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 26 #pragma ident "%Z%%M% %I% %E% SMI" 27 28 /* 29 * This file contains consumer routines of the IPv4 forwarding engine 30 */ 31 32 #include <sys/types.h> 33 #include <sys/stream.h> 34 #include <sys/stropts.h> 35 #include <sys/strlog.h> 36 #include <sys/dlpi.h> 37 #include <sys/ddi.h> 38 #include <sys/cmn_err.h> 39 #include <sys/policy.h> 40 41 #include <sys/systm.h> 42 #include <sys/strsun.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/mib2.h> 56 #include <inet/ip.h> 57 #include <inet/ip6.h> 58 #include <inet/ip_ndp.h> 59 #include <inet/arp.h> 60 #include <inet/ip_if.h> 61 #include <inet/ip_ire.h> 62 #include <inet/ip_ftable.h> 63 #include <inet/ip_rts.h> 64 #include <inet/nd.h> 65 66 #include <net/pfkeyv2.h> 67 #include <inet/ipsec_info.h> 68 #include <inet/sadb.h> 69 #include <sys/kmem.h> 70 #include <inet/tcp.h> 71 #include <inet/ipclassifier.h> 72 #include <sys/zone.h> 73 #include <net/radix.h> 74 #include <sys/tsol/label.h> 75 #include <sys/tsol/tnet.h> 76 77 #define IS_DEFAULT_ROUTE(ire) \ 78 (((ire)->ire_type & IRE_DEFAULT) || \ 79 (((ire)->ire_type & IRE_INTERFACE) && ((ire)->ire_addr == 0))) 80 81 /* 82 * structure for passing args between ire_ftable_lookup and ire_find_best_route 83 */ 84 typedef struct ire_ftable_args_s { 85 ipaddr_t ift_addr; 86 ipaddr_t ift_mask; 87 ipaddr_t ift_gateway; 88 int ift_type; 89 const ipif_t *ift_ipif; 90 zoneid_t ift_zoneid; 91 uint32_t ift_ihandle; 92 const ts_label_t *ift_tsl; 93 int ift_flags; 94 ire_t *ift_best_ire; 95 } ire_ftable_args_t; 96 97 struct radix_node_head *ip_ftable; 98 static ire_t *route_to_dst(const struct sockaddr *, zoneid_t); 99 static ire_t *ire_round_robin(irb_t *, zoneid_t, ire_ftable_args_t *); 100 static void ire_del_host_redir(ire_t *, char *); 101 static boolean_t ire_find_best_route(struct radix_node *, void *); 102 103 /* 104 * Lookup a route in forwarding table. A specific lookup is indicated by 105 * passing the required parameters and indicating the match required in the 106 * flag field. 107 * 108 * Looking for default route can be done in three ways 109 * 1) pass mask as 0 and set MATCH_IRE_MASK in flags field 110 * along with other matches. 111 * 2) pass type as IRE_DEFAULT and set MATCH_IRE_TYPE in flags 112 * field along with other matches. 113 * 3) if the destination and mask are passed as zeros. 114 * 115 * A request to return a default route if no route 116 * is found, can be specified by setting MATCH_IRE_DEFAULT 117 * in flags. 118 * 119 * It does not support recursion more than one level. It 120 * will do recursive lookup only when the lookup maps to 121 * a prefix or default route and MATCH_IRE_RECURSIVE flag is passed. 122 * 123 * If the routing table is setup to allow more than one level 124 * of recursion, the cleaning up cache table will not work resulting 125 * in invalid routing. 126 * 127 * Supports IP_BOUND_IF by following the ipif/ill when recursing. 128 * 129 * NOTE : When this function returns NULL, pire has already been released. 130 * pire is valid only when this function successfully returns an 131 * ire. 132 */ 133 ire_t * 134 ire_ftable_lookup(ipaddr_t addr, ipaddr_t mask, ipaddr_t gateway, 135 int type, const ipif_t *ipif, ire_t **pire, zoneid_t zoneid, 136 uint32_t ihandle, const ts_label_t *tsl, int flags) 137 { 138 ire_t *ire = NULL; 139 ipaddr_t gw_addr; 140 struct rt_sockaddr rdst, rmask; 141 struct rt_entry *rt; 142 ire_ftable_args_t margs; 143 boolean_t found_incomplete = B_FALSE; 144 145 ASSERT(ipif == NULL || !ipif->ipif_isv6); 146 ASSERT(!(flags & MATCH_IRE_WQ)); 147 148 /* 149 * When we return NULL from this function, we should make 150 * sure that *pire is NULL so that the callers will not 151 * wrongly REFRELE the pire. 152 */ 153 if (pire != NULL) 154 *pire = NULL; 155 /* 156 * ire_match_args() will dereference ipif MATCH_IRE_SRC or 157 * MATCH_IRE_ILL is set. 158 */ 159 if ((flags & (MATCH_IRE_SRC | MATCH_IRE_ILL | MATCH_IRE_ILL_GROUP)) && 160 (ipif == NULL)) 161 return (NULL); 162 163 (void) memset(&rdst, 0, sizeof (rdst)); 164 rdst.rt_sin_len = sizeof (rdst); 165 rdst.rt_sin_family = AF_INET; 166 rdst.rt_sin_addr.s_addr = addr; 167 168 (void) memset(&rmask, 0, sizeof (rmask)); 169 rmask.rt_sin_len = sizeof (rmask); 170 rmask.rt_sin_family = AF_INET; 171 rmask.rt_sin_addr.s_addr = mask; 172 173 (void) memset(&margs, 0, sizeof (margs)); 174 margs.ift_addr = addr; 175 margs.ift_mask = mask; 176 margs.ift_gateway = gateway; 177 margs.ift_type = type; 178 margs.ift_ipif = ipif; 179 margs.ift_zoneid = zoneid; 180 margs.ift_ihandle = ihandle; 181 margs.ift_tsl = tsl; 182 margs.ift_flags = flags; 183 184 /* 185 * The flags argument passed to ire_ftable_lookup may cause the 186 * search to return, not the longest matching prefix, but the 187 * "best matching prefix", i.e., the longest prefix that also 188 * satisfies constraints imposed via the permutation of flags 189 * passed in. To achieve this, we invoke ire_match_args() on 190 * each matching leaf in the radix tree. ire_match_args is 191 * invoked by the callback function ire_find_best_route() 192 * We hold the global tree lock in read mode when calling 193 * rn_match_args.Before dropping the global tree lock, ensure 194 * that the radix node can't be deleted by incrementing ire_refcnt. 195 */ 196 RADIX_NODE_HEAD_RLOCK(ip_ftable); 197 rt = (struct rt_entry *)ip_ftable->rnh_matchaddr_args(&rdst, ip_ftable, 198 ire_find_best_route, &margs); 199 ire = margs.ift_best_ire; 200 RADIX_NODE_HEAD_UNLOCK(ip_ftable); 201 202 if (rt == NULL) { 203 return (NULL); 204 } else { 205 ASSERT(ire != NULL); 206 } 207 208 DTRACE_PROBE2(ire__found, ire_ftable_args_t *, &margs, ire_t *, ire); 209 210 if (!IS_DEFAULT_ROUTE(ire)) 211 goto found_ire_held; 212 /* 213 * If default route is found, see if default matching criteria 214 * are satisfied. 215 */ 216 if (flags & MATCH_IRE_MASK) { 217 /* 218 * we were asked to match a 0 mask, and came back with 219 * a default route. Ok to return it. 220 */ 221 goto found_default_ire; 222 } 223 if ((flags & MATCH_IRE_TYPE) && 224 (type & (IRE_DEFAULT | IRE_INTERFACE))) { 225 /* 226 * we were asked to match a default ire type. Ok to return it. 227 */ 228 goto found_default_ire; 229 } 230 if (flags & MATCH_IRE_DEFAULT) { 231 goto found_default_ire; 232 } 233 /* 234 * we found a default route, but default matching criteria 235 * are not specified and we are not explicitly looking for 236 * default. 