1 /*- 2 * Copyright (c) 1980, 1986, 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 4. Neither the name of the University nor the names of its contributors 14 * may be used to endorse or promote products derived from this software 15 * without specific prior written permission. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 20 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 27 * SUCH DAMAGE. 28 * 29 * @(#)route.c 8.3.1.1 (Berkeley) 2/23/95 30 * $FreeBSD$ 31 */ 32 33 #include "opt_inet.h" 34 #include "opt_mrouting.h" 35 36 #include <sys/param.h> 37 #include <sys/systm.h> 38 #include <sys/malloc.h> 39 #include <sys/mbuf.h> 40 #include <sys/socket.h> 41 #include <sys/domain.h> 42 #include <sys/kernel.h> 43 44 #include <net/if.h> 45 #include <net/route.h> 46 47 #include <netinet/in.h> 48 #include <netinet/ip_mroute.h> 49 50 #include <vm/uma.h> 51 52 static struct rtstat rtstat; 53 struct radix_node_head *rt_tables[AF_MAX+1]; 54 55 static int rttrash; /* routes not in table but not freed */ 56 57 static void rt_maskedcopy(struct sockaddr *, 58 struct sockaddr *, struct sockaddr *); 59 static void rtable_init(void **); 60 61 /* compare two sockaddr structures */ 62 #define sa_equal(a1, a2) (bcmp((a1), (a2), (a1)->sa_len) == 0) 63 64 /* 65 * Convert a 'struct radix_node *' to a 'struct rtentry *'. 66 * The operation can be done safely (in this code) because a 67 * 'struct rtentry' starts with two 'struct radix_node''s, the first 68 * one representing leaf nodes in the routing tree, which is 69 * what the code in radix.c passes us as a 'struct radix_node'. 70 * 71 * But because there are a lot of assumptions in this conversion, 72 * do not cast explicitly, but always use the macro below. 73 */ 74 #define RNTORT(p) ((struct rtentry *)(p)) 75 76 static void 77 rtable_init(void **table) 78 { 79 struct domain *dom; 80 for (dom = domains; dom; dom = dom->dom_next) 81 if (dom->dom_rtattach) 82 dom->dom_rtattach(&table[dom->dom_family], 83 dom->dom_rtoffset); 84 } 85 86 static uma_zone_t rtzone; /* Routing table UMA zone. */ 87 88 static void 89 route_init(void) 90 { 91 rtzone = uma_zcreate("rtentry", sizeof(struct rtentry), NULL, NULL, 92 NULL, NULL, UMA_ALIGN_PTR, 0); 93 rn_init(); /* initialize all zeroes, all ones, mask table */ 94 rtable_init((void **)rt_tables); 95 } 96 97 /* 98 * Packet routing routines. 99 */ 100 void 101 rtalloc(struct route *ro) 102 { 103 rtalloc_ign(ro, 0UL); 104 } 105 106 void 107 rtalloc_ign(struct route *ro, u_long ignore) 108 { 109 struct rtentry *rt; 110 111 if ((rt = ro->ro_rt) != NULL) { 112 if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP) 113 return; 114 RTFREE(rt); 115 ro->ro_rt = NULL; 116 } 117 ro->ro_rt = rtalloc1(&ro->ro_dst, 1, ignore); 118 if (ro->ro_rt) 119 RT_UNLOCK(ro->ro_rt); 120 } 121 122 /* 123 * Look up the route that matches the address given 124 * Or, at least try.. Create a cloned route if needed. 125 * 126 * The returned route, if any, is locked. 127 */ 128 struct rtentry * 129 rtalloc1(struct sockaddr *dst, int report, u_long ignflags) 130 { 131 struct radix_node_head *rnh = rt_tables[dst->sa_family]; 132 struct rtentry *rt; 133 struct radix_node *rn; 134 struct rtentry *newrt; 135 struct rt_addrinfo info; 136 u_long nflags; 137 int err = 0, msgtype = RTM_MISS; 138 139 newrt = NULL; 140 /* 141 * Look up the address in the table for that Address Family 142 */ 143 if (rnh == NULL) { 144 rtstat.rts_unreach++; 145 goto miss2; 146 } 147 RADIX_NODE_HEAD_LOCK(rnh); 148 if ((rn = rnh->rnh_matchaddr(dst, rnh)) && 149 (rn->rn_flags & RNF_ROOT) == 0) { 150 /* 151 * If we find it and it's not the root node, then 152 * get a refernce on the rtentry associated. 153 */ 154 newrt = rt = RNTORT(rn); 155 nflags = rt->rt_flags & ~ignflags; 156 if (report && (nflags & RTF_CLONING)) { 157 /* 158 * We are apparently adding (report = 0 in delete). 159 * If it requires that it be cloned, do so. 160 * (This implies it wasn't a HOST route.) 161 */ 162 err = rtrequest(RTM_RESOLVE, dst, NULL, 163 NULL, 0, &newrt); 164 if (err) { 165 /* 166 * If the cloning didn't succeed, maybe 167 * what we have will do. Return that. 168 */ 169 newrt = rt; /* existing route */ 170 RT_LOCK(newrt); 171 RT_ADDREF(newrt); 172 goto miss; 173 } 174 KASSERT(newrt, ("no route and no error")); 175 RT_LOCK(newrt); 176 if (newrt->rt_flags & RTF_XRESOLVE) { 177 /* 178 * If the new route specifies it be 179 * externally resolved, then go do that. 180 */ 181 msgtype = RTM_RESOLVE; 182 goto miss; 183 } 184 /* Inform listeners of the new route. */ 185 bzero(&info, sizeof(info)); 186 info.rti_info[RTAX_DST] = rt_key(newrt); 187 info.rti_info[RTAX_NETMASK] = rt_mask(newrt); 188 info.rti_info[RTAX_GATEWAY] = newrt->rt_gateway; 189 if (newrt->rt_ifp != NULL) { 190 info.rti_info[RTAX_IFP] = 191 newrt->rt_ifp->if_addr->ifa_addr; 192 info.rti_info[RTAX_IFA] = newrt->rt_ifa->ifa_addr; 193 } 194 rt_missmsg(RTM_ADD, &info, newrt->rt_flags, 0); 195 } else { 196 KASSERT(rt == newrt, ("locking wrong route")); 197 RT_LOCK(newrt); 198 RT_ADDREF(newrt); 199 } 200 RADIX_NODE_HEAD_UNLOCK(rnh); 201 } else { 202 /* 203 * Either we hit the root or couldn't find any match, 204 * Which basically means 205 * "caint get there frm here" 206 */ 207 rtstat.rts_unreach++; 208 miss: 209 RADIX_NODE_HEAD_UNLOCK(rnh); 210 miss2: if (report) { 211 /* 212 * If required, report the failure to the supervising 213 * Authorities. 