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; 438 439 if ((flags & RTF_GATEWAY) == 0) { 440 /* 441 * If we are adding a route to an interface, 442 * and the interface is a pt to pt link 443 * we should search for the destination 444 * as our clue to the interface. Otherwise 445 * we can use the local address. 446 */ 447 ifa = NULL; 448 if (flags & RTF_HOST) 449 ifa = ifa_ifwithdstaddr(dst); 450 if (ifa == NULL) 451 ifa = ifa_ifwithaddr(gateway); 452 } else { 453 /* 454 * If we are adding a route to a remote net 455 * or host, the gateway may still be on the 456 * other end of a pt to pt link. 457 */ 458 ifa = ifa_ifwithdstaddr(gateway); 459 } 460 if (ifa == NULL) 461 ifa = ifa_ifwithnet(gateway); 462 if (ifa == NULL) { 463 struct rtentry *rt = rtalloc1(gateway, 0, 0UL); 464 if (rt == NULL) 465 return (NULL); 466 RT_REMREF(rt); 467 RT_UNLOCK(rt); 468 if ((ifa = rt->rt_ifa) == NULL) 469 return (NULL); 470 } 471 if (ifa->ifa_addr->sa_family != dst->sa_family) { 472 struct ifaddr *oifa = ifa; 473 ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp); 474 if (ifa == NULL) 475 ifa = oifa; 476 } 477 return (ifa); 478 } 479 480 static walktree_f_t rt_fixdelete; 481 static walktree_f_t rt_fixchange; 482 483 struct rtfc_arg { 484 struct rtentry *rt0; 485 struct radix_node_head *rnh; 486 }; 487 488 /* 489 * Do appropriate manipulations of a routing tree given 490 * all the bits of info needed 491 */ 492 int 493 rtrequest(int req, 494 struct sockaddr *dst, 495 struct sockaddr *gateway, 496 struct sockaddr *netmask, 497 int flags, 498 struct rtentry **ret_nrt) 499 { 500 struct rt_addrinfo info; 501 502 if (dst->sa_len == 0) 503 return(EINVAL); 504 505 bzero((caddr_t)&info, sizeof(info)); 506 info.rti_flags = flags; 507 info.rti_info[RTAX_DST] = dst; 508 info.rti_info[RTAX_GATEWAY] = gateway; 509 info.rti_info[RTAX_NETMASK] = netmask; 510 return rtrequest1(req, &info, ret_nrt); 511 } 512 513 /* 514 * These (questionable) definitions of apparent local variables apply 515 * to the next two functions. XXXXXX!!! 516 */ 517 #define dst info->rti_info[RTAX_DST] 518 #define gateway info->rti_info[RTAX_GATEWAY] 519 #define netmask info->rti_info[RTAX_NETMASK] 520 #define ifaaddr info->rti_info[RTAX_IFA] 521 #define ifpaddr info->rti_info[RTAX_IFP] 522 #define flags info->rti_flags 523 524 int 525 rt_getifa(struct rt_addrinfo *info) 526 { 527 struct ifaddr *ifa; 528 int error = 0; 529 530 /* 531 * ifp may be specified by sockaddr_dl 532 * when protocol address is ambiguous. 533 */ 534 if (info->rti_ifp == NULL && ifpaddr != NULL && 535 ifpaddr->sa_family == AF_LINK && 536 (ifa = ifa_ifwithnet(ifpaddr)) != NULL) 537 info->rti_ifp = ifa->ifa_ifp; 538 if (info->rti_ifa == NULL && ifaaddr != NULL) 539 info->rti_ifa = ifa_ifwithaddr(ifaaddr); 540 if (info->rti_ifa == NULL) { 541 struct sockaddr *sa; 542 543 sa = ifaaddr != NULL ? ifaaddr : 544 (gateway != NULL ? gateway : dst); 545 if (sa != NULL && info->rti_ifp != NULL) 546 info->rti_ifa = ifaof_ifpforaddr(sa, info->rti_ifp); 547 else if (dst != NULL && gateway != NULL) 548 info->rti_ifa = ifa_ifwithroute(flags, dst, gateway); 549 else if (sa != NULL) 550 info->rti_ifa = ifa_ifwithroute(flags, sa, sa); 551 } 552 if ((ifa = info->rti_ifa) != NULL) { 553 if (info->rti_ifp == NULL) 554 info->rti_ifp = ifa->ifa_ifp; 555 } else 556 error = ENETUNREACH; 557 return (error); 558 } 559 560 /* 561 * Expunges references to a route that's about to be reclaimed. 