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 2006 Sun Microsystems, Inc. All rights reserved. 23 * Use is subject to license terms. 24 */ 25 /* Copyright (c) 1990 Mentat Inc. */ 26 27 #pragma ident "%Z%%M% %I% %E% SMI" 28 29 30 /* 31 * This file contains routines that manipulate Internet Routing Entries (IREs). 32 */ 33 34 #include <sys/types.h> 35 #include <sys/stream.h> 36 #include <sys/stropts.h> 37 #include <sys/strsun.h> 38 #include <sys/ddi.h> 39 #include <sys/cmn_err.h> 40 #include <sys/policy.h> 41 42 #include <sys/systm.h> 43 #include <sys/kmem.h> 44 #include <sys/param.h> 45 #include <sys/socket.h> 46 #include <net/if.h> 47 #include <net/route.h> 48 #include <netinet/in.h> 49 #include <net/if_dl.h> 50 #include <netinet/ip6.h> 51 #include <netinet/icmp6.h> 52 53 #include <inet/common.h> 54 #include <inet/mi.h> 55 #include <inet/ip.h> 56 #include <inet/ip6.h> 57 #include <inet/ip_ndp.h> 58 #include <inet/arp.h> 59 #include <inet/ip_if.h> 60 #include <inet/ip_ire.h> 61 #include <inet/ip_ftable.h> 62 #include <inet/ip_rts.h> 63 #include <inet/nd.h> 64 65 #include <net/pfkeyv2.h> 66 #include <inet/ipsec_info.h> 67 #include <inet/sadb.h> 68 #include <sys/kmem.h> 69 #include <inet/tcp.h> 70 #include <inet/ipclassifier.h> 71 #include <sys/zone.h> 72 #include <sys/tsol/label.h> 73 #include <sys/tsol/tnet.h> 74 75 struct kmem_cache *rt_entry_cache; 76 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 respective tables (cache or forwarding tables). 86 * 87 * ire_mp, ire_rfq, ire_stq, ire_u *except* ire_gateway_addr[v6], ire_mask, 88 * ire_type, ire_create_time, ire_masklen, ire_ipversion, ire_flags, ire_ipif, 89 * ire_ihandle, ire_phandle, ire_nce, ire_bucket, ire_in_ill, ire_in_src_addr 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_v4/ire_add_v6 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_ident, ire_refcnt 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_max_frag, ire_frag_flag 117 * 118 * - ire_lock is used to set/read both of them together. 119 * 120 * ire_tire_mark 121 * 122 * - Set in ire_create and updated in ire_expire, which is called 123 * by only one function namely ip_trash_timer_expire. Thus only 124 * one function updates and examines the value. 125 * 126 * ire_marks 127 * - bucket lock protects this. 128 * 129 * ire_ipsec_overhead/ire_ll_hdr_length 130 * 131 * - Place holder for returning the information to the upper layers 132 * when IRE_DB_REQ comes down. 133 * 134 * 135 * ipv6_ire_default_count is protected by the bucket lock of 136 * ip_forwarding_table_v6[0][0]. 137 * 138 * ipv6_ire_default_index is not protected as it is just a hint 139 * at which default gateway to use. There is nothing 140 * wrong in using the same gateway for two different connections. 141 * 142 * As we always hold the bucket locks in all the places while accessing 143 * the above values, it is natural to use them for protecting them. 144 * 145 * We have a separate cache table and forwarding table for IPv4 and IPv6. 146 * Cache table (ip_cache_table/ip_cache_table_v6) is a pointer to an 147 * array of irb_t structure and forwarding table (ip_forwarding_table/ 148 * ip_forwarding_table_v6) is an array of pointers to array of irb_t 149 * structure. ip_forwarding_table[_v6] is allocated dynamically in 150 * ire_add_v4/v6. ire_ft_init_lock is used to serialize multiple threads 151 * initializing the same bucket. Once a bucket is initialized, it is never 152 * de-alloacted. This assumption enables us to access ip_forwarding_table[i] 153 * or ip_forwarding_table_v6[i] without any locks. 154 * 155 * Each irb_t - ire bucket structure has a lock to protect 156 * a bucket and the ires residing in the bucket have a back pointer to 157 * the bucket structure. It also has a reference count for the number 158 * of threads walking the bucket - irb_refcnt which is bumped up 159 * using the macro IRB_REFHOLD macro. The flags irb_flags can be 160 * set to IRE_MARK_CONDEMNED indicating that there are some ires 161 * in this bucket that are marked with IRE_MARK_CONDEMNED and the 162 * last thread to leave the bucket should delete the ires. Usually 163 * this is done by the IRB_REFRELE macro which is used to decrement 164 * the reference count on a bucket. 165 * 166 * IRE_REFHOLD/IRE_REFRELE macros operate on the ire which increments/ 167 * decrements the reference count, ire_refcnt, atomically on the ire. 168 * ire_refcnt is modified only using this macro. Operations on the IRE 169 * could be described as follows : 170 * 171 * CREATE an ire with reference count initialized to 1. 172 * 173 * ADDITION of an ire holds the bucket lock, checks for duplicates 174 * and then adds the ire. ire_add_v4/ire_add_v6 returns the ire after 175 * bumping up once more i.e the reference count is 2. This is to avoid 176 * an extra lookup in the functions calling ire_add which wants to 177 * work with the ire after adding. 178 * 179 * LOOKUP of an ire bumps up the reference count using IRE_REFHOLD 180 * macro. It is valid to bump up the referece count of the IRE, 181 * after the lookup has returned an ire. Following are the lookup 182 * functions that return an HELD ire : 183 * 184 * ire_lookup_local[_v6], ire_ctable_lookup[_v6], ire_ftable_lookup[_v6], 185 * ire_cache_lookup[_v6], ire_lookup_multi[_v6], ire_route_lookup[_v6], 186 * ipif_to_ire[_v6], ire_mrtun_lookup, ire_srcif_table_lookup. 187 * 188 * DELETION of an ire holds the bucket lock, removes it from the list 189 * and then decrements the reference count for having removed from the list 190 * by using the IRE_REFRELE macro. If some other thread has looked up 191 * the ire, the reference count would have been bumped up and hence 192 * this ire will not be freed once deleted. It will be freed once the 193 * reference count drops to zero. 194 * 195 * Add and Delete acquires the bucket lock as RW_WRITER, while all the 196 * lookups acquire the bucket lock as RW_READER. 197 * 198 * NOTE : The only functions that does the IRE_REFRELE when an ire is 199 * passed as an argument are : 200 * 201 * 1) ip_wput_ire : This is because it IRE_REFHOLD/RELEs the 202 * broadcast ires it looks up internally within 203 * the function. Currently, for simplicity it does 204 * not differentiate the one that is passed in and 205 * the ones it looks up internally. It always 206 * IRE_REFRELEs. 207 * 2) ire_send 208 * ire_send_v6 : As ire_send calls ip_wput_ire and other functions 209 * that take ire as an argument, it has to selectively 210 * IRE_REFRELE the ire. To maintain symmetry, 211 * ire_send_v6 does the same. 212 * 213 * Otherwise, the general rule is to do the IRE_REFRELE in the function 214 * that is passing the ire as an argument. 215 * 216 * In trying to locate ires the following points are to be noted. 217 * 218 * IRE_MARK_CONDEMNED signifies that the ire has been logically deleted and is 219 * to be ignored when walking the ires using ire_next. 220 * 221 * IRE_MARK_HIDDEN signifies that the ire is a special ire typically for the 222 * benefit of in.mpathd which needs to probe interfaces for failures. Normal 223 * applications should not be seeing this ire and hence this ire is ignored 224 * in most cases in the search using ire_next. 225 * 226 * Zones note: 227 * Walking IREs within a given zone also walks certain ires in other 228 * zones. This is done intentionally. IRE walks with a specified 229 * zoneid are used only when doing informational reports, and 230 * zone users want to see things that they can access. See block 231 * comment in ire_walk_ill_match(). 232 */ 233 234 /* This is dynamically allocated in ip_ire_init */ 235 irb_t *ip_cache_table; 236 /* This is dynamically allocated in ire_add_mrtun */ 237 irb_t *ip_mrtun_table; 238 239 uint32_t ire_handle = 1; 240 /* 241 * ire_ft_init_lock is used while initializing ip_forwarding_table 242 * dynamically in ire_add. 243 */ 244 kmutex_t ire_ft_init_lock; 245 kmutex_t ire_mrtun_lock; /* Protects creation of table and it's count */ 246 kmutex_t ire_srcif_table_lock; /* Same as above */ 247 /* 248 * The following counts are used to determine whether a walk is 249 * needed through the reverse tunnel table or through ills 250 */ 251 kmutex_t ire_handle_lock; /* Protects ire_handle */ 252 uint_t ire_mrtun_count; /* Number of ires in reverse tun table */ 253 254 /* 255 * A per-interface routing table is created ( if not present) 256 * when the first entry is added to this special routing table. 257 * This special routing table is accessed through the ill data structure. 258 * The routing table looks like cache table. For example, currently it 259 * is used by mobile-ip foreign agent to forward data that only comes from 260 * the home agent tunnel for a mobile node. Thus if the outgoing interface 261 * is a RESOLVER interface, IP may need to resolve the hardware address for 262 * the outgoing interface. The routing entries in this table are not updated 263 * in IRE_CACHE. When MCTL msg comes back from ARP, the incoming ill informa- 264 * tion is lost as the write queue is passed to ip_wput. 265 * But, before sending the packet out, the hardware information must be updated 266 * in the special forwarding table. ire_srcif_table_count keeps track of total 267 * number of ires that are in interface based tables. Each interface based 268 * table hangs off of the incoming ill and each ill_t also keeps a refcnt 269 * of ires in that table. 270 */ 271 272 uint_t ire_srcif_table_count; /* Number of ires in all srcif tables */ 273 274 /* 275 * The minimum size of IRE cache table. It will be recalcuated in 276 * ip_ire_init(). 277 */ 278 uint32_t ip_cache_table_size = IP_CACHE_TABLE_SIZE; 279 uint32_t ip6_cache_table_size = IP6_CACHE_TABLE_SIZE; 280 281 /* 282 * The size of the forwarding table. We will make sure that it is a 283 * power of 2 in ip_ire_init(). 284 */ 285 uint32_t ip6_ftable_hash_size = IP6_FTABLE_HASH_SIZE; 286 287 struct kmem_cache *ire_cache; 288 static ire_t ire_null; 289 290 ire_stats_t ire_stats_v4; /* IPv4 ire statistics */ 291 ire_stats_t ire_stats_v6; /* IPv6 ire statistics */ 292 293 /* 294 * The threshold number of IRE in a bucket when the IREs are 295 * cleaned up. This threshold is calculated later in ip_open() 296 * based on the speed of CPU and available memory. This default 297 * value is the maximum. 298 * 299 * We have two kinds of cached IRE, temporary and 300 * non-temporary. Temporary IREs are marked with 301 * IRE_MARK_TEMPORARY. They are IREs created for non 302 * TCP traffic and for forwarding purposes. All others 303 * are non-temporary IREs. We don't mark IRE created for 304 * TCP as temporary because TCP is stateful and there are 305 * info stored in the IRE which can be shared by other TCP 306 * connections to the same destination. For connected 307 * endpoint, we also don't want to mark the IRE used as 308 * temporary because the same IRE will be used frequently, 309 * otherwise, the app should not do a connect(). We change 310 * the marking at ip_bind_connected_*() if necessary. 311 * 312 * We want to keep the cache IRE hash bucket length reasonably 313 * short, otherwise IRE lookup functions will take "forever." 314 * We use the "crude" function that the IRE bucket 315 * length should be based on the CPU speed, which is 1 entry 316 * per x MHz, depending on the shift factor ip_ire_cpu_ratio 317 * (n). This means that with a 750MHz CPU, the max bucket 318 * length can be (750 >> n) entries. 319 * 320 * Note that this threshold is separate for temp and non-temp 321 * IREs. This means that the actual bucket length can be 322 * twice as that. And while we try to keep temporary IRE 323 * length at most at the threshold value, we do not attempt to 324 * make the length for non-temporary IREs fixed, for the 325 * reason stated above. Instead, we start trying to find 326 * "unused" non-temporary IREs when the bucket length reaches 327 * this threshold and clean them up. 328 * 329 * We also want to limit the amount of memory used by 330 * IREs. So if we are allowed to use ~3% of memory (M) 331 * for those IREs, each bucket should not have more than 332 * 333 * M / num of cache bucket / sizeof (ire_t) 334 * 335 * Again the above memory uses are separate for temp and 336 * non-temp cached IREs. 337 * 338 * We may also want the limit to be a function of the number 339 * of interfaces and number of CPUs. Doing the initialization 340 * in ip_open() means that every time an interface is plumbed, 341 * the max is re-calculated. Right now, we don't do anything 342 * different. In future, when we have more experience, we 343 * may want to change this behavior. 344 */ 345 uint32_t ip_ire_max_bucket_cnt = 10; 346 uint32_t ip6_ire_max_bucket_cnt = 10; 347 348 /* 349 * The minimum of the temporary IRE bucket count. We do not want 350 * the length of each bucket to be too short. This may hurt 351 * performance of some apps as the temporary IREs are removed too 352 * often. 353 */ 354 uint32_t ip_ire_min_bucket_cnt = 3; 355 uint32_t ip6_ire_min_bucket_cnt = 3; 356 357 /* 358 * The ratio of memory consumed by IRE used for temporary to available 359 * memory. This is a shift factor, so 6 means the ratio 1 to 64. This 360 * value can be changed in /etc/system. 6 is a reasonable number. 361 */ 362 uint32_t ip_ire_mem_ratio = 6; 363 /* The shift factor for CPU speed to calculate the max IRE bucket length. */ 364 uint32_t ip_ire_cpu_ratio = 7; 365 366 typedef struct nce_clookup_s { 367 ipaddr_t ncecl_addr; 368 boolean_t ncecl_found; 369 } nce_clookup_t; 370 371 /* 372 * The maximum number of buckets in IRE cache table. In future, we may 373 * want to make it a dynamic hash table. For the moment, we fix the 374 * size and allocate the table in ip_ire_init() when IP is first loaded. 375 * We take into account the amount of memory a system has. 376 */ 377 #define IP_MAX_CACHE_TABLE_SIZE 4096 378 379 static uint32_t ip_max_cache_table_size = IP_MAX_CACHE_TABLE_SIZE; 380 static uint32_t ip6_max_cache_table_size = IP_MAX_CACHE_TABLE_SIZE; 381 382 #define NUM_ILLS 3 /* To build the ILL list to unlock */ 383 384 /* Zero iulp_t for initialization. */ 385 const iulp_t ire_uinfo_null = { 0 }; 386 387 static int ire_add_v4(ire_t **ire_p, queue_t *q, mblk_t *mp, 388 ipsq_func_t func, boolean_t); 389 static int ire_add_srcif_v4(ire_t **ire_p, queue_t *q, mblk_t *mp, 390 ipsq_func_t func); 391 static ire_t *ire_update_srcif_v4(ire_t *ire); 392 static void ire_delete_v4(ire_t *ire); 393 static void ire_report_ctable(ire_t *ire, char *mp); 394 static void ire_report_mrtun_table(ire_t *ire, char *mp); 395 static void ire_report_srcif_table(ire_t *ire, char *mp); 396 static void ire_walk_ipvers(pfv_t func, void *arg, uchar_t vers, 397 zoneid_t zoneid); 398 static void ire_walk_ill_ipvers(uint_t match_flags, uint_t ire_type, 399 pfv_t func, void *arg, uchar_t vers, ill_t *ill); 400 static void ire_cache_cleanup(irb_t *irb, uint32_t threshold, int cnt); 401 extern void ill_unlock_ills(ill_t **list, int cnt); 402 static void ire_fastpath_list_add(ill_t *ill, ire_t *ire); 403 static void ip_nce_clookup_and_delete(nce_t *nce, void *arg); 404 extern void th_trace_rrecord(th_trace_t *); 405 #ifdef IRE_DEBUG 406 static void ire_trace_inactive(ire_t *); 407 #endif 408 409 /* 410 * To avoid bloating the code, we call this function instead of 411 * using the macro IRE_REFRELE. Use macro only in performance 412 * critical paths. 413 * 414 * Must not be called while holding any locks. Otherwise if this is 415 * the last reference to be released there is a chance of recursive mutex 416 * panic due to ire_refrele -> ipif_ill_refrele_tail -> qwriter_ip trying 417 * to restart an ioctl. The one exception is when the caller is sure that 418 * this is not the last reference to be released. Eg. if the caller is 419 * sure that the ire has not been deleted and won't be deleted. 420 */ 421 void 422 ire_refrele(ire_t *ire) 423 { 424 IRE_REFRELE(ire); 425 } 426 427 void 428 ire_refrele_notr(ire_t *ire) 429 { 430 IRE_REFRELE_NOTR(ire); 431 } 432 433 /* 434 * kmem_cache_alloc constructor for IRE in kma space. 435 * Note that when ire_mp is set the IRE is stored in that mblk and 436 * not in this cache. 437 */ 438 /* ARGSUSED */ 439 static int 440 ip_ire_constructor(void *buf, void *cdrarg, int kmflags) 441 { 442 ire_t *ire = buf; 443 444 ire->ire_nce = NULL; 445 446 return (0); 447 } 448 449 /* ARGSUSED1 */ 450 static void 451 ip_ire_destructor(void *buf, void *cdrarg) 452 { 453 ire_t *ire = buf; 454 455 ASSERT(ire->ire_nce == NULL); 456 } 457 458 /* 459 * This function is associated with the IP_IOC_IRE_ADVISE_NO_REPLY 460 * IOCTL. It is used by TCP (or other ULPs) to supply revised information 461 * for an existing CACHED IRE. 462 */ 463 /* ARGSUSED */ 464 int 465 ip_ire_advise(queue_t *q, mblk_t *mp, cred_t *ioc_cr) 466 { 467 uchar_t *addr_ucp; 468 ipic_t *ipic; 469 ire_t *ire; 470 ipaddr_t addr; 471 in6_addr_t v6addr; 472 irb_t *irb; 473 zoneid_t zoneid; 474 475 ASSERT(q->q_next == NULL); 476 zoneid = Q_TO_CONN(q)->conn_zoneid; 477 478 /* 479 * Check privilege using the ioctl credential; if it is NULL 480 * then this is a kernel message and therefor privileged. 481 */ 482 if (ioc_cr != NULL && secpolicy_net_config(ioc_cr, B_FALSE) != 0) 483 return (EPERM); 484 485 ipic = (ipic_t *)mp->b_rptr; 486 if (!(addr_ucp = mi_offset_param(mp, ipic->ipic_addr_offset, 487 ipic->ipic_addr_length))) { 488 return (EINVAL); 489 } 490 if (!OK_32PTR(addr_ucp)) 491 return (EINVAL); 492 switch (ipic->ipic_addr_length) { 493 case IP_ADDR_LEN: { 494 /* Extract the destination address. */ 495 addr = *(ipaddr_t *)addr_ucp; 496 /* Find the corresponding IRE. */ 497 ire = ire_cache_lookup(addr, zoneid, NULL); 498 break; 499 } 500 case IPV6_ADDR_LEN: { 501 /* Extract the destination address. */ 502 v6addr = *(in6_addr_t *)addr_ucp; 503 /* Find the corresponding IRE. */ 504 ire = ire_cache_lookup_v6(&v6addr, zoneid, NULL); 505 break; 506 } 507 default: 508 return (EINVAL); 509 } 510 511 if (ire == NULL) 512 return (ENOENT); 513 /* 514 * Update the round trip time estimate and/or the max frag size 515 * and/or the slow start threshold. 516 * 517 * We serialize multiple advises using ire_lock. 518 */ 519 mutex_enter(&ire->ire_lock); 520 if (ipic->ipic_rtt) { 521 /* 522 * If there is no old cached values, initialize them 523 * conservatively. Set them to be (1.5 * new value). 524 */ 525 if (ire->ire_uinfo.iulp_rtt != 0) { 526 ire->ire_uinfo.iulp_rtt = (ire->ire_uinfo.iulp_rtt + 527 ipic->ipic_rtt) >> 1; 528 } else { 529 ire->ire_uinfo.iulp_rtt = ipic->ipic_rtt + 530 (ipic->ipic_rtt >> 1); 531 } 532 if (ire->ire_uinfo.iulp_rtt_sd != 0) { 533 ire->ire_uinfo.iulp_rtt_sd = 534 (ire->ire_uinfo.iulp_rtt_sd + 535 ipic->ipic_rtt_sd) >> 1; 536 } else { 537 ire->ire_uinfo.iulp_rtt_sd = ipic->ipic_rtt_sd + 538 (ipic->ipic_rtt_sd >> 1); 539 } 540 } 541 if (ipic->ipic_max_frag) 542 ire->ire_max_frag = MIN(ipic->ipic_max_frag, IP_MAXPACKET); 543 if (ipic->ipic_ssthresh != 0) { 544 if (ire->ire_uinfo.iulp_ssthresh != 0) 545 ire->ire_uinfo.iulp_ssthresh = 546 (ipic->ipic_ssthresh + 547 ire->ire_uinfo.iulp_ssthresh) >> 1; 548 else 549 ire->ire_uinfo.iulp_ssthresh = ipic->ipic_ssthresh; 550 } 551 /* 552 * Don't need the ire_lock below this. ire_type does not change 553 * after initialization. ire_marks is protected by irb_lock. 554 */ 555 mutex_exit(&ire->ire_lock); 556 557 if (ipic->ipic_ire_marks != 0 && ire->ire_type == IRE_CACHE) { 558 /* 559 * Only increment the temporary IRE count if the original 560 * IRE is not already marked temporary. 561 */ 562 irb = ire->ire_bucket; 563 rw_enter(&irb->irb_lock, RW_WRITER); 564 if ((ipic->ipic_ire_marks & IRE_MARK_TEMPORARY) && 565 !(ire->ire_marks & IRE_MARK_TEMPORARY)) { 566 irb->irb_tmp_ire_cnt++; 567 } 568 ire->ire_marks |= ipic->ipic_ire_marks; 569 rw_exit(&irb->irb_lock); 570 } 571 572 ire_refrele(ire); 573 return (0); 574 } 575 576 /* 577 * This function is associated with the IP_IOC_IRE_DELETE[_NO_REPLY] 578 * IOCTL[s]. The NO_REPLY form is used by TCP to delete a route IRE 579 * for a host that is not responding. This will force an attempt to 580 * establish a new route, if available, and flush out the ARP entry so 581 * it will re-resolve. Management processes may want to use the 582 * version that generates a reply. 583 * 584 * This function does not support IPv6 since Neighbor Unreachability Detection 585 * means that negative advise like this is useless. 586 */ 587 /* ARGSUSED */ 588 int 589 ip_ire_delete(queue_t *q, mblk_t *mp, cred_t *ioc_cr) 590 { 591 uchar_t *addr_ucp; 592 ipaddr_t addr; 593 ire_t *ire; 594 ipid_t *ipid; 595 boolean_t routing_sock_info = B_FALSE; /* Sent info? */ 596 zoneid_t zoneid; 597 ire_t *gire = NULL; 598 ill_t *ill; 599 mblk_t *arp_mp; 600 601 ASSERT(q->q_next == NULL); 602 zoneid = Q_TO_CONN(q)->conn_zoneid; 603 604 /* 605 * Check privilege using the ioctl credential; if it is NULL 606 * then this is a kernel message and therefor privileged. 607 */ 608 if (ioc_cr != NULL && secpolicy_net_config(ioc_cr, B_FALSE) != 0) 609 return (EPERM); 610 611 ipid = (ipid_t *)mp->b_rptr; 612 613 /* Only actions on IRE_CACHEs are acceptable at present. */ 614 if (ipid->ipid_ire_type != IRE_CACHE) 615 return (EINVAL); 616 617 addr_ucp = mi_offset_param(mp, ipid->ipid_addr_offset, 618 ipid->ipid_addr_length); 619 if (addr_ucp == NULL || !OK_32PTR(addr_ucp)) 620 return (EINVAL); 621 switch (ipid->ipid_addr_length) { 622 case IP_ADDR_LEN: 623 /* addr_ucp points at IP addr */ 624 break; 625 case sizeof (sin_t): { 626 sin_t *sin; 627 /* 628 * got complete (sockaddr) address - increment addr_ucp to point 629 * at the ip_addr field. 630 */ 631 sin = (sin_t *)addr_ucp; 632 addr_ucp = (uchar_t *)&sin->sin_addr.s_addr; 633 break; 634 } 635 default: 636 return (EINVAL); 637 } 638 /* Extract the destination address. */ 639 bcopy(addr_ucp, &addr, IP_ADDR_LEN); 640 641 /* Try to find the CACHED IRE. */ 642 ire = ire_cache_lookup(addr, zoneid, NULL); 643 644 /* Nail it. */ 645 if (ire) { 646 /* Allow delete only on CACHE entries */ 647 if (ire->ire_type != IRE_CACHE) { 648 ire_refrele(ire); 649 return (EINVAL); 650 } 651 652 /* 653 * Verify that the IRE has been around for a while. 654 * This is to protect against transport protocols 655 * that are too eager in sending delete messages. 656 */ 657 if (gethrestime_sec() < 658 ire->ire_create_time + ip_ignore_delete_time) { 659 ire_refrele(ire); 660 return (EINVAL); 661 } 662 /* 663 * Now we have a potentially dead cache entry. We need 664 * to remove it. 665 * If this cache entry is generated from a 666 * default route (i.e., ire_cmask == 0), 667 * search the default list and mark it dead and some 668 * background process will try to activate it. 669 */ 670 if ((ire->ire_gateway_addr != 0) && (ire->ire_cmask == 0)) { 671 /* 672 * Make sure that we pick a different 673 * IRE_DEFAULT next time. 674 */ 675 ire_t *gw_ire; 676 irb_t *irb = NULL; 677 uint_t match_flags; 678 679 match_flags = (MATCH_IRE_DEFAULT | MATCH_IRE_RJ_BHOLE); 680 681 gire = ire_ftable_lookup(ire->ire_addr, 682 ire->ire_cmask, 0, 0, 683 ire->ire_ipif, NULL, zoneid, 0, NULL, match_flags); 684 685 ip3dbg(("ire_ftable_lookup() returned gire %p\n", 686 (void *)gire)); 687 688 if (gire != NULL) { 689 irb = gire->ire_bucket; 690 691 /* 692 * We grab it as writer just to serialize 693 * multiple threads trying to bump up 694 * irb_rr_origin 695 */ 696 rw_enter(&irb->irb_lock, RW_WRITER); 697 if ((gw_ire = irb->irb_rr_origin) == NULL) { 698 rw_exit(&irb->irb_lock); 699 goto done; 700 } 701 702 703 /* Skip past the potentially bad gateway */ 704 if (ire->ire_gateway_addr == 705 gw_ire->ire_gateway_addr) 706 irb->irb_rr_origin = gw_ire->ire_next; 707 708 rw_exit(&irb->irb_lock); 709 } 710 } 711 done: 712 if (gire != NULL) 713 IRE_REFRELE(gire); 714 /* report the bad route to routing sockets */ 715 ip_rts_change(RTM_LOSING, ire->ire_addr, ire->ire_gateway_addr, 716 ire->ire_mask, ire->ire_src_addr, 0, 0, 0, 717 (RTA_DST | RTA_GATEWAY | RTA_NETMASK | RTA_IFA)); 718 routing_sock_info = B_TRUE; 719 720 /* 721 * TCP is really telling us to start over completely, and it 722 * expects that we'll resend the ARP query. Tell ARP to 723 * discard the entry, if this is a local destination. 724 */ 725 ill = ire->ire_stq->q_ptr; 726 if (ire->ire_gateway_addr == 0 && 727 (arp_mp = ill_ared_alloc(ill, addr)) != NULL) { 728 putnext(ill->ill_rq, arp_mp); 729 } 730 731 ire_delete(ire); 732 ire_refrele(ire); 733 } 734 /* Also look for an IRE_HOST_REDIRECT and remove it if present */ 735 ire = ire_route_lookup(addr, 0, 0, IRE_HOST_REDIRECT, NULL, NULL, 736 ALL_ZONES, NULL, MATCH_IRE_TYPE); 737 738 /* Nail it. */ 739 if (ire) { 740 if (!routing_sock_info) { 741 ip_rts_change(RTM_LOSING, ire->ire_addr, 742 ire->ire_gateway_addr, ire->ire_mask, 743 ire->ire_src_addr, 0, 0, 0, 744 (RTA_DST | RTA_GATEWAY | RTA_NETMASK | RTA_IFA)); 745 } 746 ire_delete(ire); 747 ire_refrele(ire); 748 } 749 return (0); 750 } 751 752 /* 753 * Named Dispatch routine to produce a formatted report on all IREs. 754 * This report is accessed by using the ndd utility to "get" ND variable 755 * "ipv4_ire_status". 756 */ 757 /* ARGSUSED */ 758 int 759 ip_ire_report(queue_t *q, mblk_t *mp, caddr_t arg, cred_t *ioc_cr) 760 { 761 zoneid_t zoneid; 762 763 (void) mi_mpprintf(mp, 764 "IRE " MI_COL_HDRPAD_STR 765 /* 01234567[89ABCDEF] */ 766 "rfq " MI_COL_HDRPAD_STR 767 /* 01234567[89ABCDEF] */ 768 "stq " MI_COL_HDRPAD_STR 769 /* 01234567[89ABCDEF] */ 770 " zone " 771 /* 12345 */ 772 "addr mask " 773 /* 123.123.123.123 123.123.123.123 */ 774 "src gateway mxfrg rtt rtt_sd ssthresh ref " 775 /* 123.123.123.123 123.123.123.123 12345 12345 123456 12345678 123 */ 776 "rtomax tstamp_ok wscale_ok ecn_ok pmtud_ok sack sendpipe " 777 /* 123456 123456789 123456789 123456 12345678 1234 12345678 */ 778 "recvpipe in/out/forward type"); 779 /* 12345678 in/out/forward xxxxxxxxxx */ 780 781 /* 782 * Because of the ndd constraint, at most we can have 64K buffer 783 * to put in all IRE info. So to be more efficient, just 784 * allocate a 64K buffer here, assuming we need that large buffer. 785 * This should be OK as only root can do ndd /dev/ip. 786 */ 787 if ((mp->b_cont = allocb(ND_MAX_BUF_LEN, BPRI_HI)) == NULL) { 788 /* The following may work even if we cannot get a large buf. */ 789 (void) mi_mpprintf(mp, "<< Out of buffer >>\n"); 790 return (0); 791 } 792 793 zoneid = Q_TO_CONN(q)->conn_zoneid; 794 if (zoneid == GLOBAL_ZONEID) 795 zoneid = ALL_ZONES; 796 797 ire_walk_v4(ire_report_ftable, mp->b_cont, zoneid); 798 ire_walk_v4(ire_report_ctable, mp->b_cont, zoneid); 799 800 return (0); 801 } 802 803 804 /* ire_walk routine invoked for ip_ire_report for each cached IRE. */ 805 static void 806 ire_report_ctable(ire_t *ire, char *mp) 807 { 808 char buf1[16]; 809 char buf2[16]; 810 char buf3[16]; 811 char buf4[16]; 812 uint_t fo_pkt_count; 813 uint_t ib_pkt_count; 814 int ref; 815 uint_t print_len, buf_len; 816 817 if ((ire->ire_type & IRE_CACHETABLE) == 0) 818 return; 819 buf_len = ((mblk_t *)mp)->b_datap->db_lim - ((mblk_t *)mp)->b_wptr; 820 if (buf_len <= 0) 821 return; 822 823 /* Number of active references of this ire */ 824 ref = ire->ire_refcnt; 825 /* "inbound" to a non local address is a forward */ 826 ib_pkt_count = ire->ire_ib_pkt_count; 827 fo_pkt_count = 0; 828 if (!(ire->ire_type & (IRE_LOCAL|IRE_BROADCAST))) { 829 fo_pkt_count = ib_pkt_count; 830 ib_pkt_count = 0; 831 } 832 print_len = snprintf((char *)((mblk_t *)mp)->b_wptr, buf_len, 833 MI_COL_PTRFMT_STR MI_COL_PTRFMT_STR MI_COL_PTRFMT_STR "%5d " 834 "%s %s %s %s %05d %05ld %06ld %08d %03d %06d %09d %09d %06d %08d " 835 "%04d %08d %08d %d/%d/%d %s\n", 836 (void *)ire, (void *)ire->ire_rfq, (void *)ire->ire_stq, 837 (int)ire->ire_zoneid, 838 ip_dot_addr(ire->ire_addr, buf1), ip_dot_addr(ire->ire_mask, buf2), 839 ip_dot_addr(ire->ire_src_addr, buf3), 840 ip_dot_addr(ire->ire_gateway_addr, buf4), 841 ire->ire_max_frag, ire->ire_uinfo.iulp_rtt, 842 ire->ire_uinfo.iulp_rtt_sd, ire->ire_uinfo.iulp_ssthresh, ref, 843 ire->ire_uinfo.iulp_rtomax, 844 (ire->ire_uinfo.iulp_tstamp_ok ? 1: 0), 845 (ire->ire_uinfo.iulp_wscale_ok ? 1: 0), 846 (ire->ire_uinfo.iulp_ecn_ok ? 1: 0), 847 (ire->ire_uinfo.iulp_pmtud_ok ? 