1 /* 2 * CDDL HEADER START 3 * 4 * The contents of this file are subject to the terms of the 5 * Common Development and Distribution License (the "License"). 6 * You may not use this file except in compliance with the License. 7 * 8 * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE 9 * or http://www.opensolaris.org/os/licensing. 10 * See the License for the specific language governing permissions 11 * and limitations under the License. 12 * 13 * When distributing Covered Code, include this CDDL HEADER in each 14 * file and include the License file at usr/src/OPENSOLARIS.LICENSE. 15 * If applicable, add the following below this CDDL HEADER, with the 16 * fields enclosed by brackets "[]" replaced with your own identifying 17 * information: Portions Copyright [yyyy] [name of copyright owner] 18 * 19 * CDDL HEADER END 20 */ 21 /* 22 * Copyright 2009 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 /* Copyright (c) 1990 Mentat Inc. */ 26 27 /* 28 * This file contains routines that manipulate Internet Routing Entries (IREs). 29 */ 30 31 #include <sys/types.h> 32 #include <sys/stream.h> 33 #include <sys/stropts.h> 34 #include <sys/strsun.h> 35 #include <sys/strsubr.h> 36 #include <sys/ddi.h> 37 #include <sys/cmn_err.h> 38 #include <sys/policy.h> 39 40 #include <sys/systm.h> 41 #include <sys/kmem.h> 42 #include <sys/param.h> 43 #include <sys/socket.h> 44 #include <net/if.h> 45 #include <net/route.h> 46 #include <netinet/in.h> 47 #include <net/if_dl.h> 48 #include <netinet/ip6.h> 49 #include <netinet/icmp6.h> 50 51 #include <inet/common.h> 52 #include <inet/mi.h> 53 #include <inet/ip.h> 54 #include <inet/ip6.h> 55 #include <inet/ip_ndp.h> 56 #include <inet/arp.h> 57 #include <inet/ip_if.h> 58 #include <inet/ip_ire.h> 59 #include <inet/ip_ftable.h> 60 #include <inet/ip_rts.h> 61 #include <inet/nd.h> 62 63 #include <inet/tcp.h> 64 #include <inet/ipclassifier.h> 65 #include <sys/zone.h> 66 #include <sys/cpuvar.h> 67 68 #include <sys/tsol/label.h> 69 #include <sys/tsol/tnet.h> 70 71 struct kmem_cache *rt_entry_cache; 72 73 typedef struct nce_clookup_s { 74 ipaddr_t ncecl_addr; 75 boolean_t ncecl_found; 76 } nce_clookup_t; 77 78 /* 79 * Synchronization notes: 80 * 81 * The fields of the ire_t struct are protected in the following way : 82 * 83 * ire_next/ire_ptpn 84 * 85 * - bucket lock of the forwarding table in which is ire stored. 86 * 87 * ire_ill, ire_u *except* ire_gateway_addr[v6], ire_mask, 88 * ire_type, ire_create_time, ire_masklen, ire_ipversion, ire_flags, 89 * ire_bucket 90 * 91 * - Set in ire_create_v4/v6 and never changes after that. Thus, 92 * we don't need a lock whenever these fields are accessed. 93 * 94 * - ire_bucket and ire_masklen (also set in ire_create) is set in 95 * ire_add before inserting in the bucket and never 96 * changes after that. Thus we don't need a lock whenever these 97 * fields are accessed. 98 * 99 * ire_gateway_addr_v4[v6] 100 * 101 * - ire_gateway_addr_v4[v6] is set during ire_create and later modified 102 * by rts_setgwr[v6]. As ire_gateway_addr is a uint32_t, updates to 103 * it assumed to be atomic and hence the other parts of the code 104 * does not use any locks. ire_gateway_addr_v6 updates are not atomic 105 * and hence any access to it uses ire_lock to get/set the right value. 106 * 107 * ire_refcnt, ire_identical_ref 108 * 109 * - Updated atomically using atomic_add_32 110 * 111 * ire_ssthresh, ire_rtt_sd, ire_rtt, ire_ib_pkt_count, ire_ob_pkt_count 112 * 113 * - Assumes that 32 bit writes are atomic. No locks. ire_lock is 114 * used to serialize updates to ire_ssthresh, ire_rtt_sd, ire_rtt. 115 * 116 * ire_generation 117 * - Under ire_lock 118 * 119 * ire_nce_cache 120 * - Under ire_lock 121 * 122 * ire_dep_parent (To next IRE in recursive lookup chain) 123 * - Under ips_ire_dep_lock. Write held when modifying. Read held when 124 * walking. We also hold ire_lock when modifying to allow the data path 125 * to only acquire ire_lock. 126 * 127 * ire_dep_parent_generation (Generation number from ire_dep_parent) 128 * - Under ips_ire_dep_lock and/or ire_lock. (A read claim on the dep_lock 129 * and ire_lock held when modifying) 130 * 131 * ire_dep_children (From parent to first child) 132 * ire_dep_sib_next (linked list of siblings) 133 * ire_dep_sib_ptpn (linked list of siblings) 134 * - Under ips_ire_dep_lock. Write held when modifying. Read held when 135 * walking. 136 * 137 * As we always hold the bucket locks in all the places while accessing 138 * the above values, it is natural to use them for protecting them. 139 * 140 * We have a forwarding table for IPv4 and IPv6. The IPv6 forwarding table 141 * (ip_forwarding_table_v6) is an array of pointers to arrays of irb_t 142 * structures. ip_forwarding_table_v6 is allocated dynamically in 143 * ire_add_v6. ire_ft_init_lock is used to serialize multiple threads 144 * initializing the same bucket. Once a bucket is initialized, it is never 145 * de-alloacted. This assumption enables us to access 146 * ip_forwarding_table_v6[i] without any locks. 147 * 148 * The forwarding table for IPv4 is a radix tree whose leaves 149 * are rt_entry structures containing the irb_t for the rt_dst. The irb_t 150 * for IPv4 is dynamically allocated and freed. 151 * 152 * Each irb_t - ire bucket structure has a lock to protect 153 * a bucket and the ires residing in the bucket have a back pointer to 154 * the bucket structure. It also has a reference count for the number 155 * of threads walking the bucket - irb_refcnt which is bumped up 156 * using the irb_refhold function. The flags irb_marks can be 157 * set to IRB_MARK_CONDEMNED indicating that there are some ires 158 * in this bucket that are IRE_IS_CONDEMNED and the 159 * last thread to leave the bucket should delete the ires. Usually 160 * this is done by the irb_refrele function which is used to decrement 161 * the reference count on a bucket. See comments above irb_t structure 162 * definition in ip.h for further details. 163 * 164 * The ire_refhold/ire_refrele functions operate on the ire which increments/ 165 * decrements the reference count, ire_refcnt, atomically on the ire. 166 * ire_refcnt is modified only using those functions. Operations on the IRE 167 * could be described as follows : 168 * 169 * CREATE an ire with reference count initialized to 1. 170 * 171 * ADDITION of an ire holds the bucket lock, checks for duplicates 172 * and then adds the ire. ire_add returns the ire after 173 * bumping up once more i.e the reference count is 2. This is to avoid 174 * an extra lookup in the functions calling ire_add which wants to 175 * work with the ire after adding. 176 * 177 * LOOKUP of an ire bumps up the reference count using ire_refhold 178 * function. It is valid to bump up the referece count of the IRE, 179 * after the lookup has returned an ire. Following are the lookup 180 * functions that return an HELD ire : 181 * 182 * ire_ftable_lookup[_v6], ire_lookup_multi_ill[_v6] 183 * 184 * DELETION of an ire holds the bucket lock, removes it from the list 185 * and then decrements the reference count for having removed from the list 186 * by using the ire_refrele function. If some other thread has looked up 187 * the ire, the reference count would have been bumped up and hence 188 * this ire will not be freed once deleted. It will be freed once the 189 * reference count drops to zero. 190 * 191 * Add and Delete acquires the bucket lock as RW_WRITER, while all the 192 * lookups acquire the bucket lock as RW_READER. 193 * 194 * The general rule is to do the ire_refrele in the function 195 * that is passing the ire as an argument. 196 * 197 * In trying to locate ires the following points are to be noted. 198 * 199 * IRE_IS_CONDEMNED signifies that the ire has been logically deleted and is 200 * to be ignored when walking the ires using ire_next. 201 * 202 * Zones note: 203 * Walking IREs within a given zone also walks certain ires in other 204 * zones. This is done intentionally. IRE walks with a specified 205 * zoneid are used only when doing informational reports, and 206 * zone users want to see things that they can access. See block 207 * comment in ire_walk_ill_match(). 208 */ 209 210 /* 211 * The size of the forwarding table. We will make sure that it is a 212 * power of 2 in ip_ire_init(). 213 * Setable in /etc/system 214 */ 215 uint32_t ip6_ftable_hash_size = IP6_FTABLE_HASH_SIZE; 216 217 struct kmem_cache *ire_cache; 218 struct kmem_cache *ncec_cache; 219 struct kmem_cache *nce_cache; 220 221 static ire_t ire_null; 222 223 static ire_t *ire_add_v4(ire_t *ire); 224 static void ire_delete_v4(ire_t *ire); 225 static void ire_dep_invalidate_children(ire_t *child); 226 static void ire_walk_ipvers(pfv_t func, void *arg, uchar_t vers, 227 zoneid_t zoneid, ip_stack_t *); 228 static void ire_walk_ill_ipvers(uint_t match_flags, uint_t ire_type, 229 pfv_t func, void *arg, uchar_t vers, ill_t *ill); 230 #ifdef DEBUG 231 static void ire_trace_cleanup(const ire_t *); 232 #endif 233 234 /* 235 * Following are the functions to increment/decrement the reference 236 * count of the IREs and IRBs (ire bucket). 237 * 238 * 1) We bump up the reference count of an IRE to make sure that 239 * it does not get deleted and freed while we are using it. 240 * Typically all the lookup functions hold the bucket lock, 241 * and look for the IRE. If it finds an IRE, it bumps up the 242 * reference count before dropping the lock. Sometimes we *may* want 243 * to bump up the reference count after we *looked* up i.e without 244 * holding the bucket lock. So, the ire_refhold function does not assert 245 * on the bucket lock being held. Any thread trying to delete from 246 * the hash bucket can still do so but cannot free the IRE if 247 * ire_refcnt is not 0. 248 * 249 * 2) We bump up the reference count on the bucket where the IRE resides 250 * (IRB), when we want to prevent the IREs getting deleted from a given 251 * hash bucket. This makes life easier for ire_walk type functions which 252 * wants to walk the IRE list, call a function, but needs to drop 253 * the bucket lock to prevent recursive rw_enters. While the 254 * lock is dropped, the list could be changed by other threads or 255 * the same thread could end up deleting the ire or the ire pointed by 256 * ire_next. ire_refholding the ire or ire_next is not sufficient as 257 * a delete will still remove the ire from the bucket while we have 258 * dropped the lock and hence the ire_next would be NULL. Thus, we 259 * need a mechanism to prevent deletions from a given bucket. 260 * 261 * To prevent deletions, we bump up the reference count on the 262 * bucket. If the bucket is held, ire_delete just marks both 263 * the ire and irb as CONDEMNED. When the 264 * reference count on the bucket drops to zero, all the CONDEMNED ires 265 * are deleted. We don't have to bump up the reference count on the 266 * bucket if we are walking the bucket and never have to drop the bucket 267 * lock. Note that irb_refhold does not prevent addition of new ires 268 * in the list. It is okay because addition of new ires will not cause 269 * ire_next to point to freed memory. We do irb_refhold only when 270 * all of the 3 conditions are true : 271 * 272 * 1) The code needs to walk the IRE bucket from start to end. 273 * 2) It may have to drop the bucket lock sometimes while doing (1) 274 * 3) It does not want any ires to be deleted meanwhile. 275 */ 276 277 /* 278 * Bump up the reference count on the hash bucket - IRB to 279 * prevent ires from being deleted in this bucket. 280 */ 281 void 282 irb_refhold(irb_t *irb) 283 { 284 rw_enter(&irb->irb_lock, RW_WRITER); 285 irb->irb_refcnt++; 286 ASSERT(irb->irb_refcnt != 0); 287 rw_exit(&irb->irb_lock); 288 } 289 290 void 291 irb_refhold_locked(irb_t *irb) 292 { 293 ASSERT(RW_WRITE_HELD(&irb->irb_lock)); 294 irb->irb_refcnt++; 295 ASSERT(irb->irb_refcnt != 0); 296 } 297 298 /* 299 * Note: when IRB_MARK_DYNAMIC is not set the irb_t 300 * is statically allocated, so that when the irb_refcnt goes to 0, 301 * we simply clean up the ire list and continue. 302 */ 303 void 304 irb_refrele(irb_t *irb) 305 { 306 if (irb->irb_marks & IRB_MARK_DYNAMIC) { 307 irb_refrele_ftable(irb); 308 } else { 309 rw_enter(&irb->irb_lock, RW_WRITER); 310 ASSERT(irb->irb_refcnt != 0); 311 if (--irb->irb_refcnt == 0 && 312 (irb->irb_marks & IRB_MARK_CONDEMNED)) { 313 ire_t *ire_list; 314 315 ire_list = ire_unlink(irb); 316 rw_exit(&irb->irb_lock); 317 ASSERT(ire_list != NULL); 318 ire_cleanup(ire_list); 319 } else { 320 rw_exit(&irb->irb_lock); 321 } 322 } 323 } 324 325 326 /* 327 * Bump up the reference count on the IRE. We cannot assert that the 328 * bucket lock is being held as it is legal to bump up the reference 329 * count after the first lookup has returned the IRE without 330 * holding the lock. 331 */ 332 void 333 ire_refhold(ire_t *ire) 334 { 335 atomic_add_32(&(ire)->ire_refcnt, 1); 336 ASSERT((ire)->ire_refcnt != 0); 337 #ifdef DEBUG 338 ire_trace_ref(ire); 339 #endif 340 } 341 342 void 343 ire_refhold_notr(ire_t *ire) 344 { 345 atomic_add_32(&(ire)->ire_refcnt, 1); 346 ASSERT((ire)->ire_refcnt != 0); 347 } 348 349 void 350 ire_refhold_locked(ire_t *ire) 351 { 352 #ifdef DEBUG 353 ire_trace_ref(ire); 354 #endif 355 ire->ire_refcnt++; 356 } 357 358 /* 359 * Release a ref on an IRE. 360 * 361 * Must not be called while holding any locks. Otherwise if this is 362 * the last reference to be released there is a chance of recursive mutex 363 * panic due to ire_refrele -> ipif_ill_refrele_tail -> qwriter_ip trying 364 * to restart an ioctl. The one exception is when the caller is sure that 365 * this is not the last reference to be released. Eg. if the caller is 366 * sure that the ire has not been deleted and won't be deleted. 