562 * The route must be locked. 563 */ 564 int 565 rtexpunge(struct rtentry *rt) 566 { 567 struct radix_node *rn; 568 struct radix_node_head *rnh; 569 struct ifaddr *ifa; 570 int error = 0; 571 572 RT_LOCK_ASSERT(rt); 573 #if 0 574 /* 575 * We cannot assume anything about the reference count 576 * because protocols call us in many situations; often 577 * before unwinding references to the table entry. 578 */ 579 KASSERT(rt->rt_refcnt <= 1, ("bogus refcnt %ld", rt->rt_refcnt)); 580 #endif 581 /* 582 * Find the correct routing tree to use for this Address Family 583 */ 584 rnh = rt_tables[rt_key(rt)->sa_family]; 585 if (rnh == NULL) 586 return (EAFNOSUPPORT); 587 588 RADIX_NODE_HEAD_LOCK(rnh); 589 590 /* 591 * Remove the item from the tree; it should be there, 592 * but when callers invoke us blindly it may not (sigh). 593 */ 594 rn = rnh->rnh_deladdr(rt_key(rt), rt_mask(rt), rnh); 595 if (rn == NULL) { 596 error = ESRCH; 597 goto bad; 598 } 599 KASSERT((rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)) == 0, 600 ("unexpected flags 0x%x", rn->rn_flags)); 601 KASSERT(rt == RNTORT(rn), 602 ("lookup mismatch, rt %p rn %p", rt, rn)); 603 604 rt->rt_flags &= ~RTF_UP; 605 606 /* 607 * Now search what's left of the subtree for any cloned 608 * routes which might have been formed from this node. 609 */ 610 if ((rt->rt_flags & RTF_CLONING) && rt_mask(rt)) 611 rnh->rnh_walktree_from(rnh, rt_key(rt), rt_mask(rt), 612 rt_fixdelete, rt); 613 614 /* 615 * Remove any external references we may have. 616 * This might result in another rtentry being freed if 617 * we held its last reference. 618 */ 619 if (rt->rt_gwroute) { 620 RTFREE(rt->rt_gwroute); 621 rt->rt_gwroute = NULL; 622 } 623 624 /* 625 * Give the protocol a chance to keep things in sync. 626 */ 627 if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest) { 628 struct rt_addrinfo info; 629 630 bzero((caddr_t)&info, sizeof(info)); 631 info.rti_flags = rt->rt_flags; 632 info.rti_info[RTAX_DST] = rt_key(rt); 633 info.rti_info[RTAX_GATEWAY] = rt->rt_gateway; 634 info.rti_info[RTAX_NETMASK] = rt_mask(rt); 635 ifa->ifa_rtrequest(RTM_DELETE, rt, &info); 636 } 637 638 /* 639 * one more rtentry floating around that is not 640 * linked to the routing table. 641 */ 642 rttrash++; 643 bad: 644 RADIX_NODE_HEAD_UNLOCK(rnh); 645 return (error); 646 } 647 648 int 649 rtrequest1(int req, struct rt_addrinfo *info, struct rtentry **ret_nrt) 650 { 651 int error = 0; 652 register struct rtentry *rt; 653 register struct radix_node *rn; 654 register struct radix_node_head *rnh; 655 struct ifaddr *ifa; 656 struct sockaddr *ndst; 657 #define senderr(x) { error = x ; goto bad; } 658 659 /* 660 * Find the correct routing tree to use for this Address Family 661 */ 662 rnh = rt_tables[dst->sa_family]; 663 if (rnh == NULL) 664 return (EAFNOSUPPORT); 665 RADIX_NODE_HEAD_LOCK(rnh); 666 /* 667 * If we are adding a host route then we don't want to put 668 * a netmask in the tree, nor do we want to clone it. 669 */ 670 if (flags & RTF_HOST) { 671 netmask = NULL; 672 flags &= ~RTF_CLONING; 673 } 674 switch (req) { 675 case RTM_DELETE: 676 /* 677 * Remove the item from the tree and return it. 678 * Complain if it is not there and do no more processing. 