237 */ 238 IRE_REFRELE(ire); 239 return (NULL); 240 found_default_ire: 241 /* 242 * round-robin only if we have more than one route in the bucket. 243 */ 244 if ((ire->ire_bucket->irb_ire_cnt > 1) && 245 IS_DEFAULT_ROUTE(ire) && 246 ((flags & (MATCH_IRE_DEFAULT | MATCH_IRE_MASK)) == 247 MATCH_IRE_DEFAULT)) { 248 ire_t *next_ire; 249 250 next_ire = ire_round_robin(ire->ire_bucket, zoneid, &margs); 251 IRE_REFRELE(ire); 252 if (next_ire != NULL) { 253 ire = next_ire; 254 } else { 255 /* no route */ 256 return (NULL); 257 } 258 } 259 found_ire_held: 260 ASSERT(ire->ire_type != IRE_MIPRTUN && ire->ire_in_ill == NULL); 261 if ((flags & MATCH_IRE_RJ_BHOLE) && 262 (ire->ire_flags & (RTF_BLACKHOLE | RTF_REJECT))) { 263 return (ire); 264 } 265 /* 266 * At this point, IRE that was found must be an IRE_FORWARDTABLE 267 * type. If this is a recursive lookup and an IRE_INTERFACE type was 268 * found, return that. If it was some other IRE_FORWARDTABLE type of 269 * IRE (one of the prefix types), then it is necessary to fill in the 270 * parent IRE pointed to by pire, and then lookup the gateway address of 271 * the parent. For backwards compatiblity, if this lookup returns an 272 * IRE other than a IRE_CACHETABLE or IRE_INTERFACE, then one more level 273 * of lookup is done. 274 */ 275 if (flags & MATCH_IRE_RECURSIVE) { 276 ipif_t *gw_ipif; 277 int match_flags = MATCH_IRE_DSTONLY; 278 ire_t *save_ire; 279 280 if (ire->ire_type & IRE_INTERFACE) 281 return (ire); 282 if (pire != NULL) 283 *pire = ire; 284 /* 285 * If we can't find an IRE_INTERFACE or the caller has not 286 * asked for pire, we need to REFRELE the save_ire. 287 */ 288 save_ire = ire; 289 290 /* 291 * Currently MATCH_IRE_ILL is never used with 292 * (MATCH_IRE_RECURSIVE | MATCH_IRE_DEFAULT) while 293 * sending out packets as MATCH_IRE_ILL is used only 294 * for communicating with on-link hosts. We can't assert 295 * that here as RTM_GET calls this function with 296 * MATCH_IRE_ILL | MATCH_IRE_DEFAULT | MATCH_IRE_RECURSIVE. 297 * We have already used the MATCH_IRE_ILL in determining 298 * the right prefix route at this point. To match the 299 * behavior of how we locate routes while sending out 300 * packets, we don't want to use MATCH_IRE_ILL below 301 * while locating the interface route. 302 * 303 * ire_ftable_lookup may end up with an incomplete IRE_CACHE 304 * entry for the gateway (i.e., one for which the 305 * ire_nce->nce_state is not yet ND_REACHABLE). If the caller 306 * has specified MATCH_IRE_COMPLETE, such entries will not 307 * be returned; instead, we return the IF_RESOLVER ire. 308 */ 309 if (ire->ire_ipif != NULL) 310 match_flags |= MATCH_IRE_ILL_GROUP; 311 312 ire = ire_route_lookup(ire->ire_gateway_addr, 0, 0, 0, 313 ire->ire_ipif, NULL, zoneid, tsl, match_flags); 314 DTRACE_PROBE2(ftable__route__lookup1, (ire_t *), ire, 315 (ire_t *), save_ire); 316 if (ire == NULL || 317 ((ire->ire_type & IRE_CACHE) && ire->ire_nce && 318 ire->ire_nce->nce_state != ND_REACHABLE && 319 (flags & MATCH_IRE_COMPLETE))) { 320 /* 321 * Do not release the parent ire if MATCH_IRE_PARENT 322 * is set. Also return it via ire. 323 */ 324 if (ire != NULL) { 325 ire_refrele(ire); 326 ire = NULL; 327 found_incomplete = B_TRUE; 328 } 329 if (flags & MATCH_IRE_PARENT) { 330 if (pire != NULL) { 331 /* 332 * Need an extra REFHOLD, if the parent 333 * ire is returned via both ire and 334 * pire. 335 */ 336 IRE_REFHOLD(save_ire); 337 } 338 ire = save_ire; 339 } else { 340 ire_refrele(save_ire); 341 if (pire != NULL) 342 *pire = NULL; 343 } 344 if (!found_incomplete) 345 return (ire); 346 } 347 if (ire->ire_type & (IRE_CACHETABLE | IRE_INTERFACE)) { 348 /* 349 * If the caller did not ask for pire, release 350 * it now. 351 */ 352 if (pire == NULL) { 353 ire_refrele(save_ire); 354 } 355 return (ire); 356 } 357 match_flags |= MATCH_IRE_TYPE; 358 gw_addr = ire->ire_gateway_addr; 359 gw_ipif = ire->ire_ipif; 360 ire_refrele(ire); 361 ire = ire_route_lookup(gw_addr, 0, 0, 362 (found_incomplete? IRE_INTERFACE : 363 (IRE_CACHETABLE | IRE_INTERFACE)), 364 gw_ipif, NULL, zoneid, tsl, match_flags); 365 DTRACE_PROBE2(ftable__route__lookup2, (ire_t *), ire, 366 (ire_t *), save_ire); 367 if (ire == NULL || 368 ((ire->ire_type & IRE_CACHE) && ire->ire_nce && 369 ire->ire_nce->nce_state != ND_REACHABLE && 370 (flags & MATCH_IRE_COMPLETE))) { 371 /* 372 * Do not release the parent ire if MATCH_IRE_PARENT 373 * is set. Also return it via ire. 374 */ 375 if (ire != NULL) { 376 ire_refrele(ire); 377 ire = NULL; 378 } 379 if (flags & MATCH_IRE_PARENT) { 380 if (pire != NULL) { 381 /* 382 * Need an extra REFHOLD, if the 383 * parent ire is returned via both 384 * ire and pire. 385 */ 386 IRE_REFHOLD(save_ire); 387 } 388 ire = save_ire; 389 } else { 390 ire_refrele(save_ire); 391 if (pire != NULL) 392 *pire = NULL; 393 } 394 return (ire); 395 } else if (pire == NULL) { 396 /* 397 * If the caller did not ask for pire, release 398 * it now. 399 */ 400 ire_refrele(save_ire); 401 } 402 return (ire); 403 } 404 ASSERT(pire == NULL || *pire == NULL); 405 return (ire); 406 } 407 408 409 /* 410 * Find an IRE_OFFSUBNET IRE entry for the multicast address 'group' 411 * that goes through 'ipif'. As a fallback, a route that goes through 412 * ipif->ipif_ill can be returned. 413 */ 414 ire_t * 415 ipif_lookup_multi_ire(ipif_t *ipif, ipaddr_t group) 416 { 417 ire_t *ire; 418 ire_t *save_ire = NULL; 419 ire_t *gw_ire; 420 irb_t *irb; 421 ipaddr_t gw_addr; 422 int match_flags = MATCH_IRE_TYPE | MATCH_IRE_ILL; 423 424 ASSERT(CLASSD(group)); 425 426 ire = ire_ftable_lookup(group, 0, 0, 0, NULL, NULL, ALL_ZONES, 0, 427 NULL, MATCH_IRE_DEFAULT); 428 429 if (ire == NULL) 430 return (NULL); 431 432 irb = ire->ire_bucket; 433 ASSERT(irb); 434 435 IRB_REFHOLD(irb); 436 ire_refrele(ire); 437 for (ire = irb->irb_ire; ire != NULL; ire = ire->ire_next) { 438 if (ire->ire_addr != group || 439 ipif->ipif_zoneid != ire->ire_zoneid && 440 ire->ire_zoneid != ALL_ZONES) { 441 continue; 442 } 443 444 switch (ire->ire_type) { 445 case IRE_DEFAULT: 446 case IRE_PREFIX: 447 case IRE_HOST: 448 gw_addr = ire->ire_gateway_addr; 449 gw_ire = ire_ftable_lookup(gw_addr, 0, 0, IRE_INTERFACE, 450 ipif, NULL, ALL_ZONES, 0, NULL, match_flags); 451 452 if (gw_ire != NULL) { 453 if (save_ire != NULL) { 454 ire_refrele(save_ire); 455 } 456 IRE_REFHOLD(ire); 457 if (gw_ire->ire_ipif == ipif) { 458 ire_refrele(gw_ire); 459 460 IRB_REFRELE(irb); 461 return (ire); 462 } 463 ire_refrele(gw_ire); 464 save_ire = ire; 465 } 466 break; 467 case IRE_IF_NORESOLVER: 468 case IRE_IF_RESOLVER: 469 if (ire->ire_ipif == ipif) { 470 if (save_ire != NULL) { 471 ire_refrele(save_ire); 472 } 473 IRE_REFHOLD(ire); 474 475 IRB_REFRELE(irb); 476 return (ire); 477 } 478 break; 479 } 480 } 481 IRB_REFRELE(irb); 482 483 return (save_ire); 484 } 485 486 /* 487 * Find an IRE_INTERFACE for the multicast group. 