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 * Note: In this file a 'fib' is a "forwarding information base" * 34 * Which is the new name for an in kernel routing (next hop) table. * 35 ***********************************************************************/ 36 37 #include "opt_inet.h" 38 #include "opt_inet6.h" 39 #include "opt_route.h" 40 #include "opt_sctp.h" 41 #include "opt_mrouting.h" 42 #include "opt_mpath.h" 43 44 #include <sys/param.h> 45 #include <sys/systm.h> 46 #include <sys/syslog.h> 47 #include <sys/malloc.h> 48 #include <sys/mbuf.h> 49 #include <sys/socket.h> 50 #include <sys/sysctl.h> 51 #include <sys/syslog.h> 52 #include <sys/sysproto.h> 53 #include <sys/proc.h> 54 #include <sys/domain.h> 55 #include <sys/kernel.h> 56 57 #include <net/if.h> 58 #include <net/if_var.h> 59 #include <net/if_dl.h> 60 #include <net/route.h> 61 #include <net/vnet.h> 62 #include <net/flowtable.h> 63 64 #ifdef RADIX_MPATH 65 #include <net/radix_mpath.h> 66 #endif 67 68 #include <netinet/in.h> 69 #include <netinet/ip_mroute.h> 70 71 #include <vm/uma.h> 72 73 #define RT_MAXFIBS UINT16_MAX 74 75 /* Kernel config default option. */ 76 #ifdef ROUTETABLES 77 #if ROUTETABLES <= 0 78 #error "ROUTETABLES defined too low" 79 #endif 80 #if ROUTETABLES > RT_MAXFIBS 81 #error "ROUTETABLES defined too big" 82 #endif 83 #define RT_NUMFIBS ROUTETABLES 84 #endif /* ROUTETABLES */ 85 /* Initialize to default if not otherwise set. */ 86 #ifndef RT_NUMFIBS 87 #define RT_NUMFIBS 1 88 #endif 89 90 #if defined(INET) || defined(INET6) 91 #ifdef SCTP 92 extern void sctp_addr_change(struct ifaddr *ifa, int cmd); 93 #endif /* SCTP */ 94 #endif 95 96 97 /* This is read-only.. */ 98 u_int rt_numfibs = RT_NUMFIBS; 99 SYSCTL_UINT(_net, OID_AUTO, fibs, CTLFLAG_RDTUN, &rt_numfibs, 0, ""); 100 101 /* 102 * By default add routes to all fibs for new interfaces. 103 * Once this is set to 0 then only allocate routes on interface 104 * changes for the FIB of the caller when adding a new set of addresses 105 * to an interface. XXX this is a shotgun aproach to a problem that needs 106 * a more fine grained solution.. that will come. 107 * XXX also has the problems getting the FIB from curthread which will not 108 * always work given the fib can be overridden and prefixes can be added 109 * from the network stack context. 110 */ 111 VNET_DEFINE(u_int, rt_add_addr_allfibs) = 1; 112 SYSCTL_UINT(_net, OID_AUTO, add_addr_allfibs, CTLFLAG_RWTUN | CTLFLAG_VNET, 113 &VNET_NAME(rt_add_addr_allfibs), 0, ""); 114 115 VNET_DEFINE(struct rtstat, rtstat); 116 #define V_rtstat VNET(rtstat) 117 118 VNET_DEFINE(struct radix_node_head *, rt_tables); 119 #define V_rt_tables VNET(rt_tables) 120 121 VNET_DEFINE(int, rttrash); /* routes not in table but not freed */ 122 #define V_rttrash VNET(rttrash) 123 124 125 /* 126 * Convert a 'struct radix_node *' to a 'struct rtentry *'. 127 * The operation can be done safely (in this code) because a 128 * 'struct rtentry' starts with two 'struct radix_node''s, the first 129 * one representing leaf nodes in the routing tree, which is 130 * what the code in radix.c passes us as a 'struct radix_node'. 131 * 132 * But because there are a lot of assumptions in this conversion, 133 * do not cast explicitly, but always use the macro below. 134 */ 135 #define RNTORT(p) ((struct rtentry *)(p)) 136 137 static VNET_DEFINE(uma_zone_t, rtzone); /* Routing table UMA zone. */ 138 #define V_rtzone VNET(rtzone) 139 140 static int rtrequest1_fib_change(struct radix_node_head *, struct rt_addrinfo *, 141 struct rtentry **, u_int); 142 static void rt_setmetrics(const struct rt_addrinfo *, struct rtentry *); 143 144 struct if_mtuinfo 145 { 146 struct ifnet *ifp; 147 int mtu; 148 }; 149 150 static int if_updatemtu_cb(struct radix_node *, void *); 151 152 /* 153 * handler for net.my_fibnum 154 */ 155 static int 156 sysctl_my_fibnum(SYSCTL_HANDLER_ARGS) 157 { 158 int fibnum; 159 int error; 160 161 fibnum = curthread->td_proc->p_fibnum; 162 error = sysctl_handle_int(oidp, &fibnum, 0, req); 163 return (error); 164 } 165 166 SYSCTL_PROC(_net, OID_AUTO, my_fibnum, CTLTYPE_INT|CTLFLAG_RD, 167 NULL, 0, &sysctl_my_fibnum, "I", "default FIB of caller"); 168 169 static __inline struct radix_node_head ** 170 rt_tables_get_rnh_ptr(int table, int fam) 171 { 172 struct radix_node_head **rnh; 173 174 KASSERT(table >= 0 && table < rt_numfibs, ("%s: table out of bounds.", 175 __func__)); 176 KASSERT(fam >= 0 && fam < (AF_MAX+1), ("%s: fam out of bounds.", 177 __func__)); 178 179 /* rnh is [fib=0][af=0]. */ 180 rnh = (struct radix_node_head **)V_rt_tables; 181 /* Get the offset to the requested table and fam. */ 182 rnh += table * (AF_MAX+1) + fam; 183 184 return (rnh); 185 } 186 187 struct radix_node_head * 188 rt_tables_get_rnh(int table, int fam) 189 { 190 191 return (*rt_tables_get_rnh_ptr(table, fam)); 192 } 193 194 /* 195 * route initialization must occur before ip6_init2(), which happenas at 196 * SI_ORDER_MIDDLE. 197 */ 198 static void 199 route_init(void) 200 { 201 202 /* whack the tunable ints into line. */ 203 if (rt_numfibs > RT_MAXFIBS) 204 rt_numfibs = RT_MAXFIBS; 205 if (rt_numfibs == 0) 206 rt_numfibs = 1; 207 } 208 SYSINIT(route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0); 209 210 static int 211 rtentry_zinit(void *mem, int size, int how) 212 { 213 struct rtentry *rt = mem; 214 215 rt->rt_pksent = counter_u64_alloc(how); 216 if (rt->rt_pksent == NULL) 217 return (ENOMEM); 218 219 RT_LOCK_INIT(rt); 220 221 return (0); 222 } 223 224 static void 225 rtentry_zfini(void *mem, int size) 226 { 227 struct rtentry *rt = mem; 228 229 RT_LOCK_DESTROY(rt); 230 counter_u64_free(rt->rt_pksent); 231 } 232 233 static int 234 rtentry_ctor(void *mem, int size, void *arg, int how) 235 { 236 struct rtentry *rt = mem; 237 238 bzero(rt, offsetof(struct rtentry, rt_endzero)); 239 counter_u64_zero(rt->rt_pksent); 240 241 return (0); 242 } 243 244 static void 245 rtentry_dtor(void *mem, int size, void *arg) 246 { 247 struct rtentry *rt = mem; 248 249 RT_UNLOCK_COND(rt); 250 } 251 252 static void 253 vnet_route_init(const void *unused __unused) 254 { 255 struct domain *dom; 256 struct radix_node_head **rnh; 257 int table; 258 int fam; 259 260 V_rt_tables = malloc(rt_numfibs * (AF_MAX+1) * 261 sizeof(struct radix_node_head *), M_RTABLE, M_WAITOK|M_ZERO); 262 263 V_rtzone = uma_zcreate("rtentry", sizeof(struct rtentry), 264 rtentry_ctor, rtentry_dtor, 265 rtentry_zinit, rtentry_zfini, UMA_ALIGN_PTR, 0); 266 for (dom = domains; dom; dom = dom->dom_next) { 267 if (dom->dom_rtattach == NULL) 268 continue; 269 270 for (table = 0; table < rt_numfibs; table++) { 271 fam = dom->dom_family; 272 if (table != 0 && fam != AF_INET6 && fam != AF_INET) 273 break; 274 275 rnh = rt_tables_get_rnh_ptr(table, fam); 276 if (rnh == NULL) 277 panic("%s: rnh NULL", __func__); 278 dom->dom_rtattach((void **)rnh, 0); 279 } 280 } 281 } 282 VNET_SYSINIT(vnet_route_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FOURTH, 283 vnet_route_init, 0); 284 285 #ifdef VIMAGE 286 static void 287 vnet_route_uninit(const void *unused __unused) 288 { 289 int table; 290 int fam; 291 struct domain *dom; 292 struct radix_node_head **rnh; 293 294 for (dom = domains; dom; dom = dom->dom_next) { 295 if (dom->dom_rtdetach == NULL) 296 continue; 297 298 for (table = 0; table < rt_numfibs; table++) { 299 fam = dom->dom_family; 300 301 if (table != 0 && fam != AF_INET6 && fam != AF_INET) 302 break; 303 304 rnh = rt_tables_get_rnh_ptr(table, fam); 305 if (rnh == NULL) 306 panic("%s: rnh NULL", __func__); 307 dom->dom_rtdetach((void **)rnh, 0); 308 } 309 } 310 311 free(V_rt_tables, M_RTABLE); 312 uma_zdestroy(V_rtzone); 313 } 314 VNET_SYSUNINIT(vnet_route_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, 315 vnet_route_uninit, 0); 316 #endif 317 318 #ifndef _SYS_SYSPROTO_H_ 319 struct setfib_args { 320 int fibnum; 321 }; 322 #endif 323 int 324 sys_setfib(struct thread *td, struct setfib_args *uap) 325 { 326 if (uap->fibnum < 0 || uap->fibnum >= rt_numfibs) 327 return EINVAL; 328 td->td_proc->p_fibnum = uap->fibnum; 329 return (0); 330 } 331 332 /* 333 * Packet routing routines. 334 */ 335 void 336 rtalloc(struct route *ro) 337 { 338 339 rtalloc_ign_fib(ro, 0UL, RT_DEFAULT_FIB); 340 } 341 342 void 343 rtalloc_fib(struct route *ro, u_int fibnum) 344 { 345 rtalloc_ign_fib(ro, 0UL, fibnum); 346 } 347 348 void 349 rtalloc_ign(struct route *ro, u_long ignore) 350 { 351 struct rtentry *rt; 352 353 if ((rt = ro->ro_rt) != NULL) { 354 if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP) 355 return; 356 RTFREE(rt); 357 ro->ro_rt = NULL; 358 } 359 ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, RT_DEFAULT_FIB); 360 if (ro->ro_rt) 361 RT_UNLOCK(ro->ro_rt); 362 } 363 364 void 365 rtalloc_ign_fib(struct route *ro, u_long ignore, u_int fibnum) 366 { 367 struct rtentry *rt; 368 369 if ((rt = ro->ro_rt) != NULL) { 370 if (rt->rt_ifp != NULL && rt->rt_flags & RTF_UP) 371 return; 372 RTFREE(rt); 373 ro->ro_rt = NULL; 374 } 375 ro->ro_rt = rtalloc1_fib(&ro->ro_dst, 1, ignore, fibnum); 376 if (ro->ro_rt) 377 RT_UNLOCK(ro->ro_rt); 378 } 379 380 /* 381 * Look up the route that matches the address given 382 * Or, at least try.. Create a cloned route if needed. 383 * 384 * The returned route, if any, is locked. 385 */ 386 struct rtentry * 387 rtalloc1(struct sockaddr *dst, int report, u_long ignflags) 388 { 389 390 return (rtalloc1_fib(dst, report, ignflags, RT_DEFAULT_FIB)); 391 } 392 393 struct rtentry * 394 rtalloc1_fib(struct sockaddr *dst, int report, u_long ignflags, 395 u_int fibnum) 396 { 397 struct radix_node_head *rnh; 398 struct radix_node *rn; 399 struct rtentry *newrt; 400 struct rt_addrinfo info; 401 int err = 0, msgtype = RTM_MISS; 402 int needlock; 403 404 KASSERT((fibnum < rt_numfibs), ("rtalloc1_fib: bad fibnum")); 405 rnh = rt_tables_get_rnh(fibnum, dst->sa_family); 406 newrt = NULL; 407 if (rnh == NULL) 408 goto miss; 409 410 /* 411 * Look up the address in the table for that Address Family 412 */ 413 needlock = !(ignflags & RTF_RNH_LOCKED); 414 if (needlock) 415 RADIX_NODE_HEAD_RLOCK(rnh); 416 #ifdef INVARIANTS 417 else 418 RADIX_NODE_HEAD_LOCK_ASSERT(rnh); 419 #endif 420 rn = rnh->rnh_matchaddr(dst, rnh); 421 if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) { 422 newrt = RNTORT(rn); 423 RT_LOCK(newrt); 424 RT_ADDREF(newrt); 425 if (needlock) 426 RADIX_NODE_HEAD_RUNLOCK(rnh); 427 goto done; 428 429 } else if (needlock) 430 RADIX_NODE_HEAD_RUNLOCK(rnh); 431 432 /* 433 * Either we hit the root or couldn't find any match, 434 * Which basically means 435 * "caint get there frm here" 436 */ 437 miss: 438 V_rtstat.rts_unreach++; 439 440 if (report) { 441 /* 442 * If required, report the failure to the supervising 443 * Authorities. 444 * For a delete, this is not an error. (report == 0) 445 */ 446 bzero(&info, sizeof(info)); 447 info.rti_info[RTAX_DST] = dst; 448 rt_missmsg_fib(msgtype, &info, 0, err, fibnum); 449 } 450 done: 451 if (newrt) 452 RT_LOCK_ASSERT(newrt); 453 return (newrt); 454 } 455 456 /* 457 * Remove a reference count from an rtentry. 458 * If the count gets low enough, take it out of the routing table 459 */ 460 void 461 rtfree(struct rtentry *rt) 462 { 463 struct radix_node_head *rnh; 464 465 KASSERT(rt != NULL,("%s: NULL rt", __func__)); 466 rnh = rt_tables_get_rnh(rt->rt_fibnum, rt_key(rt)->sa_family); 467 KASSERT(rnh != NULL,("%s: NULL rnh", __func__)); 468 469 RT_LOCK_ASSERT(rt); 470 471 /* 472 * The callers should use RTFREE_LOCKED() or RTFREE(), so 473 * we should come here exactly with the last reference. 474 */ 475 RT_REMREF(rt); 476 if (rt->rt_refcnt > 0) { 477 log(LOG_DEBUG, "%s: %p has %d refs\n", __func__, rt, rt->rt_refcnt); 478 goto done; 479 } 480 481 /* 482 * On last reference give the "close method" a chance 483 * to cleanup private state. This also permits (for 484 * IPv4 and IPv6) a chance to decide if the routing table 485 * entry should be purged immediately or at a later time. 486 * When an immediate purge is to happen the close routine 487 * typically calls rtexpunge which clears the RTF_UP flag 488 * on the entry so that the code below reclaims the storage. 489 */ 490 if (rt->rt_refcnt == 0 && rnh->rnh_close) 491 rnh->rnh_close((struct radix_node *)rt, rnh); 492 493 /* 494 * If we are no longer "up" (and ref == 0) 495 * then we can free the resources associated 496 * with the route. 497 */ 498 if ((rt->rt_flags & RTF_UP) == 0) { 499 if (rt->rt_nodes->rn_flags & (RNF_ACTIVE | RNF_ROOT)) 500 panic("rtfree 2"); 501 /* 502 * the rtentry must have been removed from the routing table 503 * so it is represented in rttrash.. remove that now. 504 */ 505 V_rttrash--; 506 #ifdef DIAGNOSTIC 507 if (rt->rt_refcnt < 0) { 508 printf("rtfree: %p not freed (neg refs)\n", rt); 509 goto done; 510 } 511 #endif 512 /* 513 * release references on items we hold them on.. 514 * e.g other routes and ifaddrs. 515 */ 516 if (rt->rt_ifa) 517 ifa_free(rt->rt_ifa); 518 /* 519 * The key is separatly alloc'd so free it (see rt_setgate()). 520 * This also frees the gateway, as they are always malloc'd 521 * together. 