1: 0), 848 ire->ire_uinfo.iulp_sack, 849 ire->ire_uinfo.iulp_spipe, ire->ire_uinfo.iulp_rpipe, 850 ib_pkt_count, ire->ire_ob_pkt_count, fo_pkt_count, 851 ip_nv_lookup(ire_nv_tbl, (int)ire->ire_type)); 852 if (print_len < buf_len) { 853 ((mblk_t *)mp)->b_wptr += print_len; 854 } else { 855 ((mblk_t *)mp)->b_wptr += buf_len; 856 } 857 } 858 859 /* ARGSUSED */ 860 int 861 ip_ire_report_mrtun(queue_t *q, mblk_t *mp, caddr_t arg, cred_t *ioc_cr) 862 { 863 (void) mi_mpprintf(mp, 864 "IRE " MI_COL_HDRPAD_STR 865 /* 01234567[89ABCDEF] */ 866 "stq " MI_COL_HDRPAD_STR 867 /* 01234567[89ABCDEF] */ 868 "in_ill " MI_COL_HDRPAD_STR 869 /* 01234567[89ABCDEF] */ 870 "in_src_addr " 871 /* 123.123.123.123 */ 872 "max_frag " 873 /* 12345 */ 874 "ref "); 875 /* 123 */ 876 877 ire_walk_ill_mrtun(0, 0, ire_report_mrtun_table, mp, NULL); 878 return (0); 879 } 880 881 /* mrtun report table - supports ipv4_mrtun_ire_status ndd variable */ 882 883 static void 884 ire_report_mrtun_table(ire_t *ire, char *mp) 885 { 886 char buf1[INET_ADDRSTRLEN]; 887 int ref; 888 889 /* Number of active references of this ire */ 890 ref = ire->ire_refcnt; 891 ASSERT(ire->ire_type == IRE_MIPRTUN); 892 (void) mi_mpprintf((mblk_t *)mp, 893 MI_COL_PTRFMT_STR MI_COL_PTRFMT_STR MI_COL_PTRFMT_STR 894 "%s %05d %03d", 895 (void *)ire, (void *)ire->ire_stq, 896 (void *)ire->ire_in_ill, 897 ip_dot_addr(ire->ire_in_src_addr, buf1), 898 ire->ire_max_frag, ref); 899 } 900 901 /* 902 * Dispatch routine to format ires in interface based routine 903 */ 904 /* ARGSUSED */ 905 int 906 ip_ire_report_srcif(queue_t *q, mblk_t *mp, caddr_t arg, cred_t *ioc_cr) 907 { 908 909 /* Report all interface based ires */ 910 911 (void) mi_mpprintf(mp, 912 "IRE " MI_COL_HDRPAD_STR 913 /* 01234567[89ABCDEF] */ 914 "stq " MI_COL_HDRPAD_STR 915 /* 01234567[89ABCDEF] */ 916 "in_ill " MI_COL_HDRPAD_STR 917 /* 01234567[89ABCDEF] */ 918 "addr " 919 /* 123.123.123.123 */ 920 "gateway " 921 /* 123.123.123.123 */ 922 "max_frag " 923 /* 12345 */ 924 "ref " 925 /* 123 */ 926 "type " 927 /* ABCDEFGH */ 928 "in/out/forward"); 929 ire_walk_srcif_table_v4(ire_report_srcif_table, mp); 930 return (0); 931 } 932 933 /* Reports the interface table ires */ 934 static void 935 ire_report_srcif_table(ire_t *ire, char *mp) 936 { 937 char buf1[INET_ADDRSTRLEN]; 938 char buf2[INET_ADDRSTRLEN]; 939 int ref; 940 941 ref = ire->ire_refcnt; 942 (void) mi_mpprintf((mblk_t *)mp, 943 MI_COL_PTRFMT_STR MI_COL_PTRFMT_STR MI_COL_PTRFMT_STR 944 "%s %s %05d %03d %s %d", 945 (void *)ire, (void *)ire->ire_stq, 946 (void *)ire->ire_in_ill, 947 ip_dot_addr(ire->ire_addr, buf1), 948 ip_dot_addr(ire->ire_gateway_addr, buf2), 949 ire->ire_max_frag, ref, 950 ip_nv_lookup(ire_nv_tbl, (int)ire->ire_type), 951 ire->ire_ib_pkt_count); 952 953 } 954 /* 955 * ip_ire_req is called by ip_wput when an IRE_DB_REQ_TYPE message is handed 956 * down from the Upper Level Protocol to request a copy of the IRE (to check 957 * its type or to extract information like round-trip time estimates or the 958 * MTU.) 959 * The address is assumed to be in the ire_addr field. If no IRE is found 960 * an IRE is returned with ire_type being zero. 961 * Note that the upper lavel protocol has to check for broadcast 962 * (IRE_BROADCAST) and multicast (CLASSD(addr)). 963 * If there is a b_cont the resulting IRE_DB_TYPE mblk is placed at the 964 * end of the returned message. 965 * 966 * TCP sends down a message of this type with a connection request packet 967 * chained on. UDP and ICMP send it down to verify that a route exists for 968 * the destination address when they get connected. 969 */ 970 void 971 ip_ire_req(queue_t *q, mblk_t *mp) 972 { 973 ire_t *inire; 974 ire_t *ire; 975 mblk_t *mp1; 976 ire_t *sire = NULL; 977 zoneid_t zoneid = Q_TO_CONN(q)->conn_zoneid; 978 979 if ((mp->b_wptr - mp->b_rptr) < sizeof (ire_t) || 980 !OK_32PTR(mp->b_rptr)) { 981 freemsg(mp); 982 return; 983 } 984 inire = (ire_t *)mp->b_rptr; 985 /* 986 * Got it, now take our best shot at an IRE. 987 */ 988 if (inire->ire_ipversion == IPV6_VERSION) { 989 ire = ire_route_lookup_v6(&inire->ire_addr_v6, 0, 0, 0, 990 NULL, &sire, zoneid, NULL, 991 (MATCH_IRE_RECURSIVE | MATCH_IRE_DEFAULT)); 992 } else { 993 ASSERT(inire->ire_ipversion == IPV4_VERSION); 994 ire = ire_route_lookup(inire->ire_addr, 0, 0, 0, 995 NULL, &sire, zoneid, NULL, 996 (MATCH_IRE_RECURSIVE | MATCH_IRE_DEFAULT)); 997 } 998 999 /* 1000 * We prevent returning IRES with source address INADDR_ANY 1001 * as these were temporarily created for sending packets 1002 * from endpoints that have conn_unspec_src set. 1003 */ 1004 if (ire == NULL || 1005 (ire->ire_ipversion == IPV4_VERSION && 1006 ire->ire_src_addr == INADDR_ANY) || 1007 (ire->ire_ipversion == IPV6_VERSION && 1008 IN6_IS_ADDR_UNSPECIFIED(&ire->ire_src_addr_v6))) { 1009 inire->ire_type = 0; 1010 } else { 1011 bcopy(ire, inire, sizeof (ire_t)); 1012 /* Copy the route metrics from the parent. */ 1013 if (sire != NULL) { 1014 bcopy(&(sire->ire_uinfo), &(inire->ire_uinfo), 1015 sizeof (iulp_t)); 1016 } 1017 1018 /* 1019 * As we don't lookup global policy here, we may not 1020 * pass the right size if per-socket policy is not 1021 * present. For these cases, path mtu discovery will 1022 * do the right thing. 1023 */ 1024 inire->ire_ipsec_overhead = conn_ipsec_length(Q_TO_CONN(q)); 1025 1026 /* Pass the latest setting of the ip_path_mtu_discovery */ 1027 inire->ire_frag_flag |= (ip_path_mtu_discovery) ? IPH_DF : 0; 1028 } 1029 if (ire != NULL) 1030 ire_refrele(ire); 1031 if (sire != NULL) 1032 ire_refrele(sire); 1033 mp->b_wptr = &mp->b_rptr[sizeof (ire_t)]; 1034 mp->b_datap->db_type = IRE_DB_TYPE; 1035 1036 /* Put the IRE_DB_TYPE mblk last in the chain */ 1037 mp1 = mp->b_cont; 1038 if (mp1 != NULL) { 1039 mp->b_cont = NULL; 1040 linkb(mp1, mp); 1041 mp = mp1; 1042 } 1043 qreply(q, mp); 1044 } 1045 1046 /* 1047 * Send a packet using the specified IRE. 1048 * If ire_src_addr_v6 is all zero then discard the IRE after 1049 * the packet has been sent. 1050 */ 1051 static void 1052 ire_send(queue_t *q, mblk_t *pkt, ire_t *ire) 1053 { 1054 mblk_t *ipsec_mp; 1055 boolean_t is_secure; 1056 uint_t ifindex; 1057 ill_t *ill; 1058 zoneid_t zoneid = ire->ire_zoneid; 1059 1060 ASSERT(ire->ire_ipversion == IPV4_VERSION); 1061 ASSERT(!(ire->ire_type & IRE_LOCAL)); /* Has different ire_zoneid */ 1062 ipsec_mp = pkt; 1063 is_secure = (pkt->b_datap->db_type == M_CTL); 1064 if (is_secure) { 1065 ipsec_out_t *io; 1066 1067 pkt = pkt->b_cont; 1068 io = (ipsec_out_t *)ipsec_mp->b_rptr; 1069 if (io->ipsec_out_type == IPSEC_OUT) 1070 zoneid = io->ipsec_out_zoneid; 1071 } 1072 1073 /* If the packet originated externally then */ 1074 if (pkt->b_prev) { 1075 ire_refrele(ire); 1076 /* 1077 * Extract the ifindex from b_prev (set in ip_rput_noire). 1078 * Look up interface to see if it still exists (it could have 1079 * been unplumbed by the time the reply came back from ARP) 1080 */ 1081 ifindex = (uint_t)(uintptr_t)pkt->b_prev; 1082 ill = ill_lookup_on_ifindex(ifindex, B_FALSE, 1083 NULL, NULL, NULL, NULL); 1084 if (ill == NULL) { 1085 pkt->b_prev = NULL; 1086 pkt->b_next = NULL; 1087 freemsg(ipsec_mp); 1088 return; 1089 } 1090 q = ill->ill_rq; 1091 pkt->b_prev = NULL; 1092 /* 1093 * This packet has not gone through IPSEC processing 1094 * and hence we should not have any IPSEC message 1095 * prepended. 1096 */ 1097 ASSERT(ipsec_mp == pkt); 1098 put(q, pkt); 1099 ill_refrele(ill); 1100 } else if (pkt->b_next) { 1101 /* Packets from multicast router */ 1102 pkt->b_next = NULL; 1103 /* 1104 * We never get the IPSEC_OUT while forwarding the 1105 * packet for multicast router. 1106 */ 1107 ASSERT(ipsec_mp == pkt); 1108 ip_rput_forward(ire, (ipha_t *)pkt->b_rptr, ipsec_mp, NULL); 1109 ire_refrele(ire); 1110 } else { 1111 /* Locally originated packets */ 1112 boolean_t is_inaddr_any; 1113 ipha_t *ipha = (ipha_t *)pkt->b_rptr; 1114 1115 /* 1116 * We need to do an ire_delete below for which 1117 * we need to make sure that the IRE will be 1118 * around even after calling ip_wput_ire - 1119 * which does ire_refrele. Otherwise somebody 1120 * could potentially delete this ire and hence 1121 * free this ire and we will be calling ire_delete 1122 * on a freed ire below. 1123 */ 1124 is_inaddr_any = (ire->ire_src_addr == INADDR_ANY); 1125 if (is_inaddr_any) { 1126 IRE_REFHOLD(ire); 1127 } 1128 /* 1129 * If we were resolving a router we can not use the 1130 * routers IRE for sending the packet (since it would 1131 * violate the uniqness of the IP idents) thus we 1132 * make another pass through ip_wput to create the IRE_CACHE 1133 * for the destination. 1134 * When IRE_MARK_NOADD is set, ire_add() is not called. 1135 * Thus ip_wput() will never find a ire and result in an 1136 * infinite loop. Thus we check whether IRE_MARK_NOADD is 1137 * is set. This also implies that IRE_MARK_NOADD can only be 1138 * used to send packets to directly connected hosts. 1139 */ 1140 if (ipha->ipha_dst != ire->ire_addr && 1141 !(ire->ire_marks & IRE_MARK_NOADD)) { 1142 ire_refrele(ire); /* Held in ire_add */ 1143 if (CONN_Q(q)) { 1144 (void) ip_output(Q_TO_CONN(q), ipsec_mp, q, 1145 IRE_SEND); 1146 } else { 1147 (void) ip_output((void *)(uintptr_t)zoneid, 1148 ipsec_mp, q, IRE_SEND); 1149 } 1150 } else { 1151 if (is_secure) { 1152 ipsec_out_t *oi; 1153 ipha_t *ipha; 1154 1155 oi = (ipsec_out_t *)ipsec_mp->b_rptr; 1156 ipha = (ipha_t *)ipsec_mp->b_cont->b_rptr; 1157 if (oi->ipsec_out_proc_begin) { 1158 /* 1159 * This is the case where 1160 * ip_wput_ipsec_out could not find 1161 * the IRE and recreated a new one. 1162 * As ip_wput_ipsec_out does ire 1163 * lookups, ire_refrele for the extra 1164 * bump in ire_add. 1165 */ 1166 ire_refrele(ire); 1167 ip_wput_ipsec_out(q, ipsec_mp, ipha, 1168 NULL, NULL); 1169 } else { 1170 /* 1171 * IRE_REFRELE will be done in 1172 * ip_wput_ire. 1173 */ 1174 ip_wput_ire(q, ipsec_mp, ire, NULL, 1175 IRE_SEND, zoneid); 1176 } 1177 } else { 1178 /* 1179 * IRE_REFRELE will be done in ip_wput_ire. 1180 */ 1181 ip_wput_ire(q, ipsec_mp, ire, NULL, 1182 IRE_SEND, zoneid); 1183 } 1184 } 1185 /* 1186 * Special code to support sending a single packet with 1187 * conn_unspec_src using an IRE which has no source address. 1188 * The IRE is deleted here after sending the packet to avoid 1189 * having other code trip on it. But before we delete the 1190 * ire, somebody could have looked up this ire. 1191 * We prevent returning/using this IRE by the upper layers 1192 * by making checks to NULL source address in other places 1193 * like e.g ip_ire_append, ip_ire_req and ip_bind_connected. 1194 * Though, this does not completely prevent other threads 1195 * from using this ire, this should not cause any problems. 1196 * 1197 * NOTE : We use is_inaddr_any instead of using ire_src_addr 1198 * because for the normal case i.e !is_inaddr_any, ire_refrele 1199 * above could have potentially freed the ire. 1200 */ 1201 if (is_inaddr_any) { 1202 /* 1203 * If this IRE has been deleted by another thread, then 1204 * ire_bucket won't be NULL, but ire_ptpn will be NULL. 1205 * Thus, ire_delete will do nothing. This check 1206 * guards against calling ire_delete when the IRE was 1207 * never inserted in the table, which is handled by 1208 * ire_delete as dropping another reference. 1209 */ 1210 if (ire->ire_bucket != NULL) { 1211 ip1dbg(("ire_send: delete IRE\n")); 1212 ire_delete(ire); 1213 } 1214 ire_refrele(ire); /* Held above */ 1215 } 1216 } 1217 } 1218 1219 /* 1220 * Send a packet using the specified IRE. 1221 * If ire_src_addr_v6 is all zero then discard the IRE after 1222 * the packet has been sent. 1223 */ 1224 static void 1225 ire_send_v6(queue_t *q, mblk_t *pkt, ire_t *ire) 1226 { 1227 mblk_t *ipsec_mp; 1228 boolean_t secure; 1229 uint_t ifindex; 1230 zoneid_t zoneid = ire->ire_zoneid; 1231 1232 ASSERT(ire->ire_ipversion == IPV6_VERSION); 1233 ASSERT(!(ire->ire_type & IRE_LOCAL)); /* Has different ire_zoneid */ 1234 if (pkt->b_datap->db_type == M_CTL) { 1235 ipsec_out_t *io; 1236 1237 ipsec_mp = pkt; 1238 pkt = pkt->b_cont; 1239 secure = B_TRUE; 1240 io = (ipsec_out_t *)ipsec_mp->b_rptr; 1241 if (io->ipsec_out_type == IPSEC_OUT) 1242 zoneid = io->ipsec_out_zoneid; 1243 } else { 1244 ipsec_mp = pkt; 1245 secure = B_FALSE; 1246 } 1247 1248 /* If the packet originated externally then */ 1249 if (pkt->b_prev) { 1250 ill_t *ill; 1251 /* 1252 * Extract the ifindex from b_prev (set in ip_rput_data_v6). 1253 * Look up interface to see if it still exists (it could have 1254 * been unplumbed by the time the reply came back from the 1255 * resolver). 1256 */ 1257 ifindex = (uint_t)(uintptr_t)pkt->b_prev; 1258 ill = ill_lookup_on_ifindex(ifindex, B_TRUE, 1259 NULL, NULL, NULL, NULL); 1260 if (ill == NULL) { 1261 pkt->b_prev = NULL; 1262 pkt->b_next = NULL; 1263 freemsg(ipsec_mp); 1264 ire_refrele(ire); /* Held in ire_add */ 1265 return; 1266 } 1267 q = ill->ill_rq; 1268 pkt->b_prev = NULL; 1269 /* 1270 * This packet has not gone through IPSEC processing 1271 * and hence we should not have any IPSEC message 1272 * prepended. 1273 */ 1274 ASSERT(ipsec_mp == pkt); 1275 put(q, pkt); 1276 ill_refrele(ill); 1277 } else if (pkt->b_next) { 1278 /* Packets from multicast router */ 1279 pkt->b_next = NULL; 1280 /* 1281 * We never get the IPSEC_OUT while forwarding the 1282 * packet for multicast router. 1283 */ 1284 ASSERT(ipsec_mp == pkt); 1285 /* 1286 * XXX TODO IPv6. 1287 */ 1288 freemsg(pkt); 1289 #ifdef XXX 1290 ip_rput_forward(ire, (ipha_t *)pkt->b_rptr, pkt, NULL); 1291 #endif 1292 } else { 1293 if (secure) { 1294 ipsec_out_t *oi; 1295 ip6_t *ip6h; 1296 1297 oi = (ipsec_out_t *)ipsec_mp->b_rptr; 1298 ip6h = (ip6_t *)ipsec_mp->b_cont->b_rptr; 1299 if (oi->ipsec_out_proc_begin) { 1300 /* 1301 * This is the case where 1302 * ip_wput_ipsec_out could not find 1303 * the IRE and recreated a new one. 1304 */ 1305 ip_wput_ipsec_out_v6(q, ipsec_mp, ip6h, 1306 NULL, NULL); 1307 } else { 1308 if (CONN_Q(q)) { 1309 (void) ip_output_v6(Q_TO_CONN(q), 1310 ipsec_mp, q, IRE_SEND); 1311 } else { 1312 (void) ip_output_v6( 1313 (void *)(uintptr_t)zoneid, 1314 ipsec_mp, q, IRE_SEND); 1315 } 1316 } 1317 } else { 1318 /* 1319 * Send packets through ip_output_v6 so that any 1320 * ip6_info header can be processed again. 1321 */ 1322 if (CONN_Q(q)) { 1323 (void) ip_output_v6(Q_TO_CONN(q), ipsec_mp, q, 1324 IRE_SEND); 1325 } else { 1326 (void) ip_output_v6((void *)(uintptr_t)zoneid, 1327 ipsec_mp, q, IRE_SEND); 1328 } 1329 } 1330 /* 1331 * Special code to support sending a single packet with 1332 * conn_unspec_src using an IRE which has no source address. 1333 * The IRE is deleted here after sending the packet to avoid 1334 * having other code trip on it. But before we delete the 1335 * ire, somebody could have looked up this ire. 1336 * We prevent returning/using this IRE by the upper layers 1337 * by making checks to NULL source address in other places 1338 * like e.g ip_ire_append_v6, ip_ire_req and 1339 * ip_bind_connected_v6. Though, this does not completely 1340 * prevent other threads from using this ire, this should 1341 * not cause any problems. 1342 */ 1343 if (IN6_IS_ADDR_UNSPECIFIED(&ire->ire_src_addr_v6)) { 1344 ip1dbg(("ire_send_v6: delete IRE\n")); 1345 ire_delete(ire); 1346 } 1347 } 1348 ire_refrele(ire); /* Held in ire_add */ 1349 } 1350 1351 /* 1352 * Make sure that IRE bucket does not get too long. 1353 * This can cause lock up because ire_cache_lookup() 1354 * may take "forever" to finish. 1355 * 1356 * We just remove cnt IREs each time. This means that 1357 * the bucket length will stay approximately constant, 1358 * depending on cnt. This should be enough to defend 1359 * against DoS attack based on creating temporary IREs 1360 * (for forwarding and non-TCP traffic). 1361 * 1362 * Note that new IRE is normally added at the tail of the 1363 * bucket. This means that we are removing the "oldest" 1364 * temporary IRE added. Only if there are IREs with 1365 * the same ire_addr, do we not add it at the tail. Refer 1366 * to ire_add_v*(). It should be OK for our purpose. 1367 * 1368 * For non-temporary cached IREs, we make sure that they 1369 * have not been used for some time (defined below), they 1370 * are non-local destinations, and there is no one using 1371 * them at the moment (refcnt == 1). 1372 * 1373 * The above means that the IRE bucket length may become 1374 * very long, consisting of mostly non-temporary IREs. 1375 * This can happen when the hash function does a bad job 1376 * so that most TCP connections cluster to a specific bucket. 1377 * This "hopefully" should never happen. It can also 1378 * happen if most TCP connections have very long lives. 1379 * Even with the minimal hash table size of 256, there 1380 * has to be a lot of such connections to make the bucket 1381 * length unreasonably long. This should probably not 1382 * happen either. The third can when this can happen is 1383 * when the machine is under attack, such as SYN flooding. 1384 * TCP should already have the proper mechanism to protect 1385 * that. So we should be safe. 1386 * 1387 * This function is called by ire_add_then_send() after 1388 * a new IRE is added and the packet is sent. 1389 * 1390 * The idle cutoff interval is set to 60s. It can be 1391 * changed using /etc/system. 1392 */ 1393 uint32_t ire_idle_cutoff_interval = 60000; 1394 1395 static void 1396 ire_cache_cleanup(irb_t *irb, uint32_t threshold, int cnt) 1397 { 1398 ire_t *ire; 1399 int tmp_cnt = cnt; 1400 clock_t cut_off = drv_usectohz(ire_idle_cutoff_interval * 1000); 1401 1402 /* 1403 * irb is NULL if the IRE is not added to the hash. This 1404 * happens when IRE_MARK_NOADD is set in ire_add_then_send() 1405 * and when ires are returned from ire_update_srcif_v4() routine. 1406 */ 1407 if (irb == NULL) 1408 return; 1409 1410 IRB_REFHOLD(irb); 1411 if (irb->irb_tmp_ire_cnt > threshold) { 1412 for (ire = irb->irb_ire; ire != NULL && tmp_cnt > 0; 1413 ire = ire->ire_next) { 1414 if (ire->ire_marks & IRE_MARK_CONDEMNED) 1415 continue; 1416 if (ire->ire_marks & IRE_MARK_TEMPORARY) { 1417 ASSERT(ire->ire_type == IRE_CACHE); 1418 ire_delete(ire); 1419 tmp_cnt--; 1420 } 1421 } 1422 } 1423 if (irb->irb_ire_cnt - irb->irb_tmp_ire_cnt > threshold) { 1424 for (ire = irb->irb_ire; ire != NULL && cnt > 0; 1425 ire = ire->ire_next) { 1426 if (ire->ire_marks & IRE_MARK_CONDEMNED || 1427 ire->ire_gateway_addr == 0) { 1428 continue; 1429 } 1430 if ((ire->ire_type == IRE_CACHE) && 1431 (lbolt - ire->ire_last_used_time > cut_off) && 1432 (ire->ire_refcnt == 1)) { 1433 ire_delete(ire); 1434 cnt--; 1435 } 1436 } 1437 } 1438 IRB_REFRELE(irb); 1439 } 1440 1441 /* 1442 * ire_add_then_send is called when a new IRE has been created in order to 1443 * route an outgoing packet. Typically, it is called from ip_wput when 1444 * a response comes back down from a resolver. We add the IRE, and then 1445 * possibly run the packet through ip_wput or ip_rput, as appropriate. 1446 * However, we do not add the newly created IRE in the cache when 1447 * IRE_MARK_NOADD is set in the IRE. IRE_MARK_NOADD is set at 1448 * ip_newroute_ipif(). The ires with IRE_MARK_NOADD and ires returned 1449 * by ire_update_srcif_v4() are ire_refrele'd by ip_wput_ire() and get 1450 * deleted. 1451 * Multirouting support: the packet is silently discarded when the new IRE 1452 * holds the RTF_MULTIRT flag, but is not the first IRE to be added with the 1453 * RTF_MULTIRT flag for the same destination address. 1454 * In this case, we just want to register this additional ire without 1455 * sending the packet, as it has already been replicated through 1456 * existing multirt routes in ip_wput(). 1457 */ 1458 void 1459 ire_add_then_send(queue_t *q, ire_t *ire, mblk_t *mp) 1460 { 1461 irb_t *irb; 1462 boolean_t drop = B_FALSE; 1463 /* LINTED : set but not used in function */ 1464 boolean_t mctl_present; 1465 mblk_t *first_mp = NULL; 1466 mblk_t *save_mp = NULL; 1467 ire_t *dst_ire; 1468 ipha_t *ipha; 1469 ip6_t *ip6h; 1470 1471 if (mp != NULL) { 1472 /* 1473 * We first have to retrieve the destination address carried 1474 * by the packet. 1475 * We can't rely on ire as it can be related to a gateway. 1476 * The destination address will help in determining if 1477 * other RTF_MULTIRT ires are already registered. 1478 * 1479 * We first need to know where we are going : v4 or V6. 1480 * the ire version is enough, as there is no risk that 1481 * we resolve an IPv6 address with an IPv4 ire 1482 * or vice versa. 1483 */ 1484 if (ire->ire_ipversion == IPV4_VERSION) { 1485 EXTRACT_PKT_MP(mp, first_mp, mctl_present); 1486 ipha = (ipha_t *)mp->b_rptr; 1487 save_mp = mp; 1488 mp = first_mp; 1489 1490 dst_ire = ire_cache_lookup(ipha->ipha_dst, 1491 ire->ire_zoneid, MBLK_GETLABEL(mp)); 1492 } else { 1493 ASSERT(ire->ire_ipversion == IPV6_VERSION); 1494 /* 1495 * Get a pointer to the beginning of the IPv6 header. 1496 * Ignore leading IPsec control mblks. 1497 */ 1498 first_mp = mp; 1499 if (mp->b_datap->db_type == M_CTL) { 1500 mp = mp->b_cont; 1501 } 1502 ip6h = (ip6_t *)mp->b_rptr; 1503 save_mp = mp; 1504 mp = first_mp; 1505 dst_ire = ire_cache_lookup_v6(&ip6h->ip6_dst, 1506 ire->ire_zoneid, MBLK_GETLABEL(mp)); 1507 } 1508 if (dst_ire != NULL) { 1509 if (dst_ire->ire_flags & RTF_MULTIRT) { 1510 /* 1511 * At least one resolved multirt route 1512 * already exists for the destination, 1513 * don't sent this packet: either drop it 1514 * or complete the pending resolution, 1515 * depending on the ire. 1516 */ 1517 drop = B_TRUE; 1518 } 1519 ip1dbg(("ire_add_then_send: dst_ire %p " 1520 "[dst %08x, gw %08x], drop %d\n", 1521 (void *)dst_ire, 1522 (dst_ire->ire_ipversion == IPV4_VERSION) ? \ 1523 ntohl(dst_ire->ire_addr) : \ 1524 ntohl(V4_PART_OF_V6(dst_ire->ire_addr_v6)), 1525 (dst_ire->ire_ipversion == IPV4_VERSION) ? \ 1526 ntohl(dst_ire->ire_gateway_addr) : \ 1527 ntohl(V4_PART_OF_V6( 1528 dst_ire->ire_gateway_addr_v6)), 1529 drop)); 1530 ire_refrele(dst_ire); 1531 } 1532 } 1533 1534 if (!(ire->ire_marks & IRE_MARK_NOADD)) { 1535 /* 1536 * Regular packets with cache bound ires and 1537 * the packets from ARP response for ires which 1538 * belong to the ire_srcif_v4 table, are here. 1539 */ 1540 if (ire->ire_in_ill == NULL) { 1541 /* Add the ire */ 1542 (void) ire_add(&ire, NULL, NULL, NULL, B_FALSE); 1543 } else { 1544 /* 1545 * This must be ARP response for ire in interface based 1546 * table. Note that we don't add them in cache table, 1547 * instead we update the existing table with dlureq_mp 1548 * information. The reverse tunnel ires do not come 1549 * here, as reverse tunnel is non-resolver interface. 1550 * XXX- another design alternative was to mark the 1551 * ires in interface based table with a special mark to 1552 * make absolutely sure that we operate in right ires. 1553 * This idea was not implemented as part of code review 1554 * suggestion, as ire_in_ill suffice to distinguish 1555 * between the regular ires and interface based 1556 * ires now and thus we save a bit in the ire_marks. 1557 */ 1558 ire = ire_update_srcif_v4(ire); 1559 } 1560 1561 if (ire == NULL) { 1562 mp->b_prev = NULL; 1563 mp->b_next = NULL; 1564 MULTIRT_DEBUG_UNTAG(mp); 1565 freemsg(mp); 1566 return; 1567 } 1568 if (mp == NULL) { 1569 ire_refrele(ire); /* Held in ire_add_v4/v6 */ 1570 return; 1571 } 1572 } 1573 if (drop) { 1574 /* 1575 * If we're adding an RTF_MULTIRT ire, the resolution 1576 * is over: we just drop the packet. 1577 */ 1578 if (ire->ire_flags & RTF_MULTIRT) { 1579 if (save_mp) { 1580 save_mp->b_prev = NULL; 1581 save_mp->b_next = NULL; 1582 } 1583 MULTIRT_DEBUG_UNTAG(mp); 1584 freemsg(mp); 1585 } else { 1586 /* 1587 * Otherwise, we're adding the ire to a gateway 1588 * for a multirt route. 1589 * Invoke ip_newroute() to complete the resolution 1590 * of the route. We will then come back here and 1591 * finally drop this packet in the above code. 1592 */ 1593 if (ire->ire_ipversion == IPV4_VERSION) { 1594 /* 1595 * TODO: in order for CGTP to work in non-global 1596 * zones, ip_newroute() must create the IRE 1597 * cache in the zone indicated by 1598 * ire->ire_zoneid. 1599 */ 1600 ip_newroute(q, mp, ipha->ipha_dst, 0, 1601 (CONN_Q(q) ? Q_TO_CONN(q) : NULL), 1602 ire->ire_zoneid); 1603 } else { 1604 ASSERT(ire->ire_ipversion == IPV6_VERSION); 1605 ip_newroute_v6(q, mp, &ip6h->ip6_dst, NULL, 1606 NULL, ire->ire_zoneid); 1607 } 1608 } 1609 1610 ire_refrele(ire); /* As done by ire_send(). */ 1611 return; 1612 } 1613 /* 1614 * Need to remember ire_bucket here as ire_send*() may delete 1615 * the ire so we cannot reference it after that. 1616 */ 1617 irb = ire->ire_bucket; 1618 if (ire->ire_ipversion == IPV6_VERSION) { 1619 ire_send_v6(q, mp, ire); 1620 /* 1621 * Clean up more than 1 IRE so that the clean up does not 1622 * need to be done every time when a new IRE is added and 1623 * the threshold is reached. 1624 */ 1625 ire_cache_cleanup(irb, ip6_ire_max_bucket_cnt, 2); 1626 } else { 1627 ire_send(q, mp, ire); 1628 ire_cache_cleanup(irb, ip_ire_max_bucket_cnt, 2); 1629 } 1630 } 1631 1632 /* 1633 * Initialize the ire that is specific to IPv4 part and call 1634 * ire_init_common to finish it. 1635 */ 1636 ire_t * 1637 ire_init(ire_t *ire, uchar_t *addr, uchar_t *mask, uchar_t *src_addr, 1638 uchar_t *gateway, uchar_t *in_src_addr, uint_t *max_fragp, mblk_t *fp_mp, 1639 queue_t *rfq, queue_t *stq, ushort_t type, mblk_t *dlureq_mp, ipif_t *ipif, 1640 ill_t *in_ill, ipaddr_t cmask, uint32_t phandle, uint32_t ihandle, 1641 uint32_t flags, const iulp_t *ulp_info, tsol_gc_t *gc, tsol_gcgrp_t *gcgrp) 1642 { 1643 /* 1644 * Reject IRE security attribute creation/initialization 1645 * if system is not running in Trusted mode. 1646 */ 1647 if ((gc != NULL || gcgrp != NULL) && !is_system_labeled()) 1648 return (NULL); 1649 1650 if (fp_mp != NULL) { 1651 /* 1652 * We can't dupb() here as multiple threads could be 1653 * calling dupb on the same mp which is incorrect. 1654 * First dupb() should be called only by one thread. 1655 */ 1656 fp_mp = copyb(fp_mp); 1657 if (fp_mp == NULL) 1658 return (NULL); 1659 } 1660 1661 if (dlureq_mp != NULL) { 1662 /* 1663 * We can't dupb() here as multiple threads could be 1664 * calling dupb on the same mp which is incorrect. 1665 * First dupb() should be called only by one thread. 1666 */ 1667 dlureq_mp = copyb(dlureq_mp); 1668 if (dlureq_mp == NULL) { 1669 if (fp_mp != NULL) 1670 freeb(fp_mp); 1671 return (NULL); 1672 } 1673 } 1674 1675 /* 1676 * Check that IRE_IF_RESOLVER and IRE_IF_NORESOLVER have a 1677 * dlureq_mp which is the ill_resolver_mp for IRE_IF_RESOLVER 1678 * and DL_UNITDATA_REQ for IRE_IF_NORESOLVER. 