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