17c478bd9Sstevel@tonic-gate /* 27c478bd9Sstevel@tonic-gate * CDDL HEADER START 37c478bd9Sstevel@tonic-gate * 47c478bd9Sstevel@tonic-gate * The contents of this file are subject to the terms of the 545916cd2Sjpk * Common Development and Distribution License (the "License"). 645916cd2Sjpk * You may not use this file except in compliance with the License. 77c478bd9Sstevel@tonic-gate * 87c478bd9Sstevel@tonic-gate * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 97c478bd9Sstevel@tonic-gate * or http://www.opensolaris.org/os/licensing. 107c478bd9Sstevel@tonic-gate * See the License for the specific language governing permissions 117c478bd9Sstevel@tonic-gate * and limitations under the License. 127c478bd9Sstevel@tonic-gate * 137c478bd9Sstevel@tonic-gate * When distributing Covered Code, include this CDDL HEADER in each 147c478bd9Sstevel@tonic-gate * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 157c478bd9Sstevel@tonic-gate * If applicable, add the following below this CDDL HEADER, with the 167c478bd9Sstevel@tonic-gate * fields enclosed by brackets "[]" replaced with your own identifying 177c478bd9Sstevel@tonic-gate * information: Portions Copyright [yyyy] [name of copyright owner] 187c478bd9Sstevel@tonic-gate * 197c478bd9Sstevel@tonic-gate * CDDL HEADER END 207c478bd9Sstevel@tonic-gate */ 217c478bd9Sstevel@tonic-gate /* 229e3469d3SErik Nordmark * Copyright 2010 Sun Microsystems, Inc. All rights reserved. 237c478bd9Sstevel@tonic-gate * Use is subject to license terms. 247c478bd9Sstevel@tonic-gate */ 257c478bd9Sstevel@tonic-gate /* Copyright (c) 1990 Mentat Inc. */ 267c478bd9Sstevel@tonic-gate 277c478bd9Sstevel@tonic-gate /* 287c478bd9Sstevel@tonic-gate * This file contains routines that manipulate Internet Routing Entries (IREs). 297c478bd9Sstevel@tonic-gate */ 307c478bd9Sstevel@tonic-gate 317c478bd9Sstevel@tonic-gate #include <sys/types.h> 327c478bd9Sstevel@tonic-gate #include <sys/stream.h> 337c478bd9Sstevel@tonic-gate #include <sys/stropts.h> 34e11c3f44Smeem #include <sys/strsun.h> 35de8c4a14SErik Nordmark #include <sys/strsubr.h> 367c478bd9Sstevel@tonic-gate #include <sys/ddi.h> 377c478bd9Sstevel@tonic-gate #include <sys/cmn_err.h> 387c478bd9Sstevel@tonic-gate #include <sys/policy.h> 397c478bd9Sstevel@tonic-gate 407c478bd9Sstevel@tonic-gate #include <sys/systm.h> 417c478bd9Sstevel@tonic-gate #include <sys/kmem.h> 427c478bd9Sstevel@tonic-gate #include <sys/param.h> 437c478bd9Sstevel@tonic-gate #include <sys/socket.h> 447c478bd9Sstevel@tonic-gate #include <net/if.h> 457c478bd9Sstevel@tonic-gate #include <net/route.h> 467c478bd9Sstevel@tonic-gate #include <netinet/in.h> 477c478bd9Sstevel@tonic-gate #include <net/if_dl.h> 487c478bd9Sstevel@tonic-gate #include <netinet/ip6.h> 497c478bd9Sstevel@tonic-gate #include <netinet/icmp6.h> 507c478bd9Sstevel@tonic-gate 517c478bd9Sstevel@tonic-gate #include <inet/common.h> 527c478bd9Sstevel@tonic-gate #include <inet/mi.h> 537c478bd9Sstevel@tonic-gate #include <inet/ip.h> 547c478bd9Sstevel@tonic-gate #include <inet/ip6.h> 557c478bd9Sstevel@tonic-gate #include <inet/ip_ndp.h> 56c793af95Ssangeeta #include <inet/arp.h> 577c478bd9Sstevel@tonic-gate #include <inet/ip_if.h> 587c478bd9Sstevel@tonic-gate #include <inet/ip_ire.h> 59c793af95Ssangeeta #include <inet/ip_ftable.h> 607c478bd9Sstevel@tonic-gate #include <inet/ip_rts.h> 617c478bd9Sstevel@tonic-gate #include <inet/nd.h> 626e91bba0SGirish Moodalbail #include <inet/tunables.h> 637c478bd9Sstevel@tonic-gate 647c478bd9Sstevel@tonic-gate #include <inet/tcp.h> 657c478bd9Sstevel@tonic-gate #include <inet/ipclassifier.h> 667c478bd9Sstevel@tonic-gate #include <sys/zone.h> 67f4b3ec61Sdh155122 #include <sys/cpuvar.h> 68f4b3ec61Sdh155122 6945916cd2Sjpk #include <sys/tsol/label.h> 7045916cd2Sjpk #include <sys/tsol/tnet.h> 7145916cd2Sjpk 72c793af95Ssangeeta struct kmem_cache *rt_entry_cache; 73c793af95Ssangeeta 74bd670b35SErik Nordmark typedef struct nce_clookup_s { 75bd670b35SErik Nordmark ipaddr_t ncecl_addr; 76bd670b35SErik Nordmark boolean_t ncecl_found; 77bd670b35SErik Nordmark } nce_clookup_t; 78bd670b35SErik Nordmark 797c478bd9Sstevel@tonic-gate /* 807c478bd9Sstevel@tonic-gate * Synchronization notes: 817c478bd9Sstevel@tonic-gate * 827c478bd9Sstevel@tonic-gate * The fields of the ire_t struct are protected in the following way : 837c478bd9Sstevel@tonic-gate * 847c478bd9Sstevel@tonic-gate * ire_next/ire_ptpn 857c478bd9Sstevel@tonic-gate * 86bd670b35SErik Nordmark * - bucket lock of the forwarding table in which is ire stored. 877c478bd9Sstevel@tonic-gate * 88bd670b35SErik Nordmark * ire_ill, ire_u *except* ire_gateway_addr[v6], ire_mask, 89bd670b35SErik Nordmark * ire_type, ire_create_time, ire_masklen, ire_ipversion, ire_flags, 90bd670b35SErik Nordmark * ire_bucket 917c478bd9Sstevel@tonic-gate * 927c478bd9Sstevel@tonic-gate * - Set in ire_create_v4/v6 and never changes after that. Thus, 937c478bd9Sstevel@tonic-gate * we don't need a lock whenever these fields are accessed. 947c478bd9Sstevel@tonic-gate * 957c478bd9Sstevel@tonic-gate * - ire_bucket and ire_masklen (also set in ire_create) is set in 96bd670b35SErik Nordmark * ire_add before inserting in the bucket and never 977c478bd9Sstevel@tonic-gate * changes after that. Thus we don't need a lock whenever these 987c478bd9Sstevel@tonic-gate * fields are accessed. 997c478bd9Sstevel@tonic-gate * 1007c478bd9Sstevel@tonic-gate * ire_gateway_addr_v4[v6] 1017c478bd9Sstevel@tonic-gate * 1027c478bd9Sstevel@tonic-gate * - ire_gateway_addr_v4[v6] is set during ire_create and later modified 1037c478bd9Sstevel@tonic-gate * by rts_setgwr[v6]. As ire_gateway_addr is a uint32_t, updates to 1047c478bd9Sstevel@tonic-gate * it assumed to be atomic and hence the other parts of the code 1057c478bd9Sstevel@tonic-gate * does not use any locks. ire_gateway_addr_v6 updates are not atomic 1067c478bd9Sstevel@tonic-gate * and hence any access to it uses ire_lock to get/set the right value. 1077c478bd9Sstevel@tonic-gate * 108bd670b35SErik Nordmark * ire_refcnt, ire_identical_ref 1097c478bd9Sstevel@tonic-gate * 1107c478bd9Sstevel@tonic-gate * - Updated atomically using atomic_add_32 1117c478bd9Sstevel@tonic-gate * 1127c478bd9Sstevel@tonic-gate * ire_ssthresh, ire_rtt_sd, ire_rtt, ire_ib_pkt_count, ire_ob_pkt_count 1137c478bd9Sstevel@tonic-gate * 1147c478bd9Sstevel@tonic-gate * - Assumes that 32 bit writes are atomic. No locks. ire_lock is 1157c478bd9Sstevel@tonic-gate * used to serialize updates to ire_ssthresh, ire_rtt_sd, ire_rtt. 1167c478bd9Sstevel@tonic-gate * 117bd670b35SErik Nordmark * ire_generation 118bd670b35SErik Nordmark * - Under ire_lock 1197c478bd9Sstevel@tonic-gate * 120bd670b35SErik Nordmark * ire_nce_cache 121bd670b35SErik Nordmark * - Under ire_lock 1227c478bd9Sstevel@tonic-gate * 123bd670b35SErik Nordmark * ire_dep_parent (To next IRE in recursive lookup chain) 124bd670b35SErik Nordmark * - Under ips_ire_dep_lock. Write held when modifying. Read held when 125bd670b35SErik Nordmark * walking. We also hold ire_lock when modifying to allow the data path 126bd670b35SErik Nordmark * to only acquire ire_lock. 1277c478bd9Sstevel@tonic-gate * 128bd670b35SErik Nordmark * ire_dep_parent_generation (Generation number from ire_dep_parent) 129bd670b35SErik Nordmark * - Under ips_ire_dep_lock and/or ire_lock. (A read claim on the dep_lock 130bd670b35SErik Nordmark * and ire_lock held when modifying) 1317c478bd9Sstevel@tonic-gate * 132bd670b35SErik Nordmark * ire_dep_children (From parent to first child) 133bd670b35SErik Nordmark * ire_dep_sib_next (linked list of siblings) 134bd670b35SErik Nordmark * ire_dep_sib_ptpn (linked list of siblings) 135bd670b35SErik Nordmark * - Under ips_ire_dep_lock. Write held when modifying. Read held when 136bd670b35SErik Nordmark * walking. 1377c478bd9Sstevel@tonic-gate * 1387c478bd9Sstevel@tonic-gate * As we always hold the bucket locks in all the places while accessing 1397c478bd9Sstevel@tonic-gate * the above values, it is natural to use them for protecting them. 1407c478bd9Sstevel@tonic-gate * 141bd670b35SErik Nordmark * We have a forwarding table for IPv4 and IPv6. The IPv6 forwarding table 142219e6466Ssowmini * (ip_forwarding_table_v6) is an array of pointers to arrays of irb_t 143bd670b35SErik Nordmark * structures. ip_forwarding_table_v6 is allocated dynamically in 144f4b3ec61Sdh155122 * ire_add_v6. ire_ft_init_lock is used to serialize multiple threads 1457c478bd9Sstevel@tonic-gate * initializing the same bucket. Once a bucket is initialized, it is never 146f4b3ec61Sdh155122 * de-alloacted. This assumption enables us to access 147f4b3ec61Sdh155122 * ip_forwarding_table_v6[i] without any locks. 1487c478bd9Sstevel@tonic-gate * 149219e6466Ssowmini * The forwarding table for IPv4 is a radix tree whose leaves 150219e6466Ssowmini * are rt_entry structures containing the irb_t for the rt_dst. The irb_t 151219e6466Ssowmini * for IPv4 is dynamically allocated and freed. 152219e6466Ssowmini * 1537c478bd9Sstevel@tonic-gate * Each irb_t - ire bucket structure has a lock to protect 1547c478bd9Sstevel@tonic-gate * a bucket and the ires residing in the bucket have a back pointer to 1557c478bd9Sstevel@tonic-gate * the bucket structure. It also has a reference count for the number 1567c478bd9Sstevel@tonic-gate * of threads walking the bucket - irb_refcnt which is bumped up 157bd670b35SErik Nordmark * using the irb_refhold function. The flags irb_marks can be 158bd670b35SErik Nordmark * set to IRB_MARK_CONDEMNED indicating that there are some ires 159bd670b35SErik Nordmark * in this bucket that are IRE_IS_CONDEMNED and the 1607c478bd9Sstevel@tonic-gate * last thread to leave the bucket should delete the ires. Usually 161bd670b35SErik Nordmark * this is done by the irb_refrele function which is used to decrement 162219e6466Ssowmini * the reference count on a bucket. See comments above irb_t structure 163219e6466Ssowmini * definition in ip.h for further details. 1647c478bd9Sstevel@tonic-gate * 165bd670b35SErik Nordmark * The ire_refhold/ire_refrele functions operate on the ire which increments/ 1667c478bd9Sstevel@tonic-gate * decrements the reference count, ire_refcnt, atomically on the ire. 167bd670b35SErik Nordmark * ire_refcnt is modified only using those functions. Operations on the IRE 1687c478bd9Sstevel@tonic-gate * could be described as follows : 1697c478bd9Sstevel@tonic-gate * 1707c478bd9Sstevel@tonic-gate * CREATE an ire with reference count initialized to 1. 1717c478bd9Sstevel@tonic-gate * 1727c478bd9Sstevel@tonic-gate * ADDITION of an ire holds the bucket lock, checks for duplicates 173bd670b35SErik Nordmark * and then adds the ire. ire_add returns the ire after 1747c478bd9Sstevel@tonic-gate * bumping up once more i.e the reference count is 2. This is to avoid 1757c478bd9Sstevel@tonic-gate * an extra lookup in the functions calling ire_add which wants to 1767c478bd9Sstevel@tonic-gate * work with the ire after adding. 1777c478bd9Sstevel@tonic-gate * 178bd670b35SErik Nordmark * LOOKUP of an ire bumps up the reference count using ire_refhold 179bd670b35SErik Nordmark * function. It is valid to bump up the referece count of the IRE, 1807c478bd9Sstevel@tonic-gate * after the lookup has returned an ire. Following are the lookup 1817c478bd9Sstevel@tonic-gate * functions that return an HELD ire : 1827c478bd9Sstevel@tonic-gate * 183bd670b35SErik Nordmark * ire_ftable_lookup[_v6], ire_lookup_multi_ill[_v6] 1847c478bd9Sstevel@tonic-gate * 1857c478bd9Sstevel@tonic-gate * DELETION of an ire holds the bucket lock, removes it from the list 1867c478bd9Sstevel@tonic-gate * and then decrements the reference count for having removed from the list 187bd670b35SErik Nordmark * by using the ire_refrele function. If some other thread has looked up 1887c478bd9Sstevel@tonic-gate * the ire, the reference count would have been bumped up and hence 1897c478bd9Sstevel@tonic-gate * this ire will not be freed once deleted. It will be freed once the 1907c478bd9Sstevel@tonic-gate * reference count drops to zero. 1917c478bd9Sstevel@tonic-gate * 1927c478bd9Sstevel@tonic-gate * Add and Delete acquires the bucket lock as RW_WRITER, while all the 1937c478bd9Sstevel@tonic-gate * lookups acquire the bucket lock as RW_READER. 1947c478bd9Sstevel@tonic-gate * 195bd670b35SErik Nordmark * The general rule is to do the ire_refrele in the function 1967c478bd9Sstevel@tonic-gate * that is passing the ire as an argument. 1977c478bd9Sstevel@tonic-gate * 1987c478bd9Sstevel@tonic-gate * In trying to locate ires the following points are to be noted. 1997c478bd9Sstevel@tonic-gate * 200bd670b35SErik Nordmark * IRE_IS_CONDEMNED signifies that the ire has been logically deleted and is 2017c478bd9Sstevel@tonic-gate * to be ignored when walking the ires using ire_next. 2027c478bd9Sstevel@tonic-gate * 2037c478bd9Sstevel@tonic-gate * Zones note: 2047c478bd9Sstevel@tonic-gate * Walking IREs within a given zone also walks certain ires in other 2057c478bd9Sstevel@tonic-gate * zones. This is done intentionally. IRE walks with a specified 2067c478bd9Sstevel@tonic-gate * zoneid are used only when doing informational reports, and 2077c478bd9Sstevel@tonic-gate * zone users want to see things that they can access. See block 2087c478bd9Sstevel@tonic-gate * comment in ire_walk_ill_match(). 2097c478bd9Sstevel@tonic-gate */ 2107c478bd9Sstevel@tonic-gate 2117c478bd9Sstevel@tonic-gate /* 2127c478bd9Sstevel@tonic-gate * The size of the forwarding table. We will make sure that it is a 2137c478bd9Sstevel@tonic-gate * power of 2 in ip_ire_init(). 214f4b3ec61Sdh155122 * Setable in /etc/system 2157c478bd9Sstevel@tonic-gate */ 2167c478bd9Sstevel@tonic-gate uint32_t ip6_ftable_hash_size = IP6_FTABLE_HASH_SIZE; 2177c478bd9Sstevel@tonic-gate 2187c478bd9Sstevel@tonic-gate struct kmem_cache *ire_cache; 219bd670b35SErik Nordmark struct kmem_cache *ncec_cache; 220bd670b35SErik Nordmark struct kmem_cache *nce_cache; 221bd670b35SErik Nordmark 2227c478bd9Sstevel@tonic-gate static ire_t ire_null; 2237c478bd9Sstevel@tonic-gate 224bd670b35SErik Nordmark static ire_t *ire_add_v4(ire_t *ire); 2257c478bd9Sstevel@tonic-gate static void ire_delete_v4(ire_t *ire); 226bd670b35SErik Nordmark static void ire_dep_invalidate_children(ire_t *child); 22745916cd2Sjpk static void ire_walk_ipvers(pfv_t func, void *arg, uchar_t vers, 228f4b3ec61Sdh155122 zoneid_t zoneid, ip_stack_t *); 2297c478bd9Sstevel@tonic-gate static void ire_walk_ill_ipvers(uint_t match_flags, uint_t ire_type, 23045916cd2Sjpk pfv_t func, void *arg, uchar_t vers, ill_t *ill); 2316a8288c7Scarlsonj #ifdef DEBUG 2326a8288c7Scarlsonj static void ire_trace_cleanup(const ire_t *); 2337c478bd9Sstevel@tonic-gate #endif 234f1c454b4SSowmini Varadhan static void ire_dep_incr_generation_locked(ire_t *); 2357c478bd9Sstevel@tonic-gate 2367c478bd9Sstevel@tonic-gate /* 237bd670b35SErik Nordmark * Following are the functions to increment/decrement the reference 238bd670b35SErik Nordmark * count of the IREs and IRBs (ire bucket). 239bd670b35SErik Nordmark * 240bd670b35SErik Nordmark * 1) We bump up the reference count of an IRE to make sure that 241bd670b35SErik Nordmark * it does not get deleted and freed while we are using it. 242bd670b35SErik Nordmark * Typically all the lookup functions hold the bucket lock, 243bd670b35SErik Nordmark * and look for the IRE. If it finds an IRE, it bumps up the 244bd670b35SErik Nordmark * reference count before dropping the lock. Sometimes we *may* want 245bd670b35SErik Nordmark * to bump up the reference count after we *looked* up i.e without 246bd670b35SErik Nordmark * holding the bucket lock. So, the ire_refhold function does not assert 247bd670b35SErik Nordmark * on the bucket lock being held. Any thread trying to delete from 248bd670b35SErik Nordmark * the hash bucket can still do so but cannot free the IRE if 249bd670b35SErik Nordmark * ire_refcnt is not 0. 250bd670b35SErik Nordmark * 251bd670b35SErik Nordmark * 2) We bump up the reference count on the bucket where the IRE resides 252bd670b35SErik Nordmark * (IRB), when we want to prevent the IREs getting deleted from a given 253bd670b35SErik Nordmark * hash bucket. This makes life easier for ire_walk type functions which 254bd670b35SErik Nordmark * wants to walk the IRE list, call a function, but needs to drop 255bd670b35SErik Nordmark * the bucket lock to prevent recursive rw_enters. While the 256bd670b35SErik Nordmark * lock is dropped, the list could be changed by other threads or 257bd670b35SErik Nordmark * the same thread could end up deleting the ire or the ire pointed by 258bd670b35SErik Nordmark * ire_next. ire_refholding the ire or ire_next is not sufficient as 259bd670b35SErik Nordmark * a delete will still remove the ire from the bucket while we have 260bd670b35SErik Nordmark * dropped the lock and hence the ire_next would be NULL. Thus, we 261bd670b35SErik Nordmark * need a mechanism to prevent deletions from a given bucket. 262bd670b35SErik Nordmark * 263bd670b35SErik Nordmark * To prevent deletions, we bump up the reference count on the 264bd670b35SErik Nordmark * bucket. If the bucket is held, ire_delete just marks both 265bd670b35SErik Nordmark * the ire and irb as CONDEMNED. When the 266bd670b35SErik Nordmark * reference count on the bucket drops to zero, all the CONDEMNED ires 267bd670b35SErik Nordmark * are deleted. We don't have to bump up the reference count on the 268bd670b35SErik Nordmark * bucket if we are walking the bucket and never have to drop the bucket 269bd670b35SErik Nordmark * lock. Note that irb_refhold does not prevent addition of new ires 270bd670b35SErik Nordmark * in the list. It is okay because addition of new ires will not cause 271bd670b35SErik Nordmark * ire_next to point to freed memory. We do irb_refhold only when 272bd670b35SErik Nordmark * all of the 3 conditions are true : 273bd670b35SErik Nordmark * 274bd670b35SErik Nordmark * 1) The code needs to walk the IRE bucket from start to end. 275bd670b35SErik Nordmark * 2) It may have to drop the bucket lock sometimes while doing (1) 276bd670b35SErik Nordmark * 3) It does not want any ires to be deleted meanwhile. 277bd670b35SErik Nordmark */ 278bd670b35SErik Nordmark 279bd670b35SErik Nordmark /* 280bd670b35SErik Nordmark * Bump up the reference count on the hash bucket - IRB to 281bd670b35SErik Nordmark * prevent ires from being deleted in this bucket. 282bd670b35SErik Nordmark */ 283bd670b35SErik Nordmark void 284bd670b35SErik Nordmark irb_refhold(irb_t *irb) 285bd670b35SErik Nordmark { 286bd670b35SErik Nordmark rw_enter(&irb->irb_lock, RW_WRITER); 287bd670b35SErik Nordmark irb->irb_refcnt++; 288bd670b35SErik Nordmark ASSERT(irb->irb_refcnt != 0); 289bd670b35SErik Nordmark rw_exit(&irb->irb_lock); 290bd670b35SErik Nordmark } 291bd670b35SErik Nordmark 292bd670b35SErik Nordmark void 293bd670b35SErik Nordmark irb_refhold_locked(irb_t *irb) 294bd670b35SErik Nordmark { 295bd670b35SErik Nordmark ASSERT(RW_WRITE_HELD(&irb->irb_lock)); 296bd670b35SErik Nordmark irb->irb_refcnt++; 297bd670b35SErik Nordmark ASSERT(irb->irb_refcnt != 0); 298bd670b35SErik Nordmark } 299bd670b35SErik Nordmark 300bd670b35SErik Nordmark /* 301bd670b35SErik Nordmark * Note: when IRB_MARK_DYNAMIC is not set the irb_t 302bd670b35SErik Nordmark * is statically allocated, so that when the irb_refcnt goes to 0, 303bd670b35SErik Nordmark * we simply clean up the ire list and continue. 304bd670b35SErik Nordmark */ 305bd670b35SErik Nordmark void 306bd670b35SErik Nordmark irb_refrele(irb_t *irb) 307bd670b35SErik Nordmark { 308bd670b35SErik Nordmark if (irb->irb_marks & IRB_MARK_DYNAMIC) { 309bd670b35SErik Nordmark irb_refrele_ftable(irb); 310bd670b35SErik Nordmark } else { 311bd670b35SErik Nordmark rw_enter(&irb->irb_lock, RW_WRITER); 312bd670b35SErik Nordmark ASSERT(irb->irb_refcnt != 0); 313bd670b35SErik Nordmark if (--irb->irb_refcnt == 0 && 314bd670b35SErik Nordmark (irb->irb_marks & IRB_MARK_CONDEMNED)) { 315bd670b35SErik Nordmark ire_t *ire_list; 316bd670b35SErik Nordmark 317bd670b35SErik Nordmark ire_list = ire_unlink(irb); 318bd670b35SErik Nordmark rw_exit(&irb->irb_lock); 319bd670b35SErik Nordmark ASSERT(ire_list != NULL); 320bd670b35SErik Nordmark ire_cleanup(ire_list); 321bd670b35SErik Nordmark } else { 322bd670b35SErik Nordmark rw_exit(&irb->irb_lock); 323bd670b35SErik Nordmark } 324bd670b35SErik Nordmark } 325bd670b35SErik Nordmark } 326bd670b35SErik Nordmark 327bd670b35SErik Nordmark 328bd670b35SErik Nordmark /* 329bd670b35SErik Nordmark * Bump up the reference count on the IRE. We cannot assert that the 330bd670b35SErik Nordmark * bucket lock is being held as it is legal to bump up the reference 331bd670b35SErik Nordmark * count after the first lookup has returned the IRE without 332bd670b35SErik Nordmark * holding the lock. 333bd670b35SErik Nordmark */ 334bd670b35SErik Nordmark void 335bd670b35SErik Nordmark ire_refhold(ire_t *ire) 336bd670b35SErik Nordmark { 337bd670b35SErik Nordmark atomic_add_32(&(ire)->ire_refcnt, 1); 338bd670b35SErik Nordmark ASSERT((ire)->ire_refcnt != 0); 339bd670b35SErik Nordmark #ifdef DEBUG 340bd670b35SErik Nordmark ire_trace_ref(ire); 341bd670b35SErik Nordmark #endif 342bd670b35SErik Nordmark } 343bd670b35SErik Nordmark 344bd670b35SErik Nordmark void 345bd670b35SErik Nordmark ire_refhold_notr(ire_t *ire) 346bd670b35SErik Nordmark { 347bd670b35SErik Nordmark atomic_add_32(&(ire)->ire_refcnt, 1); 348bd670b35SErik Nordmark ASSERT((ire)->ire_refcnt != 0); 349bd670b35SErik Nordmark } 350bd670b35SErik Nordmark 351bd670b35SErik Nordmark void 352bd670b35SErik Nordmark ire_refhold_locked(ire_t *ire) 353bd670b35SErik Nordmark { 354bd670b35SErik Nordmark #ifdef DEBUG 355bd670b35SErik Nordmark ire_trace_ref(ire); 356bd670b35SErik Nordmark #endif 357bd670b35SErik Nordmark ire->ire_refcnt++; 358bd670b35SErik Nordmark } 359bd670b35SErik Nordmark 360bd670b35SErik Nordmark /* 361bd670b35SErik Nordmark * Release a ref on an IRE. 3627c478bd9Sstevel@tonic-gate * 3637c478bd9Sstevel@tonic-gate * Must not be called while holding any locks. Otherwise if this is 3647c478bd9Sstevel@tonic-gate * the last reference to be released there is a chance of recursive mutex 3657c478bd9Sstevel@tonic-gate * panic due to ire_refrele -> ipif_ill_refrele_tail -> qwriter_ip trying 3667c478bd9Sstevel@tonic-gate * to restart an ioctl. The one exception is when the caller is sure that 3677c478bd9Sstevel@tonic-gate * this is not the last reference to be released. Eg. if the caller is 3687c478bd9Sstevel@tonic-gate * sure that the ire has not been deleted and won't be deleted. 369bd670b35SErik Nordmark * 370bd670b35SErik Nordmark * In architectures e.g sun4u, where atomic_add_32_nv is just 371bd670b35SErik Nordmark * a cas, we need to maintain the right memory barrier semantics 372bd670b35SErik Nordmark * as that of mutex_exit i.e all the loads and stores should complete 373bd670b35SErik Nordmark * before the cas is executed. membar_exit() does that here. 3747c478bd9Sstevel@tonic-gate */ 3757c478bd9Sstevel@tonic-gate void 3767c478bd9Sstevel@tonic-gate ire_refrele(ire_t *ire) 3777c478bd9Sstevel@tonic-gate { 378bd670b35SErik Nordmark #ifdef DEBUG 379bd670b35SErik Nordmark ire_untrace_ref(ire); 380bd670b35SErik Nordmark #endif 381bd670b35SErik Nordmark ASSERT((ire)->ire_refcnt != 0); 382bd670b35SErik Nordmark membar_exit(); 383bd670b35SErik Nordmark if (atomic_add_32_nv(&(ire)->ire_refcnt, -1) == 0) 384bd670b35SErik Nordmark ire_inactive(ire); 3857c478bd9Sstevel@tonic-gate } 3867c478bd9Sstevel@tonic-gate 3877c478bd9Sstevel@tonic-gate void 3887c478bd9Sstevel@tonic-gate ire_refrele_notr(ire_t *ire) 3897c478bd9Sstevel@tonic-gate { 390bd670b35SErik Nordmark ASSERT((ire)->ire_refcnt != 0); 391bd670b35SErik Nordmark membar_exit(); 392bd670b35SErik Nordmark if (atomic_add_32_nv(&(ire)->ire_refcnt, -1) == 0) 393bd670b35SErik Nordmark ire_inactive(ire); 3947c478bd9Sstevel@tonic-gate } 3957c478bd9Sstevel@tonic-gate 3967c478bd9Sstevel@tonic-gate /* 3977c478bd9Sstevel@tonic-gate * This function is associated with the IP_IOC_IRE_DELETE[_NO_REPLY] 398bd670b35SErik Nordmark * IOCTL[s]. The NO_REPLY form is used by TCP to tell IP that it is 399bd670b35SErik Nordmark * having problems reaching a particular destination. 400bd670b35SErik Nordmark * This will make IP consider alternate routes (e.g., when there are 401bd670b35SErik Nordmark * muliple default routes), and it will also make IP discard any (potentially) 402bd670b35SErik Nordmark * stale redirect. 