214 * For a delete, this is not an error. (report == 0) 215 */ 216 bzero(&info, sizeof(info)); 217 info.rti_info[RTAX_DST] = dst; 218 rt_missmsg(msgtype, &info, 0, err); 219 } 220 } 221 if (newrt) 222 RT_LOCK_ASSERT(newrt); 223 return (newrt); 224 } 225 226 /* 227 * Remove a reference count from an rtentry. 228 * If the count gets low enough, take it out of the routing table 229 */ 230 void 231 rtfree(struct rtentry *rt) 232 { 233 struct radix_node_head *rnh; 234 235 /* XXX the NULL checks are probably useless */ 236 if (rt == NULL) 237 panic("rtfree: NULL rt"); 238 rnh = rt_tables[rt_key(rt)->sa_family]; 239 if (rnh == NULL) 240 panic("rtfree: NULL rnh"); 241 242 RT_LOCK_ASSERT(rt); 243 244 /* 245 * decrement the reference count by one and if it reaches 0, 246 * and there is a close function defined, call the close function 247 */ 248 RT_REMREF(rt); 249 if (rt->rt_refcnt > 0) 250 goto done; 251 252 /* 253 * On last reference give the "close method" a chance 254 * to cleanup private state. This also permits (for 255 * IPv4 and IPv6) a chance to decide if the routing table 256 * entry should be purged immediately or at a later time. 257 * When an immediate purge is to happen the close routine 258 * typically calls rtexpunge which clears the RTF_UP flag 259 * on the entry so that the code below reclaims the storage. 260 */ 261 if (rt->rt_refcnt == 0 && rnh->rnh_close) 262 rnh->rnh_close((struct radix_node *)rt, rnh); 263 264 /* 265 * If we are no longer "up" (and ref == 0) 266 * then we can free the resources associated 267 * with the route. 268 */ 269 if ((rt->rt_flags & RTF_UP) == 0) { 270 if (rt->rt_nodes->rn_flags & (RNF_ACTIVE | RNF_ROOT)) 271 panic ("rtfree 2"); 272 /* 273 * the rtentry must have been removed from the routing table 274 * so it is represented in rttrash.. remove that now. 275 */ 276 rttrash--; 277 #ifdef DIAGNOSTIC 278 if (rt->rt_refcnt < 0) { 279 printf("rtfree: %p not freed (neg refs)\n", rt); 280 goto done; 281 } 282 #endif 283 /* 284 * release references on items we hold them on.. 285 * e.g other routes and ifaddrs. 286 */ 287 if (rt->rt_ifa) 288 IFAFREE(rt->rt_ifa); 289 rt->rt_parent = NULL; /* NB: no refcnt on parent */ 290 291 /* 292 * The key is separatly alloc'd so free it (see rt_setgate()). 293 * This also frees the gateway, as they are always malloc'd 294 * together. 295 */ 296 Free(rt_key(rt)); 297 298 /* 299 * and the rtentry itself of course 300 */ 301 RT_LOCK_DESTROY(rt); 302 uma_zfree(rtzone, rt); 303 return; 304 } 305 done: 306 RT_UNLOCK(rt); 307 } 308 309 310 /* 311 * Force a routing table entry to the specified 312 * destination to go through the given gateway. 313 * Normally called as a result of a routing redirect 314 * message from the network layer. 315 */ 316 void 317 rtredirect(struct sockaddr *dst, 318 struct sockaddr *gateway, 319 struct sockaddr *netmask, 320 int flags, 321 struct sockaddr *src) 322 { 323 struct rtentry *rt; 324 int error = 0; 325 short *stat = NULL; 326 struct rt_addrinfo info; 327 struct ifaddr *ifa; 328 329 /* verify the gateway is directly reachable */ 330 if ((ifa = ifa_ifwithnet(gateway)) == NULL) { 331 error = ENETUNREACH; 332 goto out; 333 } 334 rt = rtalloc1(dst, 0, 0UL); /* NB: rt is locked */ 335 /* 336 * If the redirect isn't from our current router for this dst, 337 * it's either old or wrong. If it redirects us to ourselves, 338 * we have a routing loop, perhaps as a result of an interface 339 * going down recently. 340 */ 341 if (!(flags & RTF_DONE) && rt && 342 (!sa_equal(src, rt->rt_gateway) || rt->rt_ifa != ifa)) 343 error = EINVAL; 344 else if (ifa_ifwithaddr(gateway)) 345 error = EHOSTUNREACH; 346 if (error) 347 goto done; 348 /* 349 * Create a new entry if we just got back a wildcard entry 350 * or the the lookup failed. This is necessary for hosts 351 * which use routing redirects generated by smart gateways 352 * to dynamically build the routing tables. 353 */ 354 if (rt == NULL || (rt_mask(rt) && rt_mask(rt)->sa_len < 2)) 355 goto create; 356 /* 357 * Don't listen to the redirect if it's 358 * for a route to an interface. 