1679 */ 1680 if ((type & IRE_INTERFACE) && 1681 dlureq_mp == NULL) { 1682 ASSERT(fp_mp == NULL); 1683 ip0dbg(("ire_init: no dlureq_mp\n")); 1684 return (NULL); 1685 } 1686 1687 BUMP_IRE_STATS(ire_stats_v4, ire_stats_alloced); 1688 1689 if (addr != NULL) 1690 bcopy(addr, &ire->ire_addr, IP_ADDR_LEN); 1691 if (src_addr != NULL) 1692 bcopy(src_addr, &ire->ire_src_addr, IP_ADDR_LEN); 1693 if (mask != NULL) { 1694 bcopy(mask, &ire->ire_mask, IP_ADDR_LEN); 1695 ire->ire_masklen = ip_mask_to_plen(ire->ire_mask); 1696 } 1697 if (gateway != NULL) { 1698 bcopy(gateway, &ire->ire_gateway_addr, IP_ADDR_LEN); 1699 } 1700 if (in_src_addr != NULL) { 1701 bcopy(in_src_addr, &ire->ire_in_src_addr, IP_ADDR_LEN); 1702 } 1703 1704 if (type == IRE_CACHE) 1705 ire->ire_cmask = cmask; 1706 1707 /* ire_init_common will free the mblks upon encountering any failure */ 1708 if (!ire_init_common(ire, max_fragp, fp_mp, rfq, stq, type, dlureq_mp, 1709 ipif, in_ill, phandle, ihandle, flags, IPV4_VERSION, ulp_info, 1710 gc, gcgrp)) 1711 return (NULL); 1712 1713 return (ire); 1714 } 1715 1716 /* 1717 * Similar to ire_create except that it is called only when 1718 * we want to allocate ire as an mblk e.g. we have an external 1719 * resolver ARP. 1720 */ 1721 ire_t * 1722 ire_create_mp(uchar_t *addr, uchar_t *mask, uchar_t *src_addr, uchar_t *gateway, 1723 uchar_t *in_src_addr, uint_t max_frag, mblk_t *fp_mp, queue_t *rfq, 1724 queue_t *stq, ushort_t type, mblk_t *dlureq_mp, ipif_t *ipif, ill_t *in_ill, 1725 ipaddr_t cmask, uint32_t phandle, uint32_t ihandle, uint32_t flags, 1726 const iulp_t *ulp_info, tsol_gc_t *gc, tsol_gcgrp_t *gcgrp) 1727 { 1728 ire_t *ire, *buf; 1729 ire_t *ret_ire; 1730 mblk_t *mp; 1731 size_t bufsize; 1732 frtn_t *frtnp; 1733 ill_t *ill; 1734 1735 bufsize = sizeof (ire_t) + sizeof (frtn_t); 1736 buf = kmem_alloc(bufsize, KM_NOSLEEP); 1737 if (buf == NULL) { 1738 ip1dbg(("ire_create_mp: alloc failed\n")); 1739 return (NULL); 1740 } 1741 frtnp = (frtn_t *)(buf + 1); 1742 frtnp->free_arg = (caddr_t)buf; 1743 frtnp->free_func = ire_freemblk; 1744 1745 /* 1746 * Allocate the new IRE. The ire created will hold a ref on 1747 * an nce_t after ire_nce_init, and this ref must either be 1748 * (a) transferred to the ire_cache entry created when ire_add_v4 1749 * is called after successful arp resolution, or, 1750 * (b) released, when arp resolution fails 1751 * Case (b) is handled in ire_freemblk() which will be called 1752 * when mp is freed as a result of failed arp. 1753 */ 1754 mp = esballoc((unsigned char *)buf, bufsize, BPRI_MED, frtnp); 1755 if (mp == NULL) { 1756 ip1dbg(("ire_create_mp: alloc failed\n")); 1757 kmem_free(buf, bufsize); 1758 return (NULL); 1759 } 1760 ire = (ire_t *)mp->b_rptr; 1761 mp->b_wptr = (uchar_t *)&ire[1]; 1762 1763 /* Start clean. */ 1764 *ire = ire_null; 1765 ire->ire_mp = mp; 1766 mp->b_datap->db_type = IRE_DB_TYPE; 1767 ire->ire_marks |= IRE_MARK_UNCACHED; 1768 1769 ret_ire = ire_init(ire, addr, mask, src_addr, gateway, in_src_addr, 1770 NULL, fp_mp, rfq, stq, type, dlureq_mp, ipif, in_ill, cmask, 1771 phandle, ihandle, flags, ulp_info, gc, gcgrp); 1772 1773 if (ret_ire == NULL) { 1774 freeb(ire->ire_mp); 1775 return (NULL); 1776 } 1777 ill = ire_to_ill(ret_ire); 1778 ret_ire->ire_stq_ifindex = ill->ill_phyint->phyint_ifindex; 1779 ASSERT(ret_ire == ire); 1780 /* 1781 * ire_max_frag is normally zero here and is atomically set 1782 * under the irebucket lock in ire_add_v[46] except for the 1783 * case of IRE_MARK_NOADD. In that event the the ire_max_frag 1784 * is non-zero here. 1785 */ 1786 ire->ire_max_frag = max_frag; 1787 return (ire); 1788 } 1789 1790 /* 1791 * ire_create is called to allocate and initialize a new IRE. 1792 * 1793 * NOTE : This is called as writer sometimes though not required 1794 * by this function. 1795 */ 1796 ire_t * 1797 ire_create(uchar_t *addr, uchar_t *mask, uchar_t *src_addr, uchar_t *gateway, 1798 uchar_t *in_src_addr, uint_t *max_fragp, mblk_t *fp_mp, queue_t *rfq, 1799 queue_t *stq, ushort_t type, mblk_t *dlureq_mp, ipif_t *ipif, ill_t *in_ill, 1800 ipaddr_t cmask, uint32_t phandle, uint32_t ihandle, uint32_t flags, 1801 const iulp_t *ulp_info, tsol_gc_t *gc, tsol_gcgrp_t *gcgrp) 1802 { 1803 ire_t *ire; 1804 ire_t *ret_ire; 1805 1806 ire = kmem_cache_alloc(ire_cache, KM_NOSLEEP); 1807 if (ire == NULL) { 1808 ip1dbg(("ire_create: alloc failed\n")); 1809 return (NULL); 1810 } 1811 *ire = ire_null; 1812 1813 ret_ire = ire_init(ire, addr, mask, src_addr, gateway, in_src_addr, 1814 max_fragp, fp_mp, rfq, stq, type, dlureq_mp, ipif, in_ill, cmask, 1815 phandle, ihandle, flags, ulp_info, gc, gcgrp); 1816 1817 if (ret_ire == NULL) { 1818 kmem_cache_free(ire_cache, ire); 1819 return (NULL); 1820 } 1821 ASSERT(ret_ire == ire); 1822 return (ire); 1823 } 1824 1825 1826 /* 1827 * Common to IPv4 and IPv6 1828 */ 1829 boolean_t 1830 ire_init_common(ire_t *ire, uint_t *max_fragp, mblk_t *fp_mp, 1831 queue_t *rfq, queue_t *stq, ushort_t type, 1832 mblk_t *dlureq_mp, ipif_t *ipif, ill_t *in_ill, uint32_t phandle, 1833 uint32_t ihandle, uint32_t flags, uchar_t ipversion, 1834 const iulp_t *ulp_info, tsol_gc_t *gc, tsol_gcgrp_t *gcgrp) 1835 { 1836 ire->ire_max_fragp = max_fragp; 1837 ire->ire_frag_flag |= (ip_path_mtu_discovery) ? IPH_DF : 0; 1838 1839 ASSERT(fp_mp == NULL || fp_mp->b_datap->db_type == M_DATA); 1840 #ifdef DEBUG 1841 if (ipif != NULL) { 1842 if (ipif->ipif_isv6) 1843 ASSERT(ipversion == IPV6_VERSION); 1844 else 1845 ASSERT(ipversion == IPV4_VERSION); 1846 } 1847 #endif /* DEBUG */ 1848 1849 /* 1850 * Create/initialize IRE security attribute only in Trusted mode; 1851 * if the passed in gc/gcgrp is non-NULL, we expect that the caller 1852 * has held a reference to it and will release it when this routine 1853 * returns a failure, otherwise we own the reference. We do this 1854 * prior to initializing the rest IRE fields. 1855 * 1856 * Don't allocate ire_gw_secattr for the resolver case to prevent 1857 * memory leak (in case of external resolution failure). We'll 1858 * allocate it after a successful external resolution, in ire_add(). 1859 * Note that ire->ire_mp != NULL here means this ire is headed 1860 * to an external resolver. 1861 */ 1862 if (is_system_labeled()) { 1863 if ((type & (IRE_LOCAL | IRE_LOOPBACK | IRE_BROADCAST | 1864 IRE_INTERFACE)) != 0) { 1865 /* release references on behalf of caller */ 1866 if (gc != NULL) 1867 GC_REFRELE(gc); 1868 if (gcgrp != NULL) 1869 GCGRP_REFRELE(gcgrp); 1870 } else if ((ire->ire_mp == NULL) && 1871 tsol_ire_init_gwattr(ire, ipversion, gc, gcgrp) != 0) { 1872 /* free any caller-allocated mblks upon failure */ 1873 if (fp_mp != NULL) 1874 freeb(fp_mp); 1875 if (dlureq_mp != NULL) 1876 freeb(dlureq_mp); 1877 return (B_FALSE); 1878 } 1879 } 1880 1881 ire->ire_stq = stq; 1882 ire->ire_rfq = rfq; 1883 ire->ire_type = type; 1884 ire->ire_flags = RTF_UP | flags; 1885 ire->ire_ident = TICK_TO_MSEC(lbolt); 1886 bcopy(ulp_info, &ire->ire_uinfo, sizeof (iulp_t)); 1887 1888 ire->ire_tire_mark = ire->ire_ob_pkt_count + ire->ire_ib_pkt_count; 1889 ire->ire_last_used_time = lbolt; 1890 ire->ire_create_time = (uint32_t)gethrestime_sec(); 1891 1892 /* 1893 * If this IRE is an IRE_CACHE, inherit the handles from the 1894 * parent IREs. For others in the forwarding table, assign appropriate 1895 * new ones. 1896 * 1897 * The mutex protecting ire_handle is because ire_create is not always 1898 * called as a writer. 1899 */ 1900 if (ire->ire_type & IRE_OFFSUBNET) { 1901 mutex_enter(&ire_handle_lock); 1902 ire->ire_phandle = (uint32_t)ire_handle++; 1903 mutex_exit(&ire_handle_lock); 1904 } else if (ire->ire_type & IRE_INTERFACE) { 1905 mutex_enter(&ire_handle_lock); 1906 ire->ire_ihandle = (uint32_t)ire_handle++; 1907 mutex_exit(&ire_handle_lock); 1908 } else if (ire->ire_type == IRE_CACHE) { 1909 ire->ire_phandle = phandle; 1910 ire->ire_ihandle = ihandle; 1911 } 1912 ire->ire_in_ill = in_ill; 1913 ire->ire_ipif = ipif; 1914 if (ipif != NULL) { 1915 ire->ire_ipif_seqid = ipif->ipif_seqid; 1916 ire->ire_zoneid = ipif->ipif_zoneid; 1917 } else { 1918 ire->ire_zoneid = GLOBAL_ZONEID; 1919 } 1920 ire->ire_ipversion = ipversion; 1921 mutex_init(&ire->ire_lock, NULL, MUTEX_DEFAULT, NULL); 1922 if (ipversion == IPV4_VERSION) { 1923 if (ire_nce_init(ire, fp_mp, dlureq_mp) != 0) { 1924 /* some failure occurred. propagate error back */ 1925 return (B_FALSE); 1926 } 1927 } else { 1928 ASSERT(ipversion == IPV6_VERSION); 1929 /* 1930 * IPv6 initializes the ire_nce in ire_add_v6, 1931 * which expects to find the ire_nce to be null when 1932 * when it is called. 1933 */ 1934 if (dlureq_mp) 1935 freemsg(dlureq_mp); 1936 if (fp_mp) 1937 freemsg(fp_mp); 1938 } 1939 ire->ire_refcnt = 1; 1940 1941 #ifdef IRE_DEBUG 1942 bzero(ire->ire_trace, sizeof (th_trace_t *) * IP_TR_HASH_MAX); 1943 #endif 1944 1945 return (B_TRUE); 1946 } 1947 1948 /* 1949 * This routine is called repeatedly by ipif_up to create broadcast IREs. 1950 * It is passed a pointer to a slot in an IRE pointer array into which to 1951 * place the pointer to the new IRE, if indeed we create one. If the 1952 * IRE corresponding to the address passed in would be a duplicate of an 1953 * existing one, we don't create the new one. irep is incremented before 1954 * return only if we do create a new IRE. (Always called as writer.) 1955 * 1956 * Note that with the "match_flags" parameter, we can match on either 1957 * a particular logical interface (MATCH_IRE_IPIF) or for all logical 1958 * interfaces for a given physical interface (MATCH_IRE_ILL). Currently, 1959 * we only create broadcast ire's on a per physical interface basis. If 1960 * someone is going to be mucking with logical interfaces, it is important 1961 * to call "ipif_check_bcast_ires()" to make sure that any change to a 1962 * logical interface will not cause critical broadcast IRE's to be deleted. 1963 */ 1964 ire_t ** 1965 ire_check_and_create_bcast(ipif_t *ipif, ipaddr_t addr, ire_t **irep, 1966 int match_flags) 1967 { 1968 ire_t *ire; 1969 uint64_t check_flags = IPIF_DEPRECATED | IPIF_NOLOCAL | IPIF_ANYCAST; 1970 1971 /* 1972 * No broadcast IREs for the LOOPBACK interface 1973 * or others such as point to point and IPIF_NOXMIT. 1974 */ 1975 if (!(ipif->ipif_flags & IPIF_BROADCAST) || 1976 (ipif->ipif_flags & IPIF_NOXMIT)) 1977 return (irep); 1978 1979 /* If this would be a duplicate, don't bother. */ 1980 if ((ire = ire_ctable_lookup(addr, 0, IRE_BROADCAST, ipif, 1981 ipif->ipif_zoneid, NULL, match_flags)) != NULL) { 1982 /* 1983 * We look for non-deprecated (and non-anycast, non-nolocal) 1984 * ipifs as the best choice. ipifs with check_flags matching 1985 * (deprecated, etc) are used only if non-deprecated ipifs 1986 * are not available. if the existing ire's ipif is deprecated 1987 * and the new ipif is non-deprecated, switch to the new ipif 1988 */ 1989 if ((!(ire->ire_ipif->ipif_flags & check_flags)) || 1990 (ipif->ipif_flags & check_flags)) { 1991 ire_refrele(ire); 1992 return (irep); 1993 } 1994 /* 1995 * Bcast ires exist in pairs. Both have to be deleted, 1996 * Since we are exclusive we can make the above assertion. 1997 * The 1st has to be refrele'd since it was ctable_lookup'd. 1998 */ 1999 ASSERT(IAM_WRITER_IPIF(ipif)); 2000 ASSERT(ire->ire_next->ire_addr == ire->ire_addr); 2001 ire_delete(ire->ire_next); 2002 ire_delete(ire); 2003 ire_refrele(ire); 2004 } 2005 2006 irep = ire_create_bcast(ipif, addr, irep); 2007 2008 return (irep); 2009 } 2010 2011 uint_t ip_loopback_mtu = IP_LOOPBACK_MTU; 2012 2013 /* 2014 * This routine is called from ipif_check_bcast_ires and ire_check_bcast. 2015 * It leaves all the verifying and deleting to those routines. So it always 2016 * creates 2 bcast ires and chains them into the ire array passed in. 2017 */ 2018 ire_t ** 2019 ire_create_bcast(ipif_t *ipif, ipaddr_t addr, ire_t **irep) 2020 { 2021 *irep++ = ire_create( 2022 (uchar_t *)&addr, /* dest addr */ 2023 (uchar_t *)&ip_g_all_ones, /* mask */ 2024 (uchar_t *)&ipif->ipif_src_addr, /* source addr */ 2025 NULL, /* no gateway */ 2026 NULL, /* no in_src_addr */ 2027 &ipif->ipif_mtu, /* max frag */ 2028 NULL, /* fast path header */ 2029 ipif->ipif_rq, /* recv-from queue */ 2030 ipif->ipif_wq, /* send-to queue */ 2031 IRE_BROADCAST, 2032 ipif->ipif_bcast_mp, /* xmit header */ 2033 ipif, 2034 NULL, 2035 0, 2036 0, 2037 0, 2038 0, 2039 &ire_uinfo_null, 2040 NULL, 2041 NULL); 2042 2043 *irep++ = ire_create( 2044 (uchar_t *)&addr, /* dest address */ 2045 (uchar_t *)&ip_g_all_ones, /* mask */ 2046 (uchar_t *)&ipif->ipif_src_addr, /* source address */ 2047 NULL, /* no gateway */ 2048 NULL, /* no in_src_addr */ 2049 &ip_loopback_mtu, /* max frag size */ 2050 NULL, /* Fast Path header */ 2051 ipif->ipif_rq, /* recv-from queue */ 2052 NULL, /* no send-to queue */ 2053 IRE_BROADCAST, /* Needed for fanout in wput */ 2054 NULL, 2055 ipif, 2056 NULL, 2057 0, 2058 0, 2059 0, 2060 0, 2061 &ire_uinfo_null, 2062 NULL, 2063 NULL); 2064 2065 return (irep); 2066 } 2067 2068 /* 2069 * ire_walk routine to delete or update any IRE_CACHE that might contain 2070 * stale information. 2071 * The flags state which entries to delete or update. 2072 * Garbage collection is done separately using kmem alloc callbacks to 2073 * ip_trash_ire_reclaim. 2074 * Used for both IPv4 and IPv6. However, IPv6 only uses FLUSH_MTU_TIME 2075 * since other stale information is cleaned up using NUD. 2076 */ 2077 void 2078 ire_expire(ire_t *ire, char *arg) 2079 { 2080 int flush_flags = (int)(uintptr_t)arg; 2081 ill_t *stq_ill; 2082 2083 if ((flush_flags & FLUSH_REDIRECT_TIME) && 2084 ire->ire_type == IRE_HOST_REDIRECT) { 2085 /* Make sure we delete the corresponding IRE_CACHE */ 2086 ip1dbg(("ire_expire: all redirects\n")); 2087 ip_rts_rtmsg(RTM_DELETE, ire, 0); 2088 ire_delete(ire); 2089 atomic_dec_32(&ip_redirect_cnt); 2090 return; 2091 } 2092 if (ire->ire_type != IRE_CACHE) 2093 return; 2094 2095 if (flush_flags & FLUSH_ARP_TIME) { 2096 /* 2097 * Remove all IRE_CACHE. 2098 * Verify that create time is more than 2099 * ip_ire_arp_interval milliseconds ago. 2100 */ 2101 if (NCE_EXPIRED(ire->ire_nce)) { 2102 ire_delete(ire); 2103 return; 2104 } 2105 } 2106 2107 if (ip_path_mtu_discovery && (flush_flags & FLUSH_MTU_TIME) && 2108 (ire->ire_ipif != NULL)) { 2109 /* Increase pmtu if it is less than the interface mtu */ 2110 mutex_enter(&ire->ire_lock); 2111 /* 2112 * If the ipif is a vni (whose mtu is 0, since it's virtual) 2113 * get the mtu from the sending interfaces' ipif 2114 */ 2115 if (IS_VNI(ire->ire_ipif->ipif_ill)) { 2116 stq_ill = ire->ire_stq->q_ptr; 2117 ire->ire_max_frag = MIN(stq_ill->ill_ipif->ipif_mtu, 2118 IP_MAXPACKET); 2119 } else { 2120 ire->ire_max_frag = MIN(ire->ire_ipif->ipif_mtu, 2121 IP_MAXPACKET); 2122 } 2123 ire->ire_frag_flag |= IPH_DF; 2124 mutex_exit(&ire->ire_lock); 2125 } 2126 } 2127 2128 /* 2129 * Do fast path probing if necessary. 2130 */ 2131 void 2132 ire_fastpath(ire_t *ire) 2133 { 2134 ill_t *ill; 2135 int res; 2136 2137 if (ire->ire_nce == NULL || ire->ire_nce->nce_fp_mp != NULL || 2138 ire->ire_nce->nce_state != ND_REACHABLE || 2139 ire->ire_nce->nce_res_mp == NULL) { 2140 2141 /* 2142 * Already contains fastpath info or 2143 * doesn't have DL_UNITDATA_REQ header or 2144 * or is an incomplete ire in the ire table 2145 * or is a loopback broadcast ire i.e. no stq. 2146 */ 2147 return; 2148 } 2149 ill = ire_to_ill(ire); 2150 if (ill == NULL) 2151 return; 2152 ire_fastpath_list_add(ill, ire); 2153 res = ill_fastpath_probe(ill, ire->ire_nce->nce_res_mp); 2154 /* 2155 * EAGAIN is an indication of a transient error 2156 * i.e. allocation failure etc. leave the ire in the list it will 2157 * be updated when another probe happens for another ire if not 2158 * it will be taken out of the list when the ire is deleted. 2159 */ 2160 if (res != 0 && res != EAGAIN) 2161 ire_fastpath_list_delete(ill, ire); 2162 } 2163 2164 /* 2165 * Update all IRE's that are not in fastpath mode and 2166 * have an dlureq_mp that matches mp. mp->b_cont contains 2167 * the fastpath header. 2168 * 2169 * Returns TRUE if entry should be dequeued, or FALSE otherwise. 2170 */ 2171 boolean_t 2172 ire_fastpath_update(ire_t *ire, void *arg) 2173 { 2174 mblk_t *mp, *fp_mp; 2175 uchar_t *up, *up2; 2176 ptrdiff_t cmplen; 2177 nce_t *arpce; 2178 2179 ASSERT((ire->ire_type & (IRE_CACHE | IRE_BROADCAST | 2180 IRE_MIPRTUN)) != 0); 2181 2182 /* 2183 * Already contains fastpath info or doesn't have 2184 * DL_UNITDATA_REQ header or is an incomplete ire. 2185 */ 2186 if (ire->ire_nce == NULL || ire->ire_nce->nce_res_mp == NULL || 2187 ire->ire_nce->nce_fp_mp != NULL || 2188 ire->ire_nce->nce_state != ND_REACHABLE) 2189 return (B_TRUE); 2190 2191 ip2dbg(("ire_fastpath_update: trying\n")); 2192 mp = arg; 2193 up = mp->b_rptr; 2194 cmplen = mp->b_wptr - up; 2195 /* Serialize multiple fast path updates */ 2196 mutex_enter(&ire->ire_nce->nce_lock); 2197 up2 = ire->ire_nce->nce_res_mp->b_rptr; 2198 ASSERT(cmplen >= 0); 2199 if (ire->ire_nce->nce_res_mp->b_wptr - up2 != cmplen || 2200 bcmp(up, up2, cmplen) != 0) { 2201 mutex_exit(&ire->ire_nce->nce_lock); 2202 /* 2203 * Don't take the ire off the fastpath list yet, 2204 * since the response may come later. 2205 */ 2206 return (B_FALSE); 2207 } 2208 arpce = ire->ire_nce; 2209 /* Matched - install mp as the nce_fp_mp */ 2210 ip1dbg(("ire_fastpath_update: match\n")); 2211 fp_mp = dupb(mp->b_cont); 2212 if (fp_mp) { 2213 /* 2214 * We checked nce_fp_mp above. Check it again with the 2215 * lock. Update fp_mp only if it has not been done 2216 * already. 2217 */ 2218 if (arpce->nce_fp_mp == NULL) { 2219 /* 2220 * ire_ll_hdr_length is just an optimization to 2221 * store the length. It is used to return the 2222 * fast path header length to the upper layers. 2223 */ 2224 arpce->nce_fp_mp = fp_mp; 2225 ire->ire_ll_hdr_length = 2226 (uint_t)(fp_mp->b_wptr - fp_mp->b_rptr); 2227 } else { 2228 freeb(fp_mp); 2229 } 2230 } 2231 mutex_exit(&ire->ire_nce->nce_lock); 2232 return (B_TRUE); 2233 } 2234 2235 /* 2236 * This function handles the DL_NOTE_FASTPATH_FLUSH notification from the 2237 * driver. 2238 */ 2239 /* ARGSUSED */ 2240 void 2241 ire_fastpath_flush(ire_t *ire, void *arg) 2242 { 2243 ill_t *ill; 2244 int res; 2245 2246 /* No fastpath info? */ 2247 if (ire->ire_nce == NULL || 2248 ire->ire_nce->nce_fp_mp == NULL || ire->ire_nce->nce_res_mp == NULL) 2249 return; 2250 2251 /* 2252 * Just remove the IRE if it is for non-broadcast dest. Then 2253 * we will create another one which will have the correct 2254 * fastpath info. 2255 */ 2256 switch (ire->ire_type) { 2257 case IRE_CACHE: 2258 ire_delete(ire); 2259 break; 2260 case IRE_MIPRTUN: 2261 case IRE_BROADCAST: 2262 /* 2263 * We can't delete the ire since it is difficult to 2264 * recreate these ire's without going through the 2265 * ipif down/up dance. The nce_fp_mp is protected by the 2266 * nce_lock in the case of IRE_MIPRTUN and IRE_BROADCAST. 2267 * All access to ire->ire_nce->nce_fp_mp in the case of these 2268 * 2 ire types * is protected by nce_lock. 2269 */ 2270 mutex_enter(&ire->ire_nce->nce_lock); 2271 if (ire->ire_nce->nce_fp_mp != NULL) { 2272 freeb(ire->ire_nce->nce_fp_mp); 2273 ire->ire_nce->nce_fp_mp = NULL; 2274 mutex_exit(&ire->ire_nce->nce_lock); 2275 /* 2276 * No fastpath probe if there is no stq i.e. 2277 * i.e. the case of loopback broadcast ire. 2278 */ 2279 if (ire->ire_stq == NULL) 2280 break; 2281 ill = (ill_t *)((ire->ire_stq)->q_ptr); 2282 ire_fastpath_list_add(ill, ire); 2283 res = ill_fastpath_probe(ill, ire->ire_nce->nce_res_mp); 2284 /* 2285 * EAGAIN is an indication of a transient error 2286 * i.e. allocation failure etc. leave the ire in the 2287 * list it will be updated when another probe happens 2288 * for another ire if not it will be taken out of the 2289 * list when the ire is deleted. 2290 */ 2291 if (res != 0 && res != EAGAIN) 2292 ire_fastpath_list_delete(ill, ire); 2293 } else { 2294 mutex_exit(&ire->ire_nce->nce_lock); 2295 } 2296 break; 2297 default: 2298 /* This should not happen! */ 2299 ip0dbg(("ire_fastpath_flush: Wrong ire type %s\n", 2300 ip_nv_lookup(ire_nv_tbl, (int)ire->ire_type))); 2301 break; 2302 } 2303 } 2304 2305 /* 2306 * Drain the list of ire's waiting for fastpath response. 2307 */ 2308 void 2309 ire_fastpath_list_dispatch(ill_t *ill, boolean_t (*func)(ire_t *, void *), 2310 void *arg) 2311 { 2312 ire_t *next_ire; 2313 ire_t *current_ire; 2314 ire_t *first_ire; 2315 ire_t *prev_ire = NULL; 2316 2317 ASSERT(ill != NULL); 2318 2319 mutex_enter(&ill->ill_lock); 2320 first_ire = current_ire = (ire_t *)ill->ill_fastpath_list; 2321 while (current_ire != (ire_t *)&ill->ill_fastpath_list) { 2322 next_ire = current_ire->ire_fastpath; 2323 /* 2324 * Take it off the list if we're flushing, or if the callback 2325 * routine tells us to do so. Otherwise, leave the ire in the 2326 * fastpath list to handle any pending response from the lower 2327 * layer. We can't drain the list when the callback routine 2328 * comparison failed, because the response is asynchronous in 2329 * nature, and may not arrive in the same order as the list 2330 * insertion. 2331 */ 2332 if (func == NULL || func(current_ire, arg)) { 2333 current_ire->ire_fastpath = NULL; 2334 if (current_ire == first_ire) 2335 ill->ill_fastpath_list = first_ire = next_ire; 2336 else 2337 prev_ire->ire_fastpath = next_ire; 2338 } else { 2339 /* previous element that is still in the list */ 2340 prev_ire = current_ire; 2341 } 2342 current_ire = next_ire; 2343 } 2344 mutex_exit(&ill->ill_lock); 2345 } 2346 2347 /* 2348 * Add ire to the ire fastpath list. 2349 */ 2350 static void 2351 ire_fastpath_list_add(ill_t *ill, ire_t *ire) 2352 { 2353 ASSERT(ill != NULL); 2354 ASSERT(ire->ire_stq != NULL); 2355 2356 rw_enter(&ire->ire_bucket->irb_lock, RW_READER); 2357 mutex_enter(&ill->ill_lock); 2358 2359 /* 2360 * if ire has not been deleted and 2361 * is not already in the list add it. 2362 */ 2363 if (((ire->ire_marks & IRE_MARK_CONDEMNED) == 0) && 2364 (ire->ire_fastpath == NULL)) { 2365 ire->ire_fastpath = (ire_t *)ill->ill_fastpath_list; 2366 ill->ill_fastpath_list = ire; 2367 } 2368 2369 mutex_exit(&ill->ill_lock); 2370 rw_exit(&ire->ire_bucket->irb_lock); 2371 } 2372 2373 /* 2374 * remove ire from the ire fastpath list. 2375 */ 2376 void 2377 ire_fastpath_list_delete(ill_t *ill, ire_t *ire) 2378 { 2379 ire_t *ire_ptr; 2380 2381 ASSERT(ire->ire_stq != NULL && ill != NULL); 2382 2383 mutex_enter(&ill->ill_lock); 2384 if (ire->ire_fastpath == NULL) 2385 goto done; 2386 2387 ASSERT(ill->ill_fastpath_list != &ill->ill_fastpath_list); 2388 2389 if (ill->ill_fastpath_list == ire) { 2390 ill->ill_fastpath_list = ire->ire_fastpath; 2391 } else { 2392 ire_ptr = ill->ill_fastpath_list; 2393 while (ire_ptr != (ire_t *)&ill->ill_fastpath_list) { 2394 if (ire_ptr->ire_fastpath == ire) { 2395 ire_ptr->ire_fastpath = ire->ire_fastpath; 2396 break; 2397 } 2398 ire_ptr = ire_ptr->ire_fastpath; 2399 } 2400 } 2401 ire->ire_fastpath = NULL; 2402 done: 2403 mutex_exit(&ill->ill_lock); 2404 } 2405 2406 /* 2407 * Return any local address. We use this to target ourselves 2408 * when the src address was specified as 'default'. 2409 * Preference for IRE_LOCAL entries. 2410 */ 2411 ire_t * 2412 ire_lookup_local(zoneid_t zoneid) 2413 { 2414 ire_t *ire; 2415 irb_t *irb; 2416 ire_t *maybe = NULL; 2417 int i; 2418 2419 for (i = 0; i < ip_cache_table_size; i++) { 2420 irb = &ip_cache_table[i]; 2421 if (irb->irb_ire == NULL) 2422 continue; 2423 rw_enter(&irb->irb_lock, RW_READER); 2424 for (ire = irb->irb_ire; ire != NULL; ire = ire->ire_next) { 2425 if ((ire->ire_marks & IRE_MARK_CONDEMNED) || 2426 (ire->ire_zoneid != zoneid && 2427 ire->ire_zoneid != ALL_ZONES)) 2428 continue; 2429 switch (ire->ire_type) { 2430 case IRE_LOOPBACK: 2431 if (maybe == NULL) { 2432 IRE_REFHOLD(ire); 2433 maybe = ire; 2434 } 2435 break; 2436 case IRE_LOCAL: 2437 if (maybe != NULL) { 2438 ire_refrele(maybe); 2439 } 2440 IRE_REFHOLD(ire); 2441 rw_exit(&irb->irb_lock); 2442 return (ire); 2443 } 2444 } 2445 rw_exit(&irb->irb_lock); 2446 } 2447 return (maybe); 2448 } 2449 2450 /* 2451 * If the specified IRE is associated with a particular ILL, return 2452 * that ILL pointer (May be called as writer.). 2453 * 2454 * NOTE : This is not a generic function that can be used always. 2455 * This function always returns the ill of the outgoing packets 2456 * if this ire is used. 2457 */ 2458 ill_t * 2459 ire_to_ill(const ire_t *ire) 2460 { 2461 ill_t *ill = NULL; 2462 2463 /* 2464 * 1) For an IRE_CACHE, ire_ipif is the one where it obtained 2465 * the source address from. ire_stq is the one where the 2466 * packets will be sent out on. We return that here. 2467 * 2468 * 2) IRE_BROADCAST normally has a loopback and a non-loopback 2469 * copy and they always exist next to each other with loopback 2470 * copy being the first one. If we are called on the non-loopback 2471 * copy, return the one pointed by ire_stq. If it was called on 2472 * a loopback copy, we still return the one pointed by the next 2473 * ire's ire_stq pointer i.e the one pointed by the non-loopback 2474 * copy. We don't want use ire_ipif as it might represent the 2475 * source address (if we borrow source addresses for 2476 * IRE_BROADCASTS in the future). 2477 * However if an interface is currently coming up, the above 2478 * condition may not hold during that period since the ires 2479 * are added one at a time. Thus one of the pair could have been 2480 * added and the other not yet added. 2481 * 3) For all others return the ones pointed by ire_ipif->ipif_ill. 2482 */ 2483 2484 if (ire->ire_type == IRE_CACHE) { 2485 ill = (ill_t *)ire->ire_stq->q_ptr; 2486 } else if (ire->ire_type == IRE_BROADCAST) { 2487 if (ire->ire_stq != NULL) { 2488 ill = (ill_t *)ire->ire_stq->q_ptr; 2489 } else { 2490 ire_t *ire_next; 2491 2492 ire_next = ire->ire_next; 2493 if (ire_next != NULL && 2494 ire_next->ire_type == IRE_BROADCAST && 2495 ire_next->ire_addr == ire->ire_addr && 2496 ire_next->ire_ipif == ire->ire_ipif) { 2497 ill = (ill_t *)ire_next->ire_stq->q_ptr; 2498 } 2499 } 2500 } else if (ire->ire_ipif != NULL) { 2501 ill = ire->ire_ipif->ipif_ill; 2502 } 2503 return (ill); 2504 } 2505 2506 /* Arrange to call the specified function for every IRE in the world. */ 2507 void 2508 ire_walk(pfv_t func, void *arg) 2509 { 2510 ire_walk_ipvers(func, arg, 0, ALL_ZONES); 2511 } 2512 2513 void 2514 ire_walk_v4(pfv_t func, void *arg, zoneid_t zoneid) 2515 { 2516 ire_walk_ipvers(func, arg, IPV4_VERSION, zoneid); 2517 } 2518 2519 void 2520 ire_walk_v6(pfv_t func, void *arg, zoneid_t zoneid) 2521 { 2522 ire_walk_ipvers(func, arg, IPV6_VERSION, zoneid); 2523 } 2524 2525 /* 2526 * Walk a particular version. version == 0 means both v4 and v6. 2527 */ 2528 static void 2529 ire_walk_ipvers(pfv_t func, void *arg, uchar_t vers, zoneid_t zoneid) 2530 { 2531 if (vers != IPV6_VERSION) { 2532 /* 2533 * ip_forwarding_table variable doesn't matter for IPv4 since 2534 * ire_walk_ill_tables directly calls with the ip_ftable global 2535 */ 2536 ire_walk_ill_tables(0, 0, func, arg, IP_MASK_TABLE_SIZE, 2537 0, NULL, 2538 ip_cache_table_size, ip_cache_table, NULL, zoneid); 2539 } 2540 if (vers != IPV4_VERSION) { 2541 ire_walk_ill_tables(0, 0, func, arg, IP6_MASK_TABLE_SIZE, 2542 ip6_ftable_hash_size, ip_forwarding_table_v6, 2543 ip6_cache_table_size, ip_cache_table_v6, NULL, zoneid); 2544 } 2545 } 2546 2547 /* 2548 * Arrange to call the specified 2549 * function for every IRE that matches the ill. 2550 */ 2551 void 2552 ire_walk_ill(uint_t match_flags, uint_t ire_type, pfv_t func, void *arg, 2553 ill_t *ill) 2554 { 2555 ire_walk_ill_ipvers(match_flags, ire_type, func, arg, 0, ill); 2556 } 2557 2558 void 2559 ire_walk_ill_v4(uint_t match_flags, uint_t ire_type, pfv_t func, void *arg, 2560 ill_t *ill) 2561 { 2562 ire_walk_ill_ipvers(match_flags, ire_type, func, arg, IPV4_VERSION, 2563 ill); 2564 } 2565 2566 void 2567 ire_walk_ill_v6(uint_t match_flags, uint_t ire_type, pfv_t func, void *arg, 2568 ill_t *ill) 2569 { 2570 ire_walk_ill_ipvers(match_flags, ire_type, func, arg, IPV6_VERSION, 2571 ill); 2572 } 2573 2574 /* 2575 * Walk a particular ill and version. version == 0 means both v4 and v6. 2576 */ 2577 static void 2578 ire_walk_ill_ipvers(uint_t match_flags, uint_t ire_type, pfv_t func, 2579 void *arg, uchar_t vers, ill_t *ill) 2580 { 2581 if (vers != IPV6_VERSION) { 2582 ire_walk_ill_tables(match_flags, ire_type, func, arg, 2583 IP_MASK_TABLE_SIZE, 0, 2584 NULL, ip_cache_table_size, 2585 ip_cache_table, ill, ALL_ZONES); 2586 } 2587 if (vers != IPV4_VERSION) { 2588 ire_walk_ill_tables(match_flags, ire_type, func, arg, 2589 IP6_MASK_TABLE_SIZE, ip6_ftable_hash_size, 2590 ip_forwarding_table_v6, ip6_cache_table_size, 2591 ip_cache_table_v6, ill, ALL_ZONES); 2592 } 2593 } 2594 2595 boolean_t 2596 ire_walk_ill_match(uint_t match_flags, uint_t ire_type, ire_t *ire, 2597 ill_t *ill, zoneid_t zoneid) 2598 { 2599 ill_t *ire_stq_ill = NULL; 2600 ill_t *ire_ipif_ill = NULL; 2601 ill_group_t *ire_ill_group = NULL; 2602 2603 ASSERT(match_flags != 0 || zoneid != ALL_ZONES); 2604 /* 2605 * 1) MATCH_IRE_WQ : Used specifically to match on ire_stq. 2606 * The fast path update uses this to make sure it does not 2607 * update the fast path header of interface X with the fast 2608 * path updates it recieved on interface Y. It is similar 2609 * in handling DL_NOTE_FASTPATH_FLUSH. 