403bd670b35SErik Nordmark * Management processes may want to use the version that generates a reply. 4047c478bd9Sstevel@tonic-gate * 405bd670b35SErik Nordmark * With the use of NUD like behavior for IPv4/ARP in addition to IPv6 406bd670b35SErik Nordmark * this function shouldn't be necessary for IP to recover from a bad redirect, 407bd670b35SErik Nordmark * a bad default router (when there are multiple default routers), or 408bd670b35SErik Nordmark * a stale ND/ARP entry. But we retain it in any case. 409bd670b35SErik Nordmark * For instance, this is helpful when TCP suspects a failure before NUD does. 4107c478bd9Sstevel@tonic-gate */ 4117c478bd9Sstevel@tonic-gate int 4127c478bd9Sstevel@tonic-gate ip_ire_delete(queue_t *q, mblk_t *mp, cred_t *ioc_cr) 4137c478bd9Sstevel@tonic-gate { 4147c478bd9Sstevel@tonic-gate uchar_t *addr_ucp; 415bd670b35SErik Nordmark uint_t ipversion; 416bd670b35SErik Nordmark sin_t *sin; 417bd670b35SErik Nordmark sin6_t *sin6; 418bd670b35SErik Nordmark ipaddr_t v4addr; 419bd670b35SErik Nordmark in6_addr_t v6addr; 4207c478bd9Sstevel@tonic-gate ire_t *ire; 4217c478bd9Sstevel@tonic-gate ipid_t *ipid; 4227c478bd9Sstevel@tonic-gate zoneid_t zoneid; 423f4b3ec61Sdh155122 ip_stack_t *ipst; 4247c478bd9Sstevel@tonic-gate 4257c478bd9Sstevel@tonic-gate ASSERT(q->q_next == NULL); 426bd670b35SErik Nordmark zoneid = IPCL_ZONEID(Q_TO_CONN(q)); 427f4b3ec61Sdh155122 ipst = CONNQ_TO_IPST(q); 4287c478bd9Sstevel@tonic-gate 4297c478bd9Sstevel@tonic-gate /* 4307c478bd9Sstevel@tonic-gate * Check privilege using the ioctl credential; if it is NULL 4317c478bd9Sstevel@tonic-gate * then this is a kernel message and therefor privileged. 4327c478bd9Sstevel@tonic-gate */ 433f4b3ec61Sdh155122 if (ioc_cr != NULL && secpolicy_ip_config(ioc_cr, B_FALSE) != 0) 4347c478bd9Sstevel@tonic-gate return (EPERM); 4357c478bd9Sstevel@tonic-gate 4367c478bd9Sstevel@tonic-gate ipid = (ipid_t *)mp->b_rptr; 4377c478bd9Sstevel@tonic-gate 4387c478bd9Sstevel@tonic-gate addr_ucp = mi_offset_param(mp, ipid->ipid_addr_offset, 4397c478bd9Sstevel@tonic-gate ipid->ipid_addr_length); 4407c478bd9Sstevel@tonic-gate if (addr_ucp == NULL || !OK_32PTR(addr_ucp)) 4417c478bd9Sstevel@tonic-gate return (EINVAL); 4427c478bd9Sstevel@tonic-gate switch (ipid->ipid_addr_length) { 443bd670b35SErik Nordmark case sizeof (sin_t): 4447c478bd9Sstevel@tonic-gate /* 4457c478bd9Sstevel@tonic-gate * got complete (sockaddr) address - increment addr_ucp to point 4467c478bd9Sstevel@tonic-gate * at the ip_addr field. 4477c478bd9Sstevel@tonic-gate */ 4487c478bd9Sstevel@tonic-gate sin = (sin_t *)addr_ucp; 4497c478bd9Sstevel@tonic-gate addr_ucp = (uchar_t *)&sin->sin_addr.s_addr; 450bd670b35SErik Nordmark ipversion = IPV4_VERSION; 4517c478bd9Sstevel@tonic-gate break; 452bd670b35SErik Nordmark case sizeof (sin6_t): 453bd670b35SErik Nordmark /* 454bd670b35SErik Nordmark * got complete (sockaddr) address - increment addr_ucp to point 455bd670b35SErik Nordmark * at the ip_addr field. 456bd670b35SErik Nordmark */ 457bd670b35SErik Nordmark sin6 = (sin6_t *)addr_ucp; 458bd670b35SErik Nordmark addr_ucp = (uchar_t *)&sin6->sin6_addr; 459bd670b35SErik Nordmark ipversion = IPV6_VERSION; 460bd670b35SErik Nordmark break; 4617c478bd9Sstevel@tonic-gate default: 4627c478bd9Sstevel@tonic-gate return (EINVAL); 4637c478bd9Sstevel@tonic-gate } 464bd670b35SErik Nordmark if (ipversion == IPV4_VERSION) { 4657c478bd9Sstevel@tonic-gate /* Extract the destination address. */ 466bd670b35SErik Nordmark bcopy(addr_ucp, &v4addr, IP_ADDR_LEN); 4677c478bd9Sstevel@tonic-gate 468bd670b35SErik Nordmark ire = ire_ftable_lookup_v4(v4addr, 0, 0, 0, NULL, 469bd670b35SErik Nordmark zoneid, NULL, MATCH_IRE_DSTONLY, 0, ipst, NULL); 470bd670b35SErik Nordmark } else { 471bd670b35SErik Nordmark /* Extract the destination address. */ 472bd670b35SErik Nordmark bcopy(addr_ucp, &v6addr, IPV6_ADDR_LEN); 4737c478bd9Sstevel@tonic-gate 474bd670b35SErik Nordmark ire = ire_ftable_lookup_v6(&v6addr, NULL, NULL, 0, NULL, 475bd670b35SErik Nordmark zoneid, NULL, MATCH_IRE_DSTONLY, 0, ipst, NULL); 4767c478bd9Sstevel@tonic-gate } 4776bdb8e66Sdd193516 if (ire != NULL) { 478bd670b35SErik Nordmark if (ipversion == IPV4_VERSION) { 4797c478bd9Sstevel@tonic-gate ip_rts_change(RTM_LOSING, ire->ire_addr, 4807c478bd9Sstevel@tonic-gate ire->ire_gateway_addr, ire->ire_mask, 481bd670b35SErik Nordmark (Q_TO_CONN(q))->conn_laddr_v4, 0, 0, 0, 482bd670b35SErik Nordmark (RTA_DST | RTA_GATEWAY | RTA_NETMASK | RTA_IFA), 483bd670b35SErik Nordmark ire->ire_ipst); 4847c478bd9Sstevel@tonic-gate } 485bd670b35SErik Nordmark (void) ire_no_good(ire); 4867c478bd9Sstevel@tonic-gate ire_refrele(ire); 4877c478bd9Sstevel@tonic-gate } 4887c478bd9Sstevel@tonic-gate return (0); 4897c478bd9Sstevel@tonic-gate } 4907c478bd9Sstevel@tonic-gate 4917c478bd9Sstevel@tonic-gate /* 4927c478bd9Sstevel@tonic-gate * Initialize the ire that is specific to IPv4 part and call 4937c478bd9Sstevel@tonic-gate * ire_init_common to finish it. 494bd670b35SErik Nordmark * Returns zero or errno. 4957c478bd9Sstevel@tonic-gate */ 496bd670b35SErik Nordmark int 497bd670b35SErik Nordmark ire_init_v4(ire_t *ire, uchar_t *addr, uchar_t *mask, uchar_t *gateway, 498bd670b35SErik Nordmark ushort_t type, ill_t *ill, zoneid_t zoneid, uint_t flags, 499bd670b35SErik Nordmark tsol_gc_t *gc, ip_stack_t *ipst) 5007c478bd9Sstevel@tonic-gate { 501bd670b35SErik Nordmark int error; 502bd670b35SErik Nordmark 50345916cd2Sjpk /* 50445916cd2Sjpk * Reject IRE security attribute creation/initialization 50545916cd2Sjpk * if system is not running in Trusted mode. 50645916cd2Sjpk */ 507bd670b35SErik Nordmark if (gc != NULL && !is_system_labeled()) 508bd670b35SErik Nordmark return (EINVAL); 50945916cd2Sjpk 510f4b3ec61Sdh155122 BUMP_IRE_STATS(ipst->ips_ire_stats_v4, ire_stats_alloced); 5117c478bd9Sstevel@tonic-gate 5127c478bd9Sstevel@tonic-gate if (addr != NULL) 5137c478bd9Sstevel@tonic-gate bcopy(addr, &ire->ire_addr, IP_ADDR_LEN); 514bd670b35SErik Nordmark if (gateway != NULL) 515bd670b35SErik Nordmark bcopy(gateway, &ire->ire_gateway_addr, IP_ADDR_LEN); 516bd670b35SErik Nordmark 517bd670b35SErik Nordmark /* Make sure we don't have stray values in some fields */ 518bd670b35SErik Nordmark switch (type) { 519bd670b35SErik Nordmark case IRE_LOOPBACK: 520bd670b35SErik Nordmark case IRE_HOST: 521bd670b35SErik Nordmark case IRE_BROADCAST: 522bd670b35SErik Nordmark case IRE_LOCAL: 523bd670b35SErik Nordmark case IRE_IF_CLONE: 524bd670b35SErik Nordmark ire->ire_mask = IP_HOST_MASK; 525bd670b35SErik Nordmark ire->ire_masklen = IPV4_ABITS; 526bd670b35SErik Nordmark break; 527bd670b35SErik Nordmark case IRE_PREFIX: 528bd670b35SErik Nordmark case IRE_DEFAULT: 529bd670b35SErik Nordmark case IRE_IF_RESOLVER: 530bd670b35SErik Nordmark case IRE_IF_NORESOLVER: 5317c478bd9Sstevel@tonic-gate if (mask != NULL) { 5327c478bd9Sstevel@tonic-gate bcopy(mask, &ire->ire_mask, IP_ADDR_LEN); 5337c478bd9Sstevel@tonic-gate ire->ire_masklen = ip_mask_to_plen(ire->ire_mask); 5347c478bd9Sstevel@tonic-gate } 535bd670b35SErik Nordmark break; 536bd670b35SErik Nordmark case IRE_MULTICAST: 537bd670b35SErik Nordmark case IRE_NOROUTE: 538bd670b35SErik Nordmark ASSERT(mask == NULL); 539bd670b35SErik Nordmark break; 540bd670b35SErik Nordmark default: 541bd670b35SErik Nordmark ASSERT(0); 542bd670b35SErik Nordmark return (EINVAL); 5437c478bd9Sstevel@tonic-gate } 5447c478bd9Sstevel@tonic-gate 545bd670b35SErik Nordmark error = ire_init_common(ire, type, ill, zoneid, flags, IPV4_VERSION, 546bd670b35SErik Nordmark gc, ipst); 547bd670b35SErik Nordmark if (error != NULL) 548bd670b35SErik Nordmark return (error); 5497c478bd9Sstevel@tonic-gate 550bd670b35SErik Nordmark /* Determine which function pointers to use */ 551bd670b35SErik Nordmark ire->ire_postfragfn = ip_xmit; /* Common case */ 5527c478bd9Sstevel@tonic-gate 553bd670b35SErik Nordmark switch (ire->ire_type) { 554bd670b35SErik Nordmark case IRE_LOCAL: 555bd670b35SErik Nordmark ire->ire_sendfn = ire_send_local_v4; 556bd670b35SErik Nordmark ire->ire_recvfn = ire_recv_local_v4; 557bd670b35SErik Nordmark ASSERT(ire->ire_ill != NULL); 5581cb875aeSCathy Zhou if (ire->ire_ill->ill_flags & ILLF_NOACCEPT) 559bd670b35SErik Nordmark ire->ire_recvfn = ire_recv_noaccept_v6; 560bd670b35SErik Nordmark break; 561bd670b35SErik Nordmark case IRE_LOOPBACK: 562bd670b35SErik Nordmark ire->ire_sendfn = ire_send_local_v4; 563bd670b35SErik Nordmark ire->ire_recvfn = ire_recv_loopback_v4; 564bd670b35SErik Nordmark break; 565bd670b35SErik Nordmark case IRE_BROADCAST: 566bd670b35SErik Nordmark ire->ire_postfragfn = ip_postfrag_loopcheck; 567bd670b35SErik Nordmark ire->ire_sendfn = ire_send_broadcast_v4; 568bd670b35SErik Nordmark ire->ire_recvfn = ire_recv_broadcast_v4; 569bd670b35SErik Nordmark break; 570bd670b35SErik Nordmark case IRE_MULTICAST: 571bd670b35SErik Nordmark ire->ire_postfragfn = ip_postfrag_loopcheck; 572bd670b35SErik Nordmark ire->ire_sendfn = ire_send_multicast_v4; 573bd670b35SErik Nordmark ire->ire_recvfn = ire_recv_multicast_v4; 574bd670b35SErik Nordmark break; 575bd670b35SErik Nordmark default: 576bd670b35SErik Nordmark /* 577bd670b35SErik Nordmark * For IRE_IF_ALL and IRE_OFFLINK we forward received 578bd670b35SErik Nordmark * packets by default. 579bd670b35SErik Nordmark */ 580bd670b35SErik Nordmark ire->ire_sendfn = ire_send_wire_v4; 581bd670b35SErik Nordmark ire->ire_recvfn = ire_recv_forward_v4; 582bd670b35SErik Nordmark break; 583bd670b35SErik Nordmark } 584bd670b35SErik Nordmark if (ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) { 585bd670b35SErik Nordmark ire->ire_sendfn = ire_send_noroute_v4; 586bd670b35SErik Nordmark ire->ire_recvfn = ire_recv_noroute_v4; 587bd670b35SErik Nordmark } else if (ire->ire_flags & RTF_MULTIRT) { 588bd670b35SErik Nordmark ire->ire_postfragfn = ip_postfrag_multirt_v4; 589bd670b35SErik Nordmark ire->ire_sendfn = ire_send_multirt_v4; 590bd670b35SErik Nordmark /* Multirt receive of broadcast uses ire_recv_broadcast_v4 */ 591bd670b35SErik Nordmark if (ire->ire_type != IRE_BROADCAST) 592bd670b35SErik Nordmark ire->ire_recvfn = ire_recv_multirt_v4; 593bd670b35SErik Nordmark } 594bd670b35SErik Nordmark ire->ire_nce_capable = ire_determine_nce_capable(ire); 595bd670b35SErik Nordmark return (0); 5967c478bd9Sstevel@tonic-gate } 5977c478bd9Sstevel@tonic-gate 5987c478bd9Sstevel@tonic-gate /* 599bd670b35SErik Nordmark * Determine ire_nce_capable 6007c478bd9Sstevel@tonic-gate */ 601bd670b35SErik Nordmark boolean_t 602bd670b35SErik Nordmark ire_determine_nce_capable(ire_t *ire) 6037c478bd9Sstevel@tonic-gate { 604bd670b35SErik Nordmark int max_masklen; 6057c478bd9Sstevel@tonic-gate 606bd670b35SErik Nordmark if ((ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) || 607bd670b35SErik Nordmark (ire->ire_type & IRE_MULTICAST)) 608bd670b35SErik Nordmark return (B_TRUE); 6097c478bd9Sstevel@tonic-gate 610bd670b35SErik Nordmark if (ire->ire_ipversion == IPV4_VERSION) 611bd670b35SErik Nordmark max_masklen = IPV4_ABITS; 612bd670b35SErik Nordmark else 613bd670b35SErik Nordmark max_masklen = IPV6_ABITS; 6147c478bd9Sstevel@tonic-gate 615bd670b35SErik Nordmark if ((ire->ire_type & IRE_ONLINK) && ire->ire_masklen == max_masklen) 616bd670b35SErik Nordmark return (B_TRUE); 617bd670b35SErik Nordmark return (B_FALSE); 6187c478bd9Sstevel@tonic-gate } 6197c478bd9Sstevel@tonic-gate 6207c478bd9Sstevel@tonic-gate /* 6217c478bd9Sstevel@tonic-gate * ire_create is called to allocate and initialize a new IRE. 6227c478bd9Sstevel@tonic-gate * 6237c478bd9Sstevel@tonic-gate * NOTE : This is called as writer sometimes though not required 6247c478bd9Sstevel@tonic-gate * by this function. 6257c478bd9Sstevel@tonic-gate */ 6267c478bd9Sstevel@tonic-gate ire_t * 627bd670b35SErik Nordmark ire_create(uchar_t *addr, uchar_t *mask, uchar_t *gateway, 628bd670b35SErik Nordmark ushort_t type, ill_t *ill, zoneid_t zoneid, uint_t flags, tsol_gc_t *gc, 629bd670b35SErik Nordmark ip_stack_t *ipst) 6307c478bd9Sstevel@tonic-gate { 6317c478bd9Sstevel@tonic-gate ire_t *ire; 632bd670b35SErik Nordmark int error; 6337c478bd9Sstevel@tonic-gate 6347c478bd9Sstevel@tonic-gate ire = kmem_cache_alloc(ire_cache, KM_NOSLEEP); 6357c478bd9Sstevel@tonic-gate if (ire == NULL) { 636bd670b35SErik Nordmark DTRACE_PROBE(kmem__cache__alloc); 6377c478bd9Sstevel@tonic-gate return (NULL); 6387c478bd9Sstevel@tonic-gate } 6397c478bd9Sstevel@tonic-gate *ire = ire_null; 6407c478bd9Sstevel@tonic-gate 641bd670b35SErik Nordmark error = ire_init_v4(ire, addr, mask, gateway, type, ill, zoneid, flags, 642bd670b35SErik Nordmark gc, ipst); 643bd670b35SErik Nordmark if (error != 0) { 644bd670b35SErik Nordmark DTRACE_PROBE2(ire__init, ire_t *, ire, int, error); 6457c478bd9Sstevel@tonic-gate kmem_cache_free(ire_cache, ire); 6467c478bd9Sstevel@tonic-gate return (NULL); 6477c478bd9Sstevel@tonic-gate } 6487c478bd9Sstevel@tonic-gate return (ire); 6497c478bd9Sstevel@tonic-gate } 6507c478bd9Sstevel@tonic-gate 6517c478bd9Sstevel@tonic-gate /* 6527c478bd9Sstevel@tonic-gate * Common to IPv4 and IPv6 653bd670b35SErik Nordmark * Returns zero or errno. 6547c478bd9Sstevel@tonic-gate */ 655bd670b35SErik Nordmark int 656bd670b35SErik Nordmark ire_init_common(ire_t *ire, ushort_t type, ill_t *ill, zoneid_t zoneid, 657bd670b35SErik Nordmark uint_t flags, uchar_t ipversion, tsol_gc_t *gc, ip_stack_t *ipst) 6587c478bd9Sstevel@tonic-gate { 659bd670b35SErik Nordmark int error; 6607c478bd9Sstevel@tonic-gate 66145916cd2Sjpk #ifdef DEBUG 662bd670b35SErik Nordmark if (ill != NULL) { 663bd670b35SErik Nordmark if (ill->ill_isv6) 6647c478bd9Sstevel@tonic-gate ASSERT(ipversion == IPV6_VERSION); 6657c478bd9Sstevel@tonic-gate else 6667c478bd9Sstevel@tonic-gate ASSERT(ipversion == IPV4_VERSION); 6677c478bd9Sstevel@tonic-gate } 66845916cd2Sjpk #endif /* DEBUG */ 66945916cd2Sjpk 67045916cd2Sjpk /* 67145916cd2Sjpk * Create/initialize IRE security attribute only in Trusted mode; 672bd670b35SErik Nordmark * if the passed in gc is non-NULL, we expect that the caller 67345916cd2Sjpk * has held a reference to it and will release it when this routine 67445916cd2Sjpk * returns a failure, otherwise we own the reference. We do this 67545916cd2Sjpk * prior to initializing the rest IRE fields. 67645916cd2Sjpk */ 67745916cd2Sjpk if (is_system_labeled()) { 67845916cd2Sjpk if ((type & (IRE_LOCAL | IRE_LOOPBACK | IRE_BROADCAST | 679bd670b35SErik Nordmark IRE_IF_ALL | IRE_MULTICAST | IRE_NOROUTE)) != 0) { 68045916cd2Sjpk /* release references on behalf of caller */ 68145916cd2Sjpk if (gc != NULL) 68245916cd2Sjpk GC_REFRELE(gc); 683bd670b35SErik Nordmark } else { 684bd670b35SErik Nordmark error = tsol_ire_init_gwattr(ire, ipversion, gc); 685bd670b35SErik Nordmark if (error != 0) 686bd670b35SErik Nordmark return (error); 68745916cd2Sjpk } 68845916cd2Sjpk } 6897c478bd9Sstevel@tonic-gate 6907c478bd9Sstevel@tonic-gate ire->ire_type = type; 6917c478bd9Sstevel@tonic-gate ire->ire_flags = RTF_UP | flags; 6927c478bd9Sstevel@tonic-gate ire->ire_create_time = (uint32_t)gethrestime_sec(); 693bd670b35SErik Nordmark ire->ire_generation = IRE_GENERATION_INITIAL; 6947c478bd9Sstevel@tonic-gate 6957c478bd9Sstevel@tonic-gate /* 696bd670b35SErik Nordmark * The ill_ire_cnt isn't increased until 697bd670b35SErik Nordmark * the IRE is added to ensure that a walker will find 698bd670b35SErik Nordmark * all IREs that hold a reference on an ill. 6997c478bd9Sstevel@tonic-gate * 700bd670b35SErik Nordmark * Note that ill_ire_multicast doesn't hold a ref on the ill since 701bd670b35SErik Nordmark * ire_add() is not called for the IRE_MULTICAST. 7027c478bd9Sstevel@tonic-gate */ 703bd670b35SErik Nordmark ire->ire_ill = ill; 704bd670b35SErik Nordmark ire->ire_zoneid = zoneid; 7057c478bd9Sstevel@tonic-gate ire->ire_ipversion = ipversion; 706bd670b35SErik Nordmark 7077c478bd9Sstevel@tonic-gate mutex_init(&ire->ire_lock, NULL, MUTEX_DEFAULT, NULL); 708c793af95Ssangeeta ire->ire_refcnt = 1; 709bd670b35SErik Nordmark ire->ire_identical_ref = 1; /* Number of ire_delete's needed */ 710f4b3ec61Sdh155122 ire->ire_ipst = ipst; /* No netstack_hold */ 7116a8288c7Scarlsonj ire->ire_trace_disable = B_FALSE; 71245916cd2Sjpk 713bd670b35SErik Nordmark return (0); 7147c478bd9Sstevel@tonic-gate } 7157c478bd9Sstevel@tonic-gate 7167c478bd9Sstevel@tonic-gate /* 717bd670b35SErik Nordmark * This creates an IRE_BROADCAST based on the arguments. 718bd670b35SErik Nordmark * A mirror is ire_lookup_bcast(). 7197c478bd9Sstevel@tonic-gate * 720bd670b35SErik Nordmark * Any supression of unneeded ones is done in ire_add_v4. 721bd670b35SErik Nordmark * We add one IRE_BROADCAST per address. ire_send_broadcast_v4() 722bd670b35SErik Nordmark * takes care of generating a loopback copy of the packet. 7237c478bd9Sstevel@tonic-gate */ 7247c478bd9Sstevel@tonic-gate ire_t ** 725bd670b35SErik Nordmark ire_create_bcast(ill_t *ill, ipaddr_t addr, zoneid_t zoneid, ire_t **irep) 7267c478bd9Sstevel@tonic-gate { 727e11c3f44Smeem ip_stack_t *ipst = ill->ill_ipst; 7287c478bd9Sstevel@tonic-gate 729bd670b35SErik Nordmark ASSERT(IAM_WRITER_ILL(ill)); 730f4b3ec61Sdh155122 7317c478bd9Sstevel@tonic-gate *irep++ = ire_create( 7327c478bd9Sstevel@tonic-gate (uchar_t *)&addr, /* dest addr */ 7337c478bd9Sstevel@tonic-gate (uchar_t *)&ip_g_all_ones, /* mask */ 7347c478bd9Sstevel@tonic-gate NULL, /* no gateway */ 7357c478bd9Sstevel@tonic-gate IRE_BROADCAST, 736bd670b35SErik Nordmark ill, 737bd670b35SErik Nordmark zoneid, 738bd670b35SErik Nordmark RTF_KERNEL, 739f4b3ec61Sdh155122 NULL, 740f4b3ec61Sdh155122 ipst); 7417c478bd9Sstevel@tonic-gate 7427c478bd9Sstevel@tonic-gate return (irep); 7437c478bd9Sstevel@tonic-gate } 7447c478bd9Sstevel@tonic-gate 7457c478bd9Sstevel@tonic-gate /* 746bd670b35SErik Nordmark * This looks up an IRE_BROADCAST based on the arguments. 747bd670b35SErik Nordmark * Mirrors ire_create_bcast(). 7487c478bd9Sstevel@tonic-gate */ 7497c478bd9Sstevel@tonic-gate ire_t * 750bd670b35SErik Nordmark ire_lookup_bcast(ill_t *ill, ipaddr_t addr, zoneid_t zoneid) 7517c478bd9Sstevel@tonic-gate { 7527c478bd9Sstevel@tonic-gate ire_t *ire; 753bd670b35SErik Nordmark int match_args; 7547c478bd9Sstevel@tonic-gate 755bd670b35SErik Nordmark match_args = MATCH_IRE_TYPE | MATCH_IRE_ILL | MATCH_IRE_GW | 756bd670b35SErik Nordmark MATCH_IRE_MASK | MATCH_IRE_ZONEONLY; 757bd670b35SErik Nordmark 758bd670b35SErik Nordmark if (IS_UNDER_IPMP(ill)) 759bd670b35SErik Nordmark match_args |= MATCH_IRE_TESTHIDDEN; 760bd670b35SErik Nordmark 761bd670b35SErik Nordmark ire = ire_ftable_lookup_v4( 762bd670b35SErik Nordmark addr, /* dest addr */ 763bd670b35SErik Nordmark ip_g_all_ones, /* mask */ 764bd670b35SErik Nordmark 0, /* no gateway */ 765bd670b35SErik Nordmark IRE_BROADCAST, 766bd670b35SErik Nordmark ill, 767bd670b35SErik Nordmark zoneid, 768bd670b35SErik Nordmark NULL, 769bd670b35SErik Nordmark match_args, 770bd670b35SErik Nordmark 0, 771bd670b35SErik Nordmark ill->ill_ipst, 772bd670b35SErik Nordmark NULL); 7737c478bd9Sstevel@tonic-gate return (ire); 7747c478bd9Sstevel@tonic-gate } 7757c478bd9Sstevel@tonic-gate 7767c478bd9Sstevel@tonic-gate /* Arrange to call the specified function for every IRE in the world. */ 7777c478bd9Sstevel@tonic-gate void 778f4b3ec61Sdh155122 ire_walk(pfv_t func, void *arg, ip_stack_t *ipst) 7797c478bd9Sstevel@tonic-gate { 780f4b3ec61Sdh155122 ire_walk_ipvers(func, arg, 0, ALL_ZONES, ipst); 7817c478bd9Sstevel@tonic-gate } 7827c478bd9Sstevel@tonic-gate 7837c478bd9Sstevel@tonic-gate void 784f4b3ec61Sdh155122 ire_walk_v4(pfv_t func, void *arg, zoneid_t zoneid, ip_stack_t *ipst) 7857c478bd9Sstevel@tonic-gate { 786f4b3ec61Sdh155122 ire_walk_ipvers(func, arg, IPV4_VERSION, zoneid, ipst); 7877c478bd9Sstevel@tonic-gate } 7887c478bd9Sstevel@tonic-gate 7897c478bd9Sstevel@tonic-gate void 790f4b3ec61Sdh155122 ire_walk_v6(pfv_t func, void *arg, zoneid_t zoneid, ip_stack_t *ipst) 7917c478bd9Sstevel@tonic-gate { 792f4b3ec61Sdh155122 ire_walk_ipvers(func, arg, IPV6_VERSION, zoneid, ipst); 7937c478bd9Sstevel@tonic-gate } 7947c478bd9Sstevel@tonic-gate 7957c478bd9Sstevel@tonic-gate /* 7967c478bd9Sstevel@tonic-gate * Walk a particular version. version == 0 means both v4 and v6. 7977c478bd9Sstevel@tonic-gate */ 7987c478bd9Sstevel@tonic-gate static void 799f4b3ec61Sdh155122 ire_walk_ipvers(pfv_t func, void *arg, uchar_t vers, zoneid_t zoneid, 800f4b3ec61Sdh155122 ip_stack_t *ipst) 8017c478bd9Sstevel@tonic-gate { 8027c478bd9Sstevel@tonic-gate if (vers != IPV6_VERSION) { 803c793af95Ssangeeta /* 804c793af95Ssangeeta * ip_forwarding_table variable doesn't matter for IPv4 since 805f4b3ec61Sdh155122 * ire_walk_ill_tables uses ips_ip_ftable for IPv4. 806c793af95Ssangeeta */ 8077c478bd9Sstevel@tonic-gate ire_walk_ill_tables(0, 0, func, arg, IP_MASK_TABLE_SIZE, 808c793af95Ssangeeta 0, NULL, 809f4b3ec61Sdh155122 NULL, zoneid, ipst); 8107c478bd9Sstevel@tonic-gate } 8117c478bd9Sstevel@tonic-gate if (vers != IPV4_VERSION) { 8127c478bd9Sstevel@tonic-gate ire_walk_ill_tables(0, 0, func, arg, IP6_MASK_TABLE_SIZE, 813f4b3ec61Sdh155122 ipst->ips_ip6_ftable_hash_size, 814f4b3ec61Sdh155122 ipst->ips_ip_forwarding_table_v6, 815bd670b35SErik Nordmark NULL, zoneid, ipst); 8167c478bd9Sstevel@tonic-gate } 8177c478bd9Sstevel@tonic-gate } 8187c478bd9Sstevel@tonic-gate 8197c478bd9Sstevel@tonic-gate /* 8207924222fSmeem * Arrange to call the specified function for every IRE that matches the ill. 8217c478bd9Sstevel@tonic-gate */ 8227c478bd9Sstevel@tonic-gate void 82345916cd2Sjpk ire_walk_ill(uint_t match_flags, uint_t ire_type, pfv_t func, void *arg, 8247c478bd9Sstevel@tonic-gate ill_t *ill) 8257c478bd9Sstevel@tonic-gate { 8267924222fSmeem uchar_t vers = (ill->ill_isv6 ? IPV6_VERSION : IPV4_VERSION); 8277924222fSmeem 8287924222fSmeem ire_walk_ill_ipvers(match_flags, ire_type, func, arg, vers, ill); 8297c478bd9Sstevel@tonic-gate } 8307c478bd9Sstevel@tonic-gate 8317c478bd9Sstevel@tonic-gate /* 8327924222fSmeem * Walk a particular ill and version. 8337c478bd9Sstevel@tonic-gate */ 8347c478bd9Sstevel@tonic-gate static void 8357c478bd9Sstevel@tonic-gate ire_walk_ill_ipvers(uint_t match_flags, uint_t ire_type, pfv_t func, 83645916cd2Sjpk void *arg, uchar_t vers, ill_t *ill) 8377c478bd9Sstevel@tonic-gate { 838f4b3ec61Sdh155122 ip_stack_t *ipst = ill->ill_ipst; 839f4b3ec61Sdh155122 8407924222fSmeem if (vers == IPV4_VERSION) { 8417c478bd9Sstevel@tonic-gate ire_walk_ill_tables(match_flags, ire_type, func, arg, 842bd670b35SErik Nordmark IP_MASK_TABLE_SIZE, 843bd670b35SErik Nordmark 0, NULL, 844bd670b35SErik Nordmark ill, ALL_ZONES, ipst); 845bd670b35SErik Nordmark } 846bd670b35SErik Nordmark if (vers != IPV4_VERSION) { 8477c478bd9Sstevel@tonic-gate ire_walk_ill_tables(match_flags, ire_type, func, arg, 848f4b3ec61Sdh155122 IP6_MASK_TABLE_SIZE, ipst->ips_ip6_ftable_hash_size, 849f4b3ec61Sdh155122 ipst->ips_ip_forwarding_table_v6, 850bd670b35SErik Nordmark ill, ALL_ZONES, ipst); 8517c478bd9Sstevel@tonic-gate } 8527c478bd9Sstevel@tonic-gate } 8537c478bd9Sstevel@tonic-gate 854bd670b35SErik Nordmark /* 855bd670b35SErik Nordmark * Do the specific matching of IREs to shared-IP zones. 856bd670b35SErik Nordmark * 857bd670b35SErik Nordmark * We have the same logic as in ire_match_args but implemented slightly 858bd670b35SErik Nordmark * differently. 859bd670b35SErik Nordmark */ 860c793af95Ssangeeta boolean_t 8617c478bd9Sstevel@tonic-gate ire_walk_ill_match(uint_t match_flags, uint_t ire_type, ire_t *ire, 862f4b3ec61Sdh155122 ill_t *ill, zoneid_t zoneid, ip_stack_t *ipst) 8637c478bd9Sstevel@tonic-gate { 864188e1664SErik Nordmark ill_t *dst_ill = ire->ire_ill; 8657c478bd9Sstevel@tonic-gate 8667c478bd9Sstevel@tonic-gate ASSERT(match_flags != 0 || zoneid != ALL_ZONES); 8677c478bd9Sstevel@tonic-gate 868bd670b35SErik Nordmark if (zoneid != ALL_ZONES && zoneid != ire->ire_zoneid && 869bd670b35SErik Nordmark ire->ire_zoneid != ALL_ZONES) { 8707c478bd9Sstevel@tonic-gate /* 8717c478bd9Sstevel@tonic-gate * We're walking the IREs for a specific zone. The only relevant 8727c478bd9Sstevel@tonic-gate * IREs are: 8737c478bd9Sstevel@tonic-gate * - all IREs with a matching ire_zoneid 874bd670b35SErik Nordmark * - IRE_IF_ALL IREs for interfaces with a usable source addr 8757c478bd9Sstevel@tonic-gate * with a matching zone 876bd670b35SErik Nordmark * - IRE_OFFLINK with a gateway reachable from the zone 877bd670b35SErik Nordmark * Note that ealier we only did the IRE_OFFLINK check for 878bd670b35SErik Nordmark * IRE_DEFAULT (and only when we had multiple IRE_DEFAULTs). 