367 * 368 * In architectures e.g sun4u, where atomic_add_32_nv is just 369 * a cas, we need to maintain the right memory barrier semantics 370 * as that of mutex_exit i.e all the loads and stores should complete 371 * before the cas is executed. membar_exit() does that here. 372 */ 373 void 374 ire_refrele(ire_t *ire) 375 { 376 #ifdef DEBUG 377 ire_untrace_ref(ire); 378 #endif 379 ASSERT((ire)->ire_refcnt != 0); 380 membar_exit(); 381 if (atomic_add_32_nv(&(ire)->ire_refcnt, -1) == 0) 382 ire_inactive(ire); 383 } 384 385 void 386 ire_refrele_notr(ire_t *ire) 387 { 388 ASSERT((ire)->ire_refcnt != 0); 389 membar_exit(); 390 if (atomic_add_32_nv(&(ire)->ire_refcnt, -1) == 0) 391 ire_inactive(ire); 392 } 393 394 /* 395 * This function is associated with the IP_IOC_IRE_DELETE[_NO_REPLY] 396 * IOCTL[s]. The NO_REPLY form is used by TCP to tell IP that it is 397 * having problems reaching a particular destination. 398 * This will make IP consider alternate routes (e.g., when there are 399 * muliple default routes), and it will also make IP discard any (potentially) 400 * stale redirect. 401 * Management processes may want to use the version that generates a reply. 402 * 403 * With the use of NUD like behavior for IPv4/ARP in addition to IPv6 404 * this function shouldn't be necessary for IP to recover from a bad redirect, 405 * a bad default router (when there are multiple default routers), or 406 * a stale ND/ARP entry. But we retain it in any case. 407 * For instance, this is helpful when TCP suspects a failure before NUD does. 408 */ 409 int 410 ip_ire_delete(queue_t *q, mblk_t *mp, cred_t *ioc_cr) 411 { 412 uchar_t *addr_ucp; 413 uint_t ipversion; 414 sin_t *sin; 415 sin6_t *sin6; 416 ipaddr_t v4addr; 417 in6_addr_t v6addr; 418 ire_t *ire; 419 ipid_t *ipid; 420 zoneid_t zoneid; 421 ip_stack_t *ipst; 422 423 ASSERT(q->q_next == NULL); 424 zoneid = IPCL_ZONEID(Q_TO_CONN(q)); 425 ipst = CONNQ_TO_IPST(q); 426 427 /* 428 * Check privilege using the ioctl credential; if it is NULL 429 * then this is a kernel message and therefor privileged. 430 */ 431 if (ioc_cr != NULL && secpolicy_ip_config(ioc_cr, B_FALSE) != 0) 432 return (EPERM); 433 434 ipid = (ipid_t *)mp->b_rptr; 435 436 addr_ucp = mi_offset_param(mp, ipid->ipid_addr_offset, 437 ipid->ipid_addr_length); 438 if (addr_ucp == NULL || !OK_32PTR(addr_ucp)) 439 return (EINVAL); 440 switch (ipid->ipid_addr_length) { 441 case sizeof (sin_t): 442 /* 443 * got complete (sockaddr) address - increment addr_ucp to point 444 * at the ip_addr field. 445 */ 446 sin = (sin_t *)addr_ucp; 447 addr_ucp = (uchar_t *)&sin->sin_addr.s_addr; 448 ipversion = IPV4_VERSION; 449 break; 450 case sizeof (sin6_t): 451 /* 452 * got complete (sockaddr) address - increment addr_ucp to point 453 * at the ip_addr field. 454 */ 455 sin6 = (sin6_t *)addr_ucp; 456 addr_ucp = (uchar_t *)&sin6->sin6_addr; 457 ipversion = IPV6_VERSION; 458 break; 459 default: 460 return (EINVAL); 461 } 462 if (ipversion == IPV4_VERSION) { 463 /* Extract the destination address. */ 464 bcopy(addr_ucp, &v4addr, IP_ADDR_LEN); 465 466 ire = ire_ftable_lookup_v4(v4addr, 0, 0, 0, NULL, 467 zoneid, NULL, MATCH_IRE_DSTONLY, 0, ipst, NULL); 468 } else { 469 /* Extract the destination address. */ 470 bcopy(addr_ucp, &v6addr, IPV6_ADDR_LEN); 471 472 ire = ire_ftable_lookup_v6(&v6addr, NULL, NULL, 0, NULL, 473 zoneid, NULL, MATCH_IRE_DSTONLY, 0, ipst, NULL); 474 } 475 if (ire != NULL) { 476 if (ipversion == IPV4_VERSION) { 477 ip_rts_change(RTM_LOSING, ire->ire_addr, 478 ire->ire_gateway_addr, ire->ire_mask, 479 (Q_TO_CONN(q))->conn_laddr_v4, 0, 0, 0, 480 (RTA_DST | RTA_GATEWAY | RTA_NETMASK | RTA_IFA), 481 ire->ire_ipst); 482 } 483 (void) ire_no_good(ire); 484 ire_refrele(ire); 485 } 486 return (0); 487 } 488 489 /* 490 * Initialize the ire that is specific to IPv4 part and call 491 * ire_init_common to finish it. 492 * Returns zero or errno. 493 */ 494 int 495 ire_init_v4(ire_t *ire, uchar_t *addr, uchar_t *mask, uchar_t *gateway, 496 ushort_t type, ill_t *ill, zoneid_t zoneid, uint_t flags, 497 tsol_gc_t *gc, ip_stack_t *ipst) 498 { 499 int error; 500 501 /* 502 * Reject IRE security attribute creation/initialization 503 * if system is not running in Trusted mode. 504 */ 505 if (gc != NULL && !is_system_labeled()) 506 return (EINVAL); 507 508 BUMP_IRE_STATS(ipst->ips_ire_stats_v4, ire_stats_alloced); 509 510 if (addr != NULL) 511 bcopy(addr, &ire->ire_addr, IP_ADDR_LEN); 512 if (gateway != NULL) 513 bcopy(gateway, &ire->ire_gateway_addr, IP_ADDR_LEN); 514 515 /* Make sure we don't have stray values in some fields */ 516 switch (type) { 517 case IRE_LOOPBACK: 518 bcopy(&ire->ire_addr, &ire->ire_gateway_addr, IP_ADDR_LEN); 519 /* FALLTHRU */ 520 case IRE_HOST: 521 case IRE_BROADCAST: 522 case IRE_LOCAL: 523 case IRE_IF_CLONE: 524 ire->ire_mask = IP_HOST_MASK; 525 ire->ire_masklen = IPV4_ABITS; 526 break; 527 case IRE_PREFIX: 528 case IRE_DEFAULT: 529 case IRE_IF_RESOLVER: 530 case IRE_IF_NORESOLVER: 531 if (mask != NULL) { 532 bcopy(mask, &ire->ire_mask, IP_ADDR_LEN); 533 ire->ire_masklen = ip_mask_to_plen(ire->ire_mask); 534 } 535 break; 536 case IRE_MULTICAST: 537 case IRE_NOROUTE: 538 ASSERT(mask == NULL); 539 break; 540 default: 541 ASSERT(0); 542 return (EINVAL); 543 } 544 545 error = ire_init_common(ire, type, ill, zoneid, flags, IPV4_VERSION, 546 gc, ipst); 547 if (error != NULL) 548 return (error); 549 550 /* Determine which function pointers to use */ 551 ire->ire_postfragfn = ip_xmit; /* Common case */ 552 553 switch (ire->ire_type) { 554 case IRE_LOCAL: 555 ire->ire_sendfn = ire_send_local_v4; 556 ire->ire_recvfn = ire_recv_local_v4; 557 ASSERT(ire->ire_ill != NULL); 558 if (ire->ire_ill->ill_flags & ILLF_NOACCEPT) 559 ire->ire_recvfn = ire_recv_noaccept_v6; 560 break; 561 case IRE_LOOPBACK: 562 ire->ire_sendfn = ire_send_local_v4; 563 ire->ire_recvfn = ire_recv_loopback_v4; 564 break; 565 case IRE_BROADCAST: 566 ire->ire_postfragfn = ip_postfrag_loopcheck; 567 ire->ire_sendfn = ire_send_broadcast_v4; 568 ire->ire_recvfn = ire_recv_broadcast_v4; 569 break; 570 case IRE_MULTICAST: 571 ire->ire_postfragfn = ip_postfrag_loopcheck; 572 ire->ire_sendfn = ire_send_multicast_v4; 573 ire->ire_recvfn = ire_recv_multicast_v4; 574 break; 575 default: 576 /* 577 * For IRE_IF_ALL and IRE_OFFLINK we forward received 578 * packets by default. 579 */ 580 ire->ire_sendfn = ire_send_wire_v4; 581 ire->ire_recvfn = ire_recv_forward_v4; 582 break; 583 } 584 if (ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) { 585 ire->ire_sendfn = ire_send_noroute_v4; 586 ire->ire_recvfn = ire_recv_noroute_v4; 587 } else if (ire->ire_flags & RTF_MULTIRT) { 588 ire->ire_postfragfn = ip_postfrag_multirt_v4; 589 ire->ire_sendfn = ire_send_multirt_v4; 590 /* Multirt receive of broadcast uses ire_recv_broadcast_v4 */ 591 if (ire->ire_type != IRE_BROADCAST) 592 ire->ire_recvfn = ire_recv_multirt_v4; 593 } 594 ire->ire_nce_capable = ire_determine_nce_capable(ire); 595 return (0); 596 } 597 598 /* 599 * Determine ire_nce_capable 600 */ 601 boolean_t 602 ire_determine_nce_capable(ire_t *ire) 603 { 604 int max_masklen; 605 606 if ((ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) || 607 (ire->ire_type & IRE_MULTICAST)) 608 return (B_TRUE); 609 610 if (ire->ire_ipversion == IPV4_VERSION) 611 max_masklen = IPV4_ABITS; 612 else 613 max_masklen = IPV6_ABITS; 614 615 if ((ire->ire_type & IRE_ONLINK) && ire->ire_masklen == max_masklen) 616 return (B_TRUE); 617 return (B_FALSE); 618 } 619 620 /* 621 * ire_create is called to allocate and initialize a new IRE. 622 * 623 * NOTE : This is called as writer sometimes though not required 624 * by this function. 625 */ 626 ire_t * 627 ire_create(uchar_t *addr, uchar_t *mask, uchar_t *gateway, 628 ushort_t type, ill_t *ill, zoneid_t zoneid, uint_t flags, tsol_gc_t *gc, 629 ip_stack_t *ipst) 630 { 631 ire_t *ire; 632 int error; 633 634 ire = kmem_cache_alloc(ire_cache, KM_NOSLEEP); 635 if (ire == NULL) { 636 DTRACE_PROBE(kmem__cache__alloc); 637 return (NULL); 638 } 639 *ire = ire_null; 640 641 error = ire_init_v4(ire, addr, mask, gateway, type, ill, zoneid, flags, 642 gc, ipst); 643 if (error != 0) { 644 DTRACE_PROBE2(ire__init, ire_t *, ire, int, error); 645 kmem_cache_free(ire_cache, ire); 646 return (NULL); 647 } 648 return (ire); 649 } 650 651 /* 652 * Common to IPv4 and IPv6 653 * Returns zero or errno. 654 */ 655 int 656 ire_init_common(ire_t *ire, ushort_t type, ill_t *ill, zoneid_t zoneid, 657 uint_t flags, uchar_t ipversion, tsol_gc_t *gc, ip_stack_t *ipst) 658 { 659 int error; 660 661 #ifdef DEBUG 662 if (ill != NULL) { 663 if (ill->ill_isv6) 664 ASSERT(ipversion == IPV6_VERSION); 665 else 666 ASSERT(ipversion == IPV4_VERSION); 667 } 668 #endif /* DEBUG */ 669 670 /* 671 * Create/initialize IRE security attribute only in Trusted mode; 672 * if the passed in gc is non-NULL, we expect that the caller 673 * has held a reference to it and will release it when this routine 674 * returns a failure, otherwise we own the reference. We do this 675 * prior to initializing the rest IRE fields. 676 */ 677 if (is_system_labeled()) { 678 if ((type & (IRE_LOCAL | IRE_LOOPBACK | IRE_BROADCAST | 679 IRE_IF_ALL | IRE_MULTICAST | IRE_NOROUTE)) != 0) { 680 /* release references on behalf of caller */ 681 if (gc != NULL) 682 GC_REFRELE(gc); 683 } else { 684 error = tsol_ire_init_gwattr(ire, ipversion, gc); 685 if (error != 0) 686 return (error); 687 } 688 } 689 690 ire->ire_type = type; 691 ire->ire_flags = RTF_UP | flags; 692 ire->ire_create_time = (uint32_t)gethrestime_sec(); 693 ire->ire_generation = IRE_GENERATION_INITIAL; 694 695 /* 696 * The ill_ire_cnt isn't increased until 697 * the IRE is added to ensure that a walker will find 698 * all IREs that hold a reference on an ill. 699 * 700 * Note that ill_ire_multicast doesn't hold a ref on the ill since 701 * ire_add() is not called for the IRE_MULTICAST. 702 */ 703 ire->ire_ill = ill; 704 ire->ire_zoneid = zoneid; 705 ire->ire_ipversion = ipversion; 706 707 mutex_init(&ire->ire_lock, NULL, MUTEX_DEFAULT, NULL); 708 ire->ire_refcnt = 1; 709 ire->ire_identical_ref = 1; /* Number of ire_delete's needed */ 710 ire->ire_ipst = ipst; /* No netstack_hold */ 711 ire->ire_trace_disable = B_FALSE; 712 713 return (0); 714 } 715 716 /* 717 * This creates an IRE_BROADCAST based on the arguments. 718 * A mirror is ire_lookup_bcast(). 719 * 720 * Any supression of unneeded ones is done in ire_add_v4. 721 * We add one IRE_BROADCAST per address. ire_send_broadcast_v4() 722 * takes care of generating a loopback copy of the packet. 723 */ 724 ire_t ** 725 ire_create_bcast(ill_t *ill, ipaddr_t addr, zoneid_t zoneid, ire_t **irep) 726 { 727 ip_stack_t *ipst = ill->ill_ipst; 728 729 ASSERT(IAM_WRITER_ILL(ill)); 730 731 *irep++ = ire_create( 732 (uchar_t *)&addr, /* dest addr */ 733 (uchar_t *)&ip_g_all_ones, /* mask */ 734 NULL, /* no gateway */ 735 IRE_BROADCAST, 736 ill, 737 zoneid, 738 RTF_KERNEL, 739 NULL, 740 ipst); 741 742 return (irep); 743 } 744 745 /* 746 * This looks up an IRE_BROADCAST based on the arguments. 747 * Mirrors ire_create_bcast(). 748 */ 749 ire_t * 750 ire_lookup_bcast(ill_t *ill, ipaddr_t addr, zoneid_t zoneid) 751 { 752 ire_t *ire; 753 int match_args; 754 755 match_args = MATCH_IRE_TYPE | MATCH_IRE_ILL | MATCH_IRE_GW | 756 MATCH_IRE_MASK | MATCH_IRE_ZONEONLY; 757 758 if (IS_UNDER_IPMP(ill)) 759 match_args |= MATCH_IRE_TESTHIDDEN; 760 761 ire = ire_ftable_lookup_v4( 762 addr, /* dest addr */ 763 ip_g_all_ones, /* mask */ 764 0, /* no gateway */ 765 IRE_BROADCAST, 766 ill, 767 zoneid, 768 NULL, 769 match_args, 770 0, 771 ill->ill_ipst, 772 NULL); 773 return (ire); 774 } 775 776 /* Arrange to call the specified function for every IRE in the world. */ 777 void 778 ire_walk(pfv_t func, void *arg, ip_stack_t *ipst) 779 { 780 ire_walk_ipvers(func, arg, 0, ALL_ZONES, ipst); 781 } 782 783 void 784 ire_walk_v4(pfv_t func, void *arg, zoneid_t zoneid, ip_stack_t *ipst) 785 { 786 ire_walk_ipvers(func, arg, IPV4_VERSION, zoneid, ipst); 787 } 788 789 void 790 ire_walk_v6(pfv_t func, void *arg, zoneid_t zoneid, ip_stack_t *ipst) 791 { 792 ire_walk_ipvers(func, arg, IPV6_VERSION, zoneid, ipst); 793 } 794 795 /* 796 * Walk a particular version. version == 0 means both v4 and v6. 797 */ 798 static void 799 ire_walk_ipvers(pfv_t func, void *arg, uchar_t vers, zoneid_t zoneid, 800 ip_stack_t *ipst) 801 { 802 if (vers != IPV6_VERSION) { 803 /* 804 * ip_forwarding_table variable doesn't matter for IPv4 since 805 * ire_walk_ill_tables uses ips_ip_ftable for IPv4. 806 */ 807 ire_walk_ill_tables(0, 0, func, arg, IP_MASK_TABLE_SIZE, 808 0, NULL, 809 NULL, zoneid, ipst); 810 } 811 if (vers != IPV4_VERSION) { 812 ire_walk_ill_tables(0, 0, func, arg, IP6_MASK_TABLE_SIZE, 813 ipst->ips_ip6_ftable_hash_size, 814 ipst->ips_ip_forwarding_table_v6, 815 NULL, zoneid, ipst); 816 } 817 } 818 819 /* 820 * Arrange to call the specified function for every IRE that matches the ill. 821 */ 822 void 823 ire_walk_ill(uint_t match_flags, uint_t ire_type, pfv_t func, void *arg, 824 ill_t *ill) 825 { 826 uchar_t vers = (ill->ill_isv6 ? IPV6_VERSION : IPV4_VERSION); 827 828 ire_walk_ill_ipvers(match_flags, ire_type, func, arg, vers, ill); 829 } 830 831 /* 832 * Walk a particular ill and version. 