522 */ 523 Free(rt_key(rt)); 524 525 /* 526 * and the rtentry itself of course 527 */ 528 uma_zfree(V_rtzone, rt); 529 return; 530 } 531 done: 532 RT_UNLOCK(rt); 533 } 534 535 536 /* 537 * Force a routing table entry to the specified 538 * destination to go through the given gateway. 539 * Normally called as a result of a routing redirect 540 * message from the network layer. 541 */ 542 void 543 rtredirect(struct sockaddr *dst, 544 struct sockaddr *gateway, 545 struct sockaddr *netmask, 546 int flags, 547 struct sockaddr *src) 548 { 549 550 rtredirect_fib(dst, gateway, netmask, flags, src, RT_DEFAULT_FIB); 551 } 552 553 void 554 rtredirect_fib(struct sockaddr *dst, 555 struct sockaddr *gateway, 556 struct sockaddr *netmask, 557 int flags, 558 struct sockaddr *src, 559 u_int fibnum) 560 { 561 struct rtentry *rt, *rt0 = NULL; 562 int error = 0; 563 short *stat = NULL; 564 struct rt_addrinfo info; 565 struct ifaddr *ifa; 566 struct radix_node_head *rnh; 567 568 ifa = NULL; 569 rnh = rt_tables_get_rnh(fibnum, dst->sa_family); 570 if (rnh == NULL) { 571 error = EAFNOSUPPORT; 572 goto out; 573 } 574 575 /* verify the gateway is directly reachable */ 576 if ((ifa = ifa_ifwithnet(gateway, 0, fibnum)) == NULL) { 577 error = ENETUNREACH; 578 goto out; 579 } 580 rt = rtalloc1_fib(dst, 0, 0UL, fibnum); /* NB: rt is locked */ 581 /* 582 * If the redirect isn't from our current router for this dst, 583 * it's either old or wrong. If it redirects us to ourselves, 584 * we have a routing loop, perhaps as a result of an interface 585 * going down recently. 586 */ 587 if (!(flags & RTF_DONE) && rt && 588 (!sa_equal(src, rt->rt_gateway) || rt->rt_ifa != ifa)) 589 error = EINVAL; 590 else if (ifa_ifwithaddr_check(gateway)) 591 error = EHOSTUNREACH; 592 if (error) 593 goto done; 594 /* 595 * Create a new entry if we just got back a wildcard entry 596 * or the lookup failed. This is necessary for hosts 597 * which use routing redirects generated by smart gateways 598 * to dynamically build the routing tables. 599 */ 600 if (rt == NULL || (rt_mask(rt) && rt_mask(rt)->sa_len < 2)) 601 goto create; 602 /* 603 * Don't listen to the redirect if it's 604 * for a route to an interface. 605 */ 606 if (rt->rt_flags & RTF_GATEWAY) { 607 if (((rt->rt_flags & RTF_HOST) == 0) && (flags & RTF_HOST)) { 608 /* 609 * Changing from route to net => route to host. 610 * Create new route, rather than smashing route to net. 611 */ 612 create: 613 rt0 = rt; 614 rt = NULL; 615 616 flags |= RTF_GATEWAY | RTF_DYNAMIC; 617 bzero((caddr_t)&info, sizeof(info)); 618 info.rti_info[RTAX_DST] = dst; 619 info.rti_info[RTAX_GATEWAY] = gateway; 620 info.rti_info[RTAX_NETMASK] = netmask; 621 info.rti_ifa = ifa; 622 info.rti_flags = flags; 623 if (rt0 != NULL) 624 RT_UNLOCK(rt0); /* drop lock to avoid LOR with RNH */ 625 error = rtrequest1_fib(RTM_ADD, &info, &rt, fibnum); 626 if (rt != NULL) { 627 RT_LOCK(rt); 628 if (rt0 != NULL) 629 EVENTHANDLER_INVOKE(route_redirect_event, rt0, rt, dst); 630 flags = rt->rt_flags; 631 } 632 if (rt0 != NULL) 633 RTFREE(rt0); 634 635 stat = &V_rtstat.rts_dynamic; 636 } else { 637 struct rtentry *gwrt; 638 639 /* 640 * Smash the current notion of the gateway to 641 * this destination. Should check about netmask!!! 642 */ 643 rt->rt_flags |= RTF_MODIFIED; 644 flags |= RTF_MODIFIED; 645 stat = &V_rtstat.rts_newgateway; 646 /* 647 * add the key and gateway (in one malloc'd chunk). 648 */ 649 RT_UNLOCK(rt); 650 RADIX_NODE_HEAD_LOCK(rnh); 651 RT_LOCK(rt); 652 rt_setgate(rt, rt_key(rt), gateway); 653 gwrt = rtalloc1(gateway, 1, RTF_RNH_LOCKED); 654 RADIX_NODE_HEAD_UNLOCK(rnh); 655 EVENTHANDLER_INVOKE(route_redirect_event, rt, gwrt, dst); 656 RTFREE_LOCKED(gwrt); 657 } 658 } else 659 error = EHOSTUNREACH; 660 done: 661 if (rt) 662 RTFREE_LOCKED(rt); 663 out: 664 if (error) 665 V_rtstat.rts_badredirect++; 666 else if (stat != NULL) 667 (*stat)++; 668 bzero((caddr_t)&info, sizeof(info)); 669 info.rti_info[RTAX_DST] = dst; 670 info.rti_info[RTAX_GATEWAY] = gateway; 671 info.rti_info[RTAX_NETMASK] = netmask; 672 info.rti_info[RTAX_AUTHOR] = src; 673 rt_missmsg_fib(RTM_REDIRECT, &info, flags, error, fibnum); 674 if (ifa != NULL) 675 ifa_free(ifa); 676 } 677 678 int 679 rtioctl(u_long req, caddr_t data) 680 { 681 682 return (rtioctl_fib(req, data, RT_DEFAULT_FIB)); 683 } 684 685 /* 686 * Routing table ioctl interface. 687 */ 688 int 689 rtioctl_fib(u_long req, caddr_t data, u_int fibnum) 690 { 691 692 /* 693 * If more ioctl commands are added here, make sure the proper 694 * super-user checks are being performed because it is possible for 695 * prison-root to make it this far if raw sockets have been enabled 696 * in jails. 697 */ 698 #ifdef INET 699 /* Multicast goop, grrr... */ 700 return mrt_ioctl ? mrt_ioctl(req, data, fibnum) : EOPNOTSUPP; 701 #else /* INET */ 702 return ENXIO; 703 #endif /* INET */ 704 } 705 706 struct ifaddr * 707 ifa_ifwithroute(int flags, struct sockaddr *dst, struct sockaddr *gateway, 708 u_int fibnum) 709 { 710 struct ifaddr *ifa; 711 int not_found = 0; 712 713 if ((flags & RTF_GATEWAY) == 0) { 714 /* 715 * If we are adding a route to an interface, 716 * and the interface is a pt to pt link 717 * we should search for the destination 718 * as our clue to the interface. Otherwise 719 * we can use the local address. 720 */ 721 ifa = NULL; 722 if (flags & RTF_HOST) 723 ifa = ifa_ifwithdstaddr(dst, fibnum); 724 if (ifa == NULL) 725 ifa = ifa_ifwithaddr(gateway); 726 } else { 727 /* 728 * If we are adding a route to a remote net 729 * or host, the gateway may still be on the 730 * other end of a pt to pt link. 731 */ 732 ifa = ifa_ifwithdstaddr(gateway, fibnum); 733 } 734 if (ifa == NULL) 735 ifa = ifa_ifwithnet(gateway, 0, fibnum); 736 if (ifa == NULL) { 737 struct rtentry *rt = rtalloc1_fib(gateway, 0, RTF_RNH_LOCKED, fibnum); 738 if (rt == NULL) 739 return (NULL); 740 /* 741 * dismiss a gateway that is reachable only 742 * through the default router 743 */ 744 switch (gateway->sa_family) { 745 case AF_INET: 746 if (satosin(rt_key(rt))->sin_addr.s_addr == INADDR_ANY) 747 not_found = 1; 748 break; 749 case AF_INET6: 750 if (IN6_IS_ADDR_UNSPECIFIED(&satosin6(rt_key(rt))->sin6_addr)) 751 not_found = 1; 752 break; 753 default: 754 break; 755 } 756 if (!not_found && rt->rt_ifa != NULL) { 757 ifa = rt->rt_ifa; 758 ifa_ref(ifa); 759 } 760 RT_REMREF(rt); 761 RT_UNLOCK(rt); 762 if (not_found || ifa == NULL) 763 return (NULL); 764 } 765 if (ifa->ifa_addr->sa_family != dst->sa_family) { 766 struct ifaddr *oifa = ifa; 767 ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp); 768 if (ifa == NULL) 769 ifa = oifa; 770 else 771 ifa_free(oifa); 772 } 773 return (ifa); 774 } 775 776 /* 777 * Do