488 * Allows different routes for multicast addresses 489 * in the unicast routing table (akin to 224.0.0.0 but could be more specific) 490 * which point at different interfaces. This is used when IP_MULTICAST_IF 491 * isn't specified (when sending) and when IP_ADD_MEMBERSHIP doesn't 492 * specify the interface to join on. 493 * 494 * Supports IP_BOUND_IF by following the ipif/ill when recursing. 495 */ 496 ire_t * 497 ire_lookup_multi(ipaddr_t group, zoneid_t zoneid) 498 { 499 ire_t *ire; 500 ipif_t *ipif = NULL; 501 int match_flags = MATCH_IRE_TYPE; 502 ipaddr_t gw_addr; 503 504 ire = ire_ftable_lookup(group, 0, 0, 0, NULL, NULL, zoneid, 505 0, NULL, MATCH_IRE_DEFAULT); 506 507 /* We search a resolvable ire in case of multirouting. */ 508 if ((ire != NULL) && (ire->ire_flags & RTF_MULTIRT)) { 509 ire_t *cire = NULL; 510 /* 511 * If the route is not resolvable, the looked up ire 512 * may be changed here. In that case, ire_multirt_lookup() 513 * IRE_REFRELE the original ire and change it. 514 */ 515 (void) ire_multirt_lookup(&cire, &ire, MULTIRT_CACHEGW, NULL); 516 if (cire != NULL) 517 ire_refrele(cire); 518 } 519 if (ire == NULL) 520 return (NULL); 521 /* 522 * Make sure we follow ire_ipif. 523 * 524 * We need to determine the interface route through 525 * which the gateway will be reached. We don't really 526 * care which interface is picked if the interface is 527 * part of a group. 528 */ 529 if (ire->ire_ipif != NULL) { 530 ipif = ire->ire_ipif; 531 match_flags |= MATCH_IRE_ILL_GROUP; 532 } 533 534 switch (ire->ire_type) { 535 case IRE_DEFAULT: 536 case IRE_PREFIX: 537 case IRE_HOST: 538 gw_addr = ire->ire_gateway_addr; 539 ire_refrele(ire); 540 ire = ire_ftable_lookup(gw_addr, 0, 0, 541 IRE_INTERFACE, ipif, NULL, zoneid, 0, 542 NULL, match_flags); 543 return (ire); 544 case IRE_IF_NORESOLVER: 545 case IRE_IF_RESOLVER: 546 return (ire); 547 default: 548 ire_refrele(ire); 549 return (NULL); 550 } 551 } 552 553 /* 554 * Delete the passed in ire if the gateway addr matches 555 */ 556 void 557 ire_del_host_redir(ire_t *ire, char *gateway) 558 { 559 if ((ire->ire_type & IRE_HOST_REDIRECT) && 560 (ire->ire_gateway_addr == *(ipaddr_t *)gateway)) 561 ire_delete(ire); 562 } 563 564 /* 565 * Search for all HOST REDIRECT routes that are 566 * pointing at the specified gateway and 567 * delete them. This routine is called only 568 * when a default gateway is going away. 569 */ 570 void 571 ire_delete_host_redirects(ipaddr_t gateway) 572 { 573 struct rtfuncarg rtfarg; 574 575 (void) memset(&rtfarg, 0, sizeof (rtfarg)); 576 rtfarg.rt_func = ire_del_host_redir; 577 rtfarg.rt_arg = (void *)&gateway; 578 (void) ip_ftable->rnh_walktree(ip_ftable, rtfunc, &rtfarg); 579 } 580 581 struct ihandle_arg { 582 uint32_t ihandle; 583 ire_t *ire; 584 }; 585 586 static int 587 ire_ihandle_onlink_match(struct radix_node *rn, void *arg) 588 { 589 struct rt_entry *rt; 590 irb_t *irb; 591 ire_t *ire; 592 struct ihandle_arg *ih = arg; 593 594 rt = (struct rt_entry *)rn; 595 ASSERT(rt != NULL); 596 irb = &rt->rt_irb; 597 for (ire = irb->irb_ire; ire != NULL; ire = ire->ire_next) { 598 if ((ire->ire_type & IRE_INTERFACE) && 599 (ire->ire_ihandle == ih->ihandle)) { 600 ih->ire = ire; 601 IRE_REFHOLD(ire); 602 return (1); 603 } 604 } 605 return (0); 606 } 607 608 /* 609 * Locate the interface ire that is tied to the cache ire 'cire' via 610 * cire->ire_ihandle. 611 * 612 * We are trying to create the cache ire for an onlink destn. or 613 * gateway in 'cire'. We are called from ire_add_v4() in the IRE_IF_RESOLVER 614 * case, after the ire has come back from ARP. 615 */ 616 ire_t * 617 ire_ihandle_lookup_onlink(ire_t *cire) 618 { 619 ire_t *ire; 620 int match_flags; 621 struct ihandle_arg ih; 622 623 ASSERT(cire != NULL); 624 625 /* 626 * We don't need to specify the zoneid to ire_ftable_lookup() below 627 * because the ihandle refers to an ipif which can be in only one zone. 628 */ 629 match_flags = MATCH_IRE_TYPE | MATCH_IRE_IHANDLE | MATCH_IRE_MASK; 630 /* 631 * We know that the mask of the interface ire equals cire->ire_cmask. 632 * (When ip_newroute() created 'cire' for an on-link destn. it set its 633 * cmask from the interface ire's mask) 634 */ 635 ire = ire_ftable_lookup(cire->ire_addr, cire->ire_cmask, 0, 636 IRE_INTERFACE, NULL, NULL, ALL_ZONES, cire->ire_ihandle, 637 NULL, match_flags); 638 if (ire != NULL) 639 return (ire); 640 /* 641 * If we didn't find an interface ire above, we can't declare failure. 642 * For backwards compatibility, we need to support prefix routes 643 * pointing to next hop gateways that are not on-link. 644 * 645 * In the resolver/noresolver case, ip_newroute() thinks it is creating 646 * the cache ire for an onlink destination in 'cire'. But 'cire' is 647 * not actually onlink, because ire_ftable_lookup() cheated it, by 648 * doing ire_route_lookup() twice and returning an interface ire. 649 * 650 * Eg. default - gw1 (line 1) 651 * gw1 - gw2 (line 2) 652 * gw2 - hme0 (line 3) 653 * 654 * In the above example, ip_newroute() tried to create the cache ire 655 * 'cire' for gw1, based on the interface route in line 3. The 656 * ire_ftable_lookup() above fails, because there is no interface route 657 * to reach gw1. (it is gw2). We fall thru below. 658 * 659 * Do a brute force search based on the ihandle in a subset of the 660 * forwarding tables, corresponding to cire->ire_cmask. Otherwise 661 * things become very complex, since we don't have 'pire' in this 662 * case. (Also note that this method is not possible in the offlink 663 * case because we don't know the mask) 664 */ 665 (void) memset(&ih, 0, sizeof (ih)); 666 ih.ihandle = cire->ire_ihandle; 667 (void) ip_ftable->rnh_walktree(ip_ftable, 668 ire_ihandle_onlink_match, &ih); 669 return (ih.ire); 670 } 671 672 /* 673 * IRE iterator used by ire_ftable_lookup[_v6]() to process multiple default 674 * routes. Given a starting point in the hash list (ire_origin), walk the IREs 675 * in the bucket skipping default interface routes and deleted entries. 