1 /* 2 * Copyright (c) 2004, 2005 The DragonFly Project. All rights reserved. 3 * 4 * This code is derived from software contributed to The DragonFly Project 5 * by Jeffrey M. Hsu. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of The DragonFly Project nor the names of its 16 * contributors may be used to endorse or promote products derived 17 * from this software without specific, prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 21 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 22 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 23 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 24 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 25 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 26 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 27 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 28 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 29 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 */ 32 33 /* 34 * Copyright (c) 1980, 1986, 1991, 1993 35 * The Regents of the University of California. All rights reserved. 36 * 37 * Redistribution and use in source and binary forms, with or without 38 * modification, are permitted provided that the following conditions 39 * are met: 40 * 1. Redistributions of source code must retain the above copyright 41 * notice, this list of conditions and the following disclaimer. 42 * 2. Redistributions in binary form must reproduce the above copyright 43 * notice, this list of conditions and the following disclaimer in the 44 * documentation and/or other materials provided with the distribution. 45 * 3. All advertising materials mentioning features or use of this software 46 * must display the following acknowledgement: 47 * This product includes software developed by the University of 48 * California, Berkeley and its contributors. 49 * 4. Neither the name of the University nor the names of its contributors 50 * may be used to endorse or promote products derived from this software 51 * without specific prior written permission. 52 * 53 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 54 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 55 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 56 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 57 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 58 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 59 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 60 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 61 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 62 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 63 * SUCH DAMAGE. 64 * 65 * @(#)route.c 8.3 (Berkeley) 1/9/95 66 * $FreeBSD: src/sys/net/route.c,v 1.59.2.10 2003/01/17 08:04:00 ru Exp $ 67 * $DragonFly: src/sys/net/route.c,v 1.41 2008/11/09 10:50:15 sephe Exp $ 68 */ 69 70 #include "opt_inet.h" 71 #include "opt_mpls.h" 72 73 #include <sys/param.h> 74 #include <sys/systm.h> 75 #include <sys/malloc.h> 76 #include <sys/mbuf.h> 77 #include <sys/socket.h> 78 #include <sys/domain.h> 79 #include <sys/kernel.h> 80 #include <sys/sysctl.h> 81 #include <sys/globaldata.h> 82 #include <sys/thread.h> 83 84 #include <net/if.h> 85 #include <net/route.h> 86 #include <net/netisr.h> 87 88 #include <netinet/in.h> 89 #include <net/ip_mroute/ip_mroute.h> 90 91 #include <sys/thread2.h> 92 #include <sys/msgport2.h> 93 #include <net/netmsg2.h> 94 95 #ifdef MPLS 96 #include <netproto/mpls/mpls.h> 97 #endif 98 99 static struct rtstatistics rtstatistics_percpu[MAXCPU]; 100 #ifdef SMP 101 #define rtstat rtstatistics_percpu[mycpuid] 102 #else 103 #define rtstat rtstatistics_percpu[0] 104 #endif 105 106 struct radix_node_head *rt_tables[MAXCPU][AF_MAX+1]; 107 struct lwkt_port *rt_ports[MAXCPU]; 108 109 static void rt_maskedcopy (struct sockaddr *, struct sockaddr *, 110 struct sockaddr *); 111 static