appropriate manipulations of a routing tree given 778 * all the bits of info needed 779 */ 780 int 781 rtrequest(int req, 782 struct sockaddr *dst, 783 struct sockaddr *gateway, 784 struct sockaddr *netmask, 785 int flags, 786 struct rtentry **ret_nrt) 787 { 788 789 return (rtrequest_fib(req, dst, gateway, netmask, flags, ret_nrt, 790 RT_DEFAULT_FIB)); 791 } 792 793 int 794 rtrequest_fib(int req, 795 struct sockaddr *dst, 796 struct sockaddr *gateway, 797 struct sockaddr *netmask, 798 int flags, 799 struct rtentry **ret_nrt, 800 u_int fibnum) 801 { 802 struct rt_addrinfo info; 803 804 if (dst->sa_len == 0) 805 return(EINVAL); 806 807 bzero((caddr_t)&info, sizeof(info)); 808 info.rti_flags = flags; 809 info.rti_info[RTAX_DST] = dst; 810 info.rti_info[RTAX_GATEWAY] = gateway; 811 info.rti_info[RTAX_NETMASK] = netmask; 812 return rtrequest1_fib(req, &info, ret_nrt, fibnum); 813 } 814 815 /* 816 * These (questionable) definitions of apparent local variables apply 817 * to the next two functions. XXXXXX!!! 818 */ 819 #define dst info->rti_info[RTAX_DST] 820 #define gateway info->rti_info[RTAX_GATEWAY] 821 #define netmask info->rti_info[RTAX_NETMASK] 822 #define ifaaddr info->rti_info[RTAX_IFA] 823 #define ifpaddr info->rti_info[RTAX_IFP] 824 #define flags info->rti_flags 825 826 int 827 rt_getifa(struct rt_addrinfo *info) 828 { 829 830 return (rt_getifa_fib(info, RT_DEFAULT_FIB)); 831 } 832 833 /* 834 * Look up rt_addrinfo for a specific fib. Note that if rti_ifa is defined, 835 * it will be referenced so the caller must free it. 836 */ 837 int 838 rt_getifa_fib(struct rt_addrinfo *info, u_int fibnum) 839 { 840 struct ifaddr *ifa; 841 int error = 0; 842 843 /* 844 * ifp may be specified by sockaddr_dl 845 * when protocol address is ambiguous. 846 */ 847 if (info->rti_ifp == NULL && ifpaddr != NULL && 848 ifpaddr->sa_family == AF_LINK && 849 (ifa = ifa_ifwithnet(ifpaddr, 0, fibnum)) != NULL) { 850 info->rti_ifp = ifa->ifa_ifp; 851 ifa_free(ifa); 852 } 853 if (info->rti_ifa == NULL && ifaaddr != NULL) 854 info->rti_ifa = ifa_ifwithaddr(ifaaddr); 855 if (info->rti_ifa == NULL) { 856 struct sockaddr *sa; 857 858 sa = ifaaddr != NULL ? ifaaddr : 859 (gateway != NULL ? gateway : dst); 860 if (sa != NULL && info->rti_ifp != NULL) 861 info->rti_ifa = ifaof_ifpforaddr(sa, info->rti_ifp); 862 else if (dst != NULL && gateway != NULL) 863 info->rti_ifa = ifa_ifwithroute(flags, dst, gateway, 864 fibnum); 865 else if (sa != NULL) 866 info->rti_ifa = ifa_ifwithroute(flags, sa, sa, 867 fibnum); 868 } 869 if ((ifa = info->rti_ifa) != NULL) { 870 if (info->rti_ifp == NULL) 871 info->rti_ifp = ifa->ifa_ifp; 872 } else 873 error = ENETUNREACH; 874 return (error); 875 } 876 877 /* 878 * Expunges references to a route that's about to be reclaimed. 879 * The route must be locked. 880 */ 881 int 882 rt_expunge(struct radix_node_head *rnh, struct rtentry *rt) 883 { 884 #if !defined(RADIX_MPATH) 885 struct radix_node *rn; 886 #else 887 struct rt_addrinfo info; 888 int fib; 889 struct rtentry *rt0; 890 #endif 891 struct ifaddr *ifa; 892 int error = 0; 893 894 RT_LOCK_ASSERT(rt); 895 RADIX_NODE_HEAD_LOCK_ASSERT(rnh); 896 897 #ifdef RADIX_MPATH 898 fib = rt->rt_fibnum; 899 bzero(&info, sizeof(info)); 900 info.rti_ifp = rt->rt_ifp; 901 info.rti_flags = RTF_RNH_LOCKED; 902 info.rti_info[RTAX_DST] = rt_key(rt); 903 info.rti_info[RTAX_GATEWAY] = rt->rt_ifa->ifa_addr; 904 905 RT_UNLOCK(rt); 906 error = rtrequest1_fib(RTM_DELETE, &info, &rt0, fib); 907 908 if (error == 0 && rt0 != NULL) { 909 rt = rt0; 910 RT_LOCK(rt); 911 } else if (error != 0) { 912 RT_LOCK(rt); 913 return (error); 914 } 915 #else 916 /* 917 * Remove the item from the tree; it should be there, 918 * but when callers invoke us blindly it may not (sigh). 919 */ 920 rn = rnh->rnh_deladdr(rt_key(rt), rt_mask(rt), rnh); 921 if (rn == NULL) { 922 error = ESRCH; 923 goto bad; 924 } 925 KASSERT((rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)) == 0, 926 ("unexpected flags 0x%x", rn->rn_flags)); 927 KASSERT(rt == RNTORT(rn), 928 ("lookup mismatch, rt %p rn %p", rt, rn)); 929 #endif /* RADIX_MPATH */ 930 931 rt->rt_flags &= ~RTF_UP; 932 933 /* 934 * Give the protocol a chance to keep things in sync. 935 */ 936 if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest) { 937 struct rt_addrinfo info; 938 939 bzero((caddr_t)&info, sizeof(info)); 940 info.rti_flags = rt->rt_flags; 941 info.rti_info[RTAX_DST] = rt_key(rt); 942 info.rti_info[RTAX_GATEWAY] = rt->rt_gateway; 943 info.rti_info[RTAX_NETMASK] = rt_mask(rt); 944 ifa->ifa_rtrequest(RTM_DELETE, rt, &info); 945 } 946 947 /* 948 * one more rtentry floating around that is not 949 * linked to the routing table. 950 */ 951 V_rttrash++; 952 #if !defined(RADIX_MPATH) 953 bad: 954 #endif 955 return (error); 956 } 957 958 static int 959 if_updatemtu_cb(struct radix_node *rn, void *arg) 960 { 961 struct rtentry *rt; 962 struct if_mtuinfo *ifmtu; 963 964 rt = (struct rtentry *)rn; 965 ifmtu = (struct if_mtuinfo *)arg; 966 967 if (rt->rt_ifp != ifmtu->ifp) 968 return (0); 969 970 if (rt->rt_mtu >= ifmtu->mtu) { 971 /* We have to decrease mtu regardless of flags */ 972 rt->rt_mtu = ifmtu->mtu; 973 return (0); 974 } 975 976 /* 977 * New MTU is bigger. Check if are allowed to alter it 978 */ 979 if ((rt->rt_flags & (RTF_FIXEDMTU | RTF_GATEWAY | RTF_HOST)) != 0) { 980 981 /* 982 * Skip routes with user-supplied MTU and 983 * non-interface routes 984 */ 985 return (0); 986 } 987 988 /* We are safe to update route MTU */ 989 rt->rt_mtu = ifmtu->mtu; 990 991 return (0); 992 } 993 994 void 995 rt_updatemtu(struct ifnet *ifp) 996 { 997 struct if_mtuinfo ifmtu; 998 struct radix_node_head *rnh; 999 int i, j; 1000 1001 ifmtu.ifp = ifp; 1002 1003 /* 1004 * Try to update rt_mtu for all routes using this interface 1005 * Unfortunately the only way to do this is to traverse all 1006 * routing tables in all fibs/domains. 1007 */ 1008 for (i = 1; i <= AF_MAX; i++) { 1009 ifmtu.mtu = if_getmtu_family(ifp, i); 1010 for (j = 0; j < rt_numfibs; j++) { 1011 rnh = rt_tables_get_rnh(j, i); 1012 if (rnh == NULL) 1013 continue; 1014 RADIX_NODE_HEAD_LOCK(rnh); 1015 rnh->rnh_walktree(rnh, if_updatemtu_cb, &ifmtu); 1016 RADIX_NODE_HEAD_UNLOCK(rnh); 1017 } 1018 } 1019 } 1020 1021 1022 #if 0 1023 int p_sockaddr(char *buf, int buflen, struct sockaddr *s); 1024 int rt_print(char *buf, int buflen, struct rtentry *rt); 1025 1026 int 1027 p_sockaddr(char *buf, int buflen, struct sockaddr *s) 1028 { 1029 void *paddr = NULL; 1030 1031 switch (s->sa_family) { 1032 case AF_INET: 1033 paddr = &((struct sockaddr_in *)s)->sin_addr; 1034 break; 1035 case AF_INET6: 1036 paddr = &((struct sockaddr_in6 *)s)->sin6_addr; 1037 break; 1038 } 1039 1040 if (paddr == NULL) 1041 return (0); 1042 1043 if (inet_ntop(s->sa_family, paddr, buf, buflen) == NULL) 1044 return (0); 1045 1046 return (strlen(buf)); 1047 } 1048 1049 int 1050 rt_print(char *buf, int buflen, struct rtentry *rt) 1051 { 1052 struct sockaddr *addr, *mask; 1053 int i = 0; 1054 1055 addr = rt_key(rt); 1056 mask = rt_mask(rt); 1057 1058 i = p_sockaddr(buf, buflen, addr); 1059 if (!