676 * Returns the next IRE (unheld), or NULL when we're back to the starting point. 677 * Assumes that the caller holds a reference on the IRE bucket. 678 */ 679 ire_t * 680 ire_get_next_default_ire(ire_t *ire, ire_t *ire_origin) 681 { 682 ASSERT(ire_origin->ire_bucket != NULL); 683 ASSERT(ire != NULL); 684 685 do { 686 ire = ire->ire_next; 687 if (ire == NULL) 688 ire = ire_origin->ire_bucket->irb_ire; 689 if (ire == ire_origin) 690 return (NULL); 691 } while ((ire->ire_type & IRE_INTERFACE) || 692 (ire->ire_marks & IRE_MARK_CONDEMNED)); 693 ASSERT(ire != NULL); 694 return (ire); 695 } 696 697 static ipif_t * 698 ire_forward_src_ipif(ipaddr_t dst, ire_t *sire, ire_t *ire, ill_t *dst_ill, 699 int zoneid, ushort_t *marks) 700 { 701 ipif_t *src_ipif; 702 703 /* 704 * Pick the best source address from dst_ill. 705 * 706 * 1) If it is part of a multipathing group, we would 707 * like to spread the inbound packets across different 708 * interfaces. ipif_select_source picks a random source 709 * across the different ills in the group. 710 * 711 * 2) If it is not part of a multipathing group, we try 712 * to pick the source address from the destination 713 * route. Clustering assumes that when we have multiple 714 * prefixes hosted on an interface, the prefix of the 715 * source address matches the prefix of the destination 716 * route. We do this only if the address is not 717 * DEPRECATED. 718 * 719 * 3) If the conn is in a different zone than the ire, we 720 * need to pick a source address from the right zone. 721 * 722 * NOTE : If we hit case (1) above, the prefix of the source 723 * address picked may not match the prefix of the 724 * destination routes prefix as ipif_select_source 725 * does not look at "dst" while picking a source 726 * address. 727 * If we want the same behavior as (2), we will need 728 * to change the behavior of ipif_select_source. 729 */ 730 731 if ((sire != NULL) && (sire->ire_flags & RTF_SETSRC)) { 732 /* 733 * The RTF_SETSRC flag is set in the parent ire (sire). 734 * Check that the ipif matching the requested source 735 * address still exists. 736 */ 737 src_ipif = ipif_lookup_addr(sire->ire_src_addr, NULL, 738 zoneid, NULL, NULL, NULL, NULL); 739 return (src_ipif); 740 } 741 *marks |= IRE_MARK_USESRC_CHECK; 742 if ((dst_ill->ill_group != NULL) || 743 (ire->ire_ipif->ipif_flags & IPIF_DEPRECATED) || 744 (dst_ill->ill_usesrc_ifindex != 0)) { 745 src_ipif = ipif_select_source(dst_ill, dst, zoneid); 746 if (src_ipif == NULL) 747 return (NULL); 748 749 } else { 750 src_ipif = ire->ire_ipif; 751 ASSERT(src_ipif != NULL); 752 /* hold src_ipif for uniformity */ 753 ipif_refhold(src_ipif); 754 } 755 return (src_ipif); 756 } 757 758 /* Added to root cause a bug - should be removed later */ 759 ire_t *ire_gw_cache = NULL; 760 761 /* 762 * This function is called by ip_rput_noire() and ip_fast_forward() 763 * to resolve the route of incoming packet that needs to be forwarded. 764 * If the ire of the nexthop is not already in the cachetable, this 765 * routine will insert it to the table, but won't trigger ARP resolution yet. 766 * Thus unlike ip_newroute, this function adds incomplete ires to 767 * the cachetable. ARP resolution for these ires are delayed until 768 * after all of the packet processing is completed and its ready to 769 * be sent out on the wire, Eventually, the packet transmit routine 770 * ip_xmit_v4() attempts to send a packet to the driver. If it finds 771 * that there is no link layer information, it will do the arp 772 * resolution and queue the packet in ire->ire_nce->nce_qd_mp and 773 * then send it out once the arp resolution is over 774 * (see ip_xmit_v4()->ire_arpresolve()). This scheme is similar to 775 * the model of BSD/SunOS 4 776 * 777 * In future, the insertion of incomplete ires in the cachetable should 778 * be implemented in hostpath as well, as doing so will greatly reduce 779 * the existing complexity for code paths that depend on the context of 780 * the sender (such as IPsec). 781 * 782 * Thus this scheme of adding incomplete ires in cachetable in forwarding 783 * path can be used as a template for simplifying the hostpath. 784 */ 785 786 ire_t * 787 ire_forward(ipaddr_t dst, boolean_t *check_multirt, ire_t *supplied_ire, 788 ire_t *supplied_sire, const struct ts_label_s *tsl) 789 { 790 ipaddr_t gw = 0; 791 ire_t *ire = NULL; 792 ire_t *sire = NULL, *save_ire; 793 ill_t *dst_ill = NULL; 794 int error; 795 zoneid_t zoneid; 796 ipif_t *src_ipif = NULL; 797 mblk_t *res_mp; 798 ushort_t ire_marks = 0; 799 tsol_gcgrp_t *gcgrp = NULL; 800 tsol_gcgrp_addr_t ga; 801 802 zoneid = GLOBAL_ZONEID; 803 804 if (supplied_ire != NULL) { 805 /* We have arrived here from ipfil_sendpkt */ 806 ire = supplied_ire; 807 sire = supplied_sire; 808 goto create_irecache; 809 } 810 811 ire = ire_ftable_lookup(dst, 0, 0, 0, NULL, &sire, zoneid, 0, 812 tsl, MATCH_IRE_RECURSIVE | MATCH_IRE_DEFAULT | 813 MATCH_IRE_RJ_BHOLE | MATCH_IRE_PARENT|MATCH_IRE_SECATTR); 814 815 if (ire == NULL) { 816 ip_rts_change(RTM_MISS, dst, 0, 0, 0, 0, 0, 0, RTA_DST); 817 goto icmp_err_ret; 818 } 819 820 /* 821 * If we encounter CGTP, we should have the caller use 822 * ip_newroute to resolve multirt instead of this function. 823 * CGTP specs explicitly state that it can't be used with routers. 824 * This essentially prevents insertion of incomplete RTF_MULTIRT 825 * ires in cachetable. 826 */ 827 if (ip_cgtp_filter && 828 ((ire->ire_flags & RTF_MULTIRT) || 829 ((sire != NULL) && (sire->ire_flags & RTF_MULTIRT)))) { 830 ip3dbg(("ire_forward: packet is to be multirouted- " 831 "handing it to ip_newroute\n")); 832 if (sire != NULL) 833 ire_refrele(sire); 834 ire_refrele(ire); 835 /* 836 * Inform caller about encountering of multirt so that 837 * ip_newroute() can be called. 838 */ 839 *check_multirt = B_TRUE; 840 return (NULL); 841 } 842 843 *check_multirt = B_FALSE; 844 845 /* 846 * Verify that the returned IRE does not have either 847 * the RTF_REJECT or RTF_BLACKHOLE flags set and that the IRE is 848 * either an IRE_CACHE, IRE_IF_NORESOLVER or IRE_IF_RESOLVER. 