833 */ 834 static void 835 ire_walk_ill_ipvers(uint_t match_flags, uint_t ire_type, pfv_t func, 836 void *arg, uchar_t vers, ill_t *ill) 837 { 838 ip_stack_t *ipst = ill->ill_ipst; 839 840 if (vers == IPV4_VERSION) { 841 ire_walk_ill_tables(match_flags, ire_type, func, arg, 842 IP_MASK_TABLE_SIZE, 843 0, NULL, 844 ill, ALL_ZONES, ipst); 845 } 846 if (vers != IPV4_VERSION) { 847 ire_walk_ill_tables(match_flags, ire_type, func, arg, 848 IP6_MASK_TABLE_SIZE, ipst->ips_ip6_ftable_hash_size, 849 ipst->ips_ip_forwarding_table_v6, 850 ill, ALL_ZONES, ipst); 851 } 852 } 853 854 /* 855 * Do the specific matching of IREs to shared-IP zones. 856 * 857 * We have the same logic as in ire_match_args but implemented slightly 858 * differently. 859 */ 860 boolean_t 861 ire_walk_ill_match(uint_t match_flags, uint_t ire_type, ire_t *ire, 862 ill_t *ill, zoneid_t zoneid, ip_stack_t *ipst) 863 { 864 ill_t *dst_ill = ire->ire_ill; 865 866 ASSERT(match_flags != 0 || zoneid != ALL_ZONES); 867 868 if (zoneid != ALL_ZONES && zoneid != ire->ire_zoneid && 869 ire->ire_zoneid != ALL_ZONES) { 870 /* 871 * We're walking the IREs for a specific zone. The only relevant 872 * IREs are: 873 * - all IREs with a matching ire_zoneid 874 * - IRE_IF_ALL IREs for interfaces with a usable source addr 875 * with a matching zone 876 * - IRE_OFFLINK with a gateway reachable from the zone 877 * Note that ealier we only did the IRE_OFFLINK check for 878 * IRE_DEFAULT (and only when we had multiple IRE_DEFAULTs). 879 */ 880 if (ire->ire_type & IRE_ONLINK) { 881 uint_t ifindex; 882 883 /* 884 * Note there is no IRE_INTERFACE on vniN thus 885 * can't do an IRE lookup for a matching route. 886 */ 887 ifindex = dst_ill->ill_usesrc_ifindex; 888 if (ifindex == 0) 889 return (B_FALSE); 890 891 /* 892 * If there is a usable source address in the 893 * zone, then it's ok to return an 894 * IRE_INTERFACE 895 */ 896 if (!ipif_zone_avail(ifindex, dst_ill->ill_isv6, 897 zoneid, ipst)) { 898 return (B_FALSE); 899 } 900 } 901 if (dst_ill != NULL && (ire->ire_type & IRE_OFFLINK)) { 902 ipif_t *tipif; 903 904 mutex_enter(&dst_ill->ill_lock); 905 for (tipif = dst_ill->ill_ipif; 906 tipif != NULL; tipif = tipif->ipif_next) { 907 if (!IPIF_IS_CONDEMNED(tipif) && 908 (tipif->ipif_flags & IPIF_UP) && 909 (tipif->ipif_zoneid == zoneid || 910 tipif->ipif_zoneid == ALL_ZONES)) 911 break; 912 } 913 mutex_exit(&dst_ill->ill_lock); 914 if (tipif == NULL) { 915 return (B_FALSE); 916 } 917 } 918 } 919 /* 920 * Except for ALL_ZONES, we only match the offlink routes 921 * where ire_gateway_addr has an IRE_INTERFACE for the zoneid. 922 */ 923 if ((ire->ire_type & IRE_OFFLINK) && zoneid != ALL_ZONES) { 924 in6_addr_t gw_addr_v6; 925 926 if (ire->ire_ipversion == IPV4_VERSION) { 927 if (!ire_gateway_ok_zone_v4(ire->ire_gateway_addr, 928 zoneid, dst_ill, NULL, ipst, B_FALSE)) 929 return (B_FALSE); 930 } else { 931 ASSERT(ire->ire_ipversion == IPV6_VERSION); 932 mutex_enter(&ire->ire_lock); 933 gw_addr_v6 = ire->ire_gateway_addr_v6; 934 mutex_exit(&ire->ire_lock); 935 936 if (!ire_gateway_ok_zone_v6(&gw_addr_v6, zoneid, 937 dst_ill, NULL, ipst, B_FALSE)) 938 return (B_FALSE); 939 } 940 } 941 942 if (((!(match_flags & MATCH_IRE_TYPE)) || 943 (ire->ire_type & ire_type)) && 944 ((!(match_flags & MATCH_IRE_ILL)) || 945 (dst_ill == ill || 946 dst_ill != NULL && IS_IN_SAME_ILLGRP(dst_ill, ill)))) { 947 return (B_TRUE); 948 } 949 return (B_FALSE); 950 } 951 952 int 953 rtfunc(struct radix_node *rn, void *arg) 954 { 955 struct rtfuncarg *rtf = arg; 956 struct rt_entry *rt; 957 irb_t *irb; 958 ire_t *ire; 959 boolean_t ret; 960 961 rt = (struct rt_entry *)rn; 962 ASSERT(rt != NULL); 963 irb = &rt->rt_irb; 964 for (ire = irb->irb_ire; ire != NULL; ire = ire->ire_next) { 965 if ((rtf->rt_match_flags != 0) || 966 (rtf->rt_zoneid != ALL_ZONES)) { 967 ret = ire_walk_ill_match(rtf->rt_match_flags, 968 rtf->rt_ire_type, ire, 969 rtf->rt_ill, rtf->rt_zoneid, rtf->rt_ipst); 970 } else { 971 ret = B_TRUE; 972 } 973 if (ret) 974 (*rtf->rt_func)(ire, rtf->rt_arg); 975 } 976 return (0); 977 } 978 979 /* 980 * Walk the ftable entries that match the ill. 981 */ 982 void 983 ire_walk_ill_tables(uint_t match_flags, uint_t ire_type, pfv_t func, 984 void *arg, size_t ftbl_sz, size_t htbl_sz, irb_t **ipftbl, 985 ill_t *ill, zoneid_t zoneid, 986 ip_stack_t *ipst) 987 { 988 irb_t *irb_ptr; 989 irb_t *irb; 990 ire_t *ire; 991 int i, j; 992 boolean_t ret; 993 struct rtfuncarg rtfarg; 994 995 ASSERT((!(match_flags & MATCH_IRE_ILL)) || (ill != NULL)); 996 ASSERT(!(match_flags & MATCH_IRE_TYPE) || (ire_type != 0)); 997 998 /* knobs such that routine is called only for v6 case */ 999 if (ipftbl == ipst->ips_ip_forwarding_table_v6) { 1000 for (i = (ftbl_sz - 1); i >= 0; i--) { 1001 if ((irb_ptr = ipftbl[i]) == NULL) 1002 continue; 1003 for (j = 0; j < htbl_sz; j++) { 1004 irb = &irb_ptr[j]; 1005 if (irb->irb_ire == NULL) 1006 continue; 1007 1008 irb_refhold(irb); 1009 for (ire = irb->irb_ire; ire != NULL; 1010 ire = ire->ire_next) { 1011 if (match_flags == 0 && 1012 zoneid == ALL_ZONES) { 1013 ret = B_TRUE; 1014 } else { 1015 ret = 1016 ire_walk_ill_match( 1017 match_flags, 1018 ire_type, ire, ill, 1019 zoneid, ipst); 1020 } 1021 if (ret) 1022 (*func)(ire, arg); 1023 } 1024 irb_refrele(irb); 1025 } 1026 } 1027 } else { 1028 bzero(&rtfarg, sizeof (rtfarg)); 1029 rtfarg.rt_func = func; 1030 rtfarg.rt_arg = arg; 1031 if (match_flags != 0) { 1032 rtfarg.rt_match_flags = match_flags; 1033 } 1034 rtfarg.rt_ire_type = ire_type; 1035 rtfarg.rt_ill = ill; 1036 rtfarg.rt_zoneid = zoneid; 1037 rtfarg.rt_ipst = ipst; /* No netstack_hold */ 1038 (void) ipst->ips_ip_ftable->rnh_walktree_mt( 1039 ipst->ips_ip_ftable, 1040 rtfunc, &rtfarg, irb_refhold_rn, irb_refrele_rn); 1041 } 1042 } 1043 1044 /* 1045 * This function takes a mask and returns 1046 * number of bits set in the mask. If no 1047 * bit is set it returns 0. 1048 * Assumes a contiguous mask. 1049 */ 1050 int 1051 ip_mask_to_plen(ipaddr_t mask) 1052 { 1053 return (mask == 0 ? 0 : IP_ABITS - (ffs(ntohl(mask)) -1)); 1054 } 1055 1056 /* 1057 * Convert length for a mask to the mask. 1058 */ 1059 ipaddr_t 1060 ip_plen_to_mask(uint_t masklen) 1061 { 1062 if (masklen == 0) 1063 return (0); 1064 1065 return (htonl(IP_HOST_MASK << (IP_ABITS - masklen))); 1066 } 1067 1068 void 1069 ire_atomic_end(irb_t *irb_ptr, ire_t *ire) 1070 { 1071 ill_t *ill; 1072 1073 ill = ire->ire_ill; 1074 if (ill != NULL) 1075 mutex_exit(&ill->ill_lock); 1076 rw_exit(&irb_ptr->irb_lock); 1077 } 1078 1079 /* 1080 * ire_add_v[46] atomically make sure that the ill associated 1081 * with the new ire is not going away i.e., we check ILL_CONDEMNED. 1082 */ 1083 int 1084 ire_atomic_start(irb_t *irb_ptr, ire_t *ire) 1085 { 1086 ill_t *ill; 1087 1088 ill = ire->ire_ill; 1089 1090 rw_enter(&irb_ptr->irb_lock, RW_WRITER); 1091 if (ill != NULL) { 1092 mutex_enter(&ill->ill_lock); 1093 1094 /* 1095 * Don't allow IRE's to be created on dying ills. 1096 */ 1097 if (ill->ill_state_flags & ILL_CONDEMNED) { 1098 ire_atomic_end(irb_ptr, ire); 1099 return (ENXIO); 1100 } 1101 1102 if (IS_UNDER_IPMP(ill)) { 1103 int error = 0; 1104 mutex_enter(&ill->ill_phyint->phyint_lock); 1105 if (!ipmp_ill_is_active(ill) && 1106 IRE_HIDDEN_TYPE(ire->ire_type) && 1107 !ire->ire_testhidden) { 1108 error = EINVAL; 1109 } 1110 mutex_exit(&ill->ill_phyint->phyint_lock); 1111 if (error != 0) { 1112 ire_atomic_end(irb_ptr, ire); 1113 return (error); 1114 } 1115 } 1116 1117 } 1118 return (0); 1119 } 1120 1121 /* 1122 * Add a fully initialized IRE to the forwarding table. 1123 * This returns NULL on failure, or a held IRE on success. 1124 * Normally the returned IRE is the same as the argument. But a different 1125 * IRE will be returned if the added IRE is deemed identical to an existing 1126 * one. In that case ire_identical_ref will be increased. 1127 * The caller always needs to do an ire_refrele() on the returned IRE. 1128 */ 1129 ire_t * 1130 ire_add(ire_t *ire) 1131 { 1132 if (IRE_HIDDEN_TYPE(ire->ire_type) && 1133 ire->ire_ill != NULL && IS_UNDER_IPMP(ire->ire_ill)) { 1134 /* 1135 * IREs hosted on interfaces that are under IPMP 1136 * should be hidden so that applications don't 1137 * accidentally end up sending packets with test 1138 * addresses as their source addresses, or 1139 * sending out interfaces that are e.g. IFF_INACTIVE. 1140 * Hide them here. 1141 */ 1142 ire->ire_testhidden = B_TRUE; 1143 } 1144 1145 if (ire->ire_ipversion == IPV6_VERSION) 1146 return (ire_add_v6(ire)); 1147 else 1148 return (ire_add_v4(ire)); 1149 } 1150 1151 /* 1152 * Add a fully initialized IPv4 IRE to the forwarding table. 1153 * This returns NULL on failure, or a held IRE on success. 1154 * Normally the returned IRE is the same as the argument. But a different 1155 * IRE will be returned if the added IRE is deemed identical to an existing 1156 * one. In that case ire_identical_ref will be increased. 1157 * The caller always needs to do an ire_refrele() on the returned IRE. 1158 */ 1159 static ire_t * 1160 ire_add_v4(ire_t *ire) 1161 { 1162 ire_t *ire1; 1163 irb_t *irb_ptr; 1164 ire_t **irep; 1165 int match_flags; 1166 int error; 1167 ip_stack_t *ipst = ire->ire_ipst; 1168 1169 if (ire->ire_ill != NULL) 1170 ASSERT(!MUTEX_HELD(&ire->ire_ill->ill_lock)); 1171 ASSERT(ire->ire_ipversion == IPV4_VERSION); 1172 1173 /* Make sure the address is properly masked. */ 1174 ire->ire_addr &= ire->ire_mask; 1175 1176 match_flags = (MATCH_IRE_MASK | MATCH_IRE_TYPE | MATCH_IRE_GW); 1177 1178 if (ire->ire_ill != NULL) { 1179 match_flags |= MATCH_IRE_ILL; 1180 } 1181 irb_ptr = ire_get_bucket(ire); 1182 if (irb_ptr == NULL) { 1183 printf("no bucket for %p\n", (void *)ire); 1184 ire_delete(ire); 1185 return (NULL); 1186 } 1187 1188 /* 1189 * Start the atomic add of the ire. Grab the ill lock, 1190 * the bucket lock. Check for condemned. 1191 */ 1192 error = ire_atomic_start(irb_ptr, ire); 1193 if (error != 0) { 1194 printf("no ire_atomic_start for %p\n", (void *)ire); 1195 ire_delete(ire); 1196 irb_refrele(irb_ptr); 1197 return (NULL); 1198 } 1199 /* 1200 * If we are creating a hidden IRE, make sure we search for 1201 * hidden IREs when searching for duplicates below. 1202 * Otherwise, we might find an IRE on some other interface 1203 * that's not marked hidden. 1204 */ 1205 if (ire->ire_testhidden) 1206 match_flags |= MATCH_IRE_TESTHIDDEN; 1207 1208 /* 1209 * Atomically check for duplicate and insert in the table. 1210 */ 1211 for (ire1 = irb_ptr->irb_ire; ire1 != NULL; ire1 = ire1->ire_next) { 1212 if (IRE_IS_CONDEMNED(ire1)) 1213 continue; 1214 /* 1215 * Here we need an exact match on zoneid, i.e., 1216 * ire_match_args doesn't fit. 1217 */ 1218 if (ire1->ire_zoneid != ire->ire_zoneid) 1219 continue; 1220 1221 if (ire1->ire_type != ire->ire_type) 1222 continue; 1223 1224 /* 1225 * Note: We do not allow multiple routes that differ only 1226 * in the gateway security attributes; such routes are 1227 * considered duplicates. 1228 * To change that we explicitly have to treat them as 1229 * different here. 1230 */ 1231 if (ire_match_args(ire1, ire->ire_addr, ire->ire_mask, 1232 ire->ire_gateway_addr, ire->ire_type, ire->ire_ill, 1233 ire->ire_zoneid, NULL, match_flags)) { 1234 /* 1235 * Return the old ire after doing a REFHOLD. 1236 * As most of the callers continue to use the IRE 1237 * after adding, we return a held ire. This will 1238 * avoid a lookup in the caller again. If the callers 1239 * don't want to use it, they need to do a REFRELE. 1240 */ 1241 atomic_add_32(&ire1->ire_identical_ref, 1); 1242 DTRACE_PROBE2(ire__add__exist, ire_t *, ire1, 1243 ire_t *, ire); 1244 ire_refhold(ire1); 1245 ire_atomic_end(irb_ptr, ire); 1246 ire_delete(ire); 1247 irb_refrele(irb_ptr); 1248 return (ire1); 1249 } 1250 } 1251 1252 /* 1253 * Normally we do head insertion since most things do not care about 1254 * the order of the IREs in the bucket. Note that ip_cgtp_bcast_add 1255 * assumes we at least do head insertion so that its IRE_BROADCAST 1256 * arrive ahead of existing IRE_HOST for the same address. 1257 * However, due to shared-IP zones (and restrict_interzone_loopback) 1258 * we can have an IRE_LOCAL as well as IRE_IF_CLONE for the same 1259 * address. For that reason we do tail insertion for IRE_IF_CLONE. 1260 * Due to the IRE_BROADCAST on cgtp0, which must be last in the bucket, 1261 * we do tail insertion of IRE_BROADCASTs that do not have RTF_MULTIRT 1262 * set. 1263 */ 1264 irep = (ire_t **)irb_ptr; 1265 if ((ire->ire_type & IRE_IF_CLONE) || 1266 ((ire->ire_type & IRE_BROADCAST) && 1267 !(ire->ire_flags & RTF_MULTIRT))) { 1268 while ((ire1 = *irep) != NULL) 1269 irep = &ire1->ire_next; 1270 } 1271 /* Insert at *irep */ 1272 ire1 = *irep; 1273 if (ire1 != NULL) 1274 ire1->ire_ptpn = &ire->ire_next; 1275 ire->ire_next = ire1; 1276 /* Link the new one in. */ 1277 ire->ire_ptpn = irep; 1278 1279 /* 1280 * ire_walk routines de-reference ire_next without holding 1281 * a lock. Before we point to the new ire, we want to make 1282 * sure the store that sets the ire_next of the new ire 1283 * reaches global visibility, so that ire_walk routines 1284 * don't see a truncated list of ires i.e if the ire_next 1285 * of the new ire gets set after we do "*irep = ire" due 1286 * to re-ordering, the ire_walk thread will see a NULL 1287 * once it accesses the ire_next of the new ire. 1288 * membar_producer() makes sure that the following store 1289 * happens *after* all of the above stores. 