));
7473         if (ira->ira_mhip != NULL &&
7474             (addr = ira->ira_mhip->mhi_saddr) != NULL) {
7475                 bcopy(addr, ira->ira_l2src, alen);
7476         } else if ((ira->ira_flags & IRAF_L2SRC_LOOPBACK) &&
7477             (addr = ill->ill_phys_addr) != NULL) {
7478                 bcopy(addr, ira->ira_l2src, alen);
7479         } else {
7480                 bzero(ira->ira_l2src, alen);
7481         }
7482         ira->ira_flags |= IRAF_L2SRC_SET;
7483 }
7484 
7485 /*
7486  * check ip header length and align it.
7487  */
7488 mblk_t *
7489 ip_check_and_align_header(mblk_t *mp, uint_t min_size, ip_recv_attr_t *ira)
7490 {
7491         ill_t   *ill = ira->ira_ill;
7492         ssize_t len;
7493 
7494         len = MBLKL(mp);
7495 
7496         if (!OK_32PTR(mp->b_rptr))
7497                 IP_STAT(ill->ill_ipst, ip_notaligned);
7498         else
7499                 IP_STAT(ill->ill_ipst, ip_recv_pullup);
7500 
7501         /* Guard against bogus device drivers */
7502         if (len < 0) {
7503                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7504                 ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7505                 freemsg(mp);
7506                 return (NULL);
7507         }
7508 
7509         if (len == 0) {
7510                 /* GLD sometimes sends up mblk with b_rptr == b_wptr! */
7511                 mblk_t *mp1 = mp->b_cont;
7512 
7513                 if (!(ira->ira_flags & IRAF_L2SRC_SET))
7514                         ip_setl2src(mp, ira, ira->ira_rill);
7515                 ASSERT(ira->ira_flags & IRAF_L2SRC_SET);
7516 
7517                 freeb(mp);
7518                 mp = mp1;
7519                 if (mp == NULL)
7520                         return (NULL);
7521 
7522                 if (OK_32PTR(mp->b_rptr) && MBLKL(mp) >= min_size)
7523                         return (mp);
7524         }
7525         if (ip_pullup(mp, min_size, ira) == NULL) {
7526                 if (msgdsize(mp) < min_size) {
7527                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7528                         ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7529                 } else {
7530                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
7531                         ip_drop_input("ipIfStatsInDiscards", mp, ill);
7532                 }
7533                 freemsg(mp);
7534                 return (NULL);
7535         }
7536         return (mp);
7537 }
7538 
7539 /*
7540  * Common code for IPv4 and IPv6 to check and pullup multi-mblks
7541  */
7542 mblk_t *
7543 ip_check_length(mblk_t *mp, uchar_t *rptr, ssize_t len, uint_t pkt_len,
7544     uint_t min_size, ip_recv_attr_t *ira)
7545 {
7546         ill_t   *ill = ira->ira_ill;
7547 
7548         /*
7549          * Make sure we have data length consistent
7550          * with the IP header.
7551          */
7552         if (mp->b_cont == NULL) {
7553                 /* pkt_len is based on ipha_len, not the mblk length */
7554                 if (pkt_len < min_size) {
7555                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7556                         ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7557                         freemsg(mp);
7558                         return (NULL);
7559                 }
7560                 if (len < 0) {
7561                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
7562                         ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
7563                         freemsg(mp);
7564                         return (NULL);
7565                 }
7566                 /* Drop any pad */
7567                 mp->b_wptr = rptr + pkt_len;
7568         } else if ((len += msgdsize(mp->b_cont)) != 0) {
7569                 ASSERT(pkt_len >= min_size);
7570                 if (pkt_len < min_size) {
7571                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7572                         ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7573                         freemsg(mp);
7574                         return (NULL);
7575                 }
7576                 if (len < 0) {
7577                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInTruncatedPkts);
7578                         ip_drop_input("ipIfStatsInTruncatedPkts", mp, ill);
7579                         freemsg(mp);
7580                         return (NULL);
7581                 }
7582                 /* Drop any pad */
7583                 (void) adjmsg(mp, -len);
7584                 /*
7585                  * adjmsg may have freed an mblk from the chain, hence
7586                  * invalidate any hw checksum here. This will force IP to
7587                  * calculate the checksum in sw, but only for this packet.
7588                  */
7589                 DB_CKSUMFLAGS(mp) = 0;
7590                 IP_STAT(ill->ill_ipst, ip_multimblk);
7591         }
7592         return (mp);
7593 }
7594 
7595 /*
7596  * Check that the IPv4 opt_len is consistent with the packet and pullup
7597  * the options.
7598  */
7599 mblk_t *
7600 ip_check_optlen(mblk_t *mp, ipha_t *ipha, uint_t opt_len, uint_t pkt_len,
7601     ip_recv_attr_t *ira)
7602 {
7603         ill_t   *ill = ira->ira_ill;
7604         ssize_t len;
7605 
7606         /* Assume no IPv6 packets arrive over the IPv4 queue */
7607         if (IPH_HDR_VERSION(ipha) != IPV4_VERSION) {
7608                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7609                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInWrongIPVersion);
7610                 ip_drop_input("IPvN packet on IPv4 ill", mp, ill);
7611                 freemsg(mp);
7612                 return (NULL);
7613         }
7614 
7615         if (opt_len > (15 - IP_SIMPLE_HDR_LENGTH_IN_WORDS)) {
7616                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7617                 ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7618                 freemsg(mp);
7619                 return (NULL);
7620         }
7621         /*
7622          * Recompute complete header length and make sure we
7623          * have access to all of it.
7624          */
7625         len = ((size_t)opt_len + IP_SIMPLE_HDR_LENGTH_IN_WORDS) << 2;
7626         if (len > (mp->b_wptr - mp->b_rptr)) {
7627                 if (len > pkt_len) {
7628                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInHdrErrors);
7629                         ip_drop_input("ipIfStatsInHdrErrors", mp, ill);
7630                         freemsg(mp);
7631                         return (NULL);
7632                 }
7633                 if (ip_pullup(mp, len, ira) == NULL) {
7634                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
7635                         ip_drop_input("ipIfStatsInDiscards", mp, ill);
7636                         freemsg(mp);
7637                         return (NULL);
7638                 }
7639         }
7640         return (mp);
7641 }
7642 
7643 /*
7644  * Returns a new ire, or the same ire, or NULL.
7645  * If a different IRE is returned, then it is held; the caller
7646  * needs to release it.
7647  * In no case is there any hold/release on the ire argument.
7648  */
7649 ire_t *
7650 ip_check_multihome(void *addr, ire_t *ire, ill_t *ill)
7651 {
7652         ire_t           *new_ire;
7653         ill_t           *ire_ill;
7654         uint_t          ifindex;
7655         ip_stack_t      *ipst = ill->ill_ipst;
7656         boolean_t       strict_check = B_FALSE;
7657 
7658         /*
7659          * IPMP common case: if IRE and ILL are in the same group, there's no
7660          * issue (e.g. packet received on an underlying interface matched an
7661          * IRE_LOCAL on its associated group interface).
7662          */
7663         ASSERT(ire->ire_ill != NULL);
7664         if (IS_IN_SAME_ILLGRP(ill, ire->ire_ill))
7665                 return (ire);
7666 
7667         /*
7668          * Do another ire lookup here, using the ingress ill, to see if the
7669          * interface is in a usesrc group.
7670          * As long as the ills belong to the same group, we don't consider
7671          * them to be arriving on the wrong interface. Thus, if the switch
7672          * is doing inbound load spreading, we won't drop packets when the
7673          * ip*_strict_dst_multihoming switch is on.
7674          * We also need to check for IPIF_UNNUMBERED point2point interfaces
7675          * where the local address may not be unique. In this case we were
7676          * at the mercy of the initial ire lookup and the IRE_LOCAL it
7677          * actually returned. The new lookup, which is more specific, should
7678          * only find the IRE_LOCAL associated with the ingress ill if one
7679          * exists.
7680          */
7681         if (ire->ire_ipversion == IPV4_VERSION) {
7682                 if (ipst->ips_ip_strict_dst_multihoming)
7683                         strict_check = B_TRUE;
7684                 new_ire = ire_ftable_lookup_v4(*((ipaddr_t *)addr), 0, 0,
7685                     IRE_LOCAL, ill, ALL_ZONES, NULL,
7686                     (MATCH_IRE_TYPE|MATCH_IRE_ILL), 0, ipst, NULL);
7687         } else {
7688                 ASSERT(!IN6_IS_ADDR_MULTICAST((in6_addr_t *)addr));
7689                 if (ipst->ips_ipv6_strict_dst_multihoming)
7690                         strict_check = B_TRUE;
7691                 new_ire = ire_ftable_lookup_v6((in6_addr_t *)addr, NULL, NULL,
7692                     IRE_LOCAL, ill, ALL_ZONES, NULL,
7693                     (MATCH_IRE_TYPE|MATCH_IRE_ILL), 0, ipst, NULL);
7694         }
7695         /*
7696          * If the same ire that was returned in ip_input() is found then this
7697          * is an indication that usesrc groups are in use. The packet
7698          * arrived on a different ill in the group than the one associated with
7699          * the destination address.  If a different ire was found then the same
7700          * IP address must be hosted on multiple ills. This is possible with
7701          * unnumbered point2point interfaces. We switch to use this new ire in
7702          * order to have accurate interface statistics.
7703          */
7704         if (new_ire != NULL) {
7705                 /* Note: held in one case but not the other? Caller handles */
7706                 if (new_ire != ire)
7707                         return (new_ire);
7708                 /* Unchanged */
7709                 ire_refrele(new_ire);
7710                 return (ire);
7711         }
7712 
7713         /*
7714          * Chase pointers once and store locally.
7715          */
7716         ASSERT(ire->ire_ill != NULL);
7717         ire_ill = ire->ire_ill;
7718         ifindex = ill->ill_usesrc_ifindex;
7719 
7720         /*
7721          * Check if it's a legal address on the 'usesrc' interface.
7722          * For IPMP data addresses the IRE_LOCAL is the upper, hence we
7723          * can just check phyint_ifindex.
7724          */
7725         if (ifindex != 0 && ifindex == ire_ill->ill_phyint->phyint_ifindex) {
7726                 return (ire);
7727         }
7728 
7729         /*
7730          * If the ip*_strict_dst_multihoming switch is on then we can
7731          * only accept this packet if the interface is marked as routing.
7732          */
7733         if (!(strict_check))
7734                 return (ire);
7735 
7736         if ((ill->ill_flags & ire->ire_ill->ill_flags & ILLF_ROUTER) != 0) {
7737                 return (ire);
7738         }
7739         return (NULL);
7740 }
7741 
7742 /*
7743  * This function is used to construct a mac_header_info_s from a
7744  * DL_UNITDATA_IND message.
7745  * The address fields in the mhi structure points into the message,
7746  * thus the caller can't use those fields after freeing the message.
7747  *
7748  * We determine whether the packet received is a non-unicast packet
7749  * and in doing so, determine whether or not it is broadcast vs multicast.
7750  * For it to be a broadcast packet, we must have the appropriate mblk_t
7751  * hanging off the ill_t.  If this is either not present or doesn't match
7752  * the destination mac address in the DL_UNITDATA_IND, the packet is deemed
7753  * to be multicast.  Thus NICs that have no broadcast address (or no
7754  * capability for one, such as point to point links) cannot return as
7755  * the packet being broadcast.
7756  */
7757 void
7758 ip_dlur_to_mhi(ill_t *ill, mblk_t *mb, struct mac_header_info_s *mhip)
7759 {
7760         dl_unitdata_ind_t *ind = (dl_unitdata_ind_t *)mb->b_rptr;
7761         mblk_t *bmp;
7762         uint_t extra_offset;
7763 
7764         bzero(mhip, sizeof (struct mac_header_info_s));
7765 
7766         mhip->mhi_dsttype = MAC_ADDRTYPE_UNICAST;
7767 
7768         if (ill->ill_sap_length < 0)
7769                 extra_offset = 0;
7770         else
7771                 extra_offset = ill->ill_sap_length;
7772 
7773         mhip->mhi_daddr = (uchar_t *)ind + ind->dl_dest_addr_offset +
7774             extra_offset;
7775         mhip->mhi_saddr = (uchar_t *)ind + ind->dl_src_addr_offset +
7776             extra_offset;
7777 
7778         if (!ind->dl_group_address)
7779                 return;
7780 
7781         /* Multicast or broadcast */
7782         mhip->mhi_dsttype = MAC_ADDRTYPE_MULTICAST;
7783 
7784         if (ind->dl_dest_addr_offset > sizeof (*ind) &&
7785             ind->dl_dest_addr_offset + ind->dl_dest_addr_length < MBLKL(mb) &&
7786             (bmp = ill->ill_bcast_mp) != NULL) {
7787                 dl_unitdata_req_t *dlur;
7788                 uint8_t *bphys_addr;
7789 
7790                 dlur = (dl_unitdata_req_t *)bmp->b_rptr;
7791                 bphys_addr = (uchar_t *)dlur + dlur->dl_dest_addr_offset +
7792                     extra_offset;
7793 
7794                 if (bcmp(mhip->mhi_daddr, bphys_addr,
7795                     ind->dl_dest_addr_length) == 0)
7796                         mhip->mhi_dsttype = MAC_ADDRTYPE_BROADCAST;
7797         }
7798 }
7799 
7800 /*
7801  * This function is used to construct a mac_header_info_s from a
7802  * M_DATA fastpath message from a DLPI driver.
7803  * The address fields in the mhi structure points into the message,
7804  * thus the caller can't use those fields after freeing the message.
7805  *
7806  * We determine whether the packet received is a non-unicast packet
7807  * and in doing so, determine whether or not it is broadcast vs multicast.
7808  * For it to be a broadcast packet, we must have the appropriate mblk_t
7809  * hanging off the ill_t.  If this is either not present or doesn't match
7810  * the destination mac address in the DL_UNITDATA_IND, the packet is deemed
7811  * to be multicast.  Thus NICs that have no broadcast address (or no
7812  * capability for one, such as point to point links) cannot return as
7813  * the packet being broadcast.
7814  */
7815 void
7816 ip_mdata_to_mhi(ill_t *ill, mblk_t *mp, struct mac_header_info_s *mhip)
7817 {
7818         mblk_t *bmp;
7819         struct ether_header *pether;
7820 
7821         bzero(mhip, sizeof (struct mac_header_info_s));
7822 
7823         mhip->mhi_dsttype = MAC_ADDRTYPE_UNICAST;
7824 
7825         pether = (struct ether_header *)((char *)mp->b_rptr
7826             - sizeof (struct ether_header));
7827 
7828         /*
7829          * Make sure the interface is an ethernet type, since we don't
7830          * know the header format for anything but Ethernet. Also make
7831          * sure we are pointing correctly above db_base.
7832          */
7833         if (ill->ill_type != IFT_ETHER)
7834                 return;
7835 
7836 retry:
7837         if ((uchar_t *)pether < mp->b_datap->db_base)
7838                 return;
7839 
7840         /* Is there a VLAN tag? */
7841         if (ill->ill_isv6) {
7842                 if (pether->ether_type != htons(ETHERTYPE_IPV6)) {
7843                         pether = (struct ether_header *)((char *)pether - 4);
7844                         goto retry;
7845                 }
7846         } else {
7847                 if (pether->ether_type != htons(ETHERTYPE_IP)) {
7848                         pether = (struct ether_header *)((char *)pether - 4);
7849                         goto retry;
7850                 }
7851         }
7852         mhip->mhi_daddr = (uchar_t *)&pether->ether_dhost;
7853         mhip->mhi_saddr = (uchar_t *)&pether->ether_shost;
7854 
7855         if (!(mhip->mhi_daddr[0] & 0x01))
7856                 return;
7857 
7858         /* Multicast or broadcast */
7859         mhip->mhi_dsttype = MAC_ADDRTYPE_MULTICAST;
7860 
7861         if ((bmp = ill->ill_bcast_mp) != NULL) {
7862                 dl_unitdata_req_t *dlur;
7863                 uint8_t *bphys_addr;
7864                 uint_t  addrlen;
7865 
7866                 dlur = (dl_unitdata_req_t *)bmp->b_rptr;
7867                 addrlen = dlur->dl_dest_addr_length;
7868                 if (ill->ill_sap_length < 0) {
7869                         bphys_addr = (uchar_t *)dlur +
7870                             dlur->dl_dest_addr_offset;
7871                         addrlen += ill->ill_sap_length;
7872                 } else {
7873                         bphys_addr = (uchar_t *)dlur +
7874                             dlur->dl_dest_addr_offset +
7875                             ill->ill_sap_length;
7876                         addrlen -= ill->ill_sap_length;
7877                 }
7878                 if (bcmp(mhip->mhi_daddr, bphys_addr, addrlen) == 0)
7879                         mhip->mhi_dsttype = MAC_ADDRTYPE_BROADCAST;
7880         }
7881 }
7882 
7883 /*
7884  * Handle anything but M_DATA messages
7885  * We see the DL_UNITDATA_IND which are part
7886  * of the data path, and also the other messages from the driver.
7887  */
7888 void
7889 ip_rput_notdata(ill_t *ill, mblk_t *mp)
7890 {
7891         mblk_t          *first_mp;
7892         struct iocblk   *iocp;
7893         struct mac_header_info_s mhi;
7894 
7895         switch (DB_TYPE(mp)) {
7896         case M_PROTO:
7897         case M_PCPROTO: {
7898                 if (((dl_unitdata_ind_t *)mp->b_rptr)->dl_primitive !=
7899                     DL_UNITDATA_IND) {
7900                         /* Go handle anything other than data elsewhere. */
7901                         ip_rput_dlpi(ill, mp);
7902                         return;
7903                 }
7904 
7905                 first_mp = mp;
7906                 mp = first_mp->b_cont;
7907                 first_mp->b_cont = NULL;
7908 
7909                 if (mp == NULL) {
7910                         freeb(first_mp);
7911                         return;
7912                 }
7913                 ip_dlur_to_mhi(ill, first_mp, &mhi);
7914                 if (ill->ill_isv6)
7915                         ip_input_v6(ill, NULL, mp, &mhi);
7916                 else
7917                         ip_input(ill, NULL, mp, &mhi);
7918 
7919                 /* Ditch the DLPI header. */
7920                 freeb(first_mp);
7921                 return;
7922         }
7923         case M_IOCACK:
7924                 iocp = (struct iocblk *)mp->b_rptr;
7925                 switch (iocp->ioc_cmd) {
7926                 case DL_IOC_HDR_INFO:
7927                         ill_fastpath_ack(ill, mp);
7928                         return;
7929                 default:
7930                         putnext(ill->ill_rq, mp);
7931                         return;
7932                 }
7933                 /* FALLTHRU */
7934         case M_ERROR:
7935         case M_HANGUP:
7936                 mutex_enter(&ill->ill_lock);
7937                 if (ill->ill_state_flags & ILL_CONDEMNED) {
7938                         mutex_exit(&ill->ill_lock);
7939                         freemsg(mp);
7940                         return;
7941                 }
7942                 ill_refhold_locked(ill);
7943                 mutex_exit(&ill->ill_lock);
7944                 qwriter_ip(ill, ill->ill_rq, mp, ip_rput_other, CUR_OP,
7945                     B_FALSE);
7946                 return;
7947         case M_CTL:
7948                 putnext(ill->ill_rq, mp);
7949                 return;
7950         case M_IOCNAK:
7951                 ip1dbg(("got iocnak "));
7952                 iocp = (struct iocblk *)mp->b_rptr;
7953                 switch (iocp->ioc_cmd) {
7954                 case DL_IOC_HDR_INFO:
7955                         ip_rput_other(NULL, ill->ill_rq, mp, NULL);
7956                         return;
7957                 default:
7958                         break;
7959                 }
7960                 /* FALLTHRU */
7961         default:
7962                 putnext(ill->ill_rq, mp);
7963                 return;
7964         }
7965 }
7966 
7967 /* Read side put procedure.  Packets coming from the wire arrive here. */
7968 void
7969 ip_rput(queue_t *q, mblk_t *mp)
7970 {
7971         ill_t   *ill;
7972         union DL_primitives *dl;
7973 
7974         ill = (ill_t *)q->q_ptr;
7975 
7976         if (ill->ill_state_flags & (ILL_CONDEMNED | ILL_LL_SUBNET_PENDING)) {
7977                 /*
7978                  * If things are opening or closing, only accept high-priority
7979                  * DLPI messages.  (On open ill->ill_ipif has not yet been
7980                  * created; on close, things hanging off the ill may have been
7981                  * freed already.)
7982                  */
7983                 dl = (union DL_primitives *)mp->b_rptr;
7984                 if (DB_TYPE(mp) != M_PCPROTO ||
7985                     dl->dl_primitive == DL_UNITDATA_IND) {
7986                         inet_freemsg(mp);
7987                         return;
7988                 }
7989         }
7990         if (DB_TYPE(mp) == M_DATA) {
7991                 struct mac_header_info_s mhi;
7992 
7993                 ip_mdata_to_mhi(ill, mp, &mhi);
7994                 ip_input(ill, NULL, mp, &mhi);
7995         } else {
7996                 ip_rput_notdata(ill, mp);
7997         }
7998 }
7999 
8000 /*
8001  * Move the information to a copy.
8002  */
8003 mblk_t *
8004 ip_fix_dbref(mblk_t *mp, ip_recv_attr_t *ira)
8005 {
8006         mblk_t          *mp1;
8007         ill_t           *ill = ira->ira_ill;
8008         ip_stack_t      *ipst = ill->ill_ipst;
8009 
8010         IP_STAT(ipst, ip_db_ref);
8011 
8012         /* Make sure we have ira_l2src before we loose the original mblk */
8013         if (!(ira->ira_flags & IRAF_L2SRC_SET))
8014                 ip_setl2src(mp, ira, ira->ira_rill);
8015 
8016         mp1 = copymsg(mp);
8017         if (mp1 == NULL) {
8018                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
8019                 ip_drop_input("ipIfStatsInDiscards", mp, ill);
8020                 freemsg(mp);
8021                 return (NULL);
8022         }
8023         /* preserve the hardware checksum flags and data, if present */
8024         if (DB_CKSUMFLAGS(mp) != 0) {
8025                 DB_CKSUMFLAGS(mp1) = DB_CKSUMFLAGS(mp);
8026                 DB_CKSUMSTART(mp1) = DB_CKSUMSTART(mp);
8027                 DB_CKSUMSTUFF(mp1) = DB_CKSUMSTUFF(mp);
8028                 DB_CKSUMEND(mp1) = DB_CKSUMEND(mp);
8029                 DB_CKSUM16(mp1) = DB_CKSUM16(mp);
8030         }
8031         freemsg(mp);
8032         return (mp1);
8033 }
8034 
8035 static void
8036 ip_dlpi_error(ill_t *ill, t_uscalar_t prim, t_uscalar_t dl_err,
8037     t_uscalar_t err)
8038 {
8039         if (dl_err == DL_SYSERR) {
8040                 (void) mi_strlog(ill->ill_rq, 1, SL_CONSOLE|SL_ERROR|SL_TRACE,
8041                     "%s: %s failed: DL_SYSERR (errno %u)\n",
8042                     ill->ill_name, dl_primstr(prim), err);
8043                 return;
8044         }
8045 
8046         (void) mi_strlog(ill->ill_rq, 1, SL_CONSOLE|SL_ERROR|SL_TRACE,
8047             "%s: %s failed: %s\n", ill->ill_name, dl_primstr(prim),
8048             dl_errstr(dl_err));
8049 }
8050 
8051 /*
8052  * ip_rput_dlpi is called by ip_rput to handle all DLPI messages other
8053  * than DL_UNITDATA_IND messages. If we need to process this message
8054  * exclusively, we call qwriter_ip, in which case we also need to call
8055  * ill_refhold before that, since qwriter_ip does an ill_refrele.
8056  */
8057 void
8058 ip_rput_dlpi(ill_t *ill, mblk_t *mp)
8059 {
8060         dl_ok_ack_t     *dloa = (dl_ok_ack_t *)mp->b_rptr;
8061         dl_error_ack_t  *dlea = (dl_error_ack_t *)dloa;
8062         queue_t         *q = ill->ill_rq;
8063         t_uscalar_t     prim = dloa->dl_primitive;
8064         t_uscalar_t     reqprim = DL_PRIM_INVAL;
8065 
8066         DTRACE_PROBE3(ill__dlpi, char *, "ip_rput_dlpi",
8067             char *, dl_primstr(prim), ill_t *, ill);
8068         ip1dbg(("ip_rput_dlpi"));
8069 
8070         /*
8071          * If we received an ACK but didn't send a request for it, then it
8072          * can't be part of any pending operation; discard up-front.
8073          */
8074         switch (prim) {
8075         case DL_ERROR_ACK:
8076                 reqprim = dlea->dl_error_primitive;
8077                 ip2dbg(("ip_rput_dlpi(%s): DL_ERROR_ACK for %s (0x%x): %s "
8078                     "(0x%x), unix %u\n", ill->ill_name, dl_primstr(reqprim),
8079                     reqprim, dl_errstr(dlea->dl_errno), dlea->dl_errno,
8080                     dlea->dl_unix_errno));
8081                 break;
8082         case DL_OK_ACK:
8083                 reqprim = dloa->dl_correct_primitive;
8084                 break;
8085         case DL_INFO_ACK:
8086                 reqprim = DL_INFO_REQ;
8087                 break;
8088         case DL_BIND_ACK:
8089                 reqprim = DL_BIND_REQ;
8090                 break;
8091         case DL_PHYS_ADDR_ACK:
8092                 reqprim = DL_PHYS_ADDR_REQ;
8093                 break;
8094         case DL_NOTIFY_ACK:
8095                 reqprim = DL_NOTIFY_REQ;
8096                 break;
8097         case DL_CAPABILITY_ACK:
8098                 reqprim = DL_CAPABILITY_REQ;
8099                 break;
8100         }
8101 
8102         if (prim != DL_NOTIFY_IND) {
8103                 if (reqprim == DL_PRIM_INVAL ||
8104                     !ill_dlpi_pending(ill, reqprim)) {
8105                         /* Not a DLPI message we support or expected */
8106                         freemsg(mp);
8107                         return;
8108                 }
8109                 ip1dbg(("ip_rput: received %s for %s\n", dl_primstr(prim),
8110                     dl_primstr(reqprim)));
8111         }
8112 
8113         switch (reqprim) {
8114         case DL_UNBIND_REQ:
8115                 /*
8116                  * NOTE: we mark the unbind as complete even if we got a
8117                  * DL_ERROR_ACK, since there's not much else we can do.
8118                  */
8119                 mutex_enter(&ill->ill_lock);
8120                 ill->ill_state_flags &= ~ILL_DL_UNBIND_IN_PROGRESS;
8121                 cv_signal(&ill->ill_cv);
8122                 mutex_exit(&ill->ill_lock);
8123                 break;
8124 
8125         case DL_ENABMULTI_REQ:
8126                 if (prim == DL_OK_ACK) {
8127                         if (ill->ill_dlpi_multicast_state == IDS_INPROGRESS)
8128                                 ill->ill_dlpi_multicast_state = IDS_OK;
8129                 }
8130                 break;
8131         }
8132 
8133         /*
8134          * The message is one we're waiting for (or DL_NOTIFY_IND), but we
8135          * need to become writer to continue to process it.  Because an
8136          * exclusive operation doesn't complete until replies to all queued
8137          * DLPI messages have been received, we know we're in the middle of an
8138          * exclusive operation and pass CUR_OP (except for DL_NOTIFY_IND).
8139          *
8140          * As required by qwriter_ip(), we refhold the ill; it will refrele.
8141          * Since this is on the ill stream we unconditionally bump up the
8142          * refcount without doing ILL_CAN_LOOKUP().
8143          */
8144         ill_refhold(ill);
8145         if (prim == DL_NOTIFY_IND)
8146                 qwriter_ip(ill, q, mp, ip_rput_dlpi_writer, NEW_OP, B_FALSE);
8147         else
8148                 qwriter_ip(ill, q, mp, ip_rput_dlpi_writer, CUR_OP, B_FALSE);
8149 }
8150 
8151 /*
8152  * Handling of DLPI messages that require exclusive access to the ipsq.
8153  *
8154  * Need to do ipsq_pending_mp_get on ioctl completion, which could
8155  * happen here. (along with mi_copy_done)
8156  */
8157 /* ARGSUSED */
8158 static void
8159 ip_rput_dlpi_writer(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *dummy_arg)
8160 {
8161         dl_ok_ack_t     *dloa = (dl_ok_ack_t *)mp->b_rptr;
8162         dl_error_ack_t  *dlea = (dl_error_ack_t *)dloa;
8163         int             err = 0;
8164         ill_t           *ill = (ill_t *)q->q_ptr;
8165         ipif_t          *ipif = NULL;
8166         mblk_t          *mp1 = NULL;
8167         conn_t          *connp = NULL;
8168         t_uscalar_t     paddrreq;
8169         mblk_t          *mp_hw;
8170         boolean_t       success;
8171         boolean_t       ioctl_aborted = B_FALSE;
8172         boolean_t       log = B_TRUE;
8173 
8174         DTRACE_PROBE3(ill__dlpi, char *, "ip_rput_dlpi_writer",
8175             char *, dl_primstr(dloa->dl_primitive), ill_t *, ill);
8176 
8177         ip1dbg(("ip_rput_dlpi_writer .."));
8178         ASSERT(ipsq->ipsq_xop == ill->ill_phyint->phyint_ipsq->ipsq_xop);
8179         ASSERT(IAM_WRITER_ILL(ill));
8180 
8181         ipif = ipsq->ipsq_xop->ipx_pending_ipif;
8182         /*
8183          * The current ioctl could have been aborted by the user and a new
8184          * ioctl to bring up another ill could have started. We could still
8185          * get a response from the driver later.
8186          */
8187         if (ipif != NULL && ipif->ipif_ill != ill)
8188                 ioctl_aborted = B_TRUE;
8189 
8190         switch (dloa->dl_primitive) {
8191         case DL_ERROR_ACK:
8192                 ip1dbg(("ip_rput_dlpi_writer: got DL_ERROR_ACK for %s\n",
8193                     dl_primstr(dlea->dl_error_primitive)));
8194 
8195                 DTRACE_PROBE3(ill__dlpi, char *, "ip_rput_dlpi_writer error",
8196                     char *, dl_primstr(dlea->dl_error_primitive),
8197                     ill_t *, ill);
8198 
8199                 switch (dlea->dl_error_primitive) {
8200                 case DL_DISABMULTI_REQ:
8201                         ill_dlpi_done(ill, dlea->dl_error_primitive);
8202                         break;
8203                 case DL_PROMISCON_REQ:
8204                 case DL_PROMISCOFF_REQ:
8205                 case DL_UNBIND_REQ:
8206                 case DL_ATTACH_REQ:
8207                 case DL_INFO_REQ:
8208                         ill_dlpi_done(ill, dlea->dl_error_primitive);
8209                         break;
8210                 case DL_NOTIFY_REQ:
8211                         ill_dlpi_done(ill, DL_NOTIFY_REQ);
8212                         log = B_FALSE;
8213                         break;
8214                 case DL_PHYS_ADDR_REQ:
8215                         /*
8216                          * For IPv6 only, there are two additional
8217                          * phys_addr_req's sent to the driver to get the
8218                          * IPv6 token and lla. This allows IP to acquire
8219                          * the hardware address format for a given interface
8220                          * without having built in knowledge of the hardware
8221                          * address. ill_phys_addr_pend keeps track of the last
8222                          * DL_PAR sent so we know which response we are
8223                          * dealing with. ill_dlpi_done will update
8224                          * ill_phys_addr_pend when it sends the next req.
8225                          * We don't complete the IOCTL until all three DL_PARs
8226                          * have been attempted, so set *_len to 0 and break.
8227                          */
8228                         paddrreq = ill->ill_phys_addr_pend;
8229                         ill_dlpi_done(ill, DL_PHYS_ADDR_REQ);
8230                         if (paddrreq == DL_IPV6_TOKEN) {
8231                                 ill->ill_token_length = 0;
8232                                 log = B_FALSE;
8233                                 break;
8234                         } else if (paddrreq == DL_IPV6_LINK_LAYER_ADDR) {
8235                                 ill->ill_nd_lla_len = 0;
8236                                 log = B_FALSE;
8237                                 break;
8238                         }
8239                         /*
8240                          * Something went wrong with the DL_PHYS_ADDR_REQ.
8241                          * We presumably have an IOCTL hanging out waiting
8242                          * for completion. Find it and complete the IOCTL
8243                          * with the error noted.
8244                          * However, ill_dl_phys was called on an ill queue
8245                          * (from SIOCSLIFNAME), thus conn_pending_ill is not
8246                          * set. But the ioctl is known to be pending on ill_wq.
8247                          */
8248                         if (!ill->ill_ifname_pending)
8249                                 break;
8250                         ill->ill_ifname_pending = 0;
8251                         if (!ioctl_aborted)
8252                                 mp1 = ipsq_pending_mp_get(ipsq, &connp);
8253                         if (mp1 != NULL) {
8254                                 /*
8255                                  * This operation (SIOCSLIFNAME) must have
8256                                  * happened on the ill. Assert there is no conn
8257                                  */
8258                                 ASSERT(connp == NULL);
8259                                 q = ill->ill_wq;
8260                         }
8261                         break;
8262                 case DL_BIND_REQ:
8263                         ill_dlpi_done(ill, DL_BIND_REQ);
8264                         if (ill->ill_ifname_pending)
8265                                 break;
8266                         mutex_enter(&ill->ill_lock);
8267                         ill->ill_state_flags &= ~ILL_DOWN_IN_PROGRESS;
8268                         mutex_exit(&ill->ill_lock);
8269                         /*
8270                          * Something went wrong with the bind.  We presumably
8271                          * have an IOCTL hanging out waiting for completion.
8272                          * Find it, take down the interface that was coming
8273                          * up, and complete the IOCTL with the error noted.
8274                          */
8275                         if (!ioctl_aborted)
8276                                 mp1 = ipsq_pending_mp_get(ipsq, &connp);
8277                         if (mp1 != NULL) {
8278                                 /*
8279                                  * This might be a result of a DL_NOTE_REPLUMB
8280                                  * notification. In that case, connp is NULL.
8281                                  */
8282                                 if (connp != NULL)
8283                                         q = CONNP_TO_WQ(connp);
8284 
8285                                 (void) ipif_down(ipif, NULL, NULL);
8286                                 /* error is set below the switch */
8287                         }
8288                         break;
8289                 case DL_ENABMULTI_REQ:
8290                         ill_dlpi_done(ill, DL_ENABMULTI_REQ);
8291 
8292                         if (ill->ill_dlpi_multicast_state == IDS_INPROGRESS)
8293                                 ill->ill_dlpi_multicast_state = IDS_FAILED;
8294                         if (ill->ill_dlpi_multicast_state == IDS_FAILED) {
8295 
8296                                 printf("ip: joining multicasts failed (%d)"
8297                                     " on %s - will use link layer "
8298                                     "broadcasts for multicast\n",
8299                                     dlea->dl_errno, ill->ill_name);
8300 
8301                                 /*
8302                                  * Set up for multi_bcast; We are the
8303                                  * writer, so ok to access ill->ill_ipif
8304                                  * without any lock.
8305                                  */
8306                                 mutex_enter(&ill->ill_phyint->phyint_lock);
8307                                 ill->ill_phyint->phyint_flags |=
8308                                     PHYI_MULTI_BCAST;
8309                                 mutex_exit(&ill->ill_phyint->phyint_lock);
8310 
8311                         }
8312                         freemsg(mp);    /* Don't want to pass this up */
8313                         return;
8314                 case DL_CAPABILITY_REQ:
8315                         ip1dbg(("ip_rput_dlpi_writer: got DL_ERROR_ACK for "
8316                             "DL_CAPABILITY REQ\n"));
8317                         if (ill->ill_dlpi_capab_state == IDCS_PROBE_SENT)
8318                                 ill->ill_dlpi_capab_state = IDCS_FAILED;
8319                         ill_capability_done(ill);
8320                         freemsg(mp);
8321                         return;
8322                 }
8323                 /*
8324                  * Note the error for IOCTL completion (mp1 is set when
8325                  * ready to complete ioctl). If ill_ifname_pending_err is
8326                  * set, an error occured during plumbing (ill_ifname_pending),
8327                  * so we want to report that error.
8328                  *
8329                  * NOTE: there are two addtional DL_PHYS_ADDR_REQ's
8330                  * (DL_IPV6_TOKEN and DL_IPV6_LINK_LAYER_ADDR) that are
8331                  * expected to get errack'd if the driver doesn't support
8332                  * these flags (e.g. ethernet). log will be set to B_FALSE
8333                  * if these error conditions are encountered.
8334                  */
8335                 if (mp1 != NULL) {
8336                         if (ill->ill_ifname_pending_err != 0)  {
8337                                 err = ill->ill_ifname_pending_err;
8338                                 ill->ill_ifname_pending_err = 0;
8339                         } else {
8340                                 err = dlea->dl_unix_errno ?
8341                                     dlea->dl_unix_errno : ENXIO;
8342                         }
8343                 /*
8344                  * If we're plumbing an interface and an error hasn't already
8345                  * been saved, set ill_ifname_pending_err to the error passed
8346                  * up. Ignore the error if log is B_FALSE (see comment above).
8347                  */
8348                 } else if (log && ill->ill_ifname_pending &&
8349                     ill->ill_ifname_pending_err == 0) {
8350                         ill->ill_ifname_pending_err = dlea->dl_unix_errno ?
8351                             dlea->dl_unix_errno : ENXIO;
8352                 }
8353 
8354                 if (log)
8355                         ip_dlpi_error(ill, dlea->dl_error_primitive,
8356                             dlea->dl_errno, dlea->dl_unix_errno);
8357                 break;
8358         case DL_CAPABILITY_ACK:
8359                 ill_capability_ack(ill, mp);
8360                 /*
8361                  * The message has been handed off to ill_capability_ack
8362                  * and must not be freed below
8363                  */
8364                 mp = NULL;
8365                 break;
8366 
8367         case DL_INFO_ACK:
8368                 /* Call a routine to handle this one. */
8369                 ill_dlpi_done(ill, DL_INFO_REQ);
8370                 ip_ll_subnet_defaults(ill, mp);
8371                 ASSERT(!MUTEX_HELD(&ill->ill_phyint->phyint_ipsq->ipsq_lock));
8372                 return;
8373         case DL_BIND_ACK:
8374                 /*
8375                  * We should have an IOCTL waiting on this unless
8376                  * sent by ill_dl_phys, in which case just return
8377                  */
8378                 ill_dlpi_done(ill, DL_BIND_REQ);
8379 
8380                 if (ill->ill_ifname_pending) {
8381                         DTRACE_PROBE2(ip__rput__dlpi__ifname__pending,
8382                             ill_t *, ill, mblk_t *, mp);
8383                         break;
8384                 }
8385                 mutex_enter(&ill->ill_lock);
8386                 ill->ill_dl_up = 1;
8387                 ill->ill_state_flags &= ~ILL_DOWN_IN_PROGRESS;
8388                 mutex_exit(&ill->ill_lock);
8389 
8390                 if (!ioctl_aborted)
8391                         mp1 = ipsq_pending_mp_get(ipsq, &connp);
8392                 if (mp1 == NULL) {
8393                         DTRACE_PROBE1(ip__rput__dlpi__no__mblk, ill_t *, ill);
8394                         break;
8395                 }
8396                 /*
8397                  * mp1 was added by ill_dl_up(). if that is a result of
8398                  * a DL_NOTE_REPLUMB notification, connp could be NULL.
8399                  */
8400                 if (connp != NULL)
8401                         q = CONNP_TO_WQ(connp);
8402                 /*
8403                  * We are exclusive. So nothing can change even after
8404                  * we get the pending mp.
8405                  */
8406                 ip1dbg(("ip_rput_dlpi: bind_ack %s\n", ill->ill_name));
8407                 DTRACE_PROBE1(ip__rput__dlpi__bind__ack, ill_t *, ill);
8408                 ill_nic_event_dispatch(ill, 0, NE_UP, NULL, 0);
8409 
8410                 /*
8411                  * Now bring up the resolver; when that is complete, we'll
8412                  * create IREs.  Note that we intentionally mirror what
8413                  * ipif_up() would have done, because we got here by way of
8414                  * ill_dl_up(), which stopped ipif_up()'s processing.
8415                  */
8416                 if (ill->ill_isv6) {
8417                         /*
8418                          * v6 interfaces.
8419                          * Unlike ARP which has to do another bind
8420                          * and attach, once we get here we are
8421                          * done with NDP
8422                          */
8423                         (void) ipif_resolver_up(ipif, Res_act_initial);
8424                         if ((err = ipif_ndp_up(ipif, B_TRUE)) == 0)
8425                                 err = ipif_up_done_v6(ipif);
8426                 } else if (ill->ill_net_type == IRE_IF_RESOLVER) {
8427                         /*
8428                          * ARP and other v4 external resolvers.
8429                          * Leave the pending mblk intact so that
8430                          * the ioctl completes in ip_rput().
8431                          */
8432                         if (connp != NULL)
8433                                 mutex_enter(&connp->conn_lock);
8434                         mutex_enter(&ill->ill_lock);
8435                         success = ipsq_pending_mp_add(connp, ipif, q, mp1, 0);
8436                         mutex_exit(&ill->ill_lock);
8437                         if (connp != NULL)
8438                                 mutex_exit(&connp->conn_lock);
8439                         if (success) {
8440                                 err = ipif_resolver_up(ipif, Res_act_initial);
8441                                 if (err == EINPROGRESS) {
8442                                         freemsg(mp);
8443                                         return;
8444                                 }
8445                                 mp1 = ipsq_pending_mp_get(ipsq, &connp);
8446                         } else {
8447                                 /* The conn has started closing */
8448                                 err = EINTR;
8449                         }
8450                 } else {
8451                         /*
8452                          * This one is complete. Reply to pending ioctl.
8453                          */
8454                         (void) ipif_resolver_up(ipif, Res_act_initial);
8455                         err = ipif_up_done(ipif);
8456                 }
8457 
8458                 if ((err == 0) && (ill->ill_up_ipifs)) {
8459                         err = ill_up_ipifs(ill, q, mp1);
8460                         if (err == EINPROGRESS) {
8461                                 freemsg(mp);
8462                                 return;
8463                         }
8464                 }
8465 
8466                 /*
8467                  * If we have a moved ipif to bring up, and everything has
8468                  * succeeded to this point, bring it up on the IPMP ill.
8469                  * Otherwise, leave it down -- the admin can try to bring it
8470                  * up by hand if need be.
8471                  */
8472                 if (ill->ill_move_ipif != NULL) {
8473                         if (err != 0) {
8474                                 ill->ill_move_ipif = NULL;
8475                         } else {
8476                                 ipif = ill->ill_move_ipif;
8477                                 ill->ill_move_ipif = NULL;
8478                                 err = ipif_up(ipif, q, mp1);
8479                                 if (err == EINPROGRESS) {
8480                                         freemsg(mp);
8481                                         return;
8482                                 }
8483                         }
8484                 }
8485                 break;
8486 
8487         case DL_NOTIFY_IND: {
8488                 dl_notify_ind_t *notify = (dl_notify_ind_t *)mp->b_rptr;
8489                 uint_t orig_mtu, orig_mc_mtu;
8490 
8491                 switch (notify->dl_notification) {
8492                 case DL_NOTE_PHYS_ADDR:
8493                         err = ill_set_phys_addr(ill, mp);
8494                         break;
8495 
8496                 case DL_NOTE_REPLUMB:
8497                         /*
8498                          * Directly return after calling ill_replumb().
8499                          * Note that we should not free mp as it is reused
8500                          * in the ill_replumb() function.