8797c478bd9Sstevel@tonic-gate */ 880bd670b35SErik Nordmark if (ire->ire_type & IRE_ONLINK) { 881bd670b35SErik Nordmark uint_t ifindex; 882bd670b35SErik Nordmark 8837c478bd9Sstevel@tonic-gate /* 884bd670b35SErik Nordmark * Note there is no IRE_INTERFACE on vniN thus 885bd670b35SErik Nordmark * can't do an IRE lookup for a matching route. 8867c478bd9Sstevel@tonic-gate */ 887bd670b35SErik Nordmark ifindex = dst_ill->ill_usesrc_ifindex; 888bd670b35SErik Nordmark if (ifindex == 0) 8897c478bd9Sstevel@tonic-gate return (B_FALSE); 890bd670b35SErik Nordmark 8917c478bd9Sstevel@tonic-gate /* 892bd670b35SErik Nordmark * If there is a usable source address in the 893bd670b35SErik Nordmark * zone, then it's ok to return an 894bd670b35SErik Nordmark * IRE_INTERFACE 8957c478bd9Sstevel@tonic-gate */ 896bd670b35SErik Nordmark if (!ipif_zone_avail(ifindex, dst_ill->ill_isv6, 897bd670b35SErik Nordmark zoneid, ipst)) { 8987c478bd9Sstevel@tonic-gate return (B_FALSE); 8997c478bd9Sstevel@tonic-gate } 9007c478bd9Sstevel@tonic-gate } 901bd670b35SErik Nordmark if (dst_ill != NULL && (ire->ire_type & IRE_OFFLINK)) { 902bd670b35SErik Nordmark ipif_t *tipif; 903bd670b35SErik Nordmark 904bd670b35SErik Nordmark mutex_enter(&dst_ill->ill_lock); 905bd670b35SErik Nordmark for (tipif = dst_ill->ill_ipif; 906bd670b35SErik Nordmark tipif != NULL; tipif = tipif->ipif_next) { 907bd670b35SErik Nordmark if (!IPIF_IS_CONDEMNED(tipif) && 908bd670b35SErik Nordmark (tipif->ipif_flags & IPIF_UP) && 909bd670b35SErik Nordmark (tipif->ipif_zoneid == zoneid || 910bd670b35SErik Nordmark tipif->ipif_zoneid == ALL_ZONES)) 911bd670b35SErik Nordmark break; 912bd670b35SErik Nordmark } 913bd670b35SErik Nordmark mutex_exit(&dst_ill->ill_lock); 914bd670b35SErik Nordmark if (tipif == NULL) { 9157c478bd9Sstevel@tonic-gate return (B_FALSE); 9167c478bd9Sstevel@tonic-gate } 9177c478bd9Sstevel@tonic-gate } 918188e1664SErik Nordmark } 9197c478bd9Sstevel@tonic-gate /* 920188e1664SErik Nordmark * Except for ALL_ZONES, we only match the offlink routes 921bd670b35SErik Nordmark * where ire_gateway_addr has an IRE_INTERFACE for the zoneid. 9229e3469d3SErik Nordmark * Since we can have leftover routes after the IP addresses have 9239e3469d3SErik Nordmark * changed, the global zone will also match offlink routes where the 9249e3469d3SErik Nordmark * gateway is unreachable from any zone. 9257c478bd9Sstevel@tonic-gate */ 926188e1664SErik Nordmark if ((ire->ire_type & IRE_OFFLINK) && zoneid != ALL_ZONES) { 9277c478bd9Sstevel@tonic-gate in6_addr_t gw_addr_v6; 9289e3469d3SErik Nordmark boolean_t reach; 929e11c3f44Smeem 9307c478bd9Sstevel@tonic-gate if (ire->ire_ipversion == IPV4_VERSION) { 9319e3469d3SErik Nordmark reach = ire_gateway_ok_zone_v4(ire->ire_gateway_addr, 9329e3469d3SErik Nordmark zoneid, dst_ill, NULL, ipst, B_FALSE); 9337c478bd9Sstevel@tonic-gate } else { 9347c478bd9Sstevel@tonic-gate ASSERT(ire->ire_ipversion == IPV6_VERSION); 9357c478bd9Sstevel@tonic-gate mutex_enter(&ire->ire_lock); 9367c478bd9Sstevel@tonic-gate gw_addr_v6 = ire->ire_gateway_addr_v6; 9377c478bd9Sstevel@tonic-gate mutex_exit(&ire->ire_lock); 938bd670b35SErik Nordmark 9399e3469d3SErik Nordmark reach = ire_gateway_ok_zone_v6(&gw_addr_v6, zoneid, 9409e3469d3SErik Nordmark dst_ill, NULL, ipst, B_FALSE); 9419e3469d3SErik Nordmark } 9429e3469d3SErik Nordmark if (!reach) { 9439e3469d3SErik Nordmark if (zoneid != GLOBAL_ZONEID) 9447c478bd9Sstevel@tonic-gate return (B_FALSE); 9459e3469d3SErik Nordmark 9469e3469d3SErik Nordmark /* 9479e3469d3SErik Nordmark * Check if ALL_ZONES reachable - if not then let the 9489e3469d3SErik Nordmark * global zone see it. 9499e3469d3SErik Nordmark */ 9509e3469d3SErik Nordmark if (ire->ire_ipversion == IPV4_VERSION) { 9519e3469d3SErik Nordmark reach = ire_gateway_ok_zone_v4( 9529e3469d3SErik Nordmark ire->ire_gateway_addr, ALL_ZONES, 9539e3469d3SErik Nordmark dst_ill, NULL, ipst, B_FALSE); 9549e3469d3SErik Nordmark } else { 9559e3469d3SErik Nordmark reach = ire_gateway_ok_zone_v6(&gw_addr_v6, 9569e3469d3SErik Nordmark ALL_ZONES, dst_ill, NULL, ipst, B_FALSE); 9579e3469d3SErik Nordmark } 9589e3469d3SErik Nordmark if (reach) { 9599e3469d3SErik Nordmark /* 9609e3469d3SErik Nordmark * Some other zone could see it, hence hide it 9619e3469d3SErik Nordmark * in the global zone. 9629e3469d3SErik Nordmark */ 9639e3469d3SErik Nordmark return (B_FALSE); 9649e3469d3SErik Nordmark } 9657c478bd9Sstevel@tonic-gate } 9667c478bd9Sstevel@tonic-gate } 9677c478bd9Sstevel@tonic-gate 9687c478bd9Sstevel@tonic-gate if (((!(match_flags & MATCH_IRE_TYPE)) || 9697c478bd9Sstevel@tonic-gate (ire->ire_type & ire_type)) && 9707c478bd9Sstevel@tonic-gate ((!(match_flags & MATCH_IRE_ILL)) || 971bd670b35SErik Nordmark (dst_ill == ill || 972bd670b35SErik Nordmark dst_ill != NULL && IS_IN_SAME_ILLGRP(dst_ill, ill)))) { 9737c478bd9Sstevel@tonic-gate return (B_TRUE); 9747c478bd9Sstevel@tonic-gate } 9757c478bd9Sstevel@tonic-gate return (B_FALSE); 9767c478bd9Sstevel@tonic-gate } 9777c478bd9Sstevel@tonic-gate 978c793af95Ssangeeta int 979c793af95Ssangeeta rtfunc(struct radix_node *rn, void *arg) 980c793af95Ssangeeta { 981c793af95Ssangeeta struct rtfuncarg *rtf = arg; 982c793af95Ssangeeta struct rt_entry *rt; 983c793af95Ssangeeta irb_t *irb; 984c793af95Ssangeeta ire_t *ire; 985c793af95Ssangeeta boolean_t ret; 986c793af95Ssangeeta 987c793af95Ssangeeta rt = (struct rt_entry *)rn; 988c793af95Ssangeeta ASSERT(rt != NULL); 989c793af95Ssangeeta irb = &rt->rt_irb; 990c793af95Ssangeeta for (ire = irb->irb_ire; ire != NULL; ire = ire->ire_next) { 991c793af95Ssangeeta if ((rtf->rt_match_flags != 0) || 992c793af95Ssangeeta (rtf->rt_zoneid != ALL_ZONES)) { 993c793af95Ssangeeta ret = ire_walk_ill_match(rtf->rt_match_flags, 994c793af95Ssangeeta rtf->rt_ire_type, ire, 995f4b3ec61Sdh155122 rtf->rt_ill, rtf->rt_zoneid, rtf->rt_ipst); 996bd670b35SErik Nordmark } else { 997c793af95Ssangeeta ret = B_TRUE; 998bd670b35SErik Nordmark } 999c793af95Ssangeeta if (ret) 1000c793af95Ssangeeta (*rtf->rt_func)(ire, rtf->rt_arg); 1001c793af95Ssangeeta } 1002c793af95Ssangeeta return (0); 1003c793af95Ssangeeta } 1004c793af95Ssangeeta 10057c478bd9Sstevel@tonic-gate /* 1006bd670b35SErik Nordmark * Walk the ftable entries that match the ill. 10077c478bd9Sstevel@tonic-gate */ 1008c793af95Ssangeeta void 10097c478bd9Sstevel@tonic-gate ire_walk_ill_tables(uint_t match_flags, uint_t ire_type, pfv_t func, 101045916cd2Sjpk void *arg, size_t ftbl_sz, size_t htbl_sz, irb_t **ipftbl, 1011bd670b35SErik Nordmark ill_t *ill, zoneid_t zoneid, 1012f4b3ec61Sdh155122 ip_stack_t *ipst) 10137c478bd9Sstevel@tonic-gate { 10147c478bd9Sstevel@tonic-gate irb_t *irb_ptr; 10157c478bd9Sstevel@tonic-gate irb_t *irb; 10167c478bd9Sstevel@tonic-gate ire_t *ire; 10177c478bd9Sstevel@tonic-gate int i, j; 10187c478bd9Sstevel@tonic-gate boolean_t ret; 1019c793af95Ssangeeta struct rtfuncarg rtfarg; 10207c478bd9Sstevel@tonic-gate 1021e11c3f44Smeem ASSERT((!(match_flags & MATCH_IRE_ILL)) || (ill != NULL)); 10227c478bd9Sstevel@tonic-gate ASSERT(!(match_flags & MATCH_IRE_TYPE) || (ire_type != 0)); 1023bd670b35SErik Nordmark 1024c793af95Ssangeeta /* knobs such that routine is called only for v6 case */ 1025f4b3ec61Sdh155122 if (ipftbl == ipst->ips_ip_forwarding_table_v6) { 10267c478bd9Sstevel@tonic-gate for (i = (ftbl_sz - 1); i >= 0; i--) { 10277c478bd9Sstevel@tonic-gate if ((irb_ptr = ipftbl[i]) == NULL) 10287c478bd9Sstevel@tonic-gate continue; 10297c478bd9Sstevel@tonic-gate for (j = 0; j < htbl_sz; j++) { 10307c478bd9Sstevel@tonic-gate irb = &irb_ptr[j]; 10317c478bd9Sstevel@tonic-gate if (irb->irb_ire == NULL) 10327c478bd9Sstevel@tonic-gate continue; 1033c793af95Ssangeeta 1034bd670b35SErik Nordmark irb_refhold(irb); 10357c478bd9Sstevel@tonic-gate for (ire = irb->irb_ire; ire != NULL; 10367c478bd9Sstevel@tonic-gate ire = ire->ire_next) { 10377c478bd9Sstevel@tonic-gate if (match_flags == 0 && 10387c478bd9Sstevel@tonic-gate zoneid == ALL_ZONES) { 10397c478bd9Sstevel@tonic-gate ret = B_TRUE; 10407c478bd9Sstevel@tonic-gate } else { 1041c793af95Ssangeeta ret = 1042c793af95Ssangeeta ire_walk_ill_match( 1043c793af95Ssangeeta match_flags, 1044c793af95Ssangeeta ire_type, ire, ill, 1045f4b3ec61Sdh155122 zoneid, ipst); 10467c478bd9Sstevel@tonic-gate } 10477c478bd9Sstevel@tonic-gate if (ret) 10487c478bd9Sstevel@tonic-gate (*func)(ire, arg); 10497c478bd9Sstevel@tonic-gate } 1050bd670b35SErik Nordmark irb_refrele(irb); 10517c478bd9Sstevel@tonic-gate } 10527c478bd9Sstevel@tonic-gate } 1053c793af95Ssangeeta } else { 1054188e1664SErik Nordmark bzero(&rtfarg, sizeof (rtfarg)); 1055c793af95Ssangeeta rtfarg.rt_func = func; 1056c793af95Ssangeeta rtfarg.rt_arg = arg; 1057c793af95Ssangeeta if (match_flags != 0) { 1058c793af95Ssangeeta rtfarg.rt_match_flags = match_flags; 1059c793af95Ssangeeta } 1060c793af95Ssangeeta rtfarg.rt_ire_type = ire_type; 1061c793af95Ssangeeta rtfarg.rt_ill = ill; 1062c793af95Ssangeeta rtfarg.rt_zoneid = zoneid; 1063f4b3ec61Sdh155122 rtfarg.rt_ipst = ipst; /* No netstack_hold */ 1064f4b3ec61Sdh155122 (void) ipst->ips_ip_ftable->rnh_walktree_mt( 1065f4b3ec61Sdh155122 ipst->ips_ip_ftable, 1066f4b3ec61Sdh155122 rtfunc, &rtfarg, irb_refhold_rn, irb_refrele_rn); 1067c793af95Ssangeeta } 10687c478bd9Sstevel@tonic-gate } 10697c478bd9Sstevel@tonic-gate 10707c478bd9Sstevel@tonic-gate /* 10717c478bd9Sstevel@tonic-gate * This function takes a mask and returns 10727c478bd9Sstevel@tonic-gate * number of bits set in the mask. If no 10737c478bd9Sstevel@tonic-gate * bit is set it returns 0. 10747c478bd9Sstevel@tonic-gate * Assumes a contiguous mask. 10757c478bd9Sstevel@tonic-gate */ 10767c478bd9Sstevel@tonic-gate int 10777c478bd9Sstevel@tonic-gate ip_mask_to_plen(ipaddr_t mask) 10787c478bd9Sstevel@tonic-gate { 10797c478bd9Sstevel@tonic-gate return (mask == 0 ? 0 : IP_ABITS - (ffs(ntohl(mask)) -1)); 10807c478bd9Sstevel@tonic-gate } 10817c478bd9Sstevel@tonic-gate 10827c478bd9Sstevel@tonic-gate /* 10837c478bd9Sstevel@tonic-gate * Convert length for a mask to the mask. 10847c478bd9Sstevel@tonic-gate */ 10857c478bd9Sstevel@tonic-gate ipaddr_t 10867c478bd9Sstevel@tonic-gate ip_plen_to_mask(uint_t masklen) 10877c478bd9Sstevel@tonic-gate { 1088bd670b35SErik Nordmark if (masklen == 0) 1089bd670b35SErik Nordmark return (0); 1090bd670b35SErik Nordmark 10917c478bd9Sstevel@tonic-gate return (htonl(IP_HOST_MASK << (IP_ABITS - masklen))); 10927c478bd9Sstevel@tonic-gate } 10937c478bd9Sstevel@tonic-gate 10947c478bd9Sstevel@tonic-gate void 10957c478bd9Sstevel@tonic-gate ire_atomic_end(irb_t *irb_ptr, ire_t *ire) 10967c478bd9Sstevel@tonic-gate { 10977c478bd9Sstevel@tonic-gate ill_t *ill; 10987c478bd9Sstevel@tonic-gate 1099bd670b35SErik Nordmark ill = ire->ire_ill; 1100bd670b35SErik Nordmark if (ill != NULL) 1101bd670b35SErik Nordmark mutex_exit(&ill->ill_lock); 1102bd670b35SErik Nordmark rw_exit(&irb_ptr->irb_lock); 11037c478bd9Sstevel@tonic-gate } 11047c478bd9Sstevel@tonic-gate 1105bd670b35SErik Nordmark /* 1106bd670b35SErik Nordmark * ire_add_v[46] atomically make sure that the ill associated 1107bd670b35SErik Nordmark * with the new ire is not going away i.e., we check ILL_CONDEMNED. 1108bd670b35SErik Nordmark */ 1109bd670b35SErik Nordmark int 1110bd670b35SErik Nordmark ire_atomic_start(irb_t *irb_ptr, ire_t *ire) 1111bd670b35SErik Nordmark { 1112bd670b35SErik Nordmark ill_t *ill; 1113bd670b35SErik Nordmark 1114bd670b35SErik Nordmark ill = ire->ire_ill; 1115bd670b35SErik Nordmark 1116bd670b35SErik Nordmark rw_enter(&irb_ptr->irb_lock, RW_WRITER); 1117bd670b35SErik Nordmark if (ill != NULL) { 11187c478bd9Sstevel@tonic-gate mutex_enter(&ill->ill_lock); 1119bd670b35SErik Nordmark 1120bd670b35SErik Nordmark /* 1121bd670b35SErik Nordmark * Don't allow IRE's to be created on dying ills. 1122bd670b35SErik Nordmark */ 11237c478bd9Sstevel@tonic-gate if (ill->ill_state_flags & ILL_CONDEMNED) { 1124bd670b35SErik Nordmark ire_atomic_end(irb_ptr, ire); 1125bd670b35SErik Nordmark return (ENXIO); 11267c478bd9Sstevel@tonic-gate } 11277c478bd9Sstevel@tonic-gate 1128bd670b35SErik Nordmark if (IS_UNDER_IPMP(ill)) { 1129bd670b35SErik Nordmark int error = 0; 1130bd670b35SErik Nordmark mutex_enter(&ill->ill_phyint->phyint_lock); 1131bd670b35SErik Nordmark if (!ipmp_ill_is_active(ill) && 1132bd670b35SErik Nordmark IRE_HIDDEN_TYPE(ire->ire_type) && 1133bd670b35SErik Nordmark !ire->ire_testhidden) { 1134bd670b35SErik Nordmark error = EINVAL; 1135dc3879f9Sjarrett } 1136bd670b35SErik Nordmark mutex_exit(&ill->ill_phyint->phyint_lock); 1137dc3879f9Sjarrett if (error != 0) { 1138bd670b35SErik Nordmark ire_atomic_end(irb_ptr, ire); 1139dc3879f9Sjarrett return (error); 1140dc3879f9Sjarrett } 1141dc3879f9Sjarrett } 1142bd670b35SErik Nordmark 1143bd670b35SErik Nordmark } 1144bd670b35SErik Nordmark return (0); 11457c478bd9Sstevel@tonic-gate } 11467c478bd9Sstevel@tonic-gate 11477c478bd9Sstevel@tonic-gate /* 1148bd670b35SErik Nordmark * Add a fully initialized IRE to the forwarding table. 1149bd670b35SErik Nordmark * This returns NULL on failure, or a held IRE on success. 1150bd670b35SErik Nordmark * Normally the returned IRE is the same as the argument. But a different 1151bd670b35SErik Nordmark * IRE will be returned if the added IRE is deemed identical to an existing 1152bd670b35SErik Nordmark * one. In that case ire_identical_ref will be increased. 1153bd670b35SErik Nordmark * The caller always needs to do an ire_refrele() on the returned IRE. 11547c478bd9Sstevel@tonic-gate */ 1155bd670b35SErik Nordmark ire_t * 1156bd670b35SErik Nordmark ire_add(ire_t *ire) 1157bd670b35SErik Nordmark { 1158bd670b35SErik Nordmark if (IRE_HIDDEN_TYPE(ire->ire_type) && 1159bd670b35SErik Nordmark ire->ire_ill != NULL && IS_UNDER_IPMP(ire->ire_ill)) { 1160bd670b35SErik Nordmark /* 1161bd670b35SErik Nordmark * IREs hosted on interfaces that are under IPMP 1162bd670b35SErik Nordmark * should be hidden so that applications don't 1163bd670b35SErik Nordmark * accidentally end up sending packets with test 1164bd670b35SErik Nordmark * addresses as their source addresses, or 1165bd670b35SErik Nordmark * sending out interfaces that are e.g. IFF_INACTIVE. 1166bd670b35SErik Nordmark * Hide them here. 1167bd670b35SErik Nordmark */ 1168bd670b35SErik Nordmark ire->ire_testhidden = B_TRUE; 1169bd670b35SErik Nordmark } 1170bd670b35SErik Nordmark 11715c0b7edeSseb if (ire->ire_ipversion == IPV6_VERSION) 1172bd670b35SErik Nordmark return (ire_add_v6(ire)); 11737c478bd9Sstevel@tonic-gate else 1174bd670b35SErik Nordmark return (ire_add_v4(ire)); 11757c478bd9Sstevel@tonic-gate } 11767c478bd9Sstevel@tonic-gate 11777c478bd9Sstevel@tonic-gate /* 1178bd670b35SErik Nordmark * Add a fully initialized IPv4 IRE to the forwarding table. 1179bd670b35SErik Nordmark * This returns NULL on failure, or a held IRE on success. 1180bd670b35SErik Nordmark * Normally the returned IRE is the same as the argument. But a different 1181bd670b35SErik Nordmark * IRE will be returned if the added IRE is deemed identical to an existing 1182bd670b35SErik Nordmark * one. In that case ire_identical_ref will be increased. 1183bd670b35SErik Nordmark * The caller always needs to do an ire_refrele() on the returned IRE. 11847c478bd9Sstevel@tonic-gate */ 1185bd670b35SErik Nordmark static ire_t * 1186bd670b35SErik Nordmark ire_add_v4(ire_t *ire) 11877c478bd9Sstevel@tonic-gate { 11887c478bd9Sstevel@tonic-gate ire_t *ire1; 11897c478bd9Sstevel@tonic-gate irb_t *irb_ptr; 11907c478bd9Sstevel@tonic-gate ire_t **irep; 1191bd670b35SErik Nordmark int match_flags; 11927c478bd9Sstevel@tonic-gate int error; 1193f4b3ec61Sdh155122 ip_stack_t *ipst = ire->ire_ipst; 1194e11c3f44Smeem 1195bd670b35SErik Nordmark if (ire->ire_ill != NULL) 1196bd670b35SErik Nordmark ASSERT(!MUTEX_HELD(&ire->ire_ill->ill_lock)); 11977c478bd9Sstevel@tonic-gate ASSERT(ire->ire_ipversion == IPV4_VERSION); 11987c478bd9Sstevel@tonic-gate 11997c478bd9Sstevel@tonic-gate /* Make sure the address is properly masked. */ 12007c478bd9Sstevel@tonic-gate ire->ire_addr &= ire->ire_mask; 12017c478bd9Sstevel@tonic-gate 1202bd670b35SErik Nordmark match_flags = (MATCH_IRE_MASK | MATCH_IRE_TYPE | MATCH_IRE_GW); 1203c793af95Ssangeeta 1204bd670b35SErik Nordmark if (ire->ire_ill != NULL) { 1205bd670b35SErik Nordmark match_flags |= MATCH_IRE_ILL; 1206bd670b35SErik Nordmark } 1207bd670b35SErik Nordmark irb_ptr = ire_get_bucket(ire); 1208bd670b35SErik Nordmark if (irb_ptr == NULL) { 1209bd670b35SErik Nordmark printf("no bucket for %p\n", (void *)ire); 12107c478bd9Sstevel@tonic-gate ire_delete(ire); 1211bd670b35SErik Nordmark return (NULL); 12127c478bd9Sstevel@tonic-gate } 12137c478bd9Sstevel@tonic-gate 12147c478bd9Sstevel@tonic-gate /* 1215bd670b35SErik Nordmark * Start the atomic add of the ire. Grab the ill lock, 1216bd670b35SErik Nordmark * the bucket lock. Check for condemned. 12177c478bd9Sstevel@tonic-gate */ 1218bd670b35SErik Nordmark error = ire_atomic_start(irb_ptr, ire); 1219bd670b35SErik Nordmark if (error != 0) { 1220bd670b35SErik Nordmark printf("no ire_atomic_start for %p\n", (void *)ire); 1221bd670b35SErik Nordmark ire_delete(ire); 1222bd670b35SErik Nordmark irb_refrele(irb_ptr); 1223bd670b35SErik Nordmark return (NULL); 1224bd670b35SErik Nordmark } 12257c478bd9Sstevel@tonic-gate /* 1226e11c3f44Smeem * If we are creating a hidden IRE, make sure we search for 1227e11c3f44Smeem * hidden IREs when searching for duplicates below. 1228e11c3f44Smeem * Otherwise, we might find an IRE on some other interface 1229e11c3f44Smeem * that's not marked hidden. 12307c478bd9Sstevel@tonic-gate */ 1231bd670b35SErik Nordmark if (ire->ire_testhidden) 1232bd670b35SErik Nordmark match_flags |= MATCH_IRE_TESTHIDDEN; 12337c478bd9Sstevel@tonic-gate 12347c478bd9Sstevel@tonic-gate /* 12357c478bd9Sstevel@tonic-gate * Atomically check for duplicate and insert in the table. 12367c478bd9Sstevel@tonic-gate */ 12377c478bd9Sstevel@tonic-gate for (ire1 = irb_ptr->irb_ire; ire1 != NULL; ire1 = ire1->ire_next) { 1238bd670b35SErik Nordmark if (IRE_IS_CONDEMNED(ire1)) 12397c478bd9Sstevel@tonic-gate continue; 12407c478bd9Sstevel@tonic-gate /* 1241bd670b35SErik Nordmark * Here we need an exact match on zoneid, i.e., 1242bd670b35SErik Nordmark * ire_match_args doesn't fit. 12437c478bd9Sstevel@tonic-gate */ 12447c478bd9Sstevel@tonic-gate if (ire1->ire_zoneid != ire->ire_zoneid) 12457c478bd9Sstevel@tonic-gate continue; 1246bd670b35SErik Nordmark 1247bd670b35SErik Nordmark if (ire1->ire_type != ire->ire_type) 1248bd670b35SErik Nordmark continue; 1249bd670b35SErik Nordmark 1250bd670b35SErik Nordmark /* 1251bd670b35SErik Nordmark * Note: We do not allow multiple routes that differ only 1252bd670b35SErik Nordmark * in the gateway security attributes; such routes are 1253bd670b35SErik Nordmark * considered duplicates. 1254bd670b35SErik Nordmark * To change that we explicitly have to treat them as 1255bd670b35SErik Nordmark * different here. 1256bd670b35SErik Nordmark */ 12577c478bd9Sstevel@tonic-gate if (ire_match_args(ire1, ire->ire_addr, ire->ire_mask, 1258bd670b35SErik Nordmark ire->ire_gateway_addr, ire->ire_type, ire->ire_ill, 1259bd670b35SErik Nordmark ire->ire_zoneid, NULL, match_flags)) { 12607c478bd9Sstevel@tonic-gate /* 12617c478bd9Sstevel@tonic-gate * Return the old ire after doing a REFHOLD. 12627c478bd9Sstevel@tonic-gate * As most of the callers continue to use the IRE 12637c478bd9Sstevel@tonic-gate * after adding, we return a held ire. This will 12647c478bd9Sstevel@tonic-gate * avoid a lookup in the caller again. If the callers 12657c478bd9Sstevel@tonic-gate * don't want to use it, they need to do a REFRELE. 12667c478bd9Sstevel@tonic-gate */ 1267bd670b35SErik Nordmark atomic_add_32(&ire1->ire_identical_ref, 1); 1268bd670b35SErik Nordmark DTRACE_PROBE2(ire__add__exist, ire_t *, ire1, 1269bd670b35SErik Nordmark ire_t *, ire); 1270bd670b35SErik Nordmark ire_refhold(ire1); 12717c478bd9Sstevel@tonic-gate ire_atomic_end(irb_ptr, ire); 12727c478bd9Sstevel@tonic-gate ire_delete(ire); 1273bd670b35SErik Nordmark irb_refrele(irb_ptr); 1274bd670b35SErik Nordmark return (ire1); 12757c478bd9Sstevel@tonic-gate } 12767c478bd9Sstevel@tonic-gate } 1277e11c3f44Smeem 1278c793af95Ssangeeta /* 1279bd670b35SErik Nordmark * Normally we do head insertion since most things do not care about 1280bd670b35SErik Nordmark * the order of the IREs in the bucket. Note that ip_cgtp_bcast_add 1281bd670b35SErik Nordmark * assumes we at least do head insertion so that its IRE_BROADCAST 1282bd670b35SErik Nordmark * arrive ahead of existing IRE_HOST for the same address. 1283bd670b35SErik Nordmark * However, due to shared-IP zones (and restrict_interzone_loopback) 1284bd670b35SErik Nordmark * we can have an IRE_LOCAL as well as IRE_IF_CLONE for the same 1285bd670b35SErik Nordmark * address. For that reason we do tail insertion for IRE_IF_CLONE. 1286bd670b35SErik Nordmark * Due to the IRE_BROADCAST on cgtp0, which must be last in the bucket, 1287bd670b35SErik Nordmark * we do tail insertion of IRE_BROADCASTs that do not have RTF_MULTIRT 1288bd670b35SErik Nordmark * set. 12897c478bd9Sstevel@tonic-gate */ 12907c478bd9Sstevel@tonic-gate irep = (ire_t **)irb_ptr; 1291bd670b35SErik Nordmark if ((ire->ire_type & IRE_IF_CLONE) || 1292bd670b35SErik Nordmark ((ire->ire_type & IRE_BROADCAST) && 1293bd670b35SErik Nordmark !(ire->ire_flags & RTF_MULTIRT))) { 1294bd670b35SErik Nordmark while ((ire1 = *irep) != NULL) 12957c478bd9Sstevel@tonic-gate irep = &ire1->ire_next; 12967c478bd9Sstevel@tonic-gate } 12977c478bd9Sstevel@tonic-gate /* Insert at *irep */ 12987c478bd9Sstevel@tonic-gate ire1 = *irep; 12997c478bd9Sstevel@tonic-gate if (ire1 != NULL) 13007c478bd9Sstevel@tonic-gate ire1->ire_ptpn = &ire->ire_next; 13017c478bd9Sstevel@tonic-gate ire->ire_next = ire1; 13027c478bd9Sstevel@tonic-gate /* Link the new one in. */ 13037c478bd9Sstevel@tonic-gate ire->ire_ptpn = irep; 13047c478bd9Sstevel@tonic-gate 13057c478bd9Sstevel@tonic-gate /* 13067c478bd9Sstevel@tonic-gate * ire_walk routines de-reference ire_next without holding 13077c478bd9Sstevel@tonic-gate * a lock. Before we point to the new ire, we want to make 13087c478bd9Sstevel@tonic-gate * sure the store that sets the ire_next of the new ire 13097c478bd9Sstevel@tonic-gate * reaches global visibility, so that ire_walk routines 13107c478bd9Sstevel@tonic-gate * don't see a truncated list of ires i.e if the ire_next 13117c478bd9Sstevel@tonic-gate * of the new ire gets set after we do "*irep = ire" due 13127c478bd9Sstevel@tonic-gate * to re-ordering, the ire_walk thread will see a NULL 13137c478bd9Sstevel@tonic-gate * once it accesses the ire_next of the new ire. 13147c478bd9Sstevel@tonic-gate * membar_producer() makes sure that the following store 13157c478bd9Sstevel@tonic-gate * happens *after* all of the above stores. 