679 */ 680 rn = rnh->rnh_deladdr(dst, netmask, rnh); 681 if (rn == NULL) 682 senderr(ESRCH); 683 if (rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)) 684 panic ("rtrequest delete"); 685 rt = RNTORT(rn); 686 RT_LOCK(rt); 687 RT_ADDREF(rt); 688 rt->rt_flags &= ~RTF_UP; 689 690 /* 691 * Now search what's left of the subtree for any cloned 692 * routes which might have been formed from this node. 693 */ 694 if ((rt->rt_flags & RTF_CLONING) && 695 rt_mask(rt)) { 696 rnh->rnh_walktree_from(rnh, dst, rt_mask(rt), 697 rt_fixdelete, rt); 698 } 699 700 /* 701 * Remove any external references we may have. 702 * This might result in another rtentry being freed if 703 * we held its last reference. 704 */ 705 if (rt->rt_gwroute) { 706 RTFREE(rt->rt_gwroute); 707 rt->rt_gwroute = NULL; 708 } 709 710 /* 711 * give the protocol a chance to keep things in sync. 712 */ 713 if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest) 714 ifa->ifa_rtrequest(RTM_DELETE, rt, info); 715 716 /* 717 * One more rtentry floating around that is not 718 * linked to the routing table. rttrash will be decremented 719 * when RTFREE(rt) is eventually called. 720 */ 721 rttrash++; 722 723 /* 724 * If the caller wants it, then it can have it, 725 * but it's up to it to free the rtentry as we won't be 726 * doing it. 727 */ 728 if (ret_nrt) { 729 *ret_nrt = rt; 730 RT_UNLOCK(rt); 731 } else 732 RTFREE_LOCKED(rt); 733 break; 734 735 case RTM_RESOLVE: 736 if (ret_nrt == NULL || (rt = *ret_nrt) == NULL) 737 senderr(EINVAL); 738 ifa = rt->rt_ifa; 739 /* XXX locking? */ 740 flags = rt->rt_flags & 741 ~(RTF_CLONING | RTF_STATIC); 742 flags |= RTF_WASCLONED; 743 gateway = rt->rt_gateway; 744 if ((netmask = rt->rt_genmask) == NULL) 745 flags |= RTF_HOST; 746 goto makeroute; 747 748 case RTM_ADD: 749 if ((flags & RTF_GATEWAY) && !gateway) 750 senderr(EINVAL); 751 if (dst && gateway && (dst->sa_family != gateway->sa_family) && 752 (gateway->sa_family != AF_UNSPEC) && (gateway->sa_family != AF_LINK)) 753 senderr(EINVAL); 754 755 if (info->rti_ifa == NULL && (error = rt_getifa(info))) 756 senderr(error); 757 ifa = info->rti_ifa; 758 759 makeroute: 760 rt = uma_zalloc(rtzone, M_NOWAIT | M_ZERO); 761 if (rt == NULL) 762 senderr(ENOBUFS); 763 RT_LOCK_INIT(rt); 764 rt->rt_flags = RTF_UP | flags; 765 /* 766 * Add the gateway. Possibly re-malloc-ing the storage for it 767 * also add the rt_gwroute if possible. 768 */ 769 RT_LOCK(rt); 770 if ((error = rt_setgate(rt, dst, gateway)) != 0) { 771 RT_LOCK_DESTROY(rt); 772 uma_zfree(rtzone, rt); 773 senderr(error); 774 } 775 776 /* 777 * point to the (possibly newly malloc'd) dest address. 778 */ 779 ndst = (struct sockaddr *)rt_key(rt); 780 781 /* 782 * make sure it contains the value we want (masked if needed). 