8501                          */
8502                         err = ill_replumb(ill, mp);
8503                         return;
8504 
8505                 case DL_NOTE_FASTPATH_FLUSH:
8506                         nce_flush(ill, B_FALSE);
8507                         break;
8508 
8509                 case DL_NOTE_SDU_SIZE:
8510                 case DL_NOTE_SDU_SIZE2:
8511                         /*
8512                          * The dce and fragmentation code can cope with
8513                          * this changing while packets are being sent.
8514                          * When packets are sent ip_output will discover
8515                          * a change.
8516                          *
8517                          * Change the MTU size of the interface.
8518                          */
8519                         mutex_enter(&ill->ill_lock);
8520                         orig_mtu = ill->ill_mtu;
8521                         orig_mc_mtu = ill->ill_mc_mtu;
8522                         switch (notify->dl_notification) {
8523                         case DL_NOTE_SDU_SIZE:
8524                                 ill->ill_current_frag =
8525                                     (uint_t)notify->dl_data;
8526                                 ill->ill_mc_mtu = (uint_t)notify->dl_data;
8527                                 break;
8528                         case DL_NOTE_SDU_SIZE2:
8529                                 ill->ill_current_frag =
8530                                     (uint_t)notify->dl_data1;
8531                                 ill->ill_mc_mtu = (uint_t)notify->dl_data2;
8532                                 break;
8533                         }
8534                         if (ill->ill_current_frag > ill->ill_max_frag)
8535                                 ill->ill_max_frag = ill->ill_current_frag;
8536 
8537                         if (!(ill->ill_flags & ILLF_FIXEDMTU)) {
8538                                 ill->ill_mtu = ill->ill_current_frag;
8539 
8540                                 /*
8541                                  * If ill_user_mtu was set (via
8542                                  * SIOCSLIFLNKINFO), clamp ill_mtu at it.
8543                                  */
8544                                 if (ill->ill_user_mtu != 0 &&
8545                                     ill->ill_user_mtu < ill->ill_mtu)
8546                                         ill->ill_mtu = ill->ill_user_mtu;
8547 
8548                                 if (ill->ill_user_mtu != 0 &&
8549                                     ill->ill_user_mtu < ill->ill_mc_mtu)
8550                                         ill->ill_mc_mtu = ill->ill_user_mtu;
8551 
8552                                 if (ill->ill_isv6) {
8553                                         if (ill->ill_mtu < IPV6_MIN_MTU)
8554                                                 ill->ill_mtu = IPV6_MIN_MTU;
8555                                         if (ill->ill_mc_mtu < IPV6_MIN_MTU)
8556                                                 ill->ill_mc_mtu = IPV6_MIN_MTU;
8557                                 } else {
8558                                         if (ill->ill_mtu < IP_MIN_MTU)
8559                                                 ill->ill_mtu = IP_MIN_MTU;
8560                                         if (ill->ill_mc_mtu < IP_MIN_MTU)
8561                                                 ill->ill_mc_mtu = IP_MIN_MTU;
8562                                 }
8563                         } else if (ill->ill_mc_mtu > ill->ill_mtu) {
8564                                 ill->ill_mc_mtu = ill->ill_mtu;
8565                         }
8566 
8567                         mutex_exit(&ill->ill_lock);
8568                         /*
8569                          * Make sure all dce_generation checks find out
8570                          * that ill_mtu/ill_mc_mtu has changed.
8571                          */
8572                         if (orig_mtu != ill->ill_mtu ||
8573                             orig_mc_mtu != ill->ill_mc_mtu) {
8574                                 dce_increment_all_generations(ill->ill_isv6,
8575                                     ill->ill_ipst);
8576                         }
8577 
8578                         /*
8579                          * Refresh IPMP meta-interface MTU if necessary.
8580                          */
8581                         if (IS_UNDER_IPMP(ill))
8582                                 ipmp_illgrp_refresh_mtu(ill->ill_grp);
8583                         break;
8584 
8585                 case DL_NOTE_LINK_UP:
8586                 case DL_NOTE_LINK_DOWN: {
8587                         /*
8588                          * We are writer. ill / phyint / ipsq assocs stable.
8589                          * The RUNNING flag reflects the state of the link.
8590                          */
8591                         phyint_t *phyint = ill->ill_phyint;
8592                         uint64_t new_phyint_flags;
8593                         boolean_t changed = B_FALSE;
8594                         boolean_t went_up;
8595 
8596                         went_up = notify->dl_notification == DL_NOTE_LINK_UP;
8597                         mutex_enter(&phyint->phyint_lock);
8598 
8599                         new_phyint_flags = went_up ?
8600                             phyint->phyint_flags | PHYI_RUNNING :
8601                             phyint->phyint_flags & ~PHYI_RUNNING;
8602 
8603                         if (IS_IPMP(ill)) {
8604                                 new_phyint_flags = went_up ?
8605                                     new_phyint_flags & ~PHYI_FAILED :
8606                                     new_phyint_flags | PHYI_FAILED;
8607                         }
8608 
8609                         if (new_phyint_flags != phyint->phyint_flags) {
8610                                 phyint->phyint_flags = new_phyint_flags;
8611                                 changed = B_TRUE;
8612                         }
8613                         mutex_exit(&phyint->phyint_lock);
8614                         /*
8615                          * ill_restart_dad handles the DAD restart and routing
8616                          * socket notification logic.
8617                          */
8618                         if (changed) {
8619                                 ill_restart_dad(phyint->phyint_illv4, went_up);
8620                                 ill_restart_dad(phyint->phyint_illv6, went_up);
8621                         }
8622                         break;
8623                 }
8624                 case DL_NOTE_PROMISC_ON_PHYS: {
8625                         phyint_t *phyint = ill->ill_phyint;
8626 
8627                         mutex_enter(&phyint->phyint_lock);
8628                         phyint->phyint_flags |= PHYI_PROMISC;
8629                         mutex_exit(&phyint->phyint_lock);
8630                         break;
8631                 }
8632                 case DL_NOTE_PROMISC_OFF_PHYS: {
8633                         phyint_t *phyint = ill->ill_phyint;
8634 
8635                         mutex_enter(&phyint->phyint_lock);
8636                         phyint->phyint_flags &= ~PHYI_PROMISC;
8637                         mutex_exit(&phyint->phyint_lock);
8638                         break;
8639                 }
8640                 case DL_NOTE_CAPAB_RENEG:
8641                         /*
8642                          * Something changed on the driver side.
8643                          * It wants us to renegotiate the capabilities
8644                          * on this ill. One possible cause is the aggregation
8645                          * interface under us where a port got added or
8646                          * went away.
8647                          *
8648                          * If the capability negotiation is already done
8649                          * or is in progress, reset the capabilities and
8650                          * mark the ill's ill_capab_reneg to be B_TRUE,
8651                          * so that when the ack comes back, we can start
8652                          * the renegotiation process.
8653                          *
8654                          * Note that if ill_capab_reneg is already B_TRUE
8655                          * (ill_dlpi_capab_state is IDS_UNKNOWN in this case),
8656                          * the capability resetting request has been sent
8657                          * and the renegotiation has not been started yet;
8658                          * nothing needs to be done in this case.
8659                          */
8660                         ipsq_current_start(ipsq, ill->ill_ipif, 0);
8661                         ill_capability_reset(ill, B_TRUE);
8662                         ipsq_current_finish(ipsq);
8663                         break;
8664 
8665                 case DL_NOTE_ALLOWED_IPS:
8666                         ill_set_allowed_ips(ill, mp);
8667                         break;
8668                 default:
8669                         ip0dbg(("ip_rput_dlpi_writer: unknown notification "
8670                             "type 0x%x for DL_NOTIFY_IND\n",
8671                             notify->dl_notification));
8672                         break;
8673                 }
8674 
8675                 /*
8676                  * As this is an asynchronous operation, we
8677                  * should not call ill_dlpi_done
8678                  */
8679                 break;
8680         }
8681         case DL_NOTIFY_ACK: {
8682                 dl_notify_ack_t *noteack = (dl_notify_ack_t *)mp->b_rptr;
8683 
8684                 if (noteack->dl_notifications & DL_NOTE_LINK_UP)
8685                         ill->ill_note_link = 1;
8686                 ill_dlpi_done(ill, DL_NOTIFY_REQ);
8687                 break;
8688         }
8689         case DL_PHYS_ADDR_ACK: {
8690                 /*
8691                  * As part of plumbing the interface via SIOCSLIFNAME,
8692                  * ill_dl_phys() will queue a series of DL_PHYS_ADDR_REQs,
8693                  * whose answers we receive here.  As each answer is received,
8694                  * we call ill_dlpi_done() to dispatch the next request as
8695                  * we're processing the current one.  Once all answers have
8696                  * been received, we use ipsq_pending_mp_get() to dequeue the
8697                  * outstanding IOCTL and reply to it.  (Because ill_dl_phys()
8698                  * is invoked from an ill queue, conn_oper_pending_ill is not
8699                  * available, but we know the ioctl is pending on ill_wq.)
8700                  */
8701                 uint_t  paddrlen, paddroff;
8702                 uint8_t *addr;
8703 
8704                 paddrreq = ill->ill_phys_addr_pend;
8705                 paddrlen = ((dl_phys_addr_ack_t *)mp->b_rptr)->dl_addr_length;
8706                 paddroff = ((dl_phys_addr_ack_t *)mp->b_rptr)->dl_addr_offset;
8707                 addr = mp->b_rptr + paddroff;
8708 
8709                 ill_dlpi_done(ill, DL_PHYS_ADDR_REQ);
8710                 if (paddrreq == DL_IPV6_TOKEN) {
8711                         /*
8712                          * bcopy to low-order bits of ill_token
8713                          *
8714                          * XXX Temporary hack - currently, all known tokens
8715                          * are 64 bits, so I'll cheat for the moment.
8716                          */
8717                         bcopy(addr, &ill->ill_token.s6_addr32[2], paddrlen);
8718                         ill->ill_token_length = paddrlen;
8719                         break;
8720                 } else if (paddrreq == DL_IPV6_LINK_LAYER_ADDR) {
8721                         ASSERT(ill->ill_nd_lla_mp == NULL);
8722                         ill_set_ndmp(ill, mp, paddroff, paddrlen);
8723                         mp = NULL;
8724                         break;
8725                 } else if (paddrreq == DL_CURR_DEST_ADDR) {
8726                         ASSERT(ill->ill_dest_addr_mp == NULL);
8727                         ill->ill_dest_addr_mp = mp;
8728                         ill->ill_dest_addr = addr;
8729                         mp = NULL;
8730                         if (ill->ill_isv6) {
8731                                 ill_setdesttoken(ill);
8732                                 ipif_setdestlinklocal(ill->ill_ipif);
8733                         }
8734                         break;
8735                 }
8736 
8737                 ASSERT(paddrreq == DL_CURR_PHYS_ADDR);
8738                 ASSERT(ill->ill_phys_addr_mp == NULL);
8739                 if (!ill->ill_ifname_pending)
8740                         break;
8741                 ill->ill_ifname_pending = 0;
8742                 if (!ioctl_aborted)
8743                         mp1 = ipsq_pending_mp_get(ipsq, &connp);
8744                 if (mp1 != NULL) {
8745                         ASSERT(connp == NULL);
8746                         q = ill->ill_wq;
8747                 }
8748                 /*
8749                  * If any error acks received during the plumbing sequence,
8750                  * ill_ifname_pending_err will be set. Break out and send up
8751                  * the error to the pending ioctl.
8752                  */
8753                 if (ill->ill_ifname_pending_err != 0) {
8754                         err = ill->ill_ifname_pending_err;
8755                         ill->ill_ifname_pending_err = 0;
8756                         break;
8757                 }
8758 
8759                 ill->ill_phys_addr_mp = mp;
8760                 ill->ill_phys_addr = (paddrlen == 0 ? NULL : addr);
8761                 mp = NULL;
8762 
8763                 /*
8764                  * If paddrlen or ill_phys_addr_length is zero, the DLPI
8765                  * provider doesn't support physical addresses.  We check both
8766                  * paddrlen and ill_phys_addr_length because sppp (PPP) does
8767                  * not have physical addresses, but historically adversises a
8768                  * physical address length of 0 in its DL_INFO_ACK, but 6 in
8769                  * its DL_PHYS_ADDR_ACK.
8770                  */
8771                 if (paddrlen == 0 || ill->ill_phys_addr_length == 0) {
8772                         ill->ill_phys_addr = NULL;
8773                 } else if (paddrlen != ill->ill_phys_addr_length) {
8774                         ip0dbg(("DL_PHYS_ADDR_ACK: got addrlen %d, expected %d",
8775                             paddrlen, ill->ill_phys_addr_length));
8776                         err = EINVAL;
8777                         break;
8778                 }
8779 
8780                 if (ill->ill_nd_lla_mp == NULL) {
8781                         if ((mp_hw = copyb(ill->ill_phys_addr_mp)) == NULL) {
8782                                 err = ENOMEM;
8783                                 break;
8784                         }
8785                         ill_set_ndmp(ill, mp_hw, paddroff, paddrlen);
8786                 }
8787 
8788                 if (ill->ill_isv6) {
8789                         ill_setdefaulttoken(ill);
8790                         ipif_setlinklocal(ill->ill_ipif);
8791                 }
8792                 break;
8793         }
8794         case DL_OK_ACK:
8795                 ip2dbg(("DL_OK_ACK %s (0x%x)\n",
8796                     dl_primstr((int)dloa->dl_correct_primitive),
8797                     dloa->dl_correct_primitive));
8798                 DTRACE_PROBE3(ill__dlpi, char *, "ip_rput_dlpi_writer ok",
8799                     char *, dl_primstr(dloa->dl_correct_primitive),
8800                     ill_t *, ill);
8801 
8802                 switch (dloa->dl_correct_primitive) {
8803                 case DL_ENABMULTI_REQ:
8804                 case DL_DISABMULTI_REQ:
8805                         ill_dlpi_done(ill, dloa->dl_correct_primitive);
8806                         break;
8807                 case DL_PROMISCON_REQ:
8808                 case DL_PROMISCOFF_REQ:
8809                 case DL_UNBIND_REQ:
8810                 case DL_ATTACH_REQ:
8811                         ill_dlpi_done(ill, dloa->dl_correct_primitive);
8812                         break;
8813                 }
8814                 break;
8815         default:
8816                 break;
8817         }
8818 
8819         freemsg(mp);
8820         if (mp1 == NULL)
8821                 return;
8822 
8823         /*
8824          * The operation must complete without EINPROGRESS since
8825          * ipsq_pending_mp_get() has removed the mblk (mp1).  Otherwise,
8826          * the operation will be stuck forever inside the IPSQ.
8827          */
8828         ASSERT(err != EINPROGRESS);
8829 
8830         DTRACE_PROBE4(ipif__ioctl, char *, "ip_rput_dlpi_writer finish",
8831             int, ipsq->ipsq_xop->ipx_current_ioctl, ill_t *, ill,
8832             ipif_t *, NULL);
8833 
8834         switch (ipsq->ipsq_xop->ipx_current_ioctl) {
8835         case 0:
8836                 ipsq_current_finish(ipsq);
8837                 break;
8838 
8839         case SIOCSLIFNAME:
8840         case IF_UNITSEL: {
8841                 ill_t *ill_other = ILL_OTHER(ill);
8842 
8843                 /*
8844                  * If SIOCSLIFNAME or IF_UNITSEL is about to succeed, and the
8845                  * ill has a peer which is in an IPMP group, then place ill
8846                  * into the same group.  One catch: although ifconfig plumbs
8847                  * the appropriate IPMP meta-interface prior to plumbing this
8848                  * ill, it is possible for multiple ifconfig applications to
8849                  * race (or for another application to adjust plumbing), in
8850                  * which case the IPMP meta-interface we need will be missing.
8851                  * If so, kick the phyint out of the group.
8852                  */
8853                 if (err == 0 && ill_other != NULL && IS_UNDER_IPMP(ill_other)) {
8854                         ipmp_grp_t      *grp = ill->ill_phyint->phyint_grp;
8855                         ipmp_illgrp_t   *illg;
8856 
8857                         illg = ill->ill_isv6 ? grp->gr_v6 : grp->gr_v4;
8858                         if (illg == NULL)
8859                                 ipmp_phyint_leave_grp(ill->ill_phyint);
8860                         else
8861                                 ipmp_ill_join_illgrp(ill, illg);
8862                 }
8863 
8864                 if (ipsq->ipsq_xop->ipx_current_ioctl == IF_UNITSEL)
8865                         ip_ioctl_finish(q, mp1, err, NO_COPYOUT, ipsq);
8866                 else
8867                         ip_ioctl_finish(q, mp1, err, COPYOUT, ipsq);
8868                 break;
8869         }
8870         case SIOCLIFADDIF:
8871                 ip_ioctl_finish(q, mp1, err, COPYOUT, ipsq);
8872                 break;
8873 
8874         default:
8875                 ip_ioctl_finish(q, mp1, err, NO_COPYOUT, ipsq);
8876                 break;
8877         }
8878 }
8879 
8880 /*
8881  * ip_rput_other is called by ip_rput to handle messages modifying the global
8882  * state in IP.  If 'ipsq' is non-NULL, caller is writer on it.
8883  */
8884 /* ARGSUSED */
8885 void
8886 ip_rput_other(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *dummy_arg)
8887 {
8888         ill_t           *ill = q->q_ptr;
8889         struct iocblk   *iocp;
8890 
8891         ip1dbg(("ip_rput_other "));
8892         if (ipsq != NULL) {
8893                 ASSERT(IAM_WRITER_IPSQ(ipsq));
8894                 ASSERT(ipsq->ipsq_xop ==
8895                     ill->ill_phyint->phyint_ipsq->ipsq_xop);
8896         }
8897 
8898         switch (mp->b_datap->db_type) {
8899         case M_ERROR:
8900         case M_HANGUP:
8901                 /*
8902                  * The device has a problem.  We force the ILL down.  It can
8903                  * be brought up again manually using SIOCSIFFLAGS (via
8904                  * ifconfig or equivalent).
8905                  */
8906                 ASSERT(ipsq != NULL);
8907                 if (mp->b_rptr < mp->b_wptr)
8908                         ill->ill_error = (int)(*mp->b_rptr & 0xFF);
8909                 if (ill->ill_error == 0)
8910                         ill->ill_error = ENXIO;
8911                 if (!ill_down_start(q, mp))
8912                         return;
8913                 ipif_all_down_tail(ipsq, q, mp, NULL);
8914                 break;
8915         case M_IOCNAK: {
8916                 iocp = (struct iocblk *)mp->b_rptr;
8917 
8918                 ASSERT(iocp->ioc_cmd == DL_IOC_HDR_INFO);
8919                 /*
8920                  * If this was the first attempt, turn off the fastpath
8921                  * probing.
8922                  */
8923                 mutex_enter(&ill->ill_lock);
8924                 if (ill->ill_dlpi_fastpath_state == IDS_INPROGRESS) {
8925                         ill->ill_dlpi_fastpath_state = IDS_FAILED;
8926                         mutex_exit(&ill->ill_lock);
8927                         /*
8928                          * don't flush the nce_t entries: we use them
8929                          * as an index to the ncec itself.
8930                          */
8931                         ip1dbg(("ip_rput: DLPI fastpath off on interface %s\n",
8932                             ill->ill_name));
8933                 } else {
8934                         mutex_exit(&ill->ill_lock);
8935                 }
8936                 freemsg(mp);
8937                 break;
8938         }
8939         default:
8940                 ASSERT(0);
8941                 break;
8942         }
8943 }
8944 
8945 /*
8946  * Update any source route, record route or timestamp options
8947  * When it fails it has consumed the message and BUMPed the MIB.
8948  */
8949 boolean_t
8950 ip_forward_options(mblk_t *mp, ipha_t *ipha, ill_t *dst_ill,
8951     ip_recv_attr_t *ira)
8952 {
8953         ipoptp_t        opts;
8954         uchar_t         *opt;
8955         uint8_t         optval;
8956         uint8_t         optlen;
8957         ipaddr_t        dst;
8958         ipaddr_t        ifaddr;
8959         uint32_t        ts;
8960         timestruc_t     now;
8961         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
8962 
8963         ip2dbg(("ip_forward_options\n"));
8964         dst = ipha->ipha_dst;
8965         for (optval = ipoptp_first(&opts, ipha);
8966             optval != IPOPT_EOL;
8967             optval = ipoptp_next(&opts)) {
8968                 ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
8969                 opt = opts.ipoptp_cur;
8970                 optlen = opts.ipoptp_len;
8971                 ip2dbg(("ip_forward_options: opt %d, len %d\n",
8972                     optval, opts.ipoptp_len));
8973                 switch (optval) {
8974                         uint32_t off;
8975                 case IPOPT_SSRR:
8976                 case IPOPT_LSRR:
8977                         /* Check if adminstratively disabled */
8978                         if (!ipst->ips_ip_forward_src_routed) {
8979                                 BUMP_MIB(dst_ill->ill_ip_mib,
8980                                     ipIfStatsForwProhibits);
8981                                 ip_drop_input("ICMP_SOURCE_ROUTE_FAILED",
8982                                     mp, dst_ill);
8983                                 icmp_unreachable(mp, ICMP_SOURCE_ROUTE_FAILED,
8984                                     ira);
8985                                 return (B_FALSE);
8986                         }
8987                         if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
8988                                 /*
8989                                  * Must be partial since ip_input_options
8990                                  * checked for strict.
8991                                  */
8992                                 break;
8993                         }
8994                         off = opt[IPOPT_OFFSET];
8995                         off--;
8996                 redo_srr:
8997                         if (optlen < IP_ADDR_LEN ||
8998                             off > optlen - IP_ADDR_LEN) {
8999                                 /* End of source route */
9000                                 ip1dbg((
9001                                     "ip_forward_options: end of SR\n"));
9002                                 break;
9003                         }
9004                         /* Pick a reasonable address on the outbound if */
9005                         ASSERT(dst_ill != NULL);
9006                         if (ip_select_source_v4(dst_ill, INADDR_ANY, dst,
9007                             INADDR_ANY, ALL_ZONES, ipst, &ifaddr, NULL,
9008                             NULL) != 0) {
9009                                 /* No source! Shouldn't happen */
9010                                 ifaddr = INADDR_ANY;
9011                         }
9012                         bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
9013                         bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9014                         ip1dbg(("ip_forward_options: next hop 0x%x\n",
9015                             ntohl(dst)));
9016 
9017                         /*
9018                          * Check if our address is present more than
9019                          * once as consecutive hops in source route.
9020                          */
9021                         if (ip_type_v4(dst, ipst) == IRE_LOCAL) {
9022                                 off += IP_ADDR_LEN;
9023                                 opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9024                                 goto redo_srr;
9025                         }
9026                         ipha->ipha_dst = dst;
9027                         opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9028                         break;
9029                 case IPOPT_RR:
9030                         off = opt[IPOPT_OFFSET];
9031                         off--;
9032                         if (optlen < IP_ADDR_LEN ||
9033                             off > optlen - IP_ADDR_LEN) {
9034                                 /* No more room - ignore */
9035                                 ip1dbg((
9036                                     "ip_forward_options: end of RR\n"));
9037                                 break;
9038                         }
9039                         /* Pick a reasonable address on the outbound if */
9040                         ASSERT(dst_ill != NULL);
9041                         if (ip_select_source_v4(dst_ill, INADDR_ANY, dst,
9042                             INADDR_ANY, ALL_ZONES, ipst, &ifaddr, NULL,
9043                             NULL) != 0) {
9044                                 /* No source! Shouldn't happen */
9045                                 ifaddr = INADDR_ANY;
9046                         }
9047                         bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9048                         opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9049                         break;
9050                 case IPOPT_TS:
9051                         /* Insert timestamp if there is room */
9052                         switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9053                         case IPOPT_TS_TSONLY:
9054                                 off = IPOPT_TS_TIMELEN;
9055                                 break;
9056                         case IPOPT_TS_PRESPEC:
9057                         case IPOPT_TS_PRESPEC_RFC791:
9058                                 /* Verify that the address matched */
9059                                 off = opt[IPOPT_OFFSET] - 1;
9060                                 bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
9061                                 if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
9062                                         /* Not for us */
9063                                         break;
9064                                 }
9065                                 /* FALLTHRU */
9066                         case IPOPT_TS_TSANDADDR:
9067                                 off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
9068                                 break;
9069                         default:
9070                                 /*
9071                                  * ip_*put_options should have already
9072                                  * dropped this packet.
9073                                  */
9074                                 cmn_err(CE_PANIC, "ip_forward_options: "
9075                                     "unknown IT - bug in ip_input_options?\n");
9076                                 return (B_TRUE);        /* Keep "lint" happy */
9077                         }
9078                         if (opt[IPOPT_OFFSET] - 1 + off > optlen) {
9079                                 /* Increase overflow counter */
9080                                 off = (opt[IPOPT_POS_OV_FLG] >> 4) + 1;
9081                                 opt[IPOPT_POS_OV_FLG] =
9082                                     (uint8_t)((opt[IPOPT_POS_OV_FLG] & 0x0F) |
9083                                     (off << 4));
9084                                 break;
9085                         }
9086                         off = opt[IPOPT_OFFSET] - 1;
9087                         switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9088                         case IPOPT_TS_PRESPEC:
9089                         case IPOPT_TS_PRESPEC_RFC791:
9090                         case IPOPT_TS_TSANDADDR:
9091                                 /* Pick a reasonable addr on the outbound if */
9092                                 ASSERT(dst_ill != NULL);
9093                                 if (ip_select_source_v4(dst_ill, INADDR_ANY,
9094                                     dst, INADDR_ANY, ALL_ZONES, ipst, &ifaddr,
9095                                     NULL, NULL) != 0) {
9096                                         /* No source! Shouldn't happen */
9097                                         ifaddr = INADDR_ANY;
9098                                 }
9099                                 bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9100                                 opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9101                                 /* FALLTHRU */
9102                         case IPOPT_TS_TSONLY:
9103                                 off = opt[IPOPT_OFFSET] - 1;
9104                                 /* Compute # of milliseconds since midnight */
9105                                 gethrestime(&now);
9106                                 ts = (now.tv_sec % (24 * 60 * 60)) * 1000 +
9107                                     now.tv_nsec / (NANOSEC / MILLISEC);
9108                                 bcopy(&ts, (char *)opt + off, IPOPT_TS_TIMELEN);
9109                                 opt[IPOPT_OFFSET] += IPOPT_TS_TIMELEN;
9110                                 break;
9111                         }
9112                         break;
9113                 }
9114         }
9115         return (B_TRUE);
9116 }
9117 
9118 /*
9119  * Call ill_frag_timeout to do garbage collection. ill_frag_timeout
9120  * returns 'true' if there are still fragments left on the queue, in
9121  * which case we restart the timer.
9122  */
9123 void
9124 ill_frag_timer(void *arg)
9125 {
9126         ill_t   *ill = (ill_t *)arg;
9127         boolean_t frag_pending;
9128         ip_stack_t *ipst = ill->ill_ipst;
9129         time_t  timeout;
9130 
9131         mutex_enter(&ill->ill_lock);
9132         ASSERT(!ill->ill_fragtimer_executing);
9133         if (ill->ill_state_flags & ILL_CONDEMNED) {
9134                 ill->ill_frag_timer_id = 0;
9135                 mutex_exit(&ill->ill_lock);
9136                 return;
9137         }
9138         ill->ill_fragtimer_executing = 1;
9139         mutex_exit(&ill->ill_lock);
9140 
9141         timeout = (ill->ill_isv6 ? ipst->ips_ipv6_reassembly_timeout :
9142             ipst->ips_ip_reassembly_timeout);
9143 
9144         frag_pending = ill_frag_timeout(ill, timeout);
9145 
9146         /*
9147          * Restart the timer, if we have fragments pending or if someone
9148          * wanted us to be scheduled again.
9149          */
9150         mutex_enter(&ill->ill_lock);
9151         ill->ill_fragtimer_executing = 0;
9152         ill->ill_frag_timer_id = 0;
9153         if (frag_pending || ill->ill_fragtimer_needrestart)
9154                 ill_frag_timer_start(ill);
9155         mutex_exit(&ill->ill_lock);
9156 }
9157 
9158 void
9159 ill_frag_timer_start(ill_t *ill)
9160 {
9161         ip_stack_t *ipst = ill->ill_ipst;
9162         clock_t timeo_ms;
9163 
9164         ASSERT(MUTEX_HELD(&ill->ill_lock));
9165 
9166         /* If the ill is closing or opening don't proceed */
9167         if (ill->ill_state_flags & ILL_CONDEMNED)
9168                 return;
9169 
9170         if (ill->ill_fragtimer_executing) {
9171                 /*
9172                  * ill_frag_timer is currently executing. Just record the
9173                  * the fact that we want the timer to be restarted.
9174                  * ill_frag_timer will post a timeout before it returns,
9175                  * ensuring it will be called again.
9176                  */
9177                 ill->ill_fragtimer_needrestart = 1;
9178                 return;
9179         }
9180 
9181         if (ill->ill_frag_timer_id == 0) {
9182                 timeo_ms = (ill->ill_isv6 ? ipst->ips_ipv6_reassembly_timeout :
9183                     ipst->ips_ip_reassembly_timeout) * SECONDS;
9184 
9185                 /*
9186                  * The timer is neither running nor is the timeout handler
9187                  * executing. Post a timeout so that ill_frag_timer will be
9188                  * called
9189                  */
9190                 ill->ill_frag_timer_id = timeout(ill_frag_timer, ill,
9191                     MSEC_TO_TICK(timeo_ms >> 1));
9192                 ill->ill_fragtimer_needrestart = 0;
9193         }
9194 }
9195 
9196 /*
9197  * Update any source route, record route or timestamp options.
9198  * Check that we are at end of strict source route.
9199  * The options have already been checked for sanity in ip_input_options().
9200  */
9201 boolean_t
9202 ip_input_local_options(mblk_t *mp, ipha_t *ipha, ip_recv_attr_t *ira)
9203 {
9204         ipoptp_t        opts;
9205         uchar_t         *opt;
9206         uint8_t         optval;
9207         uint8_t         optlen;
9208         ipaddr_t        dst;
9209         ipaddr_t        ifaddr;
9210         uint32_t        ts;
9211         timestruc_t     now;
9212         ill_t           *ill = ira->ira_ill;
9213         ip_stack_t      *ipst = ill->ill_ipst;
9214 
9215         ip2dbg(("ip_input_local_options\n"));
9216 
9217         for (optval = ipoptp_first(&opts, ipha);
9218             optval != IPOPT_EOL;
9219             optval = ipoptp_next(&opts)) {
9220                 ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
9221                 opt = opts.ipoptp_cur;
9222                 optlen = opts.ipoptp_len;
9223                 ip2dbg(("ip_input_local_options: opt %d, len %d\n",
9224                     optval, optlen));
9225                 switch (optval) {
9226                         uint32_t off;
9227                 case IPOPT_SSRR:
9228                 case IPOPT_LSRR:
9229                         off = opt[IPOPT_OFFSET];
9230                         off--;
9231                         if (optlen < IP_ADDR_LEN ||
9232                             off > optlen - IP_ADDR_LEN) {
9233                                 /* End of source route */
9234                                 ip1dbg(("ip_input_local_options: end of SR\n"));
9235                                 break;
9236                         }
9237                         /*
9238                          * This will only happen if two consecutive entries
9239                          * in the source route contains our address or if
9240                          * it is a packet with a loose source route which
9241                          * reaches us before consuming the whole source route
9242                          */
9243                         ip1dbg(("ip_input_local_options: not end of SR\n"));
9244                         if (optval == IPOPT_SSRR) {
9245                                 goto bad_src_route;
9246                         }
9247                         /*
9248                          * Hack: instead of dropping the packet truncate the
9249                          * source route to what has been used by filling the
9250                          * rest with IPOPT_NOP.
9251                          */
9252                         opt[IPOPT_OLEN] = (uint8_t)off;
9253                         while (off < optlen) {
9254                                 opt[off++] = IPOPT_NOP;
9255                         }
9256                         break;
9257                 case IPOPT_RR:
9258                         off = opt[IPOPT_OFFSET];
9259                         off--;
9260                         if (optlen < IP_ADDR_LEN ||
9261                             off > optlen - IP_ADDR_LEN) {
9262                                 /* No more room - ignore */
9263                                 ip1dbg((
9264                                     "ip_input_local_options: end of RR\n"));
9265                                 break;
9266                         }
9267                         /* Pick a reasonable address on the outbound if */
9268                         if (ip_select_source_v4(ill, INADDR_ANY, ipha->ipha_dst,
9269                             INADDR_ANY, ALL_ZONES, ipst, &ifaddr, NULL,
9270                             NULL) != 0) {
9271                                 /* No source! Shouldn't happen */
9272                                 ifaddr = INADDR_ANY;
9273                         }
9274                         bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9275                         opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9276                         break;
9277                 case IPOPT_TS:
9278                         /* Insert timestamp if there is romm */
9279                         switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9280                         case IPOPT_TS_TSONLY:
9281                                 off = IPOPT_TS_TIMELEN;
9282                                 break;
9283                         case IPOPT_TS_PRESPEC:
9284                         case IPOPT_TS_PRESPEC_RFC791:
9285                                 /* Verify that the address matched */
9286                                 off = opt[IPOPT_OFFSET] - 1;
9287                                 bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
9288                                 if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
9289                                         /* Not for us */
9290                                         break;
9291                                 }
9292                                 /* FALLTHRU */
9293                         case IPOPT_TS_TSANDADDR:
9294                                 off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
9295                                 break;
9296                         default:
9297                                 /*
9298                                  * ip_*put_options should have already
9299                                  * dropped this packet.
9300                                  */
9301                                 cmn_err(CE_PANIC, "ip_input_local_options: "
9302                                     "unknown IT - bug in ip_input_options?\n");
9303                                 return (B_TRUE);        /* Keep "lint" happy */
9304                         }
9305                         if (opt[IPOPT_OFFSET] - 1 + off > optlen) {
9306                                 /* Increase overflow counter */
9307                                 off = (opt[IPOPT_POS_OV_FLG] >> 4) + 1;
9308                                 opt[IPOPT_POS_OV_FLG] =
9309                                     (uint8_t)((opt[IPOPT_POS_OV_FLG] & 0x0F) |
9310                                     (off << 4));
9311                                 break;
9312                         }
9313                         off = opt[IPOPT_OFFSET] - 1;
9314                         switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9315                         case IPOPT_TS_PRESPEC:
9316                         case IPOPT_TS_PRESPEC_RFC791:
9317                         case IPOPT_TS_TSANDADDR:
9318                                 /* Pick a reasonable addr on the outbound if */
9319                                 if (ip_select_source_v4(ill, INADDR_ANY,
9320                                     ipha->ipha_dst, INADDR_ANY, ALL_ZONES, ipst,
9321                                     &ifaddr, NULL, NULL) != 0) {
9322                                         /* No source! Shouldn't happen */
9323                                         ifaddr = INADDR_ANY;
9324                                 }
9325                                 bcopy(&ifaddr, (char *)opt + off, IP_ADDR_LEN);
9326                                 opt[IPOPT_OFFSET] += IP_ADDR_LEN;
9327                                 /* FALLTHRU */
9328                         case IPOPT_TS_TSONLY:
9329                                 off = opt[IPOPT_OFFSET] - 1;
9330                                 /* Compute # of milliseconds since midnight */
9331                                 gethrestime(&now);
9332                                 ts = (now.tv_sec % (24 * 60 * 60)) * 1000 +
9333                                     now.tv_nsec / (NANOSEC / MILLISEC);
9334                                 bcopy(&ts, (char *)opt + off, IPOPT_TS_TIMELEN);
9335                                 opt[IPOPT_OFFSET] += IPOPT_TS_TIMELEN;
9336                                 break;
9337                         }
9338                         break;
9339                 }
9340         }
9341         return (B_TRUE);
9342 
9343 bad_src_route:
9344         /* make sure we clear any indication of a hardware checksum */
9345         DB_CKSUMFLAGS(mp) = 0;
9346         ip_drop_input("ICMP_SOURCE_ROUTE_FAILED", mp, ill);
9347         icmp_unreachable(mp, ICMP_SOURCE_ROUTE_FAILED, ira);
9348         return (B_FALSE);
9349 
9350 }
9351 
9352 /*
9353  * Process IP options in an inbound packet.  Always returns the nexthop.
9354  * Normally this is the passed in nexthop, but if there is an option
9355  * that effects the nexthop (such as a source route) that will be returned.
9356  * Sets *errorp if there is an error, in which case an ICMP error has been sent
9357  * and mp freed.
9358  */
9359 ipaddr_t
9360 ip_input_options(ipha_t *ipha, ipaddr_t dst, mblk_t *mp,
9361     ip_recv_attr_t *ira, int *errorp)
9362 {
9363         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
9364         ipoptp_t        opts;
9365         uchar_t         *opt;
9366         uint8_t         optval;
9367         uint8_t         optlen;
9368         intptr_t        code = 0;
9369         ire_t           *ire;
9370 
9371         ip2dbg(("ip_input_options\n"));
9372         *errorp = 0;
9373         for (optval = ipoptp_first(&opts, ipha);
9374             optval != IPOPT_EOL;
9375             optval = ipoptp_next(&opts)) {
9376                 opt = opts.ipoptp_cur;
9377                 optlen = opts.ipoptp_len;
9378                 ip2dbg(("ip_input_options: opt %d, len %d\n",
9379                     optval, optlen));
9380                 /*
9381                  * Note: we need to verify the checksum before we
9382                  * modify anything thus this routine only extracts the next
9383                  * hop dst from any source route.
9384                  */
9385                 switch (optval) {
9386                         uint32_t off;
9387                 case IPOPT_SSRR:
9388                 case IPOPT_LSRR:
9389                         if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
9390                                 if (optval == IPOPT_SSRR) {
9391                                         ip1dbg(("ip_input_options: not next"
9392                                             " strict source route 0x%x\n",
9393                                             ntohl(dst)));
9394                                         code = (char *)&ipha->ipha_dst -
9395                                             (char *)ipha;
9396                                         goto param_prob; /* RouterReq's */
9397                                 }
9398                                 ip2dbg(("ip_input_options: "
9399                                     "not next source route 0x%x\n",
9400                                     ntohl(dst)));
9401                                 break;
9402                         }
9403 
9404                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
9405                                 ip1dbg((
9406                                     "ip_input_options: bad option offset\n"));
9407                                 code = (char *)&opt[IPOPT_OLEN] -
9408                                     (char *)ipha;
9409                                 goto param_prob;
9410                         }
9411                         off = opt[IPOPT_OFFSET];
9412                         off--;
9413                 redo_srr:
9414                         if (optlen < IP_ADDR_LEN ||
9415                             off > optlen - IP_ADDR_LEN) {
9416                                 /* End of source route */
9417                                 ip1dbg(("ip_input_options: end of SR\n"));
9418                                 break;
9419                         }
9420                         bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
9421                         ip1dbg(("ip_input_options: next hop 0x%x\n",
9422                             ntohl(dst)));
9423 
9424                         /*
9425                          * Check if our address is present more than
9426                          * once as consecutive hops in source route.
9427                          * XXX verify per-interface ip_forwarding
9428                          * for source route?
9429                          */
9430                         if (ip_type_v4(dst, ipst) == IRE_LOCAL) {
9431                                 off += IP_ADDR_LEN;
9432                                 goto redo_srr;
9433                         }
9434 
9435                         if (dst == htonl(INADDR_LOOPBACK)) {
9436                                 ip1dbg(("ip_input_options: loopback addr in "
9437                                     "source route!\n"));
9438                                 goto bad_src_route;
9439                         }
9440                         /*
9441                          * For strict: verify that dst is directly
9442                          * reachable.
9443                          */
9444                         if (optval == IPOPT_SSRR) {
9445                                 ire = ire_ftable_lookup_v4(dst, 0, 0,
9446                                     IRE_INTERFACE, NULL, ALL_ZONES,
9447                                     ira->ira_tsl,
9448                                     MATCH_IRE_TYPE | MATCH_IRE_SECATTR, 0, ipst,
9449                                     NULL);
9450                                 if (ire == NULL) {
9451                                         ip1dbg(("ip_input_options: SSRR not "
9452                                             "directly reachable: 0x%x\n",
9453                                             ntohl(dst)));
9454                                         goto bad_src_route;
9455                                 }
9456                                 ire_refrele(ire);
9457                         }
9458                         /*
9459                          * Defer update of the offset and the record route
9460                          * until the packet is forwarded.
9461                          */
9462                         break;
9463                 case IPOPT_RR:
9464                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
9465                                 ip1dbg((
9466                                     "ip_input_options: bad option offset\n"));
9467                                 code = (char *)&opt[IPOPT_OLEN] -
9468                                     (char *)ipha;
9469                                 goto param_prob;
9470                         }
9471                         break;
9472                 case IPOPT_TS:
9473                         /*
9474                          * Verify that length >= 5 and that there is either
9475                          * room for another timestamp or that the overflow
9476                          * counter is not maxed out.
9477                          */
9478                         code = (char *)&opt[IPOPT_OLEN] - (char *)ipha;
9479                         if (optlen < IPOPT_MINLEN_IT) {
9480                                 goto param_prob;
9481                         }
9482                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
9483                                 ip1dbg((
9484                                     "ip_input_options: bad option offset\n"));
9485                                 code = (char *)&opt[IPOPT_OFFSET] -
9486                                     (char *)ipha;
9487                                 goto param_prob;
9488                         }
9489                         switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
9490                         case IPOPT_TS_TSONLY:
9491                                 off = IPOPT_TS_TIMELEN;
9492                                 break;
9493                         case IPOPT_TS_TSANDADDR:
9494                         case IPOPT_TS_PRESPEC:
9495                         case IPOPT_TS_PRESPEC_RFC791:
9496                                 off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
9497                                 break;
9498                         default:
9499                                 code = (char *)&opt[IPOPT_POS_OV_FLG] -
9500                                     (char *)ipha;
9501                                 goto param_prob;
9502                         }
9503                         if (opt[IPOPT_OFFSET] - 1 + off > optlen &&
9504                             (opt[IPOPT_POS_OV_FLG] & 0xF0) == 0xF0) {
9505                                 /*
9506                                  * No room and the overflow counter is 15
9507                                  * already.
9508                                  */
9509                                 goto param_prob;
9510                         }
9511                         break;
9512                 }
9513         }
9514 
9515         if ((opts.ipoptp_flags & IPOPTP_ERROR) == 0) {
9516                 return (dst);
9517         }
9518 
9519         ip1dbg(("ip_input_options: error processing IP options."));
9520         code = (char *)&opt[IPOPT_OFFSET] - (char *)ipha;
9521 
9522 param_prob:
9523         /* make sure we clear any indication of a hardware checksum */
9524         DB_CKSUMFLAGS(mp) = 0;
9525         ip_drop_input("ICMP_PARAM_PROBLEM", mp, ira->ira_ill);
9526         icmp_param_problem(mp, (uint8_t)code, ira);
9527         *errorp = -1;
9528         return (dst);
9529 
9530 bad_src_route:
9531         /* make sure we clear any indication of a hardware checksum */
9532         DB_CKSUMFLAGS(mp) = 0;
9533         ip_drop_input("ICMP_SOURCE_ROUTE_FAILED", mp, ira->ira_ill);
9534         icmp_unreachable(mp, ICMP_SOURCE_ROUTE_FAILED, ira);
9535         *errorp = -1;
9536         return (dst);
9537 }
9538 
9539 /*
9540  * IP & ICMP info in >=14 msg's ...