13167c478bd9Sstevel@tonic-gate */ 13177c478bd9Sstevel@tonic-gate membar_producer(); 13187c478bd9Sstevel@tonic-gate *irep = ire; 13197c478bd9Sstevel@tonic-gate ire->ire_bucket = irb_ptr; 13207c478bd9Sstevel@tonic-gate /* 13217c478bd9Sstevel@tonic-gate * We return a bumped up IRE above. Keep it symmetrical 13227c478bd9Sstevel@tonic-gate * so that the callers will always have to release. This 13237c478bd9Sstevel@tonic-gate * helps the callers of this function because they continue 13247c478bd9Sstevel@tonic-gate * to use the IRE after adding and hence they don't have to 13257c478bd9Sstevel@tonic-gate * lookup again after we return the IRE. 13267c478bd9Sstevel@tonic-gate * 13277c478bd9Sstevel@tonic-gate * NOTE : We don't have to use atomics as this is appearing 13287c478bd9Sstevel@tonic-gate * in the list for the first time and no one else can bump 13297c478bd9Sstevel@tonic-gate * up the reference count on this yet. 13307c478bd9Sstevel@tonic-gate */ 1331bd670b35SErik Nordmark ire_refhold_locked(ire); 1332f4b3ec61Sdh155122 BUMP_IRE_STATS(ipst->ips_ire_stats_v4, ire_stats_inserted); 1333c793af95Ssangeeta 13347c478bd9Sstevel@tonic-gate irb_ptr->irb_ire_cnt++; 1335bd670b35SErik Nordmark if (irb_ptr->irb_marks & IRB_MARK_DYNAMIC) 1336c793af95Ssangeeta irb_ptr->irb_nire++; 1337c793af95Ssangeeta 1338bd670b35SErik Nordmark if (ire->ire_ill != NULL) { 1339bd670b35SErik Nordmark ire->ire_ill->ill_ire_cnt++; 1340bd670b35SErik Nordmark ASSERT(ire->ire_ill->ill_ire_cnt != 0); /* Wraparound */ 13417c478bd9Sstevel@tonic-gate } 13427c478bd9Sstevel@tonic-gate 13437c478bd9Sstevel@tonic-gate ire_atomic_end(irb_ptr, ire); 13447c478bd9Sstevel@tonic-gate 1345bd670b35SErik Nordmark /* Make any caching of the IREs be notified or updated */ 13467c478bd9Sstevel@tonic-gate ire_flush_cache_v4(ire, IRE_FLUSH_ADD); 1347bd670b35SErik Nordmark 1348bd670b35SErik Nordmark if (ire->ire_ill != NULL) 1349bd670b35SErik Nordmark ASSERT(!MUTEX_HELD(&ire->ire_ill->ill_lock)); 1350bd670b35SErik Nordmark irb_refrele(irb_ptr); 1351bd670b35SErik Nordmark return (ire); 13527c478bd9Sstevel@tonic-gate } 13537c478bd9Sstevel@tonic-gate 13547c478bd9Sstevel@tonic-gate /* 1355bd670b35SErik Nordmark * irb_refrele is the only caller of the function. ire_unlink calls to 13567c478bd9Sstevel@tonic-gate * do the final cleanup for this ire. 13577c478bd9Sstevel@tonic-gate */ 13587c478bd9Sstevel@tonic-gate void 13597c478bd9Sstevel@tonic-gate ire_cleanup(ire_t *ire) 13607c478bd9Sstevel@tonic-gate { 13617c478bd9Sstevel@tonic-gate ire_t *ire_next; 1362f4b3ec61Sdh155122 ip_stack_t *ipst = ire->ire_ipst; 13637c478bd9Sstevel@tonic-gate 13647c478bd9Sstevel@tonic-gate ASSERT(ire != NULL); 13657c478bd9Sstevel@tonic-gate 13667c478bd9Sstevel@tonic-gate while (ire != NULL) { 13677c478bd9Sstevel@tonic-gate ire_next = ire->ire_next; 13687c478bd9Sstevel@tonic-gate if (ire->ire_ipversion == IPV4_VERSION) { 13697c478bd9Sstevel@tonic-gate ire_delete_v4(ire); 1370f4b3ec61Sdh155122 BUMP_IRE_STATS(ipst->ips_ire_stats_v4, 1371f4b3ec61Sdh155122 ire_stats_deleted); 13727c478bd9Sstevel@tonic-gate } else { 13737c478bd9Sstevel@tonic-gate ASSERT(ire->ire_ipversion == IPV6_VERSION); 13747c478bd9Sstevel@tonic-gate ire_delete_v6(ire); 1375f4b3ec61Sdh155122 BUMP_IRE_STATS(ipst->ips_ire_stats_v6, 1376f4b3ec61Sdh155122 ire_stats_deleted); 13777c478bd9Sstevel@tonic-gate } 13787c478bd9Sstevel@tonic-gate /* 13797c478bd9Sstevel@tonic-gate * Now it's really out of the list. Before doing the 13807c478bd9Sstevel@tonic-gate * REFRELE, set ire_next to NULL as ire_inactive asserts 13817c478bd9Sstevel@tonic-gate * so. 13827c478bd9Sstevel@tonic-gate */ 13837c478bd9Sstevel@tonic-gate ire->ire_next = NULL; 1384bd670b35SErik Nordmark ire_refrele_notr(ire); 13857c478bd9Sstevel@tonic-gate ire = ire_next; 13867c478bd9Sstevel@tonic-gate } 13877c478bd9Sstevel@tonic-gate } 13887c478bd9Sstevel@tonic-gate 13897c478bd9Sstevel@tonic-gate /* 1390bd670b35SErik Nordmark * irb_refrele is the only caller of the function. It calls to unlink 13917c478bd9Sstevel@tonic-gate * all the CONDEMNED ires from this bucket. 13927c478bd9Sstevel@tonic-gate */ 13937c478bd9Sstevel@tonic-gate ire_t * 13947c478bd9Sstevel@tonic-gate ire_unlink(irb_t *irb) 13957c478bd9Sstevel@tonic-gate { 13967c478bd9Sstevel@tonic-gate ire_t *ire; 13977c478bd9Sstevel@tonic-gate ire_t *ire1; 13987c478bd9Sstevel@tonic-gate ire_t **ptpn; 13997c478bd9Sstevel@tonic-gate ire_t *ire_list = NULL; 14007c478bd9Sstevel@tonic-gate 14017c478bd9Sstevel@tonic-gate ASSERT(RW_WRITE_HELD(&irb->irb_lock)); 1402bd670b35SErik Nordmark ASSERT(((irb->irb_marks & IRB_MARK_DYNAMIC) && irb->irb_refcnt == 1) || 1403c793af95Ssangeeta (irb->irb_refcnt == 0)); 1404c793af95Ssangeeta ASSERT(irb->irb_marks & IRB_MARK_CONDEMNED); 14057c478bd9Sstevel@tonic-gate ASSERT(irb->irb_ire != NULL); 14067c478bd9Sstevel@tonic-gate 14077c478bd9Sstevel@tonic-gate for (ire = irb->irb_ire; ire != NULL; ire = ire1) { 14087c478bd9Sstevel@tonic-gate ire1 = ire->ire_next; 1409bd670b35SErik Nordmark if (IRE_IS_CONDEMNED(ire)) { 14107c478bd9Sstevel@tonic-gate ptpn = ire->ire_ptpn; 14117c478bd9Sstevel@tonic-gate ire1 = ire->ire_next; 14127c478bd9Sstevel@tonic-gate if (ire1) 14137c478bd9Sstevel@tonic-gate ire1->ire_ptpn = ptpn; 14147c478bd9Sstevel@tonic-gate *ptpn = ire1; 14157c478bd9Sstevel@tonic-gate ire->ire_ptpn = NULL; 14167c478bd9Sstevel@tonic-gate ire->ire_next = NULL; 1417bd670b35SErik Nordmark 14187c478bd9Sstevel@tonic-gate /* 1419bd670b35SErik Nordmark * We need to call ire_delete_v4 or ire_delete_v6 to 1420bd670b35SErik Nordmark * clean up dependents and the redirects pointing at 14217c478bd9Sstevel@tonic-gate * the default gateway. We need to drop the lock 14227c478bd9Sstevel@tonic-gate * as ire_flush_cache/ire_delete_host_redircts require 14237c478bd9Sstevel@tonic-gate * so. But we can't drop the lock, as ire_unlink needs 14247c478bd9Sstevel@tonic-gate * to atomically remove the ires from the list. 14257c478bd9Sstevel@tonic-gate * So, create a temporary list of CONDEMNED ires 14267c478bd9Sstevel@tonic-gate * for doing ire_delete_v4/ire_delete_v6 operations 14277c478bd9Sstevel@tonic-gate * later on. 14287c478bd9Sstevel@tonic-gate */ 14297c478bd9Sstevel@tonic-gate ire->ire_next = ire_list; 14307c478bd9Sstevel@tonic-gate ire_list = ire; 14317c478bd9Sstevel@tonic-gate } 14327c478bd9Sstevel@tonic-gate } 1433c793af95Ssangeeta irb->irb_marks &= ~IRB_MARK_CONDEMNED; 14347c478bd9Sstevel@tonic-gate return (ire_list); 14357c478bd9Sstevel@tonic-gate } 14367c478bd9Sstevel@tonic-gate 14377c478bd9Sstevel@tonic-gate /* 1438bd670b35SErik Nordmark * Clean up the radix node for this ire. Must be called by irb_refrele 1439c793af95Ssangeeta * when there are no ire's left in the bucket. Returns TRUE if the bucket 1440c793af95Ssangeeta * is deleted and freed. 1441c793af95Ssangeeta */ 1442c793af95Ssangeeta boolean_t 1443c793af95Ssangeeta irb_inactive(irb_t *irb) 1444c793af95Ssangeeta { 1445c793af95Ssangeeta struct rt_entry *rt; 1446c793af95Ssangeeta struct radix_node *rn; 1447f4b3ec61Sdh155122 ip_stack_t *ipst = irb->irb_ipst; 1448f4b3ec61Sdh155122 1449f4b3ec61Sdh155122 ASSERT(irb->irb_ipst != NULL); 1450c793af95Ssangeeta 1451c793af95Ssangeeta rt = IRB2RT(irb); 1452c793af95Ssangeeta rn = (struct radix_node *)rt; 1453c793af95Ssangeeta 1454c793af95Ssangeeta /* first remove it from the radix tree. */ 1455f4b3ec61Sdh155122 RADIX_NODE_HEAD_WLOCK(ipst->ips_ip_ftable); 1456c793af95Ssangeeta rw_enter(&irb->irb_lock, RW_WRITER); 1457c793af95Ssangeeta if (irb->irb_refcnt == 1 && irb->irb_nire == 0) { 1458f4b3ec61Sdh155122 rn = ipst->ips_ip_ftable->rnh_deladdr(rn->rn_key, rn->rn_mask, 1459f4b3ec61Sdh155122 ipst->ips_ip_ftable); 1460c793af95Ssangeeta DTRACE_PROBE1(irb__free, rt_t *, rt); 1461c793af95Ssangeeta ASSERT((void *)rn == (void *)rt); 1462c793af95Ssangeeta Free(rt, rt_entry_cache); 1463c793af95Ssangeeta /* irb_lock is freed */ 1464f4b3ec61Sdh155122 RADIX_NODE_HEAD_UNLOCK(ipst->ips_ip_ftable); 1465c793af95Ssangeeta return (B_TRUE); 1466c793af95Ssangeeta } 1467c793af95Ssangeeta rw_exit(&irb->irb_lock); 1468f4b3ec61Sdh155122 RADIX_NODE_HEAD_UNLOCK(ipst->ips_ip_ftable); 1469c793af95Ssangeeta return (B_FALSE); 14707c478bd9Sstevel@tonic-gate } 14717c478bd9Sstevel@tonic-gate 14727c478bd9Sstevel@tonic-gate /* 14737c478bd9Sstevel@tonic-gate * Delete the specified IRE. 1474bd670b35SErik Nordmark * We assume that if ire_bucket is not set then ire_ill->ill_ire_cnt was 1475bd670b35SErik Nordmark * not incremented i.e., that the insertion in the bucket and the increment 1476bd670b35SErik Nordmark * of that counter is done atomically. 14777c478bd9Sstevel@tonic-gate */ 14787c478bd9Sstevel@tonic-gate void 14797c478bd9Sstevel@tonic-gate ire_delete(ire_t *ire) 14807c478bd9Sstevel@tonic-gate { 14817c478bd9Sstevel@tonic-gate ire_t *ire1; 14827c478bd9Sstevel@tonic-gate ire_t **ptpn; 14837c478bd9Sstevel@tonic-gate irb_t *irb; 1484f4b3ec61Sdh155122 ip_stack_t *ipst = ire->ire_ipst; 14857c478bd9Sstevel@tonic-gate 1486c793af95Ssangeeta if ((irb = ire->ire_bucket) == NULL) { 14877c478bd9Sstevel@tonic-gate /* 14887c478bd9Sstevel@tonic-gate * It was never inserted in the list. Should call REFRELE 14897c478bd9Sstevel@tonic-gate * to free this IRE. 14907c478bd9Sstevel@tonic-gate */ 1491f1c454b4SSowmini Varadhan ire_make_condemned(ire); 1492bd670b35SErik Nordmark ire_refrele_notr(ire); 14937c478bd9Sstevel@tonic-gate return; 14947c478bd9Sstevel@tonic-gate } 14957c478bd9Sstevel@tonic-gate 14967c478bd9Sstevel@tonic-gate /* 1497bd670b35SErik Nordmark * Move the use counts from an IRE_IF_CLONE to its parent 1498bd670b35SErik Nordmark * IRE_INTERFACE. 1499bd670b35SErik Nordmark * We need to do this before acquiring irb_lock. 15007c478bd9Sstevel@tonic-gate */ 1501bd670b35SErik Nordmark if (ire->ire_type & IRE_IF_CLONE) { 1502bd670b35SErik Nordmark ire_t *parent; 1503bd670b35SErik Nordmark 1504bd670b35SErik Nordmark rw_enter(&ipst->ips_ire_dep_lock, RW_READER); 1505bd670b35SErik Nordmark if ((parent = ire->ire_dep_parent) != NULL) { 1506bd670b35SErik Nordmark parent->ire_ob_pkt_count += ire->ire_ob_pkt_count; 1507bd670b35SErik Nordmark parent->ire_ib_pkt_count += ire->ire_ib_pkt_count; 1508bd670b35SErik Nordmark ire->ire_ob_pkt_count = 0; 1509bd670b35SErik Nordmark ire->ire_ib_pkt_count = 0; 1510bd670b35SErik Nordmark } 1511bd670b35SErik Nordmark rw_exit(&ipst->ips_ire_dep_lock); 15127c478bd9Sstevel@tonic-gate } 15137c478bd9Sstevel@tonic-gate 1514bd670b35SErik Nordmark rw_enter(&irb->irb_lock, RW_WRITER); 15157c478bd9Sstevel@tonic-gate if (ire->ire_ptpn == NULL) { 15167c478bd9Sstevel@tonic-gate /* 15177c478bd9Sstevel@tonic-gate * Some other thread has removed us from the list. 15187c478bd9Sstevel@tonic-gate * It should have done the REFRELE for us. 15197c478bd9Sstevel@tonic-gate */ 15207c478bd9Sstevel@tonic-gate rw_exit(&irb->irb_lock); 15217c478bd9Sstevel@tonic-gate return; 15227c478bd9Sstevel@tonic-gate } 15237c478bd9Sstevel@tonic-gate 1524bd670b35SErik Nordmark if (!IRE_IS_CONDEMNED(ire)) { 1525bd670b35SErik Nordmark /* Is this an IRE representing multiple duplicate entries? */ 1526bd670b35SErik Nordmark ASSERT(ire->ire_identical_ref >= 1); 1527bd670b35SErik Nordmark if (atomic_add_32_nv(&ire->ire_identical_ref, -1) != 0) { 1528bd670b35SErik Nordmark /* Removed one of the identical parties */ 1529bd670b35SErik Nordmark rw_exit(&irb->irb_lock); 1530bd670b35SErik Nordmark return; 15319a09d68dSja97890 } 1532bd670b35SErik Nordmark 1533bd670b35SErik Nordmark irb->irb_ire_cnt--; 1534bd670b35SErik Nordmark ire_make_condemned(ire); 15359a09d68dSja97890 } 15369a09d68dSja97890 15377c478bd9Sstevel@tonic-gate if (irb->irb_refcnt != 0) { 15387c478bd9Sstevel@tonic-gate /* 15397c478bd9Sstevel@tonic-gate * The last thread to leave this bucket will 15407c478bd9Sstevel@tonic-gate * delete this ire. 15417c478bd9Sstevel@tonic-gate */ 1542c793af95Ssangeeta irb->irb_marks |= IRB_MARK_CONDEMNED; 15437c478bd9Sstevel@tonic-gate rw_exit(&irb->irb_lock); 15447c478bd9Sstevel@tonic-gate return; 15457c478bd9Sstevel@tonic-gate } 15467c478bd9Sstevel@tonic-gate 15477c478bd9Sstevel@tonic-gate /* 15487c478bd9Sstevel@tonic-gate * Normally to delete an ire, we walk the bucket. While we 15497c478bd9Sstevel@tonic-gate * walk the bucket, we normally bump up irb_refcnt and hence 15507c478bd9Sstevel@tonic-gate * we return from above where we mark CONDEMNED and the ire 15517c478bd9Sstevel@tonic-gate * gets deleted from ire_unlink. This case is where somebody 15527c478bd9Sstevel@tonic-gate * knows the ire e.g by doing a lookup, and wants to delete the 15537c478bd9Sstevel@tonic-gate * IRE. irb_refcnt would be 0 in this case if nobody is walking 15547c478bd9Sstevel@tonic-gate * the bucket. 15557c478bd9Sstevel@tonic-gate */ 15567c478bd9Sstevel@tonic-gate ptpn = ire->ire_ptpn; 15577c478bd9Sstevel@tonic-gate ire1 = ire->ire_next; 15587c478bd9Sstevel@tonic-gate if (ire1 != NULL) 15597c478bd9Sstevel@tonic-gate ire1->ire_ptpn = ptpn; 15607c478bd9Sstevel@tonic-gate ASSERT(ptpn != NULL); 15617c478bd9Sstevel@tonic-gate *ptpn = ire1; 15627c478bd9Sstevel@tonic-gate ire->ire_ptpn = NULL; 15637c478bd9Sstevel@tonic-gate ire->ire_next = NULL; 15647c478bd9Sstevel@tonic-gate if (ire->ire_ipversion == IPV6_VERSION) { 1565f4b3ec61Sdh155122 BUMP_IRE_STATS(ipst->ips_ire_stats_v6, ire_stats_deleted); 15667c478bd9Sstevel@tonic-gate } else { 1567f4b3ec61Sdh155122 BUMP_IRE_STATS(ipst->ips_ire_stats_v4, ire_stats_deleted); 15687c478bd9Sstevel@tonic-gate } 15697c478bd9Sstevel@tonic-gate rw_exit(&irb->irb_lock); 15707c478bd9Sstevel@tonic-gate 1571bd670b35SErik Nordmark /* Cleanup dependents and related stuff */ 15727c478bd9Sstevel@tonic-gate if (ire->ire_ipversion == IPV6_VERSION) { 15737c478bd9Sstevel@tonic-gate ire_delete_v6(ire); 15747c478bd9Sstevel@tonic-gate } else { 15757c478bd9Sstevel@tonic-gate ire_delete_v4(ire); 15767c478bd9Sstevel@tonic-gate } 15777c478bd9Sstevel@tonic-gate /* 15787c478bd9Sstevel@tonic-gate * We removed it from the list. Decrement the 15797c478bd9Sstevel@tonic-gate * reference count. 15807c478bd9Sstevel@tonic-gate */ 1581bd670b35SErik Nordmark ire_refrele_notr(ire); 15827c478bd9Sstevel@tonic-gate } 15837c478bd9Sstevel@tonic-gate 15847c478bd9Sstevel@tonic-gate /* 15857c478bd9Sstevel@tonic-gate * Delete the specified IRE. 15867c478bd9Sstevel@tonic-gate * All calls should use ire_delete(). 15877c478bd9Sstevel@tonic-gate * Sometimes called as writer though not required by this function. 15887c478bd9Sstevel@tonic-gate * 15897c478bd9Sstevel@tonic-gate * NOTE : This function is called only if the ire was added 15907c478bd9Sstevel@tonic-gate * in the list. 15917c478bd9Sstevel@tonic-gate */ 15927c478bd9Sstevel@tonic-gate static void 15937c478bd9Sstevel@tonic-gate ire_delete_v4(ire_t *ire) 15947c478bd9Sstevel@tonic-gate { 1595f4b3ec61Sdh155122 ip_stack_t *ipst = ire->ire_ipst; 1596f4b3ec61Sdh155122 15977c478bd9Sstevel@tonic-gate ASSERT(ire->ire_refcnt >= 1); 15987c478bd9Sstevel@tonic-gate ASSERT(ire->ire_ipversion == IPV4_VERSION); 15997c478bd9Sstevel@tonic-gate 16007c478bd9Sstevel@tonic-gate ire_flush_cache_v4(ire, IRE_FLUSH_DELETE); 16017c478bd9Sstevel@tonic-gate if (ire->ire_type == IRE_DEFAULT) { 16027c478bd9Sstevel@tonic-gate /* 16037c478bd9Sstevel@tonic-gate * when a default gateway is going away 16047c478bd9Sstevel@tonic-gate * delete all the host redirects pointing at that 16057c478bd9Sstevel@tonic-gate * gateway. 16067c478bd9Sstevel@tonic-gate */ 1607f4b3ec61Sdh155122 ire_delete_host_redirects(ire->ire_gateway_addr, ipst); 16087c478bd9Sstevel@tonic-gate } 1609bd670b35SErik Nordmark 1610bd670b35SErik Nordmark /* 1611bd670b35SErik Nordmark * If we are deleting an IRE_INTERFACE then we make sure we also 1612bd670b35SErik Nordmark * delete any IRE_IF_CLONE that has been created from it. 1613bd670b35SErik Nordmark * Those are always in ire_dep_children. 1614bd670b35SErik Nordmark */ 1615bd670b35SErik Nordmark if ((ire->ire_type & IRE_INTERFACE) && ire->ire_dep_children != NULL) 1616bd670b35SErik Nordmark ire_dep_delete_if_clone(ire); 1617bd670b35SErik Nordmark 1618bd670b35SErik Nordmark /* Remove from parent dependencies and child */ 1619bd670b35SErik Nordmark rw_enter(&ipst->ips_ire_dep_lock, RW_WRITER); 1620bd670b35SErik Nordmark if (ire->ire_dep_parent != NULL) 1621bd670b35SErik Nordmark ire_dep_remove(ire); 1622bd670b35SErik Nordmark 1623bd670b35SErik Nordmark while (ire->ire_dep_children != NULL) 1624bd670b35SErik Nordmark ire_dep_remove(ire->ire_dep_children); 1625bd670b35SErik Nordmark rw_exit(&ipst->ips_ire_dep_lock); 16267c478bd9Sstevel@tonic-gate } 16277c478bd9Sstevel@tonic-gate 16287c478bd9Sstevel@tonic-gate /* 1629bd670b35SErik Nordmark * ire_refrele is the only caller of the function. It calls 16307c478bd9Sstevel@tonic-gate * to free the ire when the reference count goes to zero. 16317c478bd9Sstevel@tonic-gate */ 16327c478bd9Sstevel@tonic-gate void 16337c478bd9Sstevel@tonic-gate ire_inactive(ire_t *ire) 16347c478bd9Sstevel@tonic-gate { 1635bd670b35SErik Nordmark ill_t *ill; 1636c793af95Ssangeeta irb_t *irb; 1637f4b3ec61Sdh155122 ip_stack_t *ipst = ire->ire_ipst; 16387c478bd9Sstevel@tonic-gate 16397c478bd9Sstevel@tonic-gate ASSERT(ire->ire_refcnt == 0); 16407c478bd9Sstevel@tonic-gate ASSERT(ire->ire_ptpn == NULL); 16417c478bd9Sstevel@tonic-gate ASSERT(ire->ire_next == NULL); 16427c478bd9Sstevel@tonic-gate 1643bd670b35SErik Nordmark /* Count how many condemned ires for kmem_cache callback */ 1644f1c454b4SSowmini Varadhan ASSERT(IRE_IS_CONDEMNED(ire)); 1645bd670b35SErik Nordmark atomic_add_32(&ipst->ips_num_ire_condemned, -1); 1646bd670b35SErik Nordmark 1647c793af95Ssangeeta if (ire->ire_gw_secattr != NULL) { 1648c793af95Ssangeeta ire_gw_secattr_free(ire->ire_gw_secattr); 1649c793af95Ssangeeta ire->ire_gw_secattr = NULL; 1650c793af95Ssangeeta } 1651c793af95Ssangeeta 1652bd670b35SErik Nordmark /* 1653bd670b35SErik Nordmark * ire_nce_cache is cleared in ire_delete, and we make sure we don't 1654bd670b35SErik Nordmark * set it once the ire is marked condemned. 1655bd670b35SErik Nordmark */ 1656bd670b35SErik Nordmark ASSERT(ire->ire_nce_cache == NULL); 16577c478bd9Sstevel@tonic-gate 16587c478bd9Sstevel@tonic-gate /* 1659bd670b35SErik Nordmark * Since any parent would have a refhold on us they would already 1660bd670b35SErik Nordmark * have been removed. 16617c478bd9Sstevel@tonic-gate */ 1662bd670b35SErik Nordmark ASSERT(ire->ire_dep_parent == NULL); 1663bd670b35SErik Nordmark ASSERT(ire->ire_dep_sib_next == NULL); 1664bd670b35SErik Nordmark ASSERT(ire->ire_dep_sib_ptpn == NULL); 16657c478bd9Sstevel@tonic-gate 1666bd670b35SErik Nordmark /* 1667bd670b35SErik Nordmark * Since any children would have a refhold on us they should have 1668bd670b35SErik Nordmark * already been removed. 1669bd670b35SErik Nordmark */ 1670bd670b35SErik Nordmark ASSERT(ire->ire_dep_children == NULL); 1671bd670b35SErik Nordmark 1672bd670b35SErik Nordmark /* 1673bd670b35SErik Nordmark * ill_ire_ref is increased when the IRE is inserted in the 1674bd670b35SErik Nordmark * bucket - not when the IRE is created. 1675bd670b35SErik Nordmark */ 1676bd670b35SErik Nordmark irb = ire->ire_bucket; 1677bd670b35SErik Nordmark ill = ire->ire_ill; 1678bd670b35SErik Nordmark if (irb != NULL && ill != NULL) { 16797c478bd9Sstevel@tonic-gate mutex_enter(&ill->ill_lock); 1680bd670b35SErik Nordmark ASSERT(ill->ill_ire_cnt != 0); 1681bd670b35SErik Nordmark DTRACE_PROBE3(ill__decr__cnt, (ill_t *), ill, 1682968d2fd1Ssowmini (char *), "ire", (void *), ire); 1683bd670b35SErik Nordmark ill->ill_ire_cnt--; 1684bd670b35SErik Nordmark if (ILL_DOWN_OK(ill)) { 16857c478bd9Sstevel@tonic-gate /* Drops the ill lock */ 16867c478bd9Sstevel@tonic-gate ipif_ill_refrele_tail(ill); 16877c478bd9Sstevel@tonic-gate } else { 16887c478bd9Sstevel@tonic-gate mutex_exit(&ill->ill_lock); 16897c478bd9Sstevel@tonic-gate } 16907c478bd9Sstevel@tonic-gate } 1691bd670b35SErik Nordmark ire->ire_ill = NULL; 1692bd670b35SErik Nordmark 16937c478bd9Sstevel@tonic-gate /* This should be true for both V4 and V6 */ 1694bd670b35SErik Nordmark if (irb != NULL && (irb->irb_marks & IRB_MARK_DYNAMIC)) { 1695c793af95Ssangeeta rw_enter(&irb->irb_lock, RW_WRITER); 1696c793af95Ssangeeta irb->irb_nire--; 1697c793af95Ssangeeta /* 1698c793af95Ssangeeta * Instead of examining the conditions for freeing 1699c793af95Ssangeeta * the radix node here, we do it by calling 1700bd670b35SErik Nordmark * irb_refrele which is a single point in the code 1701c793af95Ssangeeta * that embeds that logic. Bump up the refcnt to 1702bd670b35SErik Nordmark * be able to call irb_refrele 1703c793af95Ssangeeta */ 1704bd670b35SErik Nordmark irb_refhold_locked(irb); 1705c793af95Ssangeeta rw_exit(&irb->irb_lock); 1706bd670b35SErik Nordmark irb_refrele(irb); 1707c793af95Ssangeeta } 17087c478bd9Sstevel@tonic-gate 17096a8288c7Scarlsonj #ifdef DEBUG 17106a8288c7Scarlsonj ire_trace_cleanup(ire); 17117c478bd9Sstevel@tonic-gate #endif 17127c478bd9Sstevel@tonic-gate mutex_destroy(&ire->ire_lock); 17137c478bd9Sstevel@tonic-gate if (ire->ire_ipversion == IPV6_VERSION) { 1714f4b3ec61Sdh155122 BUMP_IRE_STATS(ipst->ips_ire_stats_v6, ire_stats_freed); 17157c478bd9Sstevel@tonic-gate } else { 1716f4b3ec61Sdh155122 BUMP_IRE_STATS(ipst->ips_ire_stats_v4, ire_stats_freed); 17177c478bd9Sstevel@tonic-gate } 17187c478bd9Sstevel@tonic-gate kmem_cache_free(ire_cache, ire); 17197c478bd9Sstevel@tonic-gate } 17207c478bd9Sstevel@tonic-gate 17217c478bd9Sstevel@tonic-gate /* 1722bd670b35SErik Nordmark * ire_update_generation is the callback function provided by 1723bd670b35SErik Nordmark * ire_get_bucket() to update the generation number of any 1724bd670b35SErik Nordmark * matching shorter route when a new route is added. 1725bd670b35SErik Nordmark * 1726bd670b35SErik Nordmark * This fucntion always returns a failure return (B_FALSE) 1727bd670b35SErik Nordmark * to force the caller (rn_matchaddr_args) 1728bd670b35SErik Nordmark * to back-track up the tree looking for shorter matches. 17297c478bd9Sstevel@tonic-gate */ 1730bd670b35SErik Nordmark /* ARGSUSED */ 1731bd670b35SErik Nordmark static boolean_t 1732bd670b35SErik Nordmark ire_update_generation(struct radix_node *rn, void *arg) 17337c478bd9Sstevel@tonic-gate { 1734bd670b35SErik Nordmark struct rt_entry *rt = (struct rt_entry *)rn; 17357c478bd9Sstevel@tonic-gate 1736bd670b35SErik Nordmark /* We need to handle all in the same bucket */ 1737bd670b35SErik Nordmark irb_increment_generation(&rt->rt_irb); 1738bd670b35SErik Nordmark return (B_FALSE); 17397c478bd9Sstevel@tonic-gate } 17407c478bd9Sstevel@tonic-gate 17417c478bd9Sstevel@tonic-gate /* 1742bd670b35SErik Nordmark * Take care of all the generation numbers in the bucket. 1743bd670b35SErik Nordmark */ 1744bd670b35SErik Nordmark void 1745bd670b35SErik Nordmark irb_increment_generation(irb_t *irb) 1746bd670b35SErik Nordmark { 1747bd670b35SErik Nordmark ire_t *ire; 1748f1c454b4SSowmini Varadhan ip_stack_t *ipst; 1749bd670b35SErik Nordmark 1750bd670b35SErik Nordmark if (irb == NULL || irb->irb_ire_cnt == 0) 1751bd670b35SErik Nordmark return; 1752bd670b35SErik Nordmark 1753f1c454b4SSowmini Varadhan ipst = irb->irb_ipst; 1754f1c454b4SSowmini Varadhan /* 1755f1c454b4SSowmini Varadhan * we cannot do an irb_refhold/irb_refrele here as the caller 1756f1c454b4SSowmini Varadhan * already has the global RADIX_NODE_HEAD_WLOCK, and the irb_refrele 1757f1c454b4SSowmini Varadhan * may result in an attempt to free the irb_t, which also needs 1758f1c454b4SSowmini Varadhan * the RADIX_NODE_HEAD lock. However, since we want to traverse the 1759f1c454b4SSowmini Varadhan * irb_ire list without fear of having a condemned ire removed from 1760f1c454b4SSowmini Varadhan * the list, we acquire the irb_lock as WRITER. Moreover, since 1761f1c454b4SSowmini Varadhan * the ire_generation increments are done under the ire_dep_lock, 1762f1c454b4SSowmini Varadhan * acquire the locks in the prescribed lock order first. 1763f1c454b4SSowmini Varadhan */ 1764f1c454b4SSowmini Varadhan rw_enter(&ipst->ips_ire_dep_lock, RW_READER); 1765f1c454b4SSowmini Varadhan rw_enter(&irb->irb_lock, RW_WRITER); 1766bd670b35SErik Nordmark for (ire = irb->irb_ire; ire != NULL; ire = ire->ire_next) { 1767bd670b35SErik Nordmark if (!IRE_IS_CONDEMNED(ire)) 1768bd670b35SErik Nordmark ire_increment_generation(ire); /* Ourselves */ 1769f1c454b4SSowmini Varadhan ire_dep_incr_generation_locked(ire); /* Dependants */ 1770bd670b35SErik Nordmark } 1771f1c454b4SSowmini Varadhan rw_exit(&irb->irb_lock); 1772f1c454b4SSowmini Varadhan rw_exit(&ipst->ips_ire_dep_lock); 1773bd670b35SErik Nordmark } 1774bd670b35SErik Nordmark 1775bd670b35SErik Nordmark /* 1776bd670b35SErik Nordmark * When an IRE is added or deleted this routine is called to make sure 1777bd670b35SErik Nordmark * any caching of IRE information is notified or updated. 17787c478bd9Sstevel@tonic-gate * 17797c478bd9Sstevel@tonic-gate * The flag argument indicates if the flush request is due to addition 1780bd670b35SErik Nordmark * of new route (IRE_FLUSH_ADD), deletion of old route (IRE_FLUSH_DELETE), 1781bd670b35SErik Nordmark * or a change to ire_gateway_addr (IRE_FLUSH_GWCHANGE). 17827c478bd9Sstevel@tonic-gate */ 17837c478bd9Sstevel@tonic-gate void 17847c478bd9Sstevel@tonic-gate ire_flush_cache_v4(ire_t *ire, int flag) 17857c478bd9Sstevel@tonic-gate { 1786bd670b35SErik Nordmark irb_t *irb = ire->ire_bucket; 1787bd670b35SErik Nordmark struct rt_entry *rt = IRB2RT(irb); 1788f4b3ec61Sdh155122 ip_stack_t *ipst = ire->ire_ipst; 17897c478bd9Sstevel@tonic-gate 1790bd670b35SErik Nordmark /* 1791bd670b35SErik Nordmark * IRE_IF_CLONE ire's don't provide any new information 1792bd670b35SErik Nordmark * than the parent from which they are cloned, so don't 1793bd670b35SErik Nordmark * perturb the generation numbers. 