void rtable_init(void); 112 static void rtable_service_loop(void *dummy); 113 static void rtinit_rtrequest_callback(int, int, struct rt_addrinfo *, 114 struct rtentry *, void *); 115 116 #ifdef SMP 117 static void rtredirect_msghandler(struct netmsg *netmsg); 118 static void rtrequest1_msghandler(struct netmsg *netmsg); 119 #endif 120 static void rtsearch_msghandler(struct netmsg *netmsg); 121 122 static void rtmask_add_msghandler(struct netmsg *netmsg); 123 124 static int rt_setshims(struct rtentry *, struct sockaddr **); 125 126 SYSCTL_NODE(_net, OID_AUTO, route, CTLFLAG_RW, 0, "Routing"); 127 128 #ifdef ROUTE_DEBUG 129 static int route_debug = 1; 130 SYSCTL_INT(_net_route, OID_AUTO, route_debug, CTLFLAG_RW, 131 &route_debug, 0, ""); 132 #endif 133 134 int route_assert_owner_access = 0; 135 SYSCTL_INT(_net_route, OID_AUTO, assert_owner_access, CTLFLAG_RW, 136 &route_assert_owner_access, 0, ""); 137 138 /* 139 * Initialize the route table(s) for protocol domains and 140 * create a helper thread which will be responsible for updating 141 * route table entries on each cpu. 142 */ 143 void 144 route_init(void) 145 { 146 int cpu; 147 thread_t rtd; 148 149 for (cpu = 0; cpu < ncpus; ++cpu) 150 bzero(&rtstatistics_percpu[cpu], sizeof(struct rtstatistics)); 151 rn_init(); /* initialize all zeroes, all ones, mask table */ 152 rtable_init(); /* call dom_rtattach() on each cpu */ 153 154 for (cpu = 0; cpu < ncpus; cpu++) { 155 lwkt_create(rtable_service_loop, NULL, &rtd, NULL, 156 0, cpu, "rtable_cpu %d", cpu); 157 rt_ports[cpu] = &rtd->td_msgport; 158 } 159 } 160 161 static void 162 rtable_init_oncpu(struct netmsg *nmsg) 163 { 164 struct domain *dom; 165 int cpu = mycpuid; 166 167 SLIST_FOREACH(dom, &domains, dom_next) { 168 if (dom->dom_rtattach) { 169 dom->dom_rtattach( 170 (void **)&rt_tables[cpu][dom->dom_family], 171 dom->dom_rtoffset); 172 } 173 } 174 ifnet_forwardmsg(&nmsg->nm_lmsg, cpu + 1); 175 } 176 177 static void 178 rtable_init(void) 179 { 180 struct netmsg nmsg; 181 182 netmsg_init(&nmsg, &curthread->td_msgport, 0, rtable_init_oncpu); 183 ifnet_domsg(&nmsg.nm_lmsg, 0); 184 } 185 186 /* 187 * Our per-cpu table management protocol thread. All route table operations 188 * are sequentially chained through all cpus starting at cpu #0 in order to 189 * maintain duplicate route tables on each cpu. Having a spearate route 190 * table management thread allows the protocol and interrupt threads to 191 * issue route table changes. 192 */ 193 static void 194 rtable_service_loop(void *dummy __unused) 195 { 196 struct netmsg *netmsg; 197 thread_t td = curthread; 198 199 while ((netmsg = lwkt_waitport(&td->td_msgport, 0)) != NULL) { 200 netmsg->nm_dispatch(netmsg); 201 } 202 } 203 204 /* 205 * Routing statistics. 206 */ 207 #ifdef SMP 208 static int 209 sysctl_rtstatistics(SYSCTL_HANDLER_ARGS) 210 { 211 int cpu, error = 0; 212 213 for (cpu = 0; cpu < ncpus; ++cpu) { 214 if ((error = SYSCTL_OUT(req, &rtstatistics_percpu[cpu], 215 sizeof(struct rtstatistics)))) 216 break; 217 if ((error = SYSCTL_IN(req, &rtstatistics_percpu[cpu], 218 sizeof(struct rtstatistics)))) 219 break; 220 } 221 222 return (error); 223 } 224 SYSCTL_PROC(_net_route, OID_AUTO, stats, (CTLTYPE_OPAQUE|CTLFLAG_RW), 225 0, 0, sysctl_rtstatistics, "S,rtstatistics", "Routing statistics"); 226 #else 227 SYSCTL_STRUCT(_net_route, OID_AUTO, stats, CTLFLAG_RW, &rtstat, rtstatistics, 228 "Routing statistics"); 229 #endif 230 231 /* 232 * Packet routing routines. 233 */ 234 235 /* 236 * Look up and fill in the "ro_rt" rtentry field in a route structure given 237 * an address in the "ro_dst" field. Always send a report on a miss and 238 * always clone routes. 239 */ 240 void 241 rtalloc(struct route *ro) 242 { 243 rtalloc_ign(ro, 0UL); 244 } 245 246 /* 247 * Look up and fill in the "ro_rt" rtentry field in a route structure given 248 * an address