9541  *  - ip fixed part (mib2_ip_t)
9542  *  - icmp fixed part (mib2_icmp_t)
9543  *  - ipAddrEntryTable (ip 20)          all IPv4 ipifs
9544  *  - ipRouteEntryTable (ip 21)         all IPv4 IREs
9545  *  - ipNetToMediaEntryTable (ip 22)    all IPv4 Neighbor Cache entries
9546  *  - ipRouteAttributeTable (ip 102)    labeled routes
9547  *  - ip multicast membership (ip_member_t)
9548  *  - ip multicast source filtering (ip_grpsrc_t)
9549  *  - igmp fixed part (struct igmpstat)
9550  *  - multicast routing stats (struct mrtstat)
9551  *  - multicast routing vifs (array of struct vifctl)
9552  *  - multicast routing routes (array of struct mfcctl)
9553  *  - ip6 fixed part (mib2_ipv6IfStatsEntry_t)
9554  *                                      One per ill plus one generic
9555  *  - icmp6 fixed part (mib2_ipv6IfIcmpEntry_t)
9556  *                                      One per ill plus one generic
9557  *  - ipv6RouteEntry                    all IPv6 IREs
9558  *  - ipv6RouteAttributeTable (ip6 102) labeled routes
9559  *  - ipv6NetToMediaEntry               all IPv6 Neighbor Cache entries
9560  *  - ipv6AddrEntry                     all IPv6 ipifs
9561  *  - ipv6 multicast membership (ipv6_member_t)
9562  *  - ipv6 multicast source filtering (ipv6_grpsrc_t)
9563  *
9564  * NOTE: original mpctl is copied for msg's 2..N, since its ctl part is
9565  * already filled in by the caller.
9566  * If legacy_req is true then MIB structures needs to be truncated to their
9567  * legacy sizes before being returned.
9568  * Return value of 0 indicates that no messages were sent and caller
9569  * should free mpctl.
9570  */
9571 int
9572 ip_snmp_get(queue_t *q, mblk_t *mpctl, int level, boolean_t legacy_req)
9573 {
9574         ip_stack_t *ipst;
9575         sctp_stack_t *sctps;
9576 
9577         if (q->q_next != NULL) {
9578                 ipst = ILLQ_TO_IPST(q);
9579         } else {
9580                 ipst = CONNQ_TO_IPST(q);
9581         }
9582         ASSERT(ipst != NULL);
9583         sctps = ipst->ips_netstack->netstack_sctp;
9584 
9585         if (mpctl == NULL || mpctl->b_cont == NULL) {
9586                 return (0);
9587         }
9588 
9589         /*
9590          * For the purposes of the (broken) packet shell use
9591          * of the level we make sure MIB2_TCP/MIB2_UDP can be used
9592          * to make TCP and UDP appear first in the list of mib items.
9593          * TBD: We could expand this and use it in netstat so that
9594          * the kernel doesn't have to produce large tables (connections,
9595          * routes, etc) when netstat only wants the statistics or a particular
9596          * table.
9597          */
9598         if (!(level == MIB2_TCP || level == MIB2_UDP)) {
9599                 if ((mpctl = icmp_snmp_get(q, mpctl)) == NULL) {
9600                         return (1);
9601                 }
9602         }
9603 
9604         if (level != MIB2_TCP) {
9605                 if ((mpctl = udp_snmp_get(q, mpctl, legacy_req)) == NULL) {
9606                         return (1);
9607                 }
9608         }
9609 
9610         if (level != MIB2_UDP) {
9611                 if ((mpctl = tcp_snmp_get(q, mpctl, legacy_req)) == NULL) {
9612                         return (1);
9613                 }
9614         }
9615 
9616         if ((mpctl = ip_snmp_get_mib2_ip_traffic_stats(q, mpctl,
9617             ipst, legacy_req)) == NULL) {
9618                 return (1);
9619         }
9620 
9621         if ((mpctl = ip_snmp_get_mib2_ip6(q, mpctl, ipst,
9622             legacy_req)) == NULL) {
9623                 return (1);
9624         }
9625 
9626         if ((mpctl = ip_snmp_get_mib2_icmp(q, mpctl, ipst)) == NULL) {
9627                 return (1);
9628         }
9629 
9630         if ((mpctl = ip_snmp_get_mib2_icmp6(q, mpctl, ipst)) == NULL) {
9631                 return (1);
9632         }
9633 
9634         if ((mpctl = ip_snmp_get_mib2_igmp(q, mpctl, ipst)) == NULL) {
9635                 return (1);
9636         }
9637 
9638         if ((mpctl = ip_snmp_get_mib2_multi(q, mpctl, ipst)) == NULL) {
9639                 return (1);
9640         }
9641 
9642         if ((mpctl = ip_snmp_get_mib2_ip_addr(q, mpctl, ipst,
9643             legacy_req)) == NULL) {
9644                 return (1);
9645         }
9646 
9647         if ((mpctl = ip_snmp_get_mib2_ip6_addr(q, mpctl, ipst,
9648             legacy_req)) == NULL) {
9649                 return (1);
9650         }
9651 
9652         if ((mpctl = ip_snmp_get_mib2_ip_group_mem(q, mpctl, ipst)) == NULL) {
9653                 return (1);
9654         }
9655 
9656         if ((mpctl = ip_snmp_get_mib2_ip6_group_mem(q, mpctl, ipst)) == NULL) {
9657                 return (1);
9658         }
9659 
9660         if ((mpctl = ip_snmp_get_mib2_ip_group_src(q, mpctl, ipst)) == NULL) {
9661                 return (1);
9662         }
9663 
9664         if ((mpctl = ip_snmp_get_mib2_ip6_group_src(q, mpctl, ipst)) == NULL) {
9665                 return (1);
9666         }
9667 
9668         if ((mpctl = ip_snmp_get_mib2_virt_multi(q, mpctl, ipst)) == NULL) {
9669                 return (1);
9670         }
9671 
9672         if ((mpctl = ip_snmp_get_mib2_multi_rtable(q, mpctl, ipst)) == NULL) {
9673                 return (1);
9674         }
9675 
9676         mpctl = ip_snmp_get_mib2_ip_route_media(q, mpctl, level, ipst);
9677         if (mpctl == NULL)
9678                 return (1);
9679 
9680         mpctl = ip_snmp_get_mib2_ip6_route_media(q, mpctl, level, ipst);
9681         if (mpctl == NULL)
9682                 return (1);
9683 
9684         if ((mpctl = sctp_snmp_get_mib2(q, mpctl, sctps)) == NULL) {
9685                 return (1);
9686         }
9687         if ((mpctl = ip_snmp_get_mib2_ip_dce(q, mpctl, ipst)) == NULL) {
9688                 return (1);
9689         }
9690         freemsg(mpctl);
9691         return (1);
9692 }
9693 
9694 /* Get global (legacy) IPv4 statistics */
9695 static mblk_t *
9696 ip_snmp_get_mib2_ip(queue_t *q, mblk_t *mpctl, mib2_ipIfStatsEntry_t *ipmib,
9697     ip_stack_t *ipst, boolean_t legacy_req)
9698 {
9699         mib2_ip_t               old_ip_mib;
9700         struct opthdr           *optp;
9701         mblk_t                  *mp2ctl;
9702         mib2_ipAddrEntry_t      mae;
9703 
9704         /*
9705          * make a copy of the original message
9706          */
9707         mp2ctl = copymsg(mpctl);
9708 
9709         /* fixed length IP structure... */
9710         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9711         optp->level = MIB2_IP;
9712         optp->name = 0;
9713         SET_MIB(old_ip_mib.ipForwarding,
9714             (WE_ARE_FORWARDING(ipst) ? 1 : 2));
9715         SET_MIB(old_ip_mib.ipDefaultTTL,
9716             (uint32_t)ipst->ips_ip_def_ttl);
9717         SET_MIB(old_ip_mib.ipReasmTimeout,
9718             ipst->ips_ip_reassembly_timeout);
9719         SET_MIB(old_ip_mib.ipAddrEntrySize,
9720             (legacy_req) ? LEGACY_MIB_SIZE(&mae, mib2_ipAddrEntry_t) :
9721             sizeof (mib2_ipAddrEntry_t));
9722         SET_MIB(old_ip_mib.ipRouteEntrySize,
9723             sizeof (mib2_ipRouteEntry_t));
9724         SET_MIB(old_ip_mib.ipNetToMediaEntrySize,
9725             sizeof (mib2_ipNetToMediaEntry_t));
9726         SET_MIB(old_ip_mib.ipMemberEntrySize, sizeof (ip_member_t));
9727         SET_MIB(old_ip_mib.ipGroupSourceEntrySize, sizeof (ip_grpsrc_t));
9728         SET_MIB(old_ip_mib.ipRouteAttributeSize,
9729             sizeof (mib2_ipAttributeEntry_t));
9730         SET_MIB(old_ip_mib.transportMLPSize, sizeof (mib2_transportMLPEntry_t));
9731         SET_MIB(old_ip_mib.ipDestEntrySize, sizeof (dest_cache_entry_t));
9732 
9733         /*
9734          * Grab the statistics from the new IP MIB
9735          */
9736         SET_MIB(old_ip_mib.ipInReceives,
9737             (uint32_t)ipmib->ipIfStatsHCInReceives);
9738         SET_MIB(old_ip_mib.ipInHdrErrors, ipmib->ipIfStatsInHdrErrors);
9739         SET_MIB(old_ip_mib.ipInAddrErrors, ipmib->ipIfStatsInAddrErrors);
9740         SET_MIB(old_ip_mib.ipForwDatagrams,
9741             (uint32_t)ipmib->ipIfStatsHCOutForwDatagrams);
9742         SET_MIB(old_ip_mib.ipInUnknownProtos,
9743             ipmib->ipIfStatsInUnknownProtos);
9744         SET_MIB(old_ip_mib.ipInDiscards, ipmib->ipIfStatsInDiscards);
9745         SET_MIB(old_ip_mib.ipInDelivers,
9746             (uint32_t)ipmib->ipIfStatsHCInDelivers);
9747         SET_MIB(old_ip_mib.ipOutRequests,
9748             (uint32_t)ipmib->ipIfStatsHCOutRequests);
9749         SET_MIB(old_ip_mib.ipOutDiscards, ipmib->ipIfStatsOutDiscards);
9750         SET_MIB(old_ip_mib.ipOutNoRoutes, ipmib->ipIfStatsOutNoRoutes);
9751         SET_MIB(old_ip_mib.ipReasmReqds, ipmib->ipIfStatsReasmReqds);
9752         SET_MIB(old_ip_mib.ipReasmOKs, ipmib->ipIfStatsReasmOKs);
9753         SET_MIB(old_ip_mib.ipReasmFails, ipmib->ipIfStatsReasmFails);
9754         SET_MIB(old_ip_mib.ipFragOKs, ipmib->ipIfStatsOutFragOKs);
9755         SET_MIB(old_ip_mib.ipFragFails, ipmib->ipIfStatsOutFragFails);
9756         SET_MIB(old_ip_mib.ipFragCreates, ipmib->ipIfStatsOutFragCreates);
9757 
9758         /* ipRoutingDiscards is not being used */
9759         SET_MIB(old_ip_mib.ipRoutingDiscards, 0);
9760         SET_MIB(old_ip_mib.tcpInErrs, ipmib->tcpIfStatsInErrs);
9761         SET_MIB(old_ip_mib.udpNoPorts, ipmib->udpIfStatsNoPorts);
9762         SET_MIB(old_ip_mib.ipInCksumErrs, ipmib->ipIfStatsInCksumErrs);
9763         SET_MIB(old_ip_mib.ipReasmDuplicates,
9764             ipmib->ipIfStatsReasmDuplicates);
9765         SET_MIB(old_ip_mib.ipReasmPartDups, ipmib->ipIfStatsReasmPartDups);
9766         SET_MIB(old_ip_mib.ipForwProhibits, ipmib->ipIfStatsForwProhibits);
9767         SET_MIB(old_ip_mib.udpInCksumErrs, ipmib->udpIfStatsInCksumErrs);
9768         SET_MIB(old_ip_mib.udpInOverflows, ipmib->udpIfStatsInOverflows);
9769         SET_MIB(old_ip_mib.rawipInOverflows,
9770             ipmib->rawipIfStatsInOverflows);
9771 
9772         SET_MIB(old_ip_mib.ipsecInSucceeded, ipmib->ipsecIfStatsInSucceeded);
9773         SET_MIB(old_ip_mib.ipsecInFailed, ipmib->ipsecIfStatsInFailed);
9774         SET_MIB(old_ip_mib.ipInIPv6, ipmib->ipIfStatsInWrongIPVersion);
9775         SET_MIB(old_ip_mib.ipOutIPv6, ipmib->ipIfStatsOutWrongIPVersion);
9776         SET_MIB(old_ip_mib.ipOutSwitchIPv6,
9777             ipmib->ipIfStatsOutSwitchIPVersion);
9778 
9779         if (!snmp_append_data(mpctl->b_cont, (char *)&old_ip_mib,
9780             (int)sizeof (old_ip_mib))) {
9781                 ip1dbg(("ip_snmp_get_mib2_ip: failed to allocate %u bytes\n",
9782                     (uint_t)sizeof (old_ip_mib)));
9783         }
9784 
9785         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
9786         ip3dbg(("ip_snmp_get_mib2_ip: level %d, name %d, len %d\n",
9787             (int)optp->level, (int)optp->name, (int)optp->len));
9788         qreply(q, mpctl);
9789         return (mp2ctl);
9790 }
9791 
9792 /* Per interface IPv4 statistics */
9793 static mblk_t *
9794 ip_snmp_get_mib2_ip_traffic_stats(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst,
9795     boolean_t legacy_req)
9796 {
9797         struct opthdr           *optp;
9798         mblk_t                  *mp2ctl;
9799         ill_t                   *ill;
9800         ill_walk_context_t      ctx;
9801         mblk_t                  *mp_tail = NULL;
9802         mib2_ipIfStatsEntry_t   global_ip_mib;
9803         mib2_ipAddrEntry_t      mae;
9804 
9805         /*
9806          * Make a copy of the original message
9807          */
9808         mp2ctl = copymsg(mpctl);
9809 
9810         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9811         optp->level = MIB2_IP;
9812         optp->name = MIB2_IP_TRAFFIC_STATS;
9813         /* Include "unknown interface" ip_mib */
9814         ipst->ips_ip_mib.ipIfStatsIPVersion = MIB2_INETADDRESSTYPE_ipv4;
9815         ipst->ips_ip_mib.ipIfStatsIfIndex =
9816             MIB2_UNKNOWN_INTERFACE; /* Flag to netstat */
9817         SET_MIB(ipst->ips_ip_mib.ipIfStatsForwarding,
9818             (ipst->ips_ip_forwarding ? 1 : 2));
9819         SET_MIB(ipst->ips_ip_mib.ipIfStatsDefaultTTL,
9820             (uint32_t)ipst->ips_ip_def_ttl);
9821         SET_MIB(ipst->ips_ip_mib.ipIfStatsEntrySize,
9822             sizeof (mib2_ipIfStatsEntry_t));
9823         SET_MIB(ipst->ips_ip_mib.ipIfStatsAddrEntrySize,
9824             sizeof (mib2_ipAddrEntry_t));
9825         SET_MIB(ipst->ips_ip_mib.ipIfStatsRouteEntrySize,
9826             sizeof (mib2_ipRouteEntry_t));
9827         SET_MIB(ipst->ips_ip_mib.ipIfStatsNetToMediaEntrySize,
9828             sizeof (mib2_ipNetToMediaEntry_t));
9829         SET_MIB(ipst->ips_ip_mib.ipIfStatsMemberEntrySize,
9830             sizeof (ip_member_t));
9831         SET_MIB(ipst->ips_ip_mib.ipIfStatsGroupSourceEntrySize,
9832             sizeof (ip_grpsrc_t));
9833 
9834         bcopy(&ipst->ips_ip_mib, &global_ip_mib, sizeof (global_ip_mib));
9835 
9836         if (legacy_req) {
9837                 SET_MIB(global_ip_mib.ipIfStatsAddrEntrySize,
9838                     LEGACY_MIB_SIZE(&mae, mib2_ipAddrEntry_t));
9839         }
9840 
9841         if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
9842             (char *)&global_ip_mib, (int)sizeof (global_ip_mib))) {
9843                 ip1dbg(("ip_snmp_get_mib2_ip_traffic_stats: "
9844                     "failed to allocate %u bytes\n",
9845                     (uint_t)sizeof (global_ip_mib)));
9846         }
9847 
9848         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
9849         ill = ILL_START_WALK_V4(&ctx, ipst);
9850         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
9851                 ill->ill_ip_mib->ipIfStatsIfIndex =
9852                     ill->ill_phyint->phyint_ifindex;
9853                 SET_MIB(ill->ill_ip_mib->ipIfStatsForwarding,
9854                     (ipst->ips_ip_forwarding ? 1 : 2));
9855                 SET_MIB(ill->ill_ip_mib->ipIfStatsDefaultTTL,
9856                     (uint32_t)ipst->ips_ip_def_ttl);
9857 
9858                 ip_mib2_add_ip_stats(&global_ip_mib, ill->ill_ip_mib);
9859                 if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
9860                     (char *)ill->ill_ip_mib,
9861                     (int)sizeof (*ill->ill_ip_mib))) {
9862                         ip1dbg(("ip_snmp_get_mib2_ip_traffic_stats: "
9863                             "failed to allocate %u bytes\n",
9864                             (uint_t)sizeof (*ill->ill_ip_mib)));
9865                 }
9866         }
9867         rw_exit(&ipst->ips_ill_g_lock);
9868 
9869         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
9870         ip3dbg(("ip_snmp_get_mib2_ip_traffic_stats: "
9871             "level %d, name %d, len %d\n",
9872             (int)optp->level, (int)optp->name, (int)optp->len));
9873         qreply(q, mpctl);
9874 
9875         if (mp2ctl == NULL)
9876                 return (NULL);
9877 
9878         return (ip_snmp_get_mib2_ip(q, mp2ctl, &global_ip_mib, ipst,
9879             legacy_req));
9880 }
9881 
9882 /* Global IPv4 ICMP statistics */
9883 static mblk_t *
9884 ip_snmp_get_mib2_icmp(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
9885 {
9886         struct opthdr           *optp;
9887         mblk_t                  *mp2ctl;
9888 
9889         /*
9890          * Make a copy of the original message
9891          */
9892         mp2ctl = copymsg(mpctl);
9893 
9894         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9895         optp->level = MIB2_ICMP;
9896         optp->name = 0;
9897         if (!snmp_append_data(mpctl->b_cont, (char *)&ipst->ips_icmp_mib,
9898             (int)sizeof (ipst->ips_icmp_mib))) {
9899                 ip1dbg(("ip_snmp_get_mib2_icmp: failed to allocate %u bytes\n",
9900                     (uint_t)sizeof (ipst->ips_icmp_mib)));
9901         }
9902         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
9903         ip3dbg(("ip_snmp_get_mib2_icmp: level %d, name %d, len %d\n",
9904             (int)optp->level, (int)optp->name, (int)optp->len));
9905         qreply(q, mpctl);
9906         return (mp2ctl);
9907 }
9908 
9909 /* Global IPv4 IGMP statistics */
9910 static mblk_t *
9911 ip_snmp_get_mib2_igmp(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
9912 {
9913         struct opthdr           *optp;
9914         mblk_t                  *mp2ctl;
9915 
9916         /*
9917          * make a copy of the original message
9918          */
9919         mp2ctl = copymsg(mpctl);
9920 
9921         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9922         optp->level = EXPER_IGMP;
9923         optp->name = 0;
9924         if (!snmp_append_data(mpctl->b_cont, (char *)&ipst->ips_igmpstat,
9925             (int)sizeof (ipst->ips_igmpstat))) {
9926                 ip1dbg(("ip_snmp_get_mib2_igmp: failed to allocate %u bytes\n",
9927                     (uint_t)sizeof (ipst->ips_igmpstat)));
9928         }
9929         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
9930         ip3dbg(("ip_snmp_get_mib2_igmp: level %d, name %d, len %d\n",
9931             (int)optp->level, (int)optp->name, (int)optp->len));
9932         qreply(q, mpctl);
9933         return (mp2ctl);
9934 }
9935 
9936 /* Global IPv4 Multicast Routing statistics */
9937 static mblk_t *
9938 ip_snmp_get_mib2_multi(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
9939 {
9940         struct opthdr           *optp;
9941         mblk_t                  *mp2ctl;
9942 
9943         /*
9944          * make a copy of the original message
9945          */
9946         mp2ctl = copymsg(mpctl);
9947 
9948         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9949         optp->level = EXPER_DVMRP;
9950         optp->name = 0;
9951         if (!ip_mroute_stats(mpctl->b_cont, ipst)) {
9952                 ip0dbg(("ip_mroute_stats: failed\n"));
9953         }
9954         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
9955         ip3dbg(("ip_snmp_get_mib2_multi: level %d, name %d, len %d\n",
9956             (int)optp->level, (int)optp->name, (int)optp->len));
9957         qreply(q, mpctl);
9958         return (mp2ctl);
9959 }
9960 
9961 /* IPv4 address information */
9962 static mblk_t *
9963 ip_snmp_get_mib2_ip_addr(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst,
9964     boolean_t legacy_req)
9965 {
9966         struct opthdr           *optp;
9967         mblk_t                  *mp2ctl;
9968         mblk_t                  *mp_tail = NULL;
9969         ill_t                   *ill;
9970         ipif_t                  *ipif;
9971         uint_t                  bitval;
9972         mib2_ipAddrEntry_t      mae;
9973         size_t                  mae_size;
9974         zoneid_t                zoneid;
9975         ill_walk_context_t      ctx;
9976 
9977         /*
9978          * make a copy of the original message
9979          */
9980         mp2ctl = copymsg(mpctl);
9981 
9982         mae_size = (legacy_req) ? LEGACY_MIB_SIZE(&mae, mib2_ipAddrEntry_t) :
9983             sizeof (mib2_ipAddrEntry_t);
9984 
9985         /* ipAddrEntryTable */
9986 
9987         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
9988         optp->level = MIB2_IP;
9989         optp->name = MIB2_IP_ADDR;
9990         zoneid = Q_TO_CONN(q)->conn_zoneid;
9991 
9992         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
9993         ill = ILL_START_WALK_V4(&ctx, ipst);
9994         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
9995                 for (ipif = ill->ill_ipif; ipif != NULL;
9996                     ipif = ipif->ipif_next) {
9997                         if (ipif->ipif_zoneid != zoneid &&
9998                             ipif->ipif_zoneid != ALL_ZONES)
9999                                 continue;
10000                         /* Sum of count from dead IRE_LO* and our current */
10001                         mae.ipAdEntInfo.ae_ibcnt = ipif->ipif_ib_pkt_count;
10002                         if (ipif->ipif_ire_local != NULL) {
10003                                 mae.ipAdEntInfo.ae_ibcnt +=
10004                                     ipif->ipif_ire_local->ire_ib_pkt_count;
10005                         }
10006                         mae.ipAdEntInfo.ae_obcnt = 0;
10007                         mae.ipAdEntInfo.ae_focnt = 0;
10008 
10009                         ipif_get_name(ipif, mae.ipAdEntIfIndex.o_bytes,
10010                             OCTET_LENGTH);
10011                         mae.ipAdEntIfIndex.o_length =
10012                             mi_strlen(mae.ipAdEntIfIndex.o_bytes);
10013                         mae.ipAdEntAddr = ipif->ipif_lcl_addr;
10014                         mae.ipAdEntNetMask = ipif->ipif_net_mask;
10015                         mae.ipAdEntInfo.ae_subnet = ipif->ipif_subnet;
10016                         mae.ipAdEntInfo.ae_subnet_len =
10017                             ip_mask_to_plen(ipif->ipif_net_mask);
10018                         mae.ipAdEntInfo.ae_src_addr = ipif->ipif_lcl_addr;
10019                         for (bitval = 1;
10020                             bitval &&
10021                             !(bitval & ipif->ipif_brd_addr);
10022                             bitval <<= 1)
10023                                 noop;
10024                         mae.ipAdEntBcastAddr = bitval;
10025                         mae.ipAdEntReasmMaxSize = IP_MAXPACKET;
10026                         mae.ipAdEntInfo.ae_mtu = ipif->ipif_ill->ill_mtu;
10027                         mae.ipAdEntInfo.ae_metric  = ipif->ipif_ill->ill_metric;
10028                         mae.ipAdEntInfo.ae_broadcast_addr =
10029                             ipif->ipif_brd_addr;
10030                         mae.ipAdEntInfo.ae_pp_dst_addr =
10031                             ipif->ipif_pp_dst_addr;
10032                         mae.ipAdEntInfo.ae_flags = ipif->ipif_flags |
10033                             ill->ill_flags | ill->ill_phyint->phyint_flags;
10034                         mae.ipAdEntRetransmitTime =
10035                             ill->ill_reachable_retrans_time;
10036 
10037                         if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10038                             (char *)&mae, (int)mae_size)) {
10039                                 ip1dbg(("ip_snmp_get_mib2_ip_addr: failed to "
10040                                     "allocate %u bytes\n", (uint_t)mae_size));
10041                         }
10042                 }
10043         }
10044         rw_exit(&ipst->ips_ill_g_lock);
10045 
10046         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10047         ip3dbg(("ip_snmp_get_mib2_ip_addr: level %d, name %d, len %d\n",
10048             (int)optp->level, (int)optp->name, (int)optp->len));
10049         qreply(q, mpctl);
10050         return (mp2ctl);
10051 }
10052 
10053 /* IPv6 address information */
10054 static mblk_t *
10055 ip_snmp_get_mib2_ip6_addr(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst,
10056     boolean_t legacy_req)
10057 {
10058         struct opthdr           *optp;
10059         mblk_t                  *mp2ctl;
10060         mblk_t                  *mp_tail = NULL;
10061         ill_t                   *ill;
10062         ipif_t                  *ipif;
10063         mib2_ipv6AddrEntry_t    mae6;
10064         size_t                  mae6_size;
10065         zoneid_t                zoneid;
10066         ill_walk_context_t      ctx;
10067 
10068         /*
10069          * make a copy of the original message
10070          */
10071         mp2ctl = copymsg(mpctl);
10072 
10073         mae6_size = (legacy_req) ?
10074             LEGACY_MIB_SIZE(&mae6, mib2_ipv6AddrEntry_t) :
10075             sizeof (mib2_ipv6AddrEntry_t);
10076 
10077         /* ipv6AddrEntryTable */
10078 
10079         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10080         optp->level = MIB2_IP6;
10081         optp->name = MIB2_IP6_ADDR;
10082         zoneid = Q_TO_CONN(q)->conn_zoneid;
10083 
10084         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10085         ill = ILL_START_WALK_V6(&ctx, ipst);
10086         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10087                 for (ipif = ill->ill_ipif; ipif != NULL;
10088                     ipif = ipif->ipif_next) {
10089                         if (ipif->ipif_zoneid != zoneid &&
10090                             ipif->ipif_zoneid != ALL_ZONES)
10091                                 continue;
10092                         /* Sum of count from dead IRE_LO* and our current */
10093                         mae6.ipv6AddrInfo.ae_ibcnt = ipif->ipif_ib_pkt_count;
10094                         if (ipif->ipif_ire_local != NULL) {
10095                                 mae6.ipv6AddrInfo.ae_ibcnt +=
10096                                     ipif->ipif_ire_local->ire_ib_pkt_count;
10097                         }
10098                         mae6.ipv6AddrInfo.ae_obcnt = 0;
10099                         mae6.ipv6AddrInfo.ae_focnt = 0;
10100 
10101                         ipif_get_name(ipif, mae6.ipv6AddrIfIndex.o_bytes,
10102                             OCTET_LENGTH);
10103                         mae6.ipv6AddrIfIndex.o_length =
10104                             mi_strlen(mae6.ipv6AddrIfIndex.o_bytes);
10105                         mae6.ipv6AddrAddress = ipif->ipif_v6lcl_addr;
10106                         mae6.ipv6AddrPfxLength =
10107                             ip_mask_to_plen_v6(&ipif->ipif_v6net_mask);
10108                         mae6.ipv6AddrInfo.ae_subnet = ipif->ipif_v6subnet;
10109                         mae6.ipv6AddrInfo.ae_subnet_len =
10110                             mae6.ipv6AddrPfxLength;
10111                         mae6.ipv6AddrInfo.ae_src_addr = ipif->ipif_v6lcl_addr;
10112 
10113                         /* Type: stateless(1), stateful(2), unknown(3) */
10114                         if (ipif->ipif_flags & IPIF_ADDRCONF)
10115                                 mae6.ipv6AddrType = 1;
10116                         else
10117                                 mae6.ipv6AddrType = 2;
10118                         /* Anycast: true(1), false(2) */
10119                         if (ipif->ipif_flags & IPIF_ANYCAST)
10120                                 mae6.ipv6AddrAnycastFlag = 1;
10121                         else
10122                                 mae6.ipv6AddrAnycastFlag = 2;
10123 
10124                         /*
10125                          * Address status: preferred(1), deprecated(2),
10126                          * invalid(3), inaccessible(4), unknown(5)
10127                          */
10128                         if (ipif->ipif_flags & IPIF_NOLOCAL)
10129                                 mae6.ipv6AddrStatus = 3;
10130                         else if (ipif->ipif_flags & IPIF_DEPRECATED)
10131                                 mae6.ipv6AddrStatus = 2;
10132                         else
10133                                 mae6.ipv6AddrStatus = 1;
10134                         mae6.ipv6AddrInfo.ae_mtu = ipif->ipif_ill->ill_mtu;
10135                         mae6.ipv6AddrInfo.ae_metric  =
10136                             ipif->ipif_ill->ill_metric;
10137                         mae6.ipv6AddrInfo.ae_pp_dst_addr =
10138                             ipif->ipif_v6pp_dst_addr;
10139                         mae6.ipv6AddrInfo.ae_flags = ipif->ipif_flags |
10140                             ill->ill_flags | ill->ill_phyint->phyint_flags;
10141                         mae6.ipv6AddrReasmMaxSize = IP_MAXPACKET;
10142                         mae6.ipv6AddrIdentifier = ill->ill_token;
10143                         mae6.ipv6AddrIdentifierLen = ill->ill_token_length;
10144                         mae6.ipv6AddrReachableTime = ill->ill_reachable_time;
10145                         mae6.ipv6AddrRetransmitTime =
10146                             ill->ill_reachable_retrans_time;
10147                         if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10148                             (char *)&mae6, (int)mae6_size)) {
10149                                 ip1dbg(("ip_snmp_get_mib2_ip6_addr: failed to "
10150                                     "allocate %u bytes\n",
10151                                     (uint_t)mae6_size));
10152                         }
10153                 }
10154         }
10155         rw_exit(&ipst->ips_ill_g_lock);
10156 
10157         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10158         ip3dbg(("ip_snmp_get_mib2_ip6_addr: level %d, name %d, len %d\n",
10159             (int)optp->level, (int)optp->name, (int)optp->len));
10160         qreply(q, mpctl);
10161         return (mp2ctl);
10162 }
10163 
10164 /* IPv4 multicast group membership. */
10165 static mblk_t *
10166 ip_snmp_get_mib2_ip_group_mem(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10167 {
10168         struct opthdr           *optp;
10169         mblk_t                  *mp2ctl;
10170         ill_t                   *ill;
10171         ipif_t                  *ipif;
10172         ilm_t                   *ilm;
10173         ip_member_t             ipm;
10174         mblk_t                  *mp_tail = NULL;
10175         ill_walk_context_t      ctx;
10176         zoneid_t                zoneid;
10177 
10178         /*
10179          * make a copy of the original message
10180          */
10181         mp2ctl = copymsg(mpctl);
10182         zoneid = Q_TO_CONN(q)->conn_zoneid;
10183 
10184         /* ipGroupMember table */
10185         optp = (struct opthdr *)&mpctl->b_rptr[
10186             sizeof (struct T_optmgmt_ack)];
10187         optp->level = MIB2_IP;
10188         optp->name = EXPER_IP_GROUP_MEMBERSHIP;
10189 
10190         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10191         ill = ILL_START_WALK_V4(&ctx, ipst);
10192         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10193                 /* Make sure the ill isn't going away. */
10194                 if (!ill_check_and_refhold(ill))
10195                         continue;
10196                 rw_exit(&ipst->ips_ill_g_lock);
10197                 rw_enter(&ill->ill_mcast_lock, RW_READER);
10198                 for (ilm = ill->ill_ilm; ilm; ilm = ilm->ilm_next) {
10199                         if (ilm->ilm_zoneid != zoneid &&
10200                             ilm->ilm_zoneid != ALL_ZONES)
10201                                 continue;
10202 
10203                         /* Is there an ipif for ilm_ifaddr? */
10204                         for (ipif = ill->ill_ipif; ipif != NULL;
10205                             ipif = ipif->ipif_next) {
10206                                 if (!IPIF_IS_CONDEMNED(ipif) &&
10207                                     ipif->ipif_lcl_addr == ilm->ilm_ifaddr &&
10208                                     ilm->ilm_ifaddr != INADDR_ANY)
10209                                         break;
10210                         }
10211                         if (ipif != NULL) {
10212                                 ipif_get_name(ipif,
10213                                     ipm.ipGroupMemberIfIndex.o_bytes,
10214                                     OCTET_LENGTH);
10215                         } else {
10216                                 ill_get_name(ill,
10217                                     ipm.ipGroupMemberIfIndex.o_bytes,
10218                                     OCTET_LENGTH);
10219                         }
10220                         ipm.ipGroupMemberIfIndex.o_length =
10221                             mi_strlen(ipm.ipGroupMemberIfIndex.o_bytes);
10222 
10223                         ipm.ipGroupMemberAddress = ilm->ilm_addr;
10224                         ipm.ipGroupMemberRefCnt = ilm->ilm_refcnt;
10225                         ipm.ipGroupMemberFilterMode = ilm->ilm_fmode;
10226                         if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10227                             (char *)&ipm, (int)sizeof (ipm))) {
10228                                 ip1dbg(("ip_snmp_get_mib2_ip_group: "
10229                                     "failed to allocate %u bytes\n",
10230                                     (uint_t)sizeof (ipm)));
10231                         }
10232                 }
10233                 rw_exit(&ill->ill_mcast_lock);
10234                 ill_refrele(ill);
10235                 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10236         }
10237         rw_exit(&ipst->ips_ill_g_lock);
10238         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10239         ip3dbg(("ip_snmp_get: level %d, name %d, len %d\n",
10240             (int)optp->level, (int)optp->name, (int)optp->len));
10241         qreply(q, mpctl);
10242         return (mp2ctl);
10243 }
10244 
10245 /* IPv6 multicast group membership. */
10246 static mblk_t *
10247 ip_snmp_get_mib2_ip6_group_mem(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10248 {
10249         struct opthdr           *optp;
10250         mblk_t                  *mp2ctl;
10251         ill_t                   *ill;
10252         ilm_t                   *ilm;
10253         ipv6_member_t           ipm6;
10254         mblk_t                  *mp_tail = NULL;
10255         ill_walk_context_t      ctx;
10256         zoneid_t                zoneid;
10257 
10258         /*
10259          * make a copy of the original message
10260          */
10261         mp2ctl = copymsg(mpctl);
10262         zoneid = Q_TO_CONN(q)->conn_zoneid;
10263 
10264         /* ip6GroupMember table */
10265         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10266         optp->level = MIB2_IP6;
10267         optp->name = EXPER_IP6_GROUP_MEMBERSHIP;
10268 
10269         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10270         ill = ILL_START_WALK_V6(&ctx, ipst);
10271         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10272                 /* Make sure the ill isn't going away. */
10273                 if (!ill_check_and_refhold(ill))
10274                         continue;
10275                 rw_exit(&ipst->ips_ill_g_lock);
10276                 /*
10277                  * Normally we don't have any members on under IPMP interfaces.
10278                  * We report them as a debugging aid.
10279                  */
10280                 rw_enter(&ill->ill_mcast_lock, RW_READER);
10281                 ipm6.ipv6GroupMemberIfIndex = ill->ill_phyint->phyint_ifindex;
10282                 for (ilm = ill->ill_ilm; ilm; ilm = ilm->ilm_next) {
10283                         if (ilm->ilm_zoneid != zoneid &&
10284                             ilm->ilm_zoneid != ALL_ZONES)
10285                                 continue;       /* not this zone */
10286                         ipm6.ipv6GroupMemberAddress = ilm->ilm_v6addr;
10287                         ipm6.ipv6GroupMemberRefCnt = ilm->ilm_refcnt;
10288                         ipm6.ipv6GroupMemberFilterMode = ilm->ilm_fmode;
10289                         if (!snmp_append_data2(mpctl->b_cont,
10290                             &mp_tail,
10291                             (char *)&ipm6, (int)sizeof (ipm6))) {
10292                                 ip1dbg(("ip_snmp_get_mib2_ip6_group: "
10293                                     "failed to allocate %u bytes\n",
10294                                     (uint_t)sizeof (ipm6)));
10295                         }
10296                 }
10297                 rw_exit(&ill->ill_mcast_lock);
10298                 ill_refrele(ill);
10299                 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10300         }
10301         rw_exit(&ipst->ips_ill_g_lock);
10302 
10303         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10304         ip3dbg(("ip_snmp_get: level %d, name %d, len %d\n",
10305             (int)optp->level, (int)optp->name, (int)optp->len));
10306         qreply(q, mpctl);
10307         return (mp2ctl);
10308 }
10309 
10310 /* IP multicast filtered sources */
10311 static mblk_t *
10312 ip_snmp_get_mib2_ip_group_src(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10313 {
10314         struct opthdr           *optp;
10315         mblk_t                  *mp2ctl;
10316         ill_t                   *ill;
10317         ipif_t                  *ipif;
10318         ilm_t                   *ilm;
10319         ip_grpsrc_t             ips;
10320         mblk_t                  *mp_tail = NULL;
10321         ill_walk_context_t      ctx;
10322         zoneid_t                zoneid;
10323         int                     i;
10324         slist_t                 *sl;
10325 
10326         /*
10327          * make a copy of the original message
10328          */
10329         mp2ctl = copymsg(mpctl);
10330         zoneid = Q_TO_CONN(q)->conn_zoneid;
10331 
10332         /* ipGroupSource table */
10333         optp = (struct opthdr *)&mpctl->b_rptr[
10334             sizeof (struct T_optmgmt_ack)];
10335         optp->level = MIB2_IP;
10336         optp->name = EXPER_IP_GROUP_SOURCES;
10337 
10338         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10339         ill = ILL_START_WALK_V4(&ctx, ipst);
10340         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10341                 /* Make sure the ill isn't going away. */
10342                 if (!ill_check_and_refhold(ill))
10343                         continue;
10344                 rw_exit(&ipst->ips_ill_g_lock);
10345                 rw_enter(&ill->ill_mcast_lock, RW_READER);
10346                 for (ilm = ill->ill_ilm; ilm; ilm = ilm->ilm_next) {
10347                         sl = ilm->ilm_filter;
10348                         if (ilm->ilm_zoneid != zoneid &&
10349                             ilm->ilm_zoneid != ALL_ZONES)
10350                                 continue;
10351                         if (SLIST_IS_EMPTY(sl))
10352                                 continue;
10353 
10354                         /* Is there an ipif for ilm_ifaddr? */
10355                         for (ipif = ill->ill_ipif; ipif != NULL;
10356                             ipif = ipif->ipif_next) {
10357                                 if (!IPIF_IS_CONDEMNED(ipif) &&
10358                                     ipif->ipif_lcl_addr == ilm->ilm_ifaddr &&
10359                                     ilm->ilm_ifaddr != INADDR_ANY)
10360                                         break;
10361                         }
10362                         if (ipif != NULL) {
10363                                 ipif_get_name(ipif,
10364                                     ips.ipGroupSourceIfIndex.o_bytes,
10365                                     OCTET_LENGTH);
10366                         } else {
10367                                 ill_get_name(ill,
10368                                     ips.ipGroupSourceIfIndex.o_bytes,
10369                                     OCTET_LENGTH);
10370                         }
10371                         ips.ipGroupSourceIfIndex.o_length =
10372                             mi_strlen(ips.ipGroupSourceIfIndex.o_bytes);
10373 
10374                         ips.ipGroupSourceGroup = ilm->ilm_addr;
10375                         for (i = 0; i < sl->sl_numsrc; i++) {
10376                                 if (!IN6_IS_ADDR_V4MAPPED(&sl->sl_addr[i]))
10377                                         continue;
10378                                 IN6_V4MAPPED_TO_IPADDR(&sl->sl_addr[i],
10379                                     ips.ipGroupSourceAddress);
10380                                 if (snmp_append_data2(mpctl->b_cont, &mp_tail,
10381                                     (char *)&ips, (int)sizeof (ips)) == 0) {
10382                                         ip1dbg(("ip_snmp_get_mib2_ip_group_src:"
10383                                             " failed to allocate %u bytes\n",
10384                                             (uint_t)sizeof (ips)));
10385                                 }
10386                         }
10387                 }
10388                 rw_exit(&ill->ill_mcast_lock);
10389                 ill_refrele(ill);
10390                 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10391         }
10392         rw_exit(&ipst->ips_ill_g_lock);
10393         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10394         ip3dbg(("ip_snmp_get: level %d, name %d, len %d\n",
10395             (int)optp->level, (int)optp->name, (int)optp->len));
10396         qreply(q, mpctl);
10397         return (mp2ctl);
10398 }
10399 
10400 /* IPv6 multicast filtered sources. */
10401 static mblk_t *
10402 ip_snmp_get_mib2_ip6_group_src(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10403 {
10404         struct opthdr           *optp;
10405         mblk_t                  *mp2ctl;
10406         ill_t                   *ill;
10407         ilm_t                   *ilm;
10408         ipv6_grpsrc_t           ips6;
10409         mblk_t                  *mp_tail = NULL;
10410         ill_walk_context_t      ctx;
10411         zoneid_t                zoneid;
10412         int                     i;
10413         slist_t                 *sl;
10414 
10415         /*
10416          * make a copy of the original message
10417          */
10418         mp2ctl = copymsg(mpctl);
10419         zoneid = Q_TO_CONN(q)->conn_zoneid;
10420 
10421         /* ip6GroupMember table */
10422         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10423         optp->level = MIB2_IP6;
10424         optp->name = EXPER_IP6_GROUP_SOURCES;
10425 
10426         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10427         ill = ILL_START_WALK_V6(&ctx, ipst);
10428         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10429                 /* Make sure the ill isn't going away. */
10430                 if (!ill_check_and_refhold(ill))
10431                         continue;
10432                 rw_exit(&ipst->ips_ill_g_lock);
10433                 /*
10434                  * Normally we don't have any members on under IPMP interfaces.
10435                  * We report them as a debugging aid.
10436                  */
10437                 rw_enter(&ill->ill_mcast_lock, RW_READER);
10438                 ips6.ipv6GroupSourceIfIndex = ill->ill_phyint->phyint_ifindex;
10439                 for (ilm = ill->ill_ilm; ilm; ilm = ilm->ilm_next) {
10440                         sl = ilm->ilm_filter;
10441                         if (ilm->ilm_zoneid != zoneid &&
10442                             ilm->ilm_zoneid != ALL_ZONES)
10443                                 continue;
10444                         if (SLIST_IS_EMPTY(sl))
10445                                 continue;
10446                         ips6.ipv6GroupSourceGroup = ilm->ilm_v6addr;
10447                         for (i = 0; i < sl->sl_numsrc; i++) {
10448                                 ips6.ipv6GroupSourceAddress = sl->sl_addr[i];
10449                                 if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10450                                     (char *)&ips6, (int)sizeof (ips6))) {
10451                                         ip1dbg(("ip_snmp_get_mib2_ip6_"
10452                                             "group_src: failed to allocate "
10453                                             "%u bytes\n",
10454                                             (uint_t)sizeof (ips6)));
10455                                 }
10456                         }
10457                 }
10458                 rw_exit(&ill->ill_mcast_lock);
10459                 ill_refrele(ill);
10460                 rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10461         }
10462         rw_exit(&ipst->ips_ill_g_lock);
10463 
10464         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10465         ip3dbg(("ip_snmp_get: level %d, name %d, len %d\n",
10466             (int)optp->level, (int)optp->name, (int)optp->len));
10467         qreply(q, mpctl);
10468         return (mp2ctl);
10469 }
10470 
10471 /* Multicast routing virtual interface table. */
10472 static mblk_t *
10473 ip_snmp_get_mib2_virt_multi(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10474 {
10475         struct opthdr           *optp;
10476         mblk_t                  *mp2ctl;
10477 
10478         /*
10479          * make a copy of the original message
10480          */
10481         mp2ctl = copymsg(mpctl);
10482 
10483         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10484         optp->level = EXPER_DVMRP;
10485         optp->name = EXPER_DVMRP_VIF;
10486         if (!ip_mroute_vif(mpctl->b_cont, ipst)) {
10487                 ip0dbg(("ip_mroute_vif: failed\n"));
10488         }
10489         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10490         ip3dbg(("ip_snmp_get_mib2_virt_multi: level %d, name %d, len %d\n",
10491             (int)optp->level, (int)optp->name, (int)optp->len));
10492         qreply(q, mpctl);
10493         return (mp2ctl);
10494 }
10495 
10496 /* Multicast routing table. */
10497 static mblk_t *
10498 ip_snmp_get_mib2_multi_rtable(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10499 {
10500         struct opthdr           *optp;
10501         mblk_t                  *mp2ctl;
10502 
10503         /*
10504          * make a copy of the original message
10505          */
10506         mp2ctl = copymsg(mpctl);
10507 
10508         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10509         optp->level = EXPER_DVMRP;
10510         optp->name = EXPER_DVMRP_MRT;
10511         if (!ip_mroute_mrt(mpctl->b_cont, ipst)) {
10512                 ip0dbg(("ip_mroute_mrt: failed\n"));
10513         }
10514         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10515         ip3dbg(("ip_snmp_get_mib2_multi_rtable: level %d, name %d, len %d\n",
10516             (int)optp->level, (int)optp->name, (int)optp->len));
10517         qreply(q, mpctl);
10518         return (mp2ctl);
10519 }
10520 
10521 /*
10522  * Return ipRouteEntryTable, ipNetToMediaEntryTable, and ipRouteAttributeTable
10523  * in one IRE walk.