359 */ 360 if (rt->rt_flags & RTF_GATEWAY) { 361 if (((rt->rt_flags & RTF_HOST) == 0) && (flags & RTF_HOST)) { 362 /* 363 * Changing from route to net => route to host. 364 * Create new route, rather than smashing route to net. 365 */ 366 create: 367 if (rt) 368 rtfree(rt); 369 flags |= RTF_GATEWAY | RTF_DYNAMIC; 370 bzero((caddr_t)&info, sizeof(info)); 371 info.rti_info[RTAX_DST] = dst; 372 info.rti_info[RTAX_GATEWAY] = gateway; 373 info.rti_info[RTAX_NETMASK] = netmask; 374 info.rti_ifa = ifa; 375 info.rti_flags = flags; 376 rt = NULL; 377 error = rtrequest1(RTM_ADD, &info, &rt); 378 if (rt != NULL) { 379 RT_LOCK(rt); 380 flags = rt->rt_flags; 381 } 382 stat = &rtstat.rts_dynamic; 383 } else { 384 /* 385 * Smash the current notion of the gateway to 386 * this destination. Should check about netmask!!! 387 */ 388 rt->rt_flags |= RTF_MODIFIED; 389 flags |= RTF_MODIFIED; 390 stat = &rtstat.rts_newgateway; 391 /* 392 * add the key and gateway (in one malloc'd chunk). 393 */ 394 rt_setgate(rt, rt_key(rt), gateway); 395 } 396 } else 397 error = EHOSTUNREACH; 398 done: 399 if (rt) 400 rtfree(rt); 401 out: 402 if (error) 403 rtstat.rts_badredirect++; 404 else if (stat != NULL) 405 (*stat)++; 406 bzero((caddr_t)&info, sizeof(info)); 407 info.rti_info[RTAX_DST] = dst; 408 info.rti_info[RTAX_GATEWAY] = gateway; 409 info.rti_info[RTAX_NETMASK] = netmask; 410 info.rti_info[RTAX_AUTHOR] = src; 411 rt_missmsg(RTM_REDIRECT, &info, flags, error); 412 } 413 414 /* 415 * Routing table ioctl interface. 416 */ 417 int 418 rtioctl(u_long req, caddr_t data) 419 { 420 421 /* 422 * If more ioctl commands are added here, make sure the proper 423 * super-user checks are being performed because it is possible for 424 * prison-root to make it this far if raw sockets have been enabled 425 * in jails. 426 */ 427 #ifdef INET 428 /* Multicast goop, grrr... */ 429 return mrt_ioctl ? mrt_ioctl(req, data) : EOPNOTSUPP; 430 #else /* INET */ 431 return ENXIO; 432 #endif /* INET */ 433 } 434 435 struct ifaddr * 436 ifa_ifwithroute(int flags, struct sockaddr *dst, struct sockaddr *gateway) 437 { 438 register struct ifaddr *ifa; 439 int not_found = 0; 440 441 if ((flags & RTF_GATEWAY) == 0) { 442 /* 443 * If we are adding a route to an interface, 444 * and the interface is a pt to pt link 445 * we should search for the destination 446 * as our clue to the interface. Otherwise 447 * we can use the local address. 448 */ 449 ifa = NULL; 450 if (flags & RTF_HOST) 451 ifa = ifa_ifwithdstaddr(dst); 452 if (ifa == NULL) 453 ifa = ifa_ifwithaddr(gateway); 454 } else { 455 /* 456 * If we are adding a route to a remote net 457 * or host, the gateway may still be on the 458 * other end of a pt to pt link. 459 */ 460 ifa = ifa_ifwithdstaddr(gateway); 461 } 462 if (ifa == NULL) 463 ifa = ifa_ifwithnet(gateway); 464 if (ifa == NULL) { 465 struct rtentry *rt = rtalloc1(gateway, 0, 0UL); 466 if (rt == NULL) 467 return (NULL); 468 /* 469 * dismiss a gateway that is reachable only 470 * through the default router 471 */ 472 switch (gateway->sa_family) { 473 case AF_INET: 474 if (satosin(rt_key(rt))->sin_addr.s_addr == INADDR_ANY) 475 not_found = 1; 476 break; 477 case AF_INET6: 478 if (IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(rt))->sin6_addr)) 479 not_found = 1; 480 break; 481 default: 482 break; 483 } 484 RT_REMREF(rt); 485 RT_UNLOCK(rt); 486 if (not_found) 487 return (NULL); 488 if ((ifa = rt->rt_ifa) == NULL) 489 return (NULL); 490 } 491 if (ifa->ifa_addr->sa_family != dst->sa_family) { 492 struct ifaddr *oifa = ifa; 493 ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp); 494 if (ifa == NULL) 495 ifa = oifa; 496 } 497 return (ifa); 498 } 499 500 static walktree_f_t rt_fixdelete; 501 static walktree_f_t rt_fixchange; 502 503 struct rtfc_arg { 504 struct rtentry *rt0; 505 struct radix_node_head *rnh; 506 }; 507 508 /* 509 * Do appropriate manipulations of a routing tree given 510 * all the bits of info needed 511 */ 512 int 513 rtrequest(int req, 514 struct sockaddr *dst, 515 struct sockaddr *gateway, 516 struct sockaddr *netmask, 517 int flags, 518 struct rtentry **ret_nrt) 519 { 520 struct rt_addrinfo info; 521 522 if (dst->sa_len == 0) 523 return(EINVAL); 524 525 bzero((caddr_t)&info, sizeof(info)); 526 info.rti_flags = flags; 527 info.rti_info[RTAX_DST] = dst; 528 info.rti_info[RTAX_GATEWAY] = gateway; 529 info.rti_info[RTAX_NETMASK] = netmask; 530 return rtrequest1(req, &info, ret_nrt); 531 } 532 533 /* 534 * These (questionable) definitions of apparent local variables apply 535 * to the next two functions. XXXXXX!!! 