783 */ 784 if (netmask) { 785 rt_maskedcopy(dst, ndst, netmask); 786 } else 787 bcopy(dst, ndst, dst->sa_len); 788 789 /* 790 * Note that we now have a reference to the ifa. 791 * This moved from below so that rnh->rnh_addaddr() can 792 * examine the ifa and ifa->ifa_ifp if it so desires. 793 */ 794 IFAREF(ifa); 795 rt->rt_ifa = ifa; 796 rt->rt_ifp = ifa->ifa_ifp; 797 798 /* XXX mtu manipulation will be done in rnh_addaddr -- itojun */ 799 rn = rnh->rnh_addaddr(ndst, netmask, rnh, rt->rt_nodes); 800 if (rn == NULL) { 801 struct rtentry *rt2; 802 /* 803 * Uh-oh, we already have one of these in the tree. 804 * We do a special hack: if the route that's already 805 * there was generated by the cloning mechanism 806 * then we just blow it away and retry the insertion 807 * of the new one. 808 */ 809 rt2 = rtalloc1(dst, 0, 0); 810 if (rt2 && rt2->rt_parent) { 811 rtexpunge(rt2); 812 RT_UNLOCK(rt2); 813 rn = rnh->rnh_addaddr(ndst, netmask, 814 rnh, rt->rt_nodes); 815 } else if (rt2) { 816 /* undo the extra ref we got */ 817 RTFREE_LOCKED(rt2); 818 } 819 } 820 821 /* 822 * If it still failed to go into the tree, 823 * then un-make it (this should be a function) 824 */ 825 if (rn == NULL) { 826 if (rt->rt_gwroute) 827 RTFREE(rt->rt_gwroute); 828 if (rt->rt_ifa) 829 IFAFREE(rt->rt_ifa); 830 Free(rt_key(rt)); 831 RT_LOCK_DESTROY(rt); 832 uma_zfree(rtzone, rt); 833 senderr(EEXIST); 834 } 835 836 rt->rt_parent = NULL; 837 838 /* 839 * If we got here from RESOLVE, then we are cloning 840 * so clone the rest, and note that we 841 * are a clone (and increment the parent's references) 842 */ 843 if (req == RTM_RESOLVE) { 844 KASSERT(ret_nrt && *ret_nrt, 845 ("no route to clone from")); 846 rt->rt_rmx = (*ret_nrt)->rt_rmx; /* copy metrics */ 847 rt->rt_rmx.rmx_pksent = 0; /* reset packet counter */ 848 if ((*ret_nrt)->rt_flags & RTF_CLONING) { 849 /* 850 * NB: We do not bump the refcnt on the parent 851 * entry under the assumption that it will 852 * remain so long as we do. This is 853 * important when deleting the parent route 854 * as this operation requires traversing 855 * the tree to delete all clones and futzing 856 * with refcnts requires us to double-lock 857 * parent through this back reference. 858 */ 859 rt->rt_parent = *ret_nrt; 860 } 861 } 862 863 /* 864 * if this protocol has something to add to this then 865 * allow it to do that as well. 866 */ 867 if (ifa->ifa_rtrequest) 868 ifa->ifa_rtrequest(req, rt, info); 869 870 /* 871 * We repeat the same procedure from rt_setgate() here because 872 * it doesn't fire when we call it there because the node 873 * hasn't been added to the tree yet. 874 */ 875 if (req == RTM_ADD && 876 !(rt->rt_flags & RTF_HOST) && rt_mask(rt) != NULL) { 877 struct rtfc_arg arg; 878 arg.rnh = rnh; 879 arg.rt0 = rt; 880 rnh->rnh_walktree_from(rnh, rt_key(rt), rt_mask(rt), 881 rt_fixchange, &arg); 882 } 883 884 /* 885 * actually return a resultant rtentry and 886 * give the caller a single reference. 