10524  */
10525 static mblk_t *
10526 ip_snmp_get_mib2_ip_route_media(queue_t *q, mblk_t *mpctl, int level,
10527     ip_stack_t *ipst)
10528 {
10529         struct opthdr   *optp;
10530         mblk_t          *mp2ctl;        /* Returned */
10531         mblk_t          *mp3ctl;        /* nettomedia */
10532         mblk_t          *mp4ctl;        /* routeattrs */
10533         iproutedata_t   ird;
10534         zoneid_t        zoneid;
10535 
10536         /*
10537          * make copies of the original message
10538          *      - mp2ctl is returned unchanged to the caller for his use
10539          *      - mpctl is sent upstream as ipRouteEntryTable
10540          *      - mp3ctl is sent upstream as ipNetToMediaEntryTable
10541          *      - mp4ctl is sent upstream as ipRouteAttributeTable
10542          */
10543         mp2ctl = copymsg(mpctl);
10544         mp3ctl = copymsg(mpctl);
10545         mp4ctl = copymsg(mpctl);
10546         if (mp3ctl == NULL || mp4ctl == NULL) {
10547                 freemsg(mp4ctl);
10548                 freemsg(mp3ctl);
10549                 freemsg(mp2ctl);
10550                 freemsg(mpctl);
10551                 return (NULL);
10552         }
10553 
10554         bzero(&ird, sizeof (ird));
10555 
10556         ird.ird_route.lp_head = mpctl->b_cont;
10557         ird.ird_netmedia.lp_head = mp3ctl->b_cont;
10558         ird.ird_attrs.lp_head = mp4ctl->b_cont;
10559         /*
10560          * If the level has been set the special EXPER_IP_AND_ALL_IRES value,
10561          * then also include ire_testhidden IREs and IRE_IF_CLONE.  This is
10562          * intended a temporary solution until a proper MIB API is provided
10563          * that provides complete filtering/caller-opt-in.
10564          */
10565         if (level == EXPER_IP_AND_ALL_IRES)
10566                 ird.ird_flags |= IRD_REPORT_ALL;
10567 
10568         zoneid = Q_TO_CONN(q)->conn_zoneid;
10569         ire_walk_v4(ip_snmp_get2_v4, &ird, zoneid, ipst);
10570 
10571         /* ipRouteEntryTable in mpctl */
10572         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10573         optp->level = MIB2_IP;
10574         optp->name = MIB2_IP_ROUTE;
10575         optp->len = msgdsize(ird.ird_route.lp_head);
10576         ip3dbg(("ip_snmp_get_mib2_ip_route_media: level %d, name %d, len %d\n",
10577             (int)optp->level, (int)optp->name, (int)optp->len));
10578         qreply(q, mpctl);
10579 
10580         /* ipNetToMediaEntryTable in mp3ctl */
10581         ncec_walk(NULL, ip_snmp_get2_v4_media, &ird, ipst);
10582 
10583         optp = (struct opthdr *)&mp3ctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10584         optp->level = MIB2_IP;
10585         optp->name = MIB2_IP_MEDIA;
10586         optp->len = msgdsize(ird.ird_netmedia.lp_head);
10587         ip3dbg(("ip_snmp_get_mib2_ip_route_media: level %d, name %d, len %d\n",
10588             (int)optp->level, (int)optp->name, (int)optp->len));
10589         qreply(q, mp3ctl);
10590 
10591         /* ipRouteAttributeTable in mp4ctl */
10592         optp = (struct opthdr *)&mp4ctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10593         optp->level = MIB2_IP;
10594         optp->name = EXPER_IP_RTATTR;
10595         optp->len = msgdsize(ird.ird_attrs.lp_head);
10596         ip3dbg(("ip_snmp_get_mib2_ip_route_media: level %d, name %d, len %d\n",
10597             (int)optp->level, (int)optp->name, (int)optp->len));
10598         if (optp->len == 0)
10599                 freemsg(mp4ctl);
10600         else
10601                 qreply(q, mp4ctl);
10602 
10603         return (mp2ctl);
10604 }
10605 
10606 /*
10607  * Return ipv6RouteEntryTable and ipv6RouteAttributeTable in one IRE walk, and
10608  * ipv6NetToMediaEntryTable in an NDP walk.
10609  */
10610 static mblk_t *
10611 ip_snmp_get_mib2_ip6_route_media(queue_t *q, mblk_t *mpctl, int level,
10612     ip_stack_t *ipst)
10613 {
10614         struct opthdr   *optp;
10615         mblk_t          *mp2ctl;        /* Returned */
10616         mblk_t          *mp3ctl;        /* nettomedia */
10617         mblk_t          *mp4ctl;        /* routeattrs */
10618         iproutedata_t   ird;
10619         zoneid_t        zoneid;
10620 
10621         /*
10622          * make copies of the original message
10623          *      - mp2ctl is returned unchanged to the caller for his use
10624          *      - mpctl is sent upstream as ipv6RouteEntryTable
10625          *      - mp3ctl is sent upstream as ipv6NetToMediaEntryTable
10626          *      - mp4ctl is sent upstream as ipv6RouteAttributeTable
10627          */
10628         mp2ctl = copymsg(mpctl);
10629         mp3ctl = copymsg(mpctl);
10630         mp4ctl = copymsg(mpctl);
10631         if (mp3ctl == NULL || mp4ctl == NULL) {
10632                 freemsg(mp4ctl);
10633                 freemsg(mp3ctl);
10634                 freemsg(mp2ctl);
10635                 freemsg(mpctl);
10636                 return (NULL);
10637         }
10638 
10639         bzero(&ird, sizeof (ird));
10640 
10641         ird.ird_route.lp_head = mpctl->b_cont;
10642         ird.ird_netmedia.lp_head = mp3ctl->b_cont;
10643         ird.ird_attrs.lp_head = mp4ctl->b_cont;
10644         /*
10645          * If the level has been set the special EXPER_IP_AND_ALL_IRES value,
10646          * then also include ire_testhidden IREs and IRE_IF_CLONE.  This is
10647          * intended a temporary solution until a proper MIB API is provided
10648          * that provides complete filtering/caller-opt-in.
10649          */
10650         if (level == EXPER_IP_AND_ALL_IRES)
10651                 ird.ird_flags |= IRD_REPORT_ALL;
10652 
10653         zoneid = Q_TO_CONN(q)->conn_zoneid;
10654         ire_walk_v6(ip_snmp_get2_v6_route, &ird, zoneid, ipst);
10655 
10656         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10657         optp->level = MIB2_IP6;
10658         optp->name = MIB2_IP6_ROUTE;
10659         optp->len = msgdsize(ird.ird_route.lp_head);
10660         ip3dbg(("ip_snmp_get_mib2_ip6_route_media: level %d, name %d, len %d\n",
10661             (int)optp->level, (int)optp->name, (int)optp->len));
10662         qreply(q, mpctl);
10663 
10664         /* ipv6NetToMediaEntryTable in mp3ctl */
10665         ncec_walk(NULL, ip_snmp_get2_v6_media, &ird, ipst);
10666 
10667         optp = (struct opthdr *)&mp3ctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10668         optp->level = MIB2_IP6;
10669         optp->name = MIB2_IP6_MEDIA;
10670         optp->len = msgdsize(ird.ird_netmedia.lp_head);
10671         ip3dbg(("ip_snmp_get_mib2_ip6_route_media: level %d, name %d, len %d\n",
10672             (int)optp->level, (int)optp->name, (int)optp->len));
10673         qreply(q, mp3ctl);
10674 
10675         /* ipv6RouteAttributeTable in mp4ctl */
10676         optp = (struct opthdr *)&mp4ctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10677         optp->level = MIB2_IP6;
10678         optp->name = EXPER_IP_RTATTR;
10679         optp->len = msgdsize(ird.ird_attrs.lp_head);
10680         ip3dbg(("ip_snmp_get_mib2_ip6_route_media: level %d, name %d, len %d\n",
10681             (int)optp->level, (int)optp->name, (int)optp->len));
10682         if (optp->len == 0)
10683                 freemsg(mp4ctl);
10684         else
10685                 qreply(q, mp4ctl);
10686 
10687         return (mp2ctl);
10688 }
10689 
10690 /*
10691  * IPv6 mib: One per ill
10692  */
10693 static mblk_t *
10694 ip_snmp_get_mib2_ip6(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst,
10695     boolean_t legacy_req)
10696 {
10697         struct opthdr           *optp;
10698         mblk_t                  *mp2ctl;
10699         ill_t                   *ill;
10700         ill_walk_context_t      ctx;
10701         mblk_t                  *mp_tail = NULL;
10702         mib2_ipv6AddrEntry_t    mae6;
10703         mib2_ipIfStatsEntry_t   *ise;
10704         size_t                  ise_size, iae_size;
10705 
10706         /*
10707          * Make a copy of the original message
10708          */
10709         mp2ctl = copymsg(mpctl);
10710 
10711         /* fixed length IPv6 structure ... */
10712 
10713         if (legacy_req) {
10714                 ise_size = LEGACY_MIB_SIZE(&ipst->ips_ip6_mib,
10715                     mib2_ipIfStatsEntry_t);
10716                 iae_size = LEGACY_MIB_SIZE(&mae6, mib2_ipv6AddrEntry_t);
10717         } else {
10718                 ise_size = sizeof (mib2_ipIfStatsEntry_t);
10719                 iae_size = sizeof (mib2_ipv6AddrEntry_t);
10720         }
10721 
10722         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10723         optp->level = MIB2_IP6;
10724         optp->name = 0;
10725         /* Include "unknown interface" ip6_mib */
10726         ipst->ips_ip6_mib.ipIfStatsIPVersion = MIB2_INETADDRESSTYPE_ipv6;
10727         ipst->ips_ip6_mib.ipIfStatsIfIndex =
10728             MIB2_UNKNOWN_INTERFACE; /* Flag to netstat */
10729         SET_MIB(ipst->ips_ip6_mib.ipIfStatsForwarding,
10730             ipst->ips_ipv6_forwarding ? 1 : 2);
10731         SET_MIB(ipst->ips_ip6_mib.ipIfStatsDefaultHopLimit,
10732             ipst->ips_ipv6_def_hops);
10733         SET_MIB(ipst->ips_ip6_mib.ipIfStatsEntrySize,
10734             sizeof (mib2_ipIfStatsEntry_t));
10735         SET_MIB(ipst->ips_ip6_mib.ipIfStatsAddrEntrySize,
10736             sizeof (mib2_ipv6AddrEntry_t));
10737         SET_MIB(ipst->ips_ip6_mib.ipIfStatsRouteEntrySize,
10738             sizeof (mib2_ipv6RouteEntry_t));
10739         SET_MIB(ipst->ips_ip6_mib.ipIfStatsNetToMediaEntrySize,
10740             sizeof (mib2_ipv6NetToMediaEntry_t));
10741         SET_MIB(ipst->ips_ip6_mib.ipIfStatsMemberEntrySize,
10742             sizeof (ipv6_member_t));
10743         SET_MIB(ipst->ips_ip6_mib.ipIfStatsGroupSourceEntrySize,
10744             sizeof (ipv6_grpsrc_t));
10745 
10746         /*
10747          * Synchronize 64- and 32-bit counters
10748          */
10749         SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsInReceives,
10750             ipIfStatsHCInReceives);
10751         SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsInDelivers,
10752             ipIfStatsHCInDelivers);
10753         SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsOutRequests,
10754             ipIfStatsHCOutRequests);
10755         SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsOutForwDatagrams,
10756             ipIfStatsHCOutForwDatagrams);
10757         SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsOutMcastPkts,
10758             ipIfStatsHCOutMcastPkts);
10759         SYNC32_MIB(&ipst->ips_ip6_mib, ipIfStatsInMcastPkts,
10760             ipIfStatsHCInMcastPkts);
10761 
10762         if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10763             (char *)&ipst->ips_ip6_mib, (int)ise_size)) {
10764                 ip1dbg(("ip_snmp_get_mib2_ip6: failed to allocate %u bytes\n",
10765                     (uint_t)ise_size));
10766         } else if (legacy_req) {
10767                 /* Adjust the EntrySize fields for legacy requests. */
10768                 ise =
10769                     (mib2_ipIfStatsEntry_t *)(mp_tail->b_wptr - (int)ise_size);
10770                 SET_MIB(ise->ipIfStatsEntrySize, ise_size);
10771                 SET_MIB(ise->ipIfStatsAddrEntrySize, iae_size);
10772         }
10773 
10774         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10775         ill = ILL_START_WALK_V6(&ctx, ipst);
10776         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10777                 ill->ill_ip_mib->ipIfStatsIfIndex =
10778                     ill->ill_phyint->phyint_ifindex;
10779                 SET_MIB(ill->ill_ip_mib->ipIfStatsForwarding,
10780                     ipst->ips_ipv6_forwarding ? 1 : 2);
10781                 SET_MIB(ill->ill_ip_mib->ipIfStatsDefaultHopLimit,
10782                     ill->ill_max_hops);
10783 
10784                 /*
10785                  * Synchronize 64- and 32-bit counters
10786                  */
10787                 SYNC32_MIB(ill->ill_ip_mib, ipIfStatsInReceives,
10788                     ipIfStatsHCInReceives);
10789                 SYNC32_MIB(ill->ill_ip_mib, ipIfStatsInDelivers,
10790                     ipIfStatsHCInDelivers);
10791                 SYNC32_MIB(ill->ill_ip_mib, ipIfStatsOutRequests,
10792                     ipIfStatsHCOutRequests);
10793                 SYNC32_MIB(ill->ill_ip_mib, ipIfStatsOutForwDatagrams,
10794                     ipIfStatsHCOutForwDatagrams);
10795                 SYNC32_MIB(ill->ill_ip_mib, ipIfStatsOutMcastPkts,
10796                     ipIfStatsHCOutMcastPkts);
10797                 SYNC32_MIB(ill->ill_ip_mib, ipIfStatsInMcastPkts,
10798                     ipIfStatsHCInMcastPkts);
10799 
10800                 if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10801                     (char *)ill->ill_ip_mib, (int)ise_size)) {
10802                         ip1dbg(("ip_snmp_get_mib2_ip6: failed to allocate "
10803                         "%u bytes\n", (uint_t)ise_size));
10804                 } else if (legacy_req) {
10805                         /* Adjust the EntrySize fields for legacy requests. */
10806                         ise = (mib2_ipIfStatsEntry_t *)(mp_tail->b_wptr -
10807                             (int)ise_size);
10808                         SET_MIB(ise->ipIfStatsEntrySize, ise_size);
10809                         SET_MIB(ise->ipIfStatsAddrEntrySize, iae_size);
10810                 }
10811         }
10812         rw_exit(&ipst->ips_ill_g_lock);
10813 
10814         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10815         ip3dbg(("ip_snmp_get_mib2_ip6: level %d, name %d, len %d\n",
10816             (int)optp->level, (int)optp->name, (int)optp->len));
10817         qreply(q, mpctl);
10818         return (mp2ctl);
10819 }
10820 
10821 /*
10822  * ICMPv6 mib: One per ill
10823  */
10824 static mblk_t *
10825 ip_snmp_get_mib2_icmp6(queue_t *q, mblk_t *mpctl, ip_stack_t *ipst)
10826 {
10827         struct opthdr           *optp;
10828         mblk_t                  *mp2ctl;
10829         ill_t                   *ill;
10830         ill_walk_context_t      ctx;
10831         mblk_t                  *mp_tail = NULL;
10832         /*
10833          * Make a copy of the original message
10834          */
10835         mp2ctl = copymsg(mpctl);
10836 
10837         /* fixed length ICMPv6 structure ... */
10838 
10839         optp = (struct opthdr *)&mpctl->b_rptr[sizeof (struct T_optmgmt_ack)];
10840         optp->level = MIB2_ICMP6;
10841         optp->name = 0;
10842         /* Include "unknown interface" icmp6_mib */
10843         ipst->ips_icmp6_mib.ipv6IfIcmpIfIndex =
10844             MIB2_UNKNOWN_INTERFACE; /* netstat flag */
10845         ipst->ips_icmp6_mib.ipv6IfIcmpEntrySize =
10846             sizeof (mib2_ipv6IfIcmpEntry_t);
10847         if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10848             (char *)&ipst->ips_icmp6_mib,
10849             (int)sizeof (ipst->ips_icmp6_mib))) {
10850                 ip1dbg(("ip_snmp_get_mib2_icmp6: failed to allocate %u bytes\n",
10851                     (uint_t)sizeof (ipst->ips_icmp6_mib)));
10852         }
10853 
10854         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
10855         ill = ILL_START_WALK_V6(&ctx, ipst);
10856         for (; ill != NULL; ill = ill_next(&ctx, ill)) {
10857                 ill->ill_icmp6_mib->ipv6IfIcmpIfIndex =
10858                     ill->ill_phyint->phyint_ifindex;
10859                 if (!snmp_append_data2(mpctl->b_cont, &mp_tail,
10860                     (char *)ill->ill_icmp6_mib,
10861                     (int)sizeof (*ill->ill_icmp6_mib))) {
10862                         ip1dbg(("ip_snmp_get_mib2_icmp6: failed to allocate "
10863                             "%u bytes\n",
10864                             (uint_t)sizeof (*ill->ill_icmp6_mib)));
10865                 }
10866         }
10867         rw_exit(&ipst->ips_ill_g_lock);
10868 
10869         optp->len = (t_uscalar_t)msgdsize(mpctl->b_cont);
10870         ip3dbg(("ip_snmp_get_mib2_icmp6: level %d, name %d, len %d\n",
10871             (int)optp->level, (int)optp->name, (int)optp->len));
10872         qreply(q, mpctl);
10873         return (mp2ctl);
10874 }
10875 
10876 /*
10877  * ire_walk routine to create both ipRouteEntryTable and
10878  * ipRouteAttributeTable in one IRE walk
10879  */
10880 static void
10881 ip_snmp_get2_v4(ire_t *ire, iproutedata_t *ird)
10882 {
10883         ill_t                           *ill;
10884         mib2_ipRouteEntry_t             *re;
10885         mib2_ipAttributeEntry_t         iaes;
10886         tsol_ire_gw_secattr_t           *attrp;
10887         tsol_gc_t                       *gc = NULL;
10888         tsol_gcgrp_t                    *gcgrp = NULL;
10889         ip_stack_t                      *ipst = ire->ire_ipst;
10890 
10891         ASSERT(ire->ire_ipversion == IPV4_VERSION);
10892 
10893         if (!(ird->ird_flags & IRD_REPORT_ALL)) {
10894                 if (ire->ire_testhidden)
10895                         return;
10896                 if (ire->ire_type & IRE_IF_CLONE)
10897                         return;
10898         }
10899 
10900         if ((re = kmem_zalloc(sizeof (*re), KM_NOSLEEP)) == NULL)
10901                 return;
10902 
10903         if ((attrp = ire->ire_gw_secattr) != NULL) {
10904                 mutex_enter(&attrp->igsa_lock);
10905                 if ((gc = attrp->igsa_gc) != NULL) {
10906                         gcgrp = gc->gc_grp;
10907                         ASSERT(gcgrp != NULL);
10908                         rw_enter(&gcgrp->gcgrp_rwlock, RW_READER);
10909                 }
10910                 mutex_exit(&attrp->igsa_lock);
10911         }
10912         /*
10913          * Return all IRE types for route table... let caller pick and choose
10914          */
10915         re->ipRouteDest = ire->ire_addr;
10916         ill = ire->ire_ill;
10917         re->ipRouteIfIndex.o_length = 0;
10918         if (ill != NULL) {
10919                 ill_get_name(ill, re->ipRouteIfIndex.o_bytes, OCTET_LENGTH);
10920                 re->ipRouteIfIndex.o_length =
10921                     mi_strlen(re->ipRouteIfIndex.o_bytes);
10922         }
10923         re->ipRouteMetric1 = -1;
10924         re->ipRouteMetric2 = -1;
10925         re->ipRouteMetric3 = -1;
10926         re->ipRouteMetric4 = -1;
10927 
10928         re->ipRouteNextHop = ire->ire_gateway_addr;
10929         /* indirect(4), direct(3), or invalid(2) */
10930         if (ire->ire_flags & (RTF_REJECT | RTF_BLACKHOLE))
10931                 re->ipRouteType = 2;
10932         else if (ire->ire_type & IRE_ONLINK)
10933                 re->ipRouteType = 3;
10934         else
10935                 re->ipRouteType = 4;
10936 
10937         re->ipRouteProto = -1;
10938         re->ipRouteAge = gethrestime_sec() - ire->ire_create_time;
10939         re->ipRouteMask = ire->ire_mask;
10940         re->ipRouteMetric5 = -1;
10941         re->ipRouteInfo.re_max_frag = ire->ire_metrics.iulp_mtu;
10942         if (ire->ire_ill != NULL && re->ipRouteInfo.re_max_frag == 0)
10943                 re->ipRouteInfo.re_max_frag = ire->ire_ill->ill_mtu;
10944 
10945         re->ipRouteInfo.re_frag_flag = 0;
10946         re->ipRouteInfo.re_rtt               = 0;
10947         re->ipRouteInfo.re_src_addr  = 0;
10948         re->ipRouteInfo.re_ref               = ire->ire_refcnt;
10949         re->ipRouteInfo.re_obpkt     = ire->ire_ob_pkt_count;
10950         re->ipRouteInfo.re_ibpkt     = ire->ire_ib_pkt_count;
10951         re->ipRouteInfo.re_flags     = ire->ire_flags;
10952 
10953         /* Add the IRE_IF_CLONE's counters to their parent IRE_INTERFACE */
10954         if (ire->ire_type & IRE_INTERFACE) {
10955                 ire_t *child;
10956 
10957                 rw_enter(&ipst->ips_ire_dep_lock, RW_READER);
10958                 child = ire->ire_dep_children;
10959                 while (child != NULL) {
10960                         re->ipRouteInfo.re_obpkt += child->ire_ob_pkt_count;
10961                         re->ipRouteInfo.re_ibpkt += child->ire_ib_pkt_count;
10962                         child = child->ire_dep_sib_next;
10963                 }
10964                 rw_exit(&ipst->ips_ire_dep_lock);
10965         }
10966 
10967         if (ire->ire_flags & RTF_DYNAMIC) {
10968                 re->ipRouteInfo.re_ire_type  = IRE_HOST_REDIRECT;
10969         } else {
10970                 re->ipRouteInfo.re_ire_type  = ire->ire_type;
10971         }
10972 
10973         if (!snmp_append_data2(ird->ird_route.lp_head, &ird->ird_route.lp_tail,
10974             (char *)re, (int)sizeof (*re))) {
10975                 ip1dbg(("ip_snmp_get2_v4: failed to allocate %u bytes\n",
10976                     (uint_t)sizeof (*re)));
10977         }
10978 
10979         if (gc != NULL) {
10980                 iaes.iae_routeidx = ird->ird_idx;
10981                 iaes.iae_doi = gc->gc_db->gcdb_doi;
10982                 iaes.iae_slrange = gc->gc_db->gcdb_slrange;
10983 
10984                 if (!snmp_append_data2(ird->ird_attrs.lp_head,
10985                     &ird->ird_attrs.lp_tail, (char *)&iaes, sizeof (iaes))) {
10986                         ip1dbg(("ip_snmp_get2_v4: failed to allocate %u "
10987                             "bytes\n", (uint_t)sizeof (iaes)));
10988                 }
10989         }
10990 
10991         /* bump route index for next pass */
10992         ird->ird_idx++;
10993 
10994         kmem_free(re, sizeof (*re));
10995         if (gcgrp != NULL)
10996                 rw_exit(&gcgrp->gcgrp_rwlock);
10997 }
10998 
10999 /*
11000  * ire_walk routine to create ipv6RouteEntryTable and ipRouteEntryTable.
11001  */
11002 static void
11003 ip_snmp_get2_v6_route(ire_t *ire, iproutedata_t *ird)
11004 {
11005         ill_t                           *ill;
11006         mib2_ipv6RouteEntry_t           *re;
11007         mib2_ipAttributeEntry_t         iaes;
11008         tsol_ire_gw_secattr_t           *attrp;
11009         tsol_gc_t                       *gc = NULL;
11010         tsol_gcgrp_t                    *gcgrp = NULL;
11011         ip_stack_t                      *ipst = ire->ire_ipst;
11012 
11013         ASSERT(ire->ire_ipversion == IPV6_VERSION);
11014 
11015         if (!(ird->ird_flags & IRD_REPORT_ALL)) {
11016                 if (ire->ire_testhidden)
11017                         return;
11018                 if (ire->ire_type & IRE_IF_CLONE)
11019                         return;
11020         }
11021 
11022         if ((re = kmem_zalloc(sizeof (*re), KM_NOSLEEP)) == NULL)
11023                 return;
11024 
11025         if ((attrp = ire->ire_gw_secattr) != NULL) {
11026                 mutex_enter(&attrp->igsa_lock);
11027                 if ((gc = attrp->igsa_gc) != NULL) {
11028                         gcgrp = gc->gc_grp;
11029                         ASSERT(gcgrp != NULL);
11030                         rw_enter(&gcgrp->gcgrp_rwlock, RW_READER);
11031                 }
11032                 mutex_exit(&attrp->igsa_lock);
11033         }
11034         /*
11035          * Return all IRE types for route table... let caller pick and choose
11036          */
11037         re->ipv6RouteDest = ire->ire_addr_v6;
11038         re->ipv6RoutePfxLength = ip_mask_to_plen_v6(&ire->ire_mask_v6);
11039         re->ipv6RouteIndex = 0;      /* Unique when multiple with same dest/plen */
11040         re->ipv6RouteIfIndex.o_length = 0;
11041         ill = ire->ire_ill;
11042         if (ill != NULL) {
11043                 ill_get_name(ill, re->ipv6RouteIfIndex.o_bytes, OCTET_LENGTH);
11044                 re->ipv6RouteIfIndex.o_length =
11045                     mi_strlen(re->ipv6RouteIfIndex.o_bytes);
11046         }
11047 
11048         ASSERT(!(ire->ire_type & IRE_BROADCAST));
11049 
11050         mutex_enter(&ire->ire_lock);
11051         re->ipv6RouteNextHop = ire->ire_gateway_addr_v6;
11052         mutex_exit(&ire->ire_lock);
11053 
11054         /* remote(4), local(3), or discard(2) */
11055         if (ire->ire_flags & (RTF_REJECT | RTF_BLACKHOLE))
11056                 re->ipv6RouteType = 2;
11057         else if (ire->ire_type & IRE_ONLINK)
11058                 re->ipv6RouteType = 3;
11059         else
11060                 re->ipv6RouteType = 4;
11061 
11062         re->ipv6RouteProtocol        = -1;
11063         re->ipv6RoutePolicy  = 0;
11064         re->ipv6RouteAge     = gethrestime_sec() - ire->ire_create_time;
11065         re->ipv6RouteNextHopRDI      = 0;
11066         re->ipv6RouteWeight  = 0;
11067         re->ipv6RouteMetric  = 0;
11068         re->ipv6RouteInfo.re_max_frag = ire->ire_metrics.iulp_mtu;
11069         if (ire->ire_ill != NULL && re->ipv6RouteInfo.re_max_frag == 0)
11070                 re->ipv6RouteInfo.re_max_frag = ire->ire_ill->ill_mtu;
11071 
11072         re->ipv6RouteInfo.re_frag_flag       = 0;
11073         re->ipv6RouteInfo.re_rtt     = 0;
11074         re->ipv6RouteInfo.re_src_addr        = ipv6_all_zeros;
11075         re->ipv6RouteInfo.re_obpkt   = ire->ire_ob_pkt_count;
11076         re->ipv6RouteInfo.re_ibpkt   = ire->ire_ib_pkt_count;
11077         re->ipv6RouteInfo.re_ref     = ire->ire_refcnt;
11078         re->ipv6RouteInfo.re_flags   = ire->ire_flags;
11079 
11080         /* Add the IRE_IF_CLONE's counters to their parent IRE_INTERFACE */
11081         if (ire->ire_type & IRE_INTERFACE) {
11082                 ire_t *child;
11083 
11084                 rw_enter(&ipst->ips_ire_dep_lock, RW_READER);
11085                 child = ire->ire_dep_children;
11086                 while (child != NULL) {
11087                         re->ipv6RouteInfo.re_obpkt += child->ire_ob_pkt_count;
11088                         re->ipv6RouteInfo.re_ibpkt += child->ire_ib_pkt_count;
11089                         child = child->ire_dep_sib_next;
11090                 }
11091                 rw_exit(&ipst->ips_ire_dep_lock);
11092         }
11093         if (ire->ire_flags & RTF_DYNAMIC) {
11094                 re->ipv6RouteInfo.re_ire_type        = IRE_HOST_REDIRECT;
11095         } else {
11096                 re->ipv6RouteInfo.re_ire_type        = ire->ire_type;
11097         }
11098 
11099         if (!snmp_append_data2(ird->ird_route.lp_head, &ird->ird_route.lp_tail,
11100             (char *)re, (int)sizeof (*re))) {
11101                 ip1dbg(("ip_snmp_get2_v6: failed to allocate %u bytes\n",
11102                     (uint_t)sizeof (*re)));
11103         }
11104 
11105         if (gc != NULL) {
11106                 iaes.iae_routeidx = ird->ird_idx;
11107                 iaes.iae_doi = gc->gc_db->gcdb_doi;
11108                 iaes.iae_slrange = gc->gc_db->gcdb_slrange;
11109 
11110                 if (!snmp_append_data2(ird->ird_attrs.lp_head,
11111                     &ird->ird_attrs.lp_tail, (char *)&iaes, sizeof (iaes))) {
11112                         ip1dbg(("ip_snmp_get2_v6: failed to allocate %u "
11113                             "bytes\n", (uint_t)sizeof (iaes)));
11114                 }
11115         }
11116 
11117         /* bump route index for next pass */
11118         ird->ird_idx++;
11119 
11120         kmem_free(re, sizeof (*re));
11121         if (gcgrp != NULL)
11122                 rw_exit(&gcgrp->gcgrp_rwlock);
11123 }
11124 
11125 /*
11126  * ncec_walk routine to create ipv6NetToMediaEntryTable
11127  */
11128 static int
11129 ip_snmp_get2_v6_media(ncec_t *ncec, iproutedata_t *ird)
11130 {
11131         ill_t                           *ill;
11132         mib2_ipv6NetToMediaEntry_t      ntme;
11133 
11134         ill = ncec->ncec_ill;
11135         /* skip arpce entries, and loopback ncec entries */
11136         if (ill->ill_isv6 == B_FALSE || ill->ill_net_type == IRE_LOOPBACK)
11137                 return (0);
11138         /*
11139          * Neighbor cache entry attached to IRE with on-link
11140          * destination.
11141          * We report all IPMP groups on ncec_ill which is normally the upper.
11142          */
11143         ntme.ipv6NetToMediaIfIndex = ill->ill_phyint->phyint_ifindex;
11144         ntme.ipv6NetToMediaNetAddress = ncec->ncec_addr;
11145         ntme.ipv6NetToMediaPhysAddress.o_length = ill->ill_phys_addr_length;
11146         if (ncec->ncec_lladdr != NULL) {
11147                 bcopy(ncec->ncec_lladdr, ntme.ipv6NetToMediaPhysAddress.o_bytes,
11148                     ntme.ipv6NetToMediaPhysAddress.o_length);
11149         }
11150         /*
11151          * Note: Returns ND_* states. Should be:
11152          * reachable(1), stale(2), delay(3), probe(4),
11153          * invalid(5), unknown(6)
11154          */
11155         ntme.ipv6NetToMediaState = ncec->ncec_state;
11156         ntme.ipv6NetToMediaLastUpdated = 0;
11157 
11158         /* other(1), dynamic(2), static(3), local(4) */
11159         if (NCE_MYADDR(ncec)) {
11160                 ntme.ipv6NetToMediaType = 4;
11161         } else if (ncec->ncec_flags & NCE_F_PUBLISH) {
11162                 ntme.ipv6NetToMediaType = 1; /* proxy */
11163         } else if (ncec->ncec_flags & NCE_F_STATIC) {
11164                 ntme.ipv6NetToMediaType = 3;
11165         } else if (ncec->ncec_flags & (NCE_F_MCAST|NCE_F_BCAST)) {
11166                 ntme.ipv6NetToMediaType = 1;
11167         } else {
11168                 ntme.ipv6NetToMediaType = 2;
11169         }
11170 
11171         if (!snmp_append_data2(ird->ird_netmedia.lp_head,
11172             &ird->ird_netmedia.lp_tail, (char *)&ntme, sizeof (ntme))) {
11173                 ip1dbg(("ip_snmp_get2_v6_media: failed to allocate %u bytes\n",
11174                     (uint_t)sizeof (ntme)));
11175         }
11176         return (0);
11177 }
11178 
11179 int
11180 nce2ace(ncec_t *ncec)
11181 {
11182         int flags = 0;
11183 
11184         if (NCE_ISREACHABLE(ncec))
11185                 flags |= ACE_F_RESOLVED;
11186         if (ncec->ncec_flags & NCE_F_AUTHORITY)
11187                 flags |= ACE_F_AUTHORITY;
11188         if (ncec->ncec_flags & NCE_F_PUBLISH)
11189                 flags |= ACE_F_PUBLISH;
11190         if ((ncec->ncec_flags & NCE_F_NONUD) != 0)
11191                 flags |= ACE_F_PERMANENT;
11192         if (NCE_MYADDR(ncec))
11193                 flags |= (ACE_F_MYADDR | ACE_F_AUTHORITY);
11194         if (ncec->ncec_flags & NCE_F_UNVERIFIED)
11195                 flags |= ACE_F_UNVERIFIED;
11196         if (ncec->ncec_flags & NCE_F_AUTHORITY)
11197                 flags |= ACE_F_AUTHORITY;
11198         if (ncec->ncec_flags & NCE_F_DELAYED)
11199                 flags |= ACE_F_DELAYED;
11200         return (flags);
11201 }
11202 
11203 /*
11204  * ncec_walk routine to create ipNetToMediaEntryTable
11205  */
11206 static int
11207 ip_snmp_get2_v4_media(ncec_t *ncec, iproutedata_t *ird)
11208 {
11209         ill_t                           *ill;
11210         mib2_ipNetToMediaEntry_t        ntme;
11211         const char                      *name = "unknown";
11212         ipaddr_t                        ncec_addr;
11213 
11214         ill = ncec->ncec_ill;
11215         if (ill->ill_isv6 || (ncec->ncec_flags & NCE_F_BCAST) ||
11216             ill->ill_net_type == IRE_LOOPBACK)
11217                 return (0);
11218 
11219         /* We report all IPMP groups on ncec_ill which is normally the upper. */
11220         name = ill->ill_name;
11221         /* Based on RFC 4293: other(1), inval(2), dyn(3), stat(4) */
11222         if (NCE_MYADDR(ncec)) {
11223                 ntme.ipNetToMediaType = 4;
11224         } else if (ncec->ncec_flags & (NCE_F_MCAST|NCE_F_BCAST|NCE_F_PUBLISH)) {
11225                 ntme.ipNetToMediaType = 1;
11226         } else {
11227                 ntme.ipNetToMediaType = 3;
11228         }
11229         ntme.ipNetToMediaIfIndex.o_length = MIN(OCTET_LENGTH, strlen(name));
11230         bcopy(name, ntme.ipNetToMediaIfIndex.o_bytes,
11231             ntme.ipNetToMediaIfIndex.o_length);
11232 
11233         IN6_V4MAPPED_TO_IPADDR(&ncec->ncec_addr, ncec_addr);
11234         bcopy(&ncec_addr, &ntme.ipNetToMediaNetAddress, sizeof (ncec_addr));
11235 
11236         ntme.ipNetToMediaInfo.ntm_mask.o_length = sizeof (ipaddr_t);
11237         ncec_addr = INADDR_BROADCAST;
11238         bcopy(&ncec_addr, ntme.ipNetToMediaInfo.ntm_mask.o_bytes,
11239             sizeof (ncec_addr));
11240         /*
11241          * map all the flags to the ACE counterpart.
11242          */
11243         ntme.ipNetToMediaInfo.ntm_flags = nce2ace(ncec);
11244 
11245         ntme.ipNetToMediaPhysAddress.o_length =
11246             MIN(OCTET_LENGTH, ill->ill_phys_addr_length);
11247 
11248         if (!NCE_ISREACHABLE(ncec))
11249                 ntme.ipNetToMediaPhysAddress.o_length = 0;
11250         else {
11251                 if (ncec->ncec_lladdr != NULL) {
11252                         bcopy(ncec->ncec_lladdr,
11253                             ntme.ipNetToMediaPhysAddress.o_bytes,
11254                             ntme.ipNetToMediaPhysAddress.o_length);
11255                 }
11256         }
11257 
11258         if (!snmp_append_data2(ird->ird_netmedia.lp_head,
11259             &ird->ird_netmedia.lp_tail, (char *)&ntme, sizeof (ntme))) {
11260                 ip1dbg(("ip_snmp_get2_v4_media: failed to allocate %u bytes\n",
11261                     (uint_t)sizeof (ntme)));
11262         }
11263         return (0);
11264 }
11265 
11266 /*
11267  * return (0) if invalid set request, 1 otherwise, including non-tcp requests
11268  */
11269 /* ARGSUSED */
11270 int
11271 ip_snmp_set(queue_t *q, int level, int name, uchar_t *ptr, int len)
11272 {
11273         switch (level) {
11274         case MIB2_IP:
11275         case MIB2_ICMP:
11276                 switch (name) {
11277                 default:
11278                         break;
11279                 }
11280                 return (1);
11281         default:
11282                 return (1);
11283         }
11284 }
11285 
11286 /*
11287  * When there exists both a 64- and 32-bit counter of a particular type
11288  * (i.e., InReceives), only the 64-bit counters are added.