1290 */ 1291 membar_producer(); 1292 *irep = ire; 1293 ire->ire_bucket = irb_ptr; 1294 /* 1295 * We return a bumped up IRE above. Keep it symmetrical 1296 * so that the callers will always have to release. This 1297 * helps the callers of this function because they continue 1298 * to use the IRE after adding and hence they don't have to 1299 * lookup again after we return the IRE. 1300 * 1301 * NOTE : We don't have to use atomics as this is appearing 1302 * in the list for the first time and no one else can bump 1303 * up the reference count on this yet. 1304 */ 1305 ire_refhold_locked(ire); 1306 BUMP_IRE_STATS(ipst->ips_ire_stats_v4, ire_stats_inserted); 1307 1308 irb_ptr->irb_ire_cnt++; 1309 if (irb_ptr->irb_marks & IRB_MARK_DYNAMIC) 1310 irb_ptr->irb_nire++; 1311 1312 if (ire->ire_ill != NULL) { 1313 ire->ire_ill->ill_ire_cnt++; 1314 ASSERT(ire->ire_ill->ill_ire_cnt != 0); /* Wraparound */ 1315 } 1316 1317 ire_atomic_end(irb_ptr, ire); 1318 1319 /* Make any caching of the IREs be notified or updated */ 1320 ire_flush_cache_v4(ire, IRE_FLUSH_ADD); 1321 1322 if (ire->ire_ill != NULL) 1323 ASSERT(!MUTEX_HELD(&ire->ire_ill->ill_lock)); 1324 irb_refrele(irb_ptr); 1325 return (ire); 1326 } 1327 1328 /* 1329 * irb_refrele is the only caller of the function. ire_unlink calls to 1330 * do the final cleanup for this ire. 1331 */ 1332 void 1333 ire_cleanup(ire_t *ire) 1334 { 1335 ire_t *ire_next; 1336 ip_stack_t *ipst = ire->ire_ipst; 1337 1338 ASSERT(ire != NULL); 1339 1340 while (ire != NULL) { 1341 ire_next = ire->ire_next; 1342 if (ire->ire_ipversion == IPV4_VERSION) { 1343 ire_delete_v4(ire); 1344 BUMP_IRE_STATS(ipst->ips_ire_stats_v4, 1345 ire_stats_deleted); 1346 } else { 1347 ASSERT(ire->ire_ipversion == IPV6_VERSION); 1348 ire_delete_v6(ire); 1349 BUMP_IRE_STATS(ipst->ips_ire_stats_v6, 1350 ire_stats_deleted); 1351 } 1352 /* 1353 * Now it's really out of the list. Before doing the 1354 * REFRELE, set ire_next to NULL as ire_inactive asserts 1355 * so. 1356 */ 1357 ire->ire_next = NULL; 1358 ire_refrele_notr(ire); 1359 ire = ire_next; 1360 } 1361 } 1362 1363 /* 1364 * irb_refrele is the only caller of the function. It calls to unlink 1365 * all the CONDEMNED ires from this bucket. 1366 */ 1367 ire_t * 1368 ire_unlink(irb_t *irb) 1369 { 1370 ire_t *ire; 1371 ire_t *ire1; 1372 ire_t **ptpn; 1373 ire_t *ire_list = NULL; 1374 1375 ASSERT(RW_WRITE_HELD(&irb->irb_lock)); 1376 ASSERT(((irb->irb_marks & IRB_MARK_DYNAMIC) && irb->irb_refcnt == 1) || 1377 (irb->irb_refcnt == 0)); 1378 ASSERT(irb->irb_marks & IRB_MARK_CONDEMNED); 1379 ASSERT(irb->irb_ire != NULL); 1380 1381 for (ire = irb->irb_ire; ire != NULL; ire = ire1) { 1382 ire1 = ire->ire_next; 1383 if (IRE_IS_CONDEMNED(ire)) { 1384 ptpn = ire->ire_ptpn; 1385 ire1 = ire->ire_next; 1386 if (ire1) 1387 ire1->ire_ptpn = ptpn; 1388 *ptpn = ire1; 1389 ire->ire_ptpn = NULL; 1390 ire->ire_next = NULL; 1391 1392 /* 1393 * We need to call ire_delete_v4 or ire_delete_v6 to 1394 * clean up dependents and the redirects pointing at 1395 * the default gateway. We need to drop the lock 1396 * as ire_flush_cache/ire_delete_host_redircts require 1397 * so. But we can't drop the lock, as ire_unlink needs 1398 * to atomically remove the ires from the list. 1399 * So, create a temporary list of CONDEMNED ires 1400 * for doing ire_delete_v4/ire_delete_v6 operations 1401 * later on. 1402 */ 1403 ire->ire_next = ire_list; 1404 ire_list = ire; 1405 } 1406 } 1407 irb->irb_marks &= ~IRB_MARK_CONDEMNED; 1408 return (ire_list); 1409 } 1410 1411 /* 1412 * Clean up the radix node for this ire. Must be called by irb_refrele 1413 * when there are no ire's left in the bucket. Returns TRUE if the bucket 1414 * is deleted and freed. 1415 */ 1416 boolean_t 1417 irb_inactive(irb_t *irb) 1418 { 1419 struct rt_entry *rt; 1420 struct radix_node *rn; 1421 ip_stack_t *ipst = irb->irb_ipst; 1422 1423 ASSERT(irb->irb_ipst != NULL); 1424 1425 rt = IRB2RT(irb); 1426 rn = (struct radix_node *)rt; 1427 1428 /* first remove it from the radix tree. */ 1429 RADIX_NODE_HEAD_WLOCK(ipst->ips_ip_ftable); 1430 rw_enter(&irb->irb_lock, RW_WRITER); 1431 if (irb->irb_refcnt == 1 && irb->irb_nire == 0) { 1432 rn = ipst->ips_ip_ftable->rnh_deladdr(rn->rn_key, rn->rn_mask, 1433 ipst->ips_ip_ftable); 1434 DTRACE_PROBE1(irb__free, rt_t *, rt); 1435 ASSERT((void *)rn == (void *)rt); 1436 Free(rt, rt_entry_cache); 1437 /* irb_lock is freed */ 1438 RADIX_NODE_HEAD_UNLOCK(ipst->ips_ip_ftable); 1439 return (B_TRUE); 1440 } 1441 rw_exit(&irb->irb_lock); 1442 RADIX_NODE_HEAD_UNLOCK(ipst->ips_ip_ftable); 1443 return (B_FALSE); 1444 } 1445 1446 /* 1447 * Delete the specified IRE. 1448 * We assume that if ire_bucket is not set then ire_ill->ill_ire_cnt was 1449 * not incremented i.e., that the insertion in the bucket and the increment 1450 * of that counter is done atomically. 1451 */ 1452 void 1453 ire_delete(ire_t *ire) 1454 { 1455 ire_t *ire1; 1456 ire_t **ptpn; 1457 irb_t *irb; 1458 nce_t *nce; 1459 ip_stack_t *ipst = ire->ire_ipst; 1460 1461 /* We can clear ire_nce_cache under ire_lock even if the IRE is used */ 1462 mutex_enter(&ire->ire_lock); 1463 nce = ire->ire_nce_cache; 1464 ire->ire_nce_cache = NULL; 1465 mutex_exit(&ire->ire_lock); 1466 if (nce != NULL) 1467 nce_refrele(nce); 1468 1469 if ((irb = ire->ire_bucket) == NULL) { 1470 /* 1471 * It was never inserted in the list. Should call REFRELE 1472 * to free this IRE. 1473 */ 1474 ire_refrele_notr(ire); 1475 return; 1476 } 1477 1478 /* 1479 * Move the use counts from an IRE_IF_CLONE to its parent 1480 * IRE_INTERFACE. 1481 * We need to do this before acquiring irb_lock. 1482 */ 1483 if (ire->ire_type & IRE_IF_CLONE) { 1484 ire_t *parent; 1485 1486 rw_enter(&ipst->ips_ire_dep_lock, RW_READER); 1487 if ((parent = ire->ire_dep_parent) != NULL) { 1488 parent->ire_ob_pkt_count += ire->ire_ob_pkt_count; 1489 parent->ire_ib_pkt_count += ire->ire_ib_pkt_count; 1490 ire->ire_ob_pkt_count = 0; 1491 ire->ire_ib_pkt_count = 0; 1492 } 1493 rw_exit(&ipst->ips_ire_dep_lock); 1494 } 1495 1496 rw_enter(&irb->irb_lock, RW_WRITER); 1497 if (ire->ire_ptpn == NULL) { 1498 /* 1499 * Some other thread has removed us from the list. 1500 * It should have done the REFRELE for us. 1501 */ 1502 rw_exit(&irb->irb_lock); 1503 return; 1504 } 1505 1506 if (!IRE_IS_CONDEMNED(ire)) { 1507 /* Is this an IRE representing multiple duplicate entries? */ 1508 ASSERT(ire->ire_identical_ref >= 1); 1509 if (atomic_add_32_nv(&ire->ire_identical_ref, -1) != 0) { 1510 /* Removed one of the identical parties */ 1511 rw_exit(&irb->irb_lock); 1512 return; 1513 } 1514 1515 irb->irb_ire_cnt--; 1516 ire_make_condemned(ire); 1517 } 1518 1519 if (irb->irb_refcnt != 0) { 1520 /* 1521 * The last thread to leave this bucket will 1522 * delete this ire. 1523 */ 1524 irb->irb_marks |= IRB_MARK_CONDEMNED; 1525 rw_exit(&irb->irb_lock); 1526 return; 1527 } 1528 1529 /* 1530 * Normally to delete an ire, we walk the bucket. While we 1531 * walk the bucket, we normally bump up irb_refcnt and hence 1532 * we return from above where we mark CONDEMNED and the ire 1533 * gets deleted from ire_unlink. This case is where somebody 1534 * knows the ire e.g by doing a lookup, and wants to delete the 1535 * IRE. irb_refcnt would be 0 in this case if nobody is walking 1536 * the bucket. 1537 */ 1538 ptpn = ire->ire_ptpn; 1539 ire1 = ire->ire_next; 1540 if (ire1 != NULL) 1541 ire1->ire_ptpn = ptpn; 1542 ASSERT(ptpn != NULL); 1543 *ptpn = ire1; 1544 ire->ire_ptpn = NULL; 1545 ire->ire_next = NULL; 1546 if (ire->ire_ipversion == IPV6_VERSION) { 1547 BUMP_IRE_STATS(ipst->ips_ire_stats_v6, ire_stats_deleted); 1548 } else { 1549 BUMP_IRE_STATS(ipst->ips_ire_stats_v4, ire_stats_deleted); 1550 } 1551 rw_exit(&irb->irb_lock); 1552 1553 /* Cleanup dependents and related stuff */ 1554 if (ire->ire_ipversion == IPV6_VERSION) { 1555 ire_delete_v6(ire); 1556 } else { 1557 ire_delete_v4(ire); 1558 } 1559 /* 1560 * We removed it from the list. Decrement the 1561 * reference count. 1562 */ 1563 ire_refrele_notr(ire); 1564 } 1565 1566 /* 1567 * Delete the specified IRE. 1568 * All calls should use ire_delete(). 1569 * Sometimes called as writer though not required by this function. 1570 * 1571 * NOTE : This function is called only if the ire was added 1572 * in the list. 1573 */ 1574 static void 1575 ire_delete_v4(ire_t *ire) 1576 { 1577 ip_stack_t *ipst = ire->ire_ipst; 1578 1579 ASSERT(ire->ire_refcnt >= 1); 1580 ASSERT(ire->ire_ipversion == IPV4_VERSION); 1581 1582 ire_flush_cache_v4(ire, IRE_FLUSH_DELETE); 1583 if (ire->ire_type == IRE_DEFAULT) { 1584 /* 1585 * when a default gateway is going away 1586 * delete all the host redirects pointing at that 1587 * gateway. 1588 */ 1589 ire_delete_host_redirects(ire->ire_gateway_addr, ipst); 1590 } 1591 1592 /* 1593 * If we are deleting an IRE_INTERFACE then we make sure we also 1594 * delete any IRE_IF_CLONE that has been created from it. 1595 * Those are always in ire_dep_children. 1596 */ 1597 if ((ire->ire_type & IRE_INTERFACE) && ire->ire_dep_children != NULL) 1598 ire_dep_delete_if_clone(ire); 1599 1600 /* Remove from parent dependencies and child */ 1601 rw_enter(&ipst->ips_ire_dep_lock, RW_WRITER); 1602 if (ire->ire_dep_parent != NULL) 1603 ire_dep_remove(ire); 1604 1605 while (ire->ire_dep_children != NULL) 1606 ire_dep_remove(ire->ire_dep_children); 1607 rw_exit(&ipst->ips_ire_dep_lock); 1608 } 1609 1610 /* 1611 * ire_refrele is the only caller of the function. It calls 1612 * to free the ire when the reference count goes to zero. 1613 */ 1614 void 1615 ire_inactive(ire_t *ire) 1616 { 1617 ill_t *ill; 1618 irb_t *irb; 1619 ip_stack_t *ipst = ire->ire_ipst; 1620 1621 ASSERT(ire->ire_refcnt == 0); 1622 ASSERT(ire->ire_ptpn == NULL); 1623 ASSERT(ire->ire_next == NULL); 1624 1625 /* Count how many condemned ires for kmem_cache callback */ 1626 if (IRE_IS_CONDEMNED(ire)) 1627 atomic_add_32(&ipst->ips_num_ire_condemned, -1); 1628 1629 if (ire->ire_gw_secattr != NULL) { 1630 ire_gw_secattr_free(ire->ire_gw_secattr); 1631 ire->ire_gw_secattr = NULL; 1632 } 1633 1634 /* 1635 * ire_nce_cache is cleared in ire_delete, and we make sure we don't 1636 * set it once the ire is marked condemned. 1637 */ 1638 ASSERT(ire->ire_nce_cache == NULL); 1639 1640 /* 1641 * Since any parent would have a refhold on us they would already 1642 * have been removed. 1643 */ 1644 ASSERT(ire->ire_dep_parent == NULL); 1645 ASSERT(ire->ire_dep_sib_next == NULL); 1646 ASSERT(ire->ire_dep_sib_ptpn == NULL); 1647 1648 /* 1649 * Since any children would have a refhold on us they should have 1650 * already been removed. 1651 */ 1652 ASSERT(ire->ire_dep_children == NULL); 1653 1654 /* 1655 * ill_ire_ref is increased when the IRE is inserted in the 1656 * bucket - not when the IRE is created. 1657 */ 1658 irb = ire->ire_bucket; 1659 ill = ire->ire_ill; 1660 if (irb != NULL && ill != NULL) { 1661 mutex_enter(&ill->ill_lock); 1662 ASSERT(ill->ill_ire_cnt != 0); 1663 DTRACE_PROBE3(ill__decr__cnt, (ill_t *), ill, 1664 (char *), "ire", (void *), ire); 1665 ill->ill_ire_cnt--; 1666 if (ILL_DOWN_OK(ill)) { 1667 /* Drops the ill lock */ 1668 ipif_ill_refrele_tail(ill); 1669 } else { 1670 mutex_exit(&ill->ill_lock); 1671 } 1672 } 1673 ire->ire_ill = NULL; 1674 1675 /* This should be true for both V4 and V6 */ 1676 if (irb != NULL && (irb->irb_marks & IRB_MARK_DYNAMIC)) { 1677 rw_enter(&irb->irb_lock, RW_WRITER); 1678 irb->irb_nire--; 1679 /* 1680 * Instead of examining the conditions for freeing 1681 * the radix node here, we do it by calling 1682 * irb_refrele which is a single point in the code 1683 * that embeds that logic. Bump up the refcnt to 1684 * be able to call irb_refrele 1685 */ 1686 irb_refhold_locked(irb); 1687 rw_exit(&irb->irb_lock); 1688 irb_refrele(irb); 1689 } 1690 1691 #ifdef DEBUG 1692 ire_trace_cleanup(ire); 1693 #endif 1694 mutex_destroy(&ire->ire_lock); 1695 if (ire->ire_ipversion == IPV6_VERSION) { 1696 BUMP_IRE_STATS(ipst->ips_ire_stats_v6, ire_stats_freed); 1697 } else { 1698 BUMP_IRE_STATS(ipst->ips_ire_stats_v4, ire_stats_freed); 1699 } 1700 kmem_cache_free(ire_cache, ire); 1701 } 1702 1703 /* 1704 * ire_update_generation is the callback function provided by 1705 * ire_get_bucket() to update the generation number of any 1706 * matching shorter route when a new route is added. 1707 * 1708 * This fucntion always returns a failure return (B_FALSE) 1709 * to force the caller (rn_matchaddr_args) 1710 * to back-track up the tree looking for shorter matches. 1711 */ 1712 /* ARGSUSED */ 1713 static boolean_t 1714 ire_update_generation(struct radix_node *rn, void *arg) 1715 { 1716 struct rt_entry *rt = (struct rt_entry *)rn; 1717 1718 /* We need to handle all in the same bucket */ 1719 irb_increment_generation(&rt->rt_irb); 1720 return (B_FALSE); 1721 } 1722 1723 /* 1724 * Take care of all the generation numbers in the bucket. 1725 */ 1726 void 1727 irb_increment_generation(irb_t *irb) 1728 { 1729 ire_t *ire; 1730 1731 if (irb == NULL || irb->irb_ire_cnt == 0) 1732 return; 1733 1734 irb_refhold(irb); 1735 for (ire = irb->irb_ire; ire != NULL; ire = ire->ire_next) { 1736 if (!IRE_IS_CONDEMNED(ire)) 1737 ire_increment_generation(ire); /* Ourselves */ 1738 ire_dep_incr_generation(ire); /* Dependants */ 1739 } 1740 irb_refrele(irb); 1741 } 1742 1743 /* 1744 * When an IRE is added or deleted this routine is called to make sure 1745 * any caching of IRE information is notified or updated. 1746 * 1747 * The flag argument indicates if the flush request is due to addition 1748 * of new route (IRE_FLUSH_ADD), deletion of old route (IRE_FLUSH_DELETE), 1749 * or a change to ire_gateway_addr (IRE_FLUSH_GWCHANGE). 1750 */ 1751 void 1752 ire_flush_cache_v4(ire_t *ire, int flag) 1753 { 1754 irb_t *irb = ire->ire_bucket; 1755 struct rt_entry *rt = IRB2RT(irb); 1756 ip_stack_t *ipst = ire->ire_ipst; 1757 1758 /* 1759 * IRE_IF_CLONE ire's don't provide any new information 1760 * than the parent from which they are cloned, so don't 1761 * perturb the generation numbers. 