887 */ 888 if (ret_nrt) { 889 *ret_nrt = rt; 890 RT_ADDREF(rt); 891 } 892 RT_UNLOCK(rt); 893 break; 894 default: 895 error = EOPNOTSUPP; 896 } 897 bad: 898 RADIX_NODE_HEAD_UNLOCK(rnh); 899 return (error); 900 #undef senderr 901 } 902 903 #undef dst 904 #undef gateway 905 #undef netmask 906 #undef ifaaddr 907 #undef ifpaddr 908 #undef flags 909 910 /* 911 * Called from rtrequest(RTM_DELETE, ...) to fix up the route's ``family'' 912 * (i.e., the routes related to it by the operation of cloning). This 913 * routine is iterated over all potential former-child-routes by way of 914 * rnh->rnh_walktree_from() above, and those that actually are children of 915 * the late parent (passed in as VP here) are themselves deleted. 916 */ 917 static int 918 rt_fixdelete(struct radix_node *rn, void *vp) 919 { 920 struct rtentry *rt = RNTORT(rn); 921 struct rtentry *rt0 = vp; 922 923 if (rt->rt_parent == rt0 && 924 !(rt->rt_flags & (RTF_PINNED | RTF_CLONING))) { 925 return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 926 rt->rt_flags, NULL); 927 } 928 return 0; 929 } 930 931 /* 932 * This routine is called from rt_setgate() to do the analogous thing for 933 * adds and changes. There is the added complication in this case of a 934 * middle insert; i.e., insertion of a new network route between an older 935 * network route and (cloned) host routes. For this reason, a simple check 936 * of rt->rt_parent is insufficient; each candidate route must be tested 937 * against the (mask, value) of the new route (passed as before in vp) 938 * to see if the new route matches it. 939 * 940 * XXX - it may be possible to do fixdelete() for changes and reserve this 941 * routine just for adds. I'm not sure why I thought it was necessary to do 942 * changes this way. 943 */ 944 945 static int 946 rt_fixchange(struct radix_node *rn, void *vp) 947 { 948 struct rtentry *rt = RNTORT(rn); 949 struct rtfc_arg *ap = vp; 950 struct rtentry *rt0 = ap->rt0; 951 struct radix_node_head *rnh = ap->rnh; 952 u_char *xk1, *xm1, *xk2, *xmp; 953 int i, len, mlen; 954 955 /* make sure we have a parent, and route is not pinned or cloning */ 956 if (!rt->rt_parent || 957 (rt->rt_flags & (RTF_PINNED | RTF_CLONING))) 958 return 0; 959 960 if (rt->rt_parent == rt0) /* parent match */ 961 goto delete_rt; 962 /* 963 * There probably is a function somewhere which does this... 964 * if not, there should be. 965 */ 966 len = imin(rt_key(rt0)->sa_len, rt_key(rt)->sa_len); 967 968 xk1 = (u_char *)rt_key(rt0); 969 xm1 = (u_char *)rt_mask(rt0); 970 xk2 = (u_char *)rt_key(rt); 971 972 /* avoid applying a less specific route */ 973 xmp = (u_char *)rt_mask(rt->rt_parent); 974 mlen = rt_key(rt->rt_parent)->sa_len; 975 if (mlen > rt_key(rt0)->sa_len) /* less specific route */ 976 return 0; 977 for (i = rnh->rnh_treetop->rn_offset; i < mlen; i++) 978 if ((xmp[i] & ~(xmp[i] ^ xm1[i])) != xmp[i]) 979 return 0; /* less specific route */ 980 981 for (i = rnh->rnh_treetop->rn_offset; i < len; i++) 982 if ((xk2[i] & xm1[i]) != xk1[i]) 983 return 0; /* no match */ 984 985 /* 986 * OK, this node is a clone, and matches the node currently being 987 * changed/added under the node's mask. So, get rid of it. 988 */ 989 delete_rt: 990 return rtrequest(RTM_DELETE, rt_key(rt), NULL, 991 rt_mask(rt), rt->rt_flags, NULL); 992 } 993 994 int 995 rt_setgate(struct rtentry *rt, struct sockaddr *dst, struct sockaddr *gate) 996 { 997 /* XXX dst may be overwritten, can we move this to below */ 998 struct radix_node_head *rnh = rt_tables[dst->sa_family]; 999 int