1794bd670b35SErik Nordmark */ 1795bd670b35SErik Nordmark if (ire->ire_type & IRE_IF_CLONE) 17967c478bd9Sstevel@tonic-gate return; 17977c478bd9Sstevel@tonic-gate 17987c478bd9Sstevel@tonic-gate /* 1799bd670b35SErik Nordmark * Ensure that an ire_add during a lookup serializes the updates of the 1800bd670b35SErik Nordmark * generation numbers under the radix head lock so that the lookup gets 1801bd670b35SErik Nordmark * either the old ire and old generation number, or a new ire and new 1802bd670b35SErik Nordmark * generation number. 18037c478bd9Sstevel@tonic-gate */ 1804bd670b35SErik Nordmark RADIX_NODE_HEAD_WLOCK(ipst->ips_ip_ftable); 1805bd670b35SErik Nordmark 1806bd670b35SErik Nordmark /* 1807bd670b35SErik Nordmark * If a route was just added, we need to notify everybody that 1808bd670b35SErik Nordmark * has cached an IRE_NOROUTE since there might now be a better 1809bd670b35SErik Nordmark * route for them. 1810bd670b35SErik Nordmark */ 18117c478bd9Sstevel@tonic-gate if (flag == IRE_FLUSH_ADD) { 1812bd670b35SErik Nordmark ire_increment_generation(ipst->ips_ire_reject_v4); 1813bd670b35SErik Nordmark ire_increment_generation(ipst->ips_ire_blackhole_v4); 1814bd670b35SErik Nordmark } 1815bd670b35SErik Nordmark 1816bd670b35SErik Nordmark /* Adding a default can't otherwise provide a better route */ 1817bd670b35SErik Nordmark if (ire->ire_type == IRE_DEFAULT && flag == IRE_FLUSH_ADD) { 1818bd670b35SErik Nordmark RADIX_NODE_HEAD_UNLOCK(ipst->ips_ip_ftable); 1819bd670b35SErik Nordmark return; 1820bd670b35SErik Nordmark } 1821bd670b35SErik Nordmark 1822bd670b35SErik Nordmark switch (flag) { 1823bd670b35SErik Nordmark case IRE_FLUSH_DELETE: 1824bd670b35SErik Nordmark case IRE_FLUSH_GWCHANGE: 18257c478bd9Sstevel@tonic-gate /* 1826bd670b35SErik Nordmark * Update ire_generation for all ire_dep_children chains 1827bd670b35SErik Nordmark * starting with this IRE 18287c478bd9Sstevel@tonic-gate */ 1829bd670b35SErik Nordmark ire_dep_incr_generation(ire); 1830bd670b35SErik Nordmark break; 1831bd670b35SErik Nordmark case IRE_FLUSH_ADD: 18327c478bd9Sstevel@tonic-gate /* 1833bd670b35SErik Nordmark * Update the generation numbers of all shorter matching routes. 1834bd670b35SErik Nordmark * ire_update_generation takes care of the dependants by 1835bd670b35SErik Nordmark * using ire_dep_incr_generation. 18367c478bd9Sstevel@tonic-gate */ 1837bd670b35SErik Nordmark (void) ipst->ips_ip_ftable->rnh_matchaddr_args(&rt->rt_dst, 1838bd670b35SErik Nordmark ipst->ips_ip_ftable, ire_update_generation, NULL); 1839bd670b35SErik Nordmark break; 18407c478bd9Sstevel@tonic-gate } 1841bd670b35SErik Nordmark RADIX_NODE_HEAD_UNLOCK(ipst->ips_ip_ftable); 18427c478bd9Sstevel@tonic-gate } 18437c478bd9Sstevel@tonic-gate 18447c478bd9Sstevel@tonic-gate /* 18457c478bd9Sstevel@tonic-gate * Matches the arguments passed with the values in the ire. 18467c478bd9Sstevel@tonic-gate * 1847bd670b35SErik Nordmark * Note: for match types that match using "ill" passed in, ill 18487c478bd9Sstevel@tonic-gate * must be checked for non-NULL before calling this routine. 18497c478bd9Sstevel@tonic-gate */ 1850c793af95Ssangeeta boolean_t 18517c478bd9Sstevel@tonic-gate ire_match_args(ire_t *ire, ipaddr_t addr, ipaddr_t mask, ipaddr_t gateway, 1852bd670b35SErik Nordmark int type, const ill_t *ill, zoneid_t zoneid, 1853bd670b35SErik Nordmark const ts_label_t *tsl, int match_flags) 18547c478bd9Sstevel@tonic-gate { 18557c478bd9Sstevel@tonic-gate ill_t *ire_ill = NULL, *dst_ill; 1856bd670b35SErik Nordmark ip_stack_t *ipst = ire->ire_ipst; 18577c478bd9Sstevel@tonic-gate 18587c478bd9Sstevel@tonic-gate ASSERT(ire->ire_ipversion == IPV4_VERSION); 18597c478bd9Sstevel@tonic-gate ASSERT((ire->ire_addr & ~ire->ire_mask) == 0); 186044b099c4SSowmini Varadhan ASSERT((!(match_flags & (MATCH_IRE_ILL|MATCH_IRE_SRC_ILL))) || 1861bd670b35SErik Nordmark (ill != NULL && !ill->ill_isv6)); 18627c478bd9Sstevel@tonic-gate 18637c478bd9Sstevel@tonic-gate /* 1864bd670b35SErik Nordmark * If MATCH_IRE_TESTHIDDEN is set, then only return the IRE if it is 1865bd670b35SErik Nordmark * in fact hidden, to ensure the caller gets the right one. 18667c478bd9Sstevel@tonic-gate */ 1867bd670b35SErik Nordmark if (ire->ire_testhidden) { 1868bd670b35SErik Nordmark if (!(match_flags & MATCH_IRE_TESTHIDDEN)) 18697c478bd9Sstevel@tonic-gate return (B_FALSE); 1870e11c3f44Smeem } 18717c478bd9Sstevel@tonic-gate 187245916cd2Sjpk if (zoneid != ALL_ZONES && zoneid != ire->ire_zoneid && 187345916cd2Sjpk ire->ire_zoneid != ALL_ZONES) { 18747c478bd9Sstevel@tonic-gate /* 1875bd670b35SErik Nordmark * If MATCH_IRE_ZONEONLY has been set and the supplied zoneid 1876bd670b35SErik Nordmark * does not match that of ire_zoneid, a failure to 18777c478bd9Sstevel@tonic-gate * match is reported at this point. Otherwise, since some IREs 18787c478bd9Sstevel@tonic-gate * that are available in the global zone can be used in local 18797c478bd9Sstevel@tonic-gate * zones, additional checks need to be performed: 18807c478bd9Sstevel@tonic-gate * 1881bd670b35SErik Nordmark * IRE_LOOPBACK 18827c478bd9Sstevel@tonic-gate * entries should never be matched in this situation. 1883bd670b35SErik Nordmark * Each zone has its own IRE_LOOPBACK. 18847c478bd9Sstevel@tonic-gate * 1885bd670b35SErik Nordmark * IRE_LOCAL 1886bd670b35SErik Nordmark * We allow them for any zoneid. ire_route_recursive 1887bd670b35SErik Nordmark * does additional checks when 1888bd670b35SErik Nordmark * ip_restrict_interzone_loopback is set. 18897c478bd9Sstevel@tonic-gate * 1890bd670b35SErik Nordmark * If ill_usesrc_ifindex is set 1891bd670b35SErik Nordmark * Then we check if the zone has a valid source address 1892bd670b35SErik Nordmark * on the usesrc ill. 18937c478bd9Sstevel@tonic-gate * 1894bd670b35SErik Nordmark * If ire_ill is set, then check that the zone has an ipif 1895bd670b35SErik Nordmark * on that ill. 1896bd670b35SErik Nordmark * 1897bd670b35SErik Nordmark * Outside of this function (in ire_round_robin) we check 1898bd670b35SErik Nordmark * that any IRE_OFFLINK has a gateway that reachable from the 1899bd670b35SErik Nordmark * zone when we have multiple choices (ECMP). 19007c478bd9Sstevel@tonic-gate */ 19017c478bd9Sstevel@tonic-gate if (match_flags & MATCH_IRE_ZONEONLY) 19027c478bd9Sstevel@tonic-gate return (B_FALSE); 1903bd670b35SErik Nordmark if (ire->ire_type & IRE_LOOPBACK) 19047c478bd9Sstevel@tonic-gate return (B_FALSE); 1905bd670b35SErik Nordmark 1906bd670b35SErik Nordmark if (ire->ire_type & IRE_LOCAL) 1907bd670b35SErik Nordmark goto matchit; 1908bd670b35SErik Nordmark 19097c478bd9Sstevel@tonic-gate /* 1910bd670b35SErik Nordmark * The normal case of IRE_ONLINK has a matching zoneid. 1911bd670b35SErik Nordmark * Here we handle the case when shared-IP zones have been 1912bd670b35SErik Nordmark * configured with IP addresses on vniN. In that case it 1913bd670b35SErik Nordmark * is ok for traffic from a zone to use IRE_ONLINK routes 1914bd670b35SErik Nordmark * if the ill has a usesrc pointing at vniN 19157c478bd9Sstevel@tonic-gate */ 1916bd670b35SErik Nordmark dst_ill = ire->ire_ill; 1917bd670b35SErik Nordmark if (ire->ire_type & IRE_ONLINK) { 1918bd670b35SErik Nordmark uint_t ifindex; 1919bd670b35SErik Nordmark 1920bd670b35SErik Nordmark /* 1921bd670b35SErik Nordmark * Note there is no IRE_INTERFACE on vniN thus 1922bd670b35SErik Nordmark * can't do an IRE lookup for a matching route. 1923bd670b35SErik Nordmark */ 1924bd670b35SErik Nordmark ifindex = dst_ill->ill_usesrc_ifindex; 1925bd670b35SErik Nordmark if (ifindex == 0) 1926bd670b35SErik Nordmark return (B_FALSE); 1927bd670b35SErik Nordmark 19287c478bd9Sstevel@tonic-gate /* 19297c478bd9Sstevel@tonic-gate * If there is a usable source address in the 1930bd670b35SErik Nordmark * zone, then it's ok to return this IRE_INTERFACE 19317c478bd9Sstevel@tonic-gate */ 1932bd670b35SErik Nordmark if (!ipif_zone_avail(ifindex, dst_ill->ill_isv6, 1933bd670b35SErik Nordmark zoneid, ipst)) { 1934bd670b35SErik Nordmark ip3dbg(("ire_match_args: no usrsrc for zone" 19357c478bd9Sstevel@tonic-gate " dst_ill %p\n", (void *)dst_ill)); 19367c478bd9Sstevel@tonic-gate return (B_FALSE); 19377c478bd9Sstevel@tonic-gate } 19387c478bd9Sstevel@tonic-gate } 1939bd670b35SErik Nordmark /* 194044b099c4SSowmini Varadhan * For example, with 1941bd670b35SErik Nordmark * route add 11.0.0.0 gw1 -ifp bge0 1942bd670b35SErik Nordmark * route add 11.0.0.0 gw2 -ifp bge1 1943bd670b35SErik Nordmark * this code would differentiate based on 1944bd670b35SErik Nordmark * where the sending zone has addresses. 1945bd670b35SErik Nordmark * Only if the zone has an address on bge0 can it use the first 1946bd670b35SErik Nordmark * route. It isn't clear if this behavior is documented 1947bd670b35SErik Nordmark * anywhere. 1948bd670b35SErik Nordmark */ 1949bd670b35SErik Nordmark if (dst_ill != NULL && (ire->ire_type & IRE_OFFLINK)) { 19507c478bd9Sstevel@tonic-gate ipif_t *tipif; 19517c478bd9Sstevel@tonic-gate 1952bd670b35SErik Nordmark mutex_enter(&dst_ill->ill_lock); 1953bd670b35SErik Nordmark for (tipif = dst_ill->ill_ipif; 19547c478bd9Sstevel@tonic-gate tipif != NULL; tipif = tipif->ipif_next) { 1955bd670b35SErik Nordmark if (!IPIF_IS_CONDEMNED(tipif) && 19567c478bd9Sstevel@tonic-gate (tipif->ipif_flags & IPIF_UP) && 195745916cd2Sjpk (tipif->ipif_zoneid == zoneid || 195845916cd2Sjpk tipif->ipif_zoneid == ALL_ZONES)) 19597c478bd9Sstevel@tonic-gate break; 19607c478bd9Sstevel@tonic-gate } 1961bd670b35SErik Nordmark mutex_exit(&dst_ill->ill_lock); 19627c478bd9Sstevel@tonic-gate if (tipif == NULL) { 19637c478bd9Sstevel@tonic-gate return (B_FALSE); 19647c478bd9Sstevel@tonic-gate } 19657c478bd9Sstevel@tonic-gate } 19667c478bd9Sstevel@tonic-gate } 19677c478bd9Sstevel@tonic-gate 1968bd670b35SErik Nordmark matchit: 1969bd670b35SErik Nordmark ire_ill = ire->ire_ill; 197044b099c4SSowmini Varadhan if (match_flags & MATCH_IRE_ILL) { 1971bd670b35SErik Nordmark 1972bd670b35SErik Nordmark /* 1973bd670b35SErik Nordmark * If asked to match an ill, we *must* match 1974bd670b35SErik Nordmark * on the ire_ill for ipmp test addresses, or 1975bd670b35SErik Nordmark * any of the ill in the group for data addresses. 1976bd670b35SErik Nordmark * If we don't, we may as well fail. 1977bd670b35SErik Nordmark * However, we need an exception for IRE_LOCALs to ensure 1978bd670b35SErik Nordmark * we loopback packets even sent to test addresses on different 1979bd670b35SErik Nordmark * interfaces in the group. 1980bd670b35SErik Nordmark */ 1981bd670b35SErik Nordmark if ((match_flags & MATCH_IRE_TESTHIDDEN) && 1982bd670b35SErik Nordmark !(ire->ire_type & IRE_LOCAL)) { 1983bd670b35SErik Nordmark if (ire->ire_ill != ill) 1984bd670b35SErik Nordmark return (B_FALSE); 1985bd670b35SErik Nordmark } else { 1986bd670b35SErik Nordmark match_flags &= ~MATCH_IRE_TESTHIDDEN; 1987bd670b35SErik Nordmark /* 1988bd670b35SErik Nordmark * We know that ill is not NULL, but ire_ill could be 1989bd670b35SErik Nordmark * NULL 1990bd670b35SErik Nordmark */ 1991bd670b35SErik Nordmark if (ire_ill == NULL || !IS_ON_SAME_LAN(ill, ire_ill)) 1992bd670b35SErik Nordmark return (B_FALSE); 1993bd670b35SErik Nordmark } 19947c478bd9Sstevel@tonic-gate } 199544b099c4SSowmini Varadhan if (match_flags & MATCH_IRE_SRC_ILL) { 199644b099c4SSowmini Varadhan if (ire_ill == NULL) 199744b099c4SSowmini Varadhan return (B_FALSE); 199844b099c4SSowmini Varadhan if (!IS_ON_SAME_LAN(ill, ire_ill)) { 199944b099c4SSowmini Varadhan if (ire_ill->ill_usesrc_ifindex == 0 || 200044b099c4SSowmini Varadhan (ire_ill->ill_usesrc_ifindex != 200144b099c4SSowmini Varadhan ill->ill_phyint->phyint_ifindex)) 200244b099c4SSowmini Varadhan return (B_FALSE); 200344b099c4SSowmini Varadhan } 200444b099c4SSowmini Varadhan } 20057c478bd9Sstevel@tonic-gate 20067c478bd9Sstevel@tonic-gate if ((ire->ire_addr == (addr & mask)) && 20077c478bd9Sstevel@tonic-gate ((!(match_flags & MATCH_IRE_GW)) || 20087c478bd9Sstevel@tonic-gate (ire->ire_gateway_addr == gateway)) && 2009*01685f97SSowmini Varadhan ((!(match_flags & MATCH_IRE_DIRECT)) || 2010*01685f97SSowmini Varadhan !(ire->ire_flags & RTF_INDIRECT)) && 2011bd670b35SErik Nordmark ((!(match_flags & MATCH_IRE_TYPE)) || (ire->ire_type & type)) && 2012bd670b35SErik Nordmark ((!(match_flags & MATCH_IRE_TESTHIDDEN)) || ire->ire_testhidden) && 2013bd670b35SErik Nordmark ((!(match_flags & MATCH_IRE_MASK)) || (ire->ire_mask == mask)) && 201445916cd2Sjpk ((!(match_flags & MATCH_IRE_SECATTR)) || 201545916cd2Sjpk (!is_system_labeled()) || 201645916cd2Sjpk (tsol_ire_match_gwattr(ire, tsl) == 0))) { 20177c478bd9Sstevel@tonic-gate /* We found the matched IRE */ 20187c478bd9Sstevel@tonic-gate return (B_TRUE); 20197c478bd9Sstevel@tonic-gate } 20207c478bd9Sstevel@tonic-gate return (B_FALSE); 20217c478bd9Sstevel@tonic-gate } 20227c478bd9Sstevel@tonic-gate 20237c478bd9Sstevel@tonic-gate /* 2024e11c3f44Smeem * Check if the IRE_LOCAL uses the same ill as another route would use. 2025aa2eaee6Snordmark * If there is no alternate route, or the alternate is a REJECT or BLACKHOLE, 2026aa2eaee6Snordmark * then we don't allow this IRE_LOCAL to be used. 2027bd670b35SErik Nordmark * We always return an IRE; will be RTF_REJECT if no route available. 2028bd670b35SErik Nordmark */ 2029bd670b35SErik Nordmark ire_t * 2030bd670b35SErik Nordmark ire_alt_local(ire_t *ire, zoneid_t zoneid, const ts_label_t *tsl, 2031bd670b35SErik Nordmark const ill_t *ill, uint_t *generationp) 2032bd670b35SErik Nordmark { 2033bd670b35SErik Nordmark ip_stack_t *ipst = ire->ire_ipst; 2034bd670b35SErik Nordmark ire_t *alt_ire; 2035bd670b35SErik Nordmark uint_t ire_type; 2036bd670b35SErik Nordmark uint_t generation; 2037bd670b35SErik Nordmark uint_t match_flags; 2038bd670b35SErik Nordmark 2039bd670b35SErik Nordmark ASSERT(ire->ire_type & IRE_LOCAL); 2040bd670b35SErik Nordmark ASSERT(ire->ire_ill != NULL); 2041bd670b35SErik Nordmark 2042bd670b35SErik Nordmark /* 2043bd670b35SErik Nordmark * Need to match on everything but local. 2044bd670b35SErik Nordmark * This might result in the creation of a IRE_IF_CLONE for the 2045bd670b35SErik Nordmark * same address as the IRE_LOCAL when restrict_interzone_loopback is 2046bd670b35SErik Nordmark * set. ire_add_*() ensures that the IRE_IF_CLONE are tail inserted 2047bd670b35SErik Nordmark * to make sure the IRE_LOCAL is always found first. 2048bd670b35SErik Nordmark */ 2049bd670b35SErik Nordmark ire_type = (IRE_ONLINK | IRE_OFFLINK) & ~(IRE_LOCAL|IRE_LOOPBACK); 2050bd670b35SErik Nordmark match_flags = MATCH_IRE_TYPE | MATCH_IRE_SECATTR; 2051bd670b35SErik Nordmark if (ill != NULL) 2052bd670b35SErik Nordmark match_flags |= MATCH_IRE_ILL; 2053bd670b35SErik Nordmark 2054bd670b35SErik Nordmark if (ire->ire_ipversion == IPV4_VERSION) { 2055bd670b35SErik Nordmark alt_ire = ire_route_recursive_v4(ire->ire_addr, ire_type, 20569e3469d3SErik Nordmark ill, zoneid, tsl, match_flags, IRR_ALLOCATE, 0, ipst, NULL, 20579e3469d3SErik Nordmark NULL, &generation); 2058bd670b35SErik Nordmark } else { 2059bd670b35SErik Nordmark alt_ire = ire_route_recursive_v6(&ire->ire_addr_v6, ire_type, 20609e3469d3SErik Nordmark ill, zoneid, tsl, match_flags, IRR_ALLOCATE, 0, ipst, NULL, 20619e3469d3SErik Nordmark NULL, &generation); 2062bd670b35SErik Nordmark } 2063bd670b35SErik Nordmark ASSERT(alt_ire != NULL); 2064bd670b35SErik Nordmark 2065bd670b35SErik Nordmark if (alt_ire->ire_ill == ire->ire_ill) { 2066bd670b35SErik Nordmark /* Going out the same ILL - ok to send to IRE_LOCAL */ 2067bd670b35SErik Nordmark ire_refrele(alt_ire); 2068bd670b35SErik Nordmark } else { 2069bd670b35SErik Nordmark /* Different ill - ignore IRE_LOCAL */ 2070bd670b35SErik Nordmark ire_refrele(ire); 2071bd670b35SErik Nordmark ire = alt_ire; 2072bd670b35SErik Nordmark if (generationp != NULL) 2073bd670b35SErik Nordmark *generationp = generation; 2074bd670b35SErik Nordmark } 2075bd670b35SErik Nordmark return (ire); 2076bd670b35SErik Nordmark } 2077bd670b35SErik Nordmark 2078bd670b35SErik Nordmark boolean_t 2079bd670b35SErik Nordmark ire_find_zoneid(struct radix_node *rn, void *arg) 2080bd670b35SErik Nordmark { 2081bd670b35SErik Nordmark struct rt_entry *rt = (struct rt_entry *)rn; 2082bd670b35SErik Nordmark irb_t *irb; 2083bd670b35SErik Nordmark ire_t *ire; 2084bd670b35SErik Nordmark ire_ftable_args_t *margs = arg; 2085bd670b35SErik Nordmark 2086bd670b35SErik Nordmark ASSERT(rt != NULL); 2087bd670b35SErik Nordmark 2088bd670b35SErik Nordmark irb = &rt->rt_irb; 2089bd670b35SErik Nordmark 2090bd670b35SErik Nordmark if (irb->irb_ire_cnt == 0) 2091bd670b35SErik Nordmark return (B_FALSE); 2092bd670b35SErik Nordmark 2093bd670b35SErik Nordmark rw_enter(&irb->irb_lock, RW_READER); 2094bd670b35SErik Nordmark for (ire = irb->irb_ire; ire != NULL; ire = ire->ire_next) { 2095bd670b35SErik Nordmark if (IRE_IS_CONDEMNED(ire)) 2096bd670b35SErik Nordmark continue; 2097bd670b35SErik Nordmark 2098188e1664SErik Nordmark if (!(ire->ire_type & IRE_INTERFACE)) 2099188e1664SErik Nordmark continue; 2100188e1664SErik Nordmark 2101bd670b35SErik Nordmark if (ire->ire_zoneid != ALL_ZONES && 2102bd670b35SErik Nordmark ire->ire_zoneid != margs->ift_zoneid) 2103bd670b35SErik Nordmark continue; 2104bd670b35SErik Nordmark 2105bd670b35SErik Nordmark if (margs->ift_ill != NULL && margs->ift_ill != ire->ire_ill) 2106bd670b35SErik Nordmark continue; 2107bd670b35SErik Nordmark 2108bd670b35SErik Nordmark if (is_system_labeled() && 2109bd670b35SErik Nordmark tsol_ire_match_gwattr(ire, margs->ift_tsl) != 0) 2110bd670b35SErik Nordmark continue; 2111bd670b35SErik Nordmark 2112bd670b35SErik Nordmark rw_exit(&irb->irb_lock); 2113bd670b35SErik Nordmark return (B_TRUE); 2114bd670b35SErik Nordmark } 2115bd670b35SErik Nordmark rw_exit(&irb->irb_lock); 2116bd670b35SErik Nordmark return (B_FALSE); 2117bd670b35SErik Nordmark } 2118bd670b35SErik Nordmark 2119bd670b35SErik Nordmark /* 2120bd670b35SErik Nordmark * Check if the zoneid (not ALL_ZONES) has an IRE_INTERFACE for the specified 2121bd670b35SErik Nordmark * gateway address. If ill is non-NULL we also match on it. 2122bd670b35SErik Nordmark * The caller must hold a read lock on RADIX_NODE_HEAD if lock_held is set. 21235597b60aSnordmark */ 21245597b60aSnordmark boolean_t 2125bd670b35SErik Nordmark ire_gateway_ok_zone_v4(ipaddr_t gateway, zoneid_t zoneid, ill_t *ill, 2126bd670b35SErik Nordmark const ts_label_t *tsl, ip_stack_t *ipst, boolean_t lock_held) 21275597b60aSnordmark { 2128bd670b35SErik Nordmark struct rt_sockaddr rdst; 2129bd670b35SErik Nordmark struct rt_entry *rt; 2130bd670b35SErik Nordmark ire_ftable_args_t margs; 2131e11c3f44Smeem 2132bd670b35SErik Nordmark ASSERT(ill == NULL || !ill->ill_isv6); 2133bd670b35SErik Nordmark if (lock_held) 2134bd670b35SErik Nordmark ASSERT(RW_READ_HELD(&ipst->ips_ip_ftable->rnh_lock)); 21357c478bd9Sstevel@tonic-gate else 2136bd670b35SErik Nordmark RADIX_NODE_HEAD_RLOCK(ipst->ips_ip_ftable); 2137bd670b35SErik Nordmark 2138188e1664SErik Nordmark bzero(&rdst, sizeof (rdst)); 2139bd670b35SErik Nordmark rdst.rt_sin_len = sizeof (rdst); 2140bd670b35SErik Nordmark rdst.rt_sin_family = AF_INET; 2141bd670b35SErik Nordmark rdst.rt_sin_addr.s_addr = gateway; 2142bd670b35SErik Nordmark 2143bd670b35SErik Nordmark /* 2144bd670b35SErik Nordmark * We only use margs for ill, zoneid, and tsl matching in 2145bd670b35SErik Nordmark * ire_find_zoneid 2146bd670b35SErik Nordmark */ 2147188e1664SErik Nordmark bzero(&margs, sizeof (margs)); 2148bd670b35SErik Nordmark margs.ift_ill = ill; 2149bd670b35SErik Nordmark margs.ift_zoneid = zoneid; 2150bd670b35SErik Nordmark margs.ift_tsl = tsl; 2151bd670b35SErik Nordmark rt = (struct rt_entry *)ipst->ips_ip_ftable->rnh_matchaddr_args(&rdst, 2152bd670b35SErik Nordmark ipst->ips_ip_ftable, ire_find_zoneid, (void *)&margs); 2153bd670b35SErik Nordmark 2154bd670b35SErik Nordmark if (!lock_held) 2155bd670b35SErik Nordmark RADIX_NODE_HEAD_UNLOCK(ipst->ips_ip_ftable); 2156bd670b35SErik Nordmark 2157bd670b35SErik Nordmark return (rt != NULL); 21587c478bd9Sstevel@tonic-gate } 21597c478bd9Sstevel@tonic-gate 21607c478bd9Sstevel@tonic-gate /* 2161bd670b35SErik Nordmark * ire_walk routine to delete a fraction of redirect IREs and IRE_CLONE_IF IREs. 2162bd670b35SErik Nordmark * The fraction argument tells us what fraction of the IREs to delete. 2163bd670b35SErik Nordmark * Common for IPv4 and IPv6. 2164bd670b35SErik Nordmark * Used when memory backpressure. 21657c478bd9Sstevel@tonic-gate */ 2166bd670b35SErik Nordmark static void 2167bd670b35SErik Nordmark ire_delete_reclaim(ire_t *ire, char *arg) 21687c478bd9Sstevel@tonic-gate { 2169bd670b35SErik Nordmark ip_stack_t *ipst = ire->ire_ipst; 2170bd670b35SErik Nordmark uint_t fraction = *(uint_t *)arg; 21717c478bd9Sstevel@tonic-gate uint_t rand; 21727c478bd9Sstevel@tonic-gate 2173bd670b35SErik Nordmark if ((ire->ire_flags & RTF_DYNAMIC) || 2174bd670b35SErik Nordmark (ire->ire_type & IRE_IF_CLONE)) { 21757c478bd9Sstevel@tonic-gate 2176bd670b35SErik Nordmark /* Pick a random number */ 2177d3d50737SRafael Vanoni rand = (uint_t)ddi_get_lbolt() + 2178bd670b35SErik Nordmark IRE_ADDR_HASH_V6(ire->ire_addr_v6, 256); 2179bd670b35SErik Nordmark 2180bd670b35SErik Nordmark /* Use truncation */ 2181bd670b35SErik Nordmark if ((rand/fraction)*fraction == rand) { 2182bd670b35SErik Nordmark IP_STAT(ipst, ip_ire_reclaim_deleted); 21837c478bd9Sstevel@tonic-gate ire_delete(ire); 21847c478bd9Sstevel@tonic-gate } 21857c478bd9Sstevel@tonic-gate } 2186bd670b35SErik Nordmark 21877c478bd9Sstevel@tonic-gate } 2188bd670b35SErik Nordmark 21897c478bd9Sstevel@tonic-gate /* 2190bd670b35SErik Nordmark * kmem_cache callback to free up memory. 2191bd670b35SErik Nordmark * 2192bd670b35SErik Nordmark * Free a fraction (ips_ip_ire_reclaim_fraction) of things IP added dynamically 2193bd670b35SErik Nordmark * (RTF_DYNAMIC and IRE_IF_CLONE). 21947c478bd9Sstevel@tonic-gate */ 2195bd670b35SErik Nordmark static void 2196bd670b35SErik Nordmark ip_ire_reclaim_stack(ip_stack_t *ipst) 2197bd670b35SErik Nordmark { 2198bd670b35SErik Nordmark uint_t fraction = ipst->ips_ip_ire_reclaim_fraction; 2199bd670b35SErik Nordmark 2200bd670b35SErik Nordmark IP_STAT(ipst, ip_ire_reclaim_calls); 2201bd670b35SErik Nordmark 2202bd670b35SErik Nordmark ire_walk(ire_delete_reclaim, &fraction, ipst); 2203bd670b35SErik Nordmark 2204bd670b35SErik Nordmark /* 2205bd670b35SErik Nordmark * Walk all CONNs that can have a reference on an ire, nce or dce. 2206bd670b35SErik Nordmark * Get them to update any stale references to drop any refholds they 2207bd670b35SErik Nordmark * have. 2208bd670b35SErik Nordmark */ 2209bd670b35SErik Nordmark ipcl_walk(conn_ixa_cleanup, (void *)B_FALSE, ipst); 2210bd670b35SErik Nordmark } 2211bd670b35SErik Nordmark 2212bd670b35SErik Nordmark /* 2213bd670b35SErik Nordmark * Called by the memory allocator subsystem directly, when the system 2214bd670b35SErik Nordmark * is running low on memory. 2215bd670b35SErik Nordmark */ 2216bd670b35SErik Nordmark /* ARGSUSED */ 2217bd670b35SErik Nordmark void 2218bd670b35SErik Nordmark ip_ire_reclaim(void *args) 2219bd670b35SErik Nordmark { 2220bd670b35SErik Nordmark netstack_handle_t nh; 2221bd670b35SErik Nordmark netstack_t *ns; 22224ba231ceSKacheong Poon ip_stack_t *ipst; 2223bd670b35SErik Nordmark 2224bd670b35SErik Nordmark netstack_next_init(&nh); 2225bd670b35SErik Nordmark while ((ns = netstack_next(&nh)) != NULL) { 22264ba231ceSKacheong Poon /* 22274ba231ceSKacheong Poon * netstack_next() can return a netstack_t with a NULL 22284ba231ceSKacheong Poon * netstack_ip at boot time. 22294ba231ceSKacheong Poon */ 22304ba231ceSKacheong Poon if ((ipst = ns->netstack_ip) == NULL) { 22314ba231ceSKacheong Poon netstack_rele(ns); 22324ba231ceSKacheong Poon continue; 22334ba231ceSKacheong Poon } 22344ba231ceSKacheong Poon ip_ire_reclaim_stack(ipst); 2235bd670b35SErik Nordmark netstack_rele(ns); 2236bd670b35SErik Nordmark } 2237bd670b35SErik Nordmark netstack_next_fini(&nh); 22387c478bd9Sstevel@tonic-gate } 22397c478bd9Sstevel@tonic-gate 22407c478bd9Sstevel@tonic-gate static void 22417c478bd9Sstevel@tonic-gate power2_roundup(uint32_t *value) 22427c478bd9Sstevel@tonic-gate { 22437c478bd9Sstevel@tonic-gate int i; 22447c478bd9Sstevel@tonic-gate 22457c478bd9Sstevel@tonic-gate for (i = 1; i < 31; i++) { 22467c478bd9Sstevel@tonic-gate if (*value <= (1 << i)) 22477c478bd9Sstevel@tonic-gate break; 22487c478bd9Sstevel@tonic-gate } 22497c478bd9Sstevel@tonic-gate *value = (1 << i); 22507c478bd9Sstevel@tonic-gate } 22517c478bd9Sstevel@tonic-gate 2252f4b3ec61Sdh155122 /* Global init for all zones */ 22537c478bd9Sstevel@tonic-gate void 2254f4b3ec61Sdh155122 ip_ire_g_init() 22557c478bd9Sstevel@tonic-gate { 22567c478bd9Sstevel@tonic-gate /* 2257bd670b35SErik Nordmark * Create kmem_caches. ip_ire_reclaim() and ip_nce_reclaim() 2258bd670b35SErik Nordmark * will give disposable IREs back to system when needed. 