(rt->rt_flags & RTF_HOST)) { 1060 buf[i++] = '/'; 1061 i += p_sockaddr(buf + i, buflen - i, mask); 1062 } 1063 1064 if (rt->rt_flags & RTF_GATEWAY) { 1065 buf[i++] = '>'; 1066 i += p_sockaddr(buf + i, buflen - i, rt->rt_gateway); 1067 } 1068 1069 return (i); 1070 } 1071 #endif 1072 1073 #ifdef RADIX_MPATH 1074 static int 1075 rn_mpath_update(int req, struct rt_addrinfo *info, 1076 struct radix_node_head *rnh, struct rtentry **ret_nrt) 1077 { 1078 /* 1079 * if we got multipath routes, we require users to specify 1080 * a matching RTAX_GATEWAY. 1081 */ 1082 struct rtentry *rt, *rto = NULL; 1083 struct radix_node *rn; 1084 int error = 0; 1085 1086 rn = rnh->rnh_lookup(dst, netmask, rnh); 1087 if (rn == NULL) 1088 return (ESRCH); 1089 rto = rt = RNTORT(rn); 1090 1091 rt = rt_mpath_matchgate(rt, gateway); 1092 if (rt == NULL) 1093 return (ESRCH); 1094 /* 1095 * this is the first entry in the chain 1096 */ 1097 if (rto == rt) { 1098 rn = rn_mpath_next((struct radix_node *)rt); 1099 /* 1100 * there is another entry, now it's active 1101 */ 1102 if (rn) { 1103 rto = RNTORT(rn); 1104 RT_LOCK(rto); 1105 rto->rt_flags |= RTF_UP; 1106 RT_UNLOCK(rto); 1107 } else if (rt->rt_flags & RTF_GATEWAY) { 1108 /* 1109 * For gateway routes, we need to 1110 * make sure that we we are deleting 1111 * the correct gateway. 1112 * rt_mpath_matchgate() does not 1113 * check the case when there is only 1114 * one route in the chain. 1115 */ 1116 if (gateway && 1117 (rt->rt_gateway->sa_len != gateway->sa_len || 1118 memcmp(rt->rt_gateway, gateway, gateway->sa_len))) 1119 error = ESRCH; 1120 else { 1121 /* 1122 * remove from tree before returning it 1123 * to the caller 1124 */ 1125 rn = rnh->rnh_deladdr(dst, netmask, rnh); 1126 KASSERT(rt == RNTORT(rn), ("radix node disappeared")); 1127 goto gwdelete; 1128 } 1129 1130 } 1131 /* 1132 * use the normal delete code to remove 1133 * the first entry 1134 */ 1135 if (req != RTM_DELETE) 1136 goto nondelete; 1137 1138 error = ENOENT; 1139 goto done; 1140 } 1141 1142 /* 1143 * if the entry is 2nd and on up 1144 */ 1145 if ((req == RTM_DELETE) && !rt_mpath_deldup(rto, rt)) 1146 panic ("rtrequest1: rt_mpath_deldup"); 1147 gwdelete: 1148 RT_LOCK(rt); 1149 RT_ADDREF(rt); 1150 if (req == RTM_DELETE) { 1151 rt->rt_flags &= ~RTF_UP; 1152 /* 1153 * One more rtentry floating around that is not 1154 * linked to the routing table. rttrash will be decremented 1155 * when RTFREE(rt) is eventually called. 1156 */ 1157 V_rttrash++; 1158 } 1159 1160 nondelete: 1161 if (req != RTM_DELETE) 1162 panic("unrecognized request %d", req); 1163 1164 1165 /* 1166 * If the caller wants it, then it can have it, 1167 * but it's up to it to free the rtentry as we won't be 1168 * doing it. 1169 */ 1170 if (ret_nrt) { 1171 *ret_nrt = rt; 1172 RT_UNLOCK(rt); 1173 } else 1174 RTFREE_LOCKED(rt); 1175 done: 1176 return (error); 1177 } 1178 #endif 1179 1180 int 1181 rtrequest1_fib(int req, struct rt_addrinfo *info, struct rtentry **ret_nrt, 1182 u_int fibnum) 1183 { 1184 int error = 0, needlock = 0; 1185 struct rtentry *rt; 1186 #ifdef FLOWTABLE 1187 struct rtentry *rt0; 1188 #endif 1189 struct radix_node *rn; 1190 struct radix_node_head *rnh; 1191 struct ifaddr *ifa; 1192 struct sockaddr *ndst; 1193 struct sockaddr_storage mdst; 1194 #define senderr(x) { error = x ; goto bad; } 1195 1196 KASSERT((fibnum < rt_numfibs), ("rtrequest1_fib: bad fibnum")); 1197 switch (dst->sa_family) { 1198 case AF_INET6: 1199 case AF_INET: 1200 /* We support multiple FIBs. */ 1201 break; 1202 default: 1203 fibnum = RT_DEFAULT_FIB; 1204 break; 1205 } 1206 1207 /* 1208 * Find the correct routing tree to use for this Address Family 1209 */ 1210 rnh = rt_tables_get_rnh(fibnum, dst->sa_family); 1211 if (rnh == NULL) 1212 return (EAFNOSUPPORT); 1213 needlock = ((flags & RTF_RNH_LOCKED) == 0); 1214 flags &= ~RTF_RNH_LOCKED; 1215 if (needlock) 1216 RADIX_NODE_HEAD_LOCK(rnh); 1217 else 1218 RADIX_NODE_HEAD_LOCK_ASSERT(rnh); 1219 /* 1220 * If we are adding a host route then we don't want to put 1221 * a netmask in the tree, nor do we want to clone it. 1222 */ 1223 if (flags & RTF_HOST) 1224 netmask = NULL; 1225 1226 switch (req) { 1227 case RTM_DELETE: 1228 if (netmask) { 1229 rt_maskedcopy(dst, (struct sockaddr *)&mdst, netmask); 1230 dst = (struct sockaddr *)&mdst; 1231 } 1232 #ifdef RADIX_MPATH 1233 if (rn_mpath_capable(rnh)) { 1234 error = rn_mpath_update(req, info, rnh, ret_nrt); 1235 /* 1236 * "bad" holds true for the success case 1237 * as well 1238 */ 1239 if (error != ENOENT) 1240 goto bad; 1241 error = 0; 1242 } 1243 #endif 1244 if ((flags & RTF_PINNED) == 0) { 1245 /* Check if target route can be deleted */ 1246 rt = (struct rtentry *)rnh->rnh_lookup(dst, 1247 netmask, rnh); 1248 if ((rt != NULL) && (rt->rt_flags & RTF_PINNED)) 1249 senderr(EADDRINUSE); 1250 } 1251 1252 /* 1253 * Remove the item from the tree and return it. 1254 * Complain if it is not there and do no more processing. 1255 */ 1256 rn = rnh->rnh_deladdr(dst, netmask, rnh); 1257 if (rn == NULL) 1258 senderr(ESRCH); 1259 if (rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)) 1260 panic ("rtrequest delete"); 1261 rt = RNTORT(rn); 1262 RT_LOCK(rt); 1263 RT_ADDREF(rt); 1264 rt->rt_flags &= ~RTF_UP; 1265 1266 /* 1267 * give the protocol a chance to keep things in sync. 1268 */ 1269 if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest) 1270 ifa->ifa_rtrequest(RTM_DELETE, rt, info); 1271 1272 /* 1273 * One more rtentry floating around that is not 1274 * linked to the routing table. rttrash will be decremented 1275 * when RTFREE(rt) is eventually called. 