dlen = SA_SIZE(dst), glen = SA_SIZE(gate); 1000 1001 RT_LOCK_ASSERT(rt); 1002 1003 /* 1004 * A host route with the destination equal to the gateway 1005 * will interfere with keeping LLINFO in the routing 1006 * table, so disallow it. 1007 */ 1008 if (((rt->rt_flags & (RTF_HOST|RTF_GATEWAY|RTF_LLINFO)) == 1009 (RTF_HOST|RTF_GATEWAY)) && 1010 dst->sa_len == gate->sa_len && 1011 bcmp(dst, gate, dst->sa_len) == 0) { 1012 /* 1013 * The route might already exist if this is an RTM_CHANGE 1014 * or a routing redirect, so try to delete it. 1015 */ 1016 if (rt_key(rt)) 1017 rtexpunge(rt); 1018 return EADDRNOTAVAIL; 1019 } 1020 1021 /* 1022 * Cloning loop avoidance in case of bad configuration. 1023 */ 1024 if (rt->rt_flags & RTF_GATEWAY) { 1025 struct rtentry *gwrt; 1026 1027 RT_UNLOCK(rt); /* XXX workaround LOR */ 1028 gwrt = rtalloc1(gate, 1, 0); 1029 if (gwrt == rt) { 1030 RT_LOCK_ASSERT(rt); 1031 RT_REMREF(rt); 1032 return (EADDRINUSE); /* failure */ 1033 } 1034 RT_LOCK(rt); 1035 /* 1036 * If there is already a gwroute, then drop it. If we 1037 * are asked to replace route with itself, then do 1038 * not leak its refcounter. 1039 */ 1040 if (rt->rt_gwroute != NULL) { 1041 if (rt->rt_gwroute == gwrt) { 1042 RT_REMREF(rt->rt_gwroute); 1043 } else 1044 RTFREE(rt->rt_gwroute); 1045 } 1046 1047 if ((rt->rt_gwroute = gwrt) != NULL) 1048 RT_UNLOCK(rt->rt_gwroute); 1049 } 1050 1051 /* 1052 * Prepare to store the gateway in rt->rt_gateway. 1053 * Both dst and gateway are stored one after the other in the same 1054 * malloc'd chunk. If we have room, we can reuse the old buffer, 1055 * rt_gateway already points to the right place. 1056 * Otherwise, malloc a new block and update the 'dst' address. 1057 */ 1058 if (rt->rt_gateway == NULL || glen > SA_SIZE(rt->rt_gateway)) { 1059 caddr_t new; 1060 1061 R_Malloc(new, caddr_t, dlen + glen); 1062 if (new == NULL) 1063 return ENOBUFS; 1064 /* 1065 * XXX note, we copy from *dst and not *rt_key(rt) because 1066 * rt_setgate() can be called to initialize a newly 1067 * allocated route entry, in which case rt_key(rt) == NULL 1068 * (and also rt->rt_gateway == NULL). 1069 * Free()/free() handle a NULL argument just fine. 1070 */ 1071 bcopy(dst, new, dlen); 1072 Free(rt_key(rt)); /* free old block, if any */ 1073 rt_key(rt) = (struct sockaddr *)new; 1074 rt->rt_gateway = (struct sockaddr *)(new + dlen); 1075 } 1076 1077 /* 1078 * Copy the new gateway value into the memory chunk. 1079 */ 1080 bcopy(gate, rt->rt_gateway, glen); 1081 1082 /* 1083 * This isn't going to do anything useful for host routes, so 1084 * don't bother. Also make sure we have a reasonable mask 1085 * (we don't yet have one during adds). 1086 */ 1087 if (!(rt->rt_flags & RTF_HOST) && rt_mask(rt) != 0) { 1088 struct rtfc_arg arg; 1089 1090 arg.rnh = rnh; 1091 arg.rt0 = rt; 1092 RT_UNLOCK(rt); /* XXX workaround LOR */ 1093 RADIX_NODE_HEAD_LOCK(rnh); 1094 RT_LOCK(rt); 1095 rnh->rnh_walktree_from(rnh, rt_key(rt), rt_mask(rt), 1096 rt_fixchange, &arg); 1097 RADIX_NODE_HEAD_UNLOCK(rnh); 1098 } 1099 1100 return 0; 1101 } 1102 1103 static void 1104 rt_maskedcopy(struct sockaddr *src, struct sockaddr *dst, struct sockaddr *netmask) 1105 { 1106 register