22597c478bd9Sstevel@tonic-gate * This needs to be done here before anything else, since 22607c478bd9Sstevel@tonic-gate * ire_add() expects the cache to be created. 22617c478bd9Sstevel@tonic-gate */ 22627c478bd9Sstevel@tonic-gate ire_cache = kmem_cache_create("ire_cache", 2263bd670b35SErik Nordmark sizeof (ire_t), 0, NULL, NULL, 2264bd670b35SErik Nordmark ip_ire_reclaim, NULL, NULL, 0); 2265bd670b35SErik Nordmark 2266bd670b35SErik Nordmark ncec_cache = kmem_cache_create("ncec_cache", 2267bd670b35SErik Nordmark sizeof (ncec_t), 0, NULL, NULL, 2268bd670b35SErik Nordmark ip_nce_reclaim, NULL, NULL, 0); 2269bd670b35SErik Nordmark nce_cache = kmem_cache_create("nce_cache", 2270bd670b35SErik Nordmark sizeof (nce_t), 0, NULL, NULL, 2271bd670b35SErik Nordmark NULL, NULL, NULL, 0); 22727c478bd9Sstevel@tonic-gate 2273f4b3ec61Sdh155122 rt_entry_cache = kmem_cache_create("rt_entry", 2274f4b3ec61Sdh155122 sizeof (struct rt_entry), 0, NULL, NULL, NULL, NULL, NULL, 0); 2275f4b3ec61Sdh155122 2276f4b3ec61Sdh155122 /* 2277f4b3ec61Sdh155122 * Have radix code setup kmem caches etc. 2278f4b3ec61Sdh155122 */ 2279f4b3ec61Sdh155122 rn_init(); 2280f4b3ec61Sdh155122 } 2281f4b3ec61Sdh155122 2282f4b3ec61Sdh155122 void 2283f4b3ec61Sdh155122 ip_ire_init(ip_stack_t *ipst) 2284f4b3ec61Sdh155122 { 2285bd670b35SErik Nordmark ire_t *ire; 2286bd670b35SErik Nordmark int error; 2287f4b3ec61Sdh155122 2288f4b3ec61Sdh155122 mutex_init(&ipst->ips_ire_ft_init_lock, NULL, MUTEX_DEFAULT, 0); 2289f4b3ec61Sdh155122 2290f4b3ec61Sdh155122 (void) rn_inithead((void **)&ipst->ips_ip_ftable, 32); 2291f4b3ec61Sdh155122 22927c478bd9Sstevel@tonic-gate /* 22937c478bd9Sstevel@tonic-gate * Make sure that the forwarding table size is a power of 2. 22947c478bd9Sstevel@tonic-gate * The IRE*_ADDR_HASH() macroes depend on that. 22957c478bd9Sstevel@tonic-gate */ 2296f4b3ec61Sdh155122 ipst->ips_ip6_ftable_hash_size = ip6_ftable_hash_size; 2297f4b3ec61Sdh155122 power2_roundup(&ipst->ips_ip6_ftable_hash_size); 2298f4b3ec61Sdh155122 2299bd670b35SErik Nordmark /* 2300bd670b35SErik Nordmark * Allocate/initialize a pair of IRE_NOROUTEs for each of IPv4 and IPv6. 2301bd670b35SErik Nordmark * The ire_reject_v* has RTF_REJECT set, and the ire_blackhole_v* has 2302bd670b35SErik Nordmark * RTF_BLACKHOLE set. We use the latter for transient errors such 2303bd670b35SErik Nordmark * as memory allocation failures and tripping on IRE_IS_CONDEMNED 2304bd670b35SErik Nordmark * entries. 2305bd670b35SErik Nordmark */ 2306bd670b35SErik Nordmark ire = kmem_cache_alloc(ire_cache, KM_SLEEP); 2307bd670b35SErik Nordmark *ire = ire_null; 2308bd670b35SErik Nordmark error = ire_init_v4(ire, 0, 0, 0, IRE_NOROUTE, NULL, ALL_ZONES, 2309bd670b35SErik Nordmark RTF_REJECT|RTF_UP, NULL, ipst); 2310bd670b35SErik Nordmark ASSERT(error == 0); 2311bd670b35SErik Nordmark ipst->ips_ire_reject_v4 = ire; 2312bd670b35SErik Nordmark 2313bd670b35SErik Nordmark ire = kmem_cache_alloc(ire_cache, KM_SLEEP); 2314bd670b35SErik Nordmark *ire = ire_null; 2315bd670b35SErik Nordmark error = ire_init_v6(ire, 0, 0, 0, IRE_NOROUTE, NULL, ALL_ZONES, 2316bd670b35SErik Nordmark RTF_REJECT|RTF_UP, NULL, ipst); 2317bd670b35SErik Nordmark ASSERT(error == 0); 2318bd670b35SErik Nordmark ipst->ips_ire_reject_v6 = ire; 2319bd670b35SErik Nordmark 2320bd670b35SErik Nordmark ire = kmem_cache_alloc(ire_cache, KM_SLEEP); 2321bd670b35SErik Nordmark *ire = ire_null; 2322bd670b35SErik Nordmark error = ire_init_v4(ire, 0, 0, 0, IRE_NOROUTE, NULL, ALL_ZONES, 2323bd670b35SErik Nordmark RTF_BLACKHOLE|RTF_UP, NULL, ipst); 2324bd670b35SErik Nordmark ASSERT(error == 0); 2325bd670b35SErik Nordmark ipst->ips_ire_blackhole_v4 = ire; 2326bd670b35SErik Nordmark 2327bd670b35SErik Nordmark ire = kmem_cache_alloc(ire_cache, KM_SLEEP); 2328bd670b35SErik Nordmark *ire = ire_null; 2329bd670b35SErik Nordmark error = ire_init_v6(ire, 0, 0, 0, IRE_NOROUTE, NULL, ALL_ZONES, 2330bd670b35SErik Nordmark RTF_BLACKHOLE|RTF_UP, NULL, ipst); 2331bd670b35SErik Nordmark ASSERT(error == 0); 2332bd670b35SErik Nordmark ipst->ips_ire_blackhole_v6 = ire; 2333bd670b35SErik Nordmark 2334bd670b35SErik Nordmark rw_init(&ipst->ips_ip6_ire_head_lock, NULL, RW_DEFAULT, NULL); 2335bd670b35SErik Nordmark rw_init(&ipst->ips_ire_dep_lock, NULL, RW_DEFAULT, NULL); 23367c478bd9Sstevel@tonic-gate } 23377c478bd9Sstevel@tonic-gate 23387c478bd9Sstevel@tonic-gate void 2339f4b3ec61Sdh155122 ip_ire_g_fini(void) 2340f4b3ec61Sdh155122 { 2341f4b3ec61Sdh155122 kmem_cache_destroy(ire_cache); 2342bd670b35SErik Nordmark kmem_cache_destroy(ncec_cache); 2343bd670b35SErik Nordmark kmem_cache_destroy(nce_cache); 2344f4b3ec61Sdh155122 kmem_cache_destroy(rt_entry_cache); 2345f4b3ec61Sdh155122 2346f4b3ec61Sdh155122 rn_fini(); 2347f4b3ec61Sdh155122 } 2348f4b3ec61Sdh155122 2349f4b3ec61Sdh155122 void 2350f4b3ec61Sdh155122 ip_ire_fini(ip_stack_t *ipst) 23517c478bd9Sstevel@tonic-gate { 23527c478bd9Sstevel@tonic-gate int i; 23537c478bd9Sstevel@tonic-gate 2354bd670b35SErik Nordmark rw_destroy(&ipst->ips_ire_dep_lock); 2355bd670b35SErik Nordmark rw_destroy(&ipst->ips_ip6_ire_head_lock); 2356bd670b35SErik Nordmark 2357b8d97ac6SThirumalai Srinivasan ire_make_condemned(ipst->ips_ire_reject_v6); 2358bd670b35SErik Nordmark ire_refrele_notr(ipst->ips_ire_reject_v6); 2359bd670b35SErik Nordmark ipst->ips_ire_reject_v6 = NULL; 2360b8d97ac6SThirumalai Srinivasan 2361b8d97ac6SThirumalai Srinivasan ire_make_condemned(ipst->ips_ire_reject_v4); 2362bd670b35SErik Nordmark ire_refrele_notr(ipst->ips_ire_reject_v4); 2363bd670b35SErik Nordmark ipst->ips_ire_reject_v4 = NULL; 2364b8d97ac6SThirumalai Srinivasan 2365b8d97ac6SThirumalai Srinivasan ire_make_condemned(ipst->ips_ire_blackhole_v6); 2366bd670b35SErik Nordmark ire_refrele_notr(ipst->ips_ire_blackhole_v6); 2367bd670b35SErik Nordmark ipst->ips_ire_blackhole_v6 = NULL; 2368b8d97ac6SThirumalai Srinivasan 2369b8d97ac6SThirumalai Srinivasan ire_make_condemned(ipst->ips_ire_blackhole_v4); 2370bd670b35SErik Nordmark ire_refrele_notr(ipst->ips_ire_blackhole_v4); 2371bd670b35SErik Nordmark ipst->ips_ire_blackhole_v4 = NULL; 2372bd670b35SErik Nordmark 2373f4b3ec61Sdh155122 /* 2374f4b3ec61Sdh155122 * Delete all IREs - assumes that the ill/ipifs have 2375bd670b35SErik Nordmark * been removed so what remains are just the ftable to handle. 2376f4b3ec61Sdh155122 */ 2377f4b3ec61Sdh155122 ire_walk(ire_delete, NULL, ipst); 2378c793af95Ssangeeta 2379f4b3ec61Sdh155122 rn_freehead(ipst->ips_ip_ftable); 2380f4b3ec61Sdh155122 ipst->ips_ip_ftable = NULL; 23817c478bd9Sstevel@tonic-gate 2382f4b3ec61Sdh155122 mutex_destroy(&ipst->ips_ire_ft_init_lock); 23837c478bd9Sstevel@tonic-gate 2384f4b3ec61Sdh155122 for (i = 0; i < IP6_MASK_TABLE_SIZE; i++) { 2385f4b3ec61Sdh155122 irb_t *ptr; 2386f4b3ec61Sdh155122 int j; 2387f4b3ec61Sdh155122 2388f4b3ec61Sdh155122 if ((ptr = ipst->ips_ip_forwarding_table_v6[i]) == NULL) 2389f4b3ec61Sdh155122 continue; 2390f4b3ec61Sdh155122 2391f4b3ec61Sdh155122 for (j = 0; j < ipst->ips_ip6_ftable_hash_size; j++) { 2392f4b3ec61Sdh155122 ASSERT(ptr[j].irb_ire == NULL); 2393f4b3ec61Sdh155122 rw_destroy(&ptr[j].irb_lock); 2394f4b3ec61Sdh155122 } 2395f4b3ec61Sdh155122 mi_free(ptr); 2396f4b3ec61Sdh155122 ipst->ips_ip_forwarding_table_v6[i] = NULL; 2397f4b3ec61Sdh155122 } 23987c478bd9Sstevel@tonic-gate } 23997c478bd9Sstevel@tonic-gate 24006a8288c7Scarlsonj #ifdef DEBUG 24017c478bd9Sstevel@tonic-gate void 24027c478bd9Sstevel@tonic-gate ire_trace_ref(ire_t *ire) 24037c478bd9Sstevel@tonic-gate { 24047c478bd9Sstevel@tonic-gate mutex_enter(&ire->ire_lock); 24056a8288c7Scarlsonj if (ire->ire_trace_disable) { 24067c478bd9Sstevel@tonic-gate mutex_exit(&ire->ire_lock); 24077c478bd9Sstevel@tonic-gate return; 24087c478bd9Sstevel@tonic-gate } 24096a8288c7Scarlsonj 24106a8288c7Scarlsonj if (th_trace_ref(ire, ire->ire_ipst)) { 24116a8288c7Scarlsonj mutex_exit(&ire->ire_lock); 24126a8288c7Scarlsonj } else { 24137c478bd9Sstevel@tonic-gate ire->ire_trace_disable = B_TRUE; 24147c478bd9Sstevel@tonic-gate mutex_exit(&ire->ire_lock); 24156a8288c7Scarlsonj ire_trace_cleanup(ire); 24167c478bd9Sstevel@tonic-gate } 24177c478bd9Sstevel@tonic-gate } 24187c478bd9Sstevel@tonic-gate 24197c478bd9Sstevel@tonic-gate void 24207c478bd9Sstevel@tonic-gate ire_untrace_ref(ire_t *ire) 24217c478bd9Sstevel@tonic-gate { 24227c478bd9Sstevel@tonic-gate mutex_enter(&ire->ire_lock); 24236a8288c7Scarlsonj if (!ire->ire_trace_disable) 24246a8288c7Scarlsonj th_trace_unref(ire); 24257c478bd9Sstevel@tonic-gate mutex_exit(&ire->ire_lock); 24267c478bd9Sstevel@tonic-gate } 24277c478bd9Sstevel@tonic-gate 24287c478bd9Sstevel@tonic-gate static void 24296a8288c7Scarlsonj ire_trace_cleanup(const ire_t *ire) 24307c478bd9Sstevel@tonic-gate { 24316a8288c7Scarlsonj th_trace_cleanup(ire, ire->ire_trace_disable); 24327c478bd9Sstevel@tonic-gate } 24336a8288c7Scarlsonj #endif /* DEBUG */ 2434c793af95Ssangeeta 2435c793af95Ssangeeta /* 2436bd670b35SErik Nordmark * Find, or create if needed, the nce_t pointer to the neighbor cache 2437bd670b35SErik Nordmark * entry ncec_t for an IPv4 address. The nce_t will be created on the ill_t 2438bd670b35SErik Nordmark * in the non-IPMP case, or on the cast-ill in the IPMP bcast/mcast case, or 2439bd670b35SErik Nordmark * on the next available under-ill (selected by the IPMP rotor) in the 2440bd670b35SErik Nordmark * unicast IPMP case. 244154da8755Ssowmini * 244254da8755Ssowmini * If a neighbor-cache entry has to be created (i.e., one does not already 2443bd670b35SErik Nordmark * exist in the nce list) the ncec_lladdr and ncec_state of the neighbor cache 2444bd670b35SErik Nordmark * entry are initialized in nce_add_v4(). The broadcast, multicast, and 2445bd670b35SErik Nordmark * link-layer type determine the contents of {ncec_state, ncec_lladdr} of 2446bd670b35SErik Nordmark * the ncec_t created. The ncec_lladdr is non-null for all link types with 2447bd670b35SErik Nordmark * non-zero ill_phys_addr_length, though the contents may be zero in cases 2448bd670b35SErik Nordmark * where the link-layer type is not known at the time of creation 2449bd670b35SErik Nordmark * (e.g., IRE_IFRESOLVER links) 2450bd670b35SErik Nordmark * 2451bd670b35SErik Nordmark * All IRE_BROADCAST entries have ncec_state = ND_REACHABLE, and the nce_lladr 2452bd670b35SErik Nordmark * has the physical broadcast address of the outgoing interface. 2453bd670b35SErik Nordmark * For unicast ire entries, 245454da8755Ssowmini * - if the outgoing interface is of type IRE_IF_RESOLVER, a newly created 2455bd670b35SErik Nordmark * ncec_t with 0 nce_lladr contents, and will be in the ND_INITIAL state. 245654da8755Ssowmini * - if the outgoing interface is a IRE_IF_NORESOLVER interface, no link 2457bd670b35SErik Nordmark * layer resolution is necessary, so that the ncec_t will be in the 2458bd670b35SErik Nordmark * ND_REACHABLE state 245954da8755Ssowmini * 246054da8755Ssowmini * The link layer information needed for broadcast addresses, and for 246154da8755Ssowmini * packets sent on IRE_IF_NORESOLVER interfaces is a constant mapping that 2462bd670b35SErik Nordmark * never needs re-verification for the lifetime of the ncec_t. These are 2463bd670b35SErik Nordmark * therefore marked NCE_F_NONUD. 246454da8755Ssowmini * 2465bd670b35SErik Nordmark * The nce returned will be created such that the nce_ill == ill that 2466bd670b35SErik Nordmark * is passed in. Note that the nce itself may not have ncec_ill == ill 2467bd670b35SErik Nordmark * where IPMP links are involved. 2468c793af95Ssangeeta */ 2469bd670b35SErik Nordmark static nce_t * 2470bd670b35SErik Nordmark ire_nce_init(ill_t *ill, const void *addr, int ire_type) 2471c793af95Ssangeeta { 2472c793af95Ssangeeta int err; 247354da8755Ssowmini nce_t *nce = NULL; 2474bd670b35SErik Nordmark uint16_t ncec_flags; 2475bd670b35SErik Nordmark uchar_t *hwaddr; 2476bd670b35SErik Nordmark boolean_t need_refrele = B_FALSE; 2477bd670b35SErik Nordmark ill_t *in_ill = ill; 2478bd670b35SErik Nordmark boolean_t is_unicast; 2479bd670b35SErik Nordmark uint_t hwaddr_len; 2480c793af95Ssangeeta 2481bd670b35SErik Nordmark is_unicast = ((ire_type & (IRE_MULTICAST|IRE_BROADCAST)) == 0); 2482bd670b35SErik Nordmark if (IS_IPMP(ill) || 2483bd670b35SErik Nordmark ((ire_type & IRE_BROADCAST) && IS_UNDER_IPMP(ill))) { 2484bd670b35SErik Nordmark if ((ill = ipmp_ill_get_xmit_ill(ill, is_unicast)) == NULL) 2485bd670b35SErik Nordmark return (NULL); 2486bd670b35SErik Nordmark need_refrele = B_TRUE; 2487bd670b35SErik Nordmark } 2488bd670b35SErik Nordmark ncec_flags = (ill->ill_flags & ILLF_NONUD) ? NCE_F_NONUD : 0; 2489c793af95Ssangeeta 2490bd670b35SErik Nordmark switch (ire_type) { 2491c793af95Ssangeeta case IRE_BROADCAST: 2492bd670b35SErik Nordmark ASSERT(!ill->ill_isv6); 2493bd670b35SErik Nordmark ncec_flags |= (NCE_F_BCAST|NCE_F_NONUD); 2494c793af95Ssangeeta break; 2495bd670b35SErik Nordmark case IRE_MULTICAST: 2496bd670b35SErik Nordmark ncec_flags |= (NCE_F_MCAST|NCE_F_NONUD); 2497bd670b35SErik Nordmark break; 2498c793af95Ssangeeta } 2499c793af95Ssangeeta 2500bd670b35SErik Nordmark if (ill->ill_net_type == IRE_IF_NORESOLVER && is_unicast) { 2501bd670b35SErik Nordmark hwaddr = ill->ill_dest_addr; 2502bd670b35SErik Nordmark } else { 2503bd670b35SErik Nordmark hwaddr = NULL; 2504b9c344b3Ssowmini } 2505bd670b35SErik Nordmark hwaddr_len = ill->ill_phys_addr_length; 2506b9c344b3Ssowmini 2507bd670b35SErik Nordmark retry: 2508bd670b35SErik Nordmark /* nce_state will be computed by nce_add_common() */ 2509bd670b35SErik Nordmark if (!ill->ill_isv6) { 2510bd670b35SErik Nordmark err = nce_lookup_then_add_v4(ill, hwaddr, hwaddr_len, addr, 2511bd670b35SErik Nordmark ncec_flags, ND_UNCHANGED, &nce); 2512bd670b35SErik Nordmark } else { 2513bd670b35SErik Nordmark err = nce_lookup_then_add_v6(ill, hwaddr, hwaddr_len, addr, 2514bd670b35SErik Nordmark ncec_flags, ND_UNCHANGED, &nce); 2515bd670b35SErik Nordmark } 2516c793af95Ssangeeta 2517c793af95Ssangeeta switch (err) { 2518c793af95Ssangeeta case 0: 2519bd670b35SErik Nordmark break; 2520c793af95Ssangeeta case EEXIST: 2521c793af95Ssangeeta /* 2522bd670b35SErik Nordmark * When subnets change or partially overlap what was once 2523bd670b35SErik Nordmark * a broadcast address could now be a unicast, or vice versa. 2524c793af95Ssangeeta */ 2525bd670b35SErik Nordmark if (((ncec_flags ^ nce->nce_common->ncec_flags) & 2526bd670b35SErik Nordmark NCE_F_BCAST) != 0) { 2527bd670b35SErik Nordmark ASSERT(!ill->ill_isv6); 2528bd670b35SErik Nordmark ncec_delete(nce->nce_common); 2529bd670b35SErik Nordmark nce_refrele(nce); 2530bd670b35SErik Nordmark goto retry; 2531bd670b35SErik Nordmark } 2532c793af95Ssangeeta break; 2533c793af95Ssangeeta default: 2534bd670b35SErik Nordmark DTRACE_PROBE2(nce__init__fail, ill_t *, ill, int, err); 2535bd670b35SErik Nordmark if (need_refrele) 2536bd670b35SErik Nordmark ill_refrele(ill); 2537bd670b35SErik Nordmark return (NULL); 2538c793af95Ssangeeta } 25399294ad19SSowmini Varadhan /* 2540bd670b35SErik Nordmark * If the ill was an under-ill of an IPMP group, we need to verify 2541bd670b35SErik Nordmark * that it is still active so that we select an active interface in 2542bd670b35SErik Nordmark * the group. However, since ipmp_ill_is_active ASSERTs for 2543bd670b35SErik Nordmark * IS_UNDER_IPMP(), we first need to verify that the ill is an 2544bd670b35SErik Nordmark * under-ill, and since this is being done in the data path, the 2545bd670b35SErik Nordmark * only way to ascertain this is by holding the ill_g_lock. 25469294ad19SSowmini Varadhan */ 2547bd670b35SErik Nordmark rw_enter(&ill->ill_ipst->ips_ill_g_lock, RW_READER); 2548bd670b35SErik Nordmark mutex_enter(&ill->ill_lock); 2549bd670b35SErik Nordmark mutex_enter(&ill->ill_phyint->phyint_lock); 2550bd670b35SErik Nordmark if (need_refrele && IS_UNDER_IPMP(ill) && !ipmp_ill_is_active(ill)) { 2551c793af95Ssangeeta /* 2552bd670b35SErik Nordmark * need_refrele implies that the under ill was selected by 2553bd670b35SErik Nordmark * ipmp_ill_get_xmit_ill() because either the in_ill was an 2554bd670b35SErik Nordmark * ipmp_ill, or we are sending a non-unicast packet on 2555bd670b35SErik Nordmark * an under_ill. However, when we get here, the ill selected by 2556bd670b35SErik Nordmark * ipmp_ill_get_xmit_ill was pulled out of the active set 2557bd670b35SErik Nordmark * (for unicast) or cast_ill nomination (for 2558bd670b35SErik Nordmark * !unicast) after it was picked as the outgoing ill. 2559bd670b35SErik Nordmark * We have to pick an active interface and/or cast_ill in the 2560bd670b35SErik Nordmark * group. 2561c793af95Ssangeeta */ 2562bd670b35SErik Nordmark mutex_exit(&ill->ill_phyint->phyint_lock); 2563bd670b35SErik Nordmark nce_delete(nce); 2564bd670b35SErik Nordmark mutex_exit(&ill->ill_lock); 2565bd670b35SErik Nordmark rw_exit(&ill->ill_ipst->ips_ill_g_lock); 2566bd670b35SErik Nordmark nce_refrele(nce); 2567bd670b35SErik Nordmark ill_refrele(ill); 2568bd670b35SErik Nordmark if ((ill = ipmp_ill_get_xmit_ill(in_ill, is_unicast)) == NULL) 2569bd670b35SErik Nordmark return (NULL); 2570bd670b35SErik Nordmark goto retry; 2571c793af95Ssangeeta } else { 2572bd670b35SErik Nordmark mutex_exit(&ill->ill_phyint->phyint_lock); 2573bd670b35SErik Nordmark mutex_exit(&ill->ill_lock); 2574bd670b35SErik Nordmark rw_exit(&ill->ill_ipst->ips_ill_g_lock); 2575bd670b35SErik Nordmark } 2576bd670b35SErik Nordmark done: 2577bd670b35SErik Nordmark ASSERT(nce->nce_ill == ill); 2578bd670b35SErik Nordmark if (need_refrele) 2579bd670b35SErik Nordmark ill_refrele(ill); 2580bd670b35SErik Nordmark return (nce); 2581bd670b35SErik Nordmark } 2582bd670b35SErik Nordmark 2583bd670b35SErik Nordmark nce_t * 2584bd670b35SErik Nordmark arp_nce_init(ill_t *ill, in_addr_t addr4, int ire_type) 2585bd670b35SErik Nordmark { 2586bd670b35SErik Nordmark return (ire_nce_init(ill, &addr4, ire_type)); 2587bd670b35SErik Nordmark } 2588bd670b35SErik Nordmark 2589bd670b35SErik Nordmark nce_t * 2590bd670b35SErik Nordmark ndp_nce_init(ill_t *ill, const in6_addr_t *addr6, int ire_type) 2591bd670b35SErik Nordmark { 2592bd670b35SErik Nordmark ASSERT((ire_type & IRE_BROADCAST) == 0); 2593bd670b35SErik Nordmark return (ire_nce_init(ill, addr6, ire_type)); 2594bd670b35SErik Nordmark } 2595bd670b35SErik Nordmark 2596bd670b35SErik Nordmark /* 2597bd670b35SErik Nordmark * The caller should hold irb_lock as a writer if the ire is in a bucket. 2598f1c454b4SSowmini Varadhan * This routine will clear ire_nce_cache, and we make sure that we can never 2599f1c454b4SSowmini Varadhan * set ire_nce_cache after the ire is marked condemned. 2600bd670b35SErik Nordmark */ 2601bd670b35SErik Nordmark void 2602bd670b35SErik Nordmark ire_make_condemned(ire_t *ire) 2603bd670b35SErik Nordmark { 2604bd670b35SErik Nordmark ip_stack_t *ipst = ire->ire_ipst; 2605f1c454b4SSowmini Varadhan nce_t *nce; 2606bd670b35SErik Nordmark 2607bd670b35SErik Nordmark mutex_enter(&ire->ire_lock); 2608bd670b35SErik Nordmark ASSERT(ire->ire_bucket == NULL || 2609bd670b35SErik Nordmark RW_WRITE_HELD(&ire->ire_bucket->irb_lock)); 2610bd670b35SErik Nordmark ASSERT(!IRE_IS_CONDEMNED(ire)); 2611bd670b35SErik Nordmark ire->ire_generation = IRE_GENERATION_CONDEMNED; 2612bd670b35SErik Nordmark /* Count how many condemned ires for kmem_cache callback */ 2613bd670b35SErik Nordmark atomic_add_32(&ipst->ips_num_ire_condemned, 1); 2614f1c454b4SSowmini Varadhan nce = ire->ire_nce_cache; 2615f1c454b4SSowmini Varadhan ire->ire_nce_cache = NULL; 2616bd670b35SErik Nordmark mutex_exit(&ire->ire_lock); 2617f1c454b4SSowmini Varadhan if (nce != NULL) 2618f1c454b4SSowmini Varadhan nce_refrele(nce); 2619bd670b35SErik Nordmark } 2620bd670b35SErik Nordmark 2621bd670b35SErik Nordmark /* 2622bd670b35SErik Nordmark * Increment the generation avoiding the special condemned value 2623bd670b35SErik Nordmark */ 2624bd670b35SErik Nordmark void 2625bd670b35SErik Nordmark ire_increment_generation(ire_t *ire) 2626bd670b35SErik Nordmark { 2627bd670b35SErik Nordmark uint_t generation; 2628bd670b35SErik Nordmark 2629bd670b35SErik Nordmark mutex_enter(&ire->ire_lock); 2630bd670b35SErik Nordmark /* 2631bd670b35SErik Nordmark * Even though the caller has a hold it can't prevent a concurrent 2632bd670b35SErik Nordmark * ire_delete marking the IRE condemned 2633bd670b35SErik Nordmark */ 2634bd670b35SErik Nordmark if (!IRE_IS_CONDEMNED(ire)) { 2635bd670b35SErik Nordmark generation = ire->ire_generation + 1; 2636bd670b35SErik Nordmark if (generation == IRE_GENERATION_CONDEMNED) 2637bd670b35SErik Nordmark generation = IRE_GENERATION_INITIAL; 2638bd670b35SErik Nordmark ASSERT(generation != IRE_GENERATION_VERIFY); 2639bd670b35SErik Nordmark ire->ire_generation = generation; 2640bd670b35SErik Nordmark } 2641bd670b35SErik Nordmark mutex_exit(&ire->ire_lock); 2642bd670b35SErik Nordmark } 2643bd670b35SErik Nordmark 2644bd670b35SErik Nordmark /* 2645bd670b35SErik Nordmark * Increment ire_generation on all the IRE_MULTICASTs 2646bd670b35SErik Nordmark * Used when the default multicast interface (as determined by 2647bd670b35SErik Nordmark * ill_lookup_multicast) might have changed. 2648bd670b35SErik Nordmark * 2649bd670b35SErik Nordmark * That includes the zoneid, IFF_ flags, the IPv6 scope of the address, and 2650bd670b35SErik Nordmark * ill unplumb. 2651bd670b35SErik Nordmark */ 2652bd670b35SErik Nordmark void 2653bd670b35SErik Nordmark ire_increment_multicast_generation(ip_stack_t *ipst, boolean_t isv6) 2654bd670b35SErik Nordmark { 2655bd670b35SErik Nordmark ill_t *ill; 2656bd670b35SErik Nordmark ill_walk_context_t ctx; 2657bd670b35SErik Nordmark 2658bd670b35SErik Nordmark rw_enter(&ipst->ips_ill_g_lock, RW_READER); 2659bd670b35SErik Nordmark if (isv6) 2660bd670b35SErik Nordmark ill = ILL_START_WALK_V6(&ctx, ipst); 2661bd670b35SErik Nordmark else 2662bd670b35SErik Nordmark ill = ILL_START_WALK_V4(&ctx, ipst); 2663bd670b35SErik Nordmark for (; ill != NULL; ill = ill_next(&ctx, ill)) { 2664bd670b35SErik Nordmark if (ILL_IS_CONDEMNED(ill)) 2665bd670b35SErik Nordmark continue; 2666bd670b35SErik Nordmark if (ill->ill_ire_multicast != NULL) 2667bd670b35SErik Nordmark ire_increment_generation(ill->ill_ire_multicast); 2668bd670b35SErik Nordmark } 2669bd670b35SErik Nordmark rw_exit(&ipst->ips_ill_g_lock); 2670bd670b35SErik Nordmark } 2671bd670b35SErik Nordmark 2672bd670b35SErik Nordmark /* 2673bd670b35SErik Nordmark * Return a held IRE_NOROUTE with RTF_REJECT set 2674bd670b35SErik Nordmark */ 2675bd670b35SErik Nordmark ire_t * 2676bd670b35SErik Nordmark ire_reject(ip_stack_t *ipst, boolean_t isv6) 2677bd670b35SErik Nordmark { 2678bd670b35SErik Nordmark ire_t *ire; 2679bd670b35SErik Nordmark 2680bd670b35SErik Nordmark if (isv6) 2681bd670b35SErik Nordmark ire = ipst->ips_ire_reject_v6; 2682bd670b35SErik Nordmark else 2683bd670b35SErik Nordmark ire = ipst->ips_ire_reject_v4; 2684bd670b35SErik Nordmark 2685bd670b35SErik Nordmark ASSERT(ire->ire_generation != IRE_GENERATION_CONDEMNED); 2686bd670b35SErik Nordmark ire_refhold(ire); 2687bd670b35SErik Nordmark return (ire); 2688bd670b35SErik Nordmark } 2689bd670b35SErik Nordmark 2690bd670b35SErik Nordmark /* 2691bd670b35SErik Nordmark * Return a held IRE_NOROUTE with RTF_BLACKHOLE set 2692bd670b35SErik Nordmark */ 2693bd670b35SErik Nordmark ire_t * 2694bd670b35SErik Nordmark ire_blackhole(ip_stack_t *ipst, boolean_t isv6) 2695bd670b35SErik Nordmark { 2696bd670b35SErik Nordmark ire_t *ire; 2697bd670b35SErik Nordmark 2698bd670b35SErik Nordmark if (isv6) 2699bd670b35SErik Nordmark ire = ipst->ips_ire_blackhole_v6; 2700bd670b35SErik Nordmark else 2701bd670b35SErik Nordmark ire = ipst->ips_ire_blackhole_v4; 2702bd670b35SErik Nordmark 2703bd670b35SErik Nordmark ASSERT(ire->ire_generation != IRE_GENERATION_CONDEMNED); 2704bd670b35SErik Nordmark ire_refhold(ire); 2705bd670b35SErik Nordmark return (ire); 2706bd670b35SErik Nordmark } 2707bd670b35SErik Nordmark 2708bd670b35SErik Nordmark /* 2709bd670b35SErik Nordmark * Return a held IRE_MULTICAST. 