1762 */ 1763 if (ire->ire_type & IRE_IF_CLONE) 1764 return; 1765 1766 /* 1767 * Ensure that an ire_add during a lookup serializes the updates of the 1768 * generation numbers under the radix head lock so that the lookup gets 1769 * either the old ire and old generation number, or a new ire and new 1770 * generation number. 1771 */ 1772 RADIX_NODE_HEAD_WLOCK(ipst->ips_ip_ftable); 1773 1774 /* 1775 * If a route was just added, we need to notify everybody that 1776 * has cached an IRE_NOROUTE since there might now be a better 1777 * route for them. 1778 */ 1779 if (flag == IRE_FLUSH_ADD) { 1780 ire_increment_generation(ipst->ips_ire_reject_v4); 1781 ire_increment_generation(ipst->ips_ire_blackhole_v4); 1782 } 1783 1784 /* Adding a default can't otherwise provide a better route */ 1785 if (ire->ire_type == IRE_DEFAULT && flag == IRE_FLUSH_ADD) { 1786 RADIX_NODE_HEAD_UNLOCK(ipst->ips_ip_ftable); 1787 return; 1788 } 1789 1790 switch (flag) { 1791 case IRE_FLUSH_DELETE: 1792 case IRE_FLUSH_GWCHANGE: 1793 /* 1794 * Update ire_generation for all ire_dep_children chains 1795 * starting with this IRE 1796 */ 1797 ire_dep_incr_generation(ire); 1798 break; 1799 case IRE_FLUSH_ADD: 1800 /* 1801 * Update the generation numbers of all shorter matching routes. 1802 * ire_update_generation takes care of the dependants by 1803 * using ire_dep_incr_generation. 1804 */ 1805 (void) ipst->ips_ip_ftable->rnh_matchaddr_args(&rt->rt_dst, 1806 ipst->ips_ip_ftable, ire_update_generation, NULL); 1807 break; 1808 } 1809 RADIX_NODE_HEAD_UNLOCK(ipst->ips_ip_ftable); 1810 } 1811 1812 /* 1813 * Matches the arguments passed with the values in the ire. 1814 * 1815 * Note: for match types that match using "ill" passed in, ill 1816 * must be checked for non-NULL before calling this routine. 1817 */ 1818 boolean_t 1819 ire_match_args(ire_t *ire, ipaddr_t addr, ipaddr_t mask, ipaddr_t gateway, 1820 int type, const ill_t *ill, zoneid_t zoneid, 1821 const ts_label_t *tsl, int match_flags) 1822 { 1823 ill_t *ire_ill = NULL, *dst_ill; 1824 ip_stack_t *ipst = ire->ire_ipst; 1825 1826 ASSERT(ire->ire_ipversion == IPV4_VERSION); 1827 ASSERT((ire->ire_addr & ~ire->ire_mask) == 0); 1828 ASSERT((!(match_flags & MATCH_IRE_ILL)) || 1829 (ill != NULL && !ill->ill_isv6)); 1830 1831 /* 1832 * If MATCH_IRE_TESTHIDDEN is set, then only return the IRE if it is 1833 * in fact hidden, to ensure the caller gets the right one. 1834 */ 1835 if (ire->ire_testhidden) { 1836 if (!(match_flags & MATCH_IRE_TESTHIDDEN)) 1837 return (B_FALSE); 1838 } 1839 1840 if (zoneid != ALL_ZONES && zoneid != ire->ire_zoneid && 1841 ire->ire_zoneid != ALL_ZONES) { 1842 /* 1843 * If MATCH_IRE_ZONEONLY has been set and the supplied zoneid 1844 * does not match that of ire_zoneid, a failure to 1845 * match is reported at this point. Otherwise, since some IREs 1846 * that are available in the global zone can be used in local 1847 * zones, additional checks need to be performed: 1848 * 1849 * IRE_LOOPBACK 1850 * entries should never be matched in this situation. 1851 * Each zone has its own IRE_LOOPBACK. 1852 * 1853 * IRE_LOCAL 1854 * We allow them for any zoneid. ire_route_recursive 1855 * does additional checks when 1856 * ip_restrict_interzone_loopback is set. 1857 * 1858 * If ill_usesrc_ifindex is set 1859 * Then we check if the zone has a valid source address 1860 * on the usesrc ill. 1861 * 1862 * If ire_ill is set, then check that the zone has an ipif 1863 * on that ill. 1864 * 1865 * Outside of this function (in ire_round_robin) we check 1866 * that any IRE_OFFLINK has a gateway that reachable from the 1867 * zone when we have multiple choices (ECMP). 1868 */ 1869 if (match_flags & MATCH_IRE_ZONEONLY) 1870 return (B_FALSE); 1871 if (ire->ire_type & IRE_LOOPBACK) 1872 return (B_FALSE); 1873 1874 if (ire->ire_type & IRE_LOCAL) 1875 goto matchit; 1876 1877 /* 1878 * The normal case of IRE_ONLINK has a matching zoneid. 1879 * Here we handle the case when shared-IP zones have been 1880 * configured with IP addresses on vniN. In that case it 1881 * is ok for traffic from a zone to use IRE_ONLINK routes 1882 * if the ill has a usesrc pointing at vniN 1883 */ 1884 dst_ill = ire->ire_ill; 1885 if (ire->ire_type & IRE_ONLINK) { 1886 uint_t ifindex; 1887 1888 /* 1889 * Note there is no IRE_INTERFACE on vniN thus 1890 * can't do an IRE lookup for a matching route. 1891 */ 1892 ifindex = dst_ill->ill_usesrc_ifindex; 1893 if (ifindex == 0) 1894 return (B_FALSE); 1895 1896 /* 1897 * If there is a usable source address in the 1898 * zone, then it's ok to return this IRE_INTERFACE 1899 */ 1900 if (!ipif_zone_avail(ifindex, dst_ill->ill_isv6, 1901 zoneid, ipst)) { 1902 ip3dbg(("ire_match_args: no usrsrc for zone" 1903 " dst_ill %p\n", (void *)dst_ill)); 1904 return (B_FALSE); 1905 } 1906 } 1907 /* 1908 * For exampe, with 1909 * route add 11.0.0.0 gw1 -ifp bge0 1910 * route add 11.0.0.0 gw2 -ifp bge1 1911 * this code would differentiate based on 1912 * where the sending zone has addresses. 1913 * Only if the zone has an address on bge0 can it use the first 1914 * route. It isn't clear if this behavior is documented 1915 * anywhere. 1916 */ 1917 if (dst_ill != NULL && (ire->ire_type & IRE_OFFLINK)) { 1918 ipif_t *tipif; 1919 1920 mutex_enter(&dst_ill->ill_lock); 1921 for (tipif = dst_ill->ill_ipif; 1922 tipif != NULL; tipif = tipif->ipif_next) { 1923 if (!IPIF_IS_CONDEMNED(tipif) && 1924 (tipif->ipif_flags & IPIF_UP) && 1925 (tipif->ipif_zoneid == zoneid || 1926 tipif->ipif_zoneid == ALL_ZONES)) 1927 break; 1928 } 1929 mutex_exit(&dst_ill->ill_lock); 1930 if (tipif == NULL) { 1931 return (B_FALSE); 1932 } 1933 } 1934 } 1935 1936 matchit: 1937 if (match_flags & MATCH_IRE_ILL) { 1938 ire_ill = ire->ire_ill; 1939 1940 /* 1941 * If asked to match an ill, we *must* match 1942 * on the ire_ill for ipmp test addresses, or 1943 * any of the ill in the group for data addresses. 1944 * If we don't, we may as well fail. 1945 * However, we need an exception for IRE_LOCALs to ensure 1946 * we loopback packets even sent to test addresses on different 1947 * interfaces in the group. 1948 */ 1949 if ((match_flags & MATCH_IRE_TESTHIDDEN) && 1950 !(ire->ire_type & IRE_LOCAL)) { 1951 if (ire->ire_ill != ill) 1952 return (B_FALSE); 1953 } else { 1954 match_flags &= ~MATCH_IRE_TESTHIDDEN; 1955 /* 1956 * We know that ill is not NULL, but ire_ill could be 1957 * NULL 1958 */ 1959 if (ire_ill == NULL || !IS_ON_SAME_LAN(ill, ire_ill)) 1960 return (B_FALSE); 1961 } 1962 } 1963 1964 if ((ire->ire_addr == (addr & mask)) && 1965 ((!(match_flags & MATCH_IRE_GW)) || 1966 (ire->ire_gateway_addr == gateway)) && 1967 ((!(match_flags & MATCH_IRE_TYPE)) || (ire->ire_type & type)) && 1968 ((!(match_flags & MATCH_IRE_TESTHIDDEN)) || ire->ire_testhidden) && 1969 ((!(match_flags & MATCH_IRE_MASK)) || (ire->ire_mask == mask)) && 1970 ((!(match_flags & MATCH_IRE_SECATTR)) || 1971 (!is_system_labeled()) || 1972 (tsol_ire_match_gwattr(ire, tsl) == 0))) { 1973 /* We found the matched IRE */ 1974 return (B_TRUE); 1975 } 1976 return (B_FALSE); 1977 } 1978 1979 /* 1980 * Check if the IRE_LOCAL uses the same ill as another route would use. 1981 * If there is no alternate route, or the alternate is a REJECT or BLACKHOLE, 1982 * then we don't allow this IRE_LOCAL to be used. 1983 * We always return an IRE; will be RTF_REJECT if no route available. 1984 */ 1985 ire_t * 1986 ire_alt_local(ire_t *ire, zoneid_t zoneid, const ts_label_t *tsl, 1987 const ill_t *ill, uint_t *generationp) 1988 { 1989 ip_stack_t *ipst = ire->ire_ipst; 1990 ire_t *alt_ire; 1991 uint_t ire_type; 1992 uint_t generation; 1993 uint_t match_flags; 1994 1995 ASSERT(ire->ire_type & IRE_LOCAL); 1996 ASSERT(ire->ire_ill != NULL); 1997 1998 /* 1999 * Need to match on everything but local. 2000 * This might result in the creation of a IRE_IF_CLONE for the 2001 * same address as the IRE_LOCAL when restrict_interzone_loopback is 2002 * set. ire_add_*() ensures that the IRE_IF_CLONE are tail inserted 2003 * to make sure the IRE_LOCAL is always found first. 2004 */ 2005 ire_type = (IRE_ONLINK | IRE_OFFLINK) & ~(IRE_LOCAL|IRE_LOOPBACK); 2006 match_flags = MATCH_IRE_TYPE | MATCH_IRE_SECATTR; 2007 if (ill != NULL) 2008 match_flags |= MATCH_IRE_ILL; 2009 2010 if (ire->ire_ipversion == IPV4_VERSION) { 2011 alt_ire = ire_route_recursive_v4(ire->ire_addr, ire_type, 2012 ill, zoneid, tsl, match_flags, B_TRUE, 0, ipst, NULL, NULL, 2013 &generation); 2014 } else { 2015 alt_ire = ire_route_recursive_v6(&ire->ire_addr_v6, ire_type, 2016 ill, zoneid, tsl, match_flags, B_TRUE, 0, ipst, NULL, NULL, 2017 &generation); 2018 } 2019 ASSERT(alt_ire != NULL); 2020 2021 if (alt_ire->ire_ill == ire->ire_ill) { 2022 /* Going out the same ILL - ok to send to IRE_LOCAL */ 2023 ire_refrele(alt_ire); 2024 } else { 2025 /* Different ill - ignore IRE_LOCAL */ 2026 ire_refrele(ire); 2027 ire = alt_ire; 2028 if (generationp != NULL) 2029 *generationp = generation; 2030 } 2031 return (ire); 2032 } 2033 2034 boolean_t 2035 ire_find_zoneid(struct radix_node *rn, void *arg) 2036 { 2037 struct rt_entry *rt = (struct rt_entry *)rn; 2038 irb_t *irb; 2039 ire_t *ire; 2040 ire_ftable_args_t *margs = arg; 2041 2042 ASSERT(rt != NULL); 2043 2044 irb = &rt->rt_irb; 2045 2046 if (irb->irb_ire_cnt == 0) 2047 return (B_FALSE); 2048 2049 rw_enter(&irb->irb_lock, RW_READER); 2050 for (ire = irb->irb_ire; ire != NULL; ire = ire->ire_next) { 2051 if (IRE_IS_CONDEMNED(ire)) 2052 continue; 2053 2054 if (!(ire->ire_type & IRE_INTERFACE)) 2055 continue; 2056 2057 if (ire->ire_zoneid != ALL_ZONES && 2058 ire->ire_zoneid != margs->ift_zoneid) 2059 continue; 2060 2061 if (margs->ift_ill != NULL && margs->ift_ill != ire->ire_ill) 2062 continue; 2063 2064 if (is_system_labeled() && 2065 tsol_ire_match_gwattr(ire, margs->ift_tsl) != 0) 2066 continue; 2067 2068 rw_exit(&irb->irb_lock); 2069 return (B_TRUE); 2070 } 2071 rw_exit(&irb->irb_lock); 2072 return (B_FALSE); 2073 } 2074 2075 /* 2076 * Check if the zoneid (not ALL_ZONES) has an IRE_INTERFACE for the specified 2077 * gateway address. If ill is non-NULL we also match on it. 2078 * The caller must hold a read lock on RADIX_NODE_HEAD if lock_held is set. 2079 */ 2080 boolean_t 2081 ire_gateway_ok_zone_v4(ipaddr_t gateway, zoneid_t zoneid, ill_t *ill, 2082 const ts_label_t *tsl, ip_stack_t *ipst, boolean_t lock_held) 2083 { 2084 struct rt_sockaddr rdst; 2085 struct rt_entry *rt; 2086 ire_ftable_args_t margs; 2087 2088 ASSERT(ill == NULL || !ill->ill_isv6); 2089 if (lock_held) 2090 ASSERT(RW_READ_HELD(&ipst->ips_ip_ftable->rnh_lock)); 2091 else 2092 RADIX_NODE_HEAD_RLOCK(ipst->ips_ip_ftable); 2093 2094 bzero(&rdst, sizeof (rdst)); 2095 rdst.rt_sin_len = sizeof (rdst); 2096 rdst.rt_sin_family = AF_INET; 2097 rdst.rt_sin_addr.s_addr = gateway; 2098 2099 /* 2100 * We only use margs for ill, zoneid, and tsl matching in 2101 * ire_find_zoneid 2102 */ 2103 bzero(&margs, sizeof (margs)); 2104 margs.ift_ill = ill; 2105 margs.ift_zoneid = zoneid; 2106 margs.ift_tsl = tsl; 2107 rt = (struct rt_entry *)ipst->ips_ip_ftable->rnh_matchaddr_args(&rdst, 2108 ipst->ips_ip_ftable, ire_find_zoneid, (void *)&margs); 2109 2110 if (!lock_held) 2111 RADIX_NODE_HEAD_UNLOCK(ipst->ips_ip_ftable); 2112 2113 return (rt != NULL); 2114 } 2115 2116 /* 2117 * ire_walk routine to delete a fraction of redirect IREs and IRE_CLONE_IF IREs. 2118 * The fraction argument tells us what fraction of the IREs to delete. 2119 * Common for IPv4 and IPv6. 2120 * Used when memory backpressure. 2121 */ 2122 static void 2123 ire_delete_reclaim(ire_t *ire, char *arg) 2124 { 2125 ip_stack_t *ipst = ire->ire_ipst; 2126 uint_t fraction = *(uint_t *)arg; 2127 uint_t rand; 2128 2129 if ((ire->ire_flags & RTF_DYNAMIC) || 2130 (ire->ire_type & IRE_IF_CLONE)) { 2131 2132 /* Pick a random number */ 2133 rand = (uint_t)ddi_get_lbolt() + 2134 IRE_ADDR_HASH_V6(ire->ire_addr_v6, 256); 2135 2136 /* Use truncation */ 2137 if ((rand/fraction)*fraction == rand) { 2138 IP_STAT(ipst, ip_ire_reclaim_deleted); 2139 ire_delete(ire); 2140 } 2141 } 2142 2143 } 2144 2145 /* 2146 * kmem_cache callback to free up memory. 2147 * 2148 * Free a fraction (ips_ip_ire_reclaim_fraction) of things IP added dynamically 2149 * (RTF_DYNAMIC and IRE_IF_CLONE). 2150 */ 2151 static void 2152 ip_ire_reclaim_stack(ip_stack_t *ipst) 2153 { 2154 uint_t fraction = ipst->ips_ip_ire_reclaim_fraction; 2155 2156 IP_STAT(ipst, ip_ire_reclaim_calls); 2157 2158 ire_walk(ire_delete_reclaim, &fraction, ipst); 2159 2160 /* 2161 * Walk all CONNs that can have a reference on an ire, nce or dce. 2162 * Get them to update any stale references to drop any refholds they 2163 * have. 2164 */ 2165 ipcl_walk(conn_ixa_cleanup, (void *)B_FALSE, ipst); 2166 } 2167 2168 /* 2169 * Called by the memory allocator subsystem directly, when the system 2170 * is running low on memory. 2171 */ 2172 /* ARGSUSED */ 2173 void 2174 ip_ire_reclaim(void *args) 2175 { 2176 netstack_handle_t nh; 2177 netstack_t *ns; 2178 2179 netstack_next_init(&nh); 2180 while ((ns = netstack_next(&nh)) != NULL) { 2181 ip_ire_reclaim_stack(ns->netstack_ip); 2182 netstack_rele(ns); 2183 } 2184 netstack_next_fini(&nh); 2185 } 2186 2187 static void 2188 power2_roundup(uint32_t *value) 2189 { 2190 int i; 2191 2192 for (i = 1; i < 31; i++) { 2193 if (*value <= (1 << i)) 2194 break; 2195 } 2196 *value = (1 << i); 2197 } 2198 2199 /* Global init for all zones */ 2200 void 2201 ip_ire_g_init() 2202 { 2203 /* 2204 * Create kmem_caches. ip_ire_reclaim() and ip_nce_reclaim() 2205 * will give disposable IREs back to system when needed. 2206 * This needs to be done here before anything else, since 2207 * ire_add() expects the cache to be created. 