1276 */ 1277 V_rttrash++; 1278 1279 /* 1280 * If the caller wants it, then it can have it, 1281 * but it's up to it to free the rtentry as we won't be 1282 * doing it. 1283 */ 1284 if (ret_nrt) { 1285 *ret_nrt = rt; 1286 RT_UNLOCK(rt); 1287 } else 1288 RTFREE_LOCKED(rt); 1289 break; 1290 case RTM_RESOLVE: 1291 /* 1292 * resolve was only used for route cloning 1293 * here for compat 1294 */ 1295 break; 1296 case RTM_ADD: 1297 if ((flags & RTF_GATEWAY) && !gateway) 1298 senderr(EINVAL); 1299 if (dst && gateway && (dst->sa_family != gateway->sa_family) && 1300 (gateway->sa_family != AF_UNSPEC) && (gateway->sa_family != AF_LINK)) 1301 senderr(EINVAL); 1302 1303 if (info->rti_ifa == NULL) { 1304 error = rt_getifa_fib(info, fibnum); 1305 if (error) 1306 senderr(error); 1307 } else 1308 ifa_ref(info->rti_ifa); 1309 ifa = info->rti_ifa; 1310 rt = uma_zalloc(V_rtzone, M_NOWAIT); 1311 if (rt == NULL) { 1312 ifa_free(ifa); 1313 senderr(ENOBUFS); 1314 } 1315 rt->rt_flags = RTF_UP | flags; 1316 rt->rt_fibnum = fibnum; 1317 /* 1318 * Add the gateway. Possibly re-malloc-ing the storage for it. 1319 */ 1320 RT_LOCK(rt); 1321 if ((error = rt_setgate(rt, dst, gateway)) != 0) { 1322 ifa_free(ifa); 1323 uma_zfree(V_rtzone, rt); 1324 senderr(error); 1325 } 1326 1327 /* 1328 * point to the (possibly newly malloc'd) dest address. 1329 */ 1330 ndst = (struct sockaddr *)rt_key(rt); 1331 1332 /* 1333 * make sure it contains the value we want (masked if needed). 1334 */ 1335 if (netmask) { 1336 rt_maskedcopy(dst, ndst, netmask); 1337 } else 1338 bcopy(dst, ndst, dst->sa_len); 1339 1340 /* 1341 * We use the ifa reference returned by rt_getifa_fib(). 1342 * This moved from below so that rnh->rnh_addaddr() can 1343 * examine the ifa and ifa->ifa_ifp if it so desires. 1344 */ 1345 rt->rt_ifa = ifa; 1346 rt->rt_ifp = ifa->ifa_ifp; 1347 rt->rt_weight = 1; 1348 1349 rt_setmetrics(info, rt); 1350 1351 #ifdef RADIX_MPATH 1352 /* do not permit exactly the same dst/mask/gw pair */ 1353 if (rn_mpath_capable(rnh) && 1354 rt_mpath_conflict(rnh, rt, netmask)) { 1355 ifa_free(rt->rt_ifa); 1356 Free(rt_key(rt)); 1357 uma_zfree(V_rtzone, rt); 1358 senderr(EEXIST); 1359 } 1360 #endif 1361 1362 #ifdef FLOWTABLE 1363 rt0 = NULL; 1364 /* "flow-table" only supports IPv6 and IPv4 at the moment. */ 1365 switch (dst->sa_family) { 1366 #ifdef INET6 1367 case AF_INET6: 1368 #endif 1369 #ifdef INET 1370 case AF_INET: 1371 #endif 1372 #if defined(INET6) || defined(INET) 1373 rn = rnh->rnh_matchaddr(dst, rnh); 1374 if (rn && ((rn->rn_flags & RNF_ROOT) == 0)) { 1375 struct sockaddr *mask; 1376 u_char *m, *n; 1377 int len; 1378 1379 /* 1380 * compare mask to see if the new route is 1381 * more specific than the existing one 1382 */ 1383 rt0 = RNTORT(rn); 1384 RT_LOCK(rt0); 1385 RT_ADDREF(rt0); 1386 RT_UNLOCK(rt0); 1387 /* 1388 * A host route is already present, so 1389 * leave the flow-table entries as is. 1390 */ 1391 if (rt0->rt_flags & RTF_HOST) { 1392 RTFREE(rt0); 1393 rt0 = NULL; 1394 } else if (!(flags & RTF_HOST) && netmask) { 1395 mask = rt_mask(rt0); 1396 len = mask->sa_len; 1397 m = (u_char *)mask; 1398 n = (u_char *)netmask; 1399 while (len-- > 0) { 1400 if (*n != *m) 1401 break; 1402 n++; 1403 m++; 1404 } 1405 if (len == 0 || (*n < *m)) { 1406 RTFREE(rt0); 1407 rt0 = NULL; 1408 } 1409 } 1410 } 1411 #endif/* INET6 || INET */ 1412 } 1413 #endif /* FLOWTABLE */ 1414 1415 /* XXX mtu manipulation will be done in rnh_addaddr -- itojun */ 1416 rn = rnh->rnh_addaddr(ndst, netmask, rnh, rt->rt_nodes); 1417 /* 1418 * If it still failed to go into the tree, 1419 * then un-make it (this should be a function) 1420 */ 1421 if (rn == NULL) { 1422 ifa_free(rt->rt_ifa); 1423 Free(rt_key(rt)); 1424 uma_zfree(V_rtzone, rt); 1425 #ifdef FLOWTABLE 1426 if (rt0 != NULL) 1427 RTFREE(rt0); 1428 #endif 1429 senderr(EEXIST); 1430 } 1431 #ifdef FLOWTABLE 1432 else if (rt0 != NULL) { 1433 flowtable_route_flush(dst->sa_family, rt0); 1434 RTFREE(rt0); 1435 } 1436 #endif 1437 1438 /* 1439 * If this protocol has something to add to this then 1440 * allow it to do that as well. 1441 */ 1442 if (ifa->ifa_rtrequest) 1443 ifa->ifa_rtrequest(req, rt, info); 1444 1445 /* 1446 * actually return a resultant rtentry and 1447 * give the caller a single reference. 1448 */ 1449 if (ret_nrt) { 1450 *ret_nrt = rt; 1451 RT_ADDREF(rt); 1452 } 1453 RT_UNLOCK(rt); 1454 break; 1455 case RTM_CHANGE: 1456 error = rtrequest1_fib_change(rnh, info, ret_nrt, fibnum); 1457 break; 1458 default: 1459 error = EOPNOTSUPP; 1460 } 1461 bad: 1462 if (needlock) 1463 RADIX_NODE_HEAD_UNLOCK(rnh); 1464 return (error); 1465 #undef senderr 1466 } 1467 1468 #undef dst 1469 #undef gateway 1470 #undef netmask 1471 #undef ifaaddr 1472 #undef ifpaddr 1473 #undef flags 1474 1475 static int 1476 rtrequest1_fib_change(struct radix_node_head *rnh, struct rt_addrinfo *info, 1477 struct rtentry **ret_nrt, u_int fibnum) 1478 { 1479 struct rtentry *rt = NULL; 1480 int error = 0; 1481 int free_ifa = 0; 1482 int family, mtu; 1483 struct if_mtuinfo ifmtu; 1484 1485 rt = (struct rtentry *)rnh->rnh_lookup(info->rti_info[RTAX_DST], 1486 info->rti_info[RTAX_NETMASK], rnh); 1487 1488 if (rt == NULL) 1489 return (ESRCH); 1490 1491 #ifdef RADIX_MPATH 1492 /* 1493 * If we got multipath routes, 1494 * we require users to specify a matching RTAX_GATEWAY. 1495 */ 1496 if (rn_mpath_capable(rnh)) { 1497 rt = rt_mpath_matchgate(rt, info->rti_info[RTAX_GATEWAY]); 1498 if (rt == NULL) 1499 return (ESRCH); 1500 } 1501 #endif 1502 1503 RT_LOCK(rt); 1504 1505 rt_setmetrics(info, rt); 1506 1507 /* 1508 * New gateway could require new ifaddr, ifp; 1509 * flags may also be different; ifp may be specified 1510 * by ll sockaddr when protocol address is ambiguous 1511 */ 1512 if (((rt->rt_flags & RTF_GATEWAY) && 1513 info->rti_info[RTAX_GATEWAY] != NULL) || 1514 info->rti_info[RTAX_IFP] != NULL || 1515 (info->rti_info[RTAX_IFA] != NULL && 1516 !sa_equal(info->rti_info[RTAX_IFA], rt->rt_ifa->ifa_addr))) { 1517 1518 error = rt_getifa_fib(info, fibnum); 1519 if (info->rti_ifa != NULL) 1520 free_ifa = 1; 1521 1522 if (error != 0) 1523 goto bad; 1524 } 1525 1526 /* Check if outgoing interface has changed */ 1527 if (info->rti_ifa != NULL && info->rti_ifa != rt->rt_ifa && 1528 rt->rt_ifa != NULL && rt->rt_ifa->ifa_rtrequest != NULL) { 1529 rt->rt_ifa->ifa_rtrequest(RTM_DELETE, rt, info); 1530 ifa_free(rt->rt_ifa); 1531 } 1532 /* Update gateway address */ 1533 if (info->rti_info[RTAX_GATEWAY] != NULL) { 1534 error = rt_setgate(rt, rt_key(rt), info->rti_info[RTAX_GATEWAY]); 1535 if (error != 0) 1536 goto bad; 1537 1538 rt->rt_flags &= ~RTF_GATEWAY; 1539 rt->rt_flags |= (RTF_GATEWAY & info->rti_flags); 1540 } 1541 1542 if (info->rti_ifa != NULL && info->rti_ifa != rt->rt_ifa) { 1543 ifa_ref(info->rti_ifa); 1544 rt->rt_ifa = info->rti_ifa; 1545 rt->rt_ifp = info->rti_ifp; 1546 } 1547 /* Allow some flags to be toggled on change. */ 1548 rt->rt_flags &= ~RTF_FMASK; 1549 rt->rt_flags |= info->rti_flags & RTF_FMASK; 1550 1551 if (rt->rt_ifa && rt->rt_ifa->ifa_rtrequest != NULL) 1552 rt->rt_ifa->ifa_rtrequest(RTM_ADD, rt, info); 1553 1554 /* Alter route MTU if necessary */ 1555 if (rt->rt_ifp != NULL) { 1556 family = info->rti_info[RTAX_DST]->sa_family; 1557 mtu = if_getmtu_family(rt->rt_ifp, family); 1558 /* Set default MTU */ 1559 if (rt->rt_mtu == 0) 1560 rt->rt_mtu = mtu; 1561 if (rt->rt_mtu != mtu) { 1562 /* Check if we really need to update */ 1563 ifmtu.ifp = rt->rt_ifp; 1564 ifmtu.mtu = mtu; 1565 if_updatemtu_cb(rt->rt_nodes, &ifmtu); 1566 } 1567 } 1568 1569 if (ret_nrt) { 1570 *ret_nrt = rt; 1571 RT_ADDREF(rt); 1572 } 1573 bad: 1574 RT_UNLOCK(rt); 1575 if (free_ifa != 0) 1576 ifa_free(info->rti_ifa); 1577 return (error); 1578 } 1579 1580 static void 1581 rt_setmetrics(const struct rt_addrinfo *info, struct rtentry *rt) 1582 { 1583 1584 if (info->rti_mflags & RTV_MTU) { 1585 if (info->rti_rmx->rmx_mtu != 0) { 1586 1587 /* 1588 * MTU was explicitly provided by user. 1589 * Keep it. 1590 */ 1591 rt->rt_flags |= RTF_FIXEDMTU; 1592 } else { 1593 1594 /* 1595 * User explicitly sets MTU to 0. 