2710bd670b35SErik Nordmark */ 2711bd670b35SErik Nordmark ire_t * 2712bd670b35SErik Nordmark ire_multicast(ill_t *ill) 2713bd670b35SErik Nordmark { 2714bd670b35SErik Nordmark ire_t *ire = ill->ill_ire_multicast; 2715bd670b35SErik Nordmark 2716bd670b35SErik Nordmark ASSERT(ire == NULL || ire->ire_generation != IRE_GENERATION_CONDEMNED); 2717bd670b35SErik Nordmark if (ire == NULL) 2718bd670b35SErik Nordmark ire = ire_blackhole(ill->ill_ipst, ill->ill_isv6); 2719bd670b35SErik Nordmark else 2720bd670b35SErik Nordmark ire_refhold(ire); 2721bd670b35SErik Nordmark return (ire); 2722bd670b35SErik Nordmark } 2723bd670b35SErik Nordmark 2724bd670b35SErik Nordmark /* 2725bd670b35SErik Nordmark * Given an IRE return its nexthop IRE. The nexthop IRE is an IRE_ONLINK 2726bd670b35SErik Nordmark * that is an exact match (i.e., a /32 for IPv4 and /128 for IPv6). 2727bd670b35SErik Nordmark * This can return an RTF_REJECT|RTF_BLACKHOLE. 2728bd670b35SErik Nordmark * The returned IRE is held. 2729bd670b35SErik Nordmark * The assumption is that ip_select_route() has been called and returned the 2730bd670b35SErik Nordmark * IRE (thus ip_select_route would have set up the ire_dep* information.) 2731bd670b35SErik Nordmark * If some IRE is deleteted then ire_dep_remove() will have been called and 2732bd670b35SErik Nordmark * we might not find a nexthop IRE, in which case we return NULL. 2733bd670b35SErik Nordmark */ 2734bd670b35SErik Nordmark ire_t * 2735bd670b35SErik Nordmark ire_nexthop(ire_t *ire) 2736bd670b35SErik Nordmark { 2737bd670b35SErik Nordmark ip_stack_t *ipst = ire->ire_ipst; 2738bd670b35SErik Nordmark 2739bd670b35SErik Nordmark /* Acquire lock to walk ire_dep_parent */ 2740bd670b35SErik Nordmark rw_enter(&ipst->ips_ire_dep_lock, RW_READER); 2741bd670b35SErik Nordmark while (ire != NULL) { 2742bd670b35SErik Nordmark if (ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) { 2743bd670b35SErik Nordmark goto done; 27449294ad19SSowmini Varadhan } 2745c793af95Ssangeeta /* 2746bd670b35SErik Nordmark * If we find an IRE_ONLINK we are done. This includes 2747bd670b35SErik Nordmark * the case of IRE_MULTICAST. 2748bd670b35SErik Nordmark * Note that in order to send packets we need a host-specific 2749bd670b35SErik Nordmark * IRE_IF_ALL first in the ire_dep_parent chain. Normally this 2750bd670b35SErik Nordmark * is done by inserting an IRE_IF_CLONE if the IRE_INTERFACE 2751bd670b35SErik Nordmark * was not host specific. 2752bd670b35SErik Nordmark * However, ip_rts_request doesn't want to send packets 2753bd670b35SErik Nordmark * hence doesn't want to allocate an IRE_IF_CLONE. Yet 2754bd670b35SErik Nordmark * it needs an IRE_IF_ALL to get to the ill. Thus 2755bd670b35SErik Nordmark * we return IRE_IF_ALL that are not host specific here. 2756c793af95Ssangeeta */ 2757bd670b35SErik Nordmark if (ire->ire_type & IRE_ONLINK) 2758bd670b35SErik Nordmark goto done; 2759bd670b35SErik Nordmark ire = ire->ire_dep_parent; 2760c793af95Ssangeeta } 2761bd670b35SErik Nordmark rw_exit(&ipst->ips_ire_dep_lock); 2762bd670b35SErik Nordmark return (NULL); 2763bd670b35SErik Nordmark 2764bd670b35SErik Nordmark done: 2765bd670b35SErik Nordmark ire_refhold(ire); 2766bd670b35SErik Nordmark rw_exit(&ipst->ips_ire_dep_lock); 2767bd670b35SErik Nordmark return (ire); 2768bd670b35SErik Nordmark } 2769bd670b35SErik Nordmark 2770bd670b35SErik Nordmark /* 2771bd670b35SErik Nordmark * Find the ill used to send packets. This will be NULL in case 2772bd670b35SErik Nordmark * of a reject or blackhole. 2773bd670b35SErik Nordmark * The returned ill is held; caller needs to do ill_refrele when done. 2774bd670b35SErik Nordmark */ 2775bd670b35SErik Nordmark ill_t * 2776bd670b35SErik Nordmark ire_nexthop_ill(ire_t *ire) 2777bd670b35SErik Nordmark { 2778bd670b35SErik Nordmark ill_t *ill; 2779bd670b35SErik Nordmark 2780bd670b35SErik Nordmark ire = ire_nexthop(ire); 2781bd670b35SErik Nordmark if (ire == NULL) 2782bd670b35SErik Nordmark return (NULL); 2783bd670b35SErik Nordmark 2784bd670b35SErik Nordmark /* ire_ill can not change for an existing ire */ 2785bd670b35SErik Nordmark ill = ire->ire_ill; 2786bd670b35SErik Nordmark if (ill != NULL) 2787bd670b35SErik Nordmark ill_refhold(ill); 2788bd670b35SErik Nordmark ire_refrele(ire); 2789bd670b35SErik Nordmark return (ill); 2790bd670b35SErik Nordmark } 2791bd670b35SErik Nordmark 2792bd670b35SErik Nordmark #ifdef DEBUG 2793bd670b35SErik Nordmark static boolean_t 2794bd670b35SErik Nordmark parent_has_child(ire_t *parent, ire_t *child) 2795bd670b35SErik Nordmark { 2796bd670b35SErik Nordmark ire_t *ire; 2797bd670b35SErik Nordmark ire_t *prev; 2798bd670b35SErik Nordmark 2799bd670b35SErik Nordmark ire = parent->ire_dep_children; 2800bd670b35SErik Nordmark prev = NULL; 2801bd670b35SErik Nordmark while (ire != NULL) { 2802bd670b35SErik Nordmark if (prev == NULL) { 2803bd670b35SErik Nordmark ASSERT(ire->ire_dep_sib_ptpn == 2804bd670b35SErik Nordmark &(parent->ire_dep_children)); 2805bd670b35SErik Nordmark } else { 2806bd670b35SErik Nordmark ASSERT(ire->ire_dep_sib_ptpn == 2807bd670b35SErik Nordmark &(prev->ire_dep_sib_next)); 2808bd670b35SErik Nordmark } 2809bd670b35SErik Nordmark if (ire == child) 2810bd670b35SErik Nordmark return (B_TRUE); 2811bd670b35SErik Nordmark prev = ire; 2812bd670b35SErik Nordmark ire = ire->ire_dep_sib_next; 2813bd670b35SErik Nordmark } 2814bd670b35SErik Nordmark return (B_FALSE); 2815bd670b35SErik Nordmark } 2816bd670b35SErik Nordmark 2817bd670b35SErik Nordmark static void 2818bd670b35SErik Nordmark ire_dep_verify(ire_t *ire) 2819bd670b35SErik Nordmark { 2820bd670b35SErik Nordmark ire_t *parent = ire->ire_dep_parent; 2821bd670b35SErik Nordmark ire_t *child = ire->ire_dep_children; 2822bd670b35SErik Nordmark 2823bd670b35SErik Nordmark ASSERT(ire->ire_ipversion == IPV4_VERSION || 2824bd670b35SErik Nordmark ire->ire_ipversion == IPV6_VERSION); 2825bd670b35SErik Nordmark if (parent != NULL) { 2826bd670b35SErik Nordmark ASSERT(parent->ire_ipversion == IPV4_VERSION || 2827bd670b35SErik Nordmark parent->ire_ipversion == IPV6_VERSION); 2828bd670b35SErik Nordmark ASSERT(parent->ire_refcnt >= 1); 2829bd670b35SErik Nordmark ASSERT(parent_has_child(parent, ire)); 2830bd670b35SErik Nordmark } 2831bd670b35SErik Nordmark if (child != NULL) { 2832bd670b35SErik Nordmark ASSERT(child->ire_ipversion == IPV4_VERSION || 2833bd670b35SErik Nordmark child->ire_ipversion == IPV6_VERSION); 2834bd670b35SErik Nordmark ASSERT(child->ire_dep_parent == ire); 2835bd670b35SErik Nordmark ASSERT(child->ire_dep_sib_ptpn != NULL); 2836bd670b35SErik Nordmark ASSERT(parent_has_child(ire, child)); 2837bd670b35SErik Nordmark } 2838bd670b35SErik Nordmark } 2839bd670b35SErik Nordmark #endif /* DEBUG */ 2840bd670b35SErik Nordmark 2841bd670b35SErik Nordmark /* 2842bd670b35SErik Nordmark * Assumes ire_dep_parent is set. Remove this child from its parent's linkage. 2843bd670b35SErik Nordmark */ 2844bd670b35SErik Nordmark void 2845bd670b35SErik Nordmark ire_dep_remove(ire_t *ire) 2846bd670b35SErik Nordmark { 2847bd670b35SErik Nordmark ip_stack_t *ipst = ire->ire_ipst; 2848bd670b35SErik Nordmark ire_t *parent = ire->ire_dep_parent; 2849bd670b35SErik Nordmark ire_t *next; 2850bd670b35SErik Nordmark nce_t *nce; 2851bd670b35SErik Nordmark 2852bd670b35SErik Nordmark ASSERT(RW_WRITE_HELD(&ipst->ips_ire_dep_lock)); 2853bd670b35SErik Nordmark ASSERT(ire->ire_dep_parent != NULL); 2854bd670b35SErik Nordmark ASSERT(ire->ire_dep_sib_ptpn != NULL); 2855bd670b35SErik Nordmark 2856bd670b35SErik Nordmark #ifdef DEBUG 2857bd670b35SErik Nordmark ire_dep_verify(ire); 2858bd670b35SErik Nordmark ire_dep_verify(parent); 2859bd670b35SErik Nordmark #endif 2860bd670b35SErik Nordmark 2861bd670b35SErik Nordmark next = ire->ire_dep_sib_next; 2862bd670b35SErik Nordmark if (next != NULL) 2863bd670b35SErik Nordmark next->ire_dep_sib_ptpn = ire->ire_dep_sib_ptpn; 2864bd670b35SErik Nordmark 2865bd670b35SErik Nordmark ASSERT(*(ire->ire_dep_sib_ptpn) == ire); 2866bd670b35SErik Nordmark *(ire->ire_dep_sib_ptpn) = ire->ire_dep_sib_next; 2867bd670b35SErik Nordmark 2868bd670b35SErik Nordmark ire->ire_dep_sib_ptpn = NULL; 2869bd670b35SErik Nordmark ire->ire_dep_sib_next = NULL; 2870bd670b35SErik Nordmark 2871bd670b35SErik Nordmark mutex_enter(&ire->ire_lock); 2872bd670b35SErik Nordmark parent = ire->ire_dep_parent; 2873bd670b35SErik Nordmark ire->ire_dep_parent = NULL; 2874bd670b35SErik Nordmark mutex_exit(&ire->ire_lock); 2875bd670b35SErik Nordmark 2876bd670b35SErik Nordmark /* 2877bd670b35SErik Nordmark * Make sure all our children, grandchildren, etc set 2878bd670b35SErik Nordmark * ire_dep_parent_generation to IRE_GENERATION_VERIFY since 2879bd670b35SErik Nordmark * we can no longer guarantee than the children have a current 2880bd670b35SErik Nordmark * ire_nce_cache and ire_nexthop_ill(). 2881bd670b35SErik Nordmark */ 2882bd670b35SErik Nordmark if (ire->ire_dep_children != NULL) 2883bd670b35SErik Nordmark ire_dep_invalidate_children(ire->ire_dep_children); 2884bd670b35SErik Nordmark 2885bd670b35SErik Nordmark /* 2886bd670b35SErik Nordmark * Since the parent is gone we make sure we clear ire_nce_cache. 2887bd670b35SErik Nordmark * We can clear it under ire_lock even if the IRE is used 2888bd670b35SErik Nordmark */ 2889bd670b35SErik Nordmark mutex_enter(&ire->ire_lock); 2890bd670b35SErik Nordmark nce = ire->ire_nce_cache; 2891bd670b35SErik Nordmark ire->ire_nce_cache = NULL; 2892bd670b35SErik Nordmark mutex_exit(&ire->ire_lock); 2893bd670b35SErik Nordmark if (nce != NULL) 2894bd670b35SErik Nordmark nce_refrele(nce); 2895bd670b35SErik Nordmark 2896bd670b35SErik Nordmark #ifdef DEBUG 2897bd670b35SErik Nordmark ire_dep_verify(ire); 2898bd670b35SErik Nordmark ire_dep_verify(parent); 2899bd670b35SErik Nordmark #endif 2900bd670b35SErik Nordmark 2901bd670b35SErik Nordmark ire_refrele_notr(parent); 2902bd670b35SErik Nordmark ire_refrele_notr(ire); 2903bd670b35SErik Nordmark } 2904bd670b35SErik Nordmark 2905bd670b35SErik Nordmark /* 2906bd670b35SErik Nordmark * Insert the child in the linkage of the parent 2907bd670b35SErik Nordmark */ 2908bd670b35SErik Nordmark static void 2909bd670b35SErik Nordmark ire_dep_parent_insert(ire_t *child, ire_t *parent) 2910bd670b35SErik Nordmark { 2911bd670b35SErik Nordmark ip_stack_t *ipst = child->ire_ipst; 2912bd670b35SErik Nordmark ire_t *next; 2913bd670b35SErik Nordmark 2914bd670b35SErik Nordmark ASSERT(RW_WRITE_HELD(&ipst->ips_ire_dep_lock)); 2915bd670b35SErik Nordmark ASSERT(child->ire_dep_parent == NULL); 2916bd670b35SErik Nordmark 2917bd670b35SErik Nordmark #ifdef DEBUG 2918bd670b35SErik Nordmark ire_dep_verify(child); 2919bd670b35SErik Nordmark ire_dep_verify(parent); 2920bd670b35SErik Nordmark #endif 2921bd670b35SErik Nordmark /* No parents => no siblings */ 2922bd670b35SErik Nordmark ASSERT(child->ire_dep_sib_ptpn == NULL); 2923bd670b35SErik Nordmark ASSERT(child->ire_dep_sib_next == NULL); 2924bd670b35SErik Nordmark 2925bd670b35SErik Nordmark ire_refhold_notr(parent); 2926bd670b35SErik Nordmark ire_refhold_notr(child); 2927bd670b35SErik Nordmark 2928bd670b35SErik Nordmark /* Head insertion */ 2929bd670b35SErik Nordmark next = parent->ire_dep_children; 2930bd670b35SErik Nordmark if (next != NULL) { 2931bd670b35SErik Nordmark ASSERT(next->ire_dep_sib_ptpn == &(parent->ire_dep_children)); 2932bd670b35SErik Nordmark child->ire_dep_sib_next = next; 2933bd670b35SErik Nordmark next->ire_dep_sib_ptpn = &(child->ire_dep_sib_next); 2934bd670b35SErik Nordmark } 2935bd670b35SErik Nordmark parent->ire_dep_children = child; 2936bd670b35SErik Nordmark child->ire_dep_sib_ptpn = &(parent->ire_dep_children); 2937bd670b35SErik Nordmark 2938bd670b35SErik Nordmark mutex_enter(&child->ire_lock); 2939bd670b35SErik Nordmark child->ire_dep_parent = parent; 2940bd670b35SErik Nordmark mutex_exit(&child->ire_lock); 2941bd670b35SErik Nordmark 2942bd670b35SErik Nordmark #ifdef DEBUG 2943bd670b35SErik Nordmark ire_dep_verify(child); 2944bd670b35SErik Nordmark ire_dep_verify(parent); 2945bd670b35SErik Nordmark #endif 2946bd670b35SErik Nordmark } 2947bd670b35SErik Nordmark 2948bd670b35SErik Nordmark 2949bd670b35SErik Nordmark /* 2950bd670b35SErik Nordmark * Given count worth of ires and generations, build ire_dep_* relationships 2951bd670b35SErik Nordmark * from ires[0] to ires[count-1]. Record generations[i+1] in 2952bd670b35SErik Nordmark * ire_dep_parent_generation for ires[i]. 2953bd670b35SErik Nordmark * We graft onto an existing parent chain by making sure that we don't 2954bd670b35SErik Nordmark * touch ire_dep_parent for ires[count-1]. 2955bd670b35SErik Nordmark * 2956bd670b35SErik Nordmark * We check for any condemned ire_generation count and return B_FALSE in 2957bd670b35SErik Nordmark * that case so that the caller can tear it apart. 2958bd670b35SErik Nordmark * 2959bd670b35SErik Nordmark * Note that generations[0] is not used. Caller handles that. 2960bd670b35SErik Nordmark */ 2961bd670b35SErik Nordmark boolean_t 2962bd670b35SErik Nordmark ire_dep_build(ire_t *ires[], uint_t generations[], uint_t count) 2963bd670b35SErik Nordmark { 2964bd670b35SErik Nordmark ire_t *ire = ires[0]; 2965bd670b35SErik Nordmark ip_stack_t *ipst; 2966bd670b35SErik Nordmark uint_t i; 2967bd670b35SErik Nordmark 2968bd670b35SErik Nordmark ASSERT(count > 0); 2969bd670b35SErik Nordmark if (count == 1) { 2970bd670b35SErik Nordmark /* No work to do */ 2971bd670b35SErik Nordmark return (B_TRUE); 2972bd670b35SErik Nordmark } 2973bd670b35SErik Nordmark ipst = ire->ire_ipst; 2974bd670b35SErik Nordmark rw_enter(&ipst->ips_ire_dep_lock, RW_WRITER); 2975bd670b35SErik Nordmark /* 2976bd670b35SErik Nordmark * Do not remove the linkage for any existing parent chain i.e., 2977bd670b35SErik Nordmark * ires[count-1] is left alone. 2978bd670b35SErik Nordmark */ 2979bd670b35SErik Nordmark for (i = 0; i < count-1; i++) { 2980bd670b35SErik Nordmark /* Remove existing parent if we need to change it */ 2981bd670b35SErik Nordmark if (ires[i]->ire_dep_parent != NULL && 2982bd670b35SErik Nordmark ires[i]->ire_dep_parent != ires[i+1]) 2983bd670b35SErik Nordmark ire_dep_remove(ires[i]); 2984bd670b35SErik Nordmark } 2985bd670b35SErik Nordmark 2986bd670b35SErik Nordmark for (i = 0; i < count - 1; i++) { 2987bd670b35SErik Nordmark ASSERT(ires[i]->ire_ipversion == IPV4_VERSION || 2988bd670b35SErik Nordmark ires[i]->ire_ipversion == IPV6_VERSION); 2989bd670b35SErik Nordmark /* Does it need to change? */ 2990bd670b35SErik Nordmark if (ires[i]->ire_dep_parent != ires[i+1]) 2991bd670b35SErik Nordmark ire_dep_parent_insert(ires[i], ires[i+1]); 2992bd670b35SErik Nordmark 2993bd670b35SErik Nordmark mutex_enter(&ires[i+1]->ire_lock); 2994bd670b35SErik Nordmark if (IRE_IS_CONDEMNED(ires[i+1])) { 2995bd670b35SErik Nordmark mutex_exit(&ires[i+1]->ire_lock); 2996bd670b35SErik Nordmark rw_exit(&ipst->ips_ire_dep_lock); 2997bd670b35SErik Nordmark return (B_FALSE); 2998bd670b35SErik Nordmark } 2999bd670b35SErik Nordmark mutex_exit(&ires[i+1]->ire_lock); 3000bd670b35SErik Nordmark 3001bd670b35SErik Nordmark mutex_enter(&ires[i]->ire_lock); 3002bd670b35SErik Nordmark ires[i]->ire_dep_parent_generation = generations[i+1]; 3003bd670b35SErik Nordmark mutex_exit(&ires[i]->ire_lock); 3004bd670b35SErik Nordmark } 3005bd670b35SErik Nordmark rw_exit(&ipst->ips_ire_dep_lock); 3006bd670b35SErik Nordmark return (B_TRUE); 3007bd670b35SErik Nordmark } 3008bd670b35SErik Nordmark 3009bd670b35SErik Nordmark /* 3010bd670b35SErik Nordmark * Given count worth of ires, unbuild ire_dep_* relationships 3011bd670b35SErik Nordmark * from ires[0] to ires[count-1]. 3012bd670b35SErik Nordmark */ 3013bd670b35SErik Nordmark void 3014bd670b35SErik Nordmark ire_dep_unbuild(ire_t *ires[], uint_t count) 3015bd670b35SErik Nordmark { 3016bd670b35SErik Nordmark ip_stack_t *ipst; 3017bd670b35SErik Nordmark uint_t i; 3018bd670b35SErik Nordmark 3019bd670b35SErik Nordmark if (count == 0) { 3020bd670b35SErik Nordmark /* No work to do */ 3021bd670b35SErik Nordmark return; 3022bd670b35SErik Nordmark } 3023bd670b35SErik Nordmark ipst = ires[0]->ire_ipst; 3024bd670b35SErik Nordmark rw_enter(&ipst->ips_ire_dep_lock, RW_WRITER); 3025bd670b35SErik Nordmark for (i = 0; i < count; i++) { 3026bd670b35SErik Nordmark ASSERT(ires[i]->ire_ipversion == IPV4_VERSION || 3027bd670b35SErik Nordmark ires[i]->ire_ipversion == IPV6_VERSION); 3028bd670b35SErik Nordmark if (ires[i]->ire_dep_parent != NULL) 3029bd670b35SErik Nordmark ire_dep_remove(ires[i]); 3030bd670b35SErik Nordmark mutex_enter(&ires[i]->ire_lock); 3031bd670b35SErik Nordmark ires[i]->ire_dep_parent_generation = IRE_GENERATION_VERIFY; 3032bd670b35SErik Nordmark mutex_exit(&ires[i]->ire_lock); 3033bd670b35SErik Nordmark } 3034bd670b35SErik Nordmark rw_exit(&ipst->ips_ire_dep_lock); 3035bd670b35SErik Nordmark } 3036bd670b35SErik Nordmark 3037bd670b35SErik Nordmark /* 3038bd670b35SErik Nordmark * Both the forwarding and the outbound code paths can trip on 3039bd670b35SErik Nordmark * a condemned NCE, in which case we call this function. 3040bd670b35SErik Nordmark * We have two different behaviors: if the NCE was UNREACHABLE 3041bd670b35SErik Nordmark * it is an indication that something failed. In that case 3042bd670b35SErik Nordmark * we see if we should look for a different IRE (for example, 3043bd670b35SErik Nordmark * delete any matching redirect IRE, or try a different 3044bd670b35SErik Nordmark * IRE_DEFAULT (ECMP)). We mark the ire as bad so a hopefully 3045bd670b35SErik Nordmark * different IRE will be picked next time we send/forward. 3046bd670b35SErik Nordmark * 3047bd670b35SErik Nordmark * If we are called by the output path then fail_if_better is set 3048bd670b35SErik Nordmark * and we return NULL if there could be a better IRE. This is because the 3049bd670b35SErik Nordmark * output path retries the IRE lookup. (The input/forward path can not retry.) 3050bd670b35SErik Nordmark * 3051bd670b35SErik Nordmark * If the NCE was not unreachable then we pick/allocate a 3052bd670b35SErik Nordmark * new (most likely ND_INITIAL) NCE and proceed with it. 3053bd670b35SErik Nordmark * 3054bd670b35SErik Nordmark * ipha/ip6h are needed for multicast packets; ipha needs to be 3055bd670b35SErik Nordmark * set for IPv4 and ip6h needs to be set for IPv6 packets. 3056bd670b35SErik Nordmark */ 3057bd670b35SErik Nordmark nce_t * 3058bd670b35SErik Nordmark ire_handle_condemned_nce(nce_t *nce, ire_t *ire, ipha_t *ipha, ip6_t *ip6h, 3059bd670b35SErik Nordmark boolean_t fail_if_better) 3060bd670b35SErik Nordmark { 3061bd670b35SErik Nordmark if (nce->nce_common->ncec_state == ND_UNREACHABLE) { 3062bd670b35SErik Nordmark if (ire_no_good(ire) && fail_if_better) { 3063bd670b35SErik Nordmark /* 3064bd670b35SErik Nordmark * Did some changes, or ECMP likely to exist. 3065bd670b35SErik Nordmark * Make ip_output look for a different IRE 3066bd670b35SErik Nordmark */ 3067bd670b35SErik Nordmark return (NULL); 3068bd670b35SErik Nordmark } 3069bd670b35SErik Nordmark } 3070bd670b35SErik Nordmark if (ire_revalidate_nce(ire) == ENETUNREACH) { 3071bd670b35SErik Nordmark /* The ire_dep_parent chain went bad, or no memory? */ 3072bd670b35SErik Nordmark (void) ire_no_good(ire); 3073bd670b35SErik Nordmark return (NULL); 3074bd670b35SErik Nordmark } 3075bd670b35SErik Nordmark if (ire->ire_ipversion == IPV4_VERSION) { 3076bd670b35SErik Nordmark ASSERT(ipha != NULL); 3077bd670b35SErik Nordmark nce = ire_to_nce(ire, ipha->ipha_dst, NULL); 3078bd670b35SErik Nordmark } else { 3079bd670b35SErik Nordmark ASSERT(ip6h != NULL); 3080bd670b35SErik Nordmark nce = ire_to_nce(ire, INADDR_ANY, &ip6h->ip6_dst); 3081bd670b35SErik Nordmark } 3082bd670b35SErik Nordmark 3083bd670b35SErik Nordmark if (nce == NULL) 3084bd670b35SErik Nordmark return (NULL); 3085bd670b35SErik Nordmark if (nce->nce_is_condemned) { 3086bd670b35SErik Nordmark nce_refrele(nce); 3087bd670b35SErik Nordmark return (NULL); 3088bd670b35SErik Nordmark } 3089bd670b35SErik Nordmark return (nce); 3090bd670b35SErik Nordmark } 3091bd670b35SErik Nordmark 3092bd670b35SErik Nordmark /* 3093bd670b35SErik Nordmark * The caller has found that the ire is bad, either due to a reference to an NCE 3094bd670b35SErik Nordmark * in ND_UNREACHABLE state, or a MULTIRT route whose gateway can't be resolved. 3095bd670b35SErik Nordmark * We update things so a subsequent attempt to send to the destination 3096bd670b35SErik Nordmark * is likely to find different IRE, or that a new NCE would be created. 3097bd670b35SErik Nordmark * 3098bd670b35SErik Nordmark * Returns B_TRUE if it is likely that a subsequent ire_ftable_lookup would 3099bd670b35SErik Nordmark * find a different route (either due to having deleted a redirect, or there 3100bd670b35SErik Nordmark * being ECMP routes.) 3101bd670b35SErik Nordmark * 3102bd670b35SErik Nordmark * If we have a redirect (RTF_DYNAMIC) we delete it. 3103bd670b35SErik Nordmark * Otherwise we increment ire_badcnt and increment the generation number so 3104bd670b35SErik Nordmark * that a cached ixa_ire will redo the route selection. ire_badcnt is taken 3105bd670b35SErik Nordmark * into account in the route selection when we have multiple choices (multiple 3106bd670b35SErik Nordmark * default routes or ECMP in general). 3107bd670b35SErik Nordmark * Any time ip_select_route find an ire with a condemned ire_nce_cache 3108bd670b35SErik Nordmark * (e.g., if no equal cost route to the bad one) ip_select_route will make 3109bd670b35SErik Nordmark * sure the NCE is revalidated to avoid getting stuck on a 3110bd670b35SErik Nordmark * NCE_F_CONDMNED ncec that caused ire_no_good to be called. 3111bd670b35SErik Nordmark */ 3112bd670b35SErik Nordmark boolean_t 3113bd670b35SErik Nordmark ire_no_good(ire_t *ire) 3114bd670b35SErik Nordmark { 3115bd670b35SErik Nordmark ip_stack_t *ipst = ire->ire_ipst; 3116bd670b35SErik Nordmark ire_t *ire2; 3117bd670b35SErik Nordmark nce_t *nce; 3118bd670b35SErik Nordmark 3119bd670b35SErik Nordmark if (ire->ire_flags & RTF_DYNAMIC) { 3120bd670b35SErik Nordmark ire_delete(ire); 3121bd670b35SErik Nordmark return (B_TRUE); 3122bd670b35SErik Nordmark } 3123bd670b35SErik Nordmark if (ire->ire_flags & RTF_INDIRECT) { 3124bd670b35SErik Nordmark /* Check if next IRE is a redirect */ 3125bd670b35SErik Nordmark rw_enter(&ipst->ips_ire_dep_lock, RW_READER); 3126bd670b35SErik Nordmark if (ire->ire_dep_parent != NULL && 3127bd670b35SErik Nordmark (ire->ire_dep_parent->ire_flags & RTF_DYNAMIC)) { 3128bd670b35SErik Nordmark ire2 = ire->ire_dep_parent; 3129bd670b35SErik Nordmark ire_refhold(ire2); 3130bd670b35SErik Nordmark } else { 3131bd670b35SErik Nordmark ire2 = NULL; 3132bd670b35SErik Nordmark } 3133bd670b35SErik Nordmark rw_exit(&ipst->ips_ire_dep_lock); 3134bd670b35SErik Nordmark if (ire2 != NULL) { 3135bd670b35SErik Nordmark ire_delete(ire2); 3136bd670b35SErik Nordmark ire_refrele(ire2); 3137bd670b35SErik Nordmark return (B_TRUE); 3138bd670b35SErik Nordmark } 3139bd670b35SErik Nordmark } 3140bd670b35SErik Nordmark /* 3141bd670b35SErik Nordmark * No redirect involved. Increment badcnt so that if we have ECMP 3142bd670b35SErik Nordmark * routes we are likely to pick a different one for the next packet. 3143bd670b35SErik Nordmark * 3144bd670b35SErik Nordmark * If the NCE is unreachable and condemned we should drop the reference 3145bd670b35SErik Nordmark * to it so that a new NCE can be created. 3146bd670b35SErik Nordmark * 3147bd670b35SErik Nordmark * Finally we increment the generation number so that any ixa_ire 3148bd670b35SErik Nordmark * cache will be revalidated. 3149bd670b35SErik Nordmark */ 3150bd670b35SErik Nordmark mutex_enter(&ire->ire_lock); 3151bd670b35SErik Nordmark ire->ire_badcnt++; 3152d3d50737SRafael Vanoni ire->ire_last_badcnt = TICK_TO_SEC(ddi_get_lbolt64()); 3153bd670b35SErik Nordmark nce = ire->ire_nce_cache; 3154bd670b35SErik Nordmark if (nce != NULL && nce->nce_is_condemned && 3155bd670b35SErik Nordmark nce->nce_common->ncec_state == ND_UNREACHABLE) 3156bd670b35SErik Nordmark ire->ire_nce_cache = NULL; 3157bd670b35SErik Nordmark else 3158bd670b35SErik Nordmark nce = NULL; 3159bd670b35SErik Nordmark mutex_exit(&ire->ire_lock); 3160bd670b35SErik Nordmark if (nce != NULL) 3161bd670b35SErik Nordmark nce_refrele(nce); 3162bd670b35SErik Nordmark 3163bd670b35SErik Nordmark ire_increment_generation(ire); 3164bd670b35SErik Nordmark ire_dep_incr_generation(ire); 3165bd670b35SErik Nordmark 3166bd670b35SErik Nordmark return (ire->ire_bucket->irb_ire_cnt > 1); 3167bd670b35SErik Nordmark } 3168bd670b35SErik Nordmark 3169bd670b35SErik Nordmark /* 3170bd670b35SErik Nordmark * Walk ire_dep_parent chain and validate that ire_dep_parent->ire_generation == 3171bd670b35SErik Nordmark * ire_dep_parent_generation. 