536 */ 537 #define dst info->rti_info[RTAX_DST] 538 #define gateway info->rti_info[RTAX_GATEWAY] 539 #define netmask info->rti_info[RTAX_NETMASK] 540 #define ifaaddr info->rti_info[RTAX_IFA] 541 #define ifpaddr info->rti_info[RTAX_IFP] 542 #define flags info->rti_flags 543 544 int 545 rt_getifa(struct rt_addrinfo *info) 546 { 547 struct ifaddr *ifa; 548 int error = 0; 549 550 /* 551 * ifp may be specified by sockaddr_dl 552 * when protocol address is ambiguous. 553 */ 554 if (info->rti_ifp == NULL && ifpaddr != NULL && 555 ifpaddr->sa_family == AF_LINK && 556 (ifa = ifa_ifwithnet(ifpaddr)) != NULL) 557 info->rti_ifp = ifa->ifa_ifp; 558 if (info->rti_ifa == NULL && ifaaddr != NULL) 559 info->rti_ifa = ifa_ifwithaddr(ifaaddr); 560 if (info->rti_ifa == NULL) { 561 struct sockaddr *sa; 562 563 sa = ifaaddr != NULL ? ifaaddr : 564 (gateway != NULL ? gateway : dst); 565 if (sa != NULL && info->rti_ifp != NULL) 566 info->rti_ifa = ifaof_ifpforaddr(sa, info->rti_ifp); 567 else if (dst != NULL && gateway != NULL) 568 info->rti_ifa = ifa_ifwithroute(flags, dst, gateway); 569 else if (sa != NULL) 570 info->rti_ifa = ifa_ifwithroute(flags, sa, sa); 571 } 572 if ((ifa = info->rti_ifa) != NULL) { 573 if (info->rti_ifp == NULL) 574 info->rti_ifp = ifa->ifa_ifp; 575 } else 576 error = ENETUNREACH; 577 return (error); 578 } 579 580 /* 581 * Expunges references to a route that's about to be reclaimed. 582 * The route must be locked. 583 */ 584 int 585 rtexpunge(struct rtentry *rt) 586 { 587 struct radix_node *rn; 588 struct radix_node_head *rnh; 589 struct ifaddr *ifa; 590 int error = 0; 591 592 RT_LOCK_ASSERT(rt); 593 #if 0 594 /* 595 * We cannot assume anything about the reference count 596 * because protocols call us in many situations; often 597 * before unwinding references to the table entry. 598 */ 599 KASSERT(rt->rt_refcnt <= 1, ("bogus refcnt %ld", rt->rt_refcnt)); 600 #endif 601 /* 602 * Find the correct routing tree to use for this Address Family 603 */ 604 rnh = rt_tables[rt_key(rt)->sa_family]; 605 if (rnh == NULL) 606 return (EAFNOSUPPORT); 607 608 RADIX_NODE_HEAD_LOCK(rnh); 609 610 /* 611 * Remove the item from the tree; it should be there, 612 * but when callers invoke us blindly it may not (sigh). 613 */ 614 rn = rnh->rnh_deladdr(rt_key(rt), rt_mask(rt), rnh); 615 if (rn == NULL) { 616 error = ESRCH; 617 goto bad; 618 } 619 KASSERT((rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)) == 0, 620 ("unexpected flags 0x%x", rn->rn_flags)); 621 KASSERT(rt == RNTORT(rn), 622 ("lookup mismatch, rt %p rn %p", rt, rn)); 623 624 rt->rt_flags &= ~RTF_UP; 625 626 /* 627 * Now search what's left of the subtree for any cloned 628 * routes which might have been formed from this node. 629 */ 630 if ((rt->rt_flags & RTF_CLONING) && rt_mask(rt)) 631 rnh->rnh_walktree_from(rnh, rt_key(rt), rt_mask(rt), 632 rt_fixdelete, rt); 633 634 /* 635 * Remove any external references we may have. 636 * This might result in another rtentry being freed if 637 * we held its last reference. 638 */ 639 if (rt->rt_gwroute) { 640 RTFREE(rt->rt_gwroute); 641 rt->rt_gwroute = NULL; 642 } 643 644 /* 645 * Give the protocol a chance to keep things in sync. 646 */ 647 if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest) { 648 struct rt_addrinfo info; 649 650 bzero((caddr_t)&info, sizeof(info)); 651 info.rti_flags = rt->rt_flags; 652 info.rti_info[RTAX_DST] = rt_key(rt); 653 info.rti_info[RTAX_GATEWAY] = rt->rt_gateway; 654 info.rti_info[RTAX_NETMASK] = rt_mask(rt); 655 ifa->ifa_rtrequest(RTM_DELETE, rt, &info); 656 } 657 658 /* 659 * one more rtentry floating around that is not 660 * linked to the routing table. 661 */ 662 rttrash++; 663 bad: 664 RADIX_NODE_HEAD_UNLOCK(rnh); 665 return (error); 666 } 667 668 int 669 rtrequest1(int req, struct rt_addrinfo *info, struct rtentry **ret_nrt) 670 { 671 int error = 0; 672 register struct rtentry *rt; 673 register struct radix_node *rn; 674 register struct radix_node_head *rnh; 675 struct ifaddr *ifa; 676 struct sockaddr *ndst; 677 #define senderr(x) { error = x ; goto bad; } 678 679 /* 680 * Find the correct routing tree to use for this Address Family 681 */ 682 rnh = rt_tables[dst->sa_family]; 683 if (rnh == NULL) 684 return (EAFNOSUPPORT); 685 RADIX_NODE_HEAD_LOCK(rnh); 686 /* 687 * If we are adding a host route then we don't want to put 688 * a netmask in the tree, nor do we want to clone it. 689 */ 690 if (flags & RTF_HOST) { 691 netmask = NULL; 692 flags &= ~RTF_CLONING; 693 } 694 switch (req) { 695 case RTM_DELETE: 696 /* 697 * Remove the item from the tree and return it. 698 * Complain if it is not there and do no more processing. 