849 */ 850 if ((ire->ire_flags & (RTF_REJECT | RTF_BLACKHOLE)) || 851 (ire->ire_type & (IRE_CACHE | IRE_INTERFACE)) == 0) { 852 ip3dbg(("ire 0x%p is not cache/resolver/noresolver\n", 853 (void *)ire)); 854 goto icmp_err_ret; 855 } 856 857 /* 858 * If we already have a fully resolved IRE CACHE of the 859 * nexthop router, just hand over the cache entry 860 * and we are done. 861 */ 862 863 if (ire->ire_type & IRE_CACHE) { 864 865 /* 866 * If we are using this ire cache entry as a 867 * gateway to forward packets, chances are we 868 * will be using it again. So turn off 869 * the temporary flag, thus reducing its 870 * chances of getting deleted frequently. 871 */ 872 if (ire->ire_marks & IRE_MARK_TEMPORARY) { 873 irb_t *irb = ire->ire_bucket; 874 rw_enter(&irb->irb_lock, RW_WRITER); 875 ire->ire_marks &= ~IRE_MARK_TEMPORARY; 876 irb->irb_tmp_ire_cnt--; 877 rw_exit(&irb->irb_lock); 878 } 879 880 if (sire != NULL) { 881 UPDATE_OB_PKT_COUNT(sire); 882 sire->ire_last_used_time = lbolt; 883 ire_refrele(sire); 884 } 885 return (ire); 886 } 887 create_irecache: 888 /* 889 * Increment the ire_ob_pkt_count field for ire if it is an 890 * INTERFACE (IF_RESOLVER or IF_NORESOLVER) IRE type, and 891 * increment the same for the parent IRE, sire, if it is some 892 * sort of prefix IRE (which includes DEFAULT, PREFIX, HOST 893 * and HOST_REDIRECT). 894 */ 895 if ((ire->ire_type & IRE_INTERFACE) != 0) { 896 UPDATE_OB_PKT_COUNT(ire); 897 ire->ire_last_used_time = lbolt; 898 } 899 900 /* 901 * sire must be either IRE_CACHETABLE OR IRE_INTERFACE type 902 */ 903 if (sire != NULL) { 904 gw = sire->ire_gateway_addr; 905 ASSERT((sire->ire_type & 906 (IRE_CACHETABLE | IRE_INTERFACE)) == 0); 907 UPDATE_OB_PKT_COUNT(sire); 908 sire->ire_last_used_time = lbolt; 909 } 910 911 /* Obtain dst_ill */ 912 dst_ill = ip_newroute_get_dst_ill(ire->ire_ipif->ipif_ill); 913 if (dst_ill == NULL) { 914 ip2dbg(("ire_forward no dst ill; ire 0x%p\n", 915 (void *)ire)); 916 goto icmp_err_ret; 917 } 918 919 ASSERT(src_ipif == NULL); 920 /* Now obtain the src_ipif */ 921 src_ipif = ire_forward_src_ipif(dst, sire, ire, dst_ill, 922 zoneid, &ire_marks); 923 if (src_ipif == NULL) 924 goto icmp_err_ret; 925 926 switch (ire->ire_type) { 927 case IRE_IF_NORESOLVER: 928 /* create ire_cache for ire_addr endpoint */ 929 case IRE_IF_RESOLVER: 930 /* 931 * We have the IRE_IF_RESOLVER of the nexthop gateway 932 * and now need to build a IRE_CACHE for it. 933 * In this case, we have the following : 934 * 935 * 1) src_ipif - used for getting a source address. 936 * 937 * 2) dst_ill - from which we derive ire_stq/ire_rfq. This 938 * means packets using the IRE_CACHE that we will build 939 * here will go out on dst_ill. 940 * 941 * 3) sire may or may not be NULL. But, the IRE_CACHE that is 942 * to be created will only be tied to the IRE_INTERFACE 943 * that was derived from the ire_ihandle field. 944 * 945 * If sire is non-NULL, it means the destination is 946 * off-link and we will first create the IRE_CACHE for the 947 * gateway. 948 */ 949 res_mp = dst_ill->ill_resolver_mp; 950 if (ire->ire_type == IRE_IF_RESOLVER && 951 (!OK_RESOLVER_MP(res_mp))) { 952 ire_refrele(ire); 953 ire = NULL; 954 goto out; 955 } 956 /* 957 * To be at this point in the code with a non-zero gw 958 * means that dst is reachable through a gateway that 959 * we have never resolved. By changing dst to the gw 960 * addr we resolve the gateway first. 961 */ 962 if (gw != INADDR_ANY) { 963 /* 964 * The source ipif that was determined above was 965 * relative to the destination address, not the 966 * gateway's. If src_ipif was not taken out of 967 * the IRE_IF_RESOLVER entry, we'll need to call 968 * ipif_select_source() again. 969 */ 970 if (src_ipif != ire->ire_ipif) { 971 ipif_refrele(src_ipif); 972 src_ipif = ipif_select_source(dst_ill, 973 gw, zoneid); 974 if (src_ipif == NULL) 975 goto icmp_err_ret; 976 } 977 dst = gw; 978 gw = INADDR_ANY; 979 } 980 /* 981 * dst has been set to the address of the nexthop. 982 * 983 * TSol note: get security attributes of the nexthop; 984 * Note that the nexthop may either be a gateway, or the 985 * packet destination itself; Detailed explanation of 986 * issues involved is provided in the IRE_IF_NORESOLVER 987 * logic in ip_newroute(). 988 */ 989 ga.ga_af = AF_INET; 990 IN6_IPADDR_TO_V4MAPPED(dst, &ga.ga_addr); 991 gcgrp = gcgrp_lookup(&ga, B_FALSE); 992 993 if (ire->ire_type == IRE_IF_NORESOLVER) 994 dst = ire->ire_addr; /* ire_cache for tunnel endpoint */ 995 996 save_ire = ire; 997 /* 998 * create an incomplete ire-cache with a null dlureq_mp. 999 * The dlureq_mp will be created in ire_arpresolve. 1000 */ 1001 ire = ire_create( 1002 (uchar_t *)&dst, /* dest address */ 1003 (uchar_t *)&ip_g_all_ones, /* mask */ 1004 (uchar_t *)&src_ipif->ipif_src_addr, /* src addr */ 1005 (uchar_t *)&gw, /* gateway address */ 1006 NULL, 1007 (save_ire->ire_type == IRE_IF_RESOLVER ? NULL: 1008 &save_ire->ire_max_frag), 1009 NULL, 1010 dst_ill->ill_rq, /* recv-from queue */ 1011 dst_ill->ill_wq, /* send-to queue */ 1012 IRE_CACHE, /* IRE type */ 1013 NULL, 1014 src_ipif, 1015 NULL, 1016 ire->ire_mask, /* Parent mask */ 1017 0, 1018 ire->ire_ihandle, /* Interface handle */ 1019 0, 1020 &(ire->ire_uinfo), 1021 NULL, 1022 gcgrp); 1023 ip1dbg(("incomplete ire_cache 0x%p\n", (void *)ire)); 1024 if (ire != NULL) { 1025 gcgrp = NULL; /* reference now held by IRE */ 1026 ire->ire_marks |= ire_marks; 1027 /* add the incomplete ire: */ 1028 error = ire_add(&ire, NULL, NULL, NULL, B_TRUE); 1029 if (error == 0 && ire != NULL) { 1030 ire->ire_max_frag = save_ire->ire_max_frag; 1031 ip1dbg(("setting max_frag to %d in ire 0x%p\n", 1032 ire->ire_max_frag, (void *)ire)); 1033 } else { 1034 ire_refrele(save_ire); 1035 goto icmp_err_ret; 1036 } 1037 } else { 1038 if (gcgrp != NULL) { 1039 GCGRP_REFRELE(gcgrp); 1040 gcgrp = NULL; 1041 } 1042 } 1043 1044 ire_refrele(save_ire); 1045 break; 1046 default: 1047 break; 1048 } 1049 1050 out: 1051 if (sire != NULL) 1052 ire_refrele(sire); 1053 if (dst_ill != NULL) 1054 ill_refrele(dst_ill); 1055 if (src_ipif != NULL) 1056 ipif_refrele(src_ipif); 1057 return (ire); 1058 icmp_err_ret: 1059 if (src_ipif != NULL) 1060 ipif_refrele(src_ipif); 1061 if (dst_ill != NULL) 1062 ill_refrele(dst_ill); 1063 if (sire != NULL) 1064 ire_refrele(sire); 1065 if (ire != NULL) { 1066 ire_refrele(ire); 1067 } 1068 /* caller needs to send icmp error message */ 1069 return (NULL); 1070 1071 } 1072 1073 /* 1074 * Obtain the rt_entry and rt_irb for the route to be added to the ip_ftable. 