2610 * 2611 * 2) MATCH_IRE_ILL/MATCH_IRE_ILL_GROUP : We match both on ill 2612 * pointed by ire_stq and ire_ipif. Only in the case of 2613 * IRE_CACHEs can ire_stq and ire_ipif be pointing to 2614 * different ills. But we want to keep this function generic 2615 * enough for future use. So, we always try to match on both. 2616 * The only caller of this function ire_walk_ill_tables, will 2617 * call "func" after we return from this function. We expect 2618 * "func" to do the right filtering of ires in this case. 2619 * 2620 * NOTE : In the case of MATCH_IRE_ILL_GROUP, groups 2621 * pointed by ire_stq and ire_ipif should always be the same. 2622 * So, we just match on only one of them. 2623 */ 2624 if (match_flags & (MATCH_IRE_ILL|MATCH_IRE_ILL_GROUP)) { 2625 if (ire->ire_stq != NULL) 2626 ire_stq_ill = (ill_t *)ire->ire_stq->q_ptr; 2627 if (ire->ire_ipif != NULL) 2628 ire_ipif_ill = ire->ire_ipif->ipif_ill; 2629 if (ire_stq_ill != NULL) 2630 ire_ill_group = ire_stq_ill->ill_group; 2631 if ((ire_ill_group == NULL) && (ire_ipif_ill != NULL)) 2632 ire_ill_group = ire_ipif_ill->ill_group; 2633 } 2634 2635 if (zoneid != ALL_ZONES) { 2636 /* 2637 * We're walking the IREs for a specific zone. The only relevant 2638 * IREs are: 2639 * - all IREs with a matching ire_zoneid 2640 * - all IRE_OFFSUBNETs as they're shared across all zones 2641 * - IRE_INTERFACE IREs for interfaces with a usable source addr 2642 * with a matching zone 2643 * - IRE_DEFAULTs with a gateway reachable from the zone 2644 * We should really match on IRE_OFFSUBNETs and IRE_DEFAULTs 2645 * using the same rule; but the above rules are consistent with 2646 * the behavior of ire_ftable_lookup[_v6]() so that all the 2647 * routes that can be matched during lookup are also matched 2648 * here. 2649 */ 2650 if (zoneid != ire->ire_zoneid && ire->ire_zoneid != ALL_ZONES) { 2651 /* 2652 * Note, IRE_INTERFACE can have the stq as NULL. For 2653 * example, if the default multicast route is tied to 2654 * the loopback address. 2655 */ 2656 if ((ire->ire_type & IRE_INTERFACE) && 2657 (ire->ire_stq != NULL)) { 2658 ire_stq_ill = (ill_t *)ire->ire_stq->q_ptr; 2659 if (ire->ire_ipversion == IPV4_VERSION) { 2660 if (!ipif_usesrc_avail(ire_stq_ill, 2661 zoneid)) 2662 /* No usable src addr in zone */ 2663 return (B_FALSE); 2664 } else if (ire_stq_ill->ill_usesrc_ifindex 2665 != 0) { 2666 /* 2667 * For IPv6 use ipif_select_source_v6() 2668 * so the right scope selection is done 2669 */ 2670 ipif_t *src_ipif; 2671 src_ipif = 2672 ipif_select_source_v6(ire_stq_ill, 2673 &ire->ire_addr_v6, RESTRICT_TO_NONE, 2674 IPV6_PREFER_SRC_DEFAULT, 2675 zoneid); 2676 if (src_ipif != NULL) { 2677 ipif_refrele(src_ipif); 2678 } else { 2679 return (B_FALSE); 2680 } 2681 } else { 2682 return (B_FALSE); 2683 } 2684 2685 } else if (!(ire->ire_type & IRE_OFFSUBNET)) { 2686 return (B_FALSE); 2687 } 2688 } 2689 2690 /* 2691 * Match all default routes from the global zone, irrespective 2692 * of reachability. For a non-global zone only match those 2693 * where ire_gateway_addr has a IRE_INTERFACE for the zoneid. 2694 */ 2695 if (ire->ire_type == IRE_DEFAULT && zoneid != GLOBAL_ZONEID) { 2696 int ire_match_flags = 0; 2697 in6_addr_t gw_addr_v6; 2698 ire_t *rire; 2699 2700 ire_match_flags |= MATCH_IRE_TYPE; 2701 if (ire->ire_ipif != NULL) { 2702 ire_match_flags |= MATCH_IRE_ILL_GROUP; 2703 } 2704 if (ire->ire_ipversion == IPV4_VERSION) { 2705 rire = ire_route_lookup(ire->ire_gateway_addr, 2706 0, 0, IRE_INTERFACE, ire->ire_ipif, NULL, 2707 zoneid, NULL, ire_match_flags); 2708 } else { 2709 ASSERT(ire->ire_ipversion == IPV6_VERSION); 2710 mutex_enter(&ire->ire_lock); 2711 gw_addr_v6 = ire->ire_gateway_addr_v6; 2712 mutex_exit(&ire->ire_lock); 2713 rire = ire_route_lookup_v6(&gw_addr_v6, 2714 NULL, NULL, IRE_INTERFACE, ire->ire_ipif, 2715 NULL, zoneid, NULL, ire_match_flags); 2716 } 2717 if (rire == NULL) { 2718 return (B_FALSE); 2719 } 2720 ire_refrele(rire); 2721 } 2722 } 2723 2724 if (((!(match_flags & MATCH_IRE_TYPE)) || 2725 (ire->ire_type & ire_type)) && 2726 ((!(match_flags & MATCH_IRE_WQ)) || 2727 (ire->ire_stq == ill->ill_wq)) && 2728 ((!(match_flags & MATCH_IRE_ILL)) || 2729 (ire_stq_ill == ill || ire_ipif_ill == ill)) && 2730 ((!(match_flags & MATCH_IRE_ILL_GROUP)) || 2731 (ire_stq_ill == ill) || (ire_ipif_ill == ill) || 2732 (ire_ill_group != NULL && 2733 ire_ill_group == ill->ill_group))) { 2734 return (B_TRUE); 2735 } 2736 return (B_FALSE); 2737 } 2738 2739 int 2740 rtfunc(struct radix_node *rn, void *arg) 2741 { 2742 struct rtfuncarg *rtf = arg; 2743 struct rt_entry *rt; 2744 irb_t *irb; 2745 ire_t *ire; 2746 boolean_t ret; 2747 2748 rt = (struct rt_entry *)rn; 2749 ASSERT(rt != NULL); 2750 irb = &rt->rt_irb; 2751 for (ire = irb->irb_ire; ire != NULL; ire = ire->ire_next) { 2752 if ((rtf->rt_match_flags != 0) || 2753 (rtf->rt_zoneid != ALL_ZONES)) { 2754 ret = ire_walk_ill_match(rtf->rt_match_flags, 2755 rtf->rt_ire_type, ire, 2756 rtf->rt_ill, rtf->rt_zoneid); 2757 } else 2758 ret = B_TRUE; 2759 if (ret) 2760 (*rtf->rt_func)(ire, rtf->rt_arg); 2761 } 2762 return (0); 2763 } 2764 2765 /* 2766 * Walk the ftable and the ctable entries that match the ill. 2767 */ 2768 void 2769 ire_walk_ill_tables(uint_t match_flags, uint_t ire_type, pfv_t func, 2770 void *arg, size_t ftbl_sz, size_t htbl_sz, irb_t **ipftbl, 2771 size_t ctbl_sz, irb_t *ipctbl, ill_t *ill, zoneid_t zoneid) 2772 { 2773 irb_t *irb_ptr; 2774 irb_t *irb; 2775 ire_t *ire; 2776 int i, j; 2777 boolean_t ret; 2778 struct rtfuncarg rtfarg; 2779 2780 ASSERT((!(match_flags & (MATCH_IRE_WQ | MATCH_IRE_ILL | 2781 MATCH_IRE_ILL_GROUP))) || (ill != NULL)); 2782 ASSERT(!(match_flags & MATCH_IRE_TYPE) || (ire_type != 0)); 2783 /* 2784 * Optimize by not looking at the forwarding table if there 2785 * is a MATCH_IRE_TYPE specified with no IRE_FORWARDTABLE 2786 * specified in ire_type. 2787 */ 2788 if (!(match_flags & MATCH_IRE_TYPE) || 2789 ((ire_type & IRE_FORWARDTABLE) != 0)) { 2790 /* knobs such that routine is called only for v6 case */ 2791 if (ipftbl == ip_forwarding_table_v6) { 2792 for (i = (ftbl_sz - 1); i >= 0; i--) { 2793 if ((irb_ptr = ipftbl[i]) == NULL) 2794 continue; 2795 for (j = 0; j < htbl_sz; j++) { 2796 irb = &irb_ptr[j]; 2797 if (irb->irb_ire == NULL) 2798 continue; 2799 2800 IRB_REFHOLD(irb); 2801 for (ire = irb->irb_ire; ire != NULL; 2802 ire = ire->ire_next) { 2803 if (match_flags == 0 && 2804 zoneid == ALL_ZONES) { 2805 ret = B_TRUE; 2806 } else { 2807 ret = 2808 ire_walk_ill_match( 2809 match_flags, 2810 ire_type, ire, ill, 2811 zoneid); 2812 } 2813 if (ret) 2814 (*func)(ire, arg); 2815 } 2816 IRB_REFRELE(irb); 2817 } 2818 } 2819 } else { 2820 (void) memset(&rtfarg, 0, sizeof (rtfarg)); 2821 rtfarg.rt_func = func; 2822 rtfarg.rt_arg = arg; 2823 if (match_flags != 0) { 2824 rtfarg.rt_match_flags = match_flags; 2825 } 2826 rtfarg.rt_ire_type = ire_type; 2827 rtfarg.rt_ill = ill; 2828 rtfarg.rt_zoneid = zoneid; 2829 (void) ip_ftable->rnh_walktree(ip_ftable, rtfunc, 2830 &rtfarg); 2831 } 2832 } 2833 2834 /* 2835 * Optimize by not looking at the cache table if there 2836 * is a MATCH_IRE_TYPE specified with no IRE_CACHETABLE 2837 * specified in ire_type. 2838 */ 2839 if (!(match_flags & MATCH_IRE_TYPE) || 2840 ((ire_type & IRE_CACHETABLE) != 0)) { 2841 for (i = 0; i < ctbl_sz; i++) { 2842 irb = &ipctbl[i]; 2843 if (irb->irb_ire == NULL) 2844 continue; 2845 IRB_REFHOLD(irb); 2846 for (ire = irb->irb_ire; ire != NULL; 2847 ire = ire->ire_next) { 2848 if (match_flags == 0 && zoneid == ALL_ZONES) { 2849 ret = B_TRUE; 2850 } else { 2851 ret = ire_walk_ill_match( 2852 match_flags, ire_type, 2853 ire, ill, zoneid); 2854 } 2855 if (ret) 2856 (*func)(ire, arg); 2857 } 2858 IRB_REFRELE(irb); 2859 } 2860 } 2861 } 2862 2863 /* 2864 * This routine walks through the ill chain to find if there is any 2865 * ire linked to the ill's interface based forwarding table 2866 * The arg could be ill or mp. This routine is called when a ill goes 2867 * down/deleted or the 'ipv4_ire_srcif_status' report is printed. 2868 */ 2869 void 2870 ire_walk_srcif_table_v4(pfv_t func, void *arg) 2871 { 2872 irb_t *irb; 2873 ire_t *ire; 2874 ill_t *ill, *next_ill; 2875 int i; 2876 int total_count; 2877 ill_walk_context_t ctx; 2878 2879 /* 2880 * Take care of ire's in other ill's per-interface forwarding 2881 * table. Check if any ire in any of the ill's ill_srcif_table 2882 * is pointing to this ill. 2883 */ 2884 mutex_enter(&ire_srcif_table_lock); 2885 if (ire_srcif_table_count == 0) { 2886 mutex_exit(&ire_srcif_table_lock); 2887 return; 2888 } 2889 mutex_exit(&ire_srcif_table_lock); 2890 2891 #ifdef DEBUG 2892 /* Keep accounting of all interface based table ires */ 2893 total_count = 0; 2894 rw_enter(&ill_g_lock, RW_READER); 2895 ill = ILL_START_WALK_V4(&ctx); 2896 while (ill != NULL) { 2897 mutex_enter(&ill->ill_lock); 2898 total_count += ill->ill_srcif_refcnt; 2899 next_ill = ill_next(&ctx, ill); 2900 mutex_exit(&ill->ill_lock); 2901 ill = next_ill; 2902 } 2903 rw_exit(&ill_g_lock); 2904 2905 /* Hold lock here to make sure ire_srcif_table_count is stable */ 2906 mutex_enter(&ire_srcif_table_lock); 2907 i = ire_srcif_table_count; 2908 mutex_exit(&ire_srcif_table_lock); 2909 ip1dbg(("ire_walk_srcif_v4: ire_srcif_table_count %d " 2910 "total ill_srcif_refcnt %d\n", i, total_count)); 2911 #endif 2912 rw_enter(&ill_g_lock, RW_READER); 2913 ill = ILL_START_WALK_V4(&ctx); 2914 while (ill != NULL) { 2915 mutex_enter(&ill->ill_lock); 2916 if ((ill->ill_srcif_refcnt == 0) || !ILL_CAN_LOOKUP(ill)) { 2917 next_ill = ill_next(&ctx, ill); 2918 mutex_exit(&ill->ill_lock); 2919 ill = next_ill; 2920 continue; 2921 } 2922 ill_refhold_locked(ill); 2923 mutex_exit(&ill->ill_lock); 2924 rw_exit(&ill_g_lock); 2925 if (ill->ill_srcif_table != NULL) { 2926 for (i = 0; i < IP_SRCIF_TABLE_SIZE; i++) { 2927 irb = &(ill->ill_srcif_table[i]); 2928 if (irb->irb_ire == NULL) 2929 continue; 2930 IRB_REFHOLD(irb); 2931 for (ire = irb->irb_ire; ire != NULL; 2932 ire = ire->ire_next) { 2933 (*func)(ire, arg); 2934 } 2935 IRB_REFRELE(irb); 2936 } 2937 } 2938 rw_enter(&ill_g_lock, RW_READER); 2939 next_ill = ill_next(&ctx, ill); 2940 ill_refrele(ill); 2941 ill = next_ill; 2942 } 2943 rw_exit(&ill_g_lock); 2944 } 2945 2946 /* 2947 * This function takes a mask and returns 2948 * number of bits set in the mask. If no 2949 * bit is set it returns 0. 2950 * Assumes a contiguous mask. 2951 */ 2952 int 2953 ip_mask_to_plen(ipaddr_t mask) 2954 { 2955 return (mask == 0 ? 0 : IP_ABITS - (ffs(ntohl(mask)) -1)); 2956 } 2957 2958 /* 2959 * Convert length for a mask to the mask. 2960 */ 2961 ipaddr_t 2962 ip_plen_to_mask(uint_t masklen) 2963 { 2964 return (htonl(IP_HOST_MASK << (IP_ABITS - masklen))); 2965 } 2966 2967 void 2968 ire_atomic_end(irb_t *irb_ptr, ire_t *ire) 2969 { 2970 ill_t *ill_list[NUM_ILLS]; 2971 2972 ill_list[0] = ire->ire_stq != NULL ? ire->ire_stq->q_ptr : NULL; 2973 ill_list[1] = ire->ire_ipif != NULL ? ire->ire_ipif->ipif_ill : NULL; 2974 ill_list[2] = ire->ire_in_ill; 2975 ill_unlock_ills(ill_list, NUM_ILLS); 2976 rw_exit(&irb_ptr->irb_lock); 2977 rw_exit(&ill_g_usesrc_lock); 2978 } 2979 2980 /* 2981 * ire_add_v[46] atomically make sure that the ipif or ill associated 2982 * with the new ire being added is stable and not IPIF_CHANGING or ILL_CHANGING 2983 * before adding the ire to the table. This ensures that we don't create 2984 * new IRE_CACHEs with stale values for parameters that are passed to 2985 * ire_create such as ire_max_frag. Note that ire_create() is passed a pointer 2986 * to the ipif_mtu, and not the value. The actual value is derived from the 2987 * parent ire or ipif under the bucket lock. 2988 */ 2989 int 2990 ire_atomic_start(irb_t *irb_ptr, ire_t *ire, queue_t *q, mblk_t *mp, 2991 ipsq_func_t func) 2992 { 2993 ill_t *stq_ill; 2994 ill_t *ipif_ill; 2995 ill_t *in_ill; 2996 ill_t *ill_list[NUM_ILLS]; 2997 int cnt = NUM_ILLS; 2998 int error = 0; 2999 ill_t *ill = NULL; 3000 3001 ill_list[0] = stq_ill = ire->ire_stq != 3002 NULL ? ire->ire_stq->q_ptr : NULL; 3003 ill_list[1] = ipif_ill = ire->ire_ipif != 3004 NULL ? ire->ire_ipif->ipif_ill : NULL; 3005 ill_list[2] = in_ill = ire->ire_in_ill; 3006 3007 ASSERT((q != NULL && mp != NULL && func != NULL) || 3008 (q == NULL && mp == NULL && func == NULL)); 3009 rw_enter(&ill_g_usesrc_lock, RW_READER); 3010 GRAB_CONN_LOCK(q); 3011 rw_enter(&irb_ptr->irb_lock, RW_WRITER); 3012 ill_lock_ills(ill_list, cnt); 3013 3014 /* 3015 * While the IRE is in the process of being added, a user may have 3016 * invoked the ifconfig usesrc option on the stq_ill to make it a 3017 * usesrc client ILL. Check for this possibility here, if it is true 3018 * then we fail adding the IRE_CACHE. Another check is to make sure 3019 * that an ipif_ill of an IRE_CACHE being added is not part of a usesrc 3020 * group. The ill_g_usesrc_lock is released in ire_atomic_end 3021 */ 3022 if ((ire->ire_type & IRE_CACHE) && 3023 (ire->ire_marks & IRE_MARK_USESRC_CHECK)) { 3024 if (stq_ill->ill_usesrc_ifindex != 0) { 3025 ASSERT(stq_ill->ill_usesrc_grp_next != NULL); 3026 if ((ipif_ill->ill_phyint->phyint_ifindex != 3027 stq_ill->ill_usesrc_ifindex) || 3028 (ipif_ill->ill_usesrc_grp_next == NULL) || 3029 (ipif_ill->ill_usesrc_ifindex != 0)) { 3030 error = EINVAL; 3031 goto done; 3032 } 3033 } else if (ipif_ill->ill_usesrc_grp_next != NULL) { 3034 error = EINVAL; 3035 goto done; 3036 } 3037 } 3038 3039 /* 3040 * IPMP flag settings happen without taking the exclusive route 3041 * in ip_sioctl_flags. So we need to make an atomic check here 3042 * for FAILED/OFFLINE/INACTIVE flags or if it has hit the 3043 * FAILBACK=no case. 3044 */ 3045 if ((stq_ill != NULL) && !IAM_WRITER_ILL(stq_ill)) { 3046 if (stq_ill->ill_state_flags & ILL_CHANGING) { 3047 ill = stq_ill; 3048 error = EAGAIN; 3049 } else if ((stq_ill->ill_phyint->phyint_flags & PHYI_OFFLINE) || 3050 (ill_is_probeonly(stq_ill) && 3051 !(ire->ire_marks & IRE_MARK_HIDDEN))) { 3052 error = EINVAL; 3053 } 3054 goto done; 3055 } 3056 3057 /* 3058 * We don't check for OFFLINE/FAILED in this case because 3059 * the source address selection logic (ipif_select_source) 3060 * may still select a source address from such an ill. The 3061 * assumption is that these addresses will be moved by in.mpathd 3062 * soon. (i.e. this is a race). However link local addresses 3063 * will not move and hence ipif_select_source_v6 tries to avoid 3064 * FAILED ills. Please see ipif_select_source_v6 for more info 3065 */ 3066 if ((ipif_ill != NULL) && !IAM_WRITER_ILL(ipif_ill) && 3067 (ipif_ill->ill_state_flags & ILL_CHANGING)) { 3068 ill = ipif_ill; 3069 error = EAGAIN; 3070 goto done; 3071 } 3072 3073 if ((in_ill != NULL) && !IAM_WRITER_ILL(in_ill) && 3074 (in_ill->ill_state_flags & ILL_CHANGING)) { 3075 ill = in_ill; 3076 error = EAGAIN; 3077 goto done; 3078 } 3079 3080 if ((ire->ire_ipif != NULL) && !IAM_WRITER_IPIF(ire->ire_ipif) && 3081 (ire->ire_ipif->ipif_state_flags & IPIF_CHANGING)) { 3082 ill = ire->ire_ipif->ipif_ill; 3083 ASSERT(ill != NULL); 3084 error = EAGAIN; 3085 goto done; 3086 } 3087 3088 done: 3089 if (error == EAGAIN && ILL_CAN_WAIT(ill, q)) { 3090 ipsq_t *ipsq = ill->ill_phyint->phyint_ipsq; 3091 mutex_enter(&ipsq->ipsq_lock); 3092 ire_atomic_end(irb_ptr, ire); 3093 ipsq_enq(ipsq, q, mp, func, NEW_OP, ill); 3094 mutex_exit(&ipsq->ipsq_lock); 3095 error = EINPROGRESS; 3096 } else if (error != 0) { 3097 ire_atomic_end(irb_ptr, ire); 3098 } 3099 3100 RELEASE_CONN_LOCK(q); 3101 return (error); 3102 } 3103 3104 /* 3105 * Add a fully initialized IRE to an appropriate table based on 3106 * ire_type. 3107 * 3108 * allow_unresolved == B_FALSE indicates a legacy code-path call 3109 * that has prohibited the addition of incomplete ire's. If this 3110 * parameter is set, and we find an nce that is in a state other 3111 * than ND_REACHABLE, we fail the add. Note that nce_state could be 3112 * something other than ND_REACHABLE if nce_reinit has just 3113 * kicked in and reset the nce. 3114 */ 3115 int 3116 ire_add(ire_t **irep, queue_t *q, mblk_t *mp, ipsq_func_t func, 3117 boolean_t allow_unresolved) 3118 { 3119 ire_t *ire1; 3120 ill_t *stq_ill = NULL; 3121 ill_t *ill; 3122 ipif_t *ipif = NULL; 3123 ill_walk_context_t ctx; 3124 ire_t *ire = *irep; 3125 int error; 3126 boolean_t ire_is_mblk = B_FALSE; 3127 tsol_gcgrp_t *gcgrp = NULL; 3128 tsol_gcgrp_addr_t ga; 3129 3130 ASSERT(ire->ire_type != IRE_MIPRTUN); 3131 3132 /* get ready for the day when original ire is not created as mblk */ 3133 if (ire->ire_mp != NULL) { 3134 ire_is_mblk = B_TRUE; 3135 /* Copy the ire to a kmem_alloc'ed area */ 3136 ire1 = kmem_cache_alloc(ire_cache, KM_NOSLEEP); 3137 if (ire1 == NULL) { 3138 ip1dbg(("ire_add: alloc failed\n")); 3139 ire_delete(ire); 3140 *irep = NULL; 3141 return (ENOMEM); 3142 } 3143 ire->ire_marks &= ~IRE_MARK_UNCACHED; 3144 *ire1 = *ire; 3145 ire1->ire_mp = NULL; 3146 ire1->ire_stq_ifindex = 0; 3147 freeb(ire->ire_mp); 3148 ire = ire1; 3149 } 3150 if (ire->ire_stq != NULL) 3151 stq_ill = (ill_t *)ire->ire_stq->q_ptr; 3152 3153 if (ire->ire_type == IRE_CACHE) { 3154 /* 3155 * If this interface is FAILED, or INACTIVE or has hit 3156 * the FAILBACK=no case, we create IRE_CACHES marked 3157 * HIDDEN for some special cases e.g. bind to 3158 * IPIF_NOFAILOVER address etc. So, if this interface 3159 * is FAILED/INACTIVE/hit FAILBACK=no case, and we are 3160 * not creating hidden ires, we should not allow that. 3161 * This happens because the state of the interface 3162 * changed while we were waiting in ARP. If this is the 3163 * daemon sending probes, the next probe will create 3164 * HIDDEN ires and we will create an ire then. This 3165 * cannot happen with NDP currently because IRE is 3166 * never queued in NDP. But it can happen in the 3167 * future when we have external resolvers with IPv6. 3168 * If the interface gets marked with OFFLINE while we 3169 * are waiting in ARP, don't add the ire. 3170 */ 3171 if ((stq_ill->ill_phyint->phyint_flags & PHYI_OFFLINE) || 3172 (ill_is_probeonly(stq_ill) && 3173 !(ire->ire_marks & IRE_MARK_HIDDEN))) { 3174 /* 3175 * We don't know whether it is a valid ipif or not. 3176 * unless we do the check below. So, set it to NULL. 3177 */ 3178 ire->ire_ipif = NULL; 3179 ire_delete(ire); 3180 *irep = NULL; 3181 return (EINVAL); 3182 } 3183 } 3184 3185 if (stq_ill != NULL && ire->ire_type == IRE_CACHE && 3186 stq_ill->ill_net_type == IRE_IF_RESOLVER) { 3187 rw_enter(&ill_g_lock, RW_READER); 3188 ill = ILL_START_WALK_ALL(&ctx); 3189 for (; ill != NULL; ill = ill_next(&ctx, ill)) { 3190 mutex_enter(&ill->ill_lock); 3191 if (ill->ill_state_flags & ILL_CONDEMNED) { 3192 mutex_exit(&ill->ill_lock); 3193 continue; 3194 } 3195 /* 3196 * We need to make sure that the ipif is a valid one 3197 * before adding the IRE_CACHE. This happens only 3198 * with IRE_CACHE when there is an external resolver. 3199 * 3200 * We can unplumb a logical interface while the 3201 * packet is waiting in ARP with the IRE. Then, 3202 * later on when we feed the IRE back, the ipif 3203 * has to be re-checked. This can't happen with 3204 * NDP currently, as we never queue the IRE with 3205 * the packet. We always try to recreate the IRE 3206 * when the resolution is completed. But, we do 3207 * it for IPv6 also here so that in future if 3208 * we have external resolvers, it will work without 3209 * any change. 3210 */ 3211 ipif = ipif_lookup_seqid(ill, ire->ire_ipif_seqid); 3212 if (ipif != NULL) { 3213 ipif_refhold_locked(ipif); 3214 mutex_exit(&ill->ill_lock); 3215 break; 3216 } 3217 mutex_exit(&ill->ill_lock); 3218 } 3219 rw_exit(&ill_g_lock); 3220 if (ipif == NULL || 3221 (ipif->ipif_isv6 && 3222 !IN6_ARE_ADDR_EQUAL(&ire->ire_src_addr_v6, 3223 &ipif->ipif_v6src_addr)) || 3224 (!ipif->ipif_isv6 && 3225 ire->ire_src_addr != ipif->ipif_src_addr) || 3226 ire->ire_zoneid != ipif->ipif_zoneid) { 3227 3228 if (ipif != NULL) 3229 ipif_refrele(ipif); 3230 ire->ire_ipif = NULL; 3231 ire_delete(ire); 3232 *irep = NULL; 3233 return (EINVAL); 3234 } 3235 3236 3237 ASSERT(ill != NULL); 3238 /* 3239 * If this group was dismantled while this packets was 3240 * queued in ARP, don't add it here. 3241 */ 3242 if (ire->ire_ipif->ipif_ill->ill_group != ill->ill_group) { 3243 /* We don't want ire_inactive bump stats for this */ 3244 ipif_refrele(ipif); 3245 ire->ire_ipif = NULL; 3246 ire_delete(ire); 3247 *irep = NULL; 3248 return (EINVAL); 3249 } 3250 3251 /* 3252 * Since we didn't attach label security attributes to the 3253 * ire for the resolver case, we need to add it now. (only 3254 * for v4 resolver and v6 xresolv case). 3255 */ 3256 if (is_system_labeled() && ire_is_mblk) { 3257 if (ire->ire_ipversion == IPV4_VERSION) { 3258 ga.ga_af = AF_INET; 3259 IN6_IPADDR_TO_V4MAPPED(ire->ire_gateway_addr != 3260 INADDR_ANY ? ire->ire_gateway_addr : 3261 ire->ire_addr, &ga.ga_addr); 3262 } else { 3263 ga.ga_af = AF_INET6; 3264 ga.ga_addr = IN6_IS_ADDR_UNSPECIFIED( 3265 &ire->ire_gateway_addr_v6) ? 3266 ire->ire_addr_v6 : 3267 ire->ire_gateway_addr_v6; 3268 } 3269 gcgrp = gcgrp_lookup(&ga, B_FALSE); 3270 error = tsol_ire_init_gwattr(ire, ire->ire_ipversion, 3271 NULL, gcgrp); 3272 if (error != 0) { 3273 if (gcgrp != NULL) { 3274 GCGRP_REFRELE(gcgrp); 3275 gcgrp = NULL; 3276 } 3277 ipif_refrele(ipif); 3278 ire->ire_ipif = NULL; 3279 ire_delete(ire); 3280 *irep = NULL; 3281 return (error); 3282 } 3283 } 3284 } 3285 3286 /* 3287 * In case ire was changed 3288 */ 3289 *irep = ire; 3290 if (ire->ire_ipversion == IPV6_VERSION) { 3291 error = ire_add_v6(irep, q, mp, func); 3292 } else { 3293 if (ire->ire_in_ill == NULL) 3294 error = ire_add_v4(irep, q, mp, func, allow_unresolved); 3295 else 3296 error = ire_add_srcif_v4(irep, q, mp, func); 3297 } 3298 if (ipif != NULL) 3299 ipif_refrele(ipif); 3300 return (error); 3301 } 3302 3303 /* 3304 * Add an initialized IRE to an appropriate table based on ire_type. 3305 * 3306 * The forward table contains IRE_PREFIX/IRE_HOST/IRE_HOST_REDIRECT 3307 * IRE_IF_RESOLVER/IRE_IF_NORESOLVER and IRE_DEFAULT. 3308 * 3309 * The cache table contains IRE_BROADCAST/IRE_LOCAL/IRE_LOOPBACK 3310 * and IRE_CACHE. 3311 * 3312 * NOTE : This function is called as writer though not required 3313 * by this function. 3314 */ 3315 static int 3316 ire_add_v4(ire_t **ire_p, queue_t *q, mblk_t *mp, ipsq_func_t func, 3317 boolean_t allow_unresolved) 3318 { 3319 ire_t *ire1; 3320 irb_t *irb_ptr; 3321 ire_t **irep; 3322 int flags; 3323 ire_t *pire = NULL; 3324 ill_t *stq_ill; 3325 ire_t *ire = *ire_p; 3326 int error; 3327 boolean_t need_refrele = B_FALSE; 3328 nce_t *nce; 3329 3330 if (ire->ire_ipif != NULL) 3331 ASSERT(!MUTEX_HELD(&ire->ire_ipif->ipif_ill->ill_lock)); 3332 if (ire->ire_stq != NULL) 3333 ASSERT(!MUTEX_HELD( 3334 &((ill_t *)(ire->ire_stq->q_ptr))->ill_lock)); 3335 ASSERT(ire->ire_ipversion == IPV4_VERSION); 3336 ASSERT(ire->ire_mp == NULL); /* Calls should go through ire_add */ 3337 ASSERT(ire->ire_in_ill == NULL); /* No srcif entries */ 3338 3339 /* Find the appropriate list head. */ 3340 switch (ire->ire_type) { 3341 case IRE_HOST: 3342 ire->ire_mask = IP_HOST_MASK; 3343 ire->ire_masklen = IP_ABITS; 3344 if ((ire->ire_flags & RTF_SETSRC) == 0) 3345 ire->ire_src_addr = 0; 3346 break; 3347 case IRE_HOST_REDIRECT: 3348 ire->ire_mask = IP_HOST_MASK; 3349 ire->ire_masklen = IP_ABITS; 3350 ire->ire_src_addr = 0; 3351 break; 3352 case IRE_CACHE: 3353 case IRE_BROADCAST: 3354 case IRE_LOCAL: 3355 case IRE_LOOPBACK: 3356 ire->ire_mask = IP_HOST_MASK; 3357 ire->ire_masklen = IP_ABITS; 3358 break; 3359 case IRE_PREFIX: 3360 if ((ire->ire_flags & RTF_SETSRC) == 0) 3361 ire->ire_src_addr = 0; 3362 break; 3363 case IRE_DEFAULT: 3364 if ((ire->ire_flags & RTF_SETSRC) == 0) 3365 ire->ire_src_addr = 0; 3366 break; 3367 case IRE_IF_RESOLVER: 3368 case IRE_IF_NORESOLVER: 3369 break; 3370 default: 3371 ip0dbg(("ire_add_v4: ire %p has unrecognized IRE type (%d)\n", 3372 (void *)ire, ire->ire_type)); 3373 ire_delete(ire); 3374 *ire_p = NULL; 3375 return (EINVAL); 3376 } 3377 3378 /* Make sure the address is properly masked. */ 3379 ire->ire_addr &= ire->ire_mask; 3380 3381 /* 3382 * ip_newroute/ip_newroute_multi are unable to prevent the deletion 3383 * of the interface route while adding an IRE_CACHE for an on-link 3384 * destination in the IRE_IF_RESOLVER case, since the ire has to 3385 * go to ARP and return. We can't do a REFHOLD on the 3386 * associated interface ire for fear of ARP freeing the message. 3387 * Here we look up the interface ire in the forwarding table and 3388 * make sure that the interface route has not been deleted. 3389 */ 3390 if (ire->ire_type == IRE_CACHE && ire->ire_gateway_addr == 0 && 3391 ((ill_t *)ire->ire_stq->q_ptr)->ill_net_type == IRE_IF_RESOLVER) { 3392 3393 ASSERT(ire->ire_max_fragp == NULL); 3394 if (CLASSD(ire->ire_addr) && !(ire->ire_flags & RTF_SETSRC)) { 3395 /* 3396 * The ihandle that we used in ip_newroute_multi 3397 * comes from the interface route corresponding 3398 * to ire_ipif. Lookup here to see if it exists 3399 * still. 3400 * If the ire has a source address assigned using 3401 * RTF_SETSRC, ire_ipif is the logical interface holding 3402 * this source address, so we can't use it to check for 3403 * the existence of the interface route. Instead we rely 3404 * on the brute force ihandle search in 3405 * ire_ihandle_lookup_onlink() below. 3406 */ 3407 pire = ipif_to_ire(ire->ire_ipif); 3408 if (pire == NULL) { 3409 ire_delete(ire); 3410 *ire_p = NULL; 3411 return (EINVAL); 3412 } else if (pire->ire_ihandle != ire->ire_ihandle) { 3413 ire_refrele(pire); 3414 ire_delete(ire); 3415 *ire_p = NULL; 3416 return (EINVAL); 3417 } 3418 } else { 3419 pire = ire_ihandle_lookup_onlink(ire); 3420 if (pire == NULL) { 3421 ire_delete(ire); 3422 *ire_p = NULL; 3423 return (EINVAL); 3424 } 3425 } 3426 /* Prevent pire from getting deleted */ 3427 IRB_REFHOLD(pire->ire_bucket); 3428 /* Has it been removed already ? */ 3429 if (pire->ire_marks & IRE_MARK_CONDEMNED) { 3430 IRB_REFRELE(pire->ire_bucket); 3431 ire_refrele(pire); 3432 ire_delete(ire); 3433 *ire_p = NULL; 3434 return (EINVAL); 3435 } 3436 } else { 3437 ASSERT(ire->ire_max_fragp != NULL); 3438 } 3439 flags = (MATCH_IRE_MASK | MATCH_IRE_TYPE | MATCH_IRE_GW); 3440 3441 if (ire->ire_ipif != NULL) { 3442 /* 3443 * We use MATCH_IRE_IPIF while adding IRE_CACHES only 3444 * for historic reasons and to maintain symmetry with 3445 * IPv6 code path. Historically this was used by 3446 * multicast code to create multiple IRE_CACHES on 3447 * a single ill with different ipifs. This was used 3448 * so that multicast packets leaving the node had the 3449 * right source address. This is no longer needed as 3450 * ip_wput initializes the address correctly. 3451 */ 3452 flags |= MATCH_IRE_IPIF; 3453 /* 3454 * If we are creating hidden ires, make sure we search on 3455 * this ill (MATCH_IRE_ILL) and a hidden ire, 3456 * while we are searching for duplicates below. Otherwise we 3457 * could potentially find an IRE on some other interface 3458 * and it may not be a IRE marked with IRE_MARK_HIDDEN. We 3459 * shouldn't do this as this will lead to an infinite loop 3460 * (if we get to ip_wput again) eventually we need an hidden 3461 * ire for this packet to go out. MATCH_IRE_ILL is explicitly 3462 * done below. 3463 */ 3464 if (ire->ire_type == IRE_CACHE && 3465 (ire->ire_marks & IRE_MARK_HIDDEN)) 3466 flags |= (MATCH_IRE_MARK_HIDDEN); 3467 } 3468 if ((ire->ire_type & IRE_CACHETABLE) == 0) { 3469 irb_ptr = ire_get_bucket(ire); 3470 need_refrele = B_TRUE; 3471 if (irb_ptr == NULL) { 3472 /* 3473 * This assumes that the ire has not added 3474 * a reference to the ipif. 3475 */ 3476 ire->ire_ipif = NULL; 3477 ire_delete(ire); 3478 if (pire != NULL) { 3479 IRB_REFRELE(pire->ire_bucket); 3480 ire_refrele(pire); 3481 } 3482 *ire_p = NULL; 3483 return (EINVAL); 3484 } 3485 } else { 3486 irb_ptr = &(ip_cache_table[IRE_ADDR_HASH(ire->ire_addr, 3487 ip_cache_table_size)]); 3488 } 3489 3490 /* 3491 * Start the atomic add of the ire. Grab the ill locks, 3492 * ill_g_usesrc_lock and the bucket lock. Check for condemned 3493 * 3494 * If ipif or ill is changing ire_atomic_start() may queue the 3495 * request and return EINPROGRESS. 3496 * To avoid lock order problems, get the ndp4.ndp_g_lock. 3497 */ 3498 mutex_enter(&ndp4.ndp_g_lock); 3499 error = ire_atomic_start(irb_ptr, ire, q, mp, func); 3500 if (error != 0) { 3501 mutex_exit(&ndp4.ndp_g_lock); 3502 /* 3503 * We don't know whether it is a valid ipif or not. 3504 * So, set it to NULL. This assumes that the ire has not added 3505 * a reference to the ipif. 3506 */ 3507 ire->ire_ipif = NULL; 3508 ire_delete(ire); 3509 if (pire != NULL) { 3510 IRB_REFRELE(pire->ire_bucket); 3511 ire_refrele(pire); 3512 } 3513 *ire_p = NULL; 3514 if (need_refrele) 3515 IRB_REFRELE(irb_ptr); 3516 return (error); 3517 } 3518 /* 3519 * To avoid creating ires having stale values for the ire_max_frag 3520 * we get the latest value atomically here. For more details 3521 * see the block comment in ip_sioctl_mtu and in DL_NOTE_SDU_CHANGE 3522 * in ip_rput_dlpi_writer 3523 */ 3524 if (ire->ire_max_fragp == NULL) { 3525 if (CLASSD(ire->ire_addr)) 3526 ire->ire_max_frag = ire->ire_ipif->ipif_mtu; 3527 else 3528 ire->ire_max_frag = pire->ire_max_frag; 3529 } else { 3530 uint_t max_frag; 3531 3532 max_frag = *ire->ire_max_fragp; 3533 ire->ire_max_fragp = NULL; 3534 ire->ire_max_frag = max_frag; 3535 } 3536 /* 3537 * Atomically check for duplicate and insert in the table. 