699 */ 700 rn = rnh->rnh_deladdr(dst, netmask, rnh); 701 if (rn == NULL) 702 senderr(ESRCH); 703 if (rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)) 704 panic ("rtrequest delete"); 705 rt = RNTORT(rn); 706 RT_LOCK(rt); 707 RT_ADDREF(rt); 708 rt->rt_flags &= ~RTF_UP; 709 710 /* 711 * Now search what's left of the subtree for any cloned 712 * routes which might have been formed from this node. 713 */ 714 if ((rt->rt_flags & RTF_CLONING) && 715 rt_mask(rt)) { 716 rnh->rnh_walktree_from(rnh, dst, rt_mask(rt), 717 rt_fixdelete, rt); 718 } 719 720 /* 721 * Remove any external references we may have. 722 * This might result in another rtentry being freed if 723 * we held its last reference. 724 */ 725 if (rt->rt_gwroute) { 726 RTFREE(rt->rt_gwroute); 727 rt->rt_gwroute = NULL; 728 } 729 730 /* 731 * give the protocol a chance to keep things in sync. 732 */ 733 if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest) 734 ifa->ifa_rtrequest(RTM_DELETE, rt, info); 735 736 /* 737 * One more rtentry floating around that is not 738 * linked to the routing table. rttrash will be decremented 739 * when RTFREE(rt) is eventually called. 740 */ 741 rttrash++; 742 743 /* 744 * If the caller wants it, then it can have it, 745 * but it's up to it to free the rtentry as we won't be 746 * doing it. 747 */ 748 if (ret_nrt) { 749 *ret_nrt = rt; 750 RT_UNLOCK(rt); 751 } else 752 RTFREE_LOCKED(rt); 753 break; 754 755 case RTM_RESOLVE: 756 if (ret_nrt == NULL || (rt = *ret_nrt) == NULL) 757 senderr(EINVAL); 758 ifa = rt->rt_ifa; 759 /* XXX locking? */ 760 flags = rt->rt_flags & 761 ~(RTF_CLONING | RTF_STATIC); 762 flags |= RTF_WASCLONED; 763 gateway = rt->rt_gateway; 764 if ((netmask = rt->rt_genmask) == NULL) 765 flags |= RTF_HOST; 766 goto makeroute; 767 768 case RTM_ADD: 769 if ((flags & RTF_GATEWAY) && !gateway) 770 senderr(EINVAL); 771 if (dst && gateway && (dst->sa_family != gateway->sa_family) && 772 (gateway->sa_family != AF_UNSPEC) && (gateway->sa_family != AF_LINK)) 773 senderr(EINVAL); 774 775 if (info->rti_ifa == NULL && (error = rt_getifa(info))) 776 senderr(error); 777 ifa = info->rti_ifa; 778 779 makeroute: 780 rt = uma_zalloc(rtzone, M_NOWAIT | M_ZERO); 781 if (rt == NULL) 782 senderr(ENOBUFS); 783 RT_LOCK_INIT(rt); 784 rt->rt_flags = RTF_UP | flags; 785 /* 786 * Add the gateway. Possibly re-malloc-ing the storage for it 787 * also add the rt_gwroute if possible. 788 */ 789 RT_LOCK(rt); 790 if ((error = rt_setgate(rt, dst, gateway)) != 0) { 791 RT_LOCK_DESTROY(rt); 792 uma_zfree(rtzone, rt); 793 senderr(error); 794 } 795 796 /* 797 * point to the (possibly newly malloc'd) dest address. 798 */ 799 ndst = (struct sockaddr *)rt_key(rt); 800 801 /* 802 * make sure it contains the value we want (masked if needed). 803 */ 804 if (netmask) { 805 rt_maskedcopy(dst, ndst, netmask); 806 } else 807 bcopy(dst, ndst, dst->sa_len); 808 809 /* 810 * Note that we now have a reference to the ifa. 811 * This moved from below so that rnh->rnh_addaddr() can 812 * examine the ifa and ifa->ifa_ifp if it so desires. 813 */ 814 IFAREF(ifa); 815 rt->rt_ifa = ifa; 816 rt->rt_ifp = ifa->ifa_ifp; 817 818 /* XXX mtu manipulation will be done in rnh_addaddr -- itojun */ 819 rn = rnh->rnh_addaddr(ndst, netmask, rnh, rt->rt_nodes); 820 if (rn == NULL) { 821 struct rtentry *rt2; 822 /* 823 * Uh-oh, we already have one of these in the tree. 824 * We do a special hack: if the route that's already 825 * there was generated by the cloning mechanism 826 * then we just blow it away and retry the insertion 827 * of the new one. 828 */ 829 rt2 = rtalloc1(dst, 0, 0); 830 if (rt2 && rt2->rt_parent) { 831 rtexpunge(rt2); 832 RT_UNLOCK(rt2); 833 rn = rnh->rnh_addaddr(ndst, netmask, 834 rnh, rt->rt_nodes); 835 } else if (rt2) { 836 /* undo the extra ref we got */ 837 RTFREE_LOCKED(rt2); 838 } 839 } 840 841 /* 842 * If it still failed to go into the tree, 843 * then un-make it (this should be a function) 844 */ 845 if (rn == NULL) { 846 if (rt->rt_gwroute) 847 RTFREE(rt->rt_gwroute); 848 if (rt->rt_ifa) 849 IFAFREE(rt->rt_ifa); 850 Free(rt_key(rt)); 851 RT_LOCK_DESTROY(rt); 852 uma_zfree(rtzone, rt); 853 senderr(EEXIST); 854 } 855 856 rt->rt_parent = NULL; 857 858 /* 859 * If we got here from RESOLVE, then we are cloning 860 * so clone the rest, and note that we 861 * are a clone (and increment the parent's references) 862 */ 863 if (req == RTM_RESOLVE) { 864 KASSERT(ret_nrt && *ret_nrt, 865 ("no route to clone from")); 866 rt->rt_rmx = (*ret_nrt)->rt_rmx; /* copy metrics */ 867 rt->rt_rmx.rmx_pksent = 0; /* reset packet counter */ 868 if ((*ret_nrt)->rt_flags & RTF_CLONING) { 869 /* 870 * NB: We do not bump the refcnt on the parent 871 * entry under the assumption that it will 872 * remain so long as we do. This is 873 * important when deleting the parent route 874 * as this operation requires traversing 875 * the tree to delete all clones and futzing 876 * with refcnts requires us to double-lock 877 * parent through this back reference. 