1075 * First attempt to add a node to the radix tree via rn_addroute. If the 1076 * route already exists, return the bucket for the existing route. 1077 * 1078 * Locking notes: Need to hold the global radix tree lock in write mode to 1079 * add a radix node. To prevent the node from being deleted, ire_get_bucket() 1080 * returns with a ref'ed irb_t. The ire itself is added in ire_add_v4() 1081 * while holding the irb_lock, but not the radix tree lock. 1082 */ 1083 irb_t * 1084 ire_get_bucket(ire_t *ire) 1085 { 1086 struct radix_node *rn; 1087 struct rt_entry *rt; 1088 struct rt_sockaddr rmask, rdst; 1089 irb_t *irb = NULL; 1090 1091 ASSERT(ip_ftable != NULL); 1092 1093 /* first try to see if route exists (based on rtalloc1) */ 1094 (void) memset(&rdst, 0, sizeof (rdst)); 1095 rdst.rt_sin_len = sizeof (rdst); 1096 rdst.rt_sin_family = AF_INET; 1097 rdst.rt_sin_addr.s_addr = ire->ire_addr; 1098 1099 (void) memset(&rmask, 0, sizeof (rmask)); 1100 rmask.rt_sin_len = sizeof (rmask); 1101 rmask.rt_sin_family = AF_INET; 1102 rmask.rt_sin_addr.s_addr = ire->ire_mask; 1103 1104 /* 1105 * add the route. based on BSD's rtrequest1(RTM_ADD) 1106 */ 1107 R_Malloc(rt, rt_entry_cache, sizeof (*rt)); 1108 (void) memset(rt, 0, sizeof (*rt)); 1109 rt->rt_nodes->rn_key = (char *)&rt->rt_dst; 1110 rt->rt_dst = rdst; 1111 irb = &rt->rt_irb; 1112 irb->irb_marks |= IRB_MARK_FTABLE; /* dynamically allocated/freed */ 1113 rw_init(&irb->irb_lock, NULL, RW_DEFAULT, NULL); 1114 RADIX_NODE_HEAD_WLOCK(ip_ftable); 1115 rn = ip_ftable->rnh_addaddr(&rt->rt_dst, &rmask, ip_ftable, 1116 (struct radix_node *)rt); 1117 if (rn == NULL) { 1118 RADIX_NODE_HEAD_UNLOCK(ip_ftable); 1119 Free(rt, rt_entry_cache); 1120 rt = NULL; 1121 irb = NULL; 1122 RADIX_NODE_HEAD_RLOCK(ip_ftable); 1123 if ((rn = ip_ftable->rnh_lookup(&rdst, &rmask, ip_ftable)) != 1124 NULL && ((rn->rn_flags & RNF_ROOT) == 0)) { 1125 /* found a non-root match */ 1126 rt = (struct rt_entry *)rn; 1127 } 1128 } 1129 if (rt != NULL) { 1130 irb = &rt->rt_irb; 1131 IRB_REFHOLD(irb); 1132 } 1133 RADIX_NODE_HEAD_UNLOCK(ip_ftable); 1134 return (irb); 1135 } 1136 1137 /* 1138 * This function is used when the caller wants to know the outbound 1139 * interface for a packet given only the address. 1140 * If this is a offlink IP address and there are multiple 1141 * routes to this destination, this routine will utilise the 1142 * first route it finds to IP address 1143 * Return values: 1144 * 0 - FAILURE 1145 * nonzero - ifindex 1146 */ 1147 uint_t 1148 ifindex_lookup(const struct sockaddr *ipaddr, zoneid_t zoneid) 1149 { 1150 uint_t ifindex = 0; 1151 ire_t *ire; 1152 ill_t *ill; 1153 1154 /* zoneid is a placeholder for future routing table per-zone project */ 1155 ASSERT(zoneid == ALL_ZONES); 1156 1157 ASSERT(ipaddr->sa_family == AF_INET || ipaddr->sa_family == AF_INET6); 1158 1159 if ((ire = route_to_dst(ipaddr, zoneid)) != NULL) { 1160 ill = ire_to_ill(ire); 1161 if (ill != NULL) 1162 ifindex = ill->ill_phyint->phyint_ifindex; 1163 ire_refrele(ire); 1164 } 1165 return (ifindex); 1166 } 1167 1168 /* 1169 * Routine to find the route to a destination. If a ifindex is supplied 1170 * it tries to match the the route to the corresponding ipif for the ifindex 1171 */ 1172 static ire_t * 1173 route_to_dst(const struct sockaddr *dst_addr, zoneid_t zoneid) 1174 { 1175 ire_t *ire = NULL; 1176 int match_flags; 1177 1178 match_flags = (MATCH_IRE_DSTONLY | MATCH_IRE_DEFAULT | 1179 MATCH_IRE_RECURSIVE | MATCH_IRE_RJ_BHOLE); 1180 1181 /* XXX pass NULL tsl for now */ 1182 1183 if (dst_addr->sa_family == AF_INET) { 1184 ire = ire_route_lookup( 1185 ((struct sockaddr_in *)dst_addr)->sin_addr.s_addr, 1186 0, 0, 0, NULL, NULL, zoneid, NULL, match_flags); 1187 } else { 1188 ire = ire_route_lookup_v6( 1189 &((struct sockaddr_in6 *)dst_addr)->sin6_addr, 1190 0, 0, 0, NULL, NULL, zoneid, NULL, match_flags); 1191 } 1192 return (ire); 1193 } 1194 1195 /* 1196 * This routine is called by IP Filter to send a packet out on the wire 1197 * to a specified V4 dst (which may be onlink or offlink). The ifindex may or 1198 * may not be 0. A non-null ifindex indicates IP Filter has stipulated 1199 * an outgoing interface and requires the nexthop to be on that interface. 1200 * IP WILL NOT DO the following to the data packet before sending it out: 1201 * a. manipulate ttl 1202 * b. checksuming 1203 * c. ipsec work 1204 * d. fragmentation 1205 * 1206 * Return values: 1207 * 0: IP was able to send of the data pkt 1208 * ECOMM: Could not send packet 1209 * ENONET No route to dst. It is up to the caller 1210 * to send icmp unreachable error message, 1211 * EINPROGRESS The macaddr of the onlink dst or that 1212 * of the offlink dst's nexthop needs to get 1213 * resolved before packet can be sent to dst. 1214 * Thus transmission is not guaranteed. 1215 * 1216 */ 1217 1218 int 1219 ipfil_sendpkt(const struct sockaddr *dst_addr, mblk_t *mp, uint_t ifindex, 1220 zoneid_t zoneid) 1221 { 1222 ire_t *ire = NULL, *sire = NULL; 1223 ire_t *ire_cache = NULL; 1224 boolean_t check_multirt = B_FALSE; 1225 int value; 1226 int match_flags; 1227 ipaddr_t dst; 1228 1229 ASSERT(mp != NULL); 1230 1231 ASSERT(dst_addr->sa_family == AF_INET || 1232 dst_addr->sa_family == AF_INET6); 1233 1234 if (dst_addr->sa_family == AF_INET) { 1235 dst = ((struct sockaddr_in *)dst_addr)->sin_addr.s_addr; 1236 } else { 1237 /* 1238 * We dont have support for V6 yet. It will be provided 1239 * once RFE 6399103 has been delivered. 1240 * Until then, for V6 dsts, IP Filter will not call 1241 * this function. Instead, IP Filter will continue to do what 1242 * has been done since S10, namely it will use 1243 * ip_nexthop(),ip_nexthop_route() to obtain the 1244 * link-layer address of a V6 dst and then process the 1245 * packet and send it out on the wire on its own. 1246 */ 1247 ip1dbg(("ipfil_sendpkt: no V6 support \n")); 1248 value = ECOMM; 1249 freemsg(mp); 1250 goto discard; 1251 } 1252 1253 /* 1254 * Lets get the ire. We might get the ire cache entry, 1255 * or the ire,sire pair needed to create the cache entry. 