2208 */ 2209 ire_cache = kmem_cache_create("ire_cache", 2210 sizeof (ire_t), 0, NULL, NULL, 2211 ip_ire_reclaim, NULL, NULL, 0); 2212 2213 ncec_cache = kmem_cache_create("ncec_cache", 2214 sizeof (ncec_t), 0, NULL, NULL, 2215 ip_nce_reclaim, NULL, NULL, 0); 2216 nce_cache = kmem_cache_create("nce_cache", 2217 sizeof (nce_t), 0, NULL, NULL, 2218 NULL, NULL, NULL, 0); 2219 2220 rt_entry_cache = kmem_cache_create("rt_entry", 2221 sizeof (struct rt_entry), 0, NULL, NULL, NULL, NULL, NULL, 0); 2222 2223 /* 2224 * Have radix code setup kmem caches etc. 2225 */ 2226 rn_init(); 2227 } 2228 2229 void 2230 ip_ire_init(ip_stack_t *ipst) 2231 { 2232 ire_t *ire; 2233 int error; 2234 2235 mutex_init(&ipst->ips_ire_ft_init_lock, NULL, MUTEX_DEFAULT, 0); 2236 2237 (void) rn_inithead((void **)&ipst->ips_ip_ftable, 32); 2238 2239 /* 2240 * Make sure that the forwarding table size is a power of 2. 2241 * The IRE*_ADDR_HASH() macroes depend on that. 2242 */ 2243 ipst->ips_ip6_ftable_hash_size = ip6_ftable_hash_size; 2244 power2_roundup(&ipst->ips_ip6_ftable_hash_size); 2245 2246 /* 2247 * Allocate/initialize a pair of IRE_NOROUTEs for each of IPv4 and IPv6. 2248 * The ire_reject_v* has RTF_REJECT set, and the ire_blackhole_v* has 2249 * RTF_BLACKHOLE set. We use the latter for transient errors such 2250 * as memory allocation failures and tripping on IRE_IS_CONDEMNED 2251 * entries. 2252 */ 2253 ire = kmem_cache_alloc(ire_cache, KM_SLEEP); 2254 *ire = ire_null; 2255 error = ire_init_v4(ire, 0, 0, 0, IRE_NOROUTE, NULL, ALL_ZONES, 2256 RTF_REJECT|RTF_UP, NULL, ipst); 2257 ASSERT(error == 0); 2258 ipst->ips_ire_reject_v4 = ire; 2259 2260 ire = kmem_cache_alloc(ire_cache, KM_SLEEP); 2261 *ire = ire_null; 2262 error = ire_init_v6(ire, 0, 0, 0, IRE_NOROUTE, NULL, ALL_ZONES, 2263 RTF_REJECT|RTF_UP, NULL, ipst); 2264 ASSERT(error == 0); 2265 ipst->ips_ire_reject_v6 = ire; 2266 2267 ire = kmem_cache_alloc(ire_cache, KM_SLEEP); 2268 *ire = ire_null; 2269 error = ire_init_v4(ire, 0, 0, 0, IRE_NOROUTE, NULL, ALL_ZONES, 2270 RTF_BLACKHOLE|RTF_UP, NULL, ipst); 2271 ASSERT(error == 0); 2272 ipst->ips_ire_blackhole_v4 = ire; 2273 2274 ire = kmem_cache_alloc(ire_cache, KM_SLEEP); 2275 *ire = ire_null; 2276 error = ire_init_v6(ire, 0, 0, 0, IRE_NOROUTE, NULL, ALL_ZONES, 2277 RTF_BLACKHOLE|RTF_UP, NULL, ipst); 2278 ASSERT(error == 0); 2279 ipst->ips_ire_blackhole_v6 = ire; 2280 2281 rw_init(&ipst->ips_ip6_ire_head_lock, NULL, RW_DEFAULT, NULL); 2282 rw_init(&ipst->ips_ire_dep_lock, NULL, RW_DEFAULT, NULL); 2283 } 2284 2285 void 2286 ip_ire_g_fini(void) 2287 { 2288 kmem_cache_destroy(ire_cache); 2289 kmem_cache_destroy(ncec_cache); 2290 kmem_cache_destroy(nce_cache); 2291 kmem_cache_destroy(rt_entry_cache); 2292 2293 rn_fini(); 2294 } 2295 2296 void 2297 ip_ire_fini(ip_stack_t *ipst) 2298 { 2299 int i; 2300 2301 rw_destroy(&ipst->ips_ire_dep_lock); 2302 rw_destroy(&ipst->ips_ip6_ire_head_lock); 2303 2304 ire_refrele_notr(ipst->ips_ire_reject_v6); 2305 ipst->ips_ire_reject_v6 = NULL; 2306 ire_refrele_notr(ipst->ips_ire_reject_v4); 2307 ipst->ips_ire_reject_v4 = NULL; 2308 ire_refrele_notr(ipst->ips_ire_blackhole_v6); 2309 ipst->ips_ire_blackhole_v6 = NULL; 2310 ire_refrele_notr(ipst->ips_ire_blackhole_v4); 2311 ipst->ips_ire_blackhole_v4 = NULL; 2312 2313 /* 2314 * Delete all IREs - assumes that the ill/ipifs have 2315 * been removed so what remains are just the ftable to handle. 2316 */ 2317 ire_walk(ire_delete, NULL, ipst); 2318 2319 rn_freehead(ipst->ips_ip_ftable); 2320 ipst->ips_ip_ftable = NULL; 2321 2322 mutex_destroy(&ipst->ips_ire_ft_init_lock); 2323 2324 for (i = 0; i < IP6_MASK_TABLE_SIZE; i++) { 2325 irb_t *ptr; 2326 int j; 2327 2328 if ((ptr = ipst->ips_ip_forwarding_table_v6[i]) == NULL) 2329 continue; 2330 2331 for (j = 0; j < ipst->ips_ip6_ftable_hash_size; j++) { 2332 ASSERT(ptr[j].irb_ire == NULL); 2333 rw_destroy(&ptr[j].irb_lock); 2334 } 2335 mi_free(ptr); 2336 ipst->ips_ip_forwarding_table_v6[i] = NULL; 2337 } 2338 } 2339 2340 #ifdef DEBUG 2341 void 2342 ire_trace_ref(ire_t *ire) 2343 { 2344 mutex_enter(&ire->ire_lock); 2345 if (ire->ire_trace_disable) { 2346 mutex_exit(&ire->ire_lock); 2347 return; 2348 } 2349 2350 if (th_trace_ref(ire, ire->ire_ipst)) { 2351 mutex_exit(&ire->ire_lock); 2352 } else { 2353 ire->ire_trace_disable = B_TRUE; 2354 mutex_exit(&ire->ire_lock); 2355 ire_trace_cleanup(ire); 2356 } 2357 } 2358 2359 void 2360 ire_untrace_ref(ire_t *ire) 2361 { 2362 mutex_enter(&ire->ire_lock); 2363 if (!ire->ire_trace_disable) 2364 th_trace_unref(ire); 2365 mutex_exit(&ire->ire_lock); 2366 } 2367 2368 static void 2369 ire_trace_cleanup(const ire_t *ire) 2370 { 2371 th_trace_cleanup(ire, ire->ire_trace_disable); 2372 } 2373 #endif /* DEBUG */ 2374 2375 /* 2376 * Find, or create if needed, the nce_t pointer to the neighbor cache 2377 * entry ncec_t for an IPv4 address. The nce_t will be created on the ill_t 2378 * in the non-IPMP case, or on the cast-ill in the IPMP bcast/mcast case, or 2379 * on the next available under-ill (selected by the IPMP rotor) in the 2380 * unicast IPMP case. 2381 * 2382 * If a neighbor-cache entry has to be created (i.e., one does not already 2383 * exist in the nce list) the ncec_lladdr and ncec_state of the neighbor cache 2384 * entry are initialized in nce_add_v4(). The broadcast, multicast, and 2385 * link-layer type determine the contents of {ncec_state, ncec_lladdr} of 2386 * the ncec_t created. The ncec_lladdr is non-null for all link types with 2387 * non-zero ill_phys_addr_length, though the contents may be zero in cases 2388 * where the link-layer type is not known at the time of creation 2389 * (e.g., IRE_IFRESOLVER links) 2390 * 2391 * All IRE_BROADCAST entries have ncec_state = ND_REACHABLE, and the nce_lladr 2392 * has the physical broadcast address of the outgoing interface. 2393 * For unicast ire entries, 2394 * - if the outgoing interface is of type IRE_IF_RESOLVER, a newly created 2395 * ncec_t with 0 nce_lladr contents, and will be in the ND_INITIAL state. 2396 * - if the outgoing interface is a IRE_IF_NORESOLVER interface, no link 2397 * layer resolution is necessary, so that the ncec_t will be in the 2398 * ND_REACHABLE state 2399 * 2400 * The link layer information needed for broadcast addresses, and for 2401 * packets sent on IRE_IF_NORESOLVER interfaces is a constant mapping that 2402 * never needs re-verification for the lifetime of the ncec_t. These are 2403 * therefore marked NCE_F_NONUD. 2404 * 2405 * The nce returned will be created such that the nce_ill == ill that 2406 * is passed in. Note that the nce itself may not have ncec_ill == ill 2407 * where IPMP links are involved. 2408 */ 2409 static nce_t * 2410 ire_nce_init(ill_t *ill, const void *addr, int ire_type) 2411 { 2412 int err; 2413 nce_t *nce = NULL; 2414 uint16_t ncec_flags; 2415 uchar_t *hwaddr; 2416 boolean_t need_refrele = B_FALSE; 2417 ill_t *in_ill = ill; 2418 boolean_t is_unicast; 2419 uint_t hwaddr_len; 2420 2421 is_unicast = ((ire_type & (IRE_MULTICAST|IRE_BROADCAST)) == 0); 2422 if (IS_IPMP(ill) || 2423 ((ire_type & IRE_BROADCAST) && IS_UNDER_IPMP(ill))) { 2424 if ((ill = ipmp_ill_get_xmit_ill(ill, is_unicast)) == NULL) 2425 return (NULL); 2426 need_refrele = B_TRUE; 2427 } 2428 ncec_flags = (ill->ill_flags & ILLF_NONUD) ? NCE_F_NONUD : 0; 2429 2430 switch (ire_type) { 2431 case IRE_BROADCAST: 2432 ASSERT(!ill->ill_isv6); 2433 ncec_flags |= (NCE_F_BCAST|NCE_F_NONUD); 2434 break; 2435 case IRE_MULTICAST: 2436 ncec_flags |= (NCE_F_MCAST|NCE_F_NONUD); 2437 break; 2438 } 2439 2440 if (ill->ill_net_type == IRE_IF_NORESOLVER && is_unicast) { 2441 hwaddr = ill->ill_dest_addr; 2442 } else { 2443 hwaddr = NULL; 2444 } 2445 hwaddr_len = ill->ill_phys_addr_length; 2446 2447 retry: 2448 /* nce_state will be computed by nce_add_common() */ 2449 if (!ill->ill_isv6) { 2450 err = nce_lookup_then_add_v4(ill, hwaddr, hwaddr_len, addr, 2451 ncec_flags, ND_UNCHANGED, &nce); 2452 } else { 2453 err = nce_lookup_then_add_v6(ill, hwaddr, hwaddr_len, addr, 2454 ncec_flags, ND_UNCHANGED, &nce); 2455 } 2456 2457 switch (err) { 2458 case 0: 2459 break; 2460 case EEXIST: 2461 /* 2462 * When subnets change or partially overlap what was once 2463 * a broadcast address could now be a unicast, or vice versa. 2464 */ 2465 if (((ncec_flags ^ nce->nce_common->ncec_flags) & 2466 NCE_F_BCAST) != 0) { 2467 ASSERT(!ill->ill_isv6); 2468 ncec_delete(nce->nce_common); 2469 nce_refrele(nce); 2470 goto retry; 2471 } 2472 break; 2473 default: 2474 DTRACE_PROBE2(nce__init__fail, ill_t *, ill, int, err); 2475 if (need_refrele) 2476 ill_refrele(ill); 2477 return (NULL); 2478 } 2479 /* 2480 * If the ill was an under-ill of an IPMP group, we need to verify 2481 * that it is still active so that we select an active interface in 2482 * the group. However, since ipmp_ill_is_active ASSERTs for 2483 * IS_UNDER_IPMP(), we first need to verify that the ill is an 2484 * under-ill, and since this is being done in the data path, the 2485 * only way to ascertain this is by holding the ill_g_lock. 2486 */ 2487 rw_enter(&ill->ill_ipst->ips_ill_g_lock, RW_READER); 2488 mutex_enter(&ill->ill_lock); 2489 mutex_enter(&ill->ill_phyint->phyint_lock); 2490 if (need_refrele && IS_UNDER_IPMP(ill) && !ipmp_ill_is_active(ill)) { 2491 /* 2492 * need_refrele implies that the under ill was selected by 2493 * ipmp_ill_get_xmit_ill() because either the in_ill was an 2494 * ipmp_ill, or we are sending a non-unicast packet on 2495 * an under_ill. However, when we get here, the ill selected by 2496 * ipmp_ill_get_xmit_ill was pulled out of the active set 2497 * (for unicast) or cast_ill nomination (for 2498 * !unicast) after it was picked as the outgoing ill. 2499 * We have to pick an active interface and/or cast_ill in the 2500 * group. 2501 */ 2502 mutex_exit(&ill->ill_phyint->phyint_lock); 2503 nce_delete(nce); 2504 mutex_exit(&ill->ill_lock); 2505 rw_exit(&ill->ill_ipst->ips_ill_g_lock); 2506 nce_refrele(nce); 2507 ill_refrele(ill); 2508 if ((ill = ipmp_ill_get_xmit_ill(in_ill, is_unicast)) == NULL) 2509 return (NULL); 2510 goto retry; 2511 } else { 2512 mutex_exit(&ill->ill_phyint->phyint_lock); 2513 mutex_exit(&ill->ill_lock); 2514 rw_exit(&ill->ill_ipst->ips_ill_g_lock); 2515 } 2516 done: 2517 ASSERT(nce->nce_ill == ill); 2518 if (need_refrele) 2519 ill_refrele(ill); 2520 return (nce); 2521 } 2522 2523 nce_t * 2524 arp_nce_init(ill_t *ill, in_addr_t addr4, int ire_type) 2525 { 2526 return (ire_nce_init(ill, &addr4, ire_type)); 2527 } 2528 2529 nce_t * 2530 ndp_nce_init(ill_t *ill, const in6_addr_t *addr6, int ire_type) 2531 { 2532 ASSERT((ire_type & IRE_BROADCAST) == 0); 2533 return (ire_nce_init(ill, addr6, ire_type)); 2534 } 2535 2536 /* 2537 * The caller should hold irb_lock as a writer if the ire is in a bucket. 2538 */ 2539 void 2540 ire_make_condemned(ire_t *ire) 2541 { 2542 ip_stack_t *ipst = ire->ire_ipst; 2543 2544 mutex_enter(&ire->ire_lock); 2545 ASSERT(ire->ire_bucket == NULL || 2546 RW_WRITE_HELD(&ire->ire_bucket->irb_lock)); 2547 ASSERT(!IRE_IS_CONDEMNED(ire)); 2548 ire->ire_generation = IRE_GENERATION_CONDEMNED; 2549 /* Count how many condemned ires for kmem_cache callback */ 2550 atomic_add_32(&ipst->ips_num_ire_condemned, 1); 2551 mutex_exit(&ire->ire_lock); 2552 } 2553 2554 /* 2555 * Increment the generation avoiding the special condemned value 2556 */ 2557 void 2558 ire_increment_generation(ire_t *ire) 2559 { 2560 uint_t generation; 2561 2562 mutex_enter(&ire->ire_lock); 2563 /* 2564 * Even though the caller has a hold it can't prevent a concurrent 2565 * ire_delete marking the IRE condemned 2566 */ 2567 if (!IRE_IS_CONDEMNED(ire)) { 2568 generation = ire->ire_generation + 1; 2569 if (generation == IRE_GENERATION_CONDEMNED) 2570 generation = IRE_GENERATION_INITIAL; 2571 ASSERT(generation != IRE_GENERATION_VERIFY); 2572 ire->ire_generation = generation; 2573 } 2574 mutex_exit(&ire->ire_lock); 2575 } 2576 2577 /* 2578 * Increment ire_generation on all the IRE_MULTICASTs 2579 * Used when the default multicast interface (as determined by 2580 * ill_lookup_multicast) might have changed. 2581 * 2582 * That includes the zoneid, IFF_ flags, the IPv6 scope of the address, and 2583 * ill unplumb. 2584 */ 2585 void 2586 ire_increment_multicast_generation(ip_stack_t *ipst, boolean_t isv6) 2587 { 2588 ill_t *ill; 2589 ill_walk_context_t ctx; 2590 2591 rw_enter(&ipst->ips_ill_g_lock, RW_READER); 2592 if (isv6) 2593 ill = ILL_START_WALK_V6(&ctx, ipst); 2594 else 2595 ill = ILL_START_WALK_V4(&ctx, ipst); 2596 for (; ill != NULL; ill = ill_next(&ctx, ill)) { 2597 if (ILL_IS_CONDEMNED(ill)) 2598 continue; 2599 if (ill->ill_ire_multicast != NULL) 2600 ire_increment_generation(ill->ill_ire_multicast); 2601 } 2602 rw_exit(&ipst->ips_ill_g_lock); 2603 } 2604 2605 /* 2606 * Return a held IRE_NOROUTE with RTF_REJECT set 2607 */ 2608 ire_t * 2609 ire_reject(ip_stack_t *ipst, boolean_t isv6) 2610 { 2611 ire_t *ire; 2612 2613 if (isv6) 2614 ire = ipst->ips_ire_reject_v6; 2615 else 2616 ire = ipst->ips_ire_reject_v4; 2617 2618 ASSERT(ire->ire_generation != IRE_GENERATION_CONDEMNED); 2619 ire_refhold(ire); 2620 return (ire); 2621 } 2622 2623 /* 2624 * Return a held IRE_NOROUTE with RTF_BLACKHOLE set 2625 */ 2626 ire_t * 2627 ire_blackhole(ip_stack_t *ipst, boolean_t isv6) 2628 { 2629 ire_t *ire; 2630 2631 if (isv6) 2632 ire = ipst->ips_ire_blackhole_v6; 2633 else 2634 ire = ipst->ips_ire_blackhole_v4; 2635 2636 ASSERT(ire->ire_generation != IRE_GENERATION_CONDEMNED); 2637 ire_refhold(ire); 2638 return (ire); 2639 } 2640 2641 /* 2642 * Return a held IRE_MULTICAST. 