3538 */ 3539 for (ire1 = irb_ptr->irb_ire; ire1 != NULL; ire1 = ire1->ire_next) { 3540 if (ire1->ire_marks & IRE_MARK_CONDEMNED) 3541 continue; 3542 if (ire->ire_ipif != NULL) { 3543 /* 3544 * We do MATCH_IRE_ILL implicitly here for IREs 3545 * with a non-null ire_ipif, including IRE_CACHEs. 3546 * As ire_ipif and ire_stq could point to two 3547 * different ills, we can't pass just ire_ipif to 3548 * ire_match_args and get a match on both ills. 3549 * This is just needed for duplicate checks here and 3550 * so we don't add an extra argument to 3551 * ire_match_args for this. Do it locally. 3552 * 3553 * NOTE : Currently there is no part of the code 3554 * that asks for both MATH_IRE_IPIF and MATCH_IRE_ILL 3555 * match for IRE_CACHEs. Thus we don't want to 3556 * extend the arguments to ire_match_args. 3557 */ 3558 if (ire1->ire_stq != ire->ire_stq) 3559 continue; 3560 /* 3561 * Multiroute IRE_CACHEs for a given destination can 3562 * have the same ire_ipif, typically if their source 3563 * address is forced using RTF_SETSRC, and the same 3564 * send-to queue. We differentiate them using the parent 3565 * handle. 3566 */ 3567 if (ire->ire_type == IRE_CACHE && 3568 (ire1->ire_flags & RTF_MULTIRT) && 3569 (ire->ire_flags & RTF_MULTIRT) && 3570 (ire1->ire_phandle != ire->ire_phandle)) 3571 continue; 3572 } 3573 if (ire1->ire_zoneid != ire->ire_zoneid) 3574 continue; 3575 if (ire_match_args(ire1, ire->ire_addr, ire->ire_mask, 3576 ire->ire_gateway_addr, ire->ire_type, ire->ire_ipif, 3577 ire->ire_zoneid, 0, NULL, flags)) { 3578 /* 3579 * Return the old ire after doing a REFHOLD. 3580 * As most of the callers continue to use the IRE 3581 * after adding, we return a held ire. This will 3582 * avoid a lookup in the caller again. If the callers 3583 * don't want to use it, they need to do a REFRELE. 3584 */ 3585 ip1dbg(("found dup ire existing %p new %p", 3586 (void *)ire1, (void *)ire)); 3587 IRE_REFHOLD(ire1); 3588 ire_atomic_end(irb_ptr, ire); 3589 mutex_exit(&ndp4.ndp_g_lock); 3590 ire_delete(ire); 3591 if (pire != NULL) { 3592 /* 3593 * Assert that it is not removed from the 3594 * list yet. 3595 */ 3596 ASSERT(pire->ire_ptpn != NULL); 3597 IRB_REFRELE(pire->ire_bucket); 3598 ire_refrele(pire); 3599 } 3600 *ire_p = ire1; 3601 if (need_refrele) 3602 IRB_REFRELE(irb_ptr); 3603 return (0); 3604 } 3605 } 3606 if (ire->ire_type & IRE_CACHE) { 3607 ASSERT(ire->ire_stq != NULL); 3608 nce = ndp_lookup_v4(ire_to_ill(ire), 3609 ((ire->ire_gateway_addr != INADDR_ANY) ? 3610 &ire->ire_gateway_addr : &ire->ire_addr), 3611 B_TRUE); 3612 if (nce != NULL) 3613 mutex_enter(&nce->nce_lock); 3614 /* 3615 * if the nce is NCE_F_CONDEMNED, or if it is not ND_REACHABLE 3616 * and the caller has prohibited the addition of incomplete 3617 * ire's, we fail the add. Note that nce_state could be 3618 * something other than ND_REACHABLE if nce_reinit has just 3619 * kicked in and reset the nce. 3620 */ 3621 if ((nce == NULL) || 3622 (nce->nce_flags & NCE_F_CONDEMNED) || 3623 (!allow_unresolved && 3624 ((nce->nce_state & ND_REACHABLE) == 0))) { 3625 if (nce != NULL) 3626 mutex_exit(&nce->nce_lock); 3627 ire_atomic_end(irb_ptr, ire); 3628 mutex_exit(&ndp4.ndp_g_lock); 3629 if (nce != NULL) 3630 NCE_REFRELE(nce); 3631 DTRACE_PROBE1(ire__no__nce, ire_t *, ire); 3632 ire_delete(ire); 3633 if (pire != NULL) { 3634 IRB_REFRELE(pire->ire_bucket); 3635 ire_refrele(pire); 3636 } 3637 *ire_p = NULL; 3638 if (need_refrele) 3639 IRB_REFRELE(irb_ptr); 3640 return (EINVAL); 3641 } else { 3642 ire->ire_nce = nce; 3643 mutex_exit(&nce->nce_lock); 3644 /* 3645 * We are associating this nce to the ire, so 3646 * change the nce ref taken in ndp_lookup_v4() from 3647 * NCE_REFHOLD to NCE_REFHOLD_NOTR 3648 */ 3649 NCE_REFHOLD_TO_REFHOLD_NOTR(ire->ire_nce); 3650 } 3651 } 3652 /* 3653 * Make it easy for ip_wput_ire() to hit multiple broadcast ires by 3654 * grouping identical addresses together on the hash chain. We also 3655 * don't want to send multiple copies out if there are two ills part 3656 * of the same group. Thus we group the ires with same addr and same 3657 * ill group together so that ip_wput_ire can easily skip all the 3658 * ires with same addr and same group after sending the first copy. 3659 * We do this only for IRE_BROADCASTs as ip_wput_ire is currently 3660 * interested in such groupings only for broadcasts. 3661 * 3662 * NOTE : If the interfaces are brought up first and then grouped, 3663 * illgrp_insert will handle it. We come here when the interfaces 3664 * are already in group and we are bringing them UP. 3665 * 3666 * Find the first entry that matches ire_addr. *irep will be null 3667 * if no match. 3668 */ 3669 irep = (ire_t **)irb_ptr; 3670 while ((ire1 = *irep) != NULL && ire->ire_addr != ire1->ire_addr) 3671 irep = &ire1->ire_next; 3672 if (ire->ire_type == IRE_BROADCAST && *irep != NULL) { 3673 /* 3674 * We found some ire (i.e *irep) with a matching addr. We 3675 * want to group ires with same addr and same ill group 3676 * together. 3677 * 3678 * First get to the entry that matches our address and 3679 * ill group i.e stop as soon as we find the first ire 3680 * matching the ill group and address. If there is only 3681 * an address match, we should walk and look for some 3682 * group match. These are some of the possible scenarios : 3683 * 3684 * 1) There are no groups at all i.e all ire's ill_group 3685 * are NULL. In that case we will essentially group 3686 * all the ires with the same addr together. Same as 3687 * the "else" block of this "if". 3688 * 3689 * 2) There are some groups and this ire's ill_group is 3690 * NULL. In this case, we will first find the group 3691 * that matches the address and a NULL group. Then 3692 * we will insert the ire at the end of that group. 3693 * 3694 * 3) There are some groups and this ires's ill_group is 3695 * non-NULL. In this case we will first find the group 3696 * that matches the address and the ill_group. Then 3697 * we will insert the ire at the end of that group. 3698 */ 3699 /* LINTED : constant in conditional context */ 3700 while (1) { 3701 ire1 = *irep; 3702 if ((ire1->ire_next == NULL) || 3703 (ire1->ire_next->ire_addr != ire->ire_addr) || 3704 (ire1->ire_type != IRE_BROADCAST) || 3705 (ire1->ire_ipif->ipif_ill->ill_group == 3706 ire->ire_ipif->ipif_ill->ill_group)) 3707 break; 3708 irep = &ire1->ire_next; 3709 } 3710 ASSERT(*irep != NULL); 3711 irep = &((*irep)->ire_next); 3712 3713 /* 3714 * Either we have hit the end of the list or the address 3715 * did not match or the group *matched*. If we found 3716 * a match on the group, skip to the end of the group. 3717 */ 3718 while (*irep != NULL) { 3719 ire1 = *irep; 3720 if ((ire1->ire_addr != ire->ire_addr) || 3721 (ire1->ire_type != IRE_BROADCAST) || 3722 (ire1->ire_ipif->ipif_ill->ill_group != 3723 ire->ire_ipif->ipif_ill->ill_group)) 3724 break; 3725 if (ire1->ire_ipif->ipif_ill->ill_group == NULL && 3726 ire1->ire_ipif == ire->ire_ipif) { 3727 irep = &ire1->ire_next; 3728 break; 3729 } 3730 irep = &ire1->ire_next; 3731 } 3732 } else if (*irep != NULL) { 3733 /* 3734 * Find the last ire which matches ire_addr. 3735 * Needed to do tail insertion among entries with the same 3736 * ire_addr. 3737 */ 3738 while (ire->ire_addr == ire1->ire_addr) { 3739 irep = &ire1->ire_next; 3740 ire1 = *irep; 3741 if (ire1 == NULL) 3742 break; 3743 } 3744 } 3745 3746 /* Insert at *irep */ 3747 ire1 = *irep; 3748 if (ire1 != NULL) 3749 ire1->ire_ptpn = &ire->ire_next; 3750 ire->ire_next = ire1; 3751 /* Link the new one in. */ 3752 ire->ire_ptpn = irep; 3753 3754 /* 3755 * ire_walk routines de-reference ire_next without holding 3756 * a lock. Before we point to the new ire, we want to make 3757 * sure the store that sets the ire_next of the new ire 3758 * reaches global visibility, so that ire_walk routines 3759 * don't see a truncated list of ires i.e if the ire_next 3760 * of the new ire gets set after we do "*irep = ire" due 3761 * to re-ordering, the ire_walk thread will see a NULL 3762 * once it accesses the ire_next of the new ire. 3763 * membar_producer() makes sure that the following store 3764 * happens *after* all of the above stores. 3765 */ 3766 membar_producer(); 3767 *irep = ire; 3768 ire->ire_bucket = irb_ptr; 3769 /* 3770 * We return a bumped up IRE above. Keep it symmetrical 3771 * so that the callers will always have to release. This 3772 * helps the callers of this function because they continue 3773 * to use the IRE after adding and hence they don't have to 3774 * lookup again after we return the IRE. 3775 * 3776 * NOTE : We don't have to use atomics as this is appearing 3777 * in the list for the first time and no one else can bump 3778 * up the reference count on this yet. 3779 */ 3780 IRE_REFHOLD_LOCKED(ire); 3781 BUMP_IRE_STATS(ire_stats_v4, ire_stats_inserted); 3782 3783 irb_ptr->irb_ire_cnt++; 3784 if (irb_ptr->irb_marks & IRB_MARK_FTABLE) 3785 irb_ptr->irb_nire++; 3786 3787 if (ire->ire_marks & IRE_MARK_TEMPORARY) 3788 irb_ptr->irb_tmp_ire_cnt++; 3789 3790 if (ire->ire_ipif != NULL) { 3791 ire->ire_ipif->ipif_ire_cnt++; 3792 if (ire->ire_stq != NULL) { 3793 stq_ill = (ill_t *)ire->ire_stq->q_ptr; 3794 stq_ill->ill_ire_cnt++; 3795 } 3796 } else { 3797 ASSERT(ire->ire_stq == NULL); 3798 } 3799 3800 ire_atomic_end(irb_ptr, ire); 3801 mutex_exit(&ndp4.ndp_g_lock); 3802 3803 if (pire != NULL) { 3804 /* Assert that it is not removed from the list yet */ 3805 ASSERT(pire->ire_ptpn != NULL); 3806 IRB_REFRELE(pire->ire_bucket); 3807 ire_refrele(pire); 3808 } 3809 3810 if (ire->ire_type != IRE_CACHE) { 3811 /* 3812 * For ire's with host mask see if there is an entry 3813 * in the cache. If there is one flush the whole cache as 3814 * there might be multiple entries due to RTF_MULTIRT (CGTP). 3815 * If no entry is found than there is no need to flush the 3816 * cache. 3817 */ 3818 if (ire->ire_mask == IP_HOST_MASK) { 3819 ire_t *lire; 3820 lire = ire_ctable_lookup(ire->ire_addr, NULL, IRE_CACHE, 3821 NULL, ALL_ZONES, NULL, MATCH_IRE_TYPE); 3822 if (lire != NULL) { 3823 ire_refrele(lire); 3824 ire_flush_cache_v4(ire, IRE_FLUSH_ADD); 3825 } 3826 } else { 3827 ire_flush_cache_v4(ire, IRE_FLUSH_ADD); 3828 } 3829 } 3830 /* 3831 * We had to delay the fast path probe until the ire is inserted 3832 * in the list. Otherwise the fast path ack won't find the ire in 3833 * the table. 3834 */ 3835 if (ire->ire_type == IRE_CACHE || ire->ire_type == IRE_BROADCAST) 3836 ire_fastpath(ire); 3837 if (ire->ire_ipif != NULL) 3838 ASSERT(!MUTEX_HELD(&ire->ire_ipif->ipif_ill->ill_lock)); 3839 *ire_p = ire; 3840 if (need_refrele) { 3841 IRB_REFRELE(irb_ptr); 3842 } 3843 return (0); 3844 } 3845 3846 /* 3847 * IRB_REFRELE is the only caller of the function. ire_unlink calls to 3848 * do the final cleanup for this ire. 3849 */ 3850 void 3851 ire_cleanup(ire_t *ire) 3852 { 3853 ire_t *ire_next; 3854 3855 ASSERT(ire != NULL); 3856 3857 while (ire != NULL) { 3858 ire_next = ire->ire_next; 3859 if (ire->ire_ipversion == IPV4_VERSION) { 3860 ire_delete_v4(ire); 3861 BUMP_IRE_STATS(ire_stats_v4, ire_stats_deleted); 3862 } else { 3863 ASSERT(ire->ire_ipversion == IPV6_VERSION); 3864 ire_delete_v6(ire); 3865 BUMP_IRE_STATS(ire_stats_v6, ire_stats_deleted); 3866 } 3867 /* 3868 * Now it's really out of the list. Before doing the 3869 * REFRELE, set ire_next to NULL as ire_inactive asserts 3870 * so. 3871 */ 3872 ire->ire_next = NULL; 3873 IRE_REFRELE_NOTR(ire); 3874 ire = ire_next; 3875 } 3876 } 3877 3878 /* 3879 * IRB_REFRELE is the only caller of the function. It calls to unlink 3880 * all the CONDEMNED ires from this bucket. 3881 */ 3882 ire_t * 3883 ire_unlink(irb_t *irb) 3884 { 3885 ire_t *ire; 3886 ire_t *ire1; 3887 ire_t **ptpn; 3888 ire_t *ire_list = NULL; 3889 3890 ASSERT(RW_WRITE_HELD(&irb->irb_lock)); 3891 ASSERT(((irb->irb_marks & IRB_MARK_FTABLE) && irb->irb_refcnt == 1) || 3892 (irb->irb_refcnt == 0)); 3893 ASSERT(irb->irb_marks & IRB_MARK_CONDEMNED); 3894 ASSERT(irb->irb_ire != NULL); 3895 3896 for (ire = irb->irb_ire; ire != NULL; ire = ire1) { 3897 ire1 = ire->ire_next; 3898 if (ire->ire_marks & IRE_MARK_CONDEMNED) { 3899 ptpn = ire->ire_ptpn; 3900 ire1 = ire->ire_next; 3901 if (ire1) 3902 ire1->ire_ptpn = ptpn; 3903 *ptpn = ire1; 3904 ire->ire_ptpn = NULL; 3905 ire->ire_next = NULL; 3906 if (ire->ire_type == IRE_DEFAULT) { 3907 /* 3908 * IRE is out of the list. We need to adjust 3909 * the accounting before the caller drops 3910 * the lock. 3911 */ 3912 if (ire->ire_ipversion == IPV6_VERSION) { 3913 ASSERT(ipv6_ire_default_count != 0); 3914 ipv6_ire_default_count--; 3915 } 3916 } 3917 /* 3918 * We need to call ire_delete_v4 or ire_delete_v6 3919 * to clean up the cache or the redirects pointing at 3920 * the default gateway. We need to drop the lock 3921 * as ire_flush_cache/ire_delete_host_redircts require 3922 * so. But we can't drop the lock, as ire_unlink needs 3923 * to atomically remove the ires from the list. 3924 * So, create a temporary list of CONDEMNED ires 3925 * for doing ire_delete_v4/ire_delete_v6 operations 3926 * later on. 3927 */ 3928 ire->ire_next = ire_list; 3929 ire_list = ire; 3930 } 3931 } 3932 irb->irb_marks &= ~IRB_MARK_CONDEMNED; 3933 return (ire_list); 3934 } 3935 3936 /* 3937 * Delete all the cache entries with this 'addr'. When IP gets a gratuitous 3938 * ARP message on any of its interface queue, it scans the nce table and 3939 * deletes and calls ndp_delete() for the appropriate nce. This action 3940 * also deletes all the neighbor/ire cache entries for that address. 3941 * This function is called from ip_arp_news in ip.c and also for 3942 * ARP ioctl processing in ip_if.c. ip_ire_clookup_and_delete returns 3943 * true if it finds a nce entry which is used by ip_arp_news to determine if 3944 * it needs to do an ire_walk_v4. The return value is also used for the 3945 * same purpose by ARP IOCTL processing * in ip_if.c when deleting 3946 * ARP entries. For SIOC*IFARP ioctls in addition to the address, 3947 * ip_if->ipif_ill also needs to be matched. 3948 */ 3949 boolean_t 3950 ip_ire_clookup_and_delete(ipaddr_t addr, ipif_t *ipif) 3951 { 3952 ill_t *ill; 3953 nce_t *nce; 3954 3955 ill = (ipif ? ipif->ipif_ill : NULL); 3956 3957 if (ill != NULL) { 3958 /* 3959 * clean up the nce (and any relevant ire's) that matches 3960 * on addr and ill. 3961 */ 3962 nce = ndp_lookup_v4(ill, &addr, B_FALSE); 3963 if (nce != NULL) { 3964 ndp_delete(nce); 3965 return (B_TRUE); 3966 } 3967 } else { 3968 /* 3969 * ill is wildcard. clean up all nce's and 3970 * ire's that match on addr 3971 */ 3972 nce_clookup_t cl; 3973 3974 cl.ncecl_addr = addr; 3975 cl.ncecl_found = B_FALSE; 3976 3977 ndp_walk_common(&ndp4, NULL, 3978 (pfi_t)ip_nce_clookup_and_delete, (uchar_t *)&cl, B_TRUE); 3979 3980 /* 3981 * ncecl_found would be set by ip_nce_clookup_and_delete if 3982 * we found a matching nce. 3983 */ 3984 return (cl.ncecl_found); 3985 } 3986 return (B_FALSE); 3987 3988 } 3989 3990 /* Delete the supplied nce if its nce_addr matches the supplied address */ 3991 static void 3992 ip_nce_clookup_and_delete(nce_t *nce, void *arg) 3993 { 3994 nce_clookup_t *cl = (nce_clookup_t *)arg; 3995 ipaddr_t nce_addr; 3996 3997 IN6_V4MAPPED_TO_IPADDR(&nce->nce_addr, nce_addr); 3998 if (nce_addr == cl->ncecl_addr) { 3999 cl->ncecl_found = B_TRUE; 4000 /* clean up the nce (and any relevant ire's) */ 4001 ndp_delete(nce); 4002 } 4003 } 4004 4005 /* 4006 * Clean up the radix node for this ire. Must be called by IRB_REFRELE 4007 * when there are no ire's left in the bucket. Returns TRUE if the bucket 4008 * is deleted and freed. 4009 */ 4010 boolean_t 4011 irb_inactive(irb_t *irb) 4012 { 4013 struct rt_entry *rt; 4014 struct radix_node *rn; 4015 4016 rt = IRB2RT(irb); 4017 rn = (struct radix_node *)rt; 4018 4019 /* first remove it from the radix tree. */ 4020 RADIX_NODE_HEAD_WLOCK(ip_ftable); 4021 rw_enter(&irb->irb_lock, RW_WRITER); 4022 if (irb->irb_refcnt == 1 && irb->irb_nire == 0) { 4023 rn = ip_ftable->rnh_deladdr(rn->rn_key, rn->rn_mask, 4024 ip_ftable); 4025 DTRACE_PROBE1(irb__free, rt_t *, rt); 4026 ASSERT((void *)rn == (void *)rt); 4027 Free(rt, rt_entry_cache); 4028 /* irb_lock is freed */ 4029 RADIX_NODE_HEAD_UNLOCK(ip_ftable); 4030 return (B_TRUE); 4031 } 4032 rw_exit(&irb->irb_lock); 4033 RADIX_NODE_HEAD_UNLOCK(ip_ftable); 4034 return (B_FALSE); 4035 } 4036 4037 /* 4038 * Delete the specified IRE. 4039 */ 4040 void 4041 ire_delete(ire_t *ire) 4042 { 4043 ire_t *ire1; 4044 ire_t **ptpn; 4045 irb_t *irb; 4046 4047 if ((irb = ire->ire_bucket) == NULL) { 4048 /* 4049 * It was never inserted in the list. Should call REFRELE 4050 * to free this IRE. 4051 */ 4052 IRE_REFRELE_NOTR(ire); 4053 return; 4054 } 4055 4056 rw_enter(&irb->irb_lock, RW_WRITER); 4057 4058 if (irb->irb_rr_origin == ire) { 4059 irb->irb_rr_origin = NULL; 4060 } 4061 4062 /* 4063 * In case of V4 we might still be waiting for fastpath ack. 4064 */ 4065 if (ire->ire_ipversion == IPV4_VERSION && ire->ire_stq != NULL) { 4066 ill_t *ill; 4067 4068 ill = ire_to_ill(ire); 4069 if (ill != NULL) 4070 ire_fastpath_list_delete(ill, ire); 4071 } 4072 4073 if (ire->ire_ptpn == NULL) { 4074 /* 4075 * Some other thread has removed us from the list. 4076 * It should have done the REFRELE for us. 4077 */ 4078 rw_exit(&irb->irb_lock); 4079 return; 4080 } 4081 4082 if (irb->irb_refcnt != 0) { 4083 /* 4084 * The last thread to leave this bucket will 4085 * delete this ire. 4086 */ 4087 if (!(ire->ire_marks & IRE_MARK_CONDEMNED)) { 4088 irb->irb_ire_cnt--; 4089 if (ire->ire_marks & IRE_MARK_TEMPORARY) 4090 irb->irb_tmp_ire_cnt--; 4091 ire->ire_marks |= IRE_MARK_CONDEMNED; 4092 } 4093 irb->irb_marks |= IRB_MARK_CONDEMNED; 4094 rw_exit(&irb->irb_lock); 4095 return; 4096 } 4097 4098 /* 4099 * Normally to delete an ire, we walk the bucket. While we 4100 * walk the bucket, we normally bump up irb_refcnt and hence 4101 * we return from above where we mark CONDEMNED and the ire 4102 * gets deleted from ire_unlink. This case is where somebody 4103 * knows the ire e.g by doing a lookup, and wants to delete the 4104 * IRE. irb_refcnt would be 0 in this case if nobody is walking 4105 * the bucket. 4106 */ 4107 ptpn = ire->ire_ptpn; 4108 ire1 = ire->ire_next; 4109 if (ire1 != NULL) 4110 ire1->ire_ptpn = ptpn; 4111 ASSERT(ptpn != NULL); 4112 *ptpn = ire1; 4113 ire->ire_ptpn = NULL; 4114 ire->ire_next = NULL; 4115 if (ire->ire_ipversion == IPV6_VERSION) { 4116 BUMP_IRE_STATS(ire_stats_v6, ire_stats_deleted); 4117 } else { 4118 BUMP_IRE_STATS(ire_stats_v4, ire_stats_deleted); 4119 } 4120 /* 4121 * ip_wput/ip_wput_v6 checks this flag to see whether 4122 * it should still use the cached ire or not. 4123 */ 4124 ire->ire_marks |= IRE_MARK_CONDEMNED; 4125 if (ire->ire_type == IRE_DEFAULT) { 4126 /* 4127 * IRE is out of the list. We need to adjust the 4128 * accounting before we drop the lock. 4129 */ 4130 if (ire->ire_ipversion == IPV6_VERSION) { 4131 ASSERT(ipv6_ire_default_count != 0); 4132 ipv6_ire_default_count--; 4133 } 4134 } 4135 irb->irb_ire_cnt--; 4136 4137 if (ire->ire_marks & IRE_MARK_TEMPORARY) 4138 irb->irb_tmp_ire_cnt--; 4139 rw_exit(&irb->irb_lock); 4140 4141 if (ire->ire_ipversion == IPV6_VERSION) { 4142 ire_delete_v6(ire); 4143 } else { 4144 ire_delete_v4(ire); 4145 } 4146 /* 4147 * We removed it from the list. Decrement the 4148 * reference count. 4149 */ 4150 IRE_REFRELE_NOTR(ire); 4151 } 4152 4153 /* 4154 * Delete the specified IRE. 4155 * All calls should use ire_delete(). 4156 * Sometimes called as writer though not required by this function. 4157 * 4158 * NOTE : This function is called only if the ire was added 4159 * in the list. 4160 */ 4161 static void 4162 ire_delete_v4(ire_t *ire) 4163 { 4164 ASSERT(ire->ire_refcnt >= 1); 4165 ASSERT(ire->ire_ipversion == IPV4_VERSION); 4166 4167 if (ire->ire_type != IRE_CACHE) 4168 ire_flush_cache_v4(ire, IRE_FLUSH_DELETE); 4169 if (ire->ire_type == IRE_DEFAULT) { 4170 /* 4171 * when a default gateway is going away 4172 * delete all the host redirects pointing at that 4173 * gateway. 4174 */ 4175 ire_delete_host_redirects(ire->ire_gateway_addr); 4176 } 4177 } 4178 4179 /* 4180 * IRE_REFRELE/ire_refrele are the only caller of the function. It calls 4181 * to free the ire when the reference count goes to zero. 4182 */ 4183 void 4184 ire_inactive(ire_t *ire) 4185 { 4186 nce_t *nce; 4187 ill_t *ill = NULL; 4188 ill_t *stq_ill = NULL; 4189 ill_t *in_ill = NULL; 4190 ipif_t *ipif; 4191 boolean_t need_wakeup = B_FALSE; 4192 irb_t *irb; 4193 4194 ASSERT(ire->ire_refcnt == 0); 4195 ASSERT(ire->ire_ptpn == NULL); 4196 ASSERT(ire->ire_next == NULL); 4197 4198 if (ire->ire_gw_secattr != NULL) { 4199 ire_gw_secattr_free(ire->ire_gw_secattr); 4200 ire->ire_gw_secattr = NULL; 4201 } 4202 4203 if (ire->ire_mp != NULL) { 4204 ASSERT(ire->ire_fastpath == NULL); 4205 ASSERT(ire->ire_bucket == NULL); 4206 mutex_destroy(&ire->ire_lock); 4207 BUMP_IRE_STATS(ire_stats_v4, ire_stats_freed); 4208 if (ire->ire_nce != NULL) 4209 NCE_REFRELE_NOTR(ire->ire_nce); 4210 freeb(ire->ire_mp); 4211 return; 4212 } 4213 4214 if ((nce = ire->ire_nce) != NULL) { 4215 NCE_REFRELE_NOTR(nce); 4216 ire->ire_nce = NULL; 4217 } 4218 4219 if (ire->ire_ipif == NULL) 4220 goto end; 4221 4222 ipif = ire->ire_ipif; 4223 ill = ipif->ipif_ill; 4224 4225 if (ire->ire_bucket == NULL) { 4226 /* The ire was never inserted in the table. */ 4227 goto end; 4228 } 4229 4230 /* 4231 * ipif_ire_cnt on this ipif goes down by 1. If the ire_stq is 4232 * non-null ill_ire_count also goes down by 1. If the in_ill is 4233 * non-null either ill_mrtun_refcnt or ill_srcif_refcnt goes down by 1. 4234 * 4235 * The ipif that is associated with an ire is ire->ire_ipif and 4236 * hence when the ire->ire_ipif->ipif_ire_cnt drops to zero we call 4237 * ipif_ill_refrele_tail. Usually stq_ill is null or the same as 4238 * ire->ire_ipif->ipif_ill. So nothing more needs to be done. Only 4239 * in the case of IRE_CACHES when IPMP is used, stq_ill can be 4240 * different. If this is different from ire->ire_ipif->ipif_ill and 4241 * if the ill_ire_cnt on the stq_ill also has dropped to zero, we call 4242 * ipif_ill_refrele_tail on the stq_ill. If mobile ip is in use 4243 * in_ill could be non-null. If it is a reverse tunnel related ire 4244 * ill_mrtun_refcnt is non-zero. If it is forward tunnel related ire 4245 * ill_srcif_refcnt is non-null. 4246 */ 4247 4248 if (ire->ire_stq != NULL) 4249 stq_ill = (ill_t *)ire->ire_stq->q_ptr; 4250 if (ire->ire_in_ill != NULL) 4251 in_ill = ire->ire_in_ill; 4252 4253 if ((stq_ill == NULL || stq_ill == ill) && (in_ill == NULL)) { 4254 /* Optimize the most common case */ 4255 mutex_enter(&ill->ill_lock); 4256 ASSERT(ipif->ipif_ire_cnt != 0); 4257 ipif->ipif_ire_cnt--; 4258 if (ipif->ipif_ire_cnt == 0) 4259 need_wakeup = B_TRUE; 4260 if (stq_ill != NULL) { 4261 ASSERT(stq_ill->ill_ire_cnt != 0); 4262 stq_ill->ill_ire_cnt--; 4263 if (stq_ill->ill_ire_cnt == 0) 4264 need_wakeup = B_TRUE; 4265 } 4266 if (need_wakeup) { 4267 /* Drops the ill lock */ 4268 ipif_ill_refrele_tail(ill); 4269 } else { 4270 mutex_exit(&ill->ill_lock); 4271 } 4272 } else { 4273 /* 4274 * We can't grab all the ill locks at the same time. 4275 * It can lead to recursive lock enter in the call to 4276 * ipif_ill_refrele_tail and later. Instead do it 1 at 4277 * a time. 4278 */ 4279 mutex_enter(&ill->ill_lock); 4280 ASSERT(ipif->ipif_ire_cnt != 0); 4281 ipif->ipif_ire_cnt--; 4282 if (ipif->ipif_ire_cnt == 0) { 4283 /* Drops the lock */ 4284 ipif_ill_refrele_tail(ill); 4285 } else { 4286 mutex_exit(&ill->ill_lock); 4287 } 4288 if (stq_ill != NULL) { 4289 mutex_enter(&stq_ill->ill_lock); 4290 ASSERT(stq_ill->ill_ire_cnt != 0); 4291 stq_ill->ill_ire_cnt--; 4292 if (stq_ill->ill_ire_cnt == 0) { 4293 /* Drops the ill lock */ 4294 ipif_ill_refrele_tail(stq_ill); 4295 } else { 4296 mutex_exit(&stq_ill->ill_lock); 4297 } 4298 } 4299 if (in_ill != NULL) { 4300 mutex_enter(&in_ill->ill_lock); 4301 if (ire->ire_type == IRE_MIPRTUN) { 4302 /* 4303 * Mobile IP reverse tunnel ire. 4304 * Decrement table count and the 4305 * ill reference count. This signifies 4306 * mipagent is deleting reverse tunnel 4307 * route for a particular mobile node. 4308 */ 4309 mutex_enter(&ire_mrtun_lock); 4310 ire_mrtun_count--; 4311 mutex_exit(&ire_mrtun_lock); 4312 ASSERT(in_ill->ill_mrtun_refcnt != 0); 4313 in_ill->ill_mrtun_refcnt--; 4314 if (in_ill->ill_mrtun_refcnt == 0) { 4315 /* Drops the ill lock */ 4316 ipif_ill_refrele_tail(in_ill); 4317 } else { 4318 mutex_exit(&in_ill->ill_lock); 4319 } 4320 } else { 4321 mutex_enter(&ire_srcif_table_lock); 4322 ire_srcif_table_count--; 4323 mutex_exit(&ire_srcif_table_lock); 4324 ASSERT(in_ill->ill_srcif_refcnt != 0); 4325 in_ill->ill_srcif_refcnt--; 4326 if (in_ill->ill_srcif_refcnt == 0) { 4327 /* Drops the ill lock */ 4328 ipif_ill_refrele_tail(in_ill); 4329 } else { 4330 mutex_exit(&in_ill->ill_lock); 4331 } 4332 } 4333 } 4334 } 4335 end: 4336 /* This should be true for both V4 and V6 */ 4337 ASSERT(ire->ire_fastpath == NULL); 4338 4339 if ((ire->ire_type & IRE_FORWARDTABLE) && 4340 (ire->ire_ipversion == IPV4_VERSION) && 4341 ((irb = ire->ire_bucket) != NULL)) { 4342 rw_enter(&irb->irb_lock, RW_WRITER); 4343 irb->irb_nire--; 4344 /* 4345 * Instead of examining the conditions for freeing 4346 * the radix node here, we do it by calling 4347 * IRB_REFRELE which is a single point in the code 4348 * that embeds that logic. Bump up the refcnt to 4349 * be able to call IRB_REFRELE 4350 */ 4351 IRB_REFHOLD_LOCKED(irb); 4352 rw_exit(&irb->irb_lock); 4353 IRB_REFRELE(irb); 4354 } 4355 ire->ire_ipif = NULL; 4356 4357 if (ire->ire_in_ill != NULL) { 4358 ire->ire_in_ill = NULL; 4359 } 4360 4361 #ifdef IRE_DEBUG 4362 ire_trace_inactive(ire); 4363 #endif 4364 mutex_destroy(&ire->ire_lock); 4365 if (ire->ire_ipversion == IPV6_VERSION) { 4366 BUMP_IRE_STATS(ire_stats_v6, ire_stats_freed); 4367 } else { 4368 BUMP_IRE_STATS(ire_stats_v4, ire_stats_freed); 4369 } 4370 ASSERT(ire->ire_mp == NULL); 4371 /* Has been allocated out of the cache */ 4372 kmem_cache_free(ire_cache, ire); 4373 } 4374 4375 /* 4376 * ire_walk routine to delete all IRE_CACHE/IRE_HOST_REDIRECT entries 4377 * that have a given gateway address. 4378 */ 4379 void 4380 ire_delete_cache_gw(ire_t *ire, char *cp) 4381 { 4382 ipaddr_t gw_addr; 4383 4384 if (!(ire->ire_type & (IRE_CACHE|IRE_HOST_REDIRECT))) 4385 return; 4386 4387 bcopy(cp, &gw_addr, sizeof (gw_addr)); 4388 if (ire->ire_gateway_addr == gw_addr) { 4389 ip1dbg(("ire_delete_cache_gw: deleted 0x%x type %d to 0x%x\n", 4390 (int)ntohl(ire->ire_addr), ire->ire_type, 4391 (int)ntohl(ire->ire_gateway_addr))); 4392 ire_delete(ire); 4393 } 4394 } 4395 4396 /* 4397 * Remove all IRE_CACHE entries that match the ire specified. 4398 * 4399 * The flag argument indicates if the flush request is due to addition 4400 * of new route (IRE_FLUSH_ADD) or deletion of old route (IRE_FLUSH_DELETE). 4401 * 4402 * This routine takes only the IREs from the forwarding table and flushes 4403 * the corresponding entries from the cache table. 4404 * 4405 * When flushing due to the deletion of an old route, it 4406 * just checks the cache handles (ire_phandle and ire_ihandle) and 4407 * deletes the ones that match. 4408 * 4409 * When flushing due to the creation of a new route, it checks 4410 * if a cache entry's address matches the one in the IRE and 4411 * that the cache entry's parent has a less specific mask than the 4412 * one in IRE. The destination of such a cache entry could be the 4413 * gateway for other cache entries, so we need to flush those as 4414 * well by looking for gateway addresses matching the IRE's address. 