11289  */
11290 void
11291 ip_mib2_add_ip_stats(mib2_ipIfStatsEntry_t *o1, mib2_ipIfStatsEntry_t *o2)
11292 {
11293         UPDATE_MIB(o1, ipIfStatsInHdrErrors, o2->ipIfStatsInHdrErrors);
11294         UPDATE_MIB(o1, ipIfStatsInTooBigErrors, o2->ipIfStatsInTooBigErrors);
11295         UPDATE_MIB(o1, ipIfStatsInNoRoutes, o2->ipIfStatsInNoRoutes);
11296         UPDATE_MIB(o1, ipIfStatsInAddrErrors, o2->ipIfStatsInAddrErrors);
11297         UPDATE_MIB(o1, ipIfStatsInUnknownProtos, o2->ipIfStatsInUnknownProtos);
11298         UPDATE_MIB(o1, ipIfStatsInTruncatedPkts, o2->ipIfStatsInTruncatedPkts);
11299         UPDATE_MIB(o1, ipIfStatsInDiscards, o2->ipIfStatsInDiscards);
11300         UPDATE_MIB(o1, ipIfStatsOutDiscards, o2->ipIfStatsOutDiscards);
11301         UPDATE_MIB(o1, ipIfStatsOutFragOKs, o2->ipIfStatsOutFragOKs);
11302         UPDATE_MIB(o1, ipIfStatsOutFragFails, o2->ipIfStatsOutFragFails);
11303         UPDATE_MIB(o1, ipIfStatsOutFragCreates, o2->ipIfStatsOutFragCreates);
11304         UPDATE_MIB(o1, ipIfStatsReasmReqds, o2->ipIfStatsReasmReqds);
11305         UPDATE_MIB(o1, ipIfStatsReasmOKs, o2->ipIfStatsReasmOKs);
11306         UPDATE_MIB(o1, ipIfStatsReasmFails, o2->ipIfStatsReasmFails);
11307         UPDATE_MIB(o1, ipIfStatsOutNoRoutes, o2->ipIfStatsOutNoRoutes);
11308         UPDATE_MIB(o1, ipIfStatsReasmDuplicates, o2->ipIfStatsReasmDuplicates);
11309         UPDATE_MIB(o1, ipIfStatsReasmPartDups, o2->ipIfStatsReasmPartDups);
11310         UPDATE_MIB(o1, ipIfStatsForwProhibits, o2->ipIfStatsForwProhibits);
11311         UPDATE_MIB(o1, udpInCksumErrs, o2->udpInCksumErrs);
11312         UPDATE_MIB(o1, udpInOverflows, o2->udpInOverflows);
11313         UPDATE_MIB(o1, rawipInOverflows, o2->rawipInOverflows);
11314         UPDATE_MIB(o1, ipIfStatsInWrongIPVersion,
11315             o2->ipIfStatsInWrongIPVersion);
11316         UPDATE_MIB(o1, ipIfStatsOutWrongIPVersion,
11317             o2->ipIfStatsInWrongIPVersion);
11318         UPDATE_MIB(o1, ipIfStatsOutSwitchIPVersion,
11319             o2->ipIfStatsOutSwitchIPVersion);
11320         UPDATE_MIB(o1, ipIfStatsHCInReceives, o2->ipIfStatsHCInReceives);
11321         UPDATE_MIB(o1, ipIfStatsHCInOctets, o2->ipIfStatsHCInOctets);
11322         UPDATE_MIB(o1, ipIfStatsHCInForwDatagrams,
11323             o2->ipIfStatsHCInForwDatagrams);
11324         UPDATE_MIB(o1, ipIfStatsHCInDelivers, o2->ipIfStatsHCInDelivers);
11325         UPDATE_MIB(o1, ipIfStatsHCOutRequests, o2->ipIfStatsHCOutRequests);
11326         UPDATE_MIB(o1, ipIfStatsHCOutForwDatagrams,
11327             o2->ipIfStatsHCOutForwDatagrams);
11328         UPDATE_MIB(o1, ipIfStatsOutFragReqds, o2->ipIfStatsOutFragReqds);
11329         UPDATE_MIB(o1, ipIfStatsHCOutTransmits, o2->ipIfStatsHCOutTransmits);
11330         UPDATE_MIB(o1, ipIfStatsHCOutOctets, o2->ipIfStatsHCOutOctets);
11331         UPDATE_MIB(o1, ipIfStatsHCInMcastPkts, o2->ipIfStatsHCInMcastPkts);
11332         UPDATE_MIB(o1, ipIfStatsHCInMcastOctets, o2->ipIfStatsHCInMcastOctets);
11333         UPDATE_MIB(o1, ipIfStatsHCOutMcastPkts, o2->ipIfStatsHCOutMcastPkts);
11334         UPDATE_MIB(o1, ipIfStatsHCOutMcastOctets,
11335             o2->ipIfStatsHCOutMcastOctets);
11336         UPDATE_MIB(o1, ipIfStatsHCInBcastPkts, o2->ipIfStatsHCInBcastPkts);
11337         UPDATE_MIB(o1, ipIfStatsHCOutBcastPkts, o2->ipIfStatsHCOutBcastPkts);
11338         UPDATE_MIB(o1, ipsecInSucceeded, o2->ipsecInSucceeded);
11339         UPDATE_MIB(o1, ipsecInFailed, o2->ipsecInFailed);
11340         UPDATE_MIB(o1, ipInCksumErrs, o2->ipInCksumErrs);
11341         UPDATE_MIB(o1, tcpInErrs, o2->tcpInErrs);
11342         UPDATE_MIB(o1, udpNoPorts, o2->udpNoPorts);
11343 }
11344 
11345 void
11346 ip_mib2_add_icmp6_stats(mib2_ipv6IfIcmpEntry_t *o1, mib2_ipv6IfIcmpEntry_t *o2)
11347 {
11348         UPDATE_MIB(o1, ipv6IfIcmpInMsgs, o2->ipv6IfIcmpInMsgs);
11349         UPDATE_MIB(o1, ipv6IfIcmpInErrors, o2->ipv6IfIcmpInErrors);
11350         UPDATE_MIB(o1, ipv6IfIcmpInDestUnreachs, o2->ipv6IfIcmpInDestUnreachs);
11351         UPDATE_MIB(o1, ipv6IfIcmpInAdminProhibs, o2->ipv6IfIcmpInAdminProhibs);
11352         UPDATE_MIB(o1, ipv6IfIcmpInTimeExcds, o2->ipv6IfIcmpInTimeExcds);
11353         UPDATE_MIB(o1, ipv6IfIcmpInParmProblems, o2->ipv6IfIcmpInParmProblems);
11354         UPDATE_MIB(o1, ipv6IfIcmpInPktTooBigs, o2->ipv6IfIcmpInPktTooBigs);
11355         UPDATE_MIB(o1, ipv6IfIcmpInEchos, o2->ipv6IfIcmpInEchos);
11356         UPDATE_MIB(o1, ipv6IfIcmpInEchoReplies, o2->ipv6IfIcmpInEchoReplies);
11357         UPDATE_MIB(o1, ipv6IfIcmpInRouterSolicits,
11358             o2->ipv6IfIcmpInRouterSolicits);
11359         UPDATE_MIB(o1, ipv6IfIcmpInRouterAdvertisements,
11360             o2->ipv6IfIcmpInRouterAdvertisements);
11361         UPDATE_MIB(o1, ipv6IfIcmpInNeighborSolicits,
11362             o2->ipv6IfIcmpInNeighborSolicits);
11363         UPDATE_MIB(o1, ipv6IfIcmpInNeighborAdvertisements,
11364             o2->ipv6IfIcmpInNeighborAdvertisements);
11365         UPDATE_MIB(o1, ipv6IfIcmpInRedirects, o2->ipv6IfIcmpInRedirects);
11366         UPDATE_MIB(o1, ipv6IfIcmpInGroupMembQueries,
11367             o2->ipv6IfIcmpInGroupMembQueries);
11368         UPDATE_MIB(o1, ipv6IfIcmpInGroupMembResponses,
11369             o2->ipv6IfIcmpInGroupMembResponses);
11370         UPDATE_MIB(o1, ipv6IfIcmpInGroupMembReductions,
11371             o2->ipv6IfIcmpInGroupMembReductions);
11372         UPDATE_MIB(o1, ipv6IfIcmpOutMsgs, o2->ipv6IfIcmpOutMsgs);
11373         UPDATE_MIB(o1, ipv6IfIcmpOutErrors, o2->ipv6IfIcmpOutErrors);
11374         UPDATE_MIB(o1, ipv6IfIcmpOutDestUnreachs,
11375             o2->ipv6IfIcmpOutDestUnreachs);
11376         UPDATE_MIB(o1, ipv6IfIcmpOutAdminProhibs,
11377             o2->ipv6IfIcmpOutAdminProhibs);
11378         UPDATE_MIB(o1, ipv6IfIcmpOutTimeExcds, o2->ipv6IfIcmpOutTimeExcds);
11379         UPDATE_MIB(o1, ipv6IfIcmpOutParmProblems,
11380             o2->ipv6IfIcmpOutParmProblems);
11381         UPDATE_MIB(o1, ipv6IfIcmpOutPktTooBigs, o2->ipv6IfIcmpOutPktTooBigs);
11382         UPDATE_MIB(o1, ipv6IfIcmpOutEchos, o2->ipv6IfIcmpOutEchos);
11383         UPDATE_MIB(o1, ipv6IfIcmpOutEchoReplies, o2->ipv6IfIcmpOutEchoReplies);
11384         UPDATE_MIB(o1, ipv6IfIcmpOutRouterSolicits,
11385             o2->ipv6IfIcmpOutRouterSolicits);
11386         UPDATE_MIB(o1, ipv6IfIcmpOutRouterAdvertisements,
11387             o2->ipv6IfIcmpOutRouterAdvertisements);
11388         UPDATE_MIB(o1, ipv6IfIcmpOutNeighborSolicits,
11389             o2->ipv6IfIcmpOutNeighborSolicits);
11390         UPDATE_MIB(o1, ipv6IfIcmpOutNeighborAdvertisements,
11391             o2->ipv6IfIcmpOutNeighborAdvertisements);
11392         UPDATE_MIB(o1, ipv6IfIcmpOutRedirects, o2->ipv6IfIcmpOutRedirects);
11393         UPDATE_MIB(o1, ipv6IfIcmpOutGroupMembQueries,
11394             o2->ipv6IfIcmpOutGroupMembQueries);
11395         UPDATE_MIB(o1, ipv6IfIcmpOutGroupMembResponses,
11396             o2->ipv6IfIcmpOutGroupMembResponses);
11397         UPDATE_MIB(o1, ipv6IfIcmpOutGroupMembReductions,
11398             o2->ipv6IfIcmpOutGroupMembReductions);
11399         UPDATE_MIB(o1, ipv6IfIcmpInOverflows, o2->ipv6IfIcmpInOverflows);
11400         UPDATE_MIB(o1, ipv6IfIcmpBadHoplimit, o2->ipv6IfIcmpBadHoplimit);
11401         UPDATE_MIB(o1, ipv6IfIcmpInBadNeighborAdvertisements,
11402             o2->ipv6IfIcmpInBadNeighborAdvertisements);
11403         UPDATE_MIB(o1, ipv6IfIcmpInBadNeighborSolicitations,
11404             o2->ipv6IfIcmpInBadNeighborSolicitations);
11405         UPDATE_MIB(o1, ipv6IfIcmpInBadRedirects, o2->ipv6IfIcmpInBadRedirects);
11406         UPDATE_MIB(o1, ipv6IfIcmpInGroupMembTotal,
11407             o2->ipv6IfIcmpInGroupMembTotal);
11408         UPDATE_MIB(o1, ipv6IfIcmpInGroupMembBadQueries,
11409             o2->ipv6IfIcmpInGroupMembBadQueries);
11410         UPDATE_MIB(o1, ipv6IfIcmpInGroupMembBadReports,
11411             o2->ipv6IfIcmpInGroupMembBadReports);
11412         UPDATE_MIB(o1, ipv6IfIcmpInGroupMembOurReports,
11413             o2->ipv6IfIcmpInGroupMembOurReports);
11414 }
11415 
11416 /*
11417  * Called before the options are updated to check if this packet will
11418  * be source routed from here.
11419  * This routine assumes that the options are well formed i.e. that they
11420  * have already been checked.
11421  */
11422 boolean_t
11423 ip_source_routed(ipha_t *ipha, ip_stack_t *ipst)
11424 {
11425         ipoptp_t        opts;
11426         uchar_t         *opt;
11427         uint8_t         optval;
11428         uint8_t         optlen;
11429         ipaddr_t        dst;
11430 
11431         if (IS_SIMPLE_IPH(ipha)) {
11432                 ip2dbg(("not source routed\n"));
11433                 return (B_FALSE);
11434         }
11435         dst = ipha->ipha_dst;
11436         for (optval = ipoptp_first(&opts, ipha);
11437             optval != IPOPT_EOL;
11438             optval = ipoptp_next(&opts)) {
11439                 ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
11440                 opt = opts.ipoptp_cur;
11441                 optlen = opts.ipoptp_len;
11442                 ip2dbg(("ip_source_routed: opt %d, len %d\n",
11443                     optval, optlen));
11444                 switch (optval) {
11445                         uint32_t off;
11446                 case IPOPT_SSRR:
11447                 case IPOPT_LSRR:
11448                         /*
11449                          * If dst is one of our addresses and there are some
11450                          * entries left in the source route return (true).
11451                          */
11452                         if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
11453                                 ip2dbg(("ip_source_routed: not next"
11454                                     " source route 0x%x\n",
11455                                     ntohl(dst)));
11456                                 return (B_FALSE);
11457                         }
11458                         off = opt[IPOPT_OFFSET];
11459                         off--;
11460                         if (optlen < IP_ADDR_LEN ||
11461                             off > optlen - IP_ADDR_LEN) {
11462                                 /* End of source route */
11463                                 ip1dbg(("ip_source_routed: end of SR\n"));
11464                                 return (B_FALSE);
11465                         }
11466                         return (B_TRUE);
11467                 }
11468         }
11469         ip2dbg(("not source routed\n"));
11470         return (B_FALSE);
11471 }
11472 
11473 /*
11474  * ip_unbind is called by the transports to remove a conn from
11475  * the fanout table.
11476  */
11477 void
11478 ip_unbind(conn_t *connp)
11479 {
11480 
11481         ASSERT(!MUTEX_HELD(&connp->conn_lock));
11482 
11483         if (is_system_labeled() && connp->conn_anon_port) {
11484                 (void) tsol_mlp_anon(crgetzone(connp->conn_cred),
11485                     connp->conn_mlp_type, connp->conn_proto,
11486                     ntohs(connp->conn_lport), B_FALSE);
11487                 connp->conn_anon_port = 0;
11488         }
11489         connp->conn_mlp_type = mlptSingle;
11490 
11491         ipcl_hash_remove(connp);
11492 }
11493 
11494 /*
11495  * Used for deciding the MSS size for the upper layer. Thus
11496  * we need to check the outbound policy values in the conn.
11497  */
11498 int
11499 conn_ipsec_length(conn_t *connp)
11500 {
11501         ipsec_latch_t *ipl;
11502 
11503         ipl = connp->conn_latch;
11504         if (ipl == NULL)
11505                 return (0);
11506 
11507         if (connp->conn_ixa->ixa_ipsec_policy == NULL)
11508                 return (0);
11509 
11510         return (connp->conn_ixa->ixa_ipsec_policy->ipsp_act->ipa_ovhd);
11511 }
11512 
11513 /*
11514  * Returns an estimate of the IPsec headers size. This is used if
11515  * we don't want to call into IPsec to get the exact size.
11516  */
11517 int
11518 ipsec_out_extra_length(ip_xmit_attr_t *ixa)
11519 {
11520         ipsec_action_t *a;
11521 
11522         if (!(ixa->ixa_flags & IXAF_IPSEC_SECURE))
11523                 return (0);
11524 
11525         a = ixa->ixa_ipsec_action;
11526         if (a == NULL) {
11527                 ASSERT(ixa->ixa_ipsec_policy != NULL);
11528                 a = ixa->ixa_ipsec_policy->ipsp_act;
11529         }
11530         ASSERT(a != NULL);
11531 
11532         return (a->ipa_ovhd);
11533 }
11534 
11535 /*
11536  * If there are any source route options, return the true final
11537  * destination. Otherwise, return the destination.
11538  */
11539 ipaddr_t
11540 ip_get_dst(ipha_t *ipha)
11541 {
11542         ipoptp_t        opts;
11543         uchar_t         *opt;
11544         uint8_t         optval;
11545         uint8_t         optlen;
11546         ipaddr_t        dst;
11547         uint32_t off;
11548 
11549         dst = ipha->ipha_dst;
11550 
11551         if (IS_SIMPLE_IPH(ipha))
11552                 return (dst);
11553 
11554         for (optval = ipoptp_first(&opts, ipha);
11555             optval != IPOPT_EOL;
11556             optval = ipoptp_next(&opts)) {
11557                 opt = opts.ipoptp_cur;
11558                 optlen = opts.ipoptp_len;
11559                 ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
11560                 switch (optval) {
11561                 case IPOPT_SSRR:
11562                 case IPOPT_LSRR:
11563                         off = opt[IPOPT_OFFSET];
11564                         /*
11565                          * If one of the conditions is true, it means
11566                          * end of options and dst already has the right
11567                          * value.
11568                          */
11569                         if (!(optlen < IP_ADDR_LEN || off > optlen - 3)) {
11570                                 off = optlen - IP_ADDR_LEN;
11571                                 bcopy(&opt[off], &dst, IP_ADDR_LEN);
11572                         }
11573                         return (dst);
11574                 default:
11575                         break;
11576                 }
11577         }
11578 
11579         return (dst);
11580 }
11581 
11582 /*
11583  * Outbound IP fragmentation routine.
11584  * Assumes the caller has checked whether or not fragmentation should
11585  * be allowed. Here we copy the DF bit from the header to all the generated
11586  * fragments.
11587  */
11588 int
11589 ip_fragment_v4(mblk_t *mp_orig, nce_t *nce, iaflags_t ixaflags,
11590     uint_t pkt_len, uint32_t max_frag, uint32_t xmit_hint, zoneid_t szone,
11591     zoneid_t nolzid, pfirepostfrag_t postfragfn, uintptr_t *ixa_cookie)
11592 {
11593         int             i1;
11594         int             hdr_len;
11595         mblk_t          *hdr_mp;
11596         ipha_t          *ipha;
11597         int             ip_data_end;
11598         int             len;
11599         mblk_t          *mp = mp_orig;
11600         int             offset;
11601         ill_t           *ill = nce->nce_ill;
11602         ip_stack_t      *ipst = ill->ill_ipst;
11603         mblk_t          *carve_mp;
11604         uint32_t        frag_flag;
11605         uint_t          priority = mp->b_band;
11606         int             error = 0;
11607 
11608         BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragReqds);
11609 
11610         if (pkt_len != msgdsize(mp)) {
11611                 ip0dbg(("Packet length mismatch: %d, %ld\n",
11612                     pkt_len, msgdsize(mp)));
11613                 freemsg(mp);
11614                 return (EINVAL);
11615         }
11616 
11617         if (max_frag == 0) {
11618                 ip1dbg(("ip_fragment_v4: max_frag is zero. Dropping packet\n"));
11619                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11620                 ip_drop_output("FragFails: zero max_frag", mp, ill);
11621                 freemsg(mp);
11622                 return (EINVAL);
11623         }
11624 
11625         ASSERT(MBLKL(mp) >= sizeof (ipha_t));
11626         ipha = (ipha_t *)mp->b_rptr;
11627         ASSERT(ntohs(ipha->ipha_length) == pkt_len);
11628         frag_flag = ntohs(ipha->ipha_fragment_offset_and_flags) & IPH_DF;
11629 
11630         /*
11631          * Establish the starting offset.  May not be zero if we are fragging
11632          * a fragment that is being forwarded.
11633          */
11634         offset = ntohs(ipha->ipha_fragment_offset_and_flags) & IPH_OFFSET;
11635 
11636         /* TODO why is this test needed? */
11637         if (((max_frag - ntohs(ipha->ipha_length)) & ~7) < 8) {
11638                 /* TODO: notify ulp somehow */
11639                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11640                 ip_drop_output("FragFails: bad starting offset", mp, ill);
11641                 freemsg(mp);
11642                 return (EINVAL);
11643         }
11644 
11645         hdr_len = IPH_HDR_LENGTH(ipha);
11646         ipha->ipha_hdr_checksum = 0;
11647 
11648         /*
11649          * Establish the number of bytes maximum per frag, after putting
11650          * in the header.
11651          */
11652         len = (max_frag - hdr_len) & ~7;
11653 
11654         /* Get a copy of the header for the trailing frags */
11655         hdr_mp = ip_fragment_copyhdr((uchar_t *)ipha, hdr_len, offset, ipst,
11656             mp);
11657         if (hdr_mp == NULL) {
11658                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11659                 ip_drop_output("FragFails: no hdr_mp", mp, ill);
11660                 freemsg(mp);
11661                 return (ENOBUFS);
11662         }
11663 
11664         /* Store the starting offset, with the MoreFrags flag. */
11665         i1 = offset | IPH_MF | frag_flag;
11666         ipha->ipha_fragment_offset_and_flags = htons((uint16_t)i1);
11667 
11668         /* Establish the ending byte offset, based on the starting offset. */
11669         offset <<= 3;
11670         ip_data_end = offset + ntohs(ipha->ipha_length) - hdr_len;
11671 
11672         /* Store the length of the first fragment in the IP header. */
11673         i1 = len + hdr_len;
11674         ASSERT(i1 <= IP_MAXPACKET);
11675         ipha->ipha_length = htons((uint16_t)i1);
11676 
11677         /*
11678          * Compute the IP header checksum for the first frag.  We have to
11679          * watch out that we stop at the end of the header.
11680          */
11681         ipha->ipha_hdr_checksum = ip_csum_hdr(ipha);
11682 
11683         /*
11684          * Now carve off the first frag.  Note that this will include the
11685          * original IP header.
11686          */
11687         if (!(mp = ip_carve_mp(&mp_orig, i1))) {
11688                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11689                 ip_drop_output("FragFails: could not carve mp", mp_orig, ill);
11690                 freeb(hdr_mp);
11691                 freemsg(mp_orig);
11692                 return (ENOBUFS);
11693         }
11694 
11695         BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragCreates);
11696 
11697         error = postfragfn(mp, nce, ixaflags, i1, xmit_hint, szone, nolzid,
11698             ixa_cookie);
11699         if (error != 0 && error != EWOULDBLOCK) {
11700                 /* No point in sending the other fragments */
11701                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11702                 ip_drop_output("FragFails: postfragfn failed", mp_orig, ill);
11703                 freeb(hdr_mp);
11704                 freemsg(mp_orig);
11705                 return (error);
11706         }
11707 
11708         /* No need to redo state machine in loop */
11709         ixaflags &= ~IXAF_REACH_CONF;
11710 
11711         /* Advance the offset to the second frag starting point. */
11712         offset += len;
11713         /*
11714          * Update hdr_len from the copied header - there might be less options
11715          * in the later fragments.
11716          */
11717         hdr_len = IPH_HDR_LENGTH(hdr_mp->b_rptr);
11718         /* Loop until done. */
11719         for (;;) {
11720                 uint16_t        offset_and_flags;
11721                 uint16_t        ip_len;
11722 
11723                 if (ip_data_end - offset > len) {
11724                         /*
11725                          * Carve off the appropriate amount from the original
11726                          * datagram.
11727                          */
11728                         if (!(carve_mp = ip_carve_mp(&mp_orig, len))) {
11729                                 mp = NULL;
11730                                 break;
11731                         }
11732                         /*
11733                          * More frags after this one.  Get another copy
11734                          * of the header.
11735                          */
11736                         if (carve_mp->b_datap->db_ref == 1 &&
11737                             hdr_mp->b_wptr - hdr_mp->b_rptr <
11738                             carve_mp->b_rptr - carve_mp->b_datap->db_base) {
11739                                 /* Inline IP header */
11740                                 carve_mp->b_rptr -= hdr_mp->b_wptr -
11741                                     hdr_mp->b_rptr;
11742                                 bcopy(hdr_mp->b_rptr, carve_mp->b_rptr,
11743                                     hdr_mp->b_wptr - hdr_mp->b_rptr);
11744                                 mp = carve_mp;
11745                         } else {
11746                                 if (!(mp = copyb(hdr_mp))) {
11747                                         freemsg(carve_mp);
11748                                         break;
11749                                 }
11750                                 /* Get priority marking, if any. */
11751                                 mp->b_band = priority;
11752                                 mp->b_cont = carve_mp;
11753                         }
11754                         ipha = (ipha_t *)mp->b_rptr;
11755                         offset_and_flags = IPH_MF;
11756                 } else {
11757                         /*
11758                          * Last frag.  Consume the header. Set len to
11759                          * the length of this last piece.
11760                          */
11761                         len = ip_data_end - offset;
11762 
11763                         /*
11764                          * Carve off the appropriate amount from the original
11765                          * datagram.
11766                          */
11767                         if (!(carve_mp = ip_carve_mp(&mp_orig, len))) {
11768                                 mp = NULL;
11769                                 break;
11770                         }
11771                         if (carve_mp->b_datap->db_ref == 1 &&
11772                             hdr_mp->b_wptr - hdr_mp->b_rptr <
11773                             carve_mp->b_rptr - carve_mp->b_datap->db_base) {
11774                                 /* Inline IP header */
11775                                 carve_mp->b_rptr -= hdr_mp->b_wptr -
11776                                     hdr_mp->b_rptr;
11777                                 bcopy(hdr_mp->b_rptr, carve_mp->b_rptr,
11778                                     hdr_mp->b_wptr - hdr_mp->b_rptr);
11779                                 mp = carve_mp;
11780                                 freeb(hdr_mp);
11781                                 hdr_mp = mp;
11782                         } else {
11783                                 mp = hdr_mp;
11784                                 /* Get priority marking, if any. */
11785                                 mp->b_band = priority;
11786                                 mp->b_cont = carve_mp;
11787                         }
11788                         ipha = (ipha_t *)mp->b_rptr;
11789                         /* A frag of a frag might have IPH_MF non-zero */
11790                         offset_and_flags =
11791                             ntohs(ipha->ipha_fragment_offset_and_flags) &
11792                             IPH_MF;
11793                 }
11794                 offset_and_flags |= (uint16_t)(offset >> 3);
11795                 offset_and_flags |= (uint16_t)frag_flag;
11796                 /* Store the offset and flags in the IP header. */
11797                 ipha->ipha_fragment_offset_and_flags = htons(offset_and_flags);
11798 
11799                 /* Store the length in the IP header. */
11800                 ip_len = (uint16_t)(len + hdr_len);
11801                 ipha->ipha_length = htons(ip_len);
11802 
11803                 /*
11804                  * Set the IP header checksum.  Note that mp is just
11805                  * the header, so this is easy to pass to ip_csum.
11806                  */
11807                 ipha->ipha_hdr_checksum = ip_csum_hdr(ipha);
11808 
11809                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragCreates);
11810 
11811                 error = postfragfn(mp, nce, ixaflags, ip_len, xmit_hint, szone,
11812                     nolzid, ixa_cookie);
11813                 /* All done if we just consumed the hdr_mp. */
11814                 if (mp == hdr_mp) {
11815                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragOKs);
11816                         return (error);
11817                 }
11818                 if (error != 0 && error != EWOULDBLOCK) {
11819                         DTRACE_PROBE2(ip__xmit__frag__fail, ill_t *, ill,
11820                             mblk_t *, hdr_mp);
11821                         /* No point in sending the other fragments */
11822                         break;
11823                 }
11824 
11825                 /* Otherwise, advance and loop. */
11826                 offset += len;
11827         }
11828         /* Clean up following allocation failure. */
11829         BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutFragFails);
11830         ip_drop_output("FragFails: loop ended", NULL, ill);
11831         if (mp != hdr_mp)
11832                 freeb(hdr_mp);
11833         if (mp != mp_orig)
11834                 freemsg(mp_orig);
11835         return (error);
11836 }
11837 
11838 /*
11839  * Copy the header plus those options which have the copy bit set
11840  */
11841 static mblk_t *
11842 ip_fragment_copyhdr(uchar_t *rptr, int hdr_len, int offset, ip_stack_t *ipst,
11843     mblk_t *src)
11844 {
11845         mblk_t  *mp;
11846         uchar_t *up;
11847 
11848         /*
11849          * Quick check if we need to look for options without the copy bit
11850          * set
11851          */
11852         mp = allocb_tmpl(ipst->ips_ip_wroff_extra + hdr_len, src);
11853         if (!mp)
11854                 return (mp);
11855         mp->b_rptr += ipst->ips_ip_wroff_extra;
11856         if (hdr_len == IP_SIMPLE_HDR_LENGTH || offset != 0) {
11857                 bcopy(rptr, mp->b_rptr, hdr_len);
11858                 mp->b_wptr += hdr_len + ipst->ips_ip_wroff_extra;
11859                 return (mp);
11860         }
11861         up  = mp->b_rptr;
11862         bcopy(rptr, up, IP_SIMPLE_HDR_LENGTH);
11863         up += IP_SIMPLE_HDR_LENGTH;
11864         rptr += IP_SIMPLE_HDR_LENGTH;
11865         hdr_len -= IP_SIMPLE_HDR_LENGTH;
11866         while (hdr_len > 0) {
11867                 uint32_t optval;
11868                 uint32_t optlen;
11869 
11870                 optval = *rptr;
11871                 if (optval == IPOPT_EOL)
11872                         break;
11873                 if (optval == IPOPT_NOP)
11874                         optlen = 1;
11875                 else
11876                         optlen = rptr[1];
11877                 if (optval & IPOPT_COPY) {
11878                         bcopy(rptr, up, optlen);
11879                         up += optlen;
11880                 }
11881                 rptr += optlen;
11882                 hdr_len -= optlen;
11883         }
11884         /*
11885          * Make sure that we drop an even number of words by filling
11886          * with EOL to the next word boundary.
11887          */
11888         for (hdr_len = up - (mp->b_rptr + IP_SIMPLE_HDR_LENGTH);
11889             hdr_len & 0x3; hdr_len++)
11890                 *up++ = IPOPT_EOL;
11891         mp->b_wptr = up;
11892         /* Update header length */
11893         mp->b_rptr[0] = (uint8_t)((IP_VERSION << 4) | ((up - mp->b_rptr) >> 2));
11894         return (mp);
11895 }
11896 
11897 /*
11898  * Update any source route, record route, or timestamp options when
11899  * sending a packet back to ourselves.
11900  * Check that we are at end of strict source route.
11901  * The options have been sanity checked by ip_output_options().
11902  */
11903 void
11904 ip_output_local_options(ipha_t *ipha, ip_stack_t *ipst)
11905 {
11906         ipoptp_t        opts;
11907         uchar_t         *opt;
11908         uint8_t         optval;
11909         uint8_t         optlen;
11910         ipaddr_t        dst;
11911         uint32_t        ts;
11912         timestruc_t     now;
11913 
11914         for (optval = ipoptp_first(&opts, ipha);
11915             optval != IPOPT_EOL;
11916             optval = ipoptp_next(&opts)) {
11917                 opt = opts.ipoptp_cur;
11918                 optlen = opts.ipoptp_len;
11919                 ASSERT((opts.ipoptp_flags & IPOPTP_ERROR) == 0);
11920                 switch (optval) {
11921                         uint32_t off;
11922                 case IPOPT_SSRR:
11923                 case IPOPT_LSRR:
11924                         off = opt[IPOPT_OFFSET];
11925                         off--;
11926                         if (optlen < IP_ADDR_LEN ||
11927                             off > optlen - IP_ADDR_LEN) {
11928                                 /* End of source route */
11929                                 break;
11930                         }
11931                         /*
11932                          * This will only happen if two consecutive entries
11933                          * in the source route contains our address or if
11934                          * it is a packet with a loose source route which
11935                          * reaches us before consuming the whole source route
11936                          */
11937 
11938                         if (optval == IPOPT_SSRR) {
11939                                 return;
11940                         }
11941                         /*
11942                          * Hack: instead of dropping the packet truncate the
11943                          * source route to what has been used by filling the
11944                          * rest with IPOPT_NOP.
11945                          */
11946                         opt[IPOPT_OLEN] = (uint8_t)off;
11947                         while (off < optlen) {
11948                                 opt[off++] = IPOPT_NOP;
11949                         }
11950                         break;
11951                 case IPOPT_RR:
11952                         off = opt[IPOPT_OFFSET];
11953                         off--;
11954                         if (optlen < IP_ADDR_LEN ||
11955                             off > optlen - IP_ADDR_LEN) {
11956                                 /* No more room - ignore */
11957                                 ip1dbg((
11958                                     "ip_output_local_options: end of RR\n"));
11959                                 break;
11960                         }
11961                         dst = htonl(INADDR_LOOPBACK);
11962                         bcopy(&dst, (char *)opt + off, IP_ADDR_LEN);
11963                         opt[IPOPT_OFFSET] += IP_ADDR_LEN;
11964                         break;
11965                 case IPOPT_TS:
11966                         /* Insert timestamp if there is romm */
11967                         switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
11968                         case IPOPT_TS_TSONLY:
11969                                 off = IPOPT_TS_TIMELEN;
11970                                 break;
11971                         case IPOPT_TS_PRESPEC:
11972                         case IPOPT_TS_PRESPEC_RFC791:
11973                                 /* Verify that the address matched */
11974                                 off = opt[IPOPT_OFFSET] - 1;
11975                                 bcopy((char *)opt + off, &dst, IP_ADDR_LEN);
11976                                 if (ip_type_v4(dst, ipst) != IRE_LOCAL) {
11977                                         /* Not for us */
11978                                         break;
11979                                 }
11980                                 /* FALLTHRU */
11981                         case IPOPT_TS_TSANDADDR:
11982                                 off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
11983                                 break;
11984                         default:
11985                                 /*
11986                                  * ip_*put_options should have already
11987                                  * dropped this packet.
11988                                  */
11989                                 cmn_err(CE_PANIC, "ip_output_local_options: "
11990                                     "unknown IT - bug in ip_output_options?\n");
11991                                 return; /* Keep "lint" happy */
11992                         }
11993                         if (opt[IPOPT_OFFSET] - 1 + off > optlen) {
11994                                 /* Increase overflow counter */
11995                                 off = (opt[IPOPT_POS_OV_FLG] >> 4) + 1;
11996                                 opt[IPOPT_POS_OV_FLG] = (uint8_t)
11997                                     (opt[IPOPT_POS_OV_FLG] & 0x0F) |
11998                                     (off << 4);
11999                                 break;
12000                         }
12001                         off = opt[IPOPT_OFFSET] - 1;
12002                         switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
12003                         case IPOPT_TS_PRESPEC:
12004                         case IPOPT_TS_PRESPEC_RFC791:
12005                         case IPOPT_TS_TSANDADDR:
12006                                 dst = htonl(INADDR_LOOPBACK);
12007                                 bcopy(&dst, (char *)opt + off, IP_ADDR_LEN);
12008                                 opt[IPOPT_OFFSET] += IP_ADDR_LEN;
12009                                 /* FALLTHRU */
12010                         case IPOPT_TS_TSONLY:
12011                                 off = opt[IPOPT_OFFSET] - 1;
12012                                 /* Compute # of milliseconds since midnight */
12013                                 gethrestime(&now);
12014                                 ts = (now.tv_sec % (24 * 60 * 60)) * 1000 +
12015                                     now.tv_nsec / (NANOSEC / MILLISEC);
12016                                 bcopy(&ts, (char *)opt + off, IPOPT_TS_TIMELEN);
12017                                 opt[IPOPT_OFFSET] += IPOPT_TS_TIMELEN;
12018                                 break;
12019                         }
12020                         break;
12021                 }
12022         }
12023 }
12024 
12025 /*
12026  * Prepend an M_DATA fastpath header, and if none present prepend a
12027  * DL_UNITDATA_REQ. Frees the mblk on failure.
12028  *
12029  * nce_dlur_mp and nce_fp_mp can not disappear once they have been set.
12030  * If there is a change to them, the nce will be deleted (condemned) and
12031  * a new nce_t will be created when packets are sent. Thus we need no locks
12032  * to access those fields.
12033  *
12034  * We preserve b_band to support IPQoS. If a DL_UNITDATA_REQ is prepended
12035  * we place b_band in dl_priority.dl_max.
12036  */
12037 static mblk_t *
12038 ip_xmit_attach_llhdr(mblk_t *mp, nce_t *nce)
12039 {
12040         uint_t  hlen;
12041         mblk_t *mp1;
12042         uint_t  priority;
12043         uchar_t *rptr;
12044 
12045         rptr = mp->b_rptr;
12046 
12047         ASSERT(DB_TYPE(mp) == M_DATA);
12048         priority = mp->b_band;
12049 
12050         ASSERT(nce != NULL);
12051         if ((mp1 = nce->nce_fp_mp) != NULL) {
12052                 hlen = MBLKL(mp1);
12053                 /*
12054                  * Check if we have enough room to prepend fastpath
12055                  * header
12056                  */
12057                 if (hlen != 0 && (rptr - mp->b_datap->db_base) >= hlen) {
12058                         rptr -= hlen;
12059                         bcopy(mp1->b_rptr, rptr, hlen);
12060                         /*
12061                          * Set the b_rptr to the start of the link layer
12062                          * header
12063                          */
12064                         mp->b_rptr = rptr;
12065                         return (mp);
12066                 }
12067                 mp1 = copyb(mp1);
12068                 if (mp1 == NULL) {
12069                         ill_t *ill = nce->nce_ill;
12070 
12071                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
12072                         ip_drop_output("ipIfStatsOutDiscards", mp, ill);
12073                         freemsg(mp);
12074                         return (NULL);
12075                 }
12076                 mp1->b_band = priority;
12077                 mp1->b_cont = mp;
12078                 DB_CKSUMSTART(mp1) = DB_CKSUMSTART(mp);
12079                 DB_CKSUMSTUFF(mp1) = DB_CKSUMSTUFF(mp);
12080                 DB_CKSUMEND(mp1) = DB_CKSUMEND(mp);
12081                 DB_CKSUMFLAGS(mp1) = DB_CKSUMFLAGS(mp);
12082                 DB_LSOMSS(mp1) = DB_LSOMSS(mp);
12083                 DTRACE_PROBE1(ip__xmit__copyb, (mblk_t *), mp1);
12084                 /*
12085                  * XXX disable ICK_VALID and compute checksum
12086                  * here; can happen if nce_fp_mp changes and
12087                  * it can't be copied now due to insufficient
12088                  * space. (unlikely, fp mp can change, but it
12089                  * does not increase in length)
12090                  */
12091                 return (mp1);
12092         }
12093         mp1 = copyb(nce->nce_dlur_mp);
12094 
12095         if (mp1 == NULL) {
12096                 ill_t *ill = nce->nce_ill;
12097 
12098                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
12099                 ip_drop_output("ipIfStatsOutDiscards", mp, ill);
12100                 freemsg(mp);
12101                 return (NULL);
12102         }
12103         mp1->b_cont = mp;
12104         if (priority != 0) {
12105                 mp1->b_band = priority;
12106                 ((dl_unitdata_req_t *)(mp1->b_rptr))->dl_priority.dl_max =
12107                     priority;
12108         }
12109         return (mp1);
12110 #undef rptr
12111 }
12112 
12113 /*
12114  * Finish the outbound IPsec processing. This function is called from
12115  * ipsec_out_process() if the IPsec packet was processed
12116  * synchronously, or from {ah,esp}_kcf_callback_outbound() if it was processed
12117  * asynchronously.
12118  *
12119  * This is common to IPv4 and IPv6.
12120  */
12121 int
12122 ip_output_post_ipsec(mblk_t *mp, ip_xmit_attr_t *ixa)
12123 {
12124         iaflags_t       ixaflags = ixa->ixa_flags;
12125         uint_t          pktlen;
12126 
12127 
12128         /* AH/ESP don't update ixa_pktlen when they modify the packet */
12129         if (ixaflags & IXAF_IS_IPV4) {
12130                 ipha_t          *ipha = (ipha_t *)mp->b_rptr;
12131 
12132                 ASSERT(IPH_HDR_VERSION(ipha) == IPV4_VERSION);
12133                 pktlen = ntohs(ipha->ipha_length);
12134         } else {
12135                 ip6_t           *ip6h = (ip6_t *)mp->b_rptr;
12136 
12137                 ASSERT(IPH_HDR_VERSION(mp->b_rptr) == IPV6_VERSION);
12138                 pktlen = ntohs(ip6h->ip6_plen) + IPV6_HDR_LEN;
12139         }
12140 
12141         /*
12142          * We release any hard reference on the SAs here to make
12143          * sure the SAs can be garbage collected. ipsr_sa has a soft reference
12144          * on the SAs.
12145          * If in the future we want the hard latching of the SAs in the
12146          * ip_xmit_attr_t then we should remove this.
12147          */
12148         if (ixa->ixa_ipsec_esp_sa != NULL) {
12149                 IPSA_REFRELE(ixa->ixa_ipsec_esp_sa);
12150                 ixa->ixa_ipsec_esp_sa = NULL;
12151         }
12152         if (ixa->ixa_ipsec_ah_sa != NULL) {
12153                 IPSA_REFRELE(ixa->ixa_ipsec_ah_sa);
12154                 ixa->ixa_ipsec_ah_sa = NULL;
12155         }
12156 
12157         /* Do we need to fragment? */
12158         if ((ixa->ixa_flags & IXAF_IPV6_ADD_FRAGHDR) ||
12159             pktlen > ixa->ixa_fragsize) {
12160                 if (ixaflags & IXAF_IS_IPV4) {
12161                         ASSERT(!(ixa->ixa_flags & IXAF_IPV6_ADD_FRAGHDR));
12162                         /*
12163                          * We check for the DF case in ipsec_out_process
12164                          * hence this only handles the non-DF case.
12165                          */
12166                         return (ip_fragment_v4(mp, ixa->ixa_nce, ixa->ixa_flags,
12167                             pktlen, ixa->ixa_fragsize,
12168                             ixa->ixa_xmit_hint, ixa->ixa_zoneid,
12169                             ixa->ixa_no_loop_zoneid, ixa->ixa_postfragfn,
12170                             &ixa->ixa_cookie));
12171                 } else {
12172                         mp = ip_fraghdr_add_v6(mp, ixa->ixa_ident, ixa);
12173                         if (mp == NULL) {
12174                                 /* MIB and ip_drop_output already done */
12175                                 return (ENOMEM);
12176                         }
12177                         pktlen += sizeof (ip6_frag_t);
12178                         if (pktlen > ixa->ixa_fragsize) {
12179                                 return (ip_fragment_v6(mp, ixa->ixa_nce,
12180                                     ixa->ixa_flags, pktlen,
12181                                     ixa->ixa_fragsize, ixa->ixa_xmit_hint,
12182                                     ixa->ixa_zoneid, ixa->ixa_no_loop_zoneid,
12183                                     ixa->ixa_postfragfn, &ixa->ixa_cookie));
12184                         }
12185                 }
12186         }
12187         return ((ixa->ixa_postfragfn)(mp, ixa->ixa_nce, ixa->ixa_flags,
12188             pktlen, ixa->ixa_xmit_hint, ixa->ixa_zoneid,
12189             ixa->ixa_no_loop_zoneid, NULL));
12190 }
12191 
12192 /*
12193  * Finish the inbound IPsec processing. This function is called from
12194  * ipsec_out_process() if the IPsec packet was processed
12195  * synchronously, or from {ah,esp}_kcf_callback_outbound() if it was processed
12196  * asynchronously.
12197  *
12198  * This is common to IPv4 and IPv6.
12199  */
12200 void
12201 ip_input_post_ipsec(mblk_t *mp, ip_recv_attr_t *ira)
12202 {
12203         iaflags_t       iraflags = ira->ira_flags;
12204 
12205         /* Length might have changed */
12206         if (iraflags & IRAF_IS_IPV4) {
12207                 ipha_t          *ipha = (ipha_t *)mp->b_rptr;
12208 
12209                 ASSERT(IPH_HDR_VERSION(ipha) == IPV4_VERSION);
12210                 ira->ira_pktlen = ntohs(ipha->ipha_length);
12211                 ira->ira_ip_hdr_length = IPH_HDR_LENGTH(ipha);
12212                 ira->ira_protocol = ipha->ipha_protocol;
12213 
12214                 ip_fanout_v4(mp, ipha, ira);
12215         } else {
12216                 ip6_t           *ip6h = (ip6_t *)mp->b_rptr;
12217                 uint8_t         *nexthdrp;
12218 
12219                 ASSERT(IPH_HDR_VERSION(mp->b_rptr) == IPV6_VERSION);
12220                 ira->ira_pktlen = ntohs(ip6h->ip6_plen) + IPV6_HDR_LEN;
12221                 if (!ip_hdr_length_nexthdr_v6(mp, ip6h, &ira->ira_ip_hdr_length,
12222                     &nexthdrp)) {
12223                         /* Malformed packet */
12224                         BUMP_MIB(ira->ira_ill->ill_ip_mib, ipIfStatsInDiscards);
12225                         ip_drop_input("ipIfStatsInDiscards", mp, ira->ira_ill);
12226                         freemsg(mp);
12227                         return;
12228                 }
12229                 ira->ira_protocol = *nexthdrp;
12230                 ip_fanout_v6(mp, ip6h, ira);
12231         }
12232 }
12233 
12234 /*
12235  * Select which AH & ESP SA's to use (if any) for the outbound packet.
12236  *
12237  * If this function returns B_TRUE, the requested SA's have been filled
12238  * into the ixa_ipsec_*_sa pointers.
12239  *
12240  * If the function returns B_FALSE, the packet has been "consumed", most
12241  * likely by an ACQUIRE sent up via PF_KEY to a key management daemon.
12242  *
12243  * The SA references created by the protocol-specific "select"
12244  * function will be released in ip_output_post_ipsec.
12245  */
12246 static boolean_t
12247 ipsec_out_select_sa(mblk_t *mp, ip_xmit_attr_t *ixa)
12248 {
12249         boolean_t need_ah_acquire = B_FALSE, need_esp_acquire = B_FALSE;
12250         ipsec_policy_t *pp;
12251         ipsec_action_t *ap;
12252 
12253         ASSERT(ixa->ixa_flags & IXAF_IPSEC_SECURE);
12254         ASSERT((ixa->ixa_ipsec_policy != NULL) ||
12255             (ixa->ixa_ipsec_action != NULL));
12256 
12257         ap = ixa->ixa_ipsec_action;
12258         if (ap == NULL) {
12259                 pp = ixa->ixa_ipsec_policy;
12260                 ASSERT(pp != NULL);
12261                 ap = pp->ipsp_act;
12262                 ASSERT(ap != NULL);
12263         }
12264 
12265         /*
12266          * We have an action.  now, let's select SA's.
12267          * A side effect of setting ixa_ipsec_*_sa is that it will
12268          * be cached in the conn_t.
12269          */
12270         if (ap->ipa_want_esp) {
12271                 if (ixa->ixa_ipsec_esp_sa == NULL) {
12272                         need_esp_acquire = !ipsec_outbound_sa(mp, ixa,
12273                             IPPROTO_ESP);
12274                 }
12275                 ASSERT(need_esp_acquire || ixa->ixa_ipsec_esp_sa != NULL);
12276         }
12277 
12278         if (ap->ipa_want_ah) {
12279                 if (ixa->ixa_ipsec_ah_sa == NULL) {
12280                         need_ah_acquire = !ipsec_outbound_sa(mp, ixa,
12281                             IPPROTO_AH);
12282                 }
12283                 ASSERT(need_ah_acquire || ixa->ixa_ipsec_ah_sa != NULL);
12284                 /*
12285                  * The ESP and AH processing order needs to be preserved
12286                  * when both protocols are required (ESP should be applied
12287                  * before AH for an outbound packet). Force an ESP ACQUIRE
12288                  * when both ESP and AH are required, and an AH ACQUIRE
12289                  * is needed.
12290                  */
12291                 if (ap->ipa_want_esp && need_ah_acquire)
12292                         need_esp_acquire = B_TRUE;
12293         }
12294 
12295         /*
12296          * Send an ACQUIRE (extended, regular, or both) if we need one.
12297          * Release SAs that got referenced, but will not be used until we
12298          * acquire _all_ of the SAs we need.