1596 * Assume rollback to default. 1597 */ 1598 rt->rt_flags &= ~RTF_FIXEDMTU; 1599 } 1600 rt->rt_mtu = info->rti_rmx->rmx_mtu; 1601 } 1602 if (info->rti_mflags & RTV_WEIGHT) 1603 rt->rt_weight = info->rti_rmx->rmx_weight; 1604 /* Kernel -> userland timebase conversion. */ 1605 if (info->rti_mflags & RTV_EXPIRE) 1606 rt->rt_expire = info->rti_rmx->rmx_expire ? 1607 info->rti_rmx->rmx_expire - time_second + time_uptime : 0; 1608 } 1609 1610 int 1611 rt_setgate(struct rtentry *rt, struct sockaddr *dst, struct sockaddr *gate) 1612 { 1613 /* XXX dst may be overwritten, can we move this to below */ 1614 int dlen = SA_SIZE(dst), glen = SA_SIZE(gate); 1615 #ifdef INVARIANTS 1616 struct radix_node_head *rnh; 1617 1618 rnh = rt_tables_get_rnh(rt->rt_fibnum, dst->sa_family); 1619 #endif 1620 1621 RT_LOCK_ASSERT(rt); 1622 RADIX_NODE_HEAD_LOCK_ASSERT(rnh); 1623 1624 /* 1625 * Prepare to store the gateway in rt->rt_gateway. 1626 * Both dst and gateway are stored one after the other in the same 1627 * malloc'd chunk. If we have room, we can reuse the old buffer, 1628 * rt_gateway already points to the right place. 1629 * Otherwise, malloc a new block and update the 'dst' address. 1630 */ 1631 if (rt->rt_gateway == NULL || glen > SA_SIZE(rt->rt_gateway)) { 1632 caddr_t new; 1633 1634 R_Malloc(new, caddr_t, dlen + glen); 1635 if (new == NULL) 1636 return ENOBUFS; 1637 /* 1638 * XXX note, we copy from *dst and not *rt_key(rt) because 1639 * rt_setgate() can be called to initialize a newly 1640 * allocated route entry, in which case rt_key(rt) == NULL 1641 * (and also rt->rt_gateway == NULL). 1642 * Free()/free() handle a NULL argument just fine. 1643 */ 1644 bcopy(dst, new, dlen); 1645 Free(rt_key(rt)); /* free old block, if any */ 1646 rt_key(rt) = (struct sockaddr *)new; 1647 rt->rt_gateway = (struct sockaddr *)(new + dlen); 1648 } 1649 1650 /* 1651 * Copy the new gateway value into the memory chunk. 1652 */ 1653 bcopy(gate, rt->rt_gateway, glen); 1654 1655 return (0); 1656 } 1657 1658 void 1659 rt_maskedcopy(struct sockaddr *src, struct sockaddr *dst, struct sockaddr *netmask) 1660 { 1661 u_char *cp1 = (u_char *)src; 1662 u_char *cp2 = (u_char *)dst; 1663 u_char *cp3 = (u_char *)netmask; 1664 u_char *cplim = cp2 + *cp3; 1665 u_char *cplim2 = cp2 + *cp1; 1666 1667 *cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */ 1668 cp3 += 2; 1669 if (cplim > cplim2) 1670 cplim = cplim2; 1671 while (cp2 < cplim) 1672 *cp2++ = *cp1++ & *cp3++; 1673 if (cp2 < cplim2) 1674 bzero((caddr_t)cp2, (unsigned)(cplim2 - cp2)); 1675 } 1676 1677 /* 1678 * Set up a routing table entry, normally 1679 * for an interface. 1680 */ 1681 #define _SOCKADDR_TMPSIZE 128 /* Not too big.. kernel stack size is limited */ 1682 static inline int 1683 rtinit1(struct ifaddr *ifa, int cmd, int flags, int fibnum) 1684 { 1685 struct sockaddr *dst; 1686 struct sockaddr *netmask; 1687 struct rtentry *rt = NULL; 1688 struct rt_addrinfo info; 1689 int error = 0; 1690 int startfib, endfib; 1691 char tempbuf[_SOCKADDR_TMPSIZE]; 1692 int didwork = 0; 1693 int a_failure = 0; 1694 static struct sockaddr_dl null_sdl = {sizeof(null_sdl), AF_LINK}; 1695 struct radix_node_head *rnh; 1696 1697 if (flags & RTF_HOST) { 1698 dst = ifa->ifa_dstaddr; 1699 netmask = NULL; 1700 } else { 1701 dst = ifa->ifa_addr; 1702 netmask = ifa->ifa_netmask; 1703 } 1704 if (dst->sa_len == 0) 1705 return(EINVAL); 1706 switch (dst->sa_family) { 1707 case AF_INET6: 1708 case AF_INET: 1709 /* We support multiple FIBs. */ 1710 break; 1711 default: 1712 fibnum = RT_DEFAULT_FIB; 1713 break; 1714 } 1715 if (fibnum == RT_ALL_FIBS) { 1716 if (V_rt_add_addr_allfibs == 0 && cmd == (int)RTM_ADD) 1717 startfib = endfib = ifa->ifa_ifp->if_fib; 1718 else { 1719 startfib = 0; 1720 endfib = rt_numfibs - 1; 1721 } 1722 } else { 1723 KASSERT((fibnum < rt_numfibs), ("rtinit1: bad fibnum")); 1724 startfib = fibnum; 1725 endfib = fibnum; 1726 } 1727 1728 /* 1729 * If it's a delete, check that if it exists, 1730 * it's on the correct interface or we might scrub 1731 * a route to another ifa which would 1732 * be confusing at best and possibly worse. 1733 */ 1734 if (cmd == RTM_DELETE) { 1735 /* 1736 * It's a delete, so it should already exist.. 1737 * If it's a net, mask off the host bits 1738 * (Assuming we have a mask) 1739 * XXX this is kinda inet specific.. 1740 */ 1741 if (netmask != NULL) { 1742 rt_maskedcopy(dst, (struct sockaddr *)tempbuf, netmask); 1743 dst = (struct sockaddr *)tempbuf; 1744 } 1745 } 1746 /* 1747 * Now go through all the requested tables (fibs) and do the 1748 * requested action. Realistically, this will either be fib 0 1749 * for protocols that don't do multiple tables or all the 1750 * tables for those that do. 1751 */ 1752 for ( fibnum = startfib; fibnum <= endfib; fibnum++) { 1753 if (cmd == RTM_DELETE) { 1754 struct radix_node *rn; 1755 /* 1756 * Look up an rtentry that is in the routing tree and 1757 * contains the correct info. 1758 */ 1759 rnh = rt_tables_get_rnh(fibnum, dst->sa_family); 1760 if (rnh == NULL) 1761 /* this table doesn't exist but others might */ 1762 continue; 1763 RADIX_NODE_HEAD_RLOCK(rnh); 1764 rn = rnh->rnh_lookup(dst, netmask, rnh); 1765 #ifdef RADIX_MPATH 1766 if (rn_mpath_capable(rnh)) { 1767 1768 if (rn == NULL) 1769 error = ESRCH; 1770 else { 1771 rt = RNTORT(rn); 1772 /* 1773 * for interface route the 1774 * rt->rt_gateway is sockaddr_intf 1775 * for cloning ARP entries, so 1776 * rt_mpath_matchgate must use the 1777 * interface address 1778 */ 1779 rt = rt_mpath_matchgate(rt, 1780 ifa->ifa_addr); 1781 if (rt == NULL) 1782 error = ESRCH; 1783 } 1784 } 1785 #endif 1786 error = (rn == NULL || 1787 (rn->rn_flags & RNF_ROOT) || 1788 RNTORT(rn)->rt_ifa != ifa); 1789 RADIX_NODE_HEAD_RUNLOCK(rnh); 1790 if (error) { 1791 /* this is only an error if bad on ALL tables */ 1792 continue; 1793 } 1794 } 1795 /* 1796 * Do the actual request 1797 */ 1798 bzero((caddr_t)&info, sizeof(info)); 1799 info.rti_ifa = ifa; 1800 info.rti_flags = flags | 1801 (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED; 1802 info.rti_info[RTAX_DST] = dst; 1803 /* 1804 * doing this for compatibility reasons 1805 */ 1806 if (cmd == RTM_ADD) 1807 info.rti_info[RTAX_GATEWAY] = 1808 (struct sockaddr *)&null_sdl; 1809 else 1810 info.rti_info[RTAX_GATEWAY] = ifa->ifa_addr; 1811 info.rti_info[RTAX_NETMASK] = netmask; 1812 error = rtrequest1_fib(cmd, &info, &rt, fibnum); 1813 1814 if ((error == EEXIST) && (cmd == RTM_ADD)) { 1815 /* 1816 * Interface route addition failed. 