878 */ 879 rt->rt_parent = *ret_nrt; 880 } 881 } 882 883 /* 884 * if this protocol has something to add to this then 885 * allow it to do that as well. 886 */ 887 if (ifa->ifa_rtrequest) 888 ifa->ifa_rtrequest(req, rt, info); 889 890 /* 891 * We repeat the same procedure from rt_setgate() here because 892 * it doesn't fire when we call it there because the node 893 * hasn't been added to the tree yet. 894 */ 895 if (req == RTM_ADD && 896 !(rt->rt_flags & RTF_HOST) && rt_mask(rt) != NULL) { 897 struct rtfc_arg arg; 898 arg.rnh = rnh; 899 arg.rt0 = rt; 900 rnh->rnh_walktree_from(rnh, rt_key(rt), rt_mask(rt), 901 rt_fixchange, &arg); 902 } 903 904 /* 905 * actually return a resultant rtentry and 906 * give the caller a single reference. 907 */ 908 if (ret_nrt) { 909 *ret_nrt = rt; 910 RT_ADDREF(rt); 911 } 912 RT_UNLOCK(rt); 913 break; 914 default: 915 error = EOPNOTSUPP; 916 } 917 bad: 918 RADIX_NODE_HEAD_UNLOCK(rnh); 919 return (error); 920 #undef senderr 921 } 922 923 #undef dst 924 #undef gateway 925 #undef netmask 926 #undef ifaaddr 927 #undef ifpaddr 928 #undef flags 929 930 /* 931 * Called from rtrequest(RTM_DELETE, ...) to fix up the route's ``family'' 932 * (i.e., the routes related to it by the operation of cloning). This 933 * routine is iterated over all potential former-child-routes by way of 934 * rnh->rnh_walktree_from() above, and those that actually are children of 935 * the late parent (passed in as VP here) are themselves deleted. 936 */ 937 static int 938 rt_fixdelete(struct radix_node *rn, void *vp) 939 { 940 struct rtentry *rt = RNTORT(rn); 941 struct rtentry *rt0 = vp; 942 943 if (rt->rt_parent == rt0 && 944 !(rt->rt_flags & (RTF_PINNED | RTF_CLONING))) { 945 return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 946 rt->rt_flags, NULL); 947 } 948 return 0; 949 } 950 951 /* 952 * This routine is called from rt_setgate() to do the analogous thing for 953 * adds and changes. There is the added complication in this case of a 954 * middle insert; i.e., insertion of a new network route between an older 955 * network route and (cloned) host routes. For this reason, a simple check 956 * of rt->rt_parent is insufficient; each candidate route must be tested 957 * against the (mask, value) of the new route (passed as before in vp) 958 * to see if the new route matches it. 959 * 960 * XXX - it may be possible to do fixdelete() for changes and reserve this 961 * routine just for adds. I'm not sure why I thought it was necessary to do 962 * changes this way. 963 */ 964 965 static int 966 rt_fixchange(struct radix_node *rn, void *vp) 967 { 968 struct rtentry *rt = RNTORT(rn); 969 struct rtfc_arg *ap = vp; 970 struct rtentry *rt0 = ap->rt0; 971 struct radix_node_head *rnh = ap->rnh; 972 u_char *xk1, *xm1, *xk2, *xmp; 973 int i, len, mlen; 974 975 /* make sure we have a parent, and route is not pinned or cloning */ 976 if (!rt->rt_parent || 977 (rt->rt_flags & (RTF_PINNED | RTF_CLONING))) 978 return 0; 979 980 if (rt->rt_parent == rt0) /* parent match */ 981 goto delete_rt; 982 /* 983 * There probably is a function somewhere which does this... 984 * if not, there should be. 985 */ 986 len = imin(rt_key(rt0)->sa_len, rt_key(rt)->sa_len); 987 988 xk1 = (u_char *)rt_key(rt0); 989 xm1 = (u_char *)rt_mask(rt0); 990 xk2 = (u_char *)rt_key(rt); 991 992 /* avoid applying a less specific route */ 993 xmp = (u_char *)rt_mask(rt->rt_parent); 994 mlen = rt_key(rt->rt_parent)->sa_len; 995 if (mlen > rt_key(rt0)->sa_len) /* less specific route */ 996 return 0; 997 for (i = rnh->rnh_treetop->rn_offset; i < mlen; i++) 998 if ((xmp[i] & ~(xmp[i] ^ xm1[i])) != xmp[i]) 999 return 0; /* less specific route */ 1000 1001 for (i = rnh->rnh_treetop->rn_offset; i < len; i++) 1002 if ((xk2[i] & xm1[i]) != xk1[i]) 1003 return 0; /* no match */ 1004 1005 /* 1006 * OK, this node is a clone, and matches the node currently being 1007 * changed/added under the node's mask. So, get rid of it. 1008 */ 1009 delete_rt: 1010 return rtrequest(RTM_DELETE, rt_key(rt), NULL, 1011 rt_mask(rt), rt->rt_flags, NULL); 1012 } 1013 1014 int 1015 rt_setgate(struct rtentry *rt, struct sockaddr *dst, struct sockaddr *gate) 1016 { 1017 /* XXX dst may be overwritten, can we move this to below */ 1018 struct radix_node_head *rnh = rt_tables[dst->sa_family]; 1019 int dlen = SA_SIZE(dst), glen = SA_SIZE(gate); 1020 1021 RT_LOCK_ASSERT(rt); 1022 1023 /* 1024 * A host route with the destination equal to the gateway 1025 * will interfere with keeping LLINFO in the routing 1026 * table, so disallow it. 1027 */ 1028 if (((rt->rt_flags & (RTF_HOST|RTF_GATEWAY|RTF_LLINFO)) == 1029 (RTF_HOST|RTF_GATEWAY)) && 1030 dst->sa_len == gate->sa_len && 1031 bcmp(dst, gate, dst->sa_len) == 0) { 1032 /* 1033 * The route might already exist if this is an RTM_CHANGE 1034 * or a routing redirect, so try to delete it. 1035 */ 1036 if (rt_key(rt)) 1037 rtexpunge(rt); 1038 return EADDRNOTAVAIL; 1039 } 1040 1041 /* 1042 * Cloning loop avoidance in case of bad configuration. 