1256 * XXX pass NULL tsl for now. 1257 */ 1258 1259 if (ifindex == 0) { 1260 /* There is no supplied index. So use the FIB info */ 1261 1262 match_flags = (MATCH_IRE_DSTONLY | MATCH_IRE_DEFAULT | 1263 MATCH_IRE_RECURSIVE | MATCH_IRE_RJ_BHOLE); 1264 ire = ire_route_lookup(dst, 1265 0, 0, 0, NULL, &sire, zoneid, MBLK_GETLABEL(mp), 1266 match_flags); 1267 } else { 1268 ipif_t *supplied_ipif; 1269 ill_t *ill; 1270 1271 /* 1272 * If supplied ifindex is non-null, the only valid 1273 * nexthop is one off of the interface corresponding 1274 * to the specified ifindex. 1275 */ 1276 1277 ill = ill_lookup_on_ifindex(ifindex, B_FALSE, 1278 NULL, NULL, NULL, NULL); 1279 if (ill != NULL) { 1280 supplied_ipif = ipif_get_next_ipif(NULL, ill); 1281 } else { 1282 ip1dbg(("ipfil_sendpkt: Could not find" 1283 " route to dst\n")); 1284 value = ECOMM; 1285 freemsg(mp); 1286 goto discard; 1287 } 1288 1289 match_flags = (MATCH_IRE_DSTONLY | MATCH_IRE_DEFAULT | 1290 MATCH_IRE_IPIF | MATCH_IRE_RECURSIVE| MATCH_IRE_RJ_BHOLE| 1291 MATCH_IRE_SECATTR); 1292 1293 ire = ire_route_lookup(dst, 0, 0, 0, supplied_ipif, 1294 &sire, zoneid, MBLK_GETLABEL(mp), match_flags); 1295 ill_refrele(ill); 1296 } 1297 1298 /* 1299 * Verify that the returned IRE is non-null and does 1300 * not have either the RTF_REJECT or RTF_BLACKHOLE 1301 * flags set and that the IRE is either an IRE_CACHE, 1302 * IRE_IF_NORESOLVER or IRE_IF_RESOLVER. 1303 */ 1304 if (ire == NULL || 1305 ((ire->ire_flags & (RTF_REJECT | RTF_BLACKHOLE)) || 1306 (ire->ire_type & (IRE_CACHE | IRE_INTERFACE)) == 0)) { 1307 /* 1308 * Either ire could not be found or we got 1309 * an invalid one 1310 */ 1311 ip1dbg(("ipfil_sendpkt: Could not find route to dst\n")); 1312 value = ENONET; 1313 freemsg(mp); 1314 goto discard; 1315 } 1316 1317 /* IP Filter and CGTP dont mix. So bail out if CGTP is on */ 1318 if (ip_cgtp_filter && 1319 ((ire->ire_flags & RTF_MULTIRT) || 1320 ((sire != NULL) && (sire->ire_flags & RTF_MULTIRT)))) { 1321 ip1dbg(("ipfil_sendpkt: IPFilter does not work with CGTP\n")); 1322 value = ECOMM; 1323 freemsg(mp); 1324 goto discard; 1325 } 1326 1327 ASSERT(ire->ire_nce != NULL); 1328 /* 1329 * If needed, we will create the ire cache entry for the 1330 * nexthop, resolve its link-layer address and then send 1331 * the packet out without ttl, checksumming, IPSec processing. 1332 */ 1333 1334 switch (ire->ire_type) { 1335 case IRE_IF_NORESOLVER: 1336 case IRE_CACHE: 1337 if (sire != NULL) { 1338 UPDATE_OB_PKT_COUNT(sire); 1339 sire->ire_last_used_time = lbolt; 1340 ire_refrele(sire); 1341 } 1342 ire_cache = ire; 1343 break; 1344 case IRE_IF_RESOLVER: 1345 /* 1346 * Call ire_forward(). This function 1347 * will, create the ire cache entry of the 1348 * the nexthop and adds this incomplete ire 1349 * to the ire cache table 1350 */ 1351 ire_cache = ire_forward(dst, &check_multirt, ire, sire, 1352 MBLK_GETLABEL(mp)); 1353 if (ire_cache == NULL) { 1354 ip1dbg(("ipfil_sendpkt: failed to create the" 1355 " ire cache entry \n")); 1356 value = ENONET; 1357 freemsg(mp); 1358 sire = NULL; 1359 ire = NULL; 1360 goto discard; 1361 } 1362 break; 1363 } 1364 /* 1365 * Now that we have the ire cache entry of the nexthop, call 1366 * ip_xmit_v4() to trigger mac addr resolution 1367 * if necessary and send it once ready. 1368 */ 1369 1370 value = ip_xmit_v4(mp, ire_cache, NULL, B_FALSE); 1371 ire_refrele(ire_cache); 1372 /* 1373 * At this point, the reference for these have already been 1374 * released within ire_forward() and/or ip_xmit_v4(). So we set 1375 * them to NULL to make sure we dont drop the references 1376 * again in case ip_xmit_v4() returns with either SEND_FAILED 1377 * or LLHDR_RESLV_FAILED 1378 */ 1379 sire = NULL; 1380 ire = NULL; 1381 1382 switch (value) { 1383 case SEND_FAILED: 1384 ip1dbg(("ipfil_sendpkt: Send failed\n")); 1385 value = ECOMM; 1386 break; 1387 case LLHDR_RESLV_FAILED: 1388 ip1dbg(("ipfil_sendpkt: Link-layer resolution" 1389 " failed\n")); 1390 value = ECOMM; 1391 break; 1392 case LOOKUP_IN_PROGRESS: 1393 return (EINPROGRESS); 1394 case SEND_PASSED: 1395 return (0); 1396 } 1397 discard: 1398 if (dst_addr->sa_family == AF_INET) { 1399 BUMP_MIB(&ip_mib, ipOutDiscards); 1400 } else { 1401 BUMP_MIB(&ip6_mib, ipv6OutDiscards); 1402 } 1403 if (ire != NULL) 1404 ire_refrele(ire); 1405 if (sire != NULL) 1406 ire_refrele(sire); 1407 return (value); 1408 } 1409 1410 /* ire_walk routine invoked for ip_ire_report for each IRE. */ 1411 void 1412 ire_report_ftable(ire_t *ire, char *m) 1413 { 1414 char buf1[16]; 1415 char buf2[16]; 1416 char buf3[16]; 1417 char buf4[16]; 1418 uint_t fo_pkt_count; 1419 uint_t ib_pkt_count; 1420 int ref; 1421 uint_t print_len, buf_len; 1422 mblk_t *mp = (mblk_t *)m; 1423 1424 if (ire->ire_type & IRE_CACHETABLE) 1425 return; 1426 buf_len = mp->b_datap->db_lim - mp->b_wptr; 1427 if (buf_len <= 0) 1428 return; 1429 1430 /* Number of active references of this ire */ 1431 ref = ire->ire_refcnt; 1432 /* "inbound" to a non local address is a forward */ 1433 ib_pkt_count = ire->ire_ib_pkt_count; 1434 fo_pkt_count = 0; 1435 if (!(ire->ire_type & (IRE_LOCAL|IRE_BROADCAST))) { 1436 fo_pkt_count = ib_pkt_count; 1437 ib_pkt_count = 0; 1438 } 1439 print_len = snprintf((char *)mp->b_wptr, buf_len, 1440 MI_COL_PTRFMT_STR MI_COL_PTRFMT_STR MI_COL_PTRFMT_STR "%5d " 1441 "%s %s %s %s %05d %05ld %06ld %08d %03d %06d %09d %09d %06d %08d " 1442 "%04d %08d %08d %d/%d/%d %s\n", 1443 (void *)ire, (void *)ire->ire_rfq, (void *)ire->ire_stq, 1444 (int)ire->ire_zoneid, 1445 ip_dot_addr(ire->ire_addr, buf1), ip_dot_addr(ire->ire_mask, buf2), 1446 ip_dot_addr(ire->ire_src_addr, buf3), 1447 ip_dot_addr(ire->ire_gateway_addr, buf4), 1448 ire->ire_max_frag, ire->ire_uinfo.iulp_rtt, 1449 ire->ire_uinfo.iulp_rtt_sd, 1450 ire->ire_uinfo.iulp_ssthresh, ref, 1451 ire->ire_uinfo.iulp_rtomax, 1452 (ire->ire_uinfo.iulp_tstamp_ok ? 1: 0), 1453 (ire->ire_uinfo.iulp_wscale_ok ? 1: 0), 1454 (ire->ire_uinfo.iulp_ecn_ok ? 1: 0), 1455 (ire->ire_uinfo.iulp_pmtud_ok ? 