u_char *cp1 = (u_char *)src; 1107 register u_char *cp2 = (u_char *)dst; 1108 register u_char *cp3 = (u_char *)netmask; 1109 u_char *cplim = cp2 + *cp3; 1110 u_char *cplim2 = cp2 + *cp1; 1111 1112 *cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */ 1113 cp3 += 2; 1114 if (cplim > cplim2) 1115 cplim = cplim2; 1116 while (cp2 < cplim) 1117 *cp2++ = *cp1++ & *cp3++; 1118 if (cp2 < cplim2) 1119 bzero((caddr_t)cp2, (unsigned)(cplim2 - cp2)); 1120 } 1121 1122 /* 1123 * Set up a routing table entry, normally 1124 * for an interface. 1125 */ 1126 int 1127 rtinit(struct ifaddr *ifa, int cmd, int flags) 1128 { 1129 struct sockaddr *dst; 1130 struct sockaddr *netmask; 1131 struct mbuf *m = NULL; 1132 struct rtentry *rt = NULL; 1133 struct rt_addrinfo info; 1134 int error; 1135 1136 if (flags & RTF_HOST) { 1137 dst = ifa->ifa_dstaddr; 1138 netmask = NULL; 1139 } else { 1140 dst = ifa->ifa_addr; 1141 netmask = ifa->ifa_netmask; 1142 } 1143 if (dst->sa_len == 0) 1144 return(EINVAL); 1145 1146 /* 1147 * If it's a delete, check that if it exists, it's on the correct 1148 * interface or we might scrub a route to another ifa which would 1149 * be confusing at best and possibly worse. 1150 */ 1151 if (cmd == RTM_DELETE) { 1152 struct sockaddr *deldst; 1153 struct radix_node_head *rnh; 1154 struct radix_node *rn; 1155 1156 /* 1157 * It's a delete, so it should already exist.. 1158 * If it's a net, mask off the host bits 1159 * (Assuming we have a mask) 1160 */ 1161 if (netmask != NULL) { 1162 m = m_get(M_DONTWAIT, MT_SONAME); 1163 if (m == NULL) 1164 return(ENOBUFS); 1165 deldst = mtod(m, struct sockaddr *); 1166 rt_maskedcopy(dst, deldst, netmask); 1167 dst = deldst; 1168 } 1169 /* 1170 * Look up an rtentry that is in the routing tree and 1171 * contains the correct info. 1172 */ 1173 if ((rnh = rt_tables[dst->sa_family]) == NULL) 1174 goto bad; 1175 RADIX_NODE_HEAD_LOCK(rnh); 1176 error = ((rn = rnh->rnh_lookup(dst, netmask, rnh)) == NULL || 1177 (rn->rn_flags & RNF_ROOT) || 1178 RNTORT(rn)->rt_ifa != ifa || 1179 !sa_equal((struct sockaddr *)rn->rn_key, dst)); 1180 RADIX_NODE_HEAD_UNLOCK(rnh); 1181 if (error) { 1182 bad: 1183 if (m) 1184 (void) m_free(m); 1185 return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH); 1186 } 1187 } 1188 /* 1189 * Do the actual request 1190 */ 1191 bzero((caddr_t)&info, sizeof(info)); 1192 info.rti_ifa = ifa; 1193 info.rti_flags = flags | ifa->ifa_flags; 1194 info.rti_info[RTAX_DST] = dst; 1195 info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr; 1196 info.rti_info[RTAX_NETMASK] = netmask; 1197 error = rtrequest1(cmd, &info, &rt); 1198 if (error == 0 && rt != NULL) { 1199 /* 1200 * notify any listening routing agents of the change 1201 */ 1202 RT_LOCK(rt); 1203 rt_newaddrmsg(cmd, ifa, error, rt); 1204 if (cmd == RTM_DELETE) { 1205 /* 1206 * If we are deleting, and we found an entry, then 1207 * it's been removed from the tree.. now throw it away. 1208 */ 1209 RTFREE_LOCKED(rt); 1210 } else { 1211 if (cmd == RTM_ADD) { 1212 /* 1213 * We just wanted to add it.. we don't actually 1214 * need a reference. 