12299          */
12300         if (need_ah_acquire || need_esp_acquire) {
12301                 if (ixa->ixa_ipsec_ah_sa != NULL) {
12302                         IPSA_REFRELE(ixa->ixa_ipsec_ah_sa);
12303                         ixa->ixa_ipsec_ah_sa = NULL;
12304                 }
12305                 if (ixa->ixa_ipsec_esp_sa != NULL) {
12306                         IPSA_REFRELE(ixa->ixa_ipsec_esp_sa);
12307                         ixa->ixa_ipsec_esp_sa = NULL;
12308                 }
12309 
12310                 sadb_acquire(mp, ixa, need_ah_acquire, need_esp_acquire);
12311                 return (B_FALSE);
12312         }
12313 
12314         return (B_TRUE);
12315 }
12316 
12317 /*
12318  * Handle IPsec output processing.
12319  * This function is only entered once for a given packet.
12320  * We try to do things synchronously, but if we need to have user-level
12321  * set up SAs, or ESP or AH uses asynchronous kEF, then the operation
12322  * will be completed
12323  *  - when the SAs are added in esp_add_sa_finish/ah_add_sa_finish
12324  *  - when asynchronous ESP is done it will do AH
12325  *
12326  * In all cases we come back in ip_output_post_ipsec() to fragment and
12327  * send out the packet.
12328  */
12329 int
12330 ipsec_out_process(mblk_t *mp, ip_xmit_attr_t *ixa)
12331 {
12332         ill_t           *ill = ixa->ixa_nce->nce_ill;
12333         ip_stack_t      *ipst = ixa->ixa_ipst;
12334         ipsec_stack_t   *ipss;
12335         ipsec_policy_t  *pp;
12336         ipsec_action_t  *ap;
12337 
12338         ASSERT(ixa->ixa_flags & IXAF_IPSEC_SECURE);
12339 
12340         ASSERT((ixa->ixa_ipsec_policy != NULL) ||
12341             (ixa->ixa_ipsec_action != NULL));
12342 
12343         ipss = ipst->ips_netstack->netstack_ipsec;
12344         if (!ipsec_loaded(ipss)) {
12345                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
12346                 ip_drop_packet(mp, B_TRUE, ill,
12347                     DROPPER(ipss, ipds_ip_ipsec_not_loaded),
12348                     &ipss->ipsec_dropper);
12349                 return (ENOTSUP);
12350         }
12351 
12352         ap = ixa->ixa_ipsec_action;
12353         if (ap == NULL) {
12354                 pp = ixa->ixa_ipsec_policy;
12355                 ASSERT(pp != NULL);
12356                 ap = pp->ipsp_act;
12357                 ASSERT(ap != NULL);
12358         }
12359 
12360         /* Handle explicit drop action and bypass. */
12361         switch (ap->ipa_act.ipa_type) {
12362         case IPSEC_ACT_DISCARD:
12363         case IPSEC_ACT_REJECT:
12364                 ip_drop_packet(mp, B_FALSE, ill,
12365                     DROPPER(ipss, ipds_spd_explicit), &ipss->ipsec_spd_dropper);
12366                 return (EHOSTUNREACH);  /* IPsec policy failure */
12367         case IPSEC_ACT_BYPASS:
12368                 return (ip_output_post_ipsec(mp, ixa));
12369         }
12370 
12371         /*
12372          * The order of processing is first insert a IP header if needed.
12373          * Then insert the ESP header and then the AH header.
12374          */
12375         if ((ixa->ixa_flags & IXAF_IS_IPV4) && ap->ipa_want_se) {
12376                 /*
12377                  * First get the outer IP header before sending
12378                  * it to ESP.
12379                  */
12380                 ipha_t *oipha, *iipha;
12381                 mblk_t *outer_mp, *inner_mp;
12382 
12383                 if ((outer_mp = allocb(sizeof (ipha_t), BPRI_HI)) == NULL) {
12384                         (void) mi_strlog(ill->ill_rq, 0,
12385                             SL_ERROR|SL_TRACE|SL_CONSOLE,
12386                             "ipsec_out_process: "
12387                             "Self-Encapsulation failed: Out of memory\n");
12388                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
12389                         ip_drop_output("ipIfStatsOutDiscards", mp, ill);
12390                         freemsg(mp);
12391                         return (ENOBUFS);
12392                 }
12393                 inner_mp = mp;
12394                 ASSERT(inner_mp->b_datap->db_type == M_DATA);
12395                 oipha = (ipha_t *)outer_mp->b_rptr;
12396                 iipha = (ipha_t *)inner_mp->b_rptr;
12397                 *oipha = *iipha;
12398                 outer_mp->b_wptr += sizeof (ipha_t);
12399                 oipha->ipha_length = htons(ntohs(iipha->ipha_length) +
12400                     sizeof (ipha_t));
12401                 oipha->ipha_protocol = IPPROTO_ENCAP;
12402                 oipha->ipha_version_and_hdr_length =
12403                     IP_SIMPLE_HDR_VERSION;
12404                 oipha->ipha_hdr_checksum = 0;
12405                 oipha->ipha_hdr_checksum = ip_csum_hdr(oipha);
12406                 outer_mp->b_cont = inner_mp;
12407                 mp = outer_mp;
12408 
12409                 ixa->ixa_flags |= IXAF_IPSEC_TUNNEL;
12410         }
12411 
12412         /* If we need to wait for a SA then we can't return any errno */
12413         if (((ap->ipa_want_ah && (ixa->ixa_ipsec_ah_sa == NULL)) ||
12414             (ap->ipa_want_esp && (ixa->ixa_ipsec_esp_sa == NULL))) &&
12415             !ipsec_out_select_sa(mp, ixa))
12416                 return (0);
12417 
12418         /*
12419          * By now, we know what SA's to use.  Toss over to ESP & AH
12420          * to do the heavy lifting.
12421          */
12422         if (ap->ipa_want_esp) {
12423                 ASSERT(ixa->ixa_ipsec_esp_sa != NULL);
12424 
12425                 mp = ixa->ixa_ipsec_esp_sa->ipsa_output_func(mp, ixa);
12426                 if (mp == NULL) {
12427                         /*
12428                          * Either it failed or is pending. In the former case
12429                          * ipIfStatsInDiscards was increased.
12430                          */
12431                         return (0);
12432                 }
12433         }
12434 
12435         if (ap->ipa_want_ah) {
12436                 ASSERT(ixa->ixa_ipsec_ah_sa != NULL);
12437 
12438                 mp = ixa->ixa_ipsec_ah_sa->ipsa_output_func(mp, ixa);
12439                 if (mp == NULL) {
12440                         /*
12441                          * Either it failed or is pending. In the former case
12442                          * ipIfStatsInDiscards was increased.
12443                          */
12444                         return (0);
12445                 }
12446         }
12447         /*
12448          * We are done with IPsec processing. Send it over
12449          * the wire.
12450          */
12451         return (ip_output_post_ipsec(mp, ixa));
12452 }
12453 
12454 /*
12455  * ioctls that go through a down/up sequence may need to wait for the down
12456  * to complete. This involves waiting for the ire and ipif refcnts to go down
12457  * to zero. Subsequently the ioctl is restarted from ipif_ill_refrele_tail.
12458  */
12459 /* ARGSUSED */
12460 void
12461 ip_reprocess_ioctl(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *dummy_arg)
12462 {
12463         struct iocblk *iocp;
12464         mblk_t *mp1;
12465         ip_ioctl_cmd_t *ipip;
12466         int err;
12467         sin_t   *sin;
12468         struct lifreq *lifr;
12469         struct ifreq *ifr;
12470 
12471         iocp = (struct iocblk *)mp->b_rptr;
12472         ASSERT(ipsq != NULL);
12473         /* Existence of mp1 verified in ip_wput_nondata */
12474         mp1 = mp->b_cont->b_cont;
12475         ipip = ip_sioctl_lookup(iocp->ioc_cmd);
12476         if (ipip->ipi_cmd == SIOCSLIFNAME || ipip->ipi_cmd == IF_UNITSEL) {
12477                 /*
12478                  * Special case where ipx_current_ipif is not set:
12479                  * ill_phyint_reinit merged the v4 and v6 into a single ipsq.
12480                  * We are here as were not able to complete the operation in
12481                  * ipif_set_values because we could not become exclusive on
12482                  * the new ipsq.
12483                  */
12484                 ill_t *ill = q->q_ptr;
12485                 ipsq_current_start(ipsq, ill->ill_ipif, ipip->ipi_cmd);
12486         }
12487         ASSERT(ipsq->ipsq_xop->ipx_current_ipif != NULL);
12488 
12489         if (ipip->ipi_cmd_type == IF_CMD) {
12490                 /* This a old style SIOC[GS]IF* command */
12491                 ifr = (struct ifreq *)mp1->b_rptr;
12492                 sin = (sin_t *)&ifr->ifr_addr;
12493         } else if (ipip->ipi_cmd_type == LIF_CMD) {
12494                 /* This a new style SIOC[GS]LIF* command */
12495                 lifr = (struct lifreq *)mp1->b_rptr;
12496                 sin = (sin_t *)&lifr->lifr_addr;
12497         } else {
12498                 sin = NULL;
12499         }
12500 
12501         err = (*ipip->ipi_func_restart)(ipsq->ipsq_xop->ipx_current_ipif, sin,
12502             q, mp, ipip, mp1->b_rptr);
12503 
12504         DTRACE_PROBE4(ipif__ioctl, char *, "ip_reprocess_ioctl finish",
12505             int, ipip->ipi_cmd,
12506             ill_t *, ipsq->ipsq_xop->ipx_current_ipif->ipif_ill,
12507             ipif_t *, ipsq->ipsq_xop->ipx_current_ipif);
12508 
12509         ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), ipsq);
12510 }
12511 
12512 /*
12513  * ioctl processing
12514  *
12515  * ioctl processing starts with ip_sioctl_copyin_setup(), which looks up
12516  * the ioctl command in the ioctl tables, determines the copyin data size
12517  * from the ipi_copyin_size field, and does an mi_copyin() of that size.
12518  *
12519  * ioctl processing then continues when the M_IOCDATA makes its way down to
12520  * ip_wput_nondata().  The ioctl is looked up again in the ioctl table, its
12521  * associated 'conn' is refheld till the end of the ioctl and the general
12522  * ioctl processing function ip_process_ioctl() is called to extract the
12523  * arguments and process the ioctl.  To simplify extraction, ioctl commands
12524  * are "typed" based on the arguments they take (e.g., LIF_CMD which takes a
12525  * `struct lifreq'), and a common extract function (e.g., ip_extract_lifreq())
12526  * is used to extract the ioctl's arguments.
12527  *
12528  * ip_process_ioctl determines if the ioctl needs to be serialized, and if
12529  * so goes thru the serialization primitive ipsq_try_enter. Then the
12530  * appropriate function to handle the ioctl is called based on the entry in
12531  * the ioctl table. ioctl completion is encapsulated in ip_ioctl_finish
12532  * which also refreleases the 'conn' that was refheld at the start of the
12533  * ioctl. Finally ipsq_exit is called if needed to exit the ipsq.
12534  *
12535  * Many exclusive ioctls go thru an internal down up sequence as part of
12536  * the operation. For example an attempt to change the IP address of an
12537  * ipif entails ipif_down, set address, ipif_up. Bringing down the interface
12538  * does all the cleanup such as deleting all ires that use this address.
12539  * Then we need to wait till all references to the interface go away.
12540  */
12541 void
12542 ip_process_ioctl(ipsq_t *ipsq, queue_t *q, mblk_t *mp, void *arg)
12543 {
12544         struct iocblk *iocp = (struct iocblk *)mp->b_rptr;
12545         ip_ioctl_cmd_t *ipip = arg;
12546         ip_extract_func_t *extract_funcp;
12547         cmd_info_t ci;
12548         int err;
12549         boolean_t entered_ipsq = B_FALSE;
12550 
12551         ip3dbg(("ip_process_ioctl: ioctl %X\n", iocp->ioc_cmd));
12552 
12553         if (ipip == NULL)
12554                 ipip = ip_sioctl_lookup(iocp->ioc_cmd);
12555 
12556         /*
12557          * SIOCLIFADDIF needs to go thru a special path since the
12558          * ill may not exist yet. This happens in the case of lo0
12559          * which is created using this ioctl.
12560          */
12561         if (ipip->ipi_cmd == SIOCLIFADDIF) {
12562                 err = ip_sioctl_addif(NULL, NULL, q, mp, NULL, NULL);
12563                 DTRACE_PROBE4(ipif__ioctl, char *, "ip_process_ioctl finish",
12564                     int, ipip->ipi_cmd, ill_t *, NULL, ipif_t *, NULL);
12565                 ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), NULL);
12566                 return;
12567         }
12568 
12569         ci.ci_ipif = NULL;
12570         switch (ipip->ipi_cmd_type) {
12571         case MISC_CMD:
12572         case MSFILT_CMD:
12573                 /*
12574                  * All MISC_CMD ioctls come in here -- e.g. SIOCGLIFCONF.
12575                  */
12576                 if (ipip->ipi_cmd == IF_UNITSEL) {
12577                         /* ioctl comes down the ill */
12578                         ci.ci_ipif = ((ill_t *)q->q_ptr)->ill_ipif;
12579                         ipif_refhold(ci.ci_ipif);
12580                 }
12581                 err = 0;
12582                 ci.ci_sin = NULL;
12583                 ci.ci_sin6 = NULL;
12584                 ci.ci_lifr = NULL;
12585                 extract_funcp = NULL;
12586                 break;
12587 
12588         case IF_CMD:
12589         case LIF_CMD:
12590                 extract_funcp = ip_extract_lifreq;
12591                 break;
12592 
12593         case ARP_CMD:
12594         case XARP_CMD:
12595                 extract_funcp = ip_extract_arpreq;
12596                 break;
12597 
12598         default:
12599                 ASSERT(0);
12600         }
12601 
12602         if (extract_funcp != NULL) {
12603                 err = (*extract_funcp)(q, mp, ipip, &ci);
12604                 if (err != 0) {
12605                         DTRACE_PROBE4(ipif__ioctl,
12606                             char *, "ip_process_ioctl finish err",
12607                             int, ipip->ipi_cmd, ill_t *, NULL, ipif_t *, NULL);
12608                         ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), NULL);
12609                         return;
12610                 }
12611 
12612                 /*
12613                  * All of the extraction functions return a refheld ipif.
12614                  */
12615                 ASSERT(ci.ci_ipif != NULL);
12616         }
12617 
12618         if (!(ipip->ipi_flags & IPI_WR)) {
12619                 /*
12620                  * A return value of EINPROGRESS means the ioctl is
12621                  * either queued and waiting for some reason or has
12622                  * already completed.
12623                  */
12624                 err = (*ipip->ipi_func)(ci.ci_ipif, ci.ci_sin, q, mp, ipip,
12625                     ci.ci_lifr);
12626                 if (ci.ci_ipif != NULL) {
12627                         DTRACE_PROBE4(ipif__ioctl,
12628                             char *, "ip_process_ioctl finish RD",
12629                             int, ipip->ipi_cmd, ill_t *, ci.ci_ipif->ipif_ill,
12630                             ipif_t *, ci.ci_ipif);
12631                         ipif_refrele(ci.ci_ipif);
12632                 } else {
12633                         DTRACE_PROBE4(ipif__ioctl,
12634                             char *, "ip_process_ioctl finish RD",
12635                             int, ipip->ipi_cmd, ill_t *, NULL, ipif_t *, NULL);
12636                 }
12637                 ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), NULL);
12638                 return;
12639         }
12640 
12641         ASSERT(ci.ci_ipif != NULL);
12642 
12643         /*
12644          * If ipsq is non-NULL, we are already being called exclusively
12645          */
12646         ASSERT(ipsq == NULL || IAM_WRITER_IPSQ(ipsq));
12647         if (ipsq == NULL) {
12648                 ipsq = ipsq_try_enter(ci.ci_ipif, NULL, q, mp, ip_process_ioctl,
12649                     NEW_OP, B_TRUE);
12650                 if (ipsq == NULL) {
12651                         ipif_refrele(ci.ci_ipif);
12652                         return;
12653                 }
12654                 entered_ipsq = B_TRUE;
12655         }
12656         /*
12657          * Release the ipif so that ipif_down and friends that wait for
12658          * references to go away are not misled about the current ipif_refcnt
12659          * values. We are writer so we can access the ipif even after releasing
12660          * the ipif.
12661          */
12662         ipif_refrele(ci.ci_ipif);
12663 
12664         ipsq_current_start(ipsq, ci.ci_ipif, ipip->ipi_cmd);
12665 
12666         /*
12667          * A return value of EINPROGRESS means the ioctl is
12668          * either queued and waiting for some reason or has
12669          * already completed.
12670          */
12671         err = (*ipip->ipi_func)(ci.ci_ipif, ci.ci_sin, q, mp, ipip, ci.ci_lifr);
12672 
12673         DTRACE_PROBE4(ipif__ioctl, char *, "ip_process_ioctl finish WR",
12674             int, ipip->ipi_cmd,
12675             ill_t *, ci.ci_ipif == NULL ? NULL : ci.ci_ipif->ipif_ill,
12676             ipif_t *, ci.ci_ipif);
12677         ip_ioctl_finish(q, mp, err, IPI2MODE(ipip), ipsq);
12678 
12679         if (entered_ipsq)
12680                 ipsq_exit(ipsq);
12681 }
12682 
12683 /*
12684  * Complete the ioctl. Typically ioctls use the mi package and need to
12685  * do mi_copyout/mi_copy_done.
12686  */
12687 void
12688 ip_ioctl_finish(queue_t *q, mblk_t *mp, int err, int mode, ipsq_t *ipsq)
12689 {
12690         conn_t  *connp = NULL;
12691 
12692         if (err == EINPROGRESS)
12693                 return;
12694 
12695         if (CONN_Q(q)) {
12696                 connp = Q_TO_CONN(q);
12697                 ASSERT(connp->conn_ref >= 2);
12698         }
12699 
12700         switch (mode) {
12701         case COPYOUT:
12702                 if (err == 0)
12703                         mi_copyout(q, mp);
12704                 else
12705                         mi_copy_done(q, mp, err);
12706                 break;
12707 
12708         case NO_COPYOUT:
12709                 mi_copy_done(q, mp, err);
12710                 break;
12711 
12712         default:
12713                 ASSERT(mode == CONN_CLOSE);     /* aborted through CONN_CLOSE */
12714                 break;
12715         }
12716 
12717         /*
12718          * The conn refhold and ioctlref placed on the conn at the start of the
12719          * ioctl are released here.
12720          */
12721         if (connp != NULL) {
12722                 CONN_DEC_IOCTLREF(connp);
12723                 CONN_OPER_PENDING_DONE(connp);
12724         }
12725 
12726         if (ipsq != NULL)
12727                 ipsq_current_finish(ipsq);
12728 }
12729 
12730 /* Handles all non data messages */
12731 void
12732 ip_wput_nondata(queue_t *q, mblk_t *mp)
12733 {
12734         mblk_t          *mp1;
12735         struct iocblk   *iocp;
12736         ip_ioctl_cmd_t  *ipip;
12737         conn_t          *connp;
12738         cred_t          *cr;
12739         char            *proto_str;
12740 
12741         if (CONN_Q(q))
12742                 connp = Q_TO_CONN(q);
12743         else
12744                 connp = NULL;
12745 
12746         switch (DB_TYPE(mp)) {
12747         case M_IOCTL:
12748                 /*
12749                  * IOCTL processing begins in ip_sioctl_copyin_setup which
12750                  * will arrange to copy in associated control structures.
12751                  */
12752                 ip_sioctl_copyin_setup(q, mp);
12753                 return;
12754         case M_IOCDATA:
12755                 /*
12756                  * Ensure that this is associated with one of our trans-
12757                  * parent ioctls.  If it's not ours, discard it if we're
12758                  * running as a driver, or pass it on if we're a module.
12759                  */
12760                 iocp = (struct iocblk *)mp->b_rptr;
12761                 ipip = ip_sioctl_lookup(iocp->ioc_cmd);
12762                 if (ipip == NULL) {
12763                         if (q->q_next == NULL) {
12764                                 goto nak;
12765                         } else {
12766                                 putnext(q, mp);
12767                         }
12768                         return;
12769                 }
12770                 if ((q->q_next != NULL) && !(ipip->ipi_flags & IPI_MODOK)) {
12771                         /*
12772                          * The ioctl is one we recognise, but is not consumed
12773                          * by IP as a module and we are a module, so we drop
12774                          */
12775                         goto nak;
12776                 }
12777 
12778                 /* IOCTL continuation following copyin or copyout. */
12779                 if (mi_copy_state(q, mp, NULL) == -1) {
12780                         /*
12781                          * The copy operation failed.  mi_copy_state already
12782                          * cleaned up, so we're out of here.
12783                          */
12784                         return;
12785                 }
12786                 /*
12787                  * If we just completed a copy in, we become writer and
12788                  * continue processing in ip_sioctl_copyin_done.  If it
12789                  * was a copy out, we call mi_copyout again.  If there is
12790                  * nothing more to copy out, it will complete the IOCTL.
12791                  */
12792                 if (MI_COPY_DIRECTION(mp) == MI_COPY_IN) {
12793                         if (!(mp1 = mp->b_cont) || !(mp1 = mp1->b_cont)) {
12794                                 mi_copy_done(q, mp, EPROTO);
12795                                 return;
12796                         }
12797                         /*
12798                          * Check for cases that need more copying.  A return
12799                          * value of 0 means a second copyin has been started,
12800                          * so we return; a return value of 1 means no more
12801                          * copying is needed, so we continue.
12802                          */
12803                         if (ipip->ipi_cmd_type == MSFILT_CMD &&
12804                             MI_COPY_COUNT(mp) == 1) {
12805                                 if (ip_copyin_msfilter(q, mp) == 0)
12806                                         return;
12807                         }
12808                         /*
12809                          * Refhold the conn, till the ioctl completes. This is
12810                          * needed in case the ioctl ends up in the pending mp
12811                          * list. Every mp in the ipx_pending_mp list must have
12812                          * a refhold on the conn to resume processing. The
12813                          * refhold is released when the ioctl completes
12814                          * (whether normally or abnormally). An ioctlref is also
12815                          * placed on the conn to prevent TCP from removing the
12816                          * queue needed to send the ioctl reply back.
12817                          * In all cases ip_ioctl_finish is called to finish
12818                          * the ioctl and release the refholds.
12819                          */
12820                         if (connp != NULL) {
12821                                 /* This is not a reentry */
12822                                 CONN_INC_REF(connp);
12823                                 CONN_INC_IOCTLREF(connp);
12824                         } else {
12825                                 if (!(ipip->ipi_flags & IPI_MODOK)) {
12826                                         mi_copy_done(q, mp, EINVAL);
12827                                         return;
12828                                 }
12829                         }
12830 
12831                         ip_process_ioctl(NULL, q, mp, ipip);
12832 
12833                 } else {
12834                         mi_copyout(q, mp);
12835                 }
12836                 return;
12837 
12838         case M_IOCNAK:
12839                 /*
12840                  * The only way we could get here is if a resolver didn't like
12841                  * an IOCTL we sent it.  This shouldn't happen.
12842                  */
12843                 (void) mi_strlog(q, 1, SL_ERROR|SL_TRACE,
12844                     "ip_wput_nondata: unexpected M_IOCNAK, ioc_cmd 0x%x",
12845                     ((struct iocblk *)mp->b_rptr)->ioc_cmd);
12846                 freemsg(mp);
12847                 return;
12848         case M_IOCACK:
12849                 /* /dev/ip shouldn't see this */
12850                 goto nak;
12851         case M_FLUSH:
12852                 if (*mp->b_rptr & FLUSHW)
12853                         flushq(q, FLUSHALL);
12854                 if (q->q_next) {
12855                         putnext(q, mp);
12856                         return;
12857                 }
12858                 if (*mp->b_rptr & FLUSHR) {
12859                         *mp->b_rptr &= ~FLUSHW;
12860                         qreply(q, mp);
12861                         return;
12862                 }
12863                 freemsg(mp);
12864                 return;
12865         case M_CTL:
12866                 break;
12867         case M_PROTO:
12868         case M_PCPROTO:
12869                 /*
12870                  * The only PROTO messages we expect are SNMP-related.
12871                  */
12872                 switch (((union T_primitives *)mp->b_rptr)->type) {
12873                 case T_SVR4_OPTMGMT_REQ:
12874                         ip2dbg(("ip_wput_nondata: T_SVR4_OPTMGMT_REQ "
12875                             "flags %x\n",
12876                             ((struct T_optmgmt_req *)mp->b_rptr)->MGMT_flags));
12877 
12878                         if (connp == NULL) {
12879                                 proto_str = "T_SVR4_OPTMGMT_REQ";
12880                                 goto protonak;
12881                         }
12882 
12883                         /*
12884                          * All Solaris components should pass a db_credp
12885                          * for this TPI message, hence we ASSERT.
12886                          * But in case there is some other M_PROTO that looks
12887                          * like a TPI message sent by some other kernel
12888                          * component, we check and return an error.
12889                          */
12890                         cr = msg_getcred(mp, NULL);
12891                         ASSERT(cr != NULL);
12892                         if (cr == NULL) {
12893                                 mp = mi_tpi_err_ack_alloc(mp, TSYSERR, EINVAL);
12894                                 if (mp != NULL)
12895                                         qreply(q, mp);
12896                                 return;
12897                         }
12898 
12899                         if (!snmpcom_req(q, mp, ip_snmp_set, ip_snmp_get, cr)) {
12900                                 proto_str = "Bad SNMPCOM request?";
12901                                 goto protonak;
12902                         }
12903                         return;
12904                 default:
12905                         ip1dbg(("ip_wput_nondata: dropping M_PROTO prim %u\n",
12906                             (int)*(uint_t *)mp->b_rptr));
12907                         freemsg(mp);
12908                         return;
12909                 }
12910         default:
12911                 break;
12912         }
12913         if (q->q_next) {
12914                 putnext(q, mp);
12915         } else
12916                 freemsg(mp);
12917         return;
12918 
12919 nak:
12920         iocp->ioc_error = EINVAL;
12921         mp->b_datap->db_type = M_IOCNAK;
12922         iocp->ioc_count = 0;
12923         qreply(q, mp);
12924         return;
12925 
12926 protonak:
12927         cmn_err(CE_NOTE, "IP doesn't process %s as a module", proto_str);
12928         if ((mp = mi_tpi_err_ack_alloc(mp, TPROTO, EINVAL)) != NULL)
12929                 qreply(q, mp);
12930 }
12931 
12932 /*
12933  * Process IP options in an outbound packet.  Verify that the nexthop in a
12934  * strict source route is onlink.
12935  * Returns non-zero if something fails in which case an ICMP error has been
12936  * sent and mp freed.
12937  *
12938  * Assumes the ULP has called ip_massage_options to move nexthop into ipha_dst.
12939  */
12940 int
12941 ip_output_options(mblk_t *mp, ipha_t *ipha, ip_xmit_attr_t *ixa, ill_t *ill)
12942 {
12943         ipoptp_t        opts;
12944         uchar_t         *opt;
12945         uint8_t         optval;
12946         uint8_t         optlen;
12947         ipaddr_t        dst;
12948         intptr_t        code = 0;
12949         ire_t           *ire;
12950         ip_stack_t      *ipst = ixa->ixa_ipst;
12951         ip_recv_attr_t  iras;
12952 
12953         ip2dbg(("ip_output_options\n"));
12954 
12955         dst = ipha->ipha_dst;
12956         for (optval = ipoptp_first(&opts, ipha);
12957             optval != IPOPT_EOL;
12958             optval = ipoptp_next(&opts)) {
12959                 opt = opts.ipoptp_cur;
12960                 optlen = opts.ipoptp_len;
12961                 ip2dbg(("ip_output_options: opt %d, len %d\n",
12962                     optval, optlen));
12963                 switch (optval) {
12964                         uint32_t off;
12965                 case IPOPT_SSRR:
12966                 case IPOPT_LSRR:
12967                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
12968                                 ip1dbg((
12969                                     "ip_output_options: bad option offset\n"));
12970                                 code = (char *)&opt[IPOPT_OLEN] -
12971                                     (char *)ipha;
12972                                 goto param_prob;
12973                         }
12974                         off = opt[IPOPT_OFFSET];
12975                         ip1dbg(("ip_output_options: next hop 0x%x\n",
12976                             ntohl(dst)));
12977                         /*
12978                          * For strict: verify that dst is directly
12979                          * reachable.
12980                          */
12981                         if (optval == IPOPT_SSRR) {
12982                                 ire = ire_ftable_lookup_v4(dst, 0, 0,
12983                                     IRE_INTERFACE, NULL, ALL_ZONES,
12984                                     ixa->ixa_tsl,
12985                                     MATCH_IRE_TYPE | MATCH_IRE_SECATTR, 0, ipst,
12986                                     NULL);
12987                                 if (ire == NULL) {
12988                                         ip1dbg(("ip_output_options: SSRR not"
12989                                             " directly reachable: 0x%x\n",
12990                                             ntohl(dst)));
12991                                         goto bad_src_route;
12992                                 }
12993                                 ire_refrele(ire);
12994                         }
12995                         break;
12996                 case IPOPT_RR:
12997                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
12998                                 ip1dbg((
12999                                     "ip_output_options: bad option offset\n"));
13000                                 code = (char *)&opt[IPOPT_OLEN] -
13001                                     (char *)ipha;
13002                                 goto param_prob;
13003                         }
13004                         break;
13005                 case IPOPT_TS:
13006                         /*
13007                          * Verify that length >=5 and that there is either
13008                          * room for another timestamp or that the overflow
13009                          * counter is not maxed out.
13010                          */
13011                         code = (char *)&opt[IPOPT_OLEN] - (char *)ipha;
13012                         if (optlen < IPOPT_MINLEN_IT) {
13013                                 goto param_prob;
13014                         }
13015                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
13016                                 ip1dbg((
13017                                     "ip_output_options: bad option offset\n"));
13018                                 code = (char *)&opt[IPOPT_OFFSET] -
13019                                     (char *)ipha;
13020                                 goto param_prob;
13021                         }
13022                         switch (opt[IPOPT_POS_OV_FLG] & 0x0F) {
13023                         case IPOPT_TS_TSONLY:
13024                                 off = IPOPT_TS_TIMELEN;
13025                                 break;
13026                         case IPOPT_TS_TSANDADDR:
13027                         case IPOPT_TS_PRESPEC:
13028                         case IPOPT_TS_PRESPEC_RFC791:
13029                                 off = IP_ADDR_LEN + IPOPT_TS_TIMELEN;
13030                                 break;
13031                         default:
13032                                 code = (char *)&opt[IPOPT_POS_OV_FLG] -
13033                                     (char *)ipha;
13034                                 goto param_prob;
13035                         }
13036                         if (opt[IPOPT_OFFSET] - 1 + off > optlen &&
13037                             (opt[IPOPT_POS_OV_FLG] & 0xF0) == 0xF0) {
13038                                 /*
13039                                  * No room and the overflow counter is 15
13040                                  * already.
13041                                  */
13042                                 goto param_prob;
13043                         }
13044                         break;
13045                 }
13046         }
13047 
13048         if ((opts.ipoptp_flags & IPOPTP_ERROR) == 0)
13049                 return (0);
13050 
13051         ip1dbg(("ip_output_options: error processing IP options."));
13052         code = (char *)&opt[IPOPT_OFFSET] - (char *)ipha;
13053 
13054 param_prob:
13055         bzero(&iras, sizeof (iras));
13056         iras.ira_ill = iras.ira_rill = ill;
13057         iras.ira_ruifindex = ill->ill_phyint->phyint_ifindex;
13058         iras.ira_rifindex = iras.ira_ruifindex;
13059         iras.ira_flags = IRAF_IS_IPV4;
13060 
13061         ip_drop_output("ip_output_options", mp, ill);
13062         icmp_param_problem(mp, (uint8_t)code, &iras);
13063         ASSERT(!(iras.ira_flags & IRAF_IPSEC_SECURE));
13064         return (-1);
13065 
13066 bad_src_route:
13067         bzero(&iras, sizeof (iras));
13068         iras.ira_ill = iras.ira_rill = ill;
13069         iras.ira_ruifindex = ill->ill_phyint->phyint_ifindex;
13070         iras.ira_rifindex = iras.ira_ruifindex;
13071         iras.ira_flags = IRAF_IS_IPV4;
13072 
13073         ip_drop_input("ICMP_SOURCE_ROUTE_FAILED", mp, ill);
13074         icmp_unreachable(mp, ICMP_SOURCE_ROUTE_FAILED, &iras);
13075         ASSERT(!(iras.ira_flags & IRAF_IPSEC_SECURE));
13076         return (-1);
13077 }
13078 
13079 /*
13080  * The maximum value of conn_drain_list_cnt is CONN_MAXDRAINCNT.
13081  * conn_drain_list_cnt can be changed by setting conn_drain_nthreads
13082  * thru /etc/system.
13083  */
13084 #define CONN_MAXDRAINCNT        64
13085 
13086 static void
13087 conn_drain_init(ip_stack_t *ipst)
13088 {
13089         int i, j;
13090         idl_tx_list_t *itl_tx;
13091 
13092         ipst->ips_conn_drain_list_cnt = conn_drain_nthreads;
13093 
13094         if ((ipst->ips_conn_drain_list_cnt == 0) ||
13095             (ipst->ips_conn_drain_list_cnt > CONN_MAXDRAINCNT)) {
13096                 /*
13097                  * Default value of the number of drainers is the
13098                  * number of cpus, subject to maximum of 8 drainers.
13099                  */
13100                 if (boot_max_ncpus != -1)
13101                         ipst->ips_conn_drain_list_cnt = MIN(boot_max_ncpus, 8);
13102                 else
13103                         ipst->ips_conn_drain_list_cnt = MIN(max_ncpus, 8);
13104         }
13105 
13106         ipst->ips_idl_tx_list =
13107             kmem_zalloc(TX_FANOUT_SIZE * sizeof (idl_tx_list_t), KM_SLEEP);
13108         for (i = 0; i < TX_FANOUT_SIZE; i++) {
13109                 itl_tx =  &ipst->ips_idl_tx_list[i];
13110                 itl_tx->txl_drain_list =
13111                     kmem_zalloc(ipst->ips_conn_drain_list_cnt *
13112                     sizeof (idl_t), KM_SLEEP);
13113                 mutex_init(&itl_tx->txl_lock, NULL, MUTEX_DEFAULT, NULL);
13114                 for (j = 0; j < ipst->ips_conn_drain_list_cnt; j++) {
13115                         mutex_init(&itl_tx->txl_drain_list[j].idl_lock, NULL,
13116                             MUTEX_DEFAULT, NULL);
13117                         itl_tx->txl_drain_list[j].idl_itl = itl_tx;
13118                 }
13119         }
13120 }
13121 
13122 static void
13123 conn_drain_fini(ip_stack_t *ipst)
13124 {
13125         int i;
13126         idl_tx_list_t *itl_tx;
13127 
13128         for (i = 0; i < TX_FANOUT_SIZE; i++) {
13129                 itl_tx =  &ipst->ips_idl_tx_list[i];
13130                 kmem_free(itl_tx->txl_drain_list,
13131                     ipst->ips_conn_drain_list_cnt * sizeof (idl_t));
13132         }
13133         kmem_free(ipst->ips_idl_tx_list,
13134             TX_FANOUT_SIZE * sizeof (idl_tx_list_t));
13135         ipst->ips_idl_tx_list = NULL;
13136 }
13137 
13138 /*
13139  * Flow control has blocked us from proceeding.  Insert the given conn in one
13140  * of the conn drain lists.  When flow control is unblocked, either ip_wsrv()
13141  * (STREAMS) or ill_flow_enable() (direct) will be called back, which in turn
13142  * will call conn_walk_drain().  See the flow control notes at the top of this
13143  * file for more details.
13144  */
13145 void
13146 conn_drain_insert(conn_t *connp, idl_tx_list_t *tx_list)
13147 {
13148         idl_t   *idl = tx_list->txl_drain_list;
13149         uint_t  index;
13150         ip_stack_t      *ipst = connp->conn_netstack->netstack_ip;
13151 
13152         mutex_enter(&connp->conn_lock);
13153         if (connp->conn_state_flags & CONN_CLOSING) {
13154                 /*
13155                  * The conn is closing as a result of which CONN_CLOSING
13156                  * is set. Return.
13157                  */
13158                 mutex_exit(&connp->conn_lock);
13159                 return;
13160         } else if (connp->conn_idl == NULL) {
13161                 /*
13162                  * Assign the next drain list round robin. We dont' use
13163                  * a lock, and thus it may not be strictly round robin.
13164                  * Atomicity of load/stores is enough to make sure that
13165                  * conn_drain_list_index is always within bounds.
13166                  */
13167                 index = tx_list->txl_drain_index;
13168                 ASSERT(index < ipst->ips_conn_drain_list_cnt);
13169                 connp->conn_idl = &tx_list->txl_drain_list[index];
13170                 index++;
13171                 if (index == ipst->ips_conn_drain_list_cnt)
13172                         index = 0;
13173                 tx_list->txl_drain_index = index;
13174         } else {
13175                 ASSERT(connp->conn_idl->idl_itl == tx_list);
13176         }
13177         mutex_exit(&connp->conn_lock);
13178 
13179         idl = connp->conn_idl;
13180         mutex_enter(&idl->idl_lock);
13181         if ((connp->conn_drain_prev != NULL) ||
13182             (connp->conn_state_flags & CONN_CLOSING)) {
13183                 /*
13184                  * The conn is either already in the drain list or closing.
13185                  * (We needed to check for CONN_CLOSING again since close can
13186                  * sneak in between dropping conn_lock and acquiring idl_lock.)
13187                  */
13188                 mutex_exit(&idl->idl_lock);
13189                 return;
13190         }
13191 
13192         /*
13193          * The conn is not in the drain list. Insert it at the
13194          * tail of the drain list. The drain list is circular
13195          * and doubly linked. idl_conn points to the 1st element
13196          * in the list.
13197          */
13198         if (idl->idl_conn == NULL) {
13199                 idl->idl_conn = connp;
13200                 connp->conn_drain_next = connp;
13201                 connp->conn_drain_prev = connp;
13202         } else {
13203                 conn_t *head = idl->idl_conn;
13204 
13205                 connp->conn_drain_next = head;
13206                 connp->conn_drain_prev = head->conn_drain_prev;
13207                 head->conn_drain_prev->conn_drain_next = connp;
13208                 head->conn_drain_prev = connp;
13209         }
13210         /*
13211          * For non streams based sockets assert flow control.
13212          */
13213         conn_setqfull(connp, NULL);
13214         mutex_exit(&idl->idl_lock);
13215 }
13216 
13217 static void
13218 conn_drain_remove(conn_t *connp)
13219 {
13220         idl_t *idl = connp->conn_idl;
13221 
13222         if (idl != NULL) {
13223                 /*
13224                  * Remove ourself from the drain list.
13225                  */
13226                 if (connp->conn_drain_next == connp) {
13227                         /* Singleton in the list */
13228                         ASSERT(connp->conn_drain_prev == connp);
13229                         idl->idl_conn = NULL;
13230                 } else {
13231                         connp->conn_drain_prev->conn_drain_next =
13232                             connp->conn_drain_next;
13233                         connp->conn_drain_next->conn_drain_prev =
13234                             connp->conn_drain_prev;
13235                         if (idl->idl_conn == connp)
13236                                 idl->idl_conn = connp->conn_drain_next;
13237                 }
13238 
13239                 /*
13240                  * NOTE: because conn_idl is associated with a specific drain
13241                  * list which in turn is tied to the index the TX ring
13242                  * (txl_cookie) hashes to, and because the TX ring can change
13243                  * over the lifetime of the conn_t, we must clear conn_idl so
13244                  * a subsequent conn_drain_insert() will set conn_idl again
13245                  * based on the latest txl_cookie.
13246                  */
13247                 connp->conn_idl = NULL;
13248         }
13249         connp->conn_drain_next = NULL;
13250         connp->conn_drain_prev = NULL;
13251 
13252         conn_clrqfull(connp, NULL);
13253         /*
13254          * For streams based sockets open up flow control.
13255          */
13256         if (!IPCL_IS_NONSTR(connp))
13257                 enableok(connp->conn_wq);
13258 }
13259 
13260 /*
13261  * This conn is closing, and we are called from ip_close. OR
13262  * this conn is draining because flow-control on the ill has been relieved.
13263  *
13264  * We must also need to remove conn's on this idl from the list, and also
13265  * inform the sockfs upcalls about the change in flow-control.
13266  */
13267 static void
13268 conn_drain(conn_t *connp, boolean_t closing)
13269 {
13270         idl_t *idl;
13271         conn_t *next_connp;
13272 
13273         /*
13274          * connp->conn_idl is stable at this point, and no lock is needed
13275          * to check it. If we are called from ip_close, close has already
13276          * set CONN_CLOSING, thus freezing the value of conn_idl, and
13277          * called us only because conn_idl is non-null. If we are called thru
13278          * service, conn_idl could be null, but it cannot change because
13279          * service is single-threaded per queue, and there cannot be another
13280          * instance of service trying to call conn_drain_insert on this conn
13281          * now.
13282          */
13283         ASSERT(!closing || connp == NULL || connp->conn_idl != NULL);
13284 
13285         /*
13286          * If the conn doesn't exist or is not on a drain list, bail.
13287          */
13288         if (connp == NULL || connp->conn_idl == NULL ||
13289             connp->conn_drain_prev == NULL) {
13290                 return;
13291         }
13292 
13293         idl = connp->conn_idl;
13294         ASSERT(MUTEX_HELD(&idl->idl_lock));
13295 
13296         if (!closing) {
13297                 next_connp = connp->conn_drain_next;
13298                 while (next_connp != connp) {
13299                         conn_t *delconnp = next_connp;
13300 
13301                         next_connp = next_connp->conn_drain_next;
13302                         conn_drain_remove(delconnp);
13303                 }
13304                 ASSERT(connp->conn_drain_next == idl->idl_conn);
13305         }
13306         conn_drain_remove(connp);
13307 }
13308 
13309 /*
13310  * Write service routine. Shared perimeter entry point.
13311  * The device queue's messages has fallen below the low water mark and STREAMS
13312  * has backenabled the ill_wq. Send sockfs notification about flow-control on
13313  * each waiting conn.
13314  */
13315 void
13316 ip_wsrv(queue_t *q)
13317 {
13318         ill_t   *ill;
13319 
13320         ill = (ill_t *)q->q_ptr;
13321         if (ill->ill_state_flags == 0) {
13322                 ip_stack_t *ipst = ill->ill_ipst;
13323 
13324                 /*
13325                  * The device flow control has opened up.
13326                  * Walk through conn drain lists and qenable the
13327                  * first conn in each list. This makes sense only
13328                  * if the stream is fully plumbed and setup.
13329                  * Hence the ill_state_flags check above.
13330                  */
13331                 ip1dbg(("ip_wsrv: walking\n"));
13332                 conn_walk_drain(ipst, &ipst->ips_idl_tx_list[0]);
13333                 enableok(ill->ill_wq);
13334         }
13335 }
13336 
13337 /*
13338  * Callback to disable flow control in IP.
13339  *
13340  * This is a mac client callback added when the DLD_CAPAB_DIRECT capability
13341  * is enabled.
13342  *
13343  * When MAC_TX() is not able to send any more packets, dld sets its queue
13344  * to QFULL and enable the STREAMS flow control. Later, when the underlying
13345  * driver is able to continue to send packets, it calls mac_tx_(ring_)update()
13346  * function and wakes up corresponding mac worker threads, which in turn
13347  * calls this callback function, and disables flow control.