1817 * Atomically delete current prefix generating 1818 * RTM_DELETE message, and retry adding 1819 * interface prefix. 1820 */ 1821 rnh = rt_tables_get_rnh(fibnum, dst->sa_family); 1822 RADIX_NODE_HEAD_LOCK(rnh); 1823 1824 /* Delete old prefix */ 1825 info.rti_ifa = NULL; 1826 info.rti_flags = RTF_RNH_LOCKED; 1827 1828 error = rtrequest1_fib(RTM_DELETE, &info, NULL, fibnum); 1829 if (error == 0) { 1830 info.rti_ifa = ifa; 1831 info.rti_flags = flags | RTF_RNH_LOCKED | 1832 (ifa->ifa_flags & ~IFA_RTSELF) | RTF_PINNED; 1833 error = rtrequest1_fib(cmd, &info, &rt, fibnum); 1834 } 1835 1836 RADIX_NODE_HEAD_UNLOCK(rnh); 1837 } 1838 1839 1840 if (error == 0 && rt != NULL) { 1841 /* 1842 * notify any listening routing agents of the change 1843 */ 1844 RT_LOCK(rt); 1845 #ifdef RADIX_MPATH 1846 /* 1847 * in case address alias finds the first address 1848 * e.g. ifconfig bge0 192.0.2.246/24 1849 * e.g. ifconfig bge0 192.0.2.247/24 1850 * the address set in the route is 192.0.2.246 1851 * so we need to replace it with 192.0.2.247 1852 */ 1853 if (memcmp(rt->rt_ifa->ifa_addr, 1854 ifa->ifa_addr, ifa->ifa_addr->sa_len)) { 1855 ifa_free(rt->rt_ifa); 1856 ifa_ref(ifa); 1857 rt->rt_ifp = ifa->ifa_ifp; 1858 rt->rt_ifa = ifa; 1859 } 1860 #endif 1861 /* 1862 * doing this for compatibility reasons 1863 */ 1864 if (cmd == RTM_ADD) { 1865 ((struct sockaddr_dl *)rt->rt_gateway)->sdl_type = 1866 rt->rt_ifp->if_type; 1867 ((struct sockaddr_dl *)rt->rt_gateway)->sdl_index = 1868 rt->rt_ifp->if_index; 1869 } 1870 RT_ADDREF(rt); 1871 RT_UNLOCK(rt); 1872 rt_newaddrmsg_fib(cmd, ifa, error, rt, fibnum); 1873 RT_LOCK(rt); 1874 RT_REMREF(rt); 1875 if (cmd == RTM_DELETE) { 1876 /* 1877 * If we are deleting, and we found an entry, 1878 * then it's been removed from the tree.. 1879 * now throw it away. 1880 */ 1881 RTFREE_LOCKED(rt); 1882 } else { 1883 if (cmd == RTM_ADD) { 1884 /* 1885 * We just wanted to add it.. 1886 * we don't actually need a reference. 1887 */ 1888 RT_REMREF(rt); 1889 } 1890 RT_UNLOCK(rt); 1891 } 1892 didwork = 1; 1893 } 1894 if (error) 1895 a_failure = error; 1896 } 1897 if (cmd == RTM_DELETE) { 1898 if (didwork) { 1899 error = 0; 1900 } else { 1901 /* we only give an error if it wasn't in any table */ 1902 error = ((flags & RTF_HOST) ? 1903 EHOSTUNREACH : ENETUNREACH); 1904 } 1905 } else { 1906 if (a_failure) { 1907 /* return an error if any of them failed */ 1908 error = a_failure; 1909 } 1910 } 1911 return (error); 1912 } 1913 1914 /* 1915 * Set up a routing table entry, normally 1916 * for an interface. 1917 */ 1918 int 1919 rtinit(struct ifaddr *ifa, int cmd, int flags) 1920 { 1921 struct sockaddr *dst; 1922 int fib = RT_DEFAULT_FIB; 1923 1924 if (flags & RTF_HOST) { 1925 dst = ifa->ifa_dstaddr; 1926 } else { 1927 dst = ifa->ifa_addr; 1928 } 1929 1930 switch (dst->sa_family) { 1931 case AF_INET6: 1932 case AF_INET: 1933 /* We do support multiple FIBs. */ 1934 fib = RT_ALL_FIBS; 1935 break; 1936 } 1937 return (rtinit1(ifa, cmd, flags, fib)); 1938 } 1939 1940 /* 1941 * Announce interface address arrival/withdraw 1942 * Returns 0 on success. 1943 */ 1944 int 1945 rt_addrmsg(int cmd, struct ifaddr *ifa, int fibnum) 1946 { 1947 1948 KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE, 1949 ("unexpected cmd %d", cmd)); 1950 1951 KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs), 1952 ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs)); 1953 1954 #if defined(INET) || defined(INET6) 1955 #ifdef SCTP 1956 /* 1957 * notify the SCTP stack 1958 * this will only get called when an address is added/deleted 1959 * XXX pass the ifaddr struct instead if ifa->ifa_addr... 1960 */ 1961 sctp_addr_change(ifa, cmd); 1962 #endif /* SCTP */ 1963 #endif 1964 return (rtsock_addrmsg(cmd, ifa, fibnum)); 1965 } 1966 1967 /* 1968 * Announce route addition/removal. 1969 * Users of this function MUST validate input data BEFORE calling. 1970 * However we have to be able to handle invalid data: 1971 * if some userland app sends us "invalid" route message (invalid mask, 1972 * no dst, wrong address families, etc...) we need to pass it back 1973 * to app (and any other rtsock consumers) with rtm_errno field set to 1974 * non-zero value. 1975 * Returns 0 on success. 1976 */ 1977 int 1978 rt_routemsg(int cmd, struct ifnet *ifp, int error, struct rtentry *rt, 1979 int fibnum) 1980 { 1981 1982 KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE, 1983 ("unexpected cmd %d", cmd)); 1984 1985 KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs), 1986 ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs)); 1987 1988 KASSERT(rt_key(rt) != NULL, (":%s: rt_key must be supplied", __func__)); 1989 1990 return (rtsock_routemsg(cmd, ifp, error, rt, fibnum)); 1991 } 1992 1993 void 1994 rt_newaddrmsg(int cmd, struct ifaddr *ifa, int error, struct rtentry *rt) 1995 { 1996 1997 rt_newaddrmsg_fib(cmd, ifa, error, rt, RT_ALL_FIBS); 1998 } 1999 2000 /* 2001 * This is called to generate messages from the routing socket 2002 * indicating a network interface has had addresses associated with it. 2003 */ 2004 void 2005 rt_newaddrmsg_fib(int cmd, struct ifaddr *ifa, int error, struct rtentry *rt, 2006 int fibnum) 2007 { 2008 2009 KASSERT(cmd == RTM_ADD || cmd == RTM_DELETE, 2010 ("unexpected cmd %u", cmd)); 2011 KASSERT(fibnum == RT_ALL_FIBS || (fibnum >= 0 && fibnum < rt_numfibs), 2012 ("%s: fib out of range 0 <=%d<%d", __func__, fibnum, rt_numfibs)); 2013 2014 if (cmd == RTM_ADD) { 2015 rt_addrmsg(cmd, ifa, fibnum); 2016 if (rt != NULL) 2017 rt_routemsg(cmd, ifa->ifa_ifp, error, rt, fibnum); 2018 } else { 2019 if (rt != NULL) 2020 rt_routemsg(cmd, ifa->ifa_ifp, error, rt, fibnum); 2021 rt_addrmsg(cmd, ifa, fibnum); 2022 } 2023 } 2024 2025