3172bd670b35SErik Nordmark * If they all match we just return ire_generation from the topmost IRE. 3173bd670b35SErik Nordmark * Otherwise we propagate the mismatch by setting all ire_dep_parent_generation 3174bd670b35SErik Nordmark * above the mismatch to IRE_GENERATION_VERIFY and also returning 3175bd670b35SErik Nordmark * IRE_GENERATION_VERIFY. 3176bd670b35SErik Nordmark */ 3177bd670b35SErik Nordmark uint_t 3178bd670b35SErik Nordmark ire_dep_validate_generations(ire_t *ire) 3179bd670b35SErik Nordmark { 3180bd670b35SErik Nordmark ip_stack_t *ipst = ire->ire_ipst; 3181bd670b35SErik Nordmark uint_t generation; 3182bd670b35SErik Nordmark ire_t *ire1; 3183bd670b35SErik Nordmark 3184bd670b35SErik Nordmark rw_enter(&ipst->ips_ire_dep_lock, RW_READER); 3185bd670b35SErik Nordmark generation = ire->ire_generation; /* Assuming things match */ 3186bd670b35SErik Nordmark for (ire1 = ire; ire1 != NULL; ire1 = ire1->ire_dep_parent) { 3187bd670b35SErik Nordmark ASSERT(ire1->ire_ipversion == IPV4_VERSION || 3188bd670b35SErik Nordmark ire1->ire_ipversion == IPV6_VERSION); 3189bd670b35SErik Nordmark if (ire1->ire_dep_parent == NULL) 3190bd670b35SErik Nordmark break; 3191bd670b35SErik Nordmark if (ire1->ire_dep_parent_generation != 3192bd670b35SErik Nordmark ire1->ire_dep_parent->ire_generation) 3193bd670b35SErik Nordmark goto mismatch; 3194bd670b35SErik Nordmark } 3195bd670b35SErik Nordmark rw_exit(&ipst->ips_ire_dep_lock); 3196bd670b35SErik Nordmark return (generation); 3197bd670b35SErik Nordmark 3198bd670b35SErik Nordmark mismatch: 3199bd670b35SErik Nordmark generation = IRE_GENERATION_VERIFY; 3200bd670b35SErik Nordmark /* Fill from top down to the mismatch with _VERIFY */ 3201bd670b35SErik Nordmark while (ire != ire1) { 3202bd670b35SErik Nordmark ASSERT(ire->ire_ipversion == IPV4_VERSION || 3203bd670b35SErik Nordmark ire->ire_ipversion == IPV6_VERSION); 3204bd670b35SErik Nordmark mutex_enter(&ire->ire_lock); 3205bd670b35SErik Nordmark ire->ire_dep_parent_generation = IRE_GENERATION_VERIFY; 3206bd670b35SErik Nordmark mutex_exit(&ire->ire_lock); 3207bd670b35SErik Nordmark ire = ire->ire_dep_parent; 3208bd670b35SErik Nordmark } 3209bd670b35SErik Nordmark rw_exit(&ipst->ips_ire_dep_lock); 3210bd670b35SErik Nordmark return (generation); 3211bd670b35SErik Nordmark } 3212bd670b35SErik Nordmark 3213bd670b35SErik Nordmark /* 3214bd670b35SErik Nordmark * Used when we need to return an ire with ire_dep_parent, but we 3215bd670b35SErik Nordmark * know the chain is invalid for instance we didn't create an IRE_IF_CLONE 3216bd670b35SErik Nordmark * Using IRE_GENERATION_VERIFY means that next time we'll redo the 3217bd670b35SErik Nordmark * recursive lookup. 3218bd670b35SErik Nordmark */ 3219bd670b35SErik Nordmark void 3220bd670b35SErik Nordmark ire_dep_invalidate_generations(ire_t *ire) 3221bd670b35SErik Nordmark { 3222bd670b35SErik Nordmark ip_stack_t *ipst = ire->ire_ipst; 3223bd670b35SErik Nordmark 3224bd670b35SErik Nordmark rw_enter(&ipst->ips_ire_dep_lock, RW_READER); 3225bd670b35SErik Nordmark while (ire != NULL) { 3226bd670b35SErik Nordmark ASSERT(ire->ire_ipversion == IPV4_VERSION || 3227bd670b35SErik Nordmark ire->ire_ipversion == IPV6_VERSION); 3228bd670b35SErik Nordmark mutex_enter(&ire->ire_lock); 3229bd670b35SErik Nordmark ire->ire_dep_parent_generation = IRE_GENERATION_VERIFY; 3230bd670b35SErik Nordmark mutex_exit(&ire->ire_lock); 3231bd670b35SErik Nordmark ire = ire->ire_dep_parent; 3232bd670b35SErik Nordmark } 3233bd670b35SErik Nordmark rw_exit(&ipst->ips_ire_dep_lock); 3234bd670b35SErik Nordmark } 3235bd670b35SErik Nordmark 3236bd670b35SErik Nordmark /* Set _VERIFY ire_dep_parent_generation for all children recursively */ 3237bd670b35SErik Nordmark static void 3238bd670b35SErik Nordmark ire_dep_invalidate_children(ire_t *child) 3239bd670b35SErik Nordmark { 3240bd670b35SErik Nordmark ip_stack_t *ipst = child->ire_ipst; 3241bd670b35SErik Nordmark 3242bd670b35SErik Nordmark ASSERT(RW_WRITE_HELD(&ipst->ips_ire_dep_lock)); 3243bd670b35SErik Nordmark /* Depth first */ 3244bd670b35SErik Nordmark if (child->ire_dep_children != NULL) 3245bd670b35SErik Nordmark ire_dep_invalidate_children(child->ire_dep_children); 3246bd670b35SErik Nordmark 3247bd670b35SErik Nordmark while (child != NULL) { 3248bd670b35SErik Nordmark mutex_enter(&child->ire_lock); 3249bd670b35SErik Nordmark child->ire_dep_parent_generation = IRE_GENERATION_VERIFY; 3250bd670b35SErik Nordmark mutex_exit(&child->ire_lock); 3251bd670b35SErik Nordmark child = child->ire_dep_sib_next; 3252bd670b35SErik Nordmark } 3253bd670b35SErik Nordmark } 3254bd670b35SErik Nordmark 3255bd670b35SErik Nordmark static void 3256bd670b35SErik Nordmark ire_dep_increment_children(ire_t *child) 3257bd670b35SErik Nordmark { 3258bd670b35SErik Nordmark ip_stack_t *ipst = child->ire_ipst; 3259bd670b35SErik Nordmark 3260bd670b35SErik Nordmark ASSERT(RW_READ_HELD(&ipst->ips_ire_dep_lock)); 3261bd670b35SErik Nordmark /* Depth first */ 3262bd670b35SErik Nordmark if (child->ire_dep_children != NULL) 3263bd670b35SErik Nordmark ire_dep_increment_children(child->ire_dep_children); 3264bd670b35SErik Nordmark 3265bd670b35SErik Nordmark while (child != NULL) { 3266bd670b35SErik Nordmark if (!IRE_IS_CONDEMNED(child)) 3267bd670b35SErik Nordmark ire_increment_generation(child); 3268bd670b35SErik Nordmark child = child->ire_dep_sib_next; 3269bd670b35SErik Nordmark } 3270bd670b35SErik Nordmark } 3271bd670b35SErik Nordmark 3272bd670b35SErik Nordmark /* 3273bd670b35SErik Nordmark * Walk all the children of this ire recursively and increment their 3274bd670b35SErik Nordmark * generation number. 3275bd670b35SErik Nordmark */ 3276f1c454b4SSowmini Varadhan static void 3277f1c454b4SSowmini Varadhan ire_dep_incr_generation_locked(ire_t *parent) 3278f1c454b4SSowmini Varadhan { 3279f1c454b4SSowmini Varadhan ASSERT(RW_READ_HELD(&parent->ire_ipst->ips_ire_dep_lock)); 3280f1c454b4SSowmini Varadhan if (parent->ire_dep_children != NULL) 3281f1c454b4SSowmini Varadhan ire_dep_increment_children(parent->ire_dep_children); 3282f1c454b4SSowmini Varadhan } 3283f1c454b4SSowmini Varadhan 3284bd670b35SErik Nordmark void 3285bd670b35SErik Nordmark ire_dep_incr_generation(ire_t *parent) 3286bd670b35SErik Nordmark { 3287bd670b35SErik Nordmark ip_stack_t *ipst = parent->ire_ipst; 3288bd670b35SErik Nordmark 3289bd670b35SErik Nordmark rw_enter(&ipst->ips_ire_dep_lock, RW_READER); 3290f1c454b4SSowmini Varadhan ire_dep_incr_generation_locked(parent); 3291bd670b35SErik Nordmark rw_exit(&ipst->ips_ire_dep_lock); 3292bd670b35SErik Nordmark } 3293bd670b35SErik Nordmark 3294bd670b35SErik Nordmark /* 3295bd670b35SErik Nordmark * Get a new ire_nce_cache for this IRE as well as its nexthop. 3296bd670b35SErik Nordmark * Returns zero if it succeeds. Can fail due to lack of memory or when 3297bd670b35SErik Nordmark * the route has become unreachable. Returns ENOMEM and ENETUNREACH in those 3298bd670b35SErik Nordmark * cases. 3299bd670b35SErik Nordmark * 3300bd670b35SErik Nordmark * In the in.mpathd case, the ire will have ire_testhidden 3301bd670b35SErik Nordmark * set; so we should create the ncec for the underlying ill. 3302bd670b35SErik Nordmark * 3303bd670b35SErik Nordmark * Note that the error returned by ire_revalidate_nce() is ignored by most 3304bd670b35SErik Nordmark * callers except ire_handle_condemned_nce(), which handles the ENETUNREACH 3305bd670b35SErik Nordmark * error to mark potentially bad ire's. For all the other callers, an 3306bd670b35SErik Nordmark * error return could indicate a transient condition like ENOMEM, or could 3307bd670b35SErik Nordmark * be the result of an interface that is going down/unplumbing. In the former 3308bd670b35SErik Nordmark * case (transient error), we would leave the old stale ire/ire_nce_cache 3309bd670b35SErik Nordmark * in place, and possibly use incorrect link-layer information to send packets 3310bd670b35SErik Nordmark * but would eventually recover. In the latter case (ill down/replumb), 3311bd670b35SErik Nordmark * ire_revalidate_nce() might return a condemned nce back, but we would then 3312bd670b35SErik Nordmark * recover in the packet output path. 3313bd670b35SErik Nordmark */ 3314bd670b35SErik Nordmark int 3315bd670b35SErik Nordmark ire_revalidate_nce(ire_t *ire) 3316bd670b35SErik Nordmark { 3317bd670b35SErik Nordmark nce_t *nce, *old_nce; 3318bd670b35SErik Nordmark ire_t *nexthop; 3319bd670b35SErik Nordmark 3320bd670b35SErik Nordmark /* 3321bd670b35SErik Nordmark * For multicast we conceptually have an NCE but we don't store it 3322bd670b35SErik Nordmark * in ire_nce_cache; when ire_to_nce is called we allocate the nce. 3323bd670b35SErik Nordmark */ 3324bd670b35SErik Nordmark if (ire->ire_type & IRE_MULTICAST) 3325bd670b35SErik Nordmark return (0); 3326bd670b35SErik Nordmark 3327bd670b35SErik Nordmark /* ire_testhidden should only be set on under-interfaces */ 3328bd670b35SErik Nordmark ASSERT(!ire->ire_testhidden || !IS_IPMP(ire->ire_ill)); 3329bd670b35SErik Nordmark 3330bd670b35SErik Nordmark nexthop = ire_nexthop(ire); 3331bd670b35SErik Nordmark if (nexthop == NULL) { 3332bd670b35SErik Nordmark /* The route is potentially bad */ 3333bd670b35SErik Nordmark (void) ire_no_good(ire); 3334bd670b35SErik Nordmark return (ENETUNREACH); 3335bd670b35SErik Nordmark } 3336bd670b35SErik Nordmark if (ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK)) { 3337bd670b35SErik Nordmark ASSERT(ire->ire_ill != NULL); 3338bd670b35SErik Nordmark 3339bd670b35SErik Nordmark if (ire->ire_ipversion == IPV4_VERSION) 3340bd670b35SErik Nordmark nce = nce_lookup_v4(ire->ire_ill, &ire->ire_addr); 3341bd670b35SErik Nordmark else 3342bd670b35SErik Nordmark nce = nce_lookup_v6(ire->ire_ill, &ire->ire_addr_v6); 3343bd670b35SErik Nordmark } else { 3344bd670b35SErik Nordmark ASSERT(nexthop->ire_type & IRE_ONLINK); 3345bd670b35SErik Nordmark if (ire->ire_ipversion == IPV4_VERSION) { 3346bd670b35SErik Nordmark nce = arp_nce_init(nexthop->ire_ill, nexthop->ire_addr, 3347bd670b35SErik Nordmark nexthop->ire_type); 3348bd670b35SErik Nordmark } else { 3349bd670b35SErik Nordmark nce = ndp_nce_init(nexthop->ire_ill, 3350bd670b35SErik Nordmark &nexthop->ire_addr_v6, nexthop->ire_type); 3351bd670b35SErik Nordmark } 3352bd670b35SErik Nordmark } 3353bd670b35SErik Nordmark if (nce == NULL) { 3354bd670b35SErik Nordmark /* 3355bd670b35SErik Nordmark * Leave the old stale one in place to avoid a NULL 3356bd670b35SErik Nordmark * ire_nce_cache. 3357bd670b35SErik Nordmark */ 3358bd670b35SErik Nordmark ire_refrele(nexthop); 3359bd670b35SErik Nordmark return (ENOMEM); 3360bd670b35SErik Nordmark } 3361bd670b35SErik Nordmark 3362bd670b35SErik Nordmark if (nexthop != ire) { 3363bd670b35SErik Nordmark /* Update the nexthop ire */ 3364bd670b35SErik Nordmark mutex_enter(&nexthop->ire_lock); 3365bd670b35SErik Nordmark old_nce = nexthop->ire_nce_cache; 3366bd670b35SErik Nordmark if (!IRE_IS_CONDEMNED(nexthop)) { 3367bd670b35SErik Nordmark nce_refhold(nce); 3368bd670b35SErik Nordmark nexthop->ire_nce_cache = nce; 3369bd670b35SErik Nordmark } else { 3370bd670b35SErik Nordmark nexthop->ire_nce_cache = NULL; 3371bd670b35SErik Nordmark } 3372bd670b35SErik Nordmark mutex_exit(&nexthop->ire_lock); 3373bd670b35SErik Nordmark if (old_nce != NULL) 3374bd670b35SErik Nordmark nce_refrele(old_nce); 3375bd670b35SErik Nordmark } 3376bd670b35SErik Nordmark ire_refrele(nexthop); 3377bd670b35SErik Nordmark 3378bd670b35SErik Nordmark mutex_enter(&ire->ire_lock); 3379bd670b35SErik Nordmark old_nce = ire->ire_nce_cache; 3380bd670b35SErik Nordmark if (!IRE_IS_CONDEMNED(ire)) { 3381bd670b35SErik Nordmark nce_refhold(nce); 3382bd670b35SErik Nordmark ire->ire_nce_cache = nce; 3383bd670b35SErik Nordmark } else { 3384bd670b35SErik Nordmark ire->ire_nce_cache = NULL; 3385bd670b35SErik Nordmark } 3386bd670b35SErik Nordmark mutex_exit(&ire->ire_lock); 3387bd670b35SErik Nordmark if (old_nce != NULL) 3388bd670b35SErik Nordmark nce_refrele(old_nce); 3389bd670b35SErik Nordmark 3390bd670b35SErik Nordmark nce_refrele(nce); 3391c793af95Ssangeeta return (0); 3392c793af95Ssangeeta } 33935b17e9bdSJon Anderson 33945b17e9bdSJon Anderson /* 3395bd670b35SErik Nordmark * Get a held nce for a given ire. 3396bd670b35SErik Nordmark * In the common case this is just from ire_nce_cache. 3397bd670b35SErik Nordmark * For IRE_MULTICAST this needs to do an explicit lookup since we do not 3398bd670b35SErik Nordmark * have an IRE_MULTICAST per address. 3399bd670b35SErik Nordmark * Note that this explicitly returns CONDEMNED NCEs. The caller needs those 3400bd670b35SErik Nordmark * so they can check whether the NCE went unreachable (as opposed to was 3401bd670b35SErik Nordmark * condemned for some other reason). 34025b17e9bdSJon Anderson */ 3403bd670b35SErik Nordmark nce_t * 3404bd670b35SErik Nordmark ire_to_nce(ire_t *ire, ipaddr_t v4nexthop, const in6_addr_t *v6nexthop) 34055b17e9bdSJon Anderson { 3406bd670b35SErik Nordmark nce_t *nce; 34075b17e9bdSJon Anderson 3408bd670b35SErik Nordmark if (ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) 3409bd670b35SErik Nordmark return (NULL); 3410bd670b35SErik Nordmark 3411bd670b35SErik Nordmark /* ire_testhidden should only be set on under-interfaces */ 3412bd670b35SErik Nordmark ASSERT(!ire->ire_testhidden || !IS_IPMP(ire->ire_ill)); 3413bd670b35SErik Nordmark 3414bd670b35SErik Nordmark mutex_enter(&ire->ire_lock); 3415bd670b35SErik Nordmark nce = ire->ire_nce_cache; 3416bd670b35SErik Nordmark if (nce != NULL) { 3417bd670b35SErik Nordmark nce_refhold(nce); 3418bd670b35SErik Nordmark mutex_exit(&ire->ire_lock); 3419bd670b35SErik Nordmark return (nce); 3420bd670b35SErik Nordmark } 3421bd670b35SErik Nordmark mutex_exit(&ire->ire_lock); 3422bd670b35SErik Nordmark 3423bd670b35SErik Nordmark if (ire->ire_type & IRE_MULTICAST) { 3424bd670b35SErik Nordmark ASSERT(ire->ire_ill != NULL); 3425bd670b35SErik Nordmark 3426bd670b35SErik Nordmark if (ire->ire_ipversion == IPV4_VERSION) { 3427bd670b35SErik Nordmark ASSERT(v6nexthop == NULL); 3428bd670b35SErik Nordmark 3429bd670b35SErik Nordmark nce = arp_nce_init(ire->ire_ill, v4nexthop, 3430bd670b35SErik Nordmark ire->ire_type); 3431bd670b35SErik Nordmark } else { 3432bd670b35SErik Nordmark ASSERT(v6nexthop != NULL); 3433bd670b35SErik Nordmark ASSERT(v4nexthop == 0); 3434bd670b35SErik Nordmark nce = ndp_nce_init(ire->ire_ill, v6nexthop, 3435bd670b35SErik Nordmark ire->ire_type); 3436bd670b35SErik Nordmark } 3437bd670b35SErik Nordmark return (nce); 3438bd670b35SErik Nordmark } 34395b17e9bdSJon Anderson return (NULL); 34405b17e9bdSJon Anderson } 34415b17e9bdSJon Anderson 3442bd670b35SErik Nordmark nce_t * 3443bd670b35SErik Nordmark ire_to_nce_pkt(ire_t *ire, mblk_t *mp) 3444bd670b35SErik Nordmark { 3445bd670b35SErik Nordmark ipha_t *ipha; 3446bd670b35SErik Nordmark ip6_t *ip6h; 34475b17e9bdSJon Anderson 3448bd670b35SErik Nordmark if (IPH_HDR_VERSION(mp->b_rptr) == IPV4_VERSION) { 3449bd670b35SErik Nordmark ipha = (ipha_t *)mp->b_rptr; 3450bd670b35SErik Nordmark return (ire_to_nce(ire, ipha->ipha_dst, NULL)); 3451bd670b35SErik Nordmark } else { 3452bd670b35SErik Nordmark ip6h = (ip6_t *)mp->b_rptr; 3453bd670b35SErik Nordmark return (ire_to_nce(ire, INADDR_ANY, &ip6h->ip6_dst)); 3454bd670b35SErik Nordmark } 34555b17e9bdSJon Anderson } 34565b17e9bdSJon Anderson 34575b17e9bdSJon Anderson /* 3458bd670b35SErik Nordmark * Given an IRE_INTERFACE (that matches more than one address) create 3459bd670b35SErik Nordmark * and return an IRE_IF_CLONE for the specific address. 3460bd670b35SErik Nordmark * Return the generation number. 3461bd670b35SErik Nordmark * Returns NULL is no memory for the IRE. 3462bd670b35SErik Nordmark * Handles both IPv4 and IPv6. 34635b17e9bdSJon Anderson */ 34645b17e9bdSJon Anderson ire_t * 3465bd670b35SErik Nordmark ire_create_if_clone(ire_t *ire_if, const in6_addr_t *addr, uint_t *generationp) 34665b17e9bdSJon Anderson { 3467bd670b35SErik Nordmark ire_t *ire; 3468bd670b35SErik Nordmark ire_t *nire; 34695b17e9bdSJon Anderson 3470bd670b35SErik Nordmark if (ire_if->ire_ipversion == IPV4_VERSION) { 3471bd670b35SErik Nordmark ipaddr_t v4addr; 3472bd670b35SErik Nordmark ipaddr_t mask = IP_HOST_MASK; 34735b17e9bdSJon Anderson 3474bd670b35SErik Nordmark ASSERT(IN6_IS_ADDR_V4MAPPED(addr)); 3475bd670b35SErik Nordmark IN6_V4MAPPED_TO_IPADDR(addr, v4addr); 3476bd670b35SErik Nordmark 3477bd670b35SErik Nordmark ire = ire_create( 3478bd670b35SErik Nordmark (uchar_t *)&v4addr, /* dest address */ 3479bd670b35SErik Nordmark (uchar_t *)&mask, /* mask */ 3480bd670b35SErik Nordmark (uchar_t *)&ire_if->ire_gateway_addr, 3481bd670b35SErik Nordmark IRE_IF_CLONE, /* IRE type */ 3482bd670b35SErik Nordmark ire_if->ire_ill, 3483bd670b35SErik Nordmark ire_if->ire_zoneid, 3484bd670b35SErik Nordmark ire_if->ire_flags | RTF_HOST, 3485bd670b35SErik Nordmark NULL, /* No security attr for IRE_IF_ALL */ 3486bd670b35SErik Nordmark ire_if->ire_ipst); 3487bd670b35SErik Nordmark } else { 3488bd670b35SErik Nordmark ASSERT(!IN6_IS_ADDR_V4MAPPED(addr)); 3489bd670b35SErik Nordmark ire = ire_create_v6( 3490bd670b35SErik Nordmark addr, /* dest address */ 3491bd670b35SErik Nordmark &ipv6_all_ones, /* mask */ 3492bd670b35SErik Nordmark &ire_if->ire_gateway_addr_v6, /* gateway addr */ 3493bd670b35SErik Nordmark IRE_IF_CLONE, /* IRE type */ 3494bd670b35SErik Nordmark ire_if->ire_ill, 3495bd670b35SErik Nordmark ire_if->ire_zoneid, 3496bd670b35SErik Nordmark ire_if->ire_flags | RTF_HOST, 3497bd670b35SErik Nordmark NULL, /* No security attr for IRE_IF_ALL */ 3498bd670b35SErik Nordmark ire_if->ire_ipst); 3499bd670b35SErik Nordmark } 3500bd670b35SErik Nordmark if (ire == NULL) 3501bd670b35SErik Nordmark return (NULL); 3502bd670b35SErik Nordmark 3503bd670b35SErik Nordmark /* Take the metrics, in particular the mtu, from the IRE_IF */ 3504bd670b35SErik Nordmark ire->ire_metrics = ire_if->ire_metrics; 3505bd670b35SErik Nordmark 3506bd670b35SErik Nordmark nire = ire_add(ire); 3507bd670b35SErik Nordmark if (nire == NULL) /* Some failure */ 3508bd670b35SErik Nordmark return (NULL); 3509bd670b35SErik Nordmark 3510bd670b35SErik Nordmark if (generationp != NULL) 3511bd670b35SErik Nordmark *generationp = nire->ire_generation; 3512bd670b35SErik Nordmark 3513bd670b35SErik Nordmark /* 3514bd670b35SErik Nordmark * Make sure races don't add a duplicate by 3515bd670b35SErik Nordmark * catching the case when an identical was returned. 3516bd670b35SErik Nordmark */ 3517bd670b35SErik Nordmark if (nire != ire) { 3518bd670b35SErik Nordmark ASSERT(nire->ire_identical_ref > 1); 3519bd670b35SErik Nordmark ire_delete(nire); 3520bd670b35SErik Nordmark } 3521bd670b35SErik Nordmark return (nire); 3522bd670b35SErik Nordmark } 3523bd670b35SErik Nordmark 3524bd670b35SErik Nordmark /* 3525bd670b35SErik Nordmark * The argument is an IRE_INTERFACE. Delete all of IRE_IF_CLONE in the 3526bd670b35SErik Nordmark * ire_dep_children (just walk the ire_dep_sib_next since they are all 3527bd670b35SErik Nordmark * immediate children.) 3528bd670b35SErik Nordmark * Since we hold a lock while we remove them we need to defer the actual 3529bd670b35SErik Nordmark * calls to ire_delete() until we have dropped the lock. This makes things 3530bd670b35SErik Nordmark * less efficient since we restart at the top after dropping the lock. But 3531bd670b35SErik Nordmark * we only run when an IRE_INTERFACE is deleted which is infrquent. 3532bd670b35SErik Nordmark * 3533bd670b35SErik Nordmark * Note that ire_dep_children can be any mixture of offlink routes and 3534bd670b35SErik Nordmark * IRE_IF_CLONE entries. 3535bd670b35SErik Nordmark */ 3536bd670b35SErik Nordmark void 3537bd670b35SErik Nordmark ire_dep_delete_if_clone(ire_t *parent) 3538bd670b35SErik Nordmark { 3539bd670b35SErik Nordmark ip_stack_t *ipst = parent->ire_ipst; 3540bd670b35SErik Nordmark ire_t *child, *next; 3541bd670b35SErik Nordmark 3542bd670b35SErik Nordmark restart: 3543bd670b35SErik Nordmark rw_enter(&ipst->ips_ire_dep_lock, RW_READER); 3544bd670b35SErik Nordmark if (parent->ire_dep_children == NULL) { 3545bd670b35SErik Nordmark rw_exit(&ipst->ips_ire_dep_lock); 3546bd670b35SErik Nordmark return; 3547bd670b35SErik Nordmark } 3548bd670b35SErik Nordmark child = parent->ire_dep_children; 3549bd670b35SErik Nordmark while (child != NULL) { 3550bd670b35SErik Nordmark next = child->ire_dep_sib_next; 3551bd670b35SErik Nordmark if ((child->ire_type & IRE_IF_CLONE) && 3552bd670b35SErik Nordmark !IRE_IS_CONDEMNED(child)) { 3553bd670b35SErik Nordmark ire_refhold(child); 3554bd670b35SErik Nordmark rw_exit(&ipst->ips_ire_dep_lock); 3555bd670b35SErik Nordmark ire_delete(child); 3556bd670b35SErik Nordmark ASSERT(IRE_IS_CONDEMNED(child)); 3557bd670b35SErik Nordmark ire_refrele(child); 3558bd670b35SErik Nordmark goto restart; 3559bd670b35SErik Nordmark } 3560bd670b35SErik Nordmark child = next; 3561bd670b35SErik Nordmark } 3562bd670b35SErik Nordmark rw_exit(&ipst->ips_ire_dep_lock); 3563bd670b35SErik Nordmark } 3564bd670b35SErik Nordmark 3565bd670b35SErik Nordmark /* 356644b099c4SSowmini Varadhan * In the preferred/strict src multihoming modes, unbound routes (i.e., 356744b099c4SSowmini Varadhan * ire_t entries with ire_unbound set to B_TRUE) are bound to an interface 356844b099c4SSowmini Varadhan * by selecting the first available interface that has an interface route for 356944b099c4SSowmini Varadhan * the ire_gateway. If that interface is subsequently brought down, ill_downi() 357044b099c4SSowmini Varadhan * will call ire_rebind() so that the unbound route can be bound to some other 357144b099c4SSowmini Varadhan * matching interface thereby preserving the intended reachability information 357244b099c4SSowmini Varadhan * from the original unbound route. 357344b099c4SSowmini Varadhan */ 357444b099c4SSowmini Varadhan void 357544b099c4SSowmini Varadhan ire_rebind(ire_t *ire) 357644b099c4SSowmini Varadhan { 357744b099c4SSowmini Varadhan ire_t *gw_ire, *new_ire; 357844b099c4SSowmini Varadhan int match_flags = MATCH_IRE_TYPE; 357944b099c4SSowmini Varadhan ill_t *gw_ill; 358044b099c4SSowmini Varadhan boolean_t isv6 = (ire->ire_ipversion == IPV6_VERSION); 358144b099c4SSowmini Varadhan ip_stack_t *ipst = ire->ire_ipst; 358244b099c4SSowmini Varadhan 358344b099c4SSowmini Varadhan ASSERT(ire->ire_unbound); 358444b099c4SSowmini Varadhan again: 358544b099c4SSowmini Varadhan if (isv6) { 358644b099c4SSowmini Varadhan gw_ire = ire_ftable_lookup_v6(&ire->ire_gateway_addr_v6, 0, 0, 358744b099c4SSowmini Varadhan IRE_INTERFACE, NULL, ALL_ZONES, NULL, match_flags, 0, 358844b099c4SSowmini Varadhan ipst, NULL); 358944b099c4SSowmini Varadhan } else { 359044b099c4SSowmini Varadhan gw_ire = ire_ftable_lookup_v4(ire->ire_gateway_addr, 0, 0, 359144b099c4SSowmini Varadhan IRE_INTERFACE, NULL, ALL_ZONES, NULL, match_flags, 0, 359244b099c4SSowmini Varadhan ipst, NULL); 359344b099c4SSowmini Varadhan } 359444b099c4SSowmini Varadhan if (gw_ire == NULL) { 359544b099c4SSowmini Varadhan /* see comments in ip_rt_add[_v6]() for IPMP */ 359644b099c4SSowmini Varadhan if (match_flags & MATCH_IRE_TESTHIDDEN) 359744b099c4SSowmini Varadhan return; 359844b099c4SSowmini Varadhan 359944b099c4SSowmini Varadhan match_flags |= MATCH_IRE_TESTHIDDEN; 360044b099c4SSowmini Varadhan goto again; 360144b099c4SSowmini Varadhan } 360244b099c4SSowmini Varadhan gw_ill = gw_ire->ire_ill; 360344b099c4SSowmini Varadhan if (isv6) { 360444b099c4SSowmini Varadhan new_ire = ire_create_v6(&ire->ire_addr_v6, &ire->ire_mask_v6, 360544b099c4SSowmini Varadhan &ire->ire_gateway_addr_v6, ire->ire_type, gw_ill, 360644b099c4SSowmini Varadhan ire->ire_zoneid, ire->ire_flags, NULL, ipst); 360744b099c4SSowmini Varadhan } else { 360844b099c4SSowmini Varadhan new_ire = ire_create((uchar_t *)&ire->ire_addr, 360944b099c4SSowmini Varadhan (uchar_t *)&ire->ire_mask, 361044b099c4SSowmini Varadhan (uchar_t *)&ire->ire_gateway_addr, ire->ire_type, gw_ill, 361144b099c4SSowmini Varadhan ire->ire_zoneid, ire->ire_flags, NULL, ipst); 361244b099c4SSowmini Varadhan } 361344b099c4SSowmini Varadhan ire_refrele(gw_ire); 361444b099c4SSowmini Varadhan if (new_ire == NULL) 361544b099c4SSowmini Varadhan return; 361644b099c4SSowmini Varadhan new_ire->ire_unbound = B_TRUE; 361744b099c4SSowmini Varadhan new_ire = ire_add(new_ire); 361844b099c4SSowmini Varadhan if (new_ire != NULL) 361944b099c4SSowmini Varadhan ire_refrele(new_ire); 362044b099c4SSowmini Varadhan } 3621