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