1043 */ 1044 if (rt->rt_flags & RTF_GATEWAY) { 1045 struct rtentry *gwrt; 1046 1047 RT_UNLOCK(rt); /* XXX workaround LOR */ 1048 gwrt = rtalloc1(gate, 1, 0); 1049 if (gwrt == rt) { 1050 RT_LOCK_ASSERT(rt); 1051 RT_REMREF(rt); 1052 return (EADDRINUSE); /* failure */ 1053 } 1054 RT_LOCK(rt); 1055 /* 1056 * If there is already a gwroute, then drop it. If we 1057 * are asked to replace route with itself, then do 1058 * not leak its refcounter. 1059 */ 1060 if (rt->rt_gwroute != NULL) { 1061 if (rt->rt_gwroute == gwrt) { 1062 RT_REMREF(rt->rt_gwroute); 1063 } else 1064 RTFREE(rt->rt_gwroute); 1065 } 1066 1067 if ((rt->rt_gwroute = gwrt) != NULL) 1068 RT_UNLOCK(rt->rt_gwroute); 1069 } 1070 1071 /* 1072 * Prepare to store the gateway in rt->rt_gateway. 1073 * Both dst and gateway are stored one after the other in the same 1074 * malloc'd chunk. If we have room, we can reuse the old buffer, 1075 * rt_gateway already points to the right place. 1076 * Otherwise, malloc a new block and update the 'dst' address. 1077 */ 1078 if (rt->rt_gateway == NULL || glen > SA_SIZE(rt->rt_gateway)) { 1079 caddr_t new; 1080 1081 R_Malloc(new, caddr_t, dlen + glen); 1082 if (new == NULL) 1083 return ENOBUFS; 1084 /* 1085 * XXX note, we copy from *dst and not *rt_key(rt) because 1086 * rt_setgate() can be called to initialize a newly 1087 * allocated route entry, in which case rt_key(rt) == NULL 1088 * (and also rt->rt_gateway == NULL). 1089 * Free()/free() handle a NULL argument just fine. 1090 */ 1091 bcopy(dst, new, dlen); 1092 Free(rt_key(rt)); /* free old block, if any */ 1093 rt_key(rt) = (struct sockaddr *)new; 1094 rt->rt_gateway = (struct sockaddr *)(new + dlen); 1095 } 1096 1097 /* 1098 * Copy the new gateway value into the memory chunk. 1099 */ 1100 bcopy(gate, rt->rt_gateway, glen); 1101 1102 /* 1103 * This isn't going to do anything useful for host routes, so 1104 * don't bother. Also make sure we have a reasonable mask 1105 * (we don't yet have one during adds). 1106 */ 1107 if (!(rt->rt_flags & RTF_HOST) && rt_mask(rt) != 0) { 1108 struct rtfc_arg arg; 1109 1110 arg.rnh = rnh; 1111 arg.rt0 = rt; 1112 RT_UNLOCK(rt); /* XXX workaround LOR */ 1113 RADIX_NODE_HEAD_LOCK(rnh); 1114 RT_LOCK(rt); 1115 rnh->rnh_walktree_from(rnh, rt_key(rt), rt_mask(rt), 1116 rt_fixchange, &arg); 1117 RADIX_NODE_HEAD_UNLOCK(rnh); 1118 } 1119 1120 return 0; 1121 } 1122 1123 static void 1124 rt_maskedcopy(struct sockaddr *src, struct sockaddr *dst, struct sockaddr *netmask) 1125 { 1126 register u_char *cp1 = (u_char *)src; 1127 register u_char *cp2 = (u_char *)dst; 1128 register u_char *cp3 = (u_char *)netmask; 1129 u_char *cplim = cp2 + *cp3; 1130 u_char *cplim2 = cp2 + *cp1; 1131 1132 *cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */ 1133 cp3 += 2; 1134 if (cplim > cplim2) 1135 cplim = cplim2; 1136 while (cp2 < cplim) 1137 *cp2++ = *cp1++ & *cp3++; 1138 if (cp2 < cplim2) 1139 bzero((caddr_t)cp2, (unsigned)(cplim2 - cp2)); 1140 } 1141 1142 /* 1143 * Set up a routing table entry, normally 1144 * for an interface. 1145 */ 1146 int 1147 rtinit(struct ifaddr *ifa, int cmd, int flags) 1148 { 1149 struct sockaddr *dst; 1150 struct sockaddr *netmask; 1151 struct mbuf *m = NULL; 1152 struct rtentry *rt = NULL; 1153 struct rt_addrinfo info; 1154 int error; 1155 1156 if (flags & RTF_HOST) { 1157 dst = ifa->ifa_dstaddr; 1158 netmask = NULL; 1159 } else { 1160 dst = ifa->ifa_addr; 1161 netmask = ifa->ifa_netmask; 1162 } 1163 if (dst->sa_len == 0) 1164 return(EINVAL); 1165 1166 /* 1167 * If it's a delete, check that if it exists, it's on the correct 1168 * interface or we might scrub a route to another ifa which would 1169 * be confusing at best and possibly worse. 1170 */ 1171 if (cmd == RTM_DELETE) { 1172 struct sockaddr *deldst; 1173 struct radix_node_head *rnh; 1174 struct radix_node *rn; 1175 1176 /* 1177 * It's a delete, so it should already exist.. 