13348  */
13349 void
13350 ill_flow_enable(void *arg, ip_mac_tx_cookie_t cookie)
13351 {
13352         ill_t *ill = (ill_t *)arg;
13353         ip_stack_t *ipst = ill->ill_ipst;
13354         idl_tx_list_t *idl_txl;
13355 
13356         idl_txl = &ipst->ips_idl_tx_list[IDLHASHINDEX(cookie)];
13357         mutex_enter(&idl_txl->txl_lock);
13358         /* add code to to set a flag to indicate idl_txl is enabled */
13359         conn_walk_drain(ipst, idl_txl);
13360         mutex_exit(&idl_txl->txl_lock);
13361 }
13362 
13363 /*
13364  * Flow control has been relieved and STREAMS has backenabled us; drain
13365  * all the conn lists on `tx_list'.
13366  */
13367 static void
13368 conn_walk_drain(ip_stack_t *ipst, idl_tx_list_t *tx_list)
13369 {
13370         int i;
13371         idl_t *idl;
13372 
13373         IP_STAT(ipst, ip_conn_walk_drain);
13374 
13375         for (i = 0; i < ipst->ips_conn_drain_list_cnt; i++) {
13376                 idl = &tx_list->txl_drain_list[i];
13377                 mutex_enter(&idl->idl_lock);
13378                 conn_drain(idl->idl_conn, B_FALSE);
13379                 mutex_exit(&idl->idl_lock);
13380         }
13381 }
13382 
13383 /*
13384  * Determine if the ill and multicast aspects of that packets
13385  * "matches" the conn.
13386  */
13387 boolean_t
13388 conn_wantpacket(conn_t *connp, ip_recv_attr_t *ira, ipha_t *ipha)
13389 {
13390         ill_t           *ill = ira->ira_rill;
13391         zoneid_t        zoneid = ira->ira_zoneid;
13392         uint_t          in_ifindex;
13393         ipaddr_t        dst, src;
13394 
13395         dst = ipha->ipha_dst;
13396         src = ipha->ipha_src;
13397 
13398         /*
13399          * conn_incoming_ifindex is set by IP_BOUND_IF which limits
13400          * unicast, broadcast and multicast reception to
13401          * conn_incoming_ifindex.
13402          * conn_wantpacket is called for unicast, broadcast and
13403          * multicast packets.
13404          */
13405         in_ifindex = connp->conn_incoming_ifindex;
13406 
13407         /* mpathd can bind to the under IPMP interface, which we allow */
13408         if (in_ifindex != 0 && in_ifindex != ill->ill_phyint->phyint_ifindex) {
13409                 if (!IS_UNDER_IPMP(ill))
13410                         return (B_FALSE);
13411 
13412                 if (in_ifindex != ipmp_ill_get_ipmp_ifindex(ill))
13413                         return (B_FALSE);
13414         }
13415 
13416         if (!IPCL_ZONE_MATCH(connp, zoneid))
13417                 return (B_FALSE);
13418 
13419         if (!(ira->ira_flags & IRAF_MULTICAST))
13420                 return (B_TRUE);
13421 
13422         if (connp->conn_multi_router) {
13423                 /* multicast packet and multicast router socket: send up */
13424                 return (B_TRUE);
13425         }
13426 
13427         if (ipha->ipha_protocol == IPPROTO_PIM ||
13428             ipha->ipha_protocol == IPPROTO_RSVP)
13429                 return (B_TRUE);
13430 
13431         return (conn_hasmembers_ill_withsrc_v4(connp, dst, src, ira->ira_ill));
13432 }
13433 
13434 void
13435 conn_setqfull(conn_t *connp, boolean_t *flow_stopped)
13436 {
13437         if (IPCL_IS_NONSTR(connp)) {
13438                 (*connp->conn_upcalls->su_txq_full)
13439                     (connp->conn_upper_handle, B_TRUE);
13440                 if (flow_stopped != NULL)
13441                         *flow_stopped = B_TRUE;
13442         } else {
13443                 queue_t *q = connp->conn_wq;
13444 
13445                 ASSERT(q != NULL);
13446                 if (!(q->q_flag & QFULL)) {
13447                         mutex_enter(QLOCK(q));
13448                         if (!(q->q_flag & QFULL)) {
13449                                 /* still need to set QFULL */
13450                                 q->q_flag |= QFULL;
13451                                 /* set flow_stopped to true under QLOCK */
13452                                 if (flow_stopped != NULL)
13453                                         *flow_stopped = B_TRUE;
13454                                 mutex_exit(QLOCK(q));
13455                         } else {
13456                                 /* flow_stopped is left unchanged */
13457                                 mutex_exit(QLOCK(q));
13458                         }
13459                 }
13460         }
13461 }
13462 
13463 void
13464 conn_clrqfull(conn_t *connp, boolean_t *flow_stopped)
13465 {
13466         if (IPCL_IS_NONSTR(connp)) {
13467                 (*connp->conn_upcalls->su_txq_full)
13468                     (connp->conn_upper_handle, B_FALSE);
13469                 if (flow_stopped != NULL)
13470                         *flow_stopped = B_FALSE;
13471         } else {
13472                 queue_t *q = connp->conn_wq;
13473 
13474                 ASSERT(q != NULL);
13475                 if (q->q_flag & QFULL) {
13476                         mutex_enter(QLOCK(q));
13477                         if (q->q_flag & QFULL) {
13478                                 q->q_flag &= ~QFULL;
13479                                 /* set flow_stopped to false under QLOCK */
13480                                 if (flow_stopped != NULL)
13481                                         *flow_stopped = B_FALSE;
13482                                 mutex_exit(QLOCK(q));
13483                                 if (q->q_flag & QWANTW)
13484                                         qbackenable(q, 0);
13485                         } else {
13486                                 /* flow_stopped is left unchanged */
13487                                 mutex_exit(QLOCK(q));
13488                         }
13489                 }
13490         }
13491 
13492         mutex_enter(&connp->conn_lock);
13493         connp->conn_blocked = B_FALSE;
13494         mutex_exit(&connp->conn_lock);
13495 }
13496 
13497 /*
13498  * Return the length in bytes of the IPv4 headers (base header, label, and
13499  * other IP options) that will be needed based on the
13500  * ip_pkt_t structure passed by the caller.
13501  *
13502  * The returned length does not include the length of the upper level
13503  * protocol (ULP) header.
13504  * The caller needs to check that the length doesn't exceed the max for IPv4.
13505  */
13506 int
13507 ip_total_hdrs_len_v4(const ip_pkt_t *ipp)
13508 {
13509         int len;
13510 
13511         len = IP_SIMPLE_HDR_LENGTH;
13512         if (ipp->ipp_fields & IPPF_LABEL_V4) {
13513                 ASSERT(ipp->ipp_label_len_v4 != 0);
13514                 /* We need to round up here */
13515                 len += (ipp->ipp_label_len_v4 + 3) & ~3;
13516         }
13517 
13518         if (ipp->ipp_fields & IPPF_IPV4_OPTIONS) {
13519                 ASSERT(ipp->ipp_ipv4_options_len != 0);
13520                 ASSERT((ipp->ipp_ipv4_options_len & 3) == 0);
13521                 len += ipp->ipp_ipv4_options_len;
13522         }
13523         return (len);
13524 }
13525 
13526 /*
13527  * All-purpose routine to build an IPv4 header with options based
13528  * on the abstract ip_pkt_t.
13529  *
13530  * The caller has to set the source and destination address as well as
13531  * ipha_length. The caller has to massage any source route and compensate
13532  * for the ULP pseudo-header checksum due to the source route.
13533  */
13534 void
13535 ip_build_hdrs_v4(uchar_t *buf, uint_t buf_len, const ip_pkt_t *ipp,
13536     uint8_t protocol)
13537 {
13538         ipha_t  *ipha = (ipha_t *)buf;
13539         uint8_t *cp;
13540 
13541         /* Initialize IPv4 header */
13542         ipha->ipha_type_of_service = ipp->ipp_type_of_service;
13543         ipha->ipha_length = 0;       /* Caller will set later */
13544         ipha->ipha_ident = 0;
13545         ipha->ipha_fragment_offset_and_flags = 0;
13546         ipha->ipha_ttl = ipp->ipp_unicast_hops;
13547         ipha->ipha_protocol = protocol;
13548         ipha->ipha_hdr_checksum = 0;
13549 
13550         if ((ipp->ipp_fields & IPPF_ADDR) &&
13551             IN6_IS_ADDR_V4MAPPED(&ipp->ipp_addr))
13552                 ipha->ipha_src = ipp->ipp_addr_v4;
13553 
13554         cp = (uint8_t *)&ipha[1];
13555         if (ipp->ipp_fields & IPPF_LABEL_V4) {
13556                 ASSERT(ipp->ipp_label_len_v4 != 0);
13557                 bcopy(ipp->ipp_label_v4, cp, ipp->ipp_label_len_v4);
13558                 cp += ipp->ipp_label_len_v4;
13559                 /* We need to round up here */
13560                 while ((uintptr_t)cp & 0x3) {
13561                         *cp++ = IPOPT_NOP;
13562                 }
13563         }
13564 
13565         if (ipp->ipp_fields & IPPF_IPV4_OPTIONS) {
13566                 ASSERT(ipp->ipp_ipv4_options_len != 0);
13567                 ASSERT((ipp->ipp_ipv4_options_len & 3) == 0);
13568                 bcopy(ipp->ipp_ipv4_options, cp, ipp->ipp_ipv4_options_len);
13569                 cp += ipp->ipp_ipv4_options_len;
13570         }
13571         ipha->ipha_version_and_hdr_length =
13572             (uint8_t)((IP_VERSION << 4) + buf_len / 4);
13573 
13574         ASSERT((int)(cp - buf) == buf_len);
13575 }
13576 
13577 /* Allocate the private structure */
13578 static int
13579 ip_priv_alloc(void **bufp)
13580 {
13581         void    *buf;
13582 
13583         if ((buf = kmem_alloc(sizeof (ip_priv_t), KM_NOSLEEP)) == NULL)
13584                 return (ENOMEM);
13585 
13586         *bufp = buf;
13587         return (0);
13588 }
13589 
13590 /* Function to delete the private structure */
13591 void
13592 ip_priv_free(void *buf)
13593 {
13594         ASSERT(buf != NULL);
13595         kmem_free(buf, sizeof (ip_priv_t));
13596 }
13597 
13598 /*
13599  * The entry point for IPPF processing.
13600  * If the classifier (IPGPC_CLASSIFY) is not loaded and configured, the
13601  * routine just returns.
13602  *
13603  * When called, ip_process generates an ipp_packet_t structure
13604  * which holds the state information for this packet and invokes the
13605  * the classifier (via ipp_packet_process). The classification, depending on
13606  * configured filters, results in a list of actions for this packet. Invoking
13607  * an action may cause the packet to be dropped, in which case we return NULL.
13608  * proc indicates the callout position for
13609  * this packet and ill is the interface this packet arrived on or will leave
13610  * on (inbound and outbound resp.).
13611  *
13612  * We do the processing on the rill (mapped to the upper if ipmp), but MIB
13613  * on the ill corrsponding to the destination IP address.
13614  */
13615 mblk_t *
13616 ip_process(ip_proc_t proc, mblk_t *mp, ill_t *rill, ill_t *ill)
13617 {
13618         ip_priv_t       *priv;
13619         ipp_action_id_t aid;
13620         int             rc = 0;
13621         ipp_packet_t    *pp;
13622 
13623         /* If the classifier is not loaded, return  */
13624         if ((aid = ipp_action_lookup(IPGPC_CLASSIFY)) == IPP_ACTION_INVAL) {
13625                 return (mp);
13626         }
13627 
13628         ASSERT(mp != NULL);
13629 
13630         /* Allocate the packet structure */
13631         rc = ipp_packet_alloc(&pp, "ip", aid);
13632         if (rc != 0)
13633                 goto drop;
13634 
13635         /* Allocate the private structure */
13636         rc = ip_priv_alloc((void **)&priv);
13637         if (rc != 0) {
13638                 ipp_packet_free(pp);
13639                 goto drop;
13640         }
13641         priv->proc = proc;
13642         priv->ill_index = ill_get_upper_ifindex(rill);
13643 
13644         ipp_packet_set_private(pp, priv, ip_priv_free);
13645         ipp_packet_set_data(pp, mp);
13646 
13647         /* Invoke the classifier */
13648         rc = ipp_packet_process(&pp);
13649         if (pp != NULL) {
13650                 mp = ipp_packet_get_data(pp);
13651                 ipp_packet_free(pp);
13652                 if (rc != 0)
13653                         goto drop;
13654                 return (mp);
13655         } else {
13656                 /* No mp to trace in ip_drop_input/ip_drop_output  */
13657                 mp = NULL;
13658         }
13659 drop:
13660         if (proc == IPP_LOCAL_IN || proc == IPP_FWD_IN) {
13661                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
13662                 ip_drop_input("ip_process", mp, ill);
13663         } else {
13664                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
13665                 ip_drop_output("ip_process", mp, ill);
13666         }
13667         freemsg(mp);
13668         return (NULL);
13669 }
13670 
13671 /*
13672  * Propagate a multicast group membership operation (add/drop) on
13673  * all the interfaces crossed by the related multirt routes.
13674  * The call is considered successful if the operation succeeds
13675  * on at least one interface.
13676  *
13677  * This assumes that a set of IRE_HOST/RTF_MULTIRT has been created for the
13678  * multicast addresses with the ire argument being the first one.
13679  * We walk the bucket to find all the of those.
13680  *
13681  * Common to IPv4 and IPv6.
13682  */
13683 static int
13684 ip_multirt_apply_membership(int (*fn)(conn_t *, boolean_t,
13685     const in6_addr_t *, ipaddr_t, uint_t, mcast_record_t, const in6_addr_t *),
13686     ire_t *ire, conn_t *connp, boolean_t checkonly, const in6_addr_t *v6group,
13687     mcast_record_t fmode, const in6_addr_t *v6src)
13688 {
13689         ire_t           *ire_gw;
13690         irb_t           *irb;
13691         int             ifindex;
13692         int             error = 0;
13693         int             result;
13694         ip_stack_t      *ipst = ire->ire_ipst;
13695         ipaddr_t        group;
13696         boolean_t       isv6;
13697         int             match_flags;
13698 
13699         if (IN6_IS_ADDR_V4MAPPED(v6group)) {
13700                 IN6_V4MAPPED_TO_IPADDR(v6group, group);
13701                 isv6 = B_FALSE;
13702         } else {
13703                 isv6 = B_TRUE;
13704         }
13705 
13706         irb = ire->ire_bucket;
13707         ASSERT(irb != NULL);
13708 
13709         result = 0;
13710         irb_refhold(irb);
13711         for (; ire != NULL; ire = ire->ire_next) {
13712                 if ((ire->ire_flags & RTF_MULTIRT) == 0)
13713                         continue;
13714 
13715                 /* We handle -ifp routes by matching on the ill if set */
13716                 match_flags = MATCH_IRE_TYPE;
13717                 if (ire->ire_ill != NULL)
13718                         match_flags |= MATCH_IRE_ILL;
13719 
13720                 if (isv6) {
13721                         if (!IN6_ARE_ADDR_EQUAL(&ire->ire_addr_v6, v6group))
13722                                 continue;
13723 
13724                         ire_gw = ire_ftable_lookup_v6(&ire->ire_gateway_addr_v6,
13725                             0, 0, IRE_INTERFACE, ire->ire_ill, ALL_ZONES, NULL,
13726                             match_flags, 0, ipst, NULL);
13727                 } else {
13728                         if (ire->ire_addr != group)
13729                                 continue;
13730 
13731                         ire_gw = ire_ftable_lookup_v4(ire->ire_gateway_addr,
13732                             0, 0, IRE_INTERFACE, ire->ire_ill, ALL_ZONES, NULL,
13733                             match_flags, 0, ipst, NULL);
13734                 }
13735                 /* No interface route exists for the gateway; skip this ire. */
13736                 if (ire_gw == NULL)
13737                         continue;
13738                 if (ire_gw->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) {
13739                         ire_refrele(ire_gw);
13740                         continue;
13741                 }
13742                 ASSERT(ire_gw->ire_ill != NULL);     /* IRE_INTERFACE */
13743                 ifindex = ire_gw->ire_ill->ill_phyint->phyint_ifindex;
13744 
13745                 /*
13746                  * The operation is considered a success if
13747                  * it succeeds at least once on any one interface.
13748                  */
13749                 error = fn(connp, checkonly, v6group, INADDR_ANY, ifindex,
13750                     fmode, v6src);
13751                 if (error == 0)
13752                         result = CGTP_MCAST_SUCCESS;
13753 
13754                 ire_refrele(ire_gw);
13755         }
13756         irb_refrele(irb);
13757         /*
13758          * Consider the call as successful if we succeeded on at least
13759          * one interface. Otherwise, return the last encountered error.
13760          */
13761         return (result == CGTP_MCAST_SUCCESS ? 0 : error);
13762 }
13763 
13764 /*
13765  * Return the expected CGTP hooks version number.
13766  */
13767 int
13768 ip_cgtp_filter_supported(void)
13769 {
13770         return (ip_cgtp_filter_rev);
13771 }
13772 
13773 /*
13774  * CGTP hooks can be registered by invoking this function.
13775  * Checks that the version number matches.
13776  */
13777 int
13778 ip_cgtp_filter_register(netstackid_t stackid, cgtp_filter_ops_t *ops)
13779 {
13780         netstack_t *ns;
13781         ip_stack_t *ipst;
13782 
13783         if (ops->cfo_filter_rev != CGTP_FILTER_REV)
13784                 return (ENOTSUP);
13785 
13786         ns = netstack_find_by_stackid(stackid);
13787         if (ns == NULL)
13788                 return (EINVAL);
13789         ipst = ns->netstack_ip;
13790         ASSERT(ipst != NULL);
13791 
13792         if (ipst->ips_ip_cgtp_filter_ops != NULL) {
13793                 netstack_rele(ns);
13794                 return (EALREADY);
13795         }
13796 
13797         ipst->ips_ip_cgtp_filter_ops = ops;
13798 
13799         ill_set_inputfn_all(ipst);
13800 
13801         netstack_rele(ns);
13802         return (0);
13803 }
13804 
13805 /*
13806  * CGTP hooks can be unregistered by invoking this function.
13807  * Returns ENXIO if there was no registration.
13808  * Returns EBUSY if the ndd variable has not been turned off.
13809  */
13810 int
13811 ip_cgtp_filter_unregister(netstackid_t stackid)
13812 {
13813         netstack_t *ns;
13814         ip_stack_t *ipst;
13815 
13816         ns = netstack_find_by_stackid(stackid);
13817         if (ns == NULL)
13818                 return (EINVAL);
13819         ipst = ns->netstack_ip;
13820         ASSERT(ipst != NULL);
13821 
13822         if (ipst->ips_ip_cgtp_filter) {
13823                 netstack_rele(ns);
13824                 return (EBUSY);
13825         }
13826 
13827         if (ipst->ips_ip_cgtp_filter_ops == NULL) {
13828                 netstack_rele(ns);
13829                 return (ENXIO);
13830         }
13831         ipst->ips_ip_cgtp_filter_ops = NULL;
13832 
13833         ill_set_inputfn_all(ipst);
13834 
13835         netstack_rele(ns);
13836         return (0);
13837 }
13838 
13839 /*
13840  * Check whether there is a CGTP filter registration.
13841  * Returns non-zero if there is a registration, otherwise returns zero.
13842  * Note: returns zero if bad stackid.
13843  */
13844 int
13845 ip_cgtp_filter_is_registered(netstackid_t stackid)
13846 {
13847         netstack_t *ns;
13848         ip_stack_t *ipst;
13849         int ret;
13850 
13851         ns = netstack_find_by_stackid(stackid);
13852         if (ns == NULL)
13853                 return (0);
13854         ipst = ns->netstack_ip;
13855         ASSERT(ipst != NULL);
13856 
13857         if (ipst->ips_ip_cgtp_filter_ops != NULL)
13858                 ret = 1;
13859         else
13860                 ret = 0;
13861 
13862         netstack_rele(ns);
13863         return (ret);
13864 }
13865 
13866 static int
13867 ip_squeue_switch(int val)
13868 {
13869         int rval;
13870 
13871         switch (val) {
13872         case IP_SQUEUE_ENTER_NODRAIN:
13873                 rval = SQ_NODRAIN;
13874                 break;
13875         case IP_SQUEUE_ENTER:
13876                 rval = SQ_PROCESS;
13877                 break;
13878         case IP_SQUEUE_FILL:
13879         default:
13880                 rval = SQ_FILL;
13881                 break;
13882         }
13883         return (rval);
13884 }
13885 
13886 static void *
13887 ip_kstat2_init(netstackid_t stackid, ip_stat_t *ip_statisticsp)
13888 {
13889         kstat_t *ksp;
13890 
13891         ip_stat_t template = {
13892                 { "ip_udp_fannorm",             KSTAT_DATA_UINT64 },
13893                 { "ip_udp_fanmb",               KSTAT_DATA_UINT64 },
13894                 { "ip_recv_pullup",             KSTAT_DATA_UINT64 },
13895                 { "ip_db_ref",                  KSTAT_DATA_UINT64 },
13896                 { "ip_notaligned",              KSTAT_DATA_UINT64 },
13897                 { "ip_multimblk",               KSTAT_DATA_UINT64 },
13898                 { "ip_opt",                     KSTAT_DATA_UINT64 },
13899                 { "ipsec_proto_ahesp",          KSTAT_DATA_UINT64 },
13900                 { "ip_conn_flputbq",            KSTAT_DATA_UINT64 },
13901                 { "ip_conn_walk_drain",         KSTAT_DATA_UINT64 },
13902                 { "ip_out_sw_cksum",            KSTAT_DATA_UINT64 },
13903                 { "ip_out_sw_cksum_bytes",      KSTAT_DATA_UINT64 },
13904                 { "ip_in_sw_cksum",             KSTAT_DATA_UINT64 },
13905                 { "ip_ire_reclaim_calls",       KSTAT_DATA_UINT64 },
13906                 { "ip_ire_reclaim_deleted",     KSTAT_DATA_UINT64 },
13907                 { "ip_nce_reclaim_calls",       KSTAT_DATA_UINT64 },
13908                 { "ip_nce_reclaim_deleted",     KSTAT_DATA_UINT64 },
13909                 { "ip_dce_reclaim_calls",       KSTAT_DATA_UINT64 },
13910                 { "ip_dce_reclaim_deleted",     KSTAT_DATA_UINT64 },
13911                 { "ip_tcp_in_full_hw_cksum_err",        KSTAT_DATA_UINT64 },
13912                 { "ip_tcp_in_part_hw_cksum_err",        KSTAT_DATA_UINT64 },
13913                 { "ip_tcp_in_sw_cksum_err",             KSTAT_DATA_UINT64 },
13914                 { "ip_udp_in_full_hw_cksum_err",        KSTAT_DATA_UINT64 },
13915                 { "ip_udp_in_part_hw_cksum_err",        KSTAT_DATA_UINT64 },
13916                 { "ip_udp_in_sw_cksum_err",     KSTAT_DATA_UINT64 },
13917                 { "conn_in_recvdstaddr",        KSTAT_DATA_UINT64 },
13918                 { "conn_in_recvopts",           KSTAT_DATA_UINT64 },
13919                 { "conn_in_recvif",             KSTAT_DATA_UINT64 },
13920                 { "conn_in_recvslla",           KSTAT_DATA_UINT64 },
13921                 { "conn_in_recvucred",          KSTAT_DATA_UINT64 },
13922                 { "conn_in_recvttl",            KSTAT_DATA_UINT64 },
13923                 { "conn_in_recvhopopts",        KSTAT_DATA_UINT64 },
13924                 { "conn_in_recvhoplimit",       KSTAT_DATA_UINT64 },
13925                 { "conn_in_recvdstopts",        KSTAT_DATA_UINT64 },
13926                 { "conn_in_recvrthdrdstopts",   KSTAT_DATA_UINT64 },
13927                 { "conn_in_recvrthdr",          KSTAT_DATA_UINT64 },
13928                 { "conn_in_recvpktinfo",        KSTAT_DATA_UINT64 },
13929                 { "conn_in_recvtclass",         KSTAT_DATA_UINT64 },
13930                 { "conn_in_timestamp",          KSTAT_DATA_UINT64 },
13931         };
13932 
13933         ksp = kstat_create_netstack("ip", 0, "ipstat", "net",
13934             KSTAT_TYPE_NAMED, sizeof (template) / sizeof (kstat_named_t),
13935             KSTAT_FLAG_VIRTUAL, stackid);
13936 
13937         if (ksp == NULL)
13938                 return (NULL);
13939 
13940         bcopy(&template, ip_statisticsp, sizeof (template));
13941         ksp->ks_data = (void *)ip_statisticsp;
13942         ksp->ks_private = (void *)(uintptr_t)stackid;
13943 
13944         kstat_install(ksp);
13945         return (ksp);
13946 }
13947 
13948 static void
13949 ip_kstat2_fini(netstackid_t stackid, kstat_t *ksp)
13950 {
13951         if (ksp != NULL) {
13952                 ASSERT(stackid == (netstackid_t)(uintptr_t)ksp->ks_private);
13953                 kstat_delete_netstack(ksp, stackid);
13954         }
13955 }
13956 
13957 static void *
13958 ip_kstat_init(netstackid_t stackid, ip_stack_t *ipst)
13959 {
13960         kstat_t *ksp;
13961 
13962         ip_named_kstat_t template = {
13963                 { "forwarding",         KSTAT_DATA_UINT32, 0 },
13964                 { "defaultTTL",         KSTAT_DATA_UINT32, 0 },
13965                 { "inReceives",         KSTAT_DATA_UINT64, 0 },
13966                 { "inHdrErrors",        KSTAT_DATA_UINT32, 0 },
13967                 { "inAddrErrors",       KSTAT_DATA_UINT32, 0 },
13968                 { "forwDatagrams",      KSTAT_DATA_UINT64, 0 },
13969                 { "inUnknownProtos",    KSTAT_DATA_UINT32, 0 },
13970                 { "inDiscards",         KSTAT_DATA_UINT32, 0 },
13971                 { "inDelivers",         KSTAT_DATA_UINT64, 0 },
13972                 { "outRequests",        KSTAT_DATA_UINT64, 0 },
13973                 { "outDiscards",        KSTAT_DATA_UINT32, 0 },
13974                 { "outNoRoutes",        KSTAT_DATA_UINT32, 0 },
13975                 { "reasmTimeout",       KSTAT_DATA_UINT32, 0 },
13976                 { "reasmReqds",         KSTAT_DATA_UINT32, 0 },
13977                 { "reasmOKs",           KSTAT_DATA_UINT32, 0 },
13978                 { "reasmFails",         KSTAT_DATA_UINT32, 0 },
13979                 { "fragOKs",            KSTAT_DATA_UINT32, 0 },
13980                 { "fragFails",          KSTAT_DATA_UINT32, 0 },
13981                 { "fragCreates",        KSTAT_DATA_UINT32, 0 },
13982                 { "addrEntrySize",      KSTAT_DATA_INT32, 0 },
13983                 { "routeEntrySize",     KSTAT_DATA_INT32, 0 },
13984                 { "netToMediaEntrySize",        KSTAT_DATA_INT32, 0 },
13985                 { "routingDiscards",    KSTAT_DATA_UINT32, 0 },
13986                 { "inErrs",             KSTAT_DATA_UINT32, 0 },
13987                 { "noPorts",            KSTAT_DATA_UINT32, 0 },
13988                 { "inCksumErrs",        KSTAT_DATA_UINT32, 0 },
13989                 { "reasmDuplicates",    KSTAT_DATA_UINT32, 0 },
13990                 { "reasmPartDups",      KSTAT_DATA_UINT32, 0 },
13991                 { "forwProhibits",      KSTAT_DATA_UINT32, 0 },
13992                 { "udpInCksumErrs",     KSTAT_DATA_UINT32, 0 },
13993                 { "udpInOverflows",     KSTAT_DATA_UINT32, 0 },
13994                 { "rawipInOverflows",   KSTAT_DATA_UINT32, 0 },
13995                 { "ipsecInSucceeded",   KSTAT_DATA_UINT32, 0 },
13996                 { "ipsecInFailed",      KSTAT_DATA_INT32, 0 },
13997                 { "memberEntrySize",    KSTAT_DATA_INT32, 0 },
13998                 { "inIPv6",             KSTAT_DATA_UINT32, 0 },
13999                 { "outIPv6",            KSTAT_DATA_UINT32, 0 },
14000                 { "outSwitchIPv6",      KSTAT_DATA_UINT32, 0 },
14001         };
14002 
14003         ksp = kstat_create_netstack("ip", 0, "ip", "mib2", KSTAT_TYPE_NAMED,
14004             NUM_OF_FIELDS(ip_named_kstat_t), 0, stackid);
14005         if (ksp == NULL || ksp->ks_data == NULL)
14006                 return (NULL);
14007 
14008         template.forwarding.value.ui32 = WE_ARE_FORWARDING(ipst) ? 1:2;
14009         template.defaultTTL.value.ui32 = (uint32_t)ipst->ips_ip_def_ttl;
14010         template.reasmTimeout.value.ui32 = ipst->ips_ip_reassembly_timeout;
14011         template.addrEntrySize.value.i32 = sizeof (mib2_ipAddrEntry_t);
14012         template.routeEntrySize.value.i32 = sizeof (mib2_ipRouteEntry_t);
14013 
14014         template.netToMediaEntrySize.value.i32 =
14015             sizeof (mib2_ipNetToMediaEntry_t);
14016 
14017         template.memberEntrySize.value.i32 = sizeof (ipv6_member_t);
14018 
14019         bcopy(&template, ksp->ks_data, sizeof (template));
14020         ksp->ks_update = ip_kstat_update;
14021         ksp->ks_private = (void *)(uintptr_t)stackid;
14022 
14023         kstat_install(ksp);
14024         return (ksp);
14025 }
14026 
14027 static void
14028 ip_kstat_fini(netstackid_t stackid, kstat_t *ksp)
14029 {
14030         if (ksp != NULL) {
14031                 ASSERT(stackid == (netstackid_t)(uintptr_t)ksp->ks_private);
14032                 kstat_delete_netstack(ksp, stackid);
14033         }
14034 }
14035 
14036 static int
14037 ip_kstat_update(kstat_t *kp, int rw)
14038 {
14039         ip_named_kstat_t *ipkp;
14040         mib2_ipIfStatsEntry_t ipmib;
14041         ill_walk_context_t ctx;
14042         ill_t *ill;
14043         netstackid_t    stackid = (zoneid_t)(uintptr_t)kp->ks_private;
14044         netstack_t      *ns;
14045         ip_stack_t      *ipst;
14046 
14047         if (kp == NULL || kp->ks_data == NULL)
14048                 return (EIO);
14049 
14050         if (rw == KSTAT_WRITE)
14051                 return (EACCES);
14052 
14053         ns = netstack_find_by_stackid(stackid);
14054         if (ns == NULL)
14055                 return (-1);
14056         ipst = ns->netstack_ip;
14057         if (ipst == NULL) {
14058                 netstack_rele(ns);
14059                 return (-1);
14060         }
14061         ipkp = (ip_named_kstat_t *)kp->ks_data;
14062 
14063         bcopy(&ipst->ips_ip_mib, &ipmib, sizeof (ipmib));
14064         rw_enter(&ipst->ips_ill_g_lock, RW_READER);
14065         ill = ILL_START_WALK_V4(&ctx, ipst);
14066         for (; ill != NULL; ill = ill_next(&ctx, ill))
14067                 ip_mib2_add_ip_stats(&ipmib, ill->ill_ip_mib);
14068         rw_exit(&ipst->ips_ill_g_lock);
14069 
14070         ipkp->forwarding.value.ui32 =                ipmib.ipIfStatsForwarding;
14071         ipkp->defaultTTL.value.ui32 =                ipmib.ipIfStatsDefaultTTL;
14072         ipkp->inReceives.value.ui64 =                ipmib.ipIfStatsHCInReceives;
14073         ipkp->inHdrErrors.value.ui32 =               ipmib.ipIfStatsInHdrErrors;
14074         ipkp->inAddrErrors.value.ui32 =              ipmib.ipIfStatsInAddrErrors;
14075         ipkp->forwDatagrams.value.ui64 = ipmib.ipIfStatsHCOutForwDatagrams;
14076         ipkp->inUnknownProtos.value.ui32 =   ipmib.ipIfStatsInUnknownProtos;
14077         ipkp->inDiscards.value.ui32 =                ipmib.ipIfStatsInDiscards;
14078         ipkp->inDelivers.value.ui64 =                ipmib.ipIfStatsHCInDelivers;
14079         ipkp->outRequests.value.ui64 =               ipmib.ipIfStatsHCOutRequests;
14080         ipkp->outDiscards.value.ui32 =               ipmib.ipIfStatsOutDiscards;
14081         ipkp->outNoRoutes.value.ui32 =               ipmib.ipIfStatsOutNoRoutes;
14082         ipkp->reasmTimeout.value.ui32 =              ipst->ips_ip_reassembly_timeout;
14083         ipkp->reasmReqds.value.ui32 =                ipmib.ipIfStatsReasmReqds;
14084         ipkp->reasmOKs.value.ui32 =          ipmib.ipIfStatsReasmOKs;
14085         ipkp->reasmFails.value.ui32 =                ipmib.ipIfStatsReasmFails;
14086         ipkp->fragOKs.value.ui32 =           ipmib.ipIfStatsOutFragOKs;
14087         ipkp->fragFails.value.ui32 =         ipmib.ipIfStatsOutFragFails;
14088         ipkp->fragCreates.value.ui32 =               ipmib.ipIfStatsOutFragCreates;
14089 
14090         ipkp->routingDiscards.value.ui32 =   0;
14091         ipkp->inErrs.value.ui32 =            ipmib.tcpIfStatsInErrs;
14092         ipkp->noPorts.value.ui32 =           ipmib.udpIfStatsNoPorts;
14093         ipkp->inCksumErrs.value.ui32 =               ipmib.ipIfStatsInCksumErrs;
14094         ipkp->reasmDuplicates.value.ui32 =   ipmib.ipIfStatsReasmDuplicates;
14095         ipkp->reasmPartDups.value.ui32 =     ipmib.ipIfStatsReasmPartDups;
14096         ipkp->forwProhibits.value.ui32 =     ipmib.ipIfStatsForwProhibits;
14097         ipkp->udpInCksumErrs.value.ui32 =    ipmib.udpIfStatsInCksumErrs;
14098         ipkp->udpInOverflows.value.ui32 =    ipmib.udpIfStatsInOverflows;
14099         ipkp->rawipInOverflows.value.ui32 =  ipmib.rawipIfStatsInOverflows;
14100         ipkp->ipsecInSucceeded.value.ui32 =  ipmib.ipsecIfStatsInSucceeded;
14101         ipkp->ipsecInFailed.value.i32 =              ipmib.ipsecIfStatsInFailed;
14102 
14103         ipkp->inIPv6.value.ui32 =    ipmib.ipIfStatsInWrongIPVersion;
14104         ipkp->outIPv6.value.ui32 =   ipmib.ipIfStatsOutWrongIPVersion;
14105         ipkp->outSwitchIPv6.value.ui32 = ipmib.ipIfStatsOutSwitchIPVersion;
14106 
14107         netstack_rele(ns);
14108 
14109         return (0);
14110 }
14111 
14112 static void *
14113 icmp_kstat_init(netstackid_t stackid)
14114 {
14115         kstat_t *ksp;
14116 
14117         icmp_named_kstat_t template = {
14118                 { "inMsgs",             KSTAT_DATA_UINT32 },
14119                 { "inErrors",           KSTAT_DATA_UINT32 },
14120                 { "inDestUnreachs",     KSTAT_DATA_UINT32 },
14121                 { "inTimeExcds",        KSTAT_DATA_UINT32 },
14122                 { "inParmProbs",        KSTAT_DATA_UINT32 },
14123                 { "inSrcQuenchs",       KSTAT_DATA_UINT32 },
14124                 { "inRedirects",        KSTAT_DATA_UINT32 },
14125                 { "inEchos",            KSTAT_DATA_UINT32 },
14126                 { "inEchoReps",         KSTAT_DATA_UINT32 },
14127                 { "inTimestamps",       KSTAT_DATA_UINT32 },
14128                 { "inTimestampReps",    KSTAT_DATA_UINT32 },
14129                 { "inAddrMasks",        KSTAT_DATA_UINT32 },
14130                 { "inAddrMaskReps",     KSTAT_DATA_UINT32 },
14131                 { "outMsgs",            KSTAT_DATA_UINT32 },
14132                 { "outErrors",          KSTAT_DATA_UINT32 },
14133                 { "outDestUnreachs",    KSTAT_DATA_UINT32 },
14134                 { "outTimeExcds",       KSTAT_DATA_UINT32 },
14135                 { "outParmProbs",       KSTAT_DATA_UINT32 },
14136                 { "outSrcQuenchs",      KSTAT_DATA_UINT32 },
14137                 { "outRedirects",       KSTAT_DATA_UINT32 },
14138                 { "outEchos",           KSTAT_DATA_UINT32 },
14139                 { "outEchoReps",        KSTAT_DATA_UINT32 },
14140                 { "outTimestamps",      KSTAT_DATA_UINT32 },
14141                 { "outTimestampReps",   KSTAT_DATA_UINT32 },
14142                 { "outAddrMasks",       KSTAT_DATA_UINT32 },
14143                 { "outAddrMaskReps",    KSTAT_DATA_UINT32 },
14144                 { "inChksumErrs",       KSTAT_DATA_UINT32 },
14145                 { "inUnknowns",         KSTAT_DATA_UINT32 },
14146                 { "inFragNeeded",       KSTAT_DATA_UINT32 },
14147                 { "outFragNeeded",      KSTAT_DATA_UINT32 },
14148                 { "outDrops",           KSTAT_DATA_UINT32 },
14149                 { "inOverFlows",        KSTAT_DATA_UINT32 },
14150                 { "inBadRedirects",     KSTAT_DATA_UINT32 },
14151         };
14152 
14153         ksp = kstat_create_netstack("ip", 0, "icmp", "mib2", KSTAT_TYPE_NAMED,
14154             NUM_OF_FIELDS(icmp_named_kstat_t), 0, stackid);
14155         if (ksp == NULL || ksp->ks_data == NULL)
14156                 return (NULL);
14157 
14158         bcopy(&template, ksp->ks_data, sizeof (template));
14159 
14160         ksp->ks_update = icmp_kstat_update;
14161         ksp->ks_private = (void *)(uintptr_t)stackid;
14162 
14163         kstat_install(ksp);
14164         return (ksp);
14165 }
14166 
14167 static void
14168 icmp_kstat_fini(netstackid_t stackid, kstat_t *ksp)
14169 {
14170         if (ksp != NULL) {
14171                 ASSERT(stackid == (netstackid_t)(uintptr_t)ksp->ks_private);
14172                 kstat_delete_netstack(ksp, stackid);
14173         }
14174 }
14175 
14176 static int
14177 icmp_kstat_update(kstat_t *kp, int rw)
14178 {
14179         icmp_named_kstat_t *icmpkp;
14180         netstackid_t    stackid = (zoneid_t)(uintptr_t)kp->ks_private;
14181         netstack_t      *ns;
14182         ip_stack_t      *ipst;
14183 
14184         if ((kp == NULL) || (kp->ks_data == NULL))
14185                 return (EIO);
14186 
14187         if (rw == KSTAT_WRITE)
14188                 return (EACCES);
14189 
14190         ns = netstack_find_by_stackid(stackid);
14191         if (ns == NULL)
14192                 return (-1);
14193         ipst = ns->netstack_ip;
14194         if (ipst == NULL) {
14195                 netstack_rele(ns);
14196                 return (-1);
14197         }
14198         icmpkp = (icmp_named_kstat_t *)kp->ks_data;
14199 
14200         icmpkp->inMsgs.value.ui32 =      ipst->ips_icmp_mib.icmpInMsgs;
14201         icmpkp->inErrors.value.ui32 =            ipst->ips_icmp_mib.icmpInErrors;
14202         icmpkp->inDestUnreachs.value.ui32 =
14203             ipst->ips_icmp_mib.icmpInDestUnreachs;
14204         icmpkp->inTimeExcds.value.ui32 =    ipst->ips_icmp_mib.icmpInTimeExcds;
14205         icmpkp->inParmProbs.value.ui32 =    ipst->ips_icmp_mib.icmpInParmProbs;
14206         icmpkp->inSrcQuenchs.value.ui32 =   ipst->ips_icmp_mib.icmpInSrcQuenchs;
14207         icmpkp->inRedirects.value.ui32 =    ipst->ips_icmp_mib.icmpInRedirects;
14208         icmpkp->inEchos.value.ui32 =     ipst->ips_icmp_mib.icmpInEchos;
14209         icmpkp->inEchoReps.value.ui32 =          ipst->ips_icmp_mib.icmpInEchoReps;
14210         icmpkp->inTimestamps.value.ui32 =   ipst->ips_icmp_mib.icmpInTimestamps;
14211         icmpkp->inTimestampReps.value.ui32 =
14212             ipst->ips_icmp_mib.icmpInTimestampReps;
14213         icmpkp->inAddrMasks.value.ui32 =    ipst->ips_icmp_mib.icmpInAddrMasks;
14214         icmpkp->inAddrMaskReps.value.ui32 =
14215             ipst->ips_icmp_mib.icmpInAddrMaskReps;
14216         icmpkp->outMsgs.value.ui32 =     ipst->ips_icmp_mib.icmpOutMsgs;
14217         icmpkp->outErrors.value.ui32 =           ipst->ips_icmp_mib.icmpOutErrors;
14218         icmpkp->outDestUnreachs.value.ui32 =
14219             ipst->ips_icmp_mib.icmpOutDestUnreachs;
14220         icmpkp->outTimeExcds.value.ui32 =   ipst->ips_icmp_mib.icmpOutTimeExcds;
14221         icmpkp->outParmProbs.value.ui32 =   ipst->ips_icmp_mib.icmpOutParmProbs;
14222         icmpkp->outSrcQuenchs.value.ui32 =
14223             ipst->ips_icmp_mib.icmpOutSrcQuenchs;
14224         icmpkp->outRedirects.value.ui32 =   ipst->ips_icmp_mib.icmpOutRedirects;
14225         icmpkp->outEchos.value.ui32 =            ipst->ips_icmp_mib.icmpOutEchos;
14226         icmpkp->outEchoReps.value.ui32 =    ipst->ips_icmp_mib.icmpOutEchoReps;
14227         icmpkp->outTimestamps.value.ui32 =
14228             ipst->ips_icmp_mib.icmpOutTimestamps;
14229         icmpkp->outTimestampReps.value.ui32 =
14230             ipst->ips_icmp_mib.icmpOutTimestampReps;
14231         icmpkp->outAddrMasks.value.ui32 =
14232             ipst->ips_icmp_mib.icmpOutAddrMasks;
14233         icmpkp->outAddrMaskReps.value.ui32 =
14234             ipst->ips_icmp_mib.icmpOutAddrMaskReps;
14235         icmpkp->inCksumErrs.value.ui32 =    ipst->ips_icmp_mib.icmpInCksumErrs;
14236         icmpkp->inUnknowns.value.ui32 =          ipst->ips_icmp_mib.icmpInUnknowns;
14237         icmpkp->inFragNeeded.value.ui32 =   ipst->ips_icmp_mib.icmpInFragNeeded;
14238         icmpkp->outFragNeeded.value.ui32 =
14239             ipst->ips_icmp_mib.icmpOutFragNeeded;
14240         icmpkp->outDrops.value.ui32 =            ipst->ips_icmp_mib.icmpOutDrops;
14241         icmpkp->inOverflows.value.ui32 =    ipst->ips_icmp_mib.icmpInOverflows;
14242         icmpkp->inBadRedirects.value.ui32 =
14243             ipst->ips_icmp_mib.icmpInBadRedirects;
14244 
14245         netstack_rele(ns);
14246         return (0);
14247 }
14248 
14249 /*
14250  * This is the fanout function for raw socket opened for SCTP.  Note
14251  * that it is called after SCTP checks that there is no socket which
14252  * wants a packet.  Then before SCTP handles this out of the blue packet,
14253  * this function is called to see if there is any raw socket for SCTP.