4415 */ 4416 void 4417 ire_flush_cache_v4(ire_t *ire, int flag) 4418 { 4419 int i; 4420 ire_t *cire; 4421 irb_t *irb; 4422 4423 if (ire->ire_type & IRE_CACHE) 4424 return; 4425 4426 /* 4427 * If a default is just created, there is no point 4428 * in going through the cache, as there will not be any 4429 * cached ires. 4430 */ 4431 if (ire->ire_type == IRE_DEFAULT && flag == IRE_FLUSH_ADD) 4432 return; 4433 if (flag == IRE_FLUSH_ADD) { 4434 /* 4435 * This selective flush is due to the addition of 4436 * new IRE. 4437 */ 4438 for (i = 0; i < ip_cache_table_size; i++) { 4439 irb = &ip_cache_table[i]; 4440 if ((cire = irb->irb_ire) == NULL) 4441 continue; 4442 IRB_REFHOLD(irb); 4443 for (cire = irb->irb_ire; cire != NULL; 4444 cire = cire->ire_next) { 4445 if (cire->ire_type != IRE_CACHE) 4446 continue; 4447 /* 4448 * If 'cire' belongs to the same subnet 4449 * as the new ire being added, and 'cire' 4450 * is derived from a prefix that is less 4451 * specific than the new ire being added, 4452 * we need to flush 'cire'; for instance, 4453 * when a new interface comes up. 4454 */ 4455 if (((cire->ire_addr & ire->ire_mask) == 4456 (ire->ire_addr & ire->ire_mask)) && 4457 (ip_mask_to_plen(cire->ire_cmask) <= 4458 ire->ire_masklen)) { 4459 ire_delete(cire); 4460 continue; 4461 } 4462 /* 4463 * This is the case when the ire_gateway_addr 4464 * of 'cire' belongs to the same subnet as 4465 * the new ire being added. 4466 * Flushing such ires is sometimes required to 4467 * avoid misrouting: say we have a machine with 4468 * two interfaces (I1 and I2), a default router 4469 * R on the I1 subnet, and a host route to an 4470 * off-link destination D with a gateway G on 4471 * the I2 subnet. 4472 * Under normal operation, we will have an 4473 * on-link cache entry for G and an off-link 4474 * cache entry for D with G as ire_gateway_addr, 4475 * traffic to D will reach its destination 4476 * through gateway G. 4477 * If the administrator does 'ifconfig I2 down', 4478 * the cache entries for D and G will be 4479 * flushed. However, G will now be resolved as 4480 * an off-link destination using R (the default 4481 * router) as gateway. Then D will also be 4482 * resolved as an off-link destination using G 4483 * as gateway - this behavior is due to 4484 * compatibility reasons, see comment in 4485 * ire_ihandle_lookup_offlink(). Traffic to D 4486 * will go to the router R and probably won't 4487 * reach the destination. 4488 * The administrator then does 'ifconfig I2 up'. 4489 * Since G is on the I2 subnet, this routine 4490 * will flush its cache entry. It must also 4491 * flush the cache entry for D, otherwise 4492 * traffic will stay misrouted until the IRE 4493 * times out. 4494 */ 4495 if ((cire->ire_gateway_addr & ire->ire_mask) == 4496 (ire->ire_addr & ire->ire_mask)) { 4497 ire_delete(cire); 4498 continue; 4499 } 4500 } 4501 IRB_REFRELE(irb); 4502 } 4503 } else { 4504 /* 4505 * delete the cache entries based on 4506 * handle in the IRE as this IRE is 4507 * being deleted/changed. 4508 */ 4509 for (i = 0; i < ip_cache_table_size; i++) { 4510 irb = &ip_cache_table[i]; 4511 if ((cire = irb->irb_ire) == NULL) 4512 continue; 4513 IRB_REFHOLD(irb); 4514 for (cire = irb->irb_ire; cire != NULL; 4515 cire = cire->ire_next) { 4516 if (cire->ire_type != IRE_CACHE) 4517 continue; 4518 if ((cire->ire_phandle == 0 || 4519 cire->ire_phandle != ire->ire_phandle) && 4520 (cire->ire_ihandle == 0 || 4521 cire->ire_ihandle != ire->ire_ihandle)) 4522 continue; 4523 ire_delete(cire); 4524 } 4525 IRB_REFRELE(irb); 4526 } 4527 } 4528 } 4529 4530 /* 4531 * Matches the arguments passed with the values in the ire. 4532 * 4533 * Note: for match types that match using "ipif" passed in, ipif 4534 * must be checked for non-NULL before calling this routine. 4535 */ 4536 boolean_t 4537 ire_match_args(ire_t *ire, ipaddr_t addr, ipaddr_t mask, ipaddr_t gateway, 4538 int type, const ipif_t *ipif, zoneid_t zoneid, uint32_t ihandle, 4539 const ts_label_t *tsl, int match_flags) 4540 { 4541 ill_t *ire_ill = NULL, *dst_ill; 4542 ill_t *ipif_ill = NULL; 4543 ill_group_t *ire_ill_group = NULL; 4544 ill_group_t *ipif_ill_group = NULL; 4545 4546 ASSERT(ire->ire_ipversion == IPV4_VERSION); 4547 ASSERT((ire->ire_addr & ~ire->ire_mask) == 0); 4548 ASSERT((!(match_flags & (MATCH_IRE_ILL|MATCH_IRE_ILL_GROUP))) || 4549 (ipif != NULL && !ipif->ipif_isv6)); 4550 ASSERT(!(match_flags & MATCH_IRE_WQ)); 4551 4552 /* 4553 * HIDDEN cache entries have to be looked up specifically with 4554 * MATCH_IRE_MARK_HIDDEN. MATCH_IRE_MARK_HIDDEN is usually set 4555 * when the interface is FAILED or INACTIVE. In that case, 4556 * any IRE_CACHES that exists should be marked with 4557 * IRE_MARK_HIDDEN. So, we don't really need to match below 4558 * for IRE_MARK_HIDDEN. But we do so for consistency. 4559 */ 4560 if (!(match_flags & MATCH_IRE_MARK_HIDDEN) && 4561 (ire->ire_marks & IRE_MARK_HIDDEN)) 4562 return (B_FALSE); 4563 4564 /* 4565 * MATCH_IRE_MARK_PRIVATE_ADDR is set when IP_NEXTHOP option 4566 * is used. In that case the routing table is bypassed and the 4567 * packets are sent directly to the specified nexthop. The 4568 * IRE_CACHE entry representing this route should be marked 4569 * with IRE_MARK_PRIVATE_ADDR. 4570 */ 4571 4572 if (!(match_flags & MATCH_IRE_MARK_PRIVATE_ADDR) && 4573 (ire->ire_marks & IRE_MARK_PRIVATE_ADDR)) 4574 return (B_FALSE); 4575 4576 if (zoneid != ALL_ZONES && zoneid != ire->ire_zoneid && 4577 ire->ire_zoneid != ALL_ZONES) { 4578 /* 4579 * If MATCH_IRE_ZONEONLY has been set and the supplied zoneid is 4580 * valid and does not match that of ire_zoneid, a failure to 4581 * match is reported at this point. Otherwise, since some IREs 4582 * that are available in the global zone can be used in local 4583 * zones, additional checks need to be performed: 4584 * 4585 * IRE_BROADCAST, IRE_CACHE and IRE_LOOPBACK 4586 * entries should never be matched in this situation. 4587 * 4588 * IRE entries that have an interface associated with them 4589 * should in general not match unless they are an IRE_LOCAL 4590 * or in the case when MATCH_IRE_DEFAULT has been set in 4591 * the caller. In the case of the former, checking of the 4592 * other fields supplied should take place. 4593 * 4594 * In the case where MATCH_IRE_DEFAULT has been set, 4595 * all of the ipif's associated with the IRE's ill are 4596 * checked to see if there is a matching zoneid. If any 4597 * one ipif has a matching zoneid, this IRE is a 4598 * potential candidate so checking of the other fields 4599 * takes place. 4600 * 4601 * In the case where the IRE_INTERFACE has a usable source 4602 * address (indicated by ill_usesrc_ifindex) in the 4603 * correct zone then it's permitted to return this IRE 4604 */ 4605 if (match_flags & MATCH_IRE_ZONEONLY) 4606 return (B_FALSE); 4607 if (ire->ire_type & (IRE_BROADCAST | IRE_CACHE | IRE_LOOPBACK)) 4608 return (B_FALSE); 4609 /* 4610 * Note, IRE_INTERFACE can have the stq as NULL. For 4611 * example, if the default multicast route is tied to 4612 * the loopback address. 4613 */ 4614 if ((ire->ire_type & IRE_INTERFACE) && 4615 (ire->ire_stq != NULL)) { 4616 dst_ill = (ill_t *)ire->ire_stq->q_ptr; 4617 /* 4618 * If there is a usable source address in the 4619 * zone, then it's ok to return an 4620 * IRE_INTERFACE 4621 */ 4622 if (ipif_usesrc_avail(dst_ill, zoneid)) { 4623 ip3dbg(("ire_match_args: dst_ill %p match %d\n", 4624 (void *)dst_ill, 4625 (ire->ire_addr == (addr & mask)))); 4626 } else { 4627 ip3dbg(("ire_match_args: src_ipif NULL" 4628 " dst_ill %p\n", (void *)dst_ill)); 4629 return (B_FALSE); 4630 } 4631 } 4632 if (ire->ire_ipif != NULL && ire->ire_type != IRE_LOCAL && 4633 !(ire->ire_type & IRE_INTERFACE)) { 4634 ipif_t *tipif; 4635 4636 if ((match_flags & MATCH_IRE_DEFAULT) == 0) { 4637 return (B_FALSE); 4638 } 4639 mutex_enter(&ire->ire_ipif->ipif_ill->ill_lock); 4640 for (tipif = ire->ire_ipif->ipif_ill->ill_ipif; 4641 tipif != NULL; tipif = tipif->ipif_next) { 4642 if (IPIF_CAN_LOOKUP(tipif) && 4643 (tipif->ipif_flags & IPIF_UP) && 4644 (tipif->ipif_zoneid == zoneid || 4645 tipif->ipif_zoneid == ALL_ZONES)) 4646 break; 4647 } 4648 mutex_exit(&ire->ire_ipif->ipif_ill->ill_lock); 4649 if (tipif == NULL) { 4650 return (B_FALSE); 4651 } 4652 } 4653 } 4654 4655 /* 4656 * For IRE_CACHES, MATCH_IRE_ILL/ILL_GROUP really means that 4657 * somebody wants to send out on a particular interface which 4658 * is given by ire_stq and hence use ire_stq to derive the ill 4659 * value. ire_ipif for IRE_CACHES is just the means of getting 4660 * a source address i.e ire_src_addr = ire->ire_ipif->ipif_src_addr. 4661 * ire_to_ill does the right thing for this. 4662 */ 4663 if (match_flags & (MATCH_IRE_ILL|MATCH_IRE_ILL_GROUP)) { 4664 ire_ill = ire_to_ill(ire); 4665 if (ire_ill != NULL) 4666 ire_ill_group = ire_ill->ill_group; 4667 ipif_ill = ipif->ipif_ill; 4668 ipif_ill_group = ipif_ill->ill_group; 4669 } 4670 4671 if ((ire->ire_addr == (addr & mask)) && 4672 ((!(match_flags & MATCH_IRE_GW)) || 4673 (ire->ire_gateway_addr == gateway)) && 4674 ((!(match_flags & MATCH_IRE_TYPE)) || 4675 (ire->ire_type & type)) && 4676 ((!(match_flags & MATCH_IRE_SRC)) || 4677 (ire->ire_src_addr == ipif->ipif_src_addr)) && 4678 ((!(match_flags & MATCH_IRE_IPIF)) || 4679 (ire->ire_ipif == ipif)) && 4680 ((!(match_flags & MATCH_IRE_MARK_HIDDEN)) || 4681 (ire->ire_type != IRE_CACHE || 4682 ire->ire_marks & IRE_MARK_HIDDEN)) && 4683 ((!(match_flags & MATCH_IRE_MARK_PRIVATE_ADDR)) || 4684 (ire->ire_type != IRE_CACHE || 4685 ire->ire_marks & IRE_MARK_PRIVATE_ADDR)) && 4686 ((!(match_flags & MATCH_IRE_ILL)) || 4687 (ire_ill == ipif_ill)) && 4688 ((!(match_flags & MATCH_IRE_IHANDLE)) || 4689 (ire->ire_ihandle == ihandle)) && 4690 ((!(match_flags & MATCH_IRE_MASK)) || 4691 (ire->ire_mask == mask)) && 4692 ((!(match_flags & MATCH_IRE_ILL_GROUP)) || 4693 (ire_ill == ipif_ill) || 4694 (ire_ill_group != NULL && 4695 ire_ill_group == ipif_ill_group)) && 4696 ((!(match_flags & MATCH_IRE_SECATTR)) || 4697 (!is_system_labeled()) || 4698 (tsol_ire_match_gwattr(ire, tsl) == 0))) { 4699 /* We found the matched IRE */ 4700 return (B_TRUE); 4701 } 4702 return (B_FALSE); 4703 } 4704 4705 4706 /* 4707 * Lookup for a route in all the tables 4708 */ 4709 ire_t * 4710 ire_route_lookup(ipaddr_t addr, ipaddr_t mask, ipaddr_t gateway, 4711 int type, const ipif_t *ipif, ire_t **pire, zoneid_t zoneid, 4712 const ts_label_t *tsl, int flags) 4713 { 4714 ire_t *ire = NULL; 4715 4716 /* 4717 * ire_match_args() will dereference ipif MATCH_IRE_SRC or 4718 * MATCH_IRE_ILL is set. 4719 */ 4720 if ((flags & (MATCH_IRE_SRC | MATCH_IRE_ILL | MATCH_IRE_ILL_GROUP)) && 4721 (ipif == NULL)) 4722 return (NULL); 4723 4724 /* 4725 * might be asking for a cache lookup, 4726 * This is not best way to lookup cache, 4727 * user should call ire_cache_lookup directly. 4728 * 4729 * If MATCH_IRE_TYPE was set, first lookup in the cache table and then 4730 * in the forwarding table, if the applicable type flags were set. 4731 */ 4732 if ((flags & MATCH_IRE_TYPE) == 0 || (type & IRE_CACHETABLE) != 0) { 4733 ire = ire_ctable_lookup(addr, gateway, type, ipif, zoneid, 4734 tsl, flags); 4735 if (ire != NULL) 4736 return (ire); 4737 } 4738 if ((flags & MATCH_IRE_TYPE) == 0 || (type & IRE_FORWARDTABLE) != 0) { 4739 ire = ire_ftable_lookup(addr, mask, gateway, type, ipif, pire, 4740 zoneid, 0, tsl, flags); 4741 } 4742 return (ire); 4743 } 4744 4745 4746 /* 4747 * Delete the IRE cache for the gateway and all IRE caches whose 4748 * ire_gateway_addr points to this gateway, and allow them to 4749 * be created on demand by ip_newroute. 4750 */ 4751 void 4752 ire_clookup_delete_cache_gw(ipaddr_t addr, zoneid_t zoneid) 4753 { 4754 irb_t *irb; 4755 ire_t *ire; 4756 4757 irb = &ip_cache_table[IRE_ADDR_HASH(addr, ip_cache_table_size)]; 4758 IRB_REFHOLD(irb); 4759 for (ire = irb->irb_ire; ire != NULL; ire = ire->ire_next) { 4760 if (ire->ire_marks & IRE_MARK_CONDEMNED) 4761 continue; 4762 4763 ASSERT(ire->ire_mask == IP_HOST_MASK); 4764 ASSERT(ire->ire_type != IRE_MIPRTUN && ire->ire_in_ill == NULL); 4765 if (ire_match_args(ire, addr, ire->ire_mask, 0, IRE_CACHE, 4766 NULL, zoneid, 0, NULL, MATCH_IRE_TYPE)) { 4767 ire_delete(ire); 4768 } 4769 } 4770 IRB_REFRELE(irb); 4771 4772 ire_walk_v4(ire_delete_cache_gw, &addr, zoneid); 4773 } 4774 4775 /* 4776 * Looks up cache table for a route. 4777 * specific lookup can be indicated by 4778 * passing the MATCH_* flags and the 4779 * necessary parameters. 4780 */ 4781 ire_t * 4782 ire_ctable_lookup(ipaddr_t addr, ipaddr_t gateway, int type, const ipif_t *ipif, 4783 zoneid_t zoneid, const ts_label_t *tsl, int flags) 4784 { 4785 irb_t *irb_ptr; 4786 ire_t *ire; 4787 4788 /* 4789 * ire_match_args() will dereference ipif MATCH_IRE_SRC or 4790 * MATCH_IRE_ILL is set. 4791 */ 4792 if ((flags & (MATCH_IRE_SRC | MATCH_IRE_ILL | MATCH_IRE_ILL_GROUP)) && 4793 (ipif == NULL)) 4794 return (NULL); 4795 4796 irb_ptr = &ip_cache_table[IRE_ADDR_HASH(addr, ip_cache_table_size)]; 4797 rw_enter(&irb_ptr->irb_lock, RW_READER); 4798 for (ire = irb_ptr->irb_ire; ire != NULL; ire = ire->ire_next) { 4799 if (ire->ire_marks & IRE_MARK_CONDEMNED) 4800 continue; 4801 ASSERT(ire->ire_mask == IP_HOST_MASK); 4802 ASSERT(ire->ire_type != IRE_MIPRTUN && ire->ire_in_ill == NULL); 4803 if (ire_match_args(ire, addr, ire->ire_mask, gateway, type, 4804 ipif, zoneid, 0, tsl, flags)) { 4805 IRE_REFHOLD(ire); 4806 rw_exit(&irb_ptr->irb_lock); 4807 return (ire); 4808 } 4809 } 4810 rw_exit(&irb_ptr->irb_lock); 4811 return (NULL); 4812 } 4813 4814 /* 4815 * Check whether the IRE_LOCAL and the IRE potentially used to transmit 4816 * (could be an IRE_CACHE, IRE_BROADCAST, or IRE_INTERFACE) are part of 4817 * the same ill group. 4818 */ 4819 boolean_t 4820 ire_local_same_ill_group(ire_t *ire_local, ire_t *xmit_ire) 4821 { 4822 ill_t *recv_ill, *xmit_ill; 4823 ill_group_t *recv_group, *xmit_group; 4824 4825 ASSERT(ire_local->ire_type == IRE_LOCAL); 4826 ASSERT(ire_local->ire_rfq != NULL); 4827 ASSERT(xmit_ire->ire_type & (IRE_CACHE|IRE_BROADCAST|IRE_INTERFACE)); 4828 ASSERT(xmit_ire->ire_stq != NULL); 4829 ASSERT(xmit_ire->ire_ipif != NULL); 4830 4831 recv_ill = ire_local->ire_rfq->q_ptr; 4832 xmit_ill = xmit_ire->ire_stq->q_ptr; 4833 4834 if (recv_ill == xmit_ill) 4835 return (B_TRUE); 4836 4837 recv_group = recv_ill->ill_group; 4838 xmit_group = xmit_ill->ill_group; 4839 4840 if (recv_group != NULL && recv_group == xmit_group) 4841 return (B_TRUE); 4842 4843 return (B_FALSE); 4844 } 4845 4846 /* 4847 * Check if the IRE_LOCAL uses the same ill (group) as another route would use. 4848 */ 4849 boolean_t 4850 ire_local_ok_across_zones(ire_t *ire_local, zoneid_t zoneid, void *addr, 4851 const ts_label_t *tsl) 4852 { 4853 ire_t *alt_ire; 4854 boolean_t rval; 4855 4856 if (ire_local->ire_ipversion == IPV4_VERSION) { 4857 alt_ire = ire_ftable_lookup(*((ipaddr_t *)addr), 0, 0, 0, NULL, 4858 NULL, zoneid, 0, tsl, 4859 MATCH_IRE_RECURSIVE | MATCH_IRE_DEFAULT | 4860 MATCH_IRE_RJ_BHOLE); 4861 } else { 4862 alt_ire = ire_ftable_lookup_v6((in6_addr_t *)addr, NULL, NULL, 4863 0, NULL, NULL, zoneid, 0, tsl, 4864 MATCH_IRE_RECURSIVE | MATCH_IRE_DEFAULT | 4865 MATCH_IRE_RJ_BHOLE); 4866 } 4867 4868 if (alt_ire == NULL) 4869 return (B_FALSE); 4870 4871 rval = ire_local_same_ill_group(ire_local, alt_ire); 4872 4873 ire_refrele(alt_ire); 4874 return (rval); 4875 } 4876 4877 /* 4878 * Lookup cache. Don't return IRE_MARK_HIDDEN entries. Callers 4879 * should use ire_ctable_lookup with MATCH_IRE_MARK_HIDDEN to get 4880 * to the hidden ones. 4881 * 4882 * In general the zoneid has to match (where ALL_ZONES match all of them). 4883 * But for IRE_LOCAL we also need to handle the case where L2 should 4884 * conceptually loop back the packet. This is necessary since neither 4885 * Ethernet drivers nor Ethernet hardware loops back packets sent to their 4886 * own MAC address. This loopback is needed when the normal 4887 * routes (ignoring IREs with different zoneids) would send out the packet on 4888 * the same ill (or ill group) as the ill with which this IRE_LOCAL is 4889 * associated. 4890 * 4891 * Earlier versions of this code always matched an IRE_LOCAL independently of 4892 * the zoneid. We preserve that earlier behavior when 4893 * ip_restrict_interzone_loopback is turned off. 4894 */ 4895 ire_t * 4896 ire_cache_lookup(ipaddr_t addr, zoneid_t zoneid, const ts_label_t *tsl) 4897 { 4898 irb_t *irb_ptr; 4899 ire_t *ire; 4900 4901 irb_ptr = &ip_cache_table[IRE_ADDR_HASH(addr, ip_cache_table_size)]; 4902 rw_enter(&irb_ptr->irb_lock, RW_READER); 4903 for (ire = irb_ptr->irb_ire; ire != NULL; ire = ire->ire_next) { 4904 if (ire->ire_marks & (IRE_MARK_CONDEMNED | 4905 IRE_MARK_HIDDEN | IRE_MARK_PRIVATE_ADDR)) { 4906 continue; 4907 } 4908 if (ire->ire_addr == addr) { 4909 /* 4910 * Finally, check if the security policy has any 4911 * restriction on using this route for the specified 4912 * message. 4913 */ 4914 if (tsl != NULL && 4915 ire->ire_gw_secattr != NULL && 4916 tsol_ire_match_gwattr(ire, tsl) != 0) { 4917 continue; 4918 } 4919 4920 if (zoneid == ALL_ZONES || ire->ire_zoneid == zoneid || 4921 ire->ire_zoneid == ALL_ZONES) { 4922 IRE_REFHOLD(ire); 4923 rw_exit(&irb_ptr->irb_lock); 4924 return (ire); 4925 } 4926 4927 if (ire->ire_type == IRE_LOCAL) { 4928 if (ip_restrict_interzone_loopback && 4929 !ire_local_ok_across_zones(ire, zoneid, 4930 &addr, tsl)) 4931 continue; 4932 4933 IRE_REFHOLD(ire); 4934 rw_exit(&irb_ptr->irb_lock); 4935 return (ire); 4936 } 4937 } 4938 } 4939 rw_exit(&irb_ptr->irb_lock); 4940 return (NULL); 4941 } 4942 4943 /* 4944 * Locate the interface ire that is tied to the cache ire 'cire' via 4945 * cire->ire_ihandle. 4946 * 4947 * We are trying to create the cache ire for an offlink destn based 4948 * on the cache ire of the gateway in 'cire'. 'pire' is the prefix ire 4949 * as found by ip_newroute(). We are called from ip_newroute() in 4950 * the IRE_CACHE case. 4951 */ 4952 ire_t * 4953 ire_ihandle_lookup_offlink(ire_t *cire, ire_t *pire) 4954 { 4955 ire_t *ire; 4956 int match_flags; 4957 ipaddr_t gw_addr; 4958 ipif_t *gw_ipif; 4959 4960 ASSERT(cire != NULL && pire != NULL); 4961 4962 /* 4963 * We don't need to specify the zoneid to ire_ftable_lookup() below 4964 * because the ihandle refers to an ipif which can be in only one zone. 4965 */ 4966 match_flags = MATCH_IRE_TYPE | MATCH_IRE_IHANDLE | MATCH_IRE_MASK; 4967 /* 4968 * ip_newroute calls ire_ftable_lookup with MATCH_IRE_ILL only 4969 * for on-link hosts. We should never be here for onlink. 4970 * Thus, use MATCH_IRE_ILL_GROUP. 4971 */ 4972 if (pire->ire_ipif != NULL) 4973 match_flags |= MATCH_IRE_ILL_GROUP; 4974 /* 4975 * We know that the mask of the interface ire equals cire->ire_cmask. 4976 * (When ip_newroute() created 'cire' for the gateway it set its 4977 * cmask from the interface ire's mask) 4978 */ 4979 ire = ire_ftable_lookup(cire->ire_addr, cire->ire_cmask, 0, 4980 IRE_INTERFACE, pire->ire_ipif, NULL, ALL_ZONES, cire->ire_ihandle, 4981 NULL, match_flags); 4982 if (ire != NULL) 4983 return (ire); 4984 /* 4985 * If we didn't find an interface ire above, we can't declare failure. 4986 * For backwards compatibility, we need to support prefix routes 4987 * pointing to next hop gateways that are not on-link. 4988 * 4989 * Assume we are trying to ping some offlink destn, and we have the 4990 * routing table below. 4991 * 4992 * Eg. default - gw1 <--- pire (line 1) 4993 * gw1 - gw2 (line 2) 4994 * gw2 - hme0 (line 3) 4995 * 4996 * If we already have a cache ire for gw1 in 'cire', the 4997 * ire_ftable_lookup above would have failed, since there is no 4998 * interface ire to reach gw1. We will fallthru below. 4999 * 5000 * Here we duplicate the steps that ire_ftable_lookup() did in 5001 * getting 'cire' from 'pire', in the MATCH_IRE_RECURSIVE case. 5002 * The differences are the following 5003 * i. We want the interface ire only, so we call ire_ftable_lookup() 5004 * instead of ire_route_lookup() 5005 * ii. We look for only prefix routes in the 1st call below. 5006 * ii. We want to match on the ihandle in the 2nd call below. 5007 */ 5008 match_flags = MATCH_IRE_TYPE; 5009 if (pire->ire_ipif != NULL) 5010 match_flags |= MATCH_IRE_ILL_GROUP; 5011 ire = ire_ftable_lookup(pire->ire_gateway_addr, 0, 0, IRE_OFFSUBNET, 5012 pire->ire_ipif, NULL, ALL_ZONES, 0, NULL, match_flags); 5013 if (ire == NULL) 5014 return (NULL); 5015 /* 5016 * At this point 'ire' corresponds to the entry shown in line 2. 5017 * gw_addr is 'gw2' in the example above. 5018 */ 5019 gw_addr = ire->ire_gateway_addr; 5020 gw_ipif = ire->ire_ipif; 5021 ire_refrele(ire); 5022 5023 match_flags |= MATCH_IRE_IHANDLE; 5024 ire = ire_ftable_lookup(gw_addr, 0, 0, IRE_INTERFACE, 5025 gw_ipif, NULL, ALL_ZONES, cire->ire_ihandle, NULL, match_flags); 5026 return (ire); 5027 } 5028 5029 /* 5030 * ire_mrtun_lookup() is called by ip_rput() when packet is to be 5031 * tunneled through reverse tunnel. This is only supported for 5032 * IPv4 packets 5033 */ 5034 5035 ire_t * 5036 ire_mrtun_lookup(ipaddr_t srcaddr, ill_t *ill) 5037 { 5038 irb_t *irb_ptr; 5039 ire_t *ire; 5040 5041 ASSERT(ill != NULL); 5042 ASSERT(!(ill->ill_isv6)); 5043 5044 if (ip_mrtun_table == NULL) 5045 return (NULL); 5046 irb_ptr = &ip_mrtun_table[IRE_ADDR_HASH(srcaddr, IP_MRTUN_TABLE_SIZE)]; 5047 rw_enter(&irb_ptr->irb_lock, RW_READER); 5048 for (ire = irb_ptr->irb_ire; ire != NULL; ire = ire->ire_next) { 5049 if (ire->ire_marks & IRE_MARK_CONDEMNED) 5050 continue; 5051 if ((ire->ire_in_src_addr == srcaddr) && 5052 ire->ire_in_ill == ill) { 5053 IRE_REFHOLD(ire); 5054 rw_exit(&irb_ptr->irb_lock); 5055 return (ire); 5056 } 5057 } 5058 rw_exit(&irb_ptr->irb_lock); 5059 return (NULL); 5060 } 5061 5062 /* 5063 * Return the IRE_LOOPBACK, IRE_IF_RESOLVER or IRE_IF_NORESOLVER 5064 * ire associated with the specified ipif. 5065 * 5066 * This might occasionally be called when IPIF_UP is not set since 5067 * the IP_MULTICAST_IF as well as creating interface routes 5068 * allows specifying a down ipif (ipif_lookup* match ipifs that are down). 5069 * 5070 * Note that if IPIF_NOLOCAL, IPIF_NOXMIT, or IPIF_DEPRECATED is set on 5071 * the ipif, this routine might return NULL. 5072 */ 5073 ire_t * 5074 ipif_to_ire(const ipif_t *ipif) 5075 { 5076 ire_t *ire; 5077 5078 ASSERT(!ipif->ipif_isv6); 5079 if (ipif->ipif_ire_type == IRE_LOOPBACK) { 5080 ire = ire_ctable_lookup(ipif->ipif_lcl_addr, 0, IRE_LOOPBACK, 5081 ipif, ALL_ZONES, NULL, (MATCH_IRE_TYPE | MATCH_IRE_IPIF)); 5082 } else if (ipif->ipif_flags & IPIF_POINTOPOINT) { 5083 /* In this case we need to lookup destination address. */ 5084 ire = ire_ftable_lookup(ipif->ipif_pp_dst_addr, IP_HOST_MASK, 0, 5085 IRE_INTERFACE, ipif, NULL, ALL_ZONES, 0, NULL, 5086 (MATCH_IRE_TYPE | MATCH_IRE_IPIF | MATCH_IRE_MASK)); 5087 } else { 5088 ire = ire_ftable_lookup(ipif->ipif_subnet, 5089 ipif->ipif_net_mask, 0, IRE_INTERFACE, ipif, NULL, 5090 ALL_ZONES, 0, NULL, (MATCH_IRE_TYPE | MATCH_IRE_IPIF | 5091 MATCH_IRE_MASK)); 5092 } 5093 return (ire); 5094 } 5095 5096 /* 5097 * ire_walk function. 5098 * Count the number of IRE_CACHE entries in different categories. 5099 */ 5100 void 5101 ire_cache_count(ire_t *ire, char *arg) 5102 { 5103 ire_cache_count_t *icc = (ire_cache_count_t *)arg; 5104 5105 if (ire->ire_type != IRE_CACHE) 5106 return; 5107 5108 icc->icc_total++; 5109 5110 if (ire->ire_ipversion == IPV6_VERSION) { 5111 mutex_enter(&ire->ire_lock); 5112 if (IN6_IS_ADDR_UNSPECIFIED(&ire->ire_gateway_addr_v6)) { 5113 mutex_exit(&ire->ire_lock); 5114 icc->icc_onlink++; 5115 return; 5116 } 5117 mutex_exit(&ire->ire_lock); 5118 } else { 5119 if (ire->ire_gateway_addr == 0) { 5120 icc->icc_onlink++; 5121 return; 5122 } 5123 } 5124 5125 ASSERT(ire->ire_ipif != NULL); 5126 if (ire->ire_max_frag < ire->ire_ipif->ipif_mtu) 5127 icc->icc_pmtu++; 5128 else if (ire->ire_tire_mark != ire->ire_ob_pkt_count + 5129 ire->ire_ib_pkt_count) 5130 icc->icc_offlink++; 5131 else 5132 icc->icc_unused++; 5133 } 5134 5135 /* 5136 * ire_walk function called by ip_trash_ire_reclaim(). 5137 * Free a fraction of the IRE_CACHE cache entries. The fractions are 5138 * different for different categories of IRE_CACHE entries. 5139 * A fraction of zero means to not free any in that category. 5140 * Use the hash bucket id plus lbolt as a random number. Thus if the fraction 5141 * is N then every Nth hash bucket chain will be freed. 5142 */ 5143 void 5144 ire_cache_reclaim(ire_t *ire, char *arg) 5145 { 5146 ire_cache_reclaim_t *icr = (ire_cache_reclaim_t *)arg; 5147 uint_t rand; 5148 5149 if (ire->ire_type != IRE_CACHE) 5150 return; 5151 5152 if (ire->ire_ipversion == IPV6_VERSION) { 5153 rand = (uint_t)lbolt + 5154 IRE_ADDR_HASH_V6(ire->ire_addr_v6, ip6_cache_table_size); 5155 mutex_enter(&ire->ire_lock); 5156 if (IN6_IS_ADDR_UNSPECIFIED(&ire->ire_gateway_addr_v6)) { 5157 mutex_exit(&ire->ire_lock); 5158 if (icr->icr_onlink != 0 && 5159 (rand/icr->icr_onlink)*icr->icr_onlink == rand) { 5160 ire_delete(ire); 5161 return; 5162 } 5163 goto done; 5164 } 5165 mutex_exit(&ire->ire_lock); 5166 } else { 5167 rand = (uint_t)lbolt + 5168 IRE_ADDR_HASH(ire->ire_addr, ip_cache_table_size); 5169 if (ire->ire_gateway_addr == 0) { 5170 if (icr->icr_onlink != 0 && 5171 (rand/icr->icr_onlink)*icr->icr_onlink == rand) { 5172 ire_delete(ire); 5173 return; 5174 } 5175 goto done; 5176 } 5177 } 5178 /* Not onlink IRE */ 5179 ASSERT(ire->ire_ipif != NULL); 5180 if (ire->ire_max_frag < ire->ire_ipif->ipif_mtu) { 5181 /* Use ptmu fraction */ 5182 if (icr->icr_pmtu != 0 && 5183 (rand/icr->icr_pmtu)*icr->icr_pmtu == rand) { 5184 ire_delete(ire); 5185 return; 5186 } 5187 } else if (ire->ire_tire_mark != ire->ire_ob_pkt_count + 5188 ire->ire_ib_pkt_count) { 5189 /* Use offlink fraction */ 5190 if (icr->icr_offlink != 0 && 5191 (rand/icr->icr_offlink)*icr->icr_offlink == rand) { 5192 ire_delete(ire); 5193 return; 5194 } 5195 } else { 5196 /* Use unused fraction */ 5197 if (icr->icr_unused != 0 && 5198 (rand/icr->icr_unused)*icr->icr_unused == rand) { 5199 ire_delete(ire); 5200 return; 5201 } 5202 } 5203 done: 5204 /* 5205 * Update tire_mark so that those that haven't been used since this 5206 * reclaim will be considered unused next time we reclaim. 5207 */ 5208 ire->ire_tire_mark = ire->ire_ob_pkt_count + ire->ire_ib_pkt_count; 5209 } 5210 5211 static void 5212 power2_roundup(uint32_t *value) 5213 { 5214 int i; 5215 5216 for (i = 1; i < 31; i++) { 5217 if (*value <= (1 << i)) 5218 break; 5219 } 5220 *value = (1 << i); 5221 } 5222 5223 void 5224 ip_ire_init() 5225 { 5226 int i; 5227 5228 mutex_init(&ire_ft_init_lock, NULL, MUTEX_DEFAULT, 0); 5229 mutex_init(&ire_handle_lock, NULL, MUTEX_DEFAULT, NULL); 5230 mutex_init(&ire_mrtun_lock, NULL, MUTEX_DEFAULT, NULL); 5231 mutex_init(&ire_srcif_table_lock, NULL, MUTEX_DEFAULT, NULL); 5232 mutex_init(&ndp4.ndp_g_lock, NULL, MUTEX_DEFAULT, NULL); 5233 5234 rn_init(); 5235 (void) rn_inithead((void **)&ip_ftable, 32); 5236 /* 5237 * mark kernel ip ftable with RNF_SUNW_FT flag. 5238 */ 5239 ip_ftable->rnh_treetop->rn_flags |= RNF_SUNW_FT; 5240 rt_entry_cache = kmem_cache_create("rt_entry", 5241 sizeof (struct rt_entry), 0, NULL, NULL, NULL, NULL, NULL, 0); 5242 5243 /* Calculate the IPv4 cache table size. */ 5244 ip_cache_table_size = MAX(ip_cache_table_size, 5245 ((kmem_avail() >> ip_ire_mem_ratio) / sizeof (ire_t) / 5246 ip_ire_max_bucket_cnt)); 5247 if (ip_cache_table_size > ip_max_cache_table_size) 5248 ip_cache_table_size = ip_max_cache_table_size; 5249 /* 5250 * Make sure that the table size is always a power of 2. The 5251 * hash macro IRE_ADDR_HASH() depends on that. 5252 */ 5253 power2_roundup(&ip_cache_table_size); 5254 5255 ip_cache_table = (irb_t *)kmem_zalloc(ip_cache_table_size * 5256 sizeof (irb_t), KM_SLEEP); 5257 5258 for (i = 0; i < ip_cache_table_size; i++) { 5259 rw_init(&ip_cache_table[i].irb_lock, NULL, 5260 RW_DEFAULT, NULL); 5261 } 5262 5263 /* Calculate the IPv6 cache table size. */ 5264 ip6_cache_table_size = MAX(ip6_cache_table_size, 5265 ((kmem_avail() >> ip_ire_mem_ratio) / sizeof (ire_t) / 5266 ip6_ire_max_bucket_cnt)); 5267 if (ip6_cache_table_size > ip6_max_cache_table_size) 5268 ip6_cache_table_size = ip6_max_cache_table_size; 5269 /* 5270 * Make sure that the table size is always a power of 2. The 5271 * hash macro IRE_ADDR_HASH_V6() depends on that. 5272 */ 5273 power2_roundup(&ip6_cache_table_size); 5274 5275 ip_cache_table_v6 = (irb_t *)kmem_zalloc(ip6_cache_table_size * 5276 sizeof (irb_t), KM_SLEEP); 5277 5278 for (i = 0; i < ip6_cache_table_size; i++) { 5279 rw_init(&ip_cache_table_v6[i].irb_lock, NULL, 5280 RW_DEFAULT, NULL); 5281 } 5282 /* 5283 * Create ire caches, ire_reclaim() 5284 * will give IRE_CACHE back to system when needed. 