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