14254  * If there is and it is bound to the correct address, the packet will
14255  * be sent to that socket.  Note that only one raw socket can be bound to
14256  * a port.  This is assured in ipcl_sctp_hash_insert();
14257  */
14258 void
14259 ip_fanout_sctp_raw(mblk_t *mp, ipha_t *ipha, ip6_t *ip6h, uint32_t ports,
14260     ip_recv_attr_t *ira)
14261 {
14262         conn_t          *connp;
14263         queue_t         *rq;
14264         boolean_t       secure;
14265         ill_t           *ill = ira->ira_ill;
14266         ip_stack_t      *ipst = ill->ill_ipst;
14267         ipsec_stack_t   *ipss = ipst->ips_netstack->netstack_ipsec;
14268         sctp_stack_t    *sctps = ipst->ips_netstack->netstack_sctp;
14269         iaflags_t       iraflags = ira->ira_flags;
14270         ill_t           *rill = ira->ira_rill;
14271 
14272         secure = iraflags & IRAF_IPSEC_SECURE;
14273 
14274         connp = ipcl_classify_raw(mp, IPPROTO_SCTP, ports, ipha, ip6h,
14275             ira, ipst);
14276         if (connp == NULL) {
14277                 /*
14278                  * Although raw sctp is not summed, OOB chunks must be.
14279                  * Drop the packet here if the sctp checksum failed.
14280                  */
14281                 if (iraflags & IRAF_SCTP_CSUM_ERR) {
14282                         SCTPS_BUMP_MIB(sctps, sctpChecksumError);
14283                         freemsg(mp);
14284                         return;
14285                 }
14286                 ira->ira_ill = ira->ira_rill = NULL;
14287                 sctp_ootb_input(mp, ira, ipst);
14288                 ira->ira_ill = ill;
14289                 ira->ira_rill = rill;
14290                 return;
14291         }
14292         rq = connp->conn_rq;
14293         if (IPCL_IS_NONSTR(connp) ? connp->conn_flow_cntrld : !canputnext(rq)) {
14294                 CONN_DEC_REF(connp);
14295                 BUMP_MIB(ill->ill_ip_mib, rawipIfStatsInOverflows);
14296                 freemsg(mp);
14297                 return;
14298         }
14299         if (((iraflags & IRAF_IS_IPV4) ?
14300             CONN_INBOUND_POLICY_PRESENT(connp, ipss) :
14301             CONN_INBOUND_POLICY_PRESENT_V6(connp, ipss)) ||
14302             secure) {
14303                 mp = ipsec_check_inbound_policy(mp, connp, ipha,
14304                     ip6h, ira);
14305                 if (mp == NULL) {
14306                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsInDiscards);
14307                         /* Note that mp is NULL */
14308                         ip_drop_input("ipIfStatsInDiscards", mp, ill);
14309                         CONN_DEC_REF(connp);
14310                         return;
14311                 }
14312         }
14313 
14314         if (iraflags & IRAF_ICMP_ERROR) {
14315                 (connp->conn_recvicmp)(connp, mp, NULL, ira);
14316         } else {
14317                 ill_t *rill = ira->ira_rill;
14318 
14319                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCInDelivers);
14320                 /* This is the SOCK_RAW, IPPROTO_SCTP case. */
14321                 ira->ira_ill = ira->ira_rill = NULL;
14322                 (connp->conn_recv)(connp, mp, NULL, ira);
14323                 ira->ira_ill = ill;
14324                 ira->ira_rill = rill;
14325         }
14326         CONN_DEC_REF(connp);
14327 }
14328 
14329 /*
14330  * Free a packet that has the link-layer dl_unitdata_req_t or fast-path
14331  * header before the ip payload.
14332  */
14333 static void
14334 ip_xmit_flowctl_drop(ill_t *ill, mblk_t *mp, boolean_t is_fp_mp, int fp_mp_len)
14335 {
14336         int len = (mp->b_wptr - mp->b_rptr);
14337         mblk_t *ip_mp;
14338 
14339         BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
14340         if (is_fp_mp || len != fp_mp_len) {
14341                 if (len > fp_mp_len) {
14342                         /*
14343                          * fastpath header and ip header in the first mblk
14344                          */
14345                         mp->b_rptr += fp_mp_len;
14346                 } else {
14347                         /*
14348                          * ip_xmit_attach_llhdr had to prepend an mblk to
14349                          * attach the fastpath header before ip header.
14350                          */
14351                         ip_mp = mp->b_cont;
14352                         freeb(mp);
14353                         mp = ip_mp;
14354                         mp->b_rptr += (fp_mp_len - len);
14355                 }
14356         } else {
14357                 ip_mp = mp->b_cont;
14358                 freeb(mp);
14359                 mp = ip_mp;
14360         }
14361         ip_drop_output("ipIfStatsOutDiscards - flow ctl", mp, ill);
14362         freemsg(mp);
14363 }
14364 
14365 /*
14366  * Normal post fragmentation function.
14367  *
14368  * Send a packet using the passed in nce. This handles both IPv4 and IPv6
14369  * using the same state machine.
14370  *
14371  * We return an error on failure. In particular we return EWOULDBLOCK
14372  * when the driver flow controls. In that case this ensures that ip_wsrv runs
14373  * (currently by canputnext failure resulting in backenabling from GLD.)
14374  * This allows the callers of conn_ip_output() to use EWOULDBLOCK as an
14375  * indication that they can flow control until ip_wsrv() tells then to restart.
14376  *
14377  * If the nce passed by caller is incomplete, this function
14378  * queues the packet and if necessary, sends ARP request and bails.
14379  * If the Neighbor Cache passed is fully resolved, we simply prepend
14380  * the link-layer header to the packet, do ipsec hw acceleration
14381  * work if necessary, and send the packet out on the wire.
14382  */
14383 /* ARGSUSED6 */
14384 int
14385 ip_xmit(mblk_t *mp, nce_t *nce, iaflags_t ixaflags, uint_t pkt_len,
14386     uint32_t xmit_hint, zoneid_t szone, zoneid_t nolzid, uintptr_t *ixacookie)
14387 {
14388         queue_t         *wq;
14389         ill_t           *ill = nce->nce_ill;
14390         ip_stack_t      *ipst = ill->ill_ipst;
14391         uint64_t        delta;
14392         boolean_t       isv6 = ill->ill_isv6;
14393         boolean_t       fp_mp;
14394         ncec_t          *ncec = nce->nce_common;
14395         int64_t         now = LBOLT_FASTPATH64;
14396         boolean_t       is_probe;
14397 
14398         DTRACE_PROBE1(ip__xmit, nce_t *, nce);
14399 
14400         ASSERT(mp != NULL);
14401         ASSERT(mp->b_datap->db_type == M_DATA);
14402         ASSERT(pkt_len == msgdsize(mp));
14403 
14404         /*
14405          * If we have already been here and are coming back after ARP/ND.
14406          * the IXAF_NO_TRACE flag is set. We skip FW_HOOKS, DTRACE and ipobs
14407          * in that case since they have seen the packet when it came here
14408          * the first time.
14409          */
14410         if (ixaflags & IXAF_NO_TRACE)
14411                 goto sendit;
14412 
14413         if (ixaflags & IXAF_IS_IPV4) {
14414                 ipha_t *ipha = (ipha_t *)mp->b_rptr;
14415 
14416                 ASSERT(!isv6);
14417                 ASSERT(pkt_len == ntohs(((ipha_t *)mp->b_rptr)->ipha_length));
14418                 if (HOOKS4_INTERESTED_PHYSICAL_OUT(ipst) &&
14419                     !(ixaflags & IXAF_NO_PFHOOK)) {
14420                         int     error;
14421 
14422                         FW_HOOKS(ipst->ips_ip4_physical_out_event,
14423                             ipst->ips_ipv4firewall_physical_out,
14424                             NULL, ill, ipha, mp, mp, 0, ipst, error);
14425                         DTRACE_PROBE1(ip4__physical__out__end,
14426                             mblk_t *, mp);
14427                         if (mp == NULL)
14428                                 return (error);
14429 
14430                         /* The length could have changed */
14431                         pkt_len = msgdsize(mp);
14432                 }
14433                 if (ipst->ips_ip4_observe.he_interested) {
14434                         /*
14435                          * Note that for TX the zoneid is the sending
14436                          * zone, whether or not MLP is in play.
14437                          * Since the szone argument is the IP zoneid (i.e.,
14438                          * zero for exclusive-IP zones) and ipobs wants
14439                          * the system zoneid, we map it here.
14440                          */
14441                         szone = IP_REAL_ZONEID(szone, ipst);
14442 
14443                         /*
14444                          * On the outbound path the destination zone will be
14445                          * unknown as we're sending this packet out on the
14446                          * wire.
14447                          */
14448                         ipobs_hook(mp, IPOBS_HOOK_OUTBOUND, szone, ALL_ZONES,
14449                             ill, ipst);
14450                 }
14451                 DTRACE_IP7(send, mblk_t *, mp,  conn_t *, NULL,
14452                     void_ip_t *, ipha,  __dtrace_ipsr_ill_t *, ill,
14453                     ipha_t *, ipha, ip6_t *, NULL, int, 0);
14454         } else {
14455                 ip6_t *ip6h = (ip6_t *)mp->b_rptr;
14456 
14457                 ASSERT(isv6);
14458                 ASSERT(pkt_len ==
14459                     ntohs(((ip6_t *)mp->b_rptr)->ip6_plen) + IPV6_HDR_LEN);
14460                 if (HOOKS6_INTERESTED_PHYSICAL_OUT(ipst) &&
14461                     !(ixaflags & IXAF_NO_PFHOOK)) {
14462                         int     error;
14463 
14464                         FW_HOOKS6(ipst->ips_ip6_physical_out_event,
14465                             ipst->ips_ipv6firewall_physical_out,
14466                             NULL, ill, ip6h, mp, mp, 0, ipst, error);
14467                         DTRACE_PROBE1(ip6__physical__out__end,
14468                             mblk_t *, mp);
14469                         if (mp == NULL)
14470                                 return (error);
14471 
14472                         /* The length could have changed */
14473                         pkt_len = msgdsize(mp);
14474                 }
14475                 if (ipst->ips_ip6_observe.he_interested) {
14476                         /* See above */
14477                         szone = IP_REAL_ZONEID(szone, ipst);
14478 
14479                         ipobs_hook(mp, IPOBS_HOOK_OUTBOUND, szone, ALL_ZONES,
14480                             ill, ipst);
14481                 }
14482                 DTRACE_IP7(send, mblk_t *, mp,  conn_t *, NULL,
14483                     void_ip_t *, ip6h,  __dtrace_ipsr_ill_t *, ill,
14484                     ipha_t *, NULL, ip6_t *, ip6h, int, 0);
14485         }
14486 
14487 sendit:
14488         /*
14489          * We check the state without a lock because the state can never
14490          * move "backwards" to initial or incomplete.
14491          */
14492         switch (ncec->ncec_state) {
14493         case ND_REACHABLE:
14494         case ND_STALE:
14495         case ND_DELAY:
14496         case ND_PROBE:
14497                 mp = ip_xmit_attach_llhdr(mp, nce);
14498                 if (mp == NULL) {
14499                         /*
14500                          * ip_xmit_attach_llhdr has increased
14501                          * ipIfStatsOutDiscards and called ip_drop_output()
14502                          */
14503                         return (ENOBUFS);
14504                 }
14505                 /*
14506                  * check if nce_fastpath completed and we tagged on a
14507                  * copy of nce_fp_mp in ip_xmit_attach_llhdr().
14508                  */
14509                 fp_mp = (mp->b_datap->db_type == M_DATA);
14510 
14511                 if (fp_mp &&
14512                     (ill->ill_capabilities & ILL_CAPAB_DLD_DIRECT)) {
14513                         ill_dld_direct_t *idd;
14514 
14515                         idd = &ill->ill_dld_capab->idc_direct;
14516                         /*
14517                          * Send the packet directly to DLD, where it
14518                          * may be queued depending on the availability
14519                          * of transmit resources at the media layer.
14520                          * Return value should be taken into
14521                          * account and flow control the TCP.
14522                          */
14523                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCOutTransmits);
14524                         UPDATE_MIB(ill->ill_ip_mib, ipIfStatsHCOutOctets,
14525                             pkt_len);
14526 
14527                         if (ixaflags & IXAF_NO_DEV_FLOW_CTL) {
14528                                 (void) idd->idd_tx_df(idd->idd_tx_dh, mp,
14529                                     (uintptr_t)xmit_hint, IP_DROP_ON_NO_DESC);
14530                         } else {
14531                                 uintptr_t cookie;
14532 
14533                                 if ((cookie = idd->idd_tx_df(idd->idd_tx_dh,
14534                                     mp, (uintptr_t)xmit_hint, 0)) != 0) {
14535                                         if (ixacookie != NULL)
14536                                                 *ixacookie = cookie;
14537                                         return (EWOULDBLOCK);
14538                                 }
14539                         }
14540                 } else {
14541                         wq = ill->ill_wq;
14542 
14543                         if (!(ixaflags & IXAF_NO_DEV_FLOW_CTL) &&
14544                             !canputnext(wq)) {
14545                                 if (ixacookie != NULL)
14546                                         *ixacookie = 0;
14547                                 ip_xmit_flowctl_drop(ill, mp, fp_mp,
14548                                     nce->nce_fp_mp != NULL ?
14549                                     MBLKL(nce->nce_fp_mp) : 0);
14550                                 return (EWOULDBLOCK);
14551                         }
14552                         BUMP_MIB(ill->ill_ip_mib, ipIfStatsHCOutTransmits);
14553                         UPDATE_MIB(ill->ill_ip_mib, ipIfStatsHCOutOctets,
14554                             pkt_len);
14555                         putnext(wq, mp);
14556                 }
14557 
14558                 /*
14559                  * The rest of this function implements Neighbor Unreachability
14560                  * detection. Determine if the ncec is eligible for NUD.
14561                  */
14562                 if (ncec->ncec_flags & NCE_F_NONUD)
14563                         return (0);
14564 
14565                 ASSERT(ncec->ncec_state != ND_INCOMPLETE);
14566 
14567                 /*
14568                  * Check for upper layer advice
14569                  */
14570                 if (ixaflags & IXAF_REACH_CONF) {
14571                         timeout_id_t tid;
14572 
14573                         /*
14574                          * It should be o.k. to check the state without
14575                          * a lock here, at most we lose an advice.
14576                          */
14577                         ncec->ncec_last = TICK_TO_MSEC(now);
14578                         if (ncec->ncec_state != ND_REACHABLE) {
14579                                 mutex_enter(&ncec->ncec_lock);
14580                                 ncec->ncec_state = ND_REACHABLE;
14581                                 tid = ncec->ncec_timeout_id;
14582                                 ncec->ncec_timeout_id = 0;
14583                                 mutex_exit(&ncec->ncec_lock);
14584                                 (void) untimeout(tid);
14585                                 if (ip_debug > 2) {
14586                                         /* ip1dbg */
14587                                         pr_addr_dbg("ip_xmit: state"
14588                                             " for %s changed to"
14589                                             " REACHABLE\n", AF_INET6,
14590                                             &ncec->ncec_addr);
14591                                 }
14592                         }
14593                         return (0);
14594                 }
14595 
14596                 delta =  TICK_TO_MSEC(now) - ncec->ncec_last;
14597                 ip1dbg(("ip_xmit: delta = %" PRId64
14598                     " ill_reachable_time = %d \n", delta,
14599                     ill->ill_reachable_time));
14600                 if (delta > (uint64_t)ill->ill_reachable_time) {
14601                         mutex_enter(&ncec->ncec_lock);
14602                         switch (ncec->ncec_state) {
14603                         case ND_REACHABLE:
14604                                 ASSERT((ncec->ncec_flags & NCE_F_NONUD) == 0);
14605                                 /* FALLTHROUGH */
14606                         case ND_STALE:
14607                                 /*
14608                                  * ND_REACHABLE is identical to
14609                                  * ND_STALE in this specific case. If
14610                                  * reachable time has expired for this
14611                                  * neighbor (delta is greater than
14612                                  * reachable time), conceptually, the
14613                                  * neighbor cache is no longer in
14614                                  * REACHABLE state, but already in
14615                                  * STALE state.  So the correct
14616                                  * transition here is to ND_DELAY.
14617                                  */
14618                                 ncec->ncec_state = ND_DELAY;
14619                                 mutex_exit(&ncec->ncec_lock);
14620                                 nce_restart_timer(ncec,
14621                                     ipst->ips_delay_first_probe_time);
14622                                 if (ip_debug > 3) {
14623                                         /* ip2dbg */
14624                                         pr_addr_dbg("ip_xmit: state"
14625                                             " for %s changed to"
14626                                             " DELAY\n", AF_INET6,
14627                                             &ncec->ncec_addr);
14628                                 }
14629                                 break;
14630                         case ND_DELAY:
14631                         case ND_PROBE:
14632                                 mutex_exit(&ncec->ncec_lock);
14633                                 /* Timers have already started */
14634                                 break;
14635                         case ND_UNREACHABLE:
14636                                 /*
14637                                  * nce_timer has detected that this ncec
14638                                  * is unreachable and initiated deleting
14639                                  * this ncec.
14640                                  * This is a harmless race where we found the
14641                                  * ncec before it was deleted and have
14642                                  * just sent out a packet using this
14643                                  * unreachable ncec.
14644                                  */
14645                                 mutex_exit(&ncec->ncec_lock);
14646                                 break;
14647                         default:
14648                                 ASSERT(0);
14649                                 mutex_exit(&ncec->ncec_lock);
14650                         }
14651                 }
14652                 return (0);
14653 
14654         case ND_INCOMPLETE:
14655                 /*
14656                  * the state could have changed since we didn't hold the lock.
14657                  * Re-verify state under lock.
14658                  */
14659                 is_probe = ipmp_packet_is_probe(mp, nce->nce_ill);
14660                 mutex_enter(&ncec->ncec_lock);
14661                 if (NCE_ISREACHABLE(ncec)) {
14662                         mutex_exit(&ncec->ncec_lock);
14663                         goto sendit;
14664                 }
14665                 /* queue the packet */
14666                 nce_queue_mp(ncec, mp, is_probe);
14667                 mutex_exit(&ncec->ncec_lock);
14668                 DTRACE_PROBE2(ip__xmit__incomplete,
14669                     (ncec_t *), ncec, (mblk_t *), mp);
14670                 return (0);
14671 
14672         case ND_INITIAL:
14673                 /*
14674                  * State could have changed since we didn't hold the lock, so
14675                  * re-verify state.
14676                  */
14677                 is_probe = ipmp_packet_is_probe(mp, nce->nce_ill);
14678                 mutex_enter(&ncec->ncec_lock);
14679                 if (NCE_ISREACHABLE(ncec))  {
14680                         mutex_exit(&ncec->ncec_lock);
14681                         goto sendit;
14682                 }
14683                 nce_queue_mp(ncec, mp, is_probe);
14684                 if (ncec->ncec_state == ND_INITIAL) {
14685                         ncec->ncec_state = ND_INCOMPLETE;
14686                         mutex_exit(&ncec->ncec_lock);
14687                         /*
14688                          * figure out the source we want to use
14689                          * and resolve it.
14690                          */
14691                         ip_ndp_resolve(ncec);
14692                 } else  {
14693                         mutex_exit(&ncec->ncec_lock);
14694                 }
14695                 return (0);
14696 
14697         case ND_UNREACHABLE:
14698                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
14699                 ip_drop_output("ipIfStatsOutDiscards - ND_UNREACHABLE",
14700                     mp, ill);
14701                 freemsg(mp);
14702                 return (0);
14703 
14704         default:
14705                 ASSERT(0);
14706                 BUMP_MIB(ill->ill_ip_mib, ipIfStatsOutDiscards);
14707                 ip_drop_output("ipIfStatsOutDiscards - ND_other",
14708                     mp, ill);
14709                 freemsg(mp);
14710                 return (ENETUNREACH);
14711         }
14712 }
14713 
14714 /*
14715  * Return B_TRUE if the buffers differ in length or content.
14716  * This is used for comparing extension header buffers.
14717  * Note that an extension header would be declared different
14718  * even if all that changed was the next header value in that header i.e.
14719  * what really changed is the next extension header.
14720  */
14721 boolean_t
14722 ip_cmpbuf(const void *abuf, uint_t alen, boolean_t b_valid, const void *bbuf,
14723     uint_t blen)
14724 {
14725         if (!b_valid)
14726                 blen = 0;
14727 
14728         if (alen != blen)
14729                 return (B_TRUE);
14730         if (alen == 0)
14731                 return (B_FALSE);       /* Both zero length */
14732         return (bcmp(abuf, bbuf, alen));
14733 }
14734 
14735 /*
14736  * Preallocate memory for ip_savebuf(). Returns B_TRUE if ok.
14737  * Return B_FALSE if memory allocation fails - don't change any state!
14738  */
14739 boolean_t
14740 ip_allocbuf(void **dstp, uint_t *dstlenp, boolean_t src_valid,
14741     const void *src, uint_t srclen)
14742 {
14743         void *dst;
14744 
14745         if (!src_valid)
14746                 srclen = 0;
14747 
14748         ASSERT(*dstlenp == 0);
14749         if (src != NULL && srclen != 0) {
14750                 dst = mi_alloc(srclen, BPRI_MED);
14751                 if (dst == NULL)
14752                         return (B_FALSE);
14753         } else {
14754                 dst = NULL;
14755         }
14756         if (*dstp != NULL)
14757                 mi_free(*dstp);
14758         *dstp = dst;
14759         *dstlenp = dst == NULL ? 0 : srclen;
14760         return (B_TRUE);
14761 }
14762 
14763 /*
14764  * Replace what is in *dst, *dstlen with the source.
14765  * Assumes ip_allocbuf has already been called.
14766  */
14767 void
14768 ip_savebuf(void **dstp, uint_t *dstlenp, boolean_t src_valid,
14769     const void *src, uint_t srclen)
14770 {
14771         if (!src_valid)
14772                 srclen = 0;
14773 
14774         ASSERT(*dstlenp == srclen);
14775         if (src != NULL && srclen != 0)
14776                 bcopy(src, *dstp, srclen);
14777 }
14778 
14779 /*
14780  * Free the storage pointed to by the members of an ip_pkt_t.
14781  */
14782 void
14783 ip_pkt_free(ip_pkt_t *ipp)
14784 {
14785         uint_t  fields = ipp->ipp_fields;
14786 
14787         if (fields & IPPF_HOPOPTS) {
14788                 kmem_free(ipp->ipp_hopopts, ipp->ipp_hopoptslen);
14789                 ipp->ipp_hopopts = NULL;
14790                 ipp->ipp_hopoptslen = 0;
14791         }
14792         if (fields & IPPF_RTHDRDSTOPTS) {
14793                 kmem_free(ipp->ipp_rthdrdstopts, ipp->ipp_rthdrdstoptslen);
14794                 ipp->ipp_rthdrdstopts = NULL;
14795                 ipp->ipp_rthdrdstoptslen = 0;
14796         }
14797         if (fields & IPPF_DSTOPTS) {
14798                 kmem_free(ipp->ipp_dstopts, ipp->ipp_dstoptslen);
14799                 ipp->ipp_dstopts = NULL;
14800                 ipp->ipp_dstoptslen = 0;
14801         }
14802         if (fields & IPPF_RTHDR) {
14803                 kmem_free(ipp->ipp_rthdr, ipp->ipp_rthdrlen);
14804                 ipp->ipp_rthdr = NULL;
14805                 ipp->ipp_rthdrlen = 0;
14806         }
14807         if (fields & IPPF_IPV4_OPTIONS) {
14808                 kmem_free(ipp->ipp_ipv4_options, ipp->ipp_ipv4_options_len);
14809                 ipp->ipp_ipv4_options = NULL;
14810                 ipp->ipp_ipv4_options_len = 0;
14811         }
14812         if (fields & IPPF_LABEL_V4) {
14813                 kmem_free(ipp->ipp_label_v4, ipp->ipp_label_len_v4);
14814                 ipp->ipp_label_v4 = NULL;
14815                 ipp->ipp_label_len_v4 = 0;
14816         }
14817         if (fields & IPPF_LABEL_V6) {
14818                 kmem_free(ipp->ipp_label_v6, ipp->ipp_label_len_v6);
14819                 ipp->ipp_label_v6 = NULL;
14820                 ipp->ipp_label_len_v6 = 0;
14821         }
14822         ipp->ipp_fields &= ~(IPPF_HOPOPTS | IPPF_RTHDRDSTOPTS | IPPF_DSTOPTS |
14823             IPPF_RTHDR | IPPF_IPV4_OPTIONS | IPPF_LABEL_V4 | IPPF_LABEL_V6);
14824 }
14825 
14826 /*
14827  * Copy from src to dst and allocate as needed.
14828  * Returns zero or ENOMEM.
14829  *
14830  * The caller must initialize dst to zero.
14831  */
14832 int
14833 ip_pkt_copy(ip_pkt_t *src, ip_pkt_t *dst, int kmflag)
14834 {
14835         uint_t  fields = src->ipp_fields;
14836 
14837         /* Start with fields that don't require memory allocation */
14838         dst->ipp_fields = fields &
14839             ~(IPPF_HOPOPTS | IPPF_RTHDRDSTOPTS | IPPF_DSTOPTS |
14840             IPPF_RTHDR | IPPF_IPV4_OPTIONS | IPPF_LABEL_V4 | IPPF_LABEL_V6);
14841 
14842         dst->ipp_addr = src->ipp_addr;
14843         dst->ipp_unicast_hops = src->ipp_unicast_hops;
14844         dst->ipp_hoplimit = src->ipp_hoplimit;
14845         dst->ipp_tclass = src->ipp_tclass;
14846         dst->ipp_type_of_service = src->ipp_type_of_service;
14847 
14848         if (!(fields & (IPPF_HOPOPTS | IPPF_RTHDRDSTOPTS | IPPF_DSTOPTS |
14849             IPPF_RTHDR | IPPF_IPV4_OPTIONS | IPPF_LABEL_V4 | IPPF_LABEL_V6)))
14850                 return (0);
14851 
14852         if (fields & IPPF_HOPOPTS) {
14853                 dst->ipp_hopopts = kmem_alloc(src->ipp_hopoptslen, kmflag);
14854                 if (dst->ipp_hopopts == NULL) {
14855                         ip_pkt_free(dst);
14856                         return (ENOMEM);
14857                 }
14858                 dst->ipp_fields |= IPPF_HOPOPTS;
14859                 bcopy(src->ipp_hopopts, dst->ipp_hopopts,
14860                     src->ipp_hopoptslen);
14861                 dst->ipp_hopoptslen = src->ipp_hopoptslen;
14862         }
14863         if (fields & IPPF_RTHDRDSTOPTS) {
14864                 dst->ipp_rthdrdstopts = kmem_alloc(src->ipp_rthdrdstoptslen,
14865                     kmflag);
14866                 if (dst->ipp_rthdrdstopts == NULL) {
14867                         ip_pkt_free(dst);
14868                         return (ENOMEM);
14869                 }
14870                 dst->ipp_fields |= IPPF_RTHDRDSTOPTS;
14871                 bcopy(src->ipp_rthdrdstopts, dst->ipp_rthdrdstopts,
14872                     src->ipp_rthdrdstoptslen);
14873                 dst->ipp_rthdrdstoptslen = src->ipp_rthdrdstoptslen;
14874         }
14875         if (fields & IPPF_DSTOPTS) {
14876                 dst->ipp_dstopts = kmem_alloc(src->ipp_dstoptslen, kmflag);
14877                 if (dst->ipp_dstopts == NULL) {
14878                         ip_pkt_free(dst);
14879                         return (ENOMEM);
14880                 }
14881                 dst->ipp_fields |= IPPF_DSTOPTS;
14882                 bcopy(src->ipp_dstopts, dst->ipp_dstopts,
14883                     src->ipp_dstoptslen);
14884                 dst->ipp_dstoptslen = src->ipp_dstoptslen;
14885         }
14886         if (fields & IPPF_RTHDR) {
14887                 dst->ipp_rthdr = kmem_alloc(src->ipp_rthdrlen, kmflag);
14888                 if (dst->ipp_rthdr == NULL) {
14889                         ip_pkt_free(dst);
14890                         return (ENOMEM);
14891                 }
14892                 dst->ipp_fields |= IPPF_RTHDR;
14893                 bcopy(src->ipp_rthdr, dst->ipp_rthdr,
14894                     src->ipp_rthdrlen);
14895                 dst->ipp_rthdrlen = src->ipp_rthdrlen;
14896         }
14897         if (fields & IPPF_IPV4_OPTIONS) {
14898                 dst->ipp_ipv4_options = kmem_alloc(src->ipp_ipv4_options_len,
14899                     kmflag);
14900                 if (dst->ipp_ipv4_options == NULL) {
14901                         ip_pkt_free(dst);
14902                         return (ENOMEM);
14903                 }
14904                 dst->ipp_fields |= IPPF_IPV4_OPTIONS;
14905                 bcopy(src->ipp_ipv4_options, dst->ipp_ipv4_options,
14906                     src->ipp_ipv4_options_len);
14907                 dst->ipp_ipv4_options_len = src->ipp_ipv4_options_len;
14908         }
14909         if (fields & IPPF_LABEL_V4) {
14910                 dst->ipp_label_v4 = kmem_alloc(src->ipp_label_len_v4, kmflag);
14911                 if (dst->ipp_label_v4 == NULL) {
14912                         ip_pkt_free(dst);
14913                         return (ENOMEM);
14914                 }
14915                 dst->ipp_fields |= IPPF_LABEL_V4;
14916                 bcopy(src->ipp_label_v4, dst->ipp_label_v4,
14917                     src->ipp_label_len_v4);
14918                 dst->ipp_label_len_v4 = src->ipp_label_len_v4;
14919         }
14920         if (fields & IPPF_LABEL_V6) {
14921                 dst->ipp_label_v6 = kmem_alloc(src->ipp_label_len_v6, kmflag);
14922                 if (dst->ipp_label_v6 == NULL) {
14923                         ip_pkt_free(dst);
14924                         return (ENOMEM);
14925                 }
14926                 dst->ipp_fields |= IPPF_LABEL_V6;
14927                 bcopy(src->ipp_label_v6, dst->ipp_label_v6,
14928                     src->ipp_label_len_v6);
14929                 dst->ipp_label_len_v6 = src->ipp_label_len_v6;
14930         }
14931         if (fields & IPPF_FRAGHDR) {
14932                 dst->ipp_fraghdr = kmem_alloc(src->ipp_fraghdrlen, kmflag);
14933                 if (dst->ipp_fraghdr == NULL) {
14934                         ip_pkt_free(dst);
14935                         return (ENOMEM);
14936                 }
14937                 dst->ipp_fields |= IPPF_FRAGHDR;
14938                 bcopy(src->ipp_fraghdr, dst->ipp_fraghdr,
14939                     src->ipp_fraghdrlen);
14940                 dst->ipp_fraghdrlen = src->ipp_fraghdrlen;
14941         }
14942         return (0);
14943 }
14944 
14945 /*
14946  * Returns INADDR_ANY if no source route
14947  */
14948 ipaddr_t
14949 ip_pkt_source_route_v4(const ip_pkt_t *ipp)
14950 {
14951         ipaddr_t        nexthop = INADDR_ANY;
14952         ipoptp_t        opts;
14953         uchar_t         *opt;
14954         uint8_t         optval;
14955         uint8_t         optlen;
14956         uint32_t        totallen;
14957 
14958         if (!(ipp->ipp_fields & IPPF_IPV4_OPTIONS))
14959                 return (INADDR_ANY);
14960 
14961         totallen = ipp->ipp_ipv4_options_len;
14962         if (totallen & 0x3)
14963                 return (INADDR_ANY);
14964 
14965         for (optval = ipoptp_first2(&opts, totallen, ipp->ipp_ipv4_options);
14966             optval != IPOPT_EOL;
14967             optval = ipoptp_next(&opts)) {
14968                 opt = opts.ipoptp_cur;
14969                 switch (optval) {
14970                         uint8_t off;
14971                 case IPOPT_SSRR:
14972                 case IPOPT_LSRR:
14973                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
14974                                 break;
14975                         }
14976                         optlen = opts.ipoptp_len;
14977                         off = opt[IPOPT_OFFSET];
14978                         off--;
14979                         if (optlen < IP_ADDR_LEN ||
14980                             off > optlen - IP_ADDR_LEN) {
14981                                 /* End of source route */
14982                                 break;
14983                         }
14984                         bcopy((char *)opt + off, &nexthop, IP_ADDR_LEN);
14985                         if (nexthop == htonl(INADDR_LOOPBACK)) {
14986                                 /* Ignore */
14987                                 nexthop = INADDR_ANY;
14988                                 break;
14989                         }
14990                         break;
14991                 }
14992         }
14993         return (nexthop);
14994 }
14995 
14996 /*
14997  * Reverse a source route.
14998  */
14999 void
15000 ip_pkt_source_route_reverse_v4(ip_pkt_t *ipp)
15001 {
15002         ipaddr_t        tmp;
15003         ipoptp_t        opts;
15004         uchar_t         *opt;
15005         uint8_t         optval;
15006         uint32_t        totallen;
15007 
15008         if (!(ipp->ipp_fields & IPPF_IPV4_OPTIONS))
15009                 return;
15010 
15011         totallen = ipp->ipp_ipv4_options_len;
15012         if (totallen & 0x3)
15013                 return;
15014 
15015         for (optval = ipoptp_first2(&opts, totallen, ipp->ipp_ipv4_options);
15016             optval != IPOPT_EOL;
15017             optval = ipoptp_next(&opts)) {
15018                 uint8_t off1, off2;
15019 
15020                 opt = opts.ipoptp_cur;
15021                 switch (optval) {
15022                 case IPOPT_SSRR:
15023                 case IPOPT_LSRR:
15024                         if ((opts.ipoptp_flags & IPOPTP_ERROR) != 0) {
15025                                 break;
15026                         }
15027                         off1 = IPOPT_MINOFF_SR - 1;
15028                         off2 = opt[IPOPT_OFFSET] - IP_ADDR_LEN - 1;
15029                         while (off2 > off1) {
15030                                 bcopy(opt + off2, &tmp, IP_ADDR_LEN);
15031                                 bcopy(opt + off1, opt + off2, IP_ADDR_LEN);
15032                                 bcopy(&tmp, opt + off2, IP_ADDR_LEN);
15033                                 off2 -= IP_ADDR_LEN;
15034                                 off1 += IP_ADDR_LEN;
15035                         }
15036                         opt[IPOPT_OFFSET] = IPOPT_MINOFF_SR;
15037                         break;
15038                 }
15039         }
15040 }
15041 
15042 /*
15043  * Returns NULL if no routing header
15044  */
15045 in6_addr_t *
15046 ip_pkt_source_route_v6(const ip_pkt_t *ipp)
15047 {
15048         in6_addr_t      *nexthop = NULL;
15049         ip6_rthdr0_t    *rthdr;
15050 
15051         if (!(ipp->ipp_fields & IPPF_RTHDR))
15052                 return (NULL);
15053 
15054         rthdr = (ip6_rthdr0_t *)ipp->ipp_rthdr;
15055         if (rthdr->ip6r0_segleft == 0)
15056                 return (NULL);
15057 
15058         nexthop = (in6_addr_t *)((char *)rthdr + sizeof (*rthdr));
15059         return (nexthop);
15060 }
15061 
15062 zoneid_t
15063 ip_get_zoneid_v4(ipaddr_t addr, mblk_t *mp, ip_recv_attr_t *ira,
15064     zoneid_t lookup_zoneid)
15065 {
15066         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
15067         ire_t           *ire;
15068         int             ire_flags = MATCH_IRE_TYPE;
15069         zoneid_t        zoneid = ALL_ZONES;
15070 
15071         if (is_system_labeled() && !tsol_can_accept_raw(mp, ira, B_FALSE))
15072                 return (ALL_ZONES);
15073 
15074         if (lookup_zoneid != ALL_ZONES)
15075                 ire_flags |= MATCH_IRE_ZONEONLY;
15076         ire = ire_ftable_lookup_v4(addr, NULL, NULL, IRE_LOCAL | IRE_LOOPBACK,
15077             NULL, lookup_zoneid, NULL, ire_flags, 0, ipst, NULL);
15078         if (ire != NULL) {
15079                 zoneid = IP_REAL_ZONEID(ire->ire_zoneid, ipst);
15080                 ire_refrele(ire);
15081         }
15082         return (zoneid);
15083 }
15084 
15085 zoneid_t
15086 ip_get_zoneid_v6(in6_addr_t *addr, mblk_t *mp, const ill_t *ill,
15087     ip_recv_attr_t *ira, zoneid_t lookup_zoneid)
15088 {
15089         ip_stack_t      *ipst = ira->ira_ill->ill_ipst;
15090         ire_t           *ire;
15091         int             ire_flags = MATCH_IRE_TYPE;
15092         zoneid_t        zoneid = ALL_ZONES;
15093 
15094         if (is_system_labeled() && !tsol_can_accept_raw(mp, ira, B_FALSE))
15095                 return (ALL_ZONES);
15096 
15097         if (IN6_IS_ADDR_LINKLOCAL(addr))
15098                 ire_flags |= MATCH_IRE_ILL;
15099 
15100         if (lookup_zoneid != ALL_ZONES)
15101                 ire_flags |= MATCH_IRE_ZONEONLY;
15102         ire = ire_ftable_lookup_v6(addr, NULL, NULL, IRE_LOCAL | IRE_LOOPBACK,
15103             ill, lookup_zoneid, NULL, ire_flags, 0, ipst, NULL);
15104         if (ire != NULL) {
15105                 zoneid = IP_REAL_ZONEID(ire->ire_zoneid, ipst);
15106                 ire_refrele(ire);
15107         }
15108         return (zoneid);
15109 }
15110 
15111 /*
15112  * IP obserability hook support functions.
15113  */
15114 static void
15115 ipobs_init(ip_stack_t *ipst)
15116 {
15117         netid_t id;
15118 
15119         id = net_getnetidbynetstackid(ipst->ips_netstack->netstack_stackid);
15120 
15121         ipst->ips_ip4_observe_pr = net_protocol_lookup(id, NHF_INET);
15122         VERIFY(ipst->ips_ip4_observe_pr != NULL);
15123 
15124         ipst->ips_ip6_observe_pr = net_protocol_lookup(id, NHF_INET6);
15125         VERIFY(ipst->ips_ip6_observe_pr != NULL);
15126 }
15127 
15128 static void
15129 ipobs_fini(ip_stack_t *ipst)
15130 {
15131 
15132         VERIFY(net_protocol_release(ipst->ips_ip4_observe_pr) == 0);
15133         VERIFY(net_protocol_release(ipst->ips_ip6_observe_pr) == 0);
15134 }
15135 
15136 /*
15137  * hook_pkt_observe_t is composed in network byte order so that the
15138  * entire mblk_t chain handed into hook_run can be used as-is.
15139  * The caveat is that use of the fields, such as the zone fields,
15140  * requires conversion into host byte order first.
15141  */
15142 void
15143 ipobs_hook(mblk_t *mp, int htype, zoneid_t zsrc, zoneid_t zdst,
15144     const ill_t *ill, ip_stack_t *ipst)
15145 {
15146         hook_pkt_observe_t *hdr;
15147         uint64_t grifindex;
15148         mblk_t *imp;
15149 
15150         imp = allocb(sizeof (*hdr), BPRI_HI);
15151         if (imp == NULL)
15152                 return;
15153 
15154         hdr = (hook_pkt_observe_t *)imp->b_rptr;
15155         /*
15156          * b_wptr is set to make the apparent size of the data in the mblk_t
15157          * to exclude the pointers at the end of hook_pkt_observer_t.
15158          */
15159         imp->b_wptr = imp->b_rptr + sizeof (dl_ipnetinfo_t);
15160         imp->b_cont = mp;
15161 
15162         ASSERT(DB_TYPE(mp) == M_DATA);
15163 
15164         if (IS_UNDER_IPMP(ill))
15165                 grifindex = ipmp_ill_get_ipmp_ifindex(ill);
15166         else
15167                 grifindex = 0;
15168 
15169         hdr->hpo_version = 1;
15170         hdr->hpo_htype = htons(htype);
15171         hdr->hpo_pktlen = htonl((ulong_t)msgdsize(mp));
15172         hdr->hpo_ifindex = htonl(ill->ill_phyint->phyint_ifindex);
15173         hdr->hpo_grifindex = htonl(grifindex);
15174         hdr->hpo_zsrc = htonl(zsrc);
15175         hdr->hpo_zdst = htonl(zdst);
15176         hdr->hpo_pkt = imp;
15177         hdr->hpo_ctx = ipst->ips_netstack;
15178 
15179         if (ill->ill_isv6) {
15180                 hdr->hpo_family = AF_INET6;
15181                 (void) hook_run(ipst->ips_ipv6_net_data->netd_hooks,
15182                     ipst->ips_ipv6observing, (hook_data_t)hdr);
15183         } else {
15184                 hdr->hpo_family = AF_INET;
15185                 (void) hook_run(ipst->ips_ipv4_net_data->netd_hooks,
15186                     ipst->ips_ipv4observing, (hook_data_t)hdr);
15187         }
15188 
15189         imp->b_cont = NULL;
15190         freemsg(imp);
15191 }
15192 
15193 /*
15194  * Utility routine that checks if `v4srcp' is a valid address on underlying
15195  * interface `ill'.  If `ipifp' is non-NULL, it's set to a held ipif
15196  * associated with `v4srcp' on success.  NOTE: if this is not called from
15197  * inside the IPSQ (ill_g_lock is not held), `ill' may be removed from the
15198  * group during or after this lookup.
15199  */
15200 boolean_t
15201 ipif_lookup_testaddr_v4(ill_t *ill, const in_addr_t *v4srcp, ipif_t **ipifp)
15202 {
15203         ipif_t *ipif;
15204 
15205         ipif = ipif_lookup_addr_exact(*v4srcp, ill, ill->ill_ipst);
15206         if (ipif != NULL) {
15207                 if (ipifp != NULL)
15208                         *ipifp = ipif;
15209                 else
15210                         ipif_refrele(ipif);
15211                 return (B_TRUE);
15212         }
15213 
15214         ip1dbg(("ipif_lookup_testaddr_v4: cannot find ipif for src %x\n",
15215             *v4srcp));
15216         return (B_FALSE);
15217 }
15218 
15219 /*
15220  * Transport protocol call back function for CPU state change.
15221  */
15222 /* ARGSUSED */
15223 static int
15224 ip_tp_cpu_update(cpu_setup_t what, int id, void *arg)
15225 {
15226         processorid_t cpu_seqid;
15227         netstack_handle_t nh;
15228         netstack_t *ns;
15229 
15230         ASSERT(MUTEX_HELD(&cpu_lock));
15231 
15232         switch (what) {
15233         case CPU_CONFIG:
15234         case CPU_ON:
15235         case CPU_INIT:
15236         case CPU_CPUPART_IN:
15237                 cpu_seqid = cpu[id]->cpu_seqid;
15238                 netstack_next_init(&nh);
15239                 while ((ns = netstack_next(&nh)) != NULL) {
15240                         tcp_stack_cpu_add(ns->netstack_tcp, cpu_seqid);
15241                         sctp_stack_cpu_add(ns->netstack_sctp, cpu_seqid);
15242                         udp_stack_cpu_add(ns->netstack_udp, cpu_seqid);
15243                         netstack_rele(ns);
15244                 }
15245                 netstack_next_fini(&nh);
15246                 break;
15247         case CPU_UNCONFIG:
15248         case CPU_OFF:
15249         case CPU_CPUPART_OUT:
15250                 /*
15251                  * Nothing to do.  We don't remove the per CPU stats from
15252                  * the IP stack even when the CPU goes offline.
15253                  */
15254                 break;
15255         default:
15256                 break;
15257         }
15258         return (0);
15259 }