5285 * This needs to be done here before anything else, since 5286 * ire_add() expects the cache to be created. 5287 */ 5288 ire_cache = kmem_cache_create("ire_cache", 5289 sizeof (ire_t), 0, ip_ire_constructor, 5290 ip_ire_destructor, ip_trash_ire_reclaim, NULL, NULL, 0); 5291 5292 /* 5293 * Initialize ip_mrtun_table to NULL now, it will be 5294 * populated by ip_rt_add if reverse tunnel is created 5295 */ 5296 ip_mrtun_table = NULL; 5297 5298 /* 5299 * Make sure that the forwarding table size is a power of 2. 5300 * The IRE*_ADDR_HASH() macroes depend on that. 5301 */ 5302 power2_roundup(&ip6_ftable_hash_size); 5303 } 5304 5305 void 5306 ip_ire_fini() 5307 { 5308 int i; 5309 5310 mutex_destroy(&ire_ft_init_lock); 5311 mutex_destroy(&ire_handle_lock); 5312 mutex_destroy(&ndp4.ndp_g_lock); 5313 5314 rn_fini(); 5315 RADIX_NODE_HEAD_DESTROY(ip_ftable); 5316 kmem_cache_destroy(rt_entry_cache); 5317 5318 for (i = 0; i < ip_cache_table_size; i++) { 5319 rw_destroy(&ip_cache_table[i].irb_lock); 5320 } 5321 kmem_free(ip_cache_table, ip_cache_table_size * sizeof (irb_t)); 5322 5323 for (i = 0; i < ip6_cache_table_size; i++) { 5324 rw_destroy(&ip_cache_table_v6[i].irb_lock); 5325 } 5326 kmem_free(ip_cache_table_v6, ip6_cache_table_size * sizeof (irb_t)); 5327 5328 if (ip_mrtun_table != NULL) { 5329 for (i = 0; i < IP_MRTUN_TABLE_SIZE; i++) { 5330 rw_destroy(&ip_mrtun_table[i].irb_lock); 5331 } 5332 kmem_free(ip_mrtun_table, IP_MRTUN_TABLE_SIZE * sizeof (irb_t)); 5333 } 5334 kmem_cache_destroy(ire_cache); 5335 } 5336 5337 int 5338 ire_add_mrtun(ire_t **ire_p, queue_t *q, mblk_t *mp, ipsq_func_t func) 5339 { 5340 ire_t *ire1; 5341 irb_t *irb_ptr; 5342 ire_t **irep; 5343 ire_t *ire; 5344 int i; 5345 uint_t max_frag; 5346 ill_t *stq_ill; 5347 int error; 5348 5349 ire = *ire_p; 5350 ASSERT(ire->ire_ipversion == IPV4_VERSION); 5351 /* Is ip_mrtun_table empty ? */ 5352 5353 if (ip_mrtun_table == NULL) { 5354 /* create the mrtun table */ 5355 mutex_enter(&ire_mrtun_lock); 5356 if (ip_mrtun_table == NULL) { 5357 ip_mrtun_table = 5358 (irb_t *)kmem_zalloc(IP_MRTUN_TABLE_SIZE * 5359 sizeof (irb_t), KM_NOSLEEP); 5360 5361 if (ip_mrtun_table == NULL) { 5362 ip2dbg(("ire_add_mrtun: allocation failure\n")); 5363 mutex_exit(&ire_mrtun_lock); 5364 ire_refrele(ire); 5365 *ire_p = NULL; 5366 return (ENOMEM); 5367 } 5368 5369 for (i = 0; i < IP_MRTUN_TABLE_SIZE; i++) { 5370 rw_init(&ip_mrtun_table[i].irb_lock, NULL, 5371 RW_DEFAULT, NULL); 5372 } 5373 ip2dbg(("ire_add_mrtun: mrtun table is created\n")); 5374 } 5375 /* some other thread got it and created the table */ 5376 mutex_exit(&ire_mrtun_lock); 5377 } 5378 5379 /* 5380 * Check for duplicate in the bucket and insert in the table 5381 */ 5382 irb_ptr = &(ip_mrtun_table[IRE_ADDR_HASH(ire->ire_in_src_addr, 5383 IP_MRTUN_TABLE_SIZE)]); 5384 5385 /* 5386 * Start the atomic add of the ire. Grab the ill locks, 5387 * ill_g_usesrc_lock and the bucket lock. 5388 * 5389 * If ipif or ill is changing ire_atomic_start() may queue the 5390 * request and return EINPROGRESS. 5391 */ 5392 error = ire_atomic_start(irb_ptr, ire, q, mp, func); 5393 if (error != 0) { 5394 /* 5395 * We don't know whether it is a valid ipif or not. 5396 * So, set it to NULL. This assumes that the ire has not added 5397 * a reference to the ipif. 5398 */ 5399 ire->ire_ipif = NULL; 5400 ire_delete(ire); 5401 ip1dbg(("ire_add_mrtun: ire_atomic_start failed\n")); 5402 *ire_p = NULL; 5403 return (error); 5404 } 5405 for (ire1 = irb_ptr->irb_ire; ire1 != NULL; ire1 = ire1->ire_next) { 5406 if (ire1->ire_marks & IRE_MARK_CONDEMNED) 5407 continue; 5408 /* has anyone inserted the route in the meanwhile ? */ 5409 if (ire1->ire_in_ill == ire->ire_in_ill && 5410 ire1->ire_in_src_addr == ire->ire_in_src_addr) { 5411 ip1dbg(("ire_add_mrtun: Duplicate entry exists\n")); 5412 IRE_REFHOLD(ire1); 5413 ire_atomic_end(irb_ptr, ire); 5414 ire_delete(ire); 5415 /* Return the old ire */ 5416 *ire_p = ire1; 5417 return (0); 5418 } 5419 } 5420 5421 /* Atomically set the ire_max_frag */ 5422 max_frag = *ire->ire_max_fragp; 5423 ire->ire_max_fragp = NULL; 5424 ire->ire_max_frag = MIN(max_frag, IP_MAXPACKET); 5425 ASSERT(ire->ire_type != IRE_CACHE); 5426 irep = (ire_t **)irb_ptr; 5427 if (*irep != NULL) { 5428 /* Find the last ire which matches ire_in_src_addr */ 5429 ire1 = *irep; 5430 while (ire1->ire_in_src_addr == ire->ire_in_src_addr) { 5431 irep = &ire1->ire_next; 5432 ire1 = *irep; 5433 if (ire1 == NULL) 5434 break; 5435 } 5436 } 5437 ire1 = *irep; 5438 if (ire1 != NULL) 5439 ire1->ire_ptpn = &ire->ire_next; 5440 ire->ire_next = ire1; 5441 /* Link the new one in. */ 5442 ire->ire_ptpn = irep; 5443 membar_producer(); 5444 *irep = ire; 5445 ire->ire_bucket = irb_ptr; 5446 IRE_REFHOLD_LOCKED(ire); 5447 5448 ip2dbg(("ire_add_mrtun: created and linked ire %p\n", (void *)*irep)); 5449 5450 /* 5451 * Protect ire_mrtun_count and ill_mrtun_refcnt from 5452 * another thread trying to add ire in the table 5453 */ 5454 mutex_enter(&ire_mrtun_lock); 5455 ire_mrtun_count++; 5456 mutex_exit(&ire_mrtun_lock); 5457 /* 5458 * ill_mrtun_refcnt is protected by the ill_lock held via 5459 * ire_atomic_start 5460 */ 5461 ire->ire_in_ill->ill_mrtun_refcnt++; 5462 5463 if (ire->ire_ipif != NULL) { 5464 ire->ire_ipif->ipif_ire_cnt++; 5465 if (ire->ire_stq != NULL) { 5466 stq_ill = (ill_t *)ire->ire_stq->q_ptr; 5467 stq_ill->ill_ire_cnt++; 5468 } 5469 } else { 5470 ASSERT(ire->ire_stq == NULL); 5471 } 5472 5473 ire_atomic_end(irb_ptr, ire); 5474 ire_fastpath(ire); 5475 *ire_p = ire; 5476 return (0); 5477 } 5478 5479 5480 /* Walks down the mrtun table */ 5481 5482 void 5483 ire_walk_ill_mrtun(uint_t match_flags, uint_t ire_type, pfv_t func, void *arg, 5484 ill_t *ill) 5485 { 5486 irb_t *irb; 5487 ire_t *ire; 5488 int i; 5489 int ret; 5490 5491 ASSERT((!(match_flags & (MATCH_IRE_WQ | MATCH_IRE_ILL | 5492 MATCH_IRE_ILL_GROUP))) || (ill != NULL)); 5493 ASSERT(match_flags == 0 || ire_type == IRE_MIPRTUN); 5494 5495 mutex_enter(&ire_mrtun_lock); 5496 if (ire_mrtun_count == 0) { 5497 mutex_exit(&ire_mrtun_lock); 5498 return; 5499 } 5500 mutex_exit(&ire_mrtun_lock); 5501 5502 ip2dbg(("ire_walk_ill_mrtun:walking the reverse tunnel table \n")); 5503 for (i = 0; i < IP_MRTUN_TABLE_SIZE; i++) { 5504 5505 irb = &(ip_mrtun_table[i]); 5506 if (irb->irb_ire == NULL) 5507 continue; 5508 IRB_REFHOLD(irb); 5509 for (ire = irb->irb_ire; ire != NULL; 5510 ire = ire->ire_next) { 5511 ASSERT(ire->ire_ipversion == IPV4_VERSION); 5512 if (match_flags != 0) { 5513 ret = ire_walk_ill_match( 5514 match_flags, ire_type, 5515 ire, ill, ALL_ZONES); 5516 } 5517 if (match_flags == 0 || ret) 5518 (*func)(ire, arg); 5519 } 5520 IRB_REFRELE(irb); 5521 } 5522 } 5523 5524 /* 5525 * Source interface based lookup routine (IPV4 only). 5526 * This routine is called only when RTA_SRCIFP bitflag is set 5527 * by routing socket while adding/deleting the route and it is 5528 * also called from ip_rput() when packets arrive from an interface 5529 * for which ill_srcif_ref_cnt is positive. This function is useful 5530 * when a packet coming from one interface must be forwarded to another 5531 * designated interface to reach the correct node. This function is also 5532 * called from ip_newroute when the link-layer address of an ire is resolved. 5533 * We need to make sure that ip_newroute searches for IRE_IF_RESOLVER type 5534 * ires--thus the ire_type parameter is needed. 5535 */ 5536 5537 ire_t * 5538 ire_srcif_table_lookup(ipaddr_t dst_addr, int ire_type, ipif_t *ipif, 5539 ill_t *in_ill, int flags) 5540 { 5541 irb_t *irb_ptr; 5542 ire_t *ire; 5543 irb_t *ire_srcif_table; 5544 5545 ASSERT(in_ill != NULL && !in_ill->ill_isv6); 5546 ASSERT(!(flags & (MATCH_IRE_ILL|MATCH_IRE_ILL_GROUP)) || 5547 (ipif != NULL && !ipif->ipif_isv6)); 5548 5549 /* 5550 * No need to lock the ill since it is refheld by the caller of this 5551 * function 5552 */ 5553 if (in_ill->ill_srcif_table == NULL) { 5554 return (NULL); 5555 } 5556 5557 if (!(flags & MATCH_IRE_TYPE)) { 5558 flags |= MATCH_IRE_TYPE; 5559 ire_type = IRE_INTERFACE; 5560 } 5561 ire_srcif_table = in_ill->ill_srcif_table; 5562 irb_ptr = &ire_srcif_table[IRE_ADDR_HASH(dst_addr, 5563 IP_SRCIF_TABLE_SIZE)]; 5564 rw_enter(&irb_ptr->irb_lock, RW_READER); 5565 for (ire = irb_ptr->irb_ire; ire != NULL; ire = ire->ire_next) { 5566 if (ire->ire_marks & IRE_MARK_CONDEMNED) 5567 continue; 5568 if (ire_match_args(ire, dst_addr, ire->ire_mask, 0, 5569 ire_type, ipif, ire->ire_zoneid, 0, NULL, flags)) { 5570 IRE_REFHOLD(ire); 5571 rw_exit(&irb_ptr->irb_lock); 5572 return (ire); 5573 } 5574 } 5575 /* Not Found */ 5576 rw_exit(&irb_ptr->irb_lock); 5577 return (NULL); 5578 } 5579 5580 5581 /* 5582 * Adds the ire into the special routing table which is hanging off of 5583 * the src_ipif->ipif_ill. It also increments the refcnt in the ill. 5584 * The forward table contains only IRE_IF_RESOLVER, IRE_IF_NORESOLVER 5585 * i,e. IRE_INTERFACE entries. Originally the dlureq_mp field is NULL 5586 * for IRE_IF_RESOLVER entry because we do not have the dst_addr's 5587 * link-layer address at the time of addition. 5588 * Upon resolving the address from ARP, dlureq_mp field is updated with 5589 * proper information in ire_update_srcif_v4. 5590 */ 5591 static int 5592 ire_add_srcif_v4(ire_t **ire_p, queue_t *q, mblk_t *mp, ipsq_func_t func) 5593 { 5594 ire_t *ire1; 5595 irb_t *ire_srcifp_table = NULL; 5596 irb_t *irb_ptr = NULL; 5597 ire_t **irep; 5598 ire_t *ire; 5599 int flags; 5600 int i; 5601 ill_t *stq_ill; 5602 uint_t max_frag; 5603 int error = 0; 5604 5605 ire = *ire_p; 5606 ASSERT(ire->ire_in_ill != NULL); 5607 ASSERT(ire->ire_ipversion == IPV4_VERSION); 5608 ASSERT(ire->ire_type == IRE_IF_NORESOLVER || 5609 ire->ire_type == IRE_IF_RESOLVER); 5610 5611 ire->ire_mask = IP_HOST_MASK; 5612 /* 5613 * Update ire_nce->nce_res_mp with NULL value upon creation; 5614 * first free the default res_mp created by ire_nce_init. 5615 */ 5616 freeb(ire->ire_nce->nce_res_mp); 5617 if (ire->ire_type == IRE_IF_RESOLVER) { 5618 /* 5619 * assign NULL now, it will be updated 5620 * with correct value upon returning from 5621 * ARP 5622 */ 5623 ire->ire_nce->nce_res_mp = NULL; 5624 } else { 5625 ire->ire_nce->nce_res_mp = ill_dlur_gen(NULL, 5626 ire->ire_ipif->ipif_ill->ill_phys_addr_length, 5627 ire->ire_ipif->ipif_ill->ill_sap, 5628 ire->ire_ipif->ipif_ill->ill_sap_length); 5629 } 5630 /* Make sure the address is properly masked. */ 5631 ire->ire_addr &= ire->ire_mask; 5632 5633 ASSERT(ire->ire_max_fragp != NULL); 5634 max_frag = *ire->ire_max_fragp; 5635 ire->ire_max_fragp = NULL; 5636 ire->ire_max_frag = MIN(max_frag, IP_MAXPACKET); 5637 5638 mutex_enter(&ire->ire_in_ill->ill_lock); 5639 if (ire->ire_in_ill->ill_srcif_table == NULL) { 5640 /* create the incoming interface based table */ 5641 ire->ire_in_ill->ill_srcif_table = 5642 (irb_t *)kmem_zalloc(IP_SRCIF_TABLE_SIZE * 5643 sizeof (irb_t), KM_NOSLEEP); 5644 if (ire->ire_in_ill->ill_srcif_table == NULL) { 5645 ip1dbg(("ire_add_srcif_v4: Allocation fail\n")); 5646 mutex_exit(&ire->ire_in_ill->ill_lock); 5647 ire_delete(ire); 5648 *ire_p = NULL; 5649 return (ENOMEM); 5650 } 5651 ire_srcifp_table = ire->ire_in_ill->ill_srcif_table; 5652 for (i = 0; i < IP_SRCIF_TABLE_SIZE; i++) { 5653 rw_init(&ire_srcifp_table[i].irb_lock, NULL, 5654 RW_DEFAULT, NULL); 5655 } 5656 ip2dbg(("ire_add_srcif_v4: table created for ill %p\n", 5657 (void *)ire->ire_in_ill)); 5658 } 5659 /* Check for duplicate and insert */ 5660 ASSERT(ire->ire_in_ill->ill_srcif_table != NULL); 5661 irb_ptr = 5662 &(ire->ire_in_ill->ill_srcif_table[IRE_ADDR_HASH(ire->ire_addr, 5663 IP_SRCIF_TABLE_SIZE)]); 5664 mutex_exit(&ire->ire_in_ill->ill_lock); 5665 flags = (MATCH_IRE_MASK | MATCH_IRE_TYPE | MATCH_IRE_GW); 5666 flags |= MATCH_IRE_IPIF; 5667 5668 /* 5669 * Start the atomic add of the ire. Grab the ill locks, 5670 * ill_g_usesrc_lock and the bucket lock. 5671 * 5672 * If ipif or ill is changing ire_atomic_start() may queue the 5673 * request and return EINPROGRESS. 5674 */ 5675 error = ire_atomic_start(irb_ptr, ire, q, mp, func); 5676 if (error != 0) { 5677 /* 5678 * We don't know whether it is a valid ipif or not. 5679 * So, set it to NULL. This assumes that the ire has not added 5680 * a reference to the ipif. 5681 */ 5682 ire->ire_ipif = NULL; 5683 ire_delete(ire); 5684 ip1dbg(("ire_add_srcif_v4: ire_atomic_start failed\n")); 5685 *ire_p = NULL; 5686 return (error); 5687 } 5688 for (ire1 = irb_ptr->irb_ire; ire1 != NULL; ire1 = ire1->ire_next) { 5689 if (ire1->ire_marks & IRE_MARK_CONDEMNED) 5690 continue; 5691 if (ire1->ire_zoneid != ire->ire_zoneid) 5692 continue; 5693 /* Has anyone inserted route in the meanwhile ? */ 5694 if (ire_match_args(ire1, ire->ire_addr, ire->ire_mask, 0, 5695 ire->ire_type, ire->ire_ipif, ire->ire_zoneid, 0, NULL, 5696 flags)) { 5697 ip1dbg(("ire_add_srcif_v4 : Duplicate entry exists\n")); 5698 IRE_REFHOLD(ire1); 5699 ire_atomic_end(irb_ptr, ire); 5700 ire_delete(ire); 5701 /* Return old ire as in ire_add_v4 */ 5702 *ire_p = ire1; 5703 return (0); 5704 } 5705 } 5706 irep = (ire_t **)irb_ptr; 5707 if (*irep != NULL) { 5708 /* Find the last ire which matches ire_addr */ 5709 ire1 = *irep; 5710 while (ire1->ire_addr == ire->ire_addr) { 5711 irep = &ire1->ire_next; 5712 ire1 = *irep; 5713 if (ire1 == NULL) 5714 break; 5715 } 5716 } 5717 ire1 = *irep; 5718 if (ire1 != NULL) 5719 ire1->ire_ptpn = &ire->ire_next; 5720 ire->ire_next = ire1; 5721 /* Link the new one in. */ 5722 ire->ire_ptpn = irep; 5723 membar_producer(); 5724 *irep = ire; 5725 ire->ire_bucket = irb_ptr; 5726 IRE_REFHOLD_LOCKED(ire); 5727 5728 /* 5729 * Protect ire_in_ill->ill_srcif_refcnt and table reference count. 5730 * Note, ire_atomic_start already grabs the ire_in_ill->ill_lock 5731 * so ill_srcif_refcnt is already protected. 5732 */ 5733 ire->ire_in_ill->ill_srcif_refcnt++; 5734 mutex_enter(&ire_srcif_table_lock); 5735 ire_srcif_table_count++; 5736 mutex_exit(&ire_srcif_table_lock); 5737 irb_ptr->irb_ire_cnt++; 5738 if (ire->ire_ipif != NULL) { 5739 ire->ire_ipif->ipif_ire_cnt++; 5740 if (ire->ire_stq != NULL) { 5741 stq_ill = (ill_t *)ire->ire_stq->q_ptr; 5742 stq_ill->ill_ire_cnt++; 5743 } 5744 } else { 5745 ASSERT(ire->ire_stq == NULL); 5746 } 5747 5748 ire_atomic_end(irb_ptr, ire); 5749 *ire_p = ire; 5750 return (0); 5751 } 5752 5753 5754 /* 5755 * This function is called by ire_add_then_send when ARP request comes 5756 * back to ip_wput->ire_add_then_send for resolved ire in the interface 5757 * based routing table. At this point, it only needs to update the resolver 5758 * information for the ire. The passed ire is returned to the caller as it 5759 * is the ire which is created as mblk. 5760 */ 5761 5762 static ire_t * 5763 ire_update_srcif_v4(ire_t *ire) 5764 { 5765 ire_t *ire1; 5766 irb_t *irb; 5767 int error; 5768 5769 ASSERT(ire->ire_type != IRE_MIPRTUN && 5770 ire->ire_ipif->ipif_net_type == IRE_IF_RESOLVER); 5771 ASSERT(ire->ire_ipversion == IPV4_VERSION); 5772 5773 /* 5774 * This ire is from ARP. Update 5775 * ire_nce->nce_res_mp info 5776 */ 5777 ire1 = ire_srcif_table_lookup(ire->ire_addr, 5778 IRE_IF_RESOLVER, ire->ire_ipif, 5779 ire->ire_in_ill, 5780 MATCH_IRE_ILL | MATCH_IRE_TYPE); 5781 if (ire1 == NULL) { 5782 /* Mobile node registration expired ? */ 5783 ire_delete(ire); 5784 return (NULL); 5785 } 5786 irb = ire1->ire_bucket; 5787 ASSERT(irb != NULL); 5788 /* 5789 * Start the atomic add of the ire. Grab the ill locks, 5790 * ill_g_usesrc_lock and the bucket lock. 5791 */ 5792 error = ire_atomic_start(irb, ire1, NULL, NULL, NULL); 5793 if (error != 0) { 5794 /* 5795 * We don't know whether it is a valid ipif or not. 5796 * So, set it to NULL. This assumes that the ire has not added 5797 * a reference to the ipif. 5798 */ 5799 ire->ire_ipif = NULL; 5800 ire_delete(ire); 5801 ip1dbg(("ire_update_srcif_v4: ire_atomic_start failed\n")); 5802 return (NULL); 5803 } 5804 ASSERT(ire->ire_max_fragp == NULL); 5805 ire->ire_max_frag = ire1->ire_max_frag; 5806 /* 5807 * Update resolver information and 5808 * send-to queue. 5809 */ 5810 ASSERT(ire->ire_nce->nce_res_mp != NULL); 5811 ire1->ire_nce->nce_res_mp = copyb(ire->ire_nce->nce_res_mp); 5812 if (ire1->ire_nce->nce_res_mp == NULL) { 5813 ip0dbg(("ire_update_srcif: copyb failed\n")); 5814 ire_refrele(ire1); 5815 ire_refrele(ire); 5816 ire_atomic_end(irb, ire1); 5817 return (NULL); 5818 } 5819 ire1->ire_stq = ire->ire_stq; 5820 5821 ASSERT(ire->ire_nce->nce_fp_mp == NULL); 5822 5823 ire_atomic_end(irb, ire1); 5824 ire_refrele(ire1); 5825 /* Return the passed ire */ 5826 return (ire); /* Update done */ 5827 } 5828 5829 5830 /* 5831 * Check if another multirt route resolution is needed. 5832 * B_TRUE is returned is there remain a resolvable route, 5833 * or if no route for that dst is resolved yet. 5834 * B_FALSE is returned if all routes for that dst are resolved 5835 * or if the remaining unresolved routes are actually not 5836 * resolvable. 5837 * This only works in the global zone. 5838 */ 5839 boolean_t 5840 ire_multirt_need_resolve(ipaddr_t dst, const ts_label_t *tsl) 5841 { 5842 ire_t *first_fire; 5843 ire_t *first_cire; 5844 ire_t *fire; 5845 ire_t *cire; 5846 irb_t *firb; 5847 irb_t *cirb; 5848 int unres_cnt = 0; 5849 boolean_t resolvable = B_FALSE; 5850 5851 /* Retrieve the first IRE_HOST that matches the destination */ 5852 first_fire = ire_ftable_lookup(dst, IP_HOST_MASK, 0, IRE_HOST, NULL, 5853 NULL, ALL_ZONES, 0, tsl, 5854 MATCH_IRE_MASK | MATCH_IRE_TYPE | MATCH_IRE_SECATTR); 5855 5856 /* No route at all */ 5857 if (first_fire == NULL) { 5858 return (B_TRUE); 5859 } 5860 5861 firb = first_fire->ire_bucket; 5862 ASSERT(firb != NULL); 5863 5864 /* Retrieve the first IRE_CACHE ire for that destination. */ 5865 first_cire = ire_cache_lookup(dst, GLOBAL_ZONEID, tsl); 5866 5867 /* No resolved route. */ 5868 if (first_cire == NULL) { 5869 ire_refrele(first_fire); 5870 return (B_TRUE); 5871 } 5872 5873 /* 5874 * At least one route is resolved. Here we look through the forward 5875 * and cache tables, to compare the number of declared routes 5876 * with the number of resolved routes. The search for a resolvable 5877 * route is performed only if at least one route remains 5878 * unresolved. 5879 */ 5880 cirb = first_cire->ire_bucket; 5881 ASSERT(cirb != NULL); 5882 5883 /* Count the number of routes to that dest that are declared. */ 5884 IRB_REFHOLD(firb); 5885 for (fire = first_fire; fire != NULL; fire = fire->ire_next) { 5886 if (!(fire->ire_flags & RTF_MULTIRT)) 5887 continue; 5888 if (fire->ire_addr != dst) 5889 continue; 5890 unres_cnt++; 5891 } 5892 IRB_REFRELE(firb); 5893 5894 /* Then subtract the number of routes to that dst that are resolved */ 5895 IRB_REFHOLD(cirb); 5896 for (cire = first_cire; cire != NULL; cire = cire->ire_next) { 5897 if (!(cire->ire_flags & RTF_MULTIRT)) 5898 continue; 5899 if (cire->ire_addr != dst) 5900 continue; 5901 if (cire->ire_marks & (IRE_MARK_CONDEMNED | IRE_MARK_HIDDEN)) 5902 continue; 5903 unres_cnt--; 5904 } 5905 IRB_REFRELE(cirb); 5906 5907 /* At least one route is unresolved; search for a resolvable route. */ 5908 if (unres_cnt > 0) 5909 resolvable = ire_multirt_lookup(&first_cire, &first_fire, 5910 MULTIRT_USESTAMP | MULTIRT_CACHEGW, tsl); 5911 5912 if (first_fire != NULL) 5913 ire_refrele(first_fire); 5914 5915 if (first_cire != NULL) 5916 ire_refrele(first_cire); 5917 5918 return (resolvable); 5919 } 5920 5921 5922 /* 5923 * Explore a forward_table bucket, starting from fire_arg. 5924 * fire_arg MUST be an IRE_HOST entry. 5925 * 5926 * Return B_TRUE and update *ire_arg and *fire_arg 5927 * if at least one resolvable route is found. *ire_arg 5928 * is the IRE entry for *fire_arg's gateway. 5929 * 5930 * Return B_FALSE otherwise (all routes are resolved or 5931 * the remaining unresolved routes are all unresolvable). 5932 * 5933 * The IRE selection relies on a priority mechanism 5934 * driven by the flags passed in by the caller. 5935 * The caller, such as ip_newroute_ipif(), can get the most 5936 * relevant ire at each stage of a multiple route resolution. 5937 * 5938 * The rules are: 5939 * 5940 * - if MULTIRT_CACHEGW is specified in flags, IRE_CACHETABLE 5941 * ires are preferred for the gateway. This gives the highest 5942 * priority to routes that can be resolved without using 5943 * a resolver. 5944 * 5945 * - if MULTIRT_CACHEGW is not specified, or if MULTIRT_CACHEGW 5946 * is specified but no IRE_CACHETABLE ire entry for the gateway 5947 * is found, the following rules apply. 5948 * 5949 * - if MULTIRT_USESTAMP is specified in flags, IRE_INTERFACE 5950 * ires for the gateway, that have not been tried since 5951 * a configurable amount of time, are preferred. 5952 * This applies when a resolver must be invoked for 5953 * a missing route, but we don't want to use the resolver 5954 * upon each packet emission. If no such resolver is found, 5955 * B_FALSE is returned. 5956 * The MULTIRT_USESTAMP flag can be combined with 5957 * MULTIRT_CACHEGW. 5958 * 5959 * - if MULTIRT_USESTAMP is not specified in flags, the first 5960 * unresolved but resolvable route is selected. 5961 * 5962 * - Otherwise, there is no resolvalble route, and 5963 * B_FALSE is returned. 5964 * 5965 * At last, MULTIRT_SETSTAMP can be specified in flags to 5966 * request the timestamp of unresolvable routes to 5967 * be refreshed. This prevents the useless exploration 5968 * of those routes for a while, when MULTIRT_USESTAMP is used. 5969 * 5970 * This only works in the global zone. 5971 */ 5972 boolean_t 5973 ire_multirt_lookup(ire_t **ire_arg, ire_t **fire_arg, uint32_t flags, 5974 const ts_label_t *tsl) 5975 { 5976 clock_t delta; 5977 ire_t *best_fire = NULL; 5978 ire_t *best_cire = NULL; 5979 ire_t *first_fire; 5980 ire_t *first_cire; 5981 ire_t *fire; 5982 ire_t *cire; 5983 irb_t *firb = NULL; 5984 irb_t *cirb = NULL; 5985 ire_t *gw_ire; 5986 boolean_t already_resolved; 5987 boolean_t res; 5988 ipaddr_t dst; 5989 ipaddr_t gw; 5990 5991 ip2dbg(("ire_multirt_lookup: *ire_arg %p, *fire_arg %p, flags %04x\n", 5992 (void *)*ire_arg, (void *)*fire_arg, flags)); 5993 5994 ASSERT(ire_arg != NULL); 5995 ASSERT(fire_arg != NULL); 5996 5997 /* Not an IRE_HOST ire; give up. */ 5998 if ((*fire_arg == NULL) || ((*fire_arg)->ire_type != IRE_HOST)) { 5999 return (B_FALSE); 6000 } 6001 6002 /* This is the first IRE_HOST ire for that destination. */ 6003 first_fire = *fire_arg; 6004 firb = first_fire->ire_bucket; 6005 ASSERT(firb != NULL); 6006 6007 dst = first_fire->ire_addr; 6008 6009 ip2dbg(("ire_multirt_lookup: dst %08x\n", ntohl(dst))); 6010 6011 /* 6012 * Retrieve the first IRE_CACHE ire for that destination; 6013 * if we don't find one, no route for that dest is 6014 * resolved yet. 6015 */ 6016 first_cire = ire_cache_lookup(dst, GLOBAL_ZONEID, tsl); 6017 if (first_cire != NULL) { 6018 cirb = first_cire->ire_bucket; 6019 } 6020 6021 ip2dbg(("ire_multirt_lookup: first_cire %p\n", (void *)first_cire)); 6022 6023 /* 6024 * Search for a resolvable route, giving the top priority 6025 * to routes that can be resolved without any call to the resolver. 6026 */ 6027 IRB_REFHOLD(firb); 6028 6029 if (!CLASSD(dst)) { 6030 /* 6031 * For all multiroute IRE_HOST ires for that destination, 6032 * check if the route via the IRE_HOST's gateway is 6033 * resolved yet. 6034 */ 6035 for (fire = first_fire; fire != NULL; fire = fire->ire_next) { 6036 6037 if (!(fire->ire_flags & RTF_MULTIRT)) 6038 continue; 6039 if (fire->ire_addr != dst) 6040 continue; 6041 6042 if (fire->ire_gw_secattr != NULL && 6043 tsol_ire_match_gwattr(fire, tsl) != 0) { 6044 continue; 6045 } 6046 6047 gw = fire->ire_gateway_addr; 6048 6049 ip2dbg(("ire_multirt_lookup: fire %p, " 6050 "ire_addr %08x, ire_gateway_addr %08x\n", 6051 (void *)fire, ntohl(fire->ire_addr), ntohl(gw))); 6052 6053 already_resolved = B_FALSE; 6054 6055 if (first_cire != NULL) { 6056 ASSERT(cirb != NULL); 6057 6058 IRB_REFHOLD(cirb); 6059 /* 6060 * For all IRE_CACHE ires for that 6061 * destination. 6062 */ 6063 for (cire = first_cire; 6064 cire != NULL; 6065 cire = cire->ire_next) { 6066 6067 if (!(cire->ire_flags & RTF_MULTIRT)) 6068 continue; 6069 if (cire->ire_addr != dst) 6070 continue; 6071 if (cire->ire_marks & 6072 (IRE_MARK_CONDEMNED | 6073 IRE_MARK_HIDDEN)) 6074 continue; 6075 6076 if (cire->ire_gw_secattr != NULL && 6077 tsol_ire_match_gwattr(cire, 6078 tsl) != 0) { 6079 continue; 6080 } 6081 6082 /* 6083 * Check if the IRE_CACHE's gateway 6084 * matches the IRE_HOST's gateway. 6085 */ 6086 if (cire->ire_gateway_addr == gw) { 6087 already_resolved = B_TRUE; 6088 break; 6089 } 6090 } 6091 IRB_REFRELE(cirb); 6092 } 6093 6094 /* 6095 * This route is already resolved; 6096 * proceed with next one. 6097 */ 6098 if (already_resolved) { 6099 ip2dbg(("ire_multirt_lookup: found cire %p, " 6100 "already resolved\n", (void *)cire)); 6101 continue; 6102 } 6103 6104 /* 6105 * The route is unresolved; is it actually 6106 * resolvable, i.e. is there a cache or a resolver 6107 * for the gateway? 6108 */ 6109 gw_ire = ire_route_lookup(gw, 0, 0, 0, NULL, NULL, 6110 ALL_ZONES, tsl, 6111 MATCH_IRE_RECURSIVE | MATCH_IRE_SECATTR); 6112 6113 ip2dbg(("ire_multirt_lookup: looked up gw_ire %p\n", 6114 (void *)gw_ire)); 6115 6116 /* 6117 * If gw_ire is typed IRE_CACHETABLE, 6118 * this route can be resolved without any call to the 6119 * resolver. If the MULTIRT_CACHEGW flag is set, 6120 * give the top priority to this ire and exit the 6121 * loop. 6122 * This is typically the case when an ARP reply 6123 * is processed through ip_wput_nondata(). 6124 */ 6125 if ((flags & MULTIRT_CACHEGW) && 6126 (gw_ire != NULL) && 6127 (gw_ire->ire_type & IRE_CACHETABLE)) { 6128 ASSERT(gw_ire->ire_nce == NULL || 6129 gw_ire->ire_nce->nce_state == ND_REACHABLE); 6130 /* 6131 * Release the resolver associated to the 6132 * previous candidate best ire, if any. 6133 */ 6134 if (best_cire != NULL) { 6135 ire_refrele(best_cire); 6136 ASSERT(best_fire != NULL); 6137 } 6138 6139 best_fire = fire; 6140 best_cire = gw_ire; 6141 6142 ip2dbg(("ire_multirt_lookup: found top prio " 6143 "best_fire %p, best_cire %p\n", 6144 (void *)best_fire, (void *)best_cire)); 6145 break; 6146 } 6147 6148 /* 6149 * Compute the time elapsed since our preceding 6150 * attempt to resolve that route. 6151 * If the MULTIRT_USESTAMP flag is set, we take that 6152 * route into account only if this time interval 6153 * exceeds ip_multirt_resolution_interval; 6154 * this prevents us from attempting to resolve a 6155 * broken route upon each sending of a packet. 6156 */ 6157 delta = lbolt - fire->ire_last_used_time; 6158 delta = TICK_TO_MSEC(delta); 6159 6160 res = (boolean_t) 6161 ((delta > ip_multirt_resolution_interval) || 6162 (!(flags & MULTIRT_USESTAMP))); 6163 6164 ip2dbg(("ire_multirt_lookup: fire %p, delta %lu, " 6165 "res %d\n", 6166 (void *)fire, delta, res)); 6167 6168 if (res) { 6169 /* 6170 * We are here if MULTIRT_USESTAMP flag is set 6171 * and the resolver for fire's gateway 6172 * has not been tried since 6173 * ip_multirt_resolution_interval, or if 6174 * MULTIRT_USESTAMP is not set but gw_ire did 6175 * not fill the conditions for MULTIRT_CACHEGW, 6176 * or if neither MULTIRT_USESTAMP nor 6177 * MULTIRT_CACHEGW are set. 6178 */ 6179 if (gw_ire != NULL) { 6180 if (best_fire == NULL) { 6181 ASSERT(best_cire == NULL); 6182 6183 best_fire = fire; 6184 best_cire = gw_ire; 6185 6186 ip2dbg(("ire_multirt_lookup:" 6187 "found candidate " 6188 "best_fire %p, " 6189 "best_cire %p\n", 6190 (void *)best_fire, 6191 (void *)best_cire)); 6192 6193 /* 6194 * If MULTIRT_CACHEGW is not 6195 * set, we ignore the top 6196 * priority ires that can 6197 * be resolved without any 6198 * call to the resolver; 6199 * In that case, there is 6200 * actually no need 6201 * to continue the loop. 6202 */ 6203 if (!(flags & 6204 MULTIRT_CACHEGW)) { 6205 break; 6206 } 6207 continue; 6208 } 6209 } else { 6210 /* 6211 * No resolver for the gateway: the 6212 * route is not resolvable. 6213 * If the MULTIRT_SETSTAMP flag is 6214 * set, we stamp the IRE_HOST ire, 6215 * so we will not select it again 6216 * during this resolution interval. 