1: 0), 1456 ire->ire_uinfo.iulp_sack, 1457 ire->ire_uinfo.iulp_spipe, ire->ire_uinfo.iulp_rpipe, 1458 ib_pkt_count, ire->ire_ob_pkt_count, fo_pkt_count, 1459 ip_nv_lookup(ire_nv_tbl, (int)ire->ire_type)); 1460 if (print_len < buf_len) { 1461 mp->b_wptr += print_len; 1462 } else { 1463 mp->b_wptr += buf_len; 1464 } 1465 } 1466 1467 /* 1468 * callback function provided by ire_ftable_lookup when calling 1469 * rn_match_args(). Invoke ire_match_args on each matching leaf node in 1470 * the radix tree. 1471 */ 1472 boolean_t 1473 ire_find_best_route(struct radix_node *rn, void *arg) 1474 { 1475 struct rt_entry *rt = (struct rt_entry *)rn; 1476 irb_t *irb_ptr; 1477 ire_t *ire; 1478 ire_ftable_args_t *margs = arg; 1479 ipaddr_t match_mask; 1480 1481 ASSERT(rt != NULL); 1482 1483 irb_ptr = &rt->rt_irb; 1484 1485 if (irb_ptr->irb_ire_cnt == 0) 1486 return (B_FALSE); 1487 1488 rw_enter(&irb_ptr->irb_lock, RW_READER); 1489 for (ire = irb_ptr->irb_ire; ire != NULL; ire = ire->ire_next) { 1490 if (ire->ire_marks & IRE_MARK_CONDEMNED) 1491 continue; 1492 if (margs->ift_flags & MATCH_IRE_MASK) 1493 match_mask = margs->ift_mask; 1494 else 1495 match_mask = ire->ire_mask; 1496 1497 if (ire_match_args(ire, margs->ift_addr, match_mask, 1498 margs->ift_gateway, margs->ift_type, margs->ift_ipif, 1499 margs->ift_zoneid, margs->ift_ihandle, margs->ift_tsl, 1500 margs->ift_flags)) { 1501 IRE_REFHOLD(ire); 1502 rw_exit(&irb_ptr->irb_lock); 1503 margs->ift_best_ire = ire; 1504 return (B_TRUE); 1505 } 1506 } 1507 rw_exit(&irb_ptr->irb_lock); 1508 return (B_FALSE); 1509 } 1510 1511 /* 1512 * ftable irb_t structures are dynamically allocated, and we need to 1513 * check if the irb_t (and associated ftable tree attachment) needs to 1514 * be cleaned up when the irb_refcnt goes to 0. The conditions that need 1515 * be verified are: 1516 * - no other walkers of the irebucket, i.e., quiescent irb_refcnt, 1517 * - no other threads holding references to ire's in the bucket, 1518 * i.e., irb_nire == 0 1519 * - no active ire's in the bucket, i.e., irb_ire_cnt == 0 1520 * - need to hold the global tree lock and irb_lock in write mode. 1521 */ 1522 void 1523 irb_refrele_ftable(irb_t *irb) 1524 { 1525 for (;;) { 1526 rw_enter(&irb->irb_lock, RW_WRITER); 1527 ASSERT(irb->irb_refcnt != 0); 1528 if (irb->irb_refcnt != 1) { 1529 /* 1530 * Someone has a reference to this radix node 1531 * or there is some bucket walker. 1532 */ 1533 irb->irb_refcnt--; 1534 rw_exit(&irb->irb_lock); 1535 return; 1536 } else { 1537 /* 1538 * There is no other walker, nor is there any 1539 * other thread that holds a direct ref to this 1540 * radix node. Do the clean up if needed. Call 1541 * to ire_unlink will clear the IRB_MARK_CONDEMNED flag 1542 */ 1543 if (irb->irb_marks & IRB_MARK_CONDEMNED) { 1544 ire_t *ire_list; 1545 1546 ire_list = ire_unlink(irb); 1547 rw_exit(&irb->irb_lock); 1548 1549 if (ire_list != NULL) 1550 ire_cleanup(ire_list); 1551 /* 1552 * more CONDEMNED entries could have 1553 * been added while we dropped the lock, 1554 * so we have to re-check. 1555 */ 1556 continue; 1557 } 1558 1559 /* 1560 * Now check if there are still any ires 1561 * associated with this radix node. 1562 */ 1563 if (irb->irb_nire != 0) { 1564 /* 1565 * someone is still holding on 1566 * to ires in this bucket 1567 */ 1568 irb->irb_refcnt--; 1569 rw_exit(&irb->irb_lock); 1570 return; 1571 } else { 1572 /* 1573 * Everything is clear. Zero walkers, 1574 * Zero threads with a ref to this 1575 * radix node, Zero ires associated with 1576 * this radix node. Due to lock order, 1577 * check the above conditions again 1578 * after grabbing all locks in the right order 1579 */ 1580 rw_exit(&irb->irb_lock); 1581 if (irb_inactive(irb)) 1582 return; 1583 /* 1584 * irb_inactive could not free the irb. 1585 * See if there are any walkers, if not 1586 * try to clean up again. 1587 */ 1588 } 1589 } 1590 } 1591 } 1592 1593 /* 1594 * IRE iterator used by ire_ftable_lookup() to process multiple default 1595 * routes. Given a starting point in the hash list (ire_origin), walk the IREs 1596 * in the bucket skipping default interface routes and deleted entries. 1597 * Returns the next IRE (unheld), or NULL when we're back to the starting point. 1598 * Assumes that the caller holds a reference on the IRE bucket. 1599 * 1600 * In the absence of good IRE_DEFAULT routes, this function will return 1601 * the first IRE_INTERFACE route found (if any). 1602 */ 1603 ire_t * 1604 ire_round_robin(irb_t *irb_ptr, zoneid_t zoneid, ire_ftable_args_t *margs) 1605 { 1606 ire_t *ire_origin; 1607 ire_t *ire, *maybe_ire = NULL; 1608 1609 rw_enter(&irb_ptr->irb_lock, RW_WRITER); 1610 ire_origin = irb_ptr->irb_rr_origin; 1611 if (ire_origin != NULL) 1612 ire_origin = ire_origin->ire_next; 1613 1614 if (ire_origin == NULL) { 1615 /* 1616 * first time through routine, or we dropped off the end 1617 * of list. 1618 */ 1619 ire_origin = irb_ptr->irb_ire; 1620 } 1621 irb_ptr->irb_rr_origin = ire_origin; 1622 IRB_REFHOLD_LOCKED(irb_ptr); 1623 rw_exit(&irb_ptr->irb_lock); 1624 1625 /* 1626 * Round-robin the routers list looking for a route that 1627 * matches the passed in parameters. 1628 * We start with the ire we found above and we walk the hash 1629 * list until we're back where we started. It doesn't matter if 1630 * routes are added or deleted by other threads - we know this 1631 * ire will stay in the list because we hold a reference on the 1632 * ire bucket. 1633 */ 1634 ire = ire_origin; 1635 while (ire != NULL) { 1636 int match_flags = 0; 1637 ire_t *rire; 1638 1639 if (ire->ire_marks & IRE_MARK_CONDEMNED) 1640 goto next_ire; 1641 1642 if (!ire_match_args(ire, margs->ift_addr, (ipaddr_t)0, 1643 margs->ift_gateway, margs->ift_type, margs->ift_ipif, 1644 margs->ift_zoneid, margs->ift_ihandle, margs->ift_tsl, 1645 margs->ift_flags)) 1646 goto next_ire; 1647 1648 if (ire->ire_type & IRE_INTERFACE) { 1649 /* 1650 * keep looking to see if there is a non-interface 1651 * default ire, but save this one as a last resort. 1652 */ 1653 if (maybe_ire == NULL) 1654 maybe_ire = ire; 1655 goto next_ire; 1656 } 1657 1658 if (zoneid == ALL_ZONES) { 1659 IRE_REFHOLD(ire); 1660 IRB_REFRELE(irb_ptr); 1661 return (ire); 1662 } 1663 /* 1664 * When we're in a local zone, we're only 1665 * interested in routers that are 1666 * reachable through ipifs within our zone. 1667 */ 1668 if (ire->ire_ipif != NULL) { 1669 match_flags |= MATCH_IRE_ILL_GROUP; 1670 } 1671 rire = ire_route_lookup(ire->ire_gateway_addr, 1672 0, 0, 0, ire->ire_ipif, NULL, zoneid, margs->ift_tsl, 1673 match_flags); 1674 if (rire != NULL) { 1675 ire_refrele(rire); 1676 IRE_REFHOLD(ire); 1677 IRB_REFRELE(irb_ptr); 1678 return (ire); 1679 } 1680 next_ire: 1681 ire = (ire->ire_next ? ire->ire_next : irb_ptr->irb_ire); 1682 if (ire == ire_origin) 1683 break; 1684 } 1685 if (maybe_ire != NULL) 1686 IRE_REFHOLD(maybe_ire); 1687 IRB_REFRELE(irb_ptr); 1688 return (maybe_ire); 1689 } 1690