2643 */ 2644 ire_t * 2645 ire_multicast(ill_t *ill) 2646 { 2647 ire_t *ire = ill->ill_ire_multicast; 2648 2649 ASSERT(ire == NULL || ire->ire_generation != IRE_GENERATION_CONDEMNED); 2650 if (ire == NULL) 2651 ire = ire_blackhole(ill->ill_ipst, ill->ill_isv6); 2652 else 2653 ire_refhold(ire); 2654 return (ire); 2655 } 2656 2657 /* 2658 * Given an IRE return its nexthop IRE. The nexthop IRE is an IRE_ONLINK 2659 * that is an exact match (i.e., a /32 for IPv4 and /128 for IPv6). 2660 * This can return an RTF_REJECT|RTF_BLACKHOLE. 2661 * The returned IRE is held. 2662 * The assumption is that ip_select_route() has been called and returned the 2663 * IRE (thus ip_select_route would have set up the ire_dep* information.) 2664 * If some IRE is deleteted then ire_dep_remove() will have been called and 2665 * we might not find a nexthop IRE, in which case we return NULL. 2666 */ 2667 ire_t * 2668 ire_nexthop(ire_t *ire) 2669 { 2670 ip_stack_t *ipst = ire->ire_ipst; 2671 2672 /* Acquire lock to walk ire_dep_parent */ 2673 rw_enter(&ipst->ips_ire_dep_lock, RW_READER); 2674 while (ire != NULL) { 2675 if (ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) { 2676 goto done; 2677 } 2678 /* 2679 * If we find an IRE_ONLINK we are done. This includes 2680 * the case of IRE_MULTICAST. 2681 * Note that in order to send packets we need a host-specific 2682 * IRE_IF_ALL first in the ire_dep_parent chain. Normally this 2683 * is done by inserting an IRE_IF_CLONE if the IRE_INTERFACE 2684 * was not host specific. 2685 * However, ip_rts_request doesn't want to send packets 2686 * hence doesn't want to allocate an IRE_IF_CLONE. Yet 2687 * it needs an IRE_IF_ALL to get to the ill. Thus 2688 * we return IRE_IF_ALL that are not host specific here. 2689 */ 2690 if (ire->ire_type & IRE_ONLINK) 2691 goto done; 2692 ire = ire->ire_dep_parent; 2693 } 2694 rw_exit(&ipst->ips_ire_dep_lock); 2695 return (NULL); 2696 2697 done: 2698 ire_refhold(ire); 2699 rw_exit(&ipst->ips_ire_dep_lock); 2700 return (ire); 2701 } 2702 2703 /* 2704 * Find the ill used to send packets. This will be NULL in case 2705 * of a reject or blackhole. 2706 * The returned ill is held; caller needs to do ill_refrele when done. 2707 */ 2708 ill_t * 2709 ire_nexthop_ill(ire_t *ire) 2710 { 2711 ill_t *ill; 2712 2713 ire = ire_nexthop(ire); 2714 if (ire == NULL) 2715 return (NULL); 2716 2717 /* ire_ill can not change for an existing ire */ 2718 ill = ire->ire_ill; 2719 if (ill != NULL) 2720 ill_refhold(ill); 2721 ire_refrele(ire); 2722 return (ill); 2723 } 2724 2725 #ifdef DEBUG 2726 static boolean_t 2727 parent_has_child(ire_t *parent, ire_t *child) 2728 { 2729 ire_t *ire; 2730 ire_t *prev; 2731 2732 ire = parent->ire_dep_children; 2733 prev = NULL; 2734 while (ire != NULL) { 2735 if (prev == NULL) { 2736 ASSERT(ire->ire_dep_sib_ptpn == 2737 &(parent->ire_dep_children)); 2738 } else { 2739 ASSERT(ire->ire_dep_sib_ptpn == 2740 &(prev->ire_dep_sib_next)); 2741 } 2742 if (ire == child) 2743 return (B_TRUE); 2744 prev = ire; 2745 ire = ire->ire_dep_sib_next; 2746 } 2747 return (B_FALSE); 2748 } 2749 2750 static void 2751 ire_dep_verify(ire_t *ire) 2752 { 2753 ire_t *parent = ire->ire_dep_parent; 2754 ire_t *child = ire->ire_dep_children; 2755 2756 ASSERT(ire->ire_ipversion == IPV4_VERSION || 2757 ire->ire_ipversion == IPV6_VERSION); 2758 if (parent != NULL) { 2759 ASSERT(parent->ire_ipversion == IPV4_VERSION || 2760 parent->ire_ipversion == IPV6_VERSION); 2761 ASSERT(parent->ire_refcnt >= 1); 2762 ASSERT(parent_has_child(parent, ire)); 2763 } 2764 if (child != NULL) { 2765 ASSERT(child->ire_ipversion == IPV4_VERSION || 2766 child->ire_ipversion == IPV6_VERSION); 2767 ASSERT(child->ire_dep_parent == ire); 2768 ASSERT(child->ire_dep_sib_ptpn != NULL); 2769 ASSERT(parent_has_child(ire, child)); 2770 } 2771 } 2772 #endif /* DEBUG */ 2773 2774 /* 2775 * Assumes ire_dep_parent is set. Remove this child from its parent's linkage. 2776 */ 2777 void 2778 ire_dep_remove(ire_t *ire) 2779 { 2780 ip_stack_t *ipst = ire->ire_ipst; 2781 ire_t *parent = ire->ire_dep_parent; 2782 ire_t *next; 2783 nce_t *nce; 2784 2785 ASSERT(RW_WRITE_HELD(&ipst->ips_ire_dep_lock)); 2786 ASSERT(ire->ire_dep_parent != NULL); 2787 ASSERT(ire->ire_dep_sib_ptpn != NULL); 2788 2789 #ifdef DEBUG 2790 ire_dep_verify(ire); 2791 ire_dep_verify(parent); 2792 #endif 2793 2794 next = ire->ire_dep_sib_next; 2795 if (next != NULL) 2796 next->ire_dep_sib_ptpn = ire->ire_dep_sib_ptpn; 2797 2798 ASSERT(*(ire->ire_dep_sib_ptpn) == ire); 2799 *(ire->ire_dep_sib_ptpn) = ire->ire_dep_sib_next; 2800 2801 ire->ire_dep_sib_ptpn = NULL; 2802 ire->ire_dep_sib_next = NULL; 2803 2804 mutex_enter(&ire->ire_lock); 2805 parent = ire->ire_dep_parent; 2806 ire->ire_dep_parent = NULL; 2807 mutex_exit(&ire->ire_lock); 2808 2809 /* 2810 * Make sure all our children, grandchildren, etc set 2811 * ire_dep_parent_generation to IRE_GENERATION_VERIFY since 2812 * we can no longer guarantee than the children have a current 2813 * ire_nce_cache and ire_nexthop_ill(). 2814 */ 2815 if (ire->ire_dep_children != NULL) 2816 ire_dep_invalidate_children(ire->ire_dep_children); 2817 2818 /* 2819 * Since the parent is gone we make sure we clear ire_nce_cache. 2820 * We can clear it under ire_lock even if the IRE is used 2821 */ 2822 mutex_enter(&ire->ire_lock); 2823 nce = ire->ire_nce_cache; 2824 ire->ire_nce_cache = NULL; 2825 mutex_exit(&ire->ire_lock); 2826 if (nce != NULL) 2827 nce_refrele(nce); 2828 2829 #ifdef DEBUG 2830 ire_dep_verify(ire); 2831 ire_dep_verify(parent); 2832 #endif 2833 2834 ire_refrele_notr(parent); 2835 ire_refrele_notr(ire); 2836 } 2837 2838 /* 2839 * Insert the child in the linkage of the parent 2840 */ 2841 static void 2842 ire_dep_parent_insert(ire_t *child, ire_t *parent) 2843 { 2844 ip_stack_t *ipst = child->ire_ipst; 2845 ire_t *next; 2846 2847 ASSERT(RW_WRITE_HELD(&ipst->ips_ire_dep_lock)); 2848 ASSERT(child->ire_dep_parent == NULL); 2849 2850 #ifdef DEBUG 2851 ire_dep_verify(child); 2852 ire_dep_verify(parent); 2853 #endif 2854 /* No parents => no siblings */ 2855 ASSERT(child->ire_dep_sib_ptpn == NULL); 2856 ASSERT(child->ire_dep_sib_next == NULL); 2857 2858 ire_refhold_notr(parent); 2859 ire_refhold_notr(child); 2860 2861 /* Head insertion */ 2862 next = parent->ire_dep_children; 2863 if (next != NULL) { 2864 ASSERT(next->ire_dep_sib_ptpn == &(parent->ire_dep_children)); 2865 child->ire_dep_sib_next = next; 2866 next->ire_dep_sib_ptpn = &(child->ire_dep_sib_next); 2867 } 2868 parent->ire_dep_children = child; 2869 child->ire_dep_sib_ptpn = &(parent->ire_dep_children); 2870 2871 mutex_enter(&child->ire_lock); 2872 child->ire_dep_parent = parent; 2873 mutex_exit(&child->ire_lock); 2874 2875 #ifdef DEBUG 2876 ire_dep_verify(child); 2877 ire_dep_verify(parent); 2878 #endif 2879 } 2880 2881 2882 /* 2883 * Given count worth of ires and generations, build ire_dep_* relationships 2884 * from ires[0] to ires[count-1]. Record generations[i+1] in 2885 * ire_dep_parent_generation for ires[i]. 2886 * We graft onto an existing parent chain by making sure that we don't 2887 * touch ire_dep_parent for ires[count-1]. 2888 * 2889 * We check for any condemned ire_generation count and return B_FALSE in 2890 * that case so that the caller can tear it apart. 2891 * 2892 * Note that generations[0] is not used. Caller handles that. 2893 */ 2894 boolean_t 2895 ire_dep_build(ire_t *ires[], uint_t generations[], uint_t count) 2896 { 2897 ire_t *ire = ires[0]; 2898 ip_stack_t *ipst; 2899 uint_t i; 2900 2901 ASSERT(count > 0); 2902 if (count == 1) { 2903 /* No work to do */ 2904 return (B_TRUE); 2905 } 2906 ipst = ire->ire_ipst; 2907 rw_enter(&ipst->ips_ire_dep_lock, RW_WRITER); 2908 /* 2909 * Do not remove the linkage for any existing parent chain i.e., 2910 * ires[count-1] is left alone. 2911 */ 2912 for (i = 0; i < count-1; i++) { 2913 /* Remove existing parent if we need to change it */ 2914 if (ires[i]->ire_dep_parent != NULL && 2915 ires[i]->ire_dep_parent != ires[i+1]) 2916 ire_dep_remove(ires[i]); 2917 } 2918 2919 for (i = 0; i < count - 1; i++) { 2920 ASSERT(ires[i]->ire_ipversion == IPV4_VERSION || 2921 ires[i]->ire_ipversion == IPV6_VERSION); 2922 /* Does it need to change? */ 2923 if (ires[i]->ire_dep_parent != ires[i+1]) 2924 ire_dep_parent_insert(ires[i], ires[i+1]); 2925 2926 mutex_enter(&ires[i+1]->ire_lock); 2927 if (IRE_IS_CONDEMNED(ires[i+1])) { 2928 mutex_exit(&ires[i+1]->ire_lock); 2929 rw_exit(&ipst->ips_ire_dep_lock); 2930 return (B_FALSE); 2931 } 2932 mutex_exit(&ires[i+1]->ire_lock); 2933 2934 mutex_enter(&ires[i]->ire_lock); 2935 ires[i]->ire_dep_parent_generation = generations[i+1]; 2936 mutex_exit(&ires[i]->ire_lock); 2937 } 2938 rw_exit(&ipst->ips_ire_dep_lock); 2939 return (B_TRUE); 2940 } 2941 2942 /* 2943 * Given count worth of ires, unbuild ire_dep_* relationships 2944 * from ires[0] to ires[count-1]. 2945 */ 2946 void 2947 ire_dep_unbuild(ire_t *ires[], uint_t count) 2948 { 2949 ip_stack_t *ipst; 2950 uint_t i; 2951 2952 if (count == 0) { 2953 /* No work to do */ 2954 return; 2955 } 2956 ipst = ires[0]->ire_ipst; 2957 rw_enter(&ipst->ips_ire_dep_lock, RW_WRITER); 2958 for (i = 0; i < count; i++) { 2959 ASSERT(ires[i]->ire_ipversion == IPV4_VERSION || 2960 ires[i]->ire_ipversion == IPV6_VERSION); 2961 if (ires[i]->ire_dep_parent != NULL) 2962 ire_dep_remove(ires[i]); 2963 mutex_enter(&ires[i]->ire_lock); 2964 ires[i]->ire_dep_parent_generation = IRE_GENERATION_VERIFY; 2965 mutex_exit(&ires[i]->ire_lock); 2966 } 2967 rw_exit(&ipst->ips_ire_dep_lock); 2968 } 2969 2970 /* 2971 * Both the forwarding and the outbound code paths can trip on 2972 * a condemned NCE, in which case we call this function. 2973 * We have two different behaviors: if the NCE was UNREACHABLE 2974 * it is an indication that something failed. In that case 2975 * we see if we should look for a different IRE (for example, 2976 * delete any matching redirect IRE, or try a different 2977 * IRE_DEFAULT (ECMP)). We mark the ire as bad so a hopefully 2978 * different IRE will be picked next time we send/forward. 2979 * 2980 * If we are called by the output path then fail_if_better is set 2981 * and we return NULL if there could be a better IRE. This is because the 2982 * output path retries the IRE lookup. (The input/forward path can not retry.) 2983 * 2984 * If the NCE was not unreachable then we pick/allocate a 2985 * new (most likely ND_INITIAL) NCE and proceed with it. 2986 * 2987 * ipha/ip6h are needed for multicast packets; ipha needs to be 2988 * set for IPv4 and ip6h needs to be set for IPv6 packets. 2989 */ 2990 nce_t * 2991 ire_handle_condemned_nce(nce_t *nce, ire_t *ire, ipha_t *ipha, ip6_t *ip6h, 2992 boolean_t fail_if_better) 2993 { 2994 if (nce->nce_common->ncec_state == ND_UNREACHABLE) { 2995 if (ire_no_good(ire) && fail_if_better) { 2996 /* 2997 * Did some changes, or ECMP likely to exist. 2998 * Make ip_output look for a different IRE 2999 */ 3000 return (NULL); 3001 } 3002 } 3003 if (ire_revalidate_nce(ire) == ENETUNREACH) { 3004 /* The ire_dep_parent chain went bad, or no memory? */ 3005 (void) ire_no_good(ire); 3006 return (NULL); 3007 } 3008 if (ire->ire_ipversion == IPV4_VERSION) { 3009 ASSERT(ipha != NULL); 3010 nce = ire_to_nce(ire, ipha->ipha_dst, NULL); 3011 } else { 3012 ASSERT(ip6h != NULL); 3013 nce = ire_to_nce(ire, INADDR_ANY, &ip6h->ip6_dst); 3014 } 3015 3016 if (nce == NULL) 3017 return (NULL); 3018 if (nce->nce_is_condemned) { 3019 nce_refrele(nce); 3020 return (NULL); 3021 } 3022 return (nce); 3023 } 3024 3025 /* 3026 * The caller has found that the ire is bad, either due to a reference to an NCE 3027 * in ND_UNREACHABLE state, or a MULTIRT route whose gateway can't be resolved. 3028 * We update things so a subsequent attempt to send to the destination 3029 * is likely to find different IRE, or that a new NCE would be created. 3030 * 3031 * Returns B_TRUE if it is likely that a subsequent ire_ftable_lookup would 3032 * find a different route (either due to having deleted a redirect, or there 3033 * being ECMP routes.) 3034 * 3035 * If we have a redirect (RTF_DYNAMIC) we delete it. 3036 * Otherwise we increment ire_badcnt and increment the generation number so 3037 * that a cached ixa_ire will redo the route selection. ire_badcnt is taken 3038 * into account in the route selection when we have multiple choices (multiple 3039 * default routes or ECMP in general). 3040 * Any time ip_select_route find an ire with a condemned ire_nce_cache 3041 * (e.g., if no equal cost route to the bad one) ip_select_route will make 3042 * sure the NCE is revalidated to avoid getting stuck on a 3043 * NCE_F_CONDMNED ncec that caused ire_no_good to be called. 3044 */ 3045 boolean_t 3046 ire_no_good(ire_t *ire) 3047 { 3048 ip_stack_t *ipst = ire->ire_ipst; 3049 ire_t *ire2; 3050 nce_t *nce; 3051 3052 if (ire->ire_flags & RTF_DYNAMIC) { 3053 ire_delete(ire); 3054 return (B_TRUE); 3055 } 3056 if (ire->ire_flags & RTF_INDIRECT) { 3057 /* Check if next IRE is a redirect */ 3058 rw_enter(&ipst->ips_ire_dep_lock, RW_READER); 3059 if (ire->ire_dep_parent != NULL && 3060 (ire->ire_dep_parent->ire_flags & RTF_DYNAMIC)) { 3061 ire2 = ire->ire_dep_parent; 3062 ire_refhold(ire2); 3063 } else { 3064 ire2 = NULL; 3065 } 3066 rw_exit(&ipst->ips_ire_dep_lock); 3067 if (ire2 != NULL) { 3068 ire_delete(ire2); 3069 ire_refrele(ire2); 3070 return (B_TRUE); 3071 } 3072 } 3073 /* 3074 * No redirect involved. Increment badcnt so that if we have ECMP 3075 * routes we are likely to pick a different one for the next packet. 3076 * 3077 * If the NCE is unreachable and condemned we should drop the reference 3078 * to it so that a new NCE can be created. 