1178 * If it's a net, mask off the host bits 1179 * (Assuming we have a mask) 1180 */ 1181 if (netmask != NULL) { 1182 m = m_get(M_DONTWAIT, MT_SONAME); 1183 if (m == NULL) 1184 return(ENOBUFS); 1185 deldst = mtod(m, struct sockaddr *); 1186 rt_maskedcopy(dst, deldst, netmask); 1187 dst = deldst; 1188 } 1189 /* 1190 * Look up an rtentry that is in the routing tree and 1191 * contains the correct info. 1192 */ 1193 if ((rnh = rt_tables[dst->sa_family]) == NULL) 1194 goto bad; 1195 RADIX_NODE_HEAD_LOCK(rnh); 1196 error = ((rn = rnh->rnh_lookup(dst, netmask, rnh)) == NULL || 1197 (rn->rn_flags & RNF_ROOT) || 1198 RNTORT(rn)->rt_ifa != ifa || 1199 !sa_equal((struct sockaddr *)rn->rn_key, dst)); 1200 RADIX_NODE_HEAD_UNLOCK(rnh); 1201 if (error) { 1202 bad: 1203 if (m) 1204 (void) m_free(m); 1205 return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH); 1206 } 1207 } 1208 /* 1209 * Do the actual request 1210 */ 1211 bzero((caddr_t)&info, sizeof(info)); 1212 info.rti_ifa = ifa; 1213 info.rti_flags = flags | ifa->ifa_flags; 1214 info.rti_info[RTAX_DST] = dst; 1215 info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr; 1216 info.rti_info[RTAX_NETMASK] = netmask; 1217 error = rtrequest1(cmd, &info, &rt); 1218 if (error == 0 && rt != NULL) { 1219 /* 1220 * notify any listening routing agents of the change 1221 */ 1222 RT_LOCK(rt); 1223 rt_newaddrmsg(cmd, ifa, error, rt); 1224 if (cmd == RTM_DELETE) { 1225 /* 1226 * If we are deleting, and we found an entry, then 1227 * it's been removed from the tree.. now throw it away. 1228 */ 1229 RTFREE_LOCKED(rt); 1230 } else { 1231 if (cmd == RTM_ADD) { 1232 /* 1233 * We just wanted to add it.. we don't actually 1234 * need a reference. 1235 */ 1236 RT_REMREF(rt); 1237 } 1238 RT_UNLOCK(rt); 1239 } 1240 } 1241 if (m) 1242 (void) m_free(m); 1243 return (error); 1244 } 1245 1246 /* 1247 * rt_check() is invoked on each layer 2 output path, prior to 1248 * encapsulating outbound packets. 1249 * 1250 * The function is mostly used to find a routing entry for the gateway, 1251 * which in some protocol families could also point to the link-level 1252 * address for the gateway itself (the side effect of revalidating the 1253 * route to the destination is rather pointless at this stage, we did it 1254 * already a moment before in the pr_output() routine to locate the ifp 1255 * and gateway to use). 1256 * 1257 * When we remove the layer-3 to layer-2 mapping tables from the 1258 * routing table, this function can be removed. 1259 * 1260 * === On input === 1261 * *dst is the address of the NEXT HOP (which coincides with the 1262 * final destination if directly reachable); 1263 * *lrt0 points to the cached route to the final destination; 1264 * *lrt is not meaningful; 1265 * 1266 * === Operation === 1267 * If the route is marked down try to find a new route. If the route 1268 * to the gateway is gone, try to setup a new route. Otherwise, 1269 * if the route is marked for packets to be rejected, enforce that. 1270 * 1271 * === On return === 1272 * *dst is unchanged; 1273 * *lrt0 points to the (possibly new) route to the final destination 1274 * *lrt points to the route to the next hop 1275 * 1276 * Their values are meaningful ONLY if no error is returned. 1277 */ 1278 int 1279 rt_check(struct rtentry **lrt, struct rtentry **lrt0, struct sockaddr *dst) 1280 { 1281 #define senderr(x) { error = x ; goto bad; } 1282 struct rtentry *rt; 1283 struct rtentry *rt0; 1284 int error; 1285 1286 KASSERT(*lrt0 != NULL, ("rt_check")); 1287 rt = rt0 = *lrt0; 1288 1289 /* NB: the locking here is tortuous... */ 1290 RT_LOCK(rt); 1291 if ((rt->rt_flags & RTF_UP) == 0) { 1292 RT_UNLOCK(rt); 1293 rt = rtalloc1(dst, 1, 0UL); 1294 if (rt != NULL) { 1295 RT_REMREF(rt); 1296 /* XXX what about if change? */ 1297 } else 1298 senderr(EHOSTUNREACH); 1299 rt0 = rt; 1300 } 1301 /* XXX BSD/OS checks dst->sa_family != AF_NS */ 1302 if (rt->rt_flags & RTF_GATEWAY) { 1303 if (rt->rt_gwroute == NULL) 1304 goto lookup; 1305 rt = rt->rt_gwroute; 1306 RT_LOCK(rt); /* NB: gwroute */ 1307 if ((rt->rt_flags & RTF_UP) == 0) { 1308 rtfree(rt); /* unlock gwroute */ 1309 rt = rt0; 1310 lookup: 1311 RT_UNLOCK(rt0); 1312 rt = rtalloc1(rt->rt_gateway, 1, 0UL); 1313 if (rt == rt0) { 1314 rt0->rt_gwroute = NULL; 1315 RT_REMREF(rt0); 1316 RT_UNLOCK(rt0); 1317 senderr(ENETUNREACH); 1318 } 1319 RT_LOCK(rt0); 1320 rt0->rt_gwroute = rt; 1321 if (rt == NULL) { 1322 RT_UNLOCK(rt0); 1323 senderr(EHOSTUNREACH); 1324 } 1325 } 1326 RT_UNLOCK(rt0); 1327 } 1328 /* XXX why are we inspecting rmx_expire? */ 1329 error = (rt->rt_flags & RTF_REJECT) && 1330 (rt->rt_rmx.rmx_expire == 0 || 1331 time_uptime < rt->rt_rmx.rmx_expire); 1332 if (error) { 1333 RT_UNLOCK(rt); 1334 senderr(rt == rt0 ? EHOSTDOWN : EHOSTUNREACH); 1335 } 1336 1337 *lrt = rt; 1338 *lrt0 = rt0; 1339 return (0); 1340 bad: 1341 /* NB: lrt and lrt0 should not be interpreted if error is non-zero */ 1342 return (error); 1343 #undef senderr 1344 } 1345 1346 /* This must be before ip6_init2(), which is now SI_ORDER_MIDDLE */ 1347 SYSINIT(route, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0); 1348