6217 */ 6218 if (flags & MULTIRT_SETSTAMP) 6219 fire->ire_last_used_time = 6220 lbolt; 6221 } 6222 } 6223 6224 if (gw_ire != NULL) 6225 ire_refrele(gw_ire); 6226 } 6227 } else { /* CLASSD(dst) */ 6228 6229 for (fire = first_fire; 6230 fire != NULL; 6231 fire = fire->ire_next) { 6232 6233 if (!(fire->ire_flags & RTF_MULTIRT)) 6234 continue; 6235 if (fire->ire_addr != dst) 6236 continue; 6237 6238 if (fire->ire_gw_secattr != NULL && 6239 tsol_ire_match_gwattr(fire, tsl) != 0) { 6240 continue; 6241 } 6242 6243 already_resolved = B_FALSE; 6244 6245 gw = fire->ire_gateway_addr; 6246 6247 gw_ire = ire_ftable_lookup(gw, 0, 0, IRE_INTERFACE, 6248 NULL, NULL, ALL_ZONES, 0, tsl, 6249 MATCH_IRE_RECURSIVE | MATCH_IRE_TYPE | 6250 MATCH_IRE_SECATTR); 6251 6252 /* No resolver for the gateway; we skip this ire. */ 6253 if (gw_ire == NULL) { 6254 continue; 6255 } 6256 ASSERT(gw_ire->ire_nce == NULL || 6257 gw_ire->ire_nce->nce_state == ND_REACHABLE); 6258 6259 if (first_cire != NULL) { 6260 6261 IRB_REFHOLD(cirb); 6262 /* 6263 * For all IRE_CACHE ires for that 6264 * destination. 6265 */ 6266 for (cire = first_cire; 6267 cire != NULL; 6268 cire = cire->ire_next) { 6269 6270 if (!(cire->ire_flags & RTF_MULTIRT)) 6271 continue; 6272 if (cire->ire_addr != dst) 6273 continue; 6274 if (cire->ire_marks & 6275 (IRE_MARK_CONDEMNED | 6276 IRE_MARK_HIDDEN)) 6277 continue; 6278 6279 if (cire->ire_gw_secattr != NULL && 6280 tsol_ire_match_gwattr(cire, 6281 tsl) != 0) { 6282 continue; 6283 } 6284 6285 /* 6286 * Cache entries are linked to the 6287 * parent routes using the parent handle 6288 * (ire_phandle). If no cache entry has 6289 * the same handle as fire, fire is 6290 * still unresolved. 6291 */ 6292 ASSERT(cire->ire_phandle != 0); 6293 if (cire->ire_phandle == 6294 fire->ire_phandle) { 6295 already_resolved = B_TRUE; 6296 break; 6297 } 6298 } 6299 IRB_REFRELE(cirb); 6300 } 6301 6302 /* 6303 * This route is already resolved; proceed with 6304 * next one. 6305 */ 6306 if (already_resolved) { 6307 ire_refrele(gw_ire); 6308 continue; 6309 } 6310 6311 /* 6312 * Compute the time elapsed since our preceding 6313 * attempt to resolve that route. 6314 * If the MULTIRT_USESTAMP flag is set, we take 6315 * that route into account only if this time 6316 * interval exceeds ip_multirt_resolution_interval; 6317 * this prevents us from attempting to resolve a 6318 * broken route upon each sending of a packet. 6319 */ 6320 delta = lbolt - fire->ire_last_used_time; 6321 delta = TICK_TO_MSEC(delta); 6322 6323 res = (boolean_t) 6324 ((delta > ip_multirt_resolution_interval) || 6325 (!(flags & MULTIRT_USESTAMP))); 6326 6327 ip3dbg(("ire_multirt_lookup: fire %p, delta %lx, " 6328 "flags %04x, res %d\n", 6329 (void *)fire, delta, flags, res)); 6330 6331 if (res) { 6332 if (best_cire != NULL) { 6333 /* 6334 * Release the resolver associated 6335 * to the preceding candidate best 6336 * ire, if any. 6337 */ 6338 ire_refrele(best_cire); 6339 ASSERT(best_fire != NULL); 6340 } 6341 best_fire = fire; 6342 best_cire = gw_ire; 6343 continue; 6344 } 6345 6346 ire_refrele(gw_ire); 6347 } 6348 } 6349 6350 if (best_fire != NULL) { 6351 IRE_REFHOLD(best_fire); 6352 } 6353 IRB_REFRELE(firb); 6354 6355 /* Release the first IRE_CACHE we initially looked up, if any. */ 6356 if (first_cire != NULL) 6357 ire_refrele(first_cire); 6358 6359 /* Found a resolvable route. */ 6360 if (best_fire != NULL) { 6361 ASSERT(best_cire != NULL); 6362 6363 if (*fire_arg != NULL) 6364 ire_refrele(*fire_arg); 6365 if (*ire_arg != NULL) 6366 ire_refrele(*ire_arg); 6367 6368 /* 6369 * Update the passed-in arguments with the 6370 * resolvable multirt route we found. 6371 */ 6372 *fire_arg = best_fire; 6373 *ire_arg = best_cire; 6374 6375 ip2dbg(("ire_multirt_lookup: returning B_TRUE, " 6376 "*fire_arg %p, *ire_arg %p\n", 6377 (void *)best_fire, (void *)best_cire)); 6378 6379 return (B_TRUE); 6380 } 6381 6382 ASSERT(best_cire == NULL); 6383 6384 ip2dbg(("ire_multirt_lookup: returning B_FALSE, *fire_arg %p, " 6385 "*ire_arg %p\n", 6386 (void *)*fire_arg, (void *)*ire_arg)); 6387 6388 /* No resolvable route. */ 6389 return (B_FALSE); 6390 } 6391 6392 /* 6393 * The purpose of the next two functions is to provide some external access to 6394 * routing/l2 lookup functionality while hiding the implementation of routing 6395 * and interface data structures (IRE/ILL). Thus, interfaces are passed/ 6396 * returned by name instead of by ILL reference. These functions are used by 6397 * IP Filter. 6398 * Return a link layer header suitable for an IP packet being sent to the 6399 * dst_addr IP address. The interface associated with the route is put into 6400 * ifname, which must be a buffer of LIFNAMSIZ bytes. The dst_addr is the 6401 * packet's ultimate destination address, not a router address. 6402 * 6403 * This function is used when the caller wants to know the outbound interface 6404 * and MAC header for a packet given only the address. 6405 */ 6406 mblk_t * 6407 ip_nexthop_route(const struct sockaddr *target, char *ifname) 6408 { 6409 struct nce_s *nce; 6410 ire_t *dir; 6411 ill_t *ill; 6412 mblk_t *mp, *tmp_mp; 6413 6414 /* parameter sanity */ 6415 if (ifname == NULL || target == NULL) 6416 return (NULL); 6417 6418 /* Find the route entry, if it exists. */ 6419 switch (target->sa_family) { 6420 case AF_INET: 6421 dir = ire_route_lookup( 6422 ((struct sockaddr_in *)target)->sin_addr.s_addr, 6423 0xffffffff, 6424 0, 0, NULL, NULL, ALL_ZONES, NULL, 6425 MATCH_IRE_DSTONLY|MATCH_IRE_DEFAULT|MATCH_IRE_RECURSIVE); 6426 break; 6427 case AF_INET6: 6428 dir = ire_route_lookup_v6( 6429 &((struct sockaddr_in6 *)target)->sin6_addr, 6430 NULL, 6431 0, 0, NULL, NULL, ALL_ZONES, NULL, 6432 MATCH_IRE_DSTONLY|MATCH_IRE_DEFAULT|MATCH_IRE_RECURSIVE); 6433 if ((dir != NULL) && (dir->ire_nce == NULL)) { 6434 ire_refrele(dir); 6435 dir = NULL; 6436 } 6437 break; 6438 default: 6439 dir = NULL; 6440 break; 6441 } 6442 6443 if (dir == NULL) { 6444 return (NULL); 6445 } 6446 6447 /* Map the IRE to an ILL so we can fill in ifname. */ 6448 ill = ire_to_ill(dir); 6449 if (ill == NULL) { 6450 ire_refrele(dir); 6451 return (NULL); 6452 } 6453 (void) strncpy(ifname, ill->ill_name, LIFNAMSIZ); 6454 6455 if ((dir->ire_type & (IRE_CACHE|IRE_BROADCAST)) == 0) { 6456 mp = copyb(ill->ill_resolver_mp); 6457 ire_refrele(dir); 6458 return (mp); 6459 } 6460 6461 /* Return a copy of the header to the caller. */ 6462 switch (target->sa_family) { 6463 case AF_INET : 6464 if (dir->ire_nce != NULL && 6465 dir->ire_nce->nce_state == ND_REACHABLE) { 6466 if (dir->ire_nce->nce_fp_mp != NULL) 6467 tmp_mp = dir->ire_nce->nce_fp_mp; 6468 else 6469 tmp_mp = dir->ire_nce->nce_fp_mp; 6470 if ((mp = dupb(tmp_mp)) == NULL) 6471 mp = copyb(tmp_mp); 6472 } else { 6473 mp = copyb(ill->ill_resolver_mp); 6474 } 6475 break; 6476 case AF_INET6 : 6477 nce = dir->ire_nce; 6478 if (nce->nce_fp_mp != NULL) { 6479 if ((mp = dupb(nce->nce_fp_mp)) == NULL) 6480 mp = copyb(nce->nce_fp_mp); 6481 } else if (nce->nce_res_mp != NULL) { 6482 if ((mp = dupb(nce->nce_res_mp)) == NULL) 6483 mp = copyb(nce->nce_res_mp); 6484 } else { 6485 mp = NULL; 6486 } 6487 break; 6488 } 6489 6490 ire_refrele(dir); 6491 return (mp); 6492 } 6493 6494 6495 /* 6496 * Return a link layer header suitable for an IP packet being sent to the 6497 * dst_addr IP address on the specified output interface. The dst_addr 6498 * may be the packet's ultimate destination or a predetermined next hop 6499 * router's address. 6500 * ifname must be nul-terminated. 6501 * 6502 * This function is used when the caller knows the outbound interface (usually 6503 * because it was specified by policy) and only needs the MAC header for a 6504 * packet. 6505 */ 6506 mblk_t * 6507 ip_nexthop(const struct sockaddr *target, const char *ifname) 6508 { 6509 struct nce_s *nce; 6510 ill_walk_context_t ctx; 6511 t_uscalar_t sap; 6512 ire_t *dir; 6513 ill_t *ill; 6514 mblk_t *mp; 6515 6516 /* parameter sanity */ 6517 if (ifname == NULL || target == NULL) 6518 return (NULL); 6519 6520 switch (target->sa_family) { 6521 case AF_INET : 6522 sap = IP_DL_SAP; 6523 break; 6524 case AF_INET6 : 6525 sap = IP6_DL_SAP; 6526 break; 6527 default: 6528 return (NULL); 6529 } 6530 6531 /* Lock ill_g_lock before walking through the list */ 6532 rw_enter(&ill_g_lock, RW_READER); 6533 /* 6534 * Can we find the interface name among those currently configured? 6535 */ 6536 for (ill = ILL_START_WALK_ALL(&ctx); ill != NULL; 6537 ill = ill_next(&ctx, ill)) { 6538 if ((strcmp(ifname, ill->ill_name) == 0) && 6539 (ill->ill_sap == sap)) 6540 break; 6541 } 6542 if (ill == NULL || ill->ill_ipif == NULL) { 6543 rw_exit(&ill_g_lock); 6544 return (NULL); 6545 } 6546 6547 mutex_enter(&ill->ill_lock); 6548 if (!ILL_CAN_LOOKUP(ill)) { 6549 mutex_exit(&ill->ill_lock); 6550 rw_exit(&ill_g_lock); 6551 return (NULL); 6552 } 6553 ill_refhold_locked(ill); 6554 mutex_exit(&ill->ill_lock); 6555 rw_exit(&ill_g_lock); 6556 6557 /* Find the resolver entry, if it exists. */ 6558 switch (target->sa_family) { 6559 case AF_INET: 6560 dir = ire_route_lookup( 6561 ((struct sockaddr_in *)target)->sin_addr.s_addr, 6562 0xffffffff, 6563 0, 0, ill->ill_ipif, NULL, ALL_ZONES, NULL, 6564 MATCH_IRE_DSTONLY|MATCH_IRE_DEFAULT| 6565 MATCH_IRE_RECURSIVE|MATCH_IRE_IPIF); 6566 if ((dir != NULL) && dir->ire_nce != NULL && 6567 dir->ire_nce->nce_state != ND_REACHABLE) { 6568 ire_refrele(dir); 6569 dir = NULL; 6570 } 6571 break; 6572 case AF_INET6: 6573 dir = ire_route_lookup_v6( 6574 &((struct sockaddr_in6 *)target)->sin6_addr, NULL, 6575 0, 0, ill->ill_ipif, NULL, ALL_ZONES, NULL, 6576 MATCH_IRE_DSTONLY|MATCH_IRE_DEFAULT| 6577 MATCH_IRE_RECURSIVE|MATCH_IRE_IPIF); 6578 if ((dir != NULL) && (dir->ire_nce == NULL)) { 6579 ire_refrele(dir); 6580 dir = NULL; 6581 } 6582 break; 6583 default: 6584 dir = NULL; 6585 break; 6586 } 6587 6588 if (dir == NULL) { 6589 return (NULL); 6590 } 6591 6592 if ((dir->ire_type & (IRE_CACHE|IRE_BROADCAST)) == 0) { 6593 mp = copyb(ill->ill_resolver_mp); 6594 ill_refrele(ill); 6595 ire_refrele(dir); 6596 return (mp); 6597 } 6598 6599 /* Return a copy of the header to the caller. */ 6600 switch (target->sa_family) { 6601 case AF_INET : 6602 if (dir->ire_nce->nce_fp_mp != NULL) { 6603 if ((mp = dupb(dir->ire_nce->nce_fp_mp)) == NULL) 6604 mp = copyb(dir->ire_nce->nce_fp_mp); 6605 } else if (dir->ire_nce->nce_res_mp != NULL) { 6606 if ((mp = dupb(dir->ire_nce->nce_res_mp)) == NULL) 6607 mp = copyb(dir->ire_nce->nce_res_mp); 6608 } else { 6609 mp = copyb(ill->ill_resolver_mp); 6610 } 6611 break; 6612 case AF_INET6 : 6613 nce = dir->ire_nce; 6614 if (nce->nce_fp_mp != NULL) { 6615 if ((mp = dupb(nce->nce_fp_mp)) == NULL) 6616 mp = copyb(nce->nce_fp_mp); 6617 } else if (nce->nce_res_mp != NULL) { 6618 if ((mp = dupb(nce->nce_res_mp)) == NULL) 6619 mp = copyb(nce->nce_res_mp); 6620 } else { 6621 mp = NULL; 6622 } 6623 break; 6624 } 6625 6626 ire_refrele(dir); 6627 ill_refrele(ill); 6628 return (mp); 6629 } 6630 6631 /* 6632 * IRE iterator for inbound and loopback broadcast processing. 6633 * Given an IRE_BROADCAST ire, walk the ires with the same destination 6634 * address, but skip over the passed-in ire. Returns the next ire without 6635 * a hold - assumes that the caller holds a reference on the IRE bucket. 6636 */ 6637 ire_t * 6638 ire_get_next_bcast_ire(ire_t *curr, ire_t *ire) 6639 { 6640 ill_t *ill; 6641 6642 if (curr == NULL) { 6643 for (curr = ire->ire_bucket->irb_ire; curr != NULL; 6644 curr = curr->ire_next) { 6645 if (curr->ire_addr == ire->ire_addr) 6646 break; 6647 } 6648 } else { 6649 curr = curr->ire_next; 6650 } 6651 ill = ire_to_ill(ire); 6652 for (; curr != NULL; curr = curr->ire_next) { 6653 if (curr->ire_addr != ire->ire_addr) { 6654 /* 6655 * All the IREs to a given destination are contiguous; 6656 * break out once the address doesn't match. 6657 */ 6658 break; 6659 } 6660 if (curr == ire) { 6661 /* skip over the passed-in ire */ 6662 continue; 6663 } 6664 if ((curr->ire_stq != NULL && ire->ire_stq == NULL) || 6665 (curr->ire_stq == NULL && ire->ire_stq != NULL)) { 6666 /* 6667 * If the passed-in ire is loopback, skip over 6668 * non-loopback ires and vice versa. 6669 */ 6670 continue; 6671 } 6672 if (ire_to_ill(curr) != ill) { 6673 /* skip over IREs going through a different interface */ 6674 continue; 6675 } 6676 if (curr->ire_marks & IRE_MARK_CONDEMNED) { 6677 /* skip over deleted IREs */ 6678 continue; 6679 } 6680 return (curr); 6681 } 6682 return (NULL); 6683 } 6684 6685 #ifdef IRE_DEBUG 6686 th_trace_t * 6687 th_trace_ire_lookup(ire_t *ire) 6688 { 6689 int bucket_id; 6690 th_trace_t *th_trace; 6691 6692 ASSERT(MUTEX_HELD(&ire->ire_lock)); 6693 6694 bucket_id = IP_TR_HASH(curthread); 6695 ASSERT(bucket_id < IP_TR_HASH_MAX); 6696 6697 for (th_trace = ire->ire_trace[bucket_id]; th_trace != NULL; 6698 th_trace = th_trace->th_next) { 6699 if (th_trace->th_id == curthread) 6700 return (th_trace); 6701 } 6702 return (NULL); 6703 } 6704 6705 void 6706 ire_trace_ref(ire_t *ire) 6707 { 6708 int bucket_id; 6709 th_trace_t *th_trace; 6710 6711 /* 6712 * Attempt to locate the trace buffer for the curthread. 6713 * If it does not exist, then allocate a new trace buffer 6714 * and link it in list of trace bufs for this ipif, at the head 6715 */ 6716 mutex_enter(&ire->ire_lock); 6717 if (ire->ire_trace_disable == B_TRUE) { 6718 mutex_exit(&ire->ire_lock); 6719 return; 6720 } 6721 th_trace = th_trace_ire_lookup(ire); 6722 if (th_trace == NULL) { 6723 bucket_id = IP_TR_HASH(curthread); 6724 th_trace = (th_trace_t *)kmem_zalloc(sizeof (th_trace_t), 6725 KM_NOSLEEP); 6726 if (th_trace == NULL) { 6727 ire->ire_trace_disable = B_TRUE; 6728 mutex_exit(&ire->ire_lock); 6729 ire_trace_inactive(ire); 6730 return; 6731 } 6732 6733 th_trace->th_id = curthread; 6734 th_trace->th_next = ire->ire_trace[bucket_id]; 6735 th_trace->th_prev = &ire->ire_trace[bucket_id]; 6736 if (th_trace->th_next != NULL) 6737 th_trace->th_next->th_prev = &th_trace->th_next; 6738 ire->ire_trace[bucket_id] = th_trace; 6739 } 6740 ASSERT(th_trace->th_refcnt < TR_BUF_MAX - 1); 6741 th_trace->th_refcnt++; 6742 th_trace_rrecord(th_trace); 6743 mutex_exit(&ire->ire_lock); 6744 } 6745 6746 void 6747 ire_trace_free(th_trace_t *th_trace) 6748 { 6749 /* unlink th_trace and free it */ 6750 *th_trace->th_prev = th_trace->th_next; 6751 if (th_trace->th_next != NULL) 6752 th_trace->th_next->th_prev = th_trace->th_prev; 6753 th_trace->th_next = NULL; 6754 th_trace->th_prev = NULL; 6755 kmem_free(th_trace, sizeof (th_trace_t)); 6756 } 6757 6758 void 6759 ire_untrace_ref(ire_t *ire) 6760 { 6761 th_trace_t *th_trace; 6762 6763 mutex_enter(&ire->ire_lock); 6764 6765 if (ire->ire_trace_disable == B_TRUE) { 6766 mutex_exit(&ire->ire_lock); 6767 return; 6768 } 6769 6770 th_trace = th_trace_ire_lookup(ire); 6771 ASSERT(th_trace != NULL && th_trace->th_refcnt > 0); 6772 th_trace_rrecord(th_trace); 6773 th_trace->th_refcnt--; 6774 6775 if (th_trace->th_refcnt == 0) 6776 ire_trace_free(th_trace); 6777 6778 mutex_exit(&ire->ire_lock); 6779 } 6780 6781 static void 6782 ire_trace_inactive(ire_t *ire) 6783 { 6784 th_trace_t *th_trace; 6785 int i; 6786 6787 mutex_enter(&ire->ire_lock); 6788 for (i = 0; i < IP_TR_HASH_MAX; i++) { 6789 while (ire->ire_trace[i] != NULL) { 6790 th_trace = ire->ire_trace[i]; 6791 6792 /* unlink th_trace and free it */ 6793 ire->ire_trace[i] = th_trace->th_next; 6794 if (th_trace->th_next != NULL) 6795 th_trace->th_next->th_prev = 6796 &ire->ire_trace[i]; 6797 6798 th_trace->th_next = NULL; 6799 th_trace->th_prev = NULL; 6800 kmem_free(th_trace, sizeof (th_trace_t)); 6801 } 6802 } 6803 6804 mutex_exit(&ire->ire_lock); 6805 } 6806 6807 /* ARGSUSED */ 6808 void 6809 ire_thread_exit(ire_t *ire, caddr_t arg) 6810 { 6811 th_trace_t *th_trace; 6812 6813 mutex_enter(&ire->ire_lock); 6814 th_trace = th_trace_ire_lookup(ire); 6815 if (th_trace == NULL) { 6816 mutex_exit(&ire->ire_lock); 6817 return; 6818 } 6819 ASSERT(th_trace->th_refcnt == 0); 6820 6821 ire_trace_free(th_trace); 6822 mutex_exit(&ire->ire_lock); 6823 } 6824 6825 #endif 6826 6827 /* 6828 * Generate a message chain with an arp request to resolve the in_ire. 6829 * It is assumed that in_ire itself is currently in the ire cache table, 6830 * so we create a fake_ire filled with enough information about ire_addr etc. 6831 * to retrieve in_ire when the DL_UNITDATA response from the resolver 6832 * comes back. The fake_ire itself is created by calling esballoc with 6833 * the fr_rtnp (free routine) set to ire_freemblk. This routine will be 6834 * invoked when the mblk containing fake_ire is freed. 6835 */ 6836 void 6837 ire_arpresolve(ire_t *in_ire, ill_t *dst_ill) 6838 { 6839 areq_t *areq; 6840 ipaddr_t *addrp; 6841 mblk_t *ire_mp, *dlureq_mp; 6842 ire_t *ire, *buf; 6843 size_t bufsize; 6844 frtn_t *frtnp; 6845 ill_t *ill; 6846 6847 /* 6848 * Construct message chain for the resolver 6849 * of the form: 6850 * ARP_REQ_MBLK-->IRE_MBLK 6851 * 6852 * NOTE : If the response does not 6853 * come back, ARP frees the packet. For this reason, 6854 * we can't REFHOLD the bucket of save_ire to prevent 6855 * deletions. We may not be able to REFRELE the bucket 6856 * if the response never comes back. Thus, before 6857 * adding the ire, ire_add_v4 will make sure that the 6858 * interface route does not get deleted. This is the 6859 * only case unlike ip_newroute_v6, ip_newroute_ipif_v6 6860 * where we can always prevent deletions because of 6861 * the synchronous nature of adding IRES i.e 6862 * ire_add_then_send is called after creating the IRE. 6863 */ 6864 6865 /* 6866 * We use esballoc to allocate the second part(the ire_t size mblk) 6867 * of the message chain depicted above. THis mblk will be freed 6868 * by arp when there is a timeout, and otherwise passed to IP 6869 * and IP will * free it after processing the ARP response. 6870 */ 6871 6872 bufsize = sizeof (ire_t) + sizeof (frtn_t); 6873 buf = kmem_alloc(bufsize, KM_NOSLEEP); 6874 if (buf == NULL) { 6875 ip1dbg(("ire_arpresolver:alloc buffer failed\n ")); 6876 return; 6877 } 6878 frtnp = (frtn_t *)(buf + 1); 6879 frtnp->free_arg = (caddr_t)buf; 6880 frtnp->free_func = ire_freemblk; 6881 6882 ire_mp = esballoc((unsigned char *)buf, bufsize, BPRI_MED, frtnp); 6883 6884 if (ire_mp == NULL) { 6885 ip1dbg(("ire_arpresolve: esballoc failed\n")); 6886 kmem_free(buf, bufsize); 6887 return; 6888 } 6889 ASSERT(in_ire->ire_nce != NULL); 6890 dlureq_mp = copyb(dst_ill->ill_resolver_mp); 6891 if (dlureq_mp == NULL) { 6892 kmem_free(buf, bufsize); 6893 return; 6894 } 6895 6896 ire_mp->b_datap->db_type = IRE_ARPRESOLVE_TYPE; 6897 ire = (ire_t *)buf; 6898 /* 6899 * keep enough info in the fake ire so that we can pull up 6900 * the incomplete ire (in_ire) after result comes back from 6901 * arp and make it complete. 6902 */ 6903 *ire = ire_null; 6904 ire->ire_u = in_ire->ire_u; 6905 ire->ire_ipif_seqid = in_ire->ire_ipif_seqid; 6906 ire->ire_ipif = in_ire->ire_ipif; 6907 ire->ire_stq = in_ire->ire_stq; 6908 ill = ire_to_ill(ire); 6909 ire->ire_stq_ifindex = ill->ill_phyint->phyint_ifindex; 6910 ire->ire_zoneid = in_ire->ire_zoneid; 6911 /* 6912 * ire_freemblk will be called when ire_mp is freed, both for 6913 * successful and failed arp resolution. IRE_MARK_UNCACHED will be set 6914 * when the arp resolution failed. 6915 */ 6916 ire->ire_marks |= IRE_MARK_UNCACHED; 6917 ire->ire_mp = ire_mp; 6918 ire_mp->b_wptr = (uchar_t *)&ire[1]; 6919 ire_mp->b_cont = NULL; 6920 ASSERT(dlureq_mp != NULL); 6921 linkb(dlureq_mp, ire_mp); 6922 6923 /* 6924 * Fill in the source and dest addrs for the resolver. 6925 * NOTE: this depends on memory layouts imposed by 6926 * ill_init(). 6927 */ 6928 areq = (areq_t *)dlureq_mp->b_rptr; 6929 addrp = (ipaddr_t *)((char *)areq + areq->areq_sender_addr_offset); 6930 *addrp = ire->ire_src_addr; 6931 6932 addrp = (ipaddr_t *)((char *)areq + areq->areq_target_addr_offset); 6933 if (ire->ire_gateway_addr != INADDR_ANY) { 6934 *addrp = ire->ire_gateway_addr; 6935 } else { 6936 *addrp = ire->ire_addr; 6937 } 6938 6939 /* Up to the resolver. */ 6940 if (canputnext(dst_ill->ill_rq)) { 6941 putnext(dst_ill->ill_rq, dlureq_mp); 6942 } else { 6943 /* Prepare for cleanup */ 6944 freemsg(dlureq_mp); 6945 } 6946 } 6947 6948 /* 6949 * Esballoc free function for AR_ENTRY_QUERY request to clean up any 6950 * unresolved ire_t and/or nce_t structures when ARP resolution fails. 6951 * 6952 * This function can be called by ARP via free routine for ire_mp or 6953 * by IPv4(both host and forwarding path) via ire_delete 6954 * in case ARP resolution fails. 6955 * NOTE: Since IP is MT, ARP can call into IP but not vice versa 6956 * (for IP to talk to ARP, it still has to send AR* messages). 6957 * 6958 * Note that the ARP/IP merge should replace the functioanlity by providing 6959 * direct function calls to clean up unresolved entries in ire/nce lists. 6960 */ 6961 void 6962 ire_freemblk(ire_t *ire_mp) 6963 { 6964 nce_t *nce = NULL; 6965 ill_t *ill; 6966 6967 ASSERT(ire_mp != NULL); 6968 6969 if ((ire_mp->ire_addr == NULL) && (ire_mp->ire_gateway_addr == NULL)) { 6970 ip1dbg(("ire_freemblk(0x%p) ire_addr is NULL\n", 6971 (void *)ire_mp)); 6972 goto cleanup; 6973 } 6974 if ((ire_mp->ire_marks & IRE_MARK_UNCACHED) == 0) { 6975 goto cleanup; /* everything succeeded. just free and return */ 6976 } 6977 6978 /* 6979 * the arp information corresponding to this ire_mp was not 6980 * transferred to a ire_cache entry. Need 6981 * to clean up incomplete ire's and nce, if necessary. 6982 */ 6983 ASSERT(ire_mp->ire_stq != NULL); 6984 ASSERT(ire_mp->ire_stq_ifindex != 0); 6985 /* 6986 * Get any nce's corresponding to this ire_mp. We first have to 6987 * make sure that the ill is still around. 6988 */ 6989 ill = ill_lookup_on_ifindex(ire_mp->ire_stq_ifindex, B_FALSE, 6990 NULL, NULL, NULL, NULL); 6991 if (ill == NULL || (ire_mp->ire_stq != ill->ill_wq) || 6992 (ill->ill_state_flags & ILL_CONDEMNED)) { 6993 /* 6994 * ill went away. no nce to clean up. 6995 * Note that the ill_state_flags could be set to 6996 * ILL_CONDEMNED after this point, but if we know 6997 * that it is CONDEMNED now, we just bail out quickly. 6998 */ 6999 if (ill != NULL) 7000 ill_refrele(ill); 7001 goto cleanup; 7002 } 7003 nce = ndp_lookup_v4(ill, 7004 ((ire_mp->ire_gateway_addr != INADDR_ANY) ? 7005 &ire_mp->ire_gateway_addr : &ire_mp->ire_addr), 7006 B_FALSE); 7007 ill_refrele(ill); 7008 7009 if ((nce != NULL) && (nce->nce_state != ND_REACHABLE)) { 7010 /* 7011 * some incomplete nce was found. 7012 */ 7013 DTRACE_PROBE2(ire__freemblk__arp__resolv__fail, 7014 nce_t *, nce, ire_t *, ire_mp); 7015 /* 7016 * Send the icmp_unreachable messages for the queued mblks in 7017 * ire->ire_nce->nce_qd_mp, since ARP resolution failed 7018 * for this ire 7019 */ 7020 arp_resolv_failed(nce); 7021 /* 7022 * Delete the nce and clean up all ire's pointing at this nce 7023 * in the cachetable 7024 */ 7025 ndp_delete(nce); 7026 } 7027 if (nce != NULL) 7028 NCE_REFRELE(nce); /* release the ref taken by ndp_lookup_v4 */ 7029 7030 cleanup: 7031 /* 7032 * Get rid of the ire buffer 7033 * We call kmem_free here(instead of ire_delete()), since 7034 * this is the freeb's callback. 7035 */ 7036 kmem_free(ire_mp, sizeof (ire_t) + sizeof (frtn_t)); 7037 } 7038 7039 7040 /* 7041 * create the neighbor cache entry nce_t for IRE_CACHE and 7042 * non-loopback IRE_BROADCAST ire's. Note that IRE_BROADCAST 7043 * (non-loopback) entries have the nce_res_mp set to the 7044 * template passed in (generated from ill_bcast_mp); IRE_CACHE ire's 7045 * contain the information for the nexthop (ire_gateway_addr) in the 7046 * case of indirect routes, and for the dst itself (ire_addr) in the 7047 * case of direct routes, with the nce_res_mp containing a template 7048 * DL_UNITDATA request. 7049 * 7050 * This function always consumes res_mp and fp_mp. 7051 * 7052 * The actual association of the ire_nce to the nce created here is 7053 * typically done in ire_add_v4 for IRE_CACHE entries. Exceptions 7054 * to this rule are SO_DONTROUTE ire's (IRE_MARK_NO_ADD), for which 7055 * the ire_nce assignment is done in ire_add_then_send, and mobile-ip 7056 * where the assignment is done in ire_add_mrtun(). 7057 */ 7058 int 7059 ire_nce_init(ire_t *ire, mblk_t *fp_mp, mblk_t *res_mp) 7060 { 7061 in_addr_t addr4, mask4; 7062 int err; 7063 nce_t *arpce = NULL; 7064 ill_t *ire_ill; 7065 uint16_t nce_state, nce_flags; 7066 7067 if (ire->ire_stq == NULL) { 7068 if (res_mp) 7069 freemsg(res_mp); 7070 if (fp_mp) 7071 freemsg(fp_mp); 7072 return (0); /* no need to create nce for local/loopback */ 7073 } 7074 7075 mask4 = IP_HOST_MASK; 7076 switch (ire->ire_type) { 7077 case IRE_CACHE: 7078 if (ire->ire_gateway_addr != INADDR_ANY) 7079 addr4 = ire->ire_gateway_addr; /* 'G' route */ 7080 else 7081 addr4 = ire->ire_addr; /* direct route */ 7082 break; 7083 case IRE_BROADCAST: 7084 addr4 = ire->ire_addr; 7085 break; 7086 default: 7087 if (res_mp) 7088 freemsg(res_mp); 7089 if (fp_mp) 7090 freemsg(fp_mp); 7091 return (0); 7092 } 7093 7094 /* 7095 * ire_ipif is picked based on RTF_SETSRC, usesrc etc. 7096 * rules in ire_forward_src_ipif. We want the dlureq_mp 7097 * for the outgoing interface, which we get from the ire_stq. 7098 */ 7099 ire_ill = ire_to_ill(ire); 7100 7101 /* 7102 * if we are creating an nce for the first time, and this is 7103 * a NORESOLVER interface, atomically create the nce in the 7104 * REACHABLE state; else create it in the ND_INITIAL state. 7105 */ 7106 if (ire_ill->ill_net_type == IRE_IF_NORESOLVER) { 7107 nce_state = ND_REACHABLE; 7108 nce_flags = NCE_F_PERMANENT; 7109 } else { 7110 if (fp_mp != NULL) 7111 nce_state = ND_REACHABLE; 7112 else 7113 nce_state = ND_INITIAL; 7114 nce_flags = 0; 7115 } 7116 7117 err = ndp_lookup_then_add(ire_ill, NULL, 7118 &addr4, &mask4, NULL, 0, nce_flags, nce_state, &arpce, 7119 fp_mp, res_mp); 7120 7121 ip1dbg(("ire 0x%p addr 0x%lx mask 0x%lx type 0x%x; " 7122 "found nce 0x%p err %d\n", (void *)ire, (ulong_t)addr4, 7123 (ulong_t)mask4, ire->ire_type, (void *)arpce, err)); 7124 7125 switch (err) { 7126 case 0: 7127 break; 7128 case EEXIST: 7129 /* 7130 * return a pointer to an existing nce_t; 7131 * note that the ire-nce mapping is many-one, i.e., 7132 * multiple ire's could point to the same nce_t; 7133 */ 7134 if (fp_mp != NULL) { 7135 freemsg(fp_mp); 7136 } 7137 if (res_mp != NULL) { 7138 freemsg(res_mp); 7139 } 7140 break; 7141 default: 7142 DTRACE_PROBE2(nce__init__fail, ill_t *, ire_ill, int, err); 7143 if (res_mp) 7144 freemsg(res_mp); 7145 if (fp_mp) 7146 freemsg(fp_mp); 7147 return (EINVAL); 7148 } 7149 #if DEBUG 7150 /* 7151 * if an nce_fp_mp was passed in, we should be picking up an 7152 * existing nce_t in the ND_REACHABLE state. 7153 */ 7154 mutex_enter(&arpce->nce_lock); 7155 ASSERT(arpce->nce_fp_mp == NULL || arpce->nce_state == ND_REACHABLE); 7156 mutex_exit(&arpce->nce_lock); 7157 #endif 7158 if (ire->ire_type == IRE_BROADCAST) { 7159 /* 7160 * Two bcast ires are created for each interface; 7161 * 1. loopback copy (which does not have an 7162 * ire_stq, and therefore has no ire_nce), and, 7163 * 2. the non-loopback copy, which has the nce_res_mp 7164 * initialized to a copy of the ill_bcast_mp, and 7165 * is marked as ND_REACHABLE at this point. 7166 * This nce does not undergo any further state changes, 7167 * and exists as long as the interface is plumbed. 7168 * Note: we do the ire_nce assignment here for IRE_BROADCAST 7169 * because some functions like ill_mark_bcast() inline the 7170 * ire_add functionality; 7171 */ 7172 mutex_enter(&arpce->nce_lock); 7173 arpce->nce_state = ND_REACHABLE; 7174 arpce->nce_flags |= NCE_F_PERMANENT; 7175 arpce->nce_last = TICK_TO_MSEC(lbolt64); 7176 ire->ire_nce = arpce; 7177 mutex_exit(&arpce->nce_lock); 7178 /* 7179 * We are associating this nce to the ire, 7180 * so change the nce ref taken in 7181 * ndp_lookup_then_add_v4() from 7182 * NCE_REFHOLD to NCE_REFHOLD_NOTR 7183 */ 7184 NCE_REFHOLD_TO_REFHOLD_NOTR(ire->ire_nce); 7185 } else { 7186 if (NCE_EXPIRED(arpce)) 7187 arpce = nce_reinit(arpce); 7188 if (arpce != NULL) { 7189 /* 7190 * We are not using this nce_t just yet so release 7191 * the ref taken in ndp_lookup_then_add_v4() 7192 */ 7193 NCE_REFRELE(arpce); 7194 } else { 7195 ip0dbg(("can't reinit arpce for ill 0x%p;\n", 7196 (void *)ire_ill)); 7197 } 7198 } 7199 return (0); 7200 } 7201