3079 * 3080 * Finally we increment the generation number so that any ixa_ire 3081 * cache will be revalidated. 3082 */ 3083 mutex_enter(&ire->ire_lock); 3084 ire->ire_badcnt++; 3085 ire->ire_last_badcnt = TICK_TO_SEC(ddi_get_lbolt64()); 3086 nce = ire->ire_nce_cache; 3087 if (nce != NULL && nce->nce_is_condemned && 3088 nce->nce_common->ncec_state == ND_UNREACHABLE) 3089 ire->ire_nce_cache = NULL; 3090 else 3091 nce = NULL; 3092 mutex_exit(&ire->ire_lock); 3093 if (nce != NULL) 3094 nce_refrele(nce); 3095 3096 ire_increment_generation(ire); 3097 ire_dep_incr_generation(ire); 3098 3099 return (ire->ire_bucket->irb_ire_cnt > 1); 3100 } 3101 3102 /* 3103 * Walk ire_dep_parent chain and validate that ire_dep_parent->ire_generation == 3104 * ire_dep_parent_generation. 3105 * If they all match we just return ire_generation from the topmost IRE. 3106 * Otherwise we propagate the mismatch by setting all ire_dep_parent_generation 3107 * above the mismatch to IRE_GENERATION_VERIFY and also returning 3108 * IRE_GENERATION_VERIFY. 3109 */ 3110 uint_t 3111 ire_dep_validate_generations(ire_t *ire) 3112 { 3113 ip_stack_t *ipst = ire->ire_ipst; 3114 uint_t generation; 3115 ire_t *ire1; 3116 3117 rw_enter(&ipst->ips_ire_dep_lock, RW_READER); 3118 generation = ire->ire_generation; /* Assuming things match */ 3119 for (ire1 = ire; ire1 != NULL; ire1 = ire1->ire_dep_parent) { 3120 ASSERT(ire1->ire_ipversion == IPV4_VERSION || 3121 ire1->ire_ipversion == IPV6_VERSION); 3122 if (ire1->ire_dep_parent == NULL) 3123 break; 3124 if (ire1->ire_dep_parent_generation != 3125 ire1->ire_dep_parent->ire_generation) 3126 goto mismatch; 3127 } 3128 rw_exit(&ipst->ips_ire_dep_lock); 3129 return (generation); 3130 3131 mismatch: 3132 generation = IRE_GENERATION_VERIFY; 3133 /* Fill from top down to the mismatch with _VERIFY */ 3134 while (ire != ire1) { 3135 ASSERT(ire->ire_ipversion == IPV4_VERSION || 3136 ire->ire_ipversion == IPV6_VERSION); 3137 mutex_enter(&ire->ire_lock); 3138 ire->ire_dep_parent_generation = IRE_GENERATION_VERIFY; 3139 mutex_exit(&ire->ire_lock); 3140 ire = ire->ire_dep_parent; 3141 } 3142 rw_exit(&ipst->ips_ire_dep_lock); 3143 return (generation); 3144 } 3145 3146 /* 3147 * Used when we need to return an ire with ire_dep_parent, but we 3148 * know the chain is invalid for instance we didn't create an IRE_IF_CLONE 3149 * Using IRE_GENERATION_VERIFY means that next time we'll redo the 3150 * recursive lookup. 3151 */ 3152 void 3153 ire_dep_invalidate_generations(ire_t *ire) 3154 { 3155 ip_stack_t *ipst = ire->ire_ipst; 3156 3157 rw_enter(&ipst->ips_ire_dep_lock, RW_READER); 3158 while (ire != NULL) { 3159 ASSERT(ire->ire_ipversion == IPV4_VERSION || 3160 ire->ire_ipversion == IPV6_VERSION); 3161 mutex_enter(&ire->ire_lock); 3162 ire->ire_dep_parent_generation = IRE_GENERATION_VERIFY; 3163 mutex_exit(&ire->ire_lock); 3164 ire = ire->ire_dep_parent; 3165 } 3166 rw_exit(&ipst->ips_ire_dep_lock); 3167 } 3168 3169 /* Set _VERIFY ire_dep_parent_generation for all children recursively */ 3170 static void 3171 ire_dep_invalidate_children(ire_t *child) 3172 { 3173 ip_stack_t *ipst = child->ire_ipst; 3174 3175 ASSERT(RW_WRITE_HELD(&ipst->ips_ire_dep_lock)); 3176 /* Depth first */ 3177 if (child->ire_dep_children != NULL) 3178 ire_dep_invalidate_children(child->ire_dep_children); 3179 3180 while (child != NULL) { 3181 mutex_enter(&child->ire_lock); 3182 child->ire_dep_parent_generation = IRE_GENERATION_VERIFY; 3183 mutex_exit(&child->ire_lock); 3184 child = child->ire_dep_sib_next; 3185 } 3186 } 3187 3188 static void 3189 ire_dep_increment_children(ire_t *child) 3190 { 3191 ip_stack_t *ipst = child->ire_ipst; 3192 3193 ASSERT(RW_READ_HELD(&ipst->ips_ire_dep_lock)); 3194 /* Depth first */ 3195 if (child->ire_dep_children != NULL) 3196 ire_dep_increment_children(child->ire_dep_children); 3197 3198 while (child != NULL) { 3199 if (!IRE_IS_CONDEMNED(child)) 3200 ire_increment_generation(child); 3201 child = child->ire_dep_sib_next; 3202 } 3203 } 3204 3205 /* 3206 * Walk all the children of this ire recursively and increment their 3207 * generation number. 3208 */ 3209 void 3210 ire_dep_incr_generation(ire_t *parent) 3211 { 3212 ip_stack_t *ipst = parent->ire_ipst; 3213 3214 rw_enter(&ipst->ips_ire_dep_lock, RW_READER); 3215 if (parent->ire_dep_children != NULL) 3216 ire_dep_increment_children(parent->ire_dep_children); 3217 rw_exit(&ipst->ips_ire_dep_lock); 3218 } 3219 3220 /* 3221 * Get a new ire_nce_cache for this IRE as well as its nexthop. 3222 * Returns zero if it succeeds. Can fail due to lack of memory or when 3223 * the route has become unreachable. Returns ENOMEM and ENETUNREACH in those 3224 * cases. 3225 * 3226 * In the in.mpathd case, the ire will have ire_testhidden 3227 * set; so we should create the ncec for the underlying ill. 3228 * 3229 * Note that the error returned by ire_revalidate_nce() is ignored by most 3230 * callers except ire_handle_condemned_nce(), which handles the ENETUNREACH 3231 * error to mark potentially bad ire's. For all the other callers, an 3232 * error return could indicate a transient condition like ENOMEM, or could 3233 * be the result of an interface that is going down/unplumbing. In the former 3234 * case (transient error), we would leave the old stale ire/ire_nce_cache 3235 * in place, and possibly use incorrect link-layer information to send packets 3236 * but would eventually recover. In the latter case (ill down/replumb), 3237 * ire_revalidate_nce() might return a condemned nce back, but we would then 3238 * recover in the packet output path. 3239 */ 3240 int 3241 ire_revalidate_nce(ire_t *ire) 3242 { 3243 nce_t *nce, *old_nce; 3244 ire_t *nexthop; 3245 3246 /* 3247 * For multicast we conceptually have an NCE but we don't store it 3248 * in ire_nce_cache; when ire_to_nce is called we allocate the nce. 3249 */ 3250 if (ire->ire_type & IRE_MULTICAST) 3251 return (0); 3252 3253 /* ire_testhidden should only be set on under-interfaces */ 3254 ASSERT(!ire->ire_testhidden || !IS_IPMP(ire->ire_ill)); 3255 3256 nexthop = ire_nexthop(ire); 3257 if (nexthop == NULL) { 3258 /* The route is potentially bad */ 3259 (void) ire_no_good(ire); 3260 return (ENETUNREACH); 3261 } 3262 if (ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK)) { 3263 ASSERT(ire->ire_ill != NULL); 3264 3265 if (ire->ire_ipversion == IPV4_VERSION) 3266 nce = nce_lookup_v4(ire->ire_ill, &ire->ire_addr); 3267 else 3268 nce = nce_lookup_v6(ire->ire_ill, &ire->ire_addr_v6); 3269 } else { 3270 ASSERT(nexthop->ire_type & IRE_ONLINK); 3271 if (ire->ire_ipversion == IPV4_VERSION) { 3272 nce = arp_nce_init(nexthop->ire_ill, nexthop->ire_addr, 3273 nexthop->ire_type); 3274 } else { 3275 nce = ndp_nce_init(nexthop->ire_ill, 3276 &nexthop->ire_addr_v6, nexthop->ire_type); 3277 } 3278 } 3279 if (nce == NULL) { 3280 /* 3281 * Leave the old stale one in place to avoid a NULL 3282 * ire_nce_cache. 3283 */ 3284 ire_refrele(nexthop); 3285 return (ENOMEM); 3286 } 3287 3288 if (nexthop != ire) { 3289 /* Update the nexthop ire */ 3290 mutex_enter(&nexthop->ire_lock); 3291 old_nce = nexthop->ire_nce_cache; 3292 if (!IRE_IS_CONDEMNED(nexthop)) { 3293 nce_refhold(nce); 3294 nexthop->ire_nce_cache = nce; 3295 } else { 3296 nexthop->ire_nce_cache = NULL; 3297 } 3298 mutex_exit(&nexthop->ire_lock); 3299 if (old_nce != NULL) 3300 nce_refrele(old_nce); 3301 } 3302 ire_refrele(nexthop); 3303 3304 mutex_enter(&ire->ire_lock); 3305 old_nce = ire->ire_nce_cache; 3306 if (!IRE_IS_CONDEMNED(ire)) { 3307 nce_refhold(nce); 3308 ire->ire_nce_cache = nce; 3309 } else { 3310 ire->ire_nce_cache = NULL; 3311 } 3312 mutex_exit(&ire->ire_lock); 3313 if (old_nce != NULL) 3314 nce_refrele(old_nce); 3315 3316 nce_refrele(nce); 3317 return (0); 3318 } 3319 3320 /* 3321 * Get a held nce for a given ire. 3322 * In the common case this is just from ire_nce_cache. 3323 * For IRE_MULTICAST this needs to do an explicit lookup since we do not 3324 * have an IRE_MULTICAST per address. 3325 * Note that this explicitly returns CONDEMNED NCEs. The caller needs those 3326 * so they can check whether the NCE went unreachable (as opposed to was 3327 * condemned for some other reason). 3328 */ 3329 nce_t * 3330 ire_to_nce(ire_t *ire, ipaddr_t v4nexthop, const in6_addr_t *v6nexthop) 3331 { 3332 nce_t *nce; 3333 3334 if (ire->ire_flags & (RTF_REJECT|RTF_BLACKHOLE)) 3335 return (NULL); 3336 3337 /* ire_testhidden should only be set on under-interfaces */ 3338 ASSERT(!ire->ire_testhidden || !IS_IPMP(ire->ire_ill)); 3339 3340 mutex_enter(&ire->ire_lock); 3341 nce = ire->ire_nce_cache; 3342 if (nce != NULL) { 3343 nce_refhold(nce); 3344 mutex_exit(&ire->ire_lock); 3345 return (nce); 3346 } 3347 mutex_exit(&ire->ire_lock); 3348 3349 if (ire->ire_type & IRE_MULTICAST) { 3350 ASSERT(ire->ire_ill != NULL); 3351 3352 if (ire->ire_ipversion == IPV4_VERSION) { 3353 ASSERT(v6nexthop == NULL); 3354 3355 nce = arp_nce_init(ire->ire_ill, v4nexthop, 3356 ire->ire_type); 3357 } else { 3358 ASSERT(v6nexthop != NULL); 3359 ASSERT(v4nexthop == 0); 3360 nce = ndp_nce_init(ire->ire_ill, v6nexthop, 3361 ire->ire_type); 3362 } 3363 return (nce); 3364 } 3365 return (NULL); 3366 } 3367 3368 nce_t * 3369 ire_to_nce_pkt(ire_t *ire, mblk_t *mp) 3370 { 3371 ipha_t *ipha; 3372 ip6_t *ip6h; 3373 3374 if (IPH_HDR_VERSION(mp->b_rptr) == IPV4_VERSION) { 3375 ipha = (ipha_t *)mp->b_rptr; 3376 return (ire_to_nce(ire, ipha->ipha_dst, NULL)); 3377 } else { 3378 ip6h = (ip6_t *)mp->b_rptr; 3379 return (ire_to_nce(ire, INADDR_ANY, &ip6h->ip6_dst)); 3380 } 3381 } 3382 3383 /* 3384 * Given an IRE_INTERFACE (that matches more than one address) create 3385 * and return an IRE_IF_CLONE for the specific address. 3386 * Return the generation number. 3387 * Returns NULL is no memory for the IRE. 3388 * Handles both IPv4 and IPv6. 3389 */ 3390 ire_t * 3391 ire_create_if_clone(ire_t *ire_if, const in6_addr_t *addr, uint_t *generationp) 3392 { 3393 ire_t *ire; 3394 ire_t *nire; 3395 3396 if (ire_if->ire_ipversion == IPV4_VERSION) { 3397 ipaddr_t v4addr; 3398 ipaddr_t mask = IP_HOST_MASK; 3399 3400 ASSERT(IN6_IS_ADDR_V4MAPPED(addr)); 3401 IN6_V4MAPPED_TO_IPADDR(addr, v4addr); 3402 3403 ire = ire_create( 3404 (uchar_t *)&v4addr, /* dest address */ 3405 (uchar_t *)&mask, /* mask */ 3406 (uchar_t *)&ire_if->ire_gateway_addr, 3407 IRE_IF_CLONE, /* IRE type */ 3408 ire_if->ire_ill, 3409 ire_if->ire_zoneid, 3410 ire_if->ire_flags | RTF_HOST, 3411 NULL, /* No security attr for IRE_IF_ALL */ 3412 ire_if->ire_ipst); 3413 } else { 3414 ASSERT(!IN6_IS_ADDR_V4MAPPED(addr)); 3415 ire = ire_create_v6( 3416 addr, /* dest address */ 3417 &ipv6_all_ones, /* mask */ 3418 &ire_if->ire_gateway_addr_v6, /* gateway addr */ 3419 IRE_IF_CLONE, /* IRE type */ 3420 ire_if->ire_ill, 3421 ire_if->ire_zoneid, 3422 ire_if->ire_flags | RTF_HOST, 3423 NULL, /* No security attr for IRE_IF_ALL */ 3424 ire_if->ire_ipst); 3425 } 3426 if (ire == NULL) 3427 return (NULL); 3428 3429 /* Take the metrics, in particular the mtu, from the IRE_IF */ 3430 ire->ire_metrics = ire_if->ire_metrics; 3431 3432 nire = ire_add(ire); 3433 if (nire == NULL) /* Some failure */ 3434 return (NULL); 3435 3436 if (generationp != NULL) 3437 *generationp = nire->ire_generation; 3438 3439 /* 3440 * Make sure races don't add a duplicate by 3441 * catching the case when an identical was returned. 3442 */ 3443 if (nire != ire) { 3444 ASSERT(nire->ire_identical_ref > 1); 3445 ire_delete(nire); 3446 } 3447 return (nire); 3448 } 3449 3450 /* 3451 * The argument is an IRE_INTERFACE. Delete all of IRE_IF_CLONE in the 3452 * ire_dep_children (just walk the ire_dep_sib_next since they are all 3453 * immediate children.) 3454 * Since we hold a lock while we remove them we need to defer the actual 3455 * calls to ire_delete() until we have dropped the lock. This makes things 3456 * less efficient since we restart at the top after dropping the lock. But 3457 * we only run when an IRE_INTERFACE is deleted which is infrquent. 3458 * 3459 * Note that ire_dep_children can be any mixture of offlink routes and 3460 * IRE_IF_CLONE entries. 3461 */ 3462 void 3463 ire_dep_delete_if_clone(ire_t *parent) 3464 { 3465 ip_stack_t *ipst = parent->ire_ipst; 3466 ire_t *child, *next; 3467 3468 restart: 3469 rw_enter(&ipst->ips_ire_dep_lock, RW_READER); 3470 if (parent->ire_dep_children == NULL) { 3471 rw_exit(&ipst->ips_ire_dep_lock); 3472 return; 3473 } 3474 child = parent->ire_dep_children; 3475 while (child != NULL) { 3476 next = child->ire_dep_sib_next; 3477 if ((child->ire_type & IRE_IF_CLONE) && 3478 !IRE_IS_CONDEMNED(child)) { 3479 ire_refhold(child); 3480 rw_exit(&ipst->ips_ire_dep_lock); 3481 ire_delete(child); 3482 ASSERT(IRE_IS_CONDEMNED(child)); 3483 ire_refrele(child); 3484 goto restart; 3485 } 3486 child = next; 3487 } 3488 rw_exit(&ipst->ips_ire_dep_lock); 3489 } 3490 3491 /* 3492 * ire_pref() is used in recursive route-resolution for a destination to 3493 * determine the preference of an ire, where "preference" is determined 3494 * based on the level of indirection to the destination of the ire. 3495 * A higher preference indicates that fewer lookups are needed to complete 3496 * recursive route lookup. Thus 3497 * ire_pref(RTF_INDIRECT) < ire_pref(IRE_IF_RESOLVER) < ire_pref(IRE_PREF_CLONE) 3498 */ 3499 int 3500 ire_pref(ire_t *ire) 3501 { 3502 if (ire->ire_flags & RTF_INDIRECT) 3503 return (1); 3504 if (ire->ire_type & IRE_OFFLINK) 3505 return (2); 3506 if (ire->ire_type & (IRE_IF_RESOLVER|IRE_IF_NORESOLVER)) 3507 return (3); 3508 if (ire->ire_type & IRE_IF_CLONE) 3509 return (4); 3510 if (ire->ire_type & (IRE_LOCAL|IRE_LOOPBACK|IRE_BROADCAST)) 3511 return (5); 3512 return (-1); /* unknown ire_type */ 3513 } 3514