1215 */ 1216 RT_REMREF(rt); 1217 } 1218 RT_UNLOCK(rt); 1219 } 1220 } 1221 if (m) 1222 (void) m_free(m); 1223 return (error); 1224 } 1225 1226 /* 1227 * rt_check() is invoked on each layer 2 output path, prior to 1228 * encapsulating outbound packets. 1229 * 1230 * The function is mostly used to find a routing entry for the gateway, 1231 * which in some protocol families could also point to the link-level 1232 * address for the gateway itself (the side effect of revalidating the 1233 * route to the destination is rather pointless at this stage, we did it 1234 * already a moment before in the pr_output() routine to locate the ifp 1235 * and gateway to use). 1236 * 1237 * When we remove the layer-3 to layer-2 mapping tables from the 1238 * routing table, this function can be removed. 1239 * 1240 * === On input === 1241 * *dst is the address of the NEXT HOP (which coincides with the 1242 * final destination if directly reachable); 1243 * *lrt0 points to the cached route to the final destination; 1244 * *lrt is not meaningful; 1245 * 1246 * === Operation === 1247 * If the route is marked down try to find a new route. If the route 1248 * to the gateway is gone, try to setup a new route. Otherwise, 1249 * if the route is marked for packets to be rejected, enforce that. 1250 * 1251 * === On return === 1252 * *dst is unchanged; 1253 * *lrt0 points to the (possibly new) route to the final destination 1254 * *lrt points to the route to the next hop 1255 * 1256 * Their values are meaningful ONLY if no error is returned. 1257 */ 1258 int 1259 rt_check(struct rtentry **lrt, struct rtentry **lrt0, struct sockaddr *dst) 1260 { 1261 #define senderr(x) { error = x ; goto bad; } 1262 struct rtentry *rt; 1263 struct rtentry *rt0; 1264 int error; 1265 1266 KASSERT(*lrt0 != NULL, ("rt_check")); 1267 rt = rt0 = *lrt0; 1268 1269 /* NB: the locking here is tortuous... */ 1270 RT_LOCK(rt); 1271 if ((rt->rt_flags & RTF_UP) == 0) { 1272 RT_UNLOCK(rt); 1273 rt = rtalloc1(dst, 1, 0UL); 1274 if (rt != NULL) { 1275 RT_REMREF(rt); 1276 /* XXX what about if change? */ 1277 } else 1278 senderr(EHOSTUNREACH); 1279 rt0 = rt; 1280 } 1281 /* XXX BSD/OS checks dst->sa_family != AF_NS */ 1282 if (rt->rt_flags & RTF_GATEWAY) { 1283 if (rt->rt_gwroute == NULL) 1284 goto lookup; 1285 rt = rt->rt_gwroute; 1286 RT_LOCK(rt); /* NB: gwroute */ 1287 if ((rt->rt_flags & RTF_UP) == 0) { 1288 rtfree(rt); /* unlock gwroute */ 1289 rt = rt0; 1290 lookup: 1291 RT_UNLOCK(rt0); 1292 rt = rtalloc1(rt->rt_gateway, 1, 0UL); 1293 RT_LOCK(rt0); 1294 rt0->rt_gwroute = rt; 1295 if (rt == NULL) { 1296 RT_UNLOCK(rt0); 1297 senderr(EHOSTUNREACH); 1298 } 1299 } 1300 RT_UNLOCK(rt0); 1301 } 1302 /* XXX why are we inspecting rmx_expire? */ 1303 error = (rt->rt_flags & RTF_REJECT) && 1304 (rt->rt_rmx.rmx_expire == 0 || 1305 time_uptime < rt->rt_rmx.rmx_expire); 1306 if (error) { 1307 RT_UNLOCK(rt); 1308 senderr(rt == rt0 ? EHOSTDOWN : EHOSTUNREACH); 1309 } 1310 1311 *lrt = rt; 1312 *lrt0 = rt0; 1313 return (0); 1314 bad: 1315 /* NB: lrt and lrt0 should not be interpreted if error is non-zero */ 1316 return (error); 1317 #undef senderr 1318 } 1319 1320 /* This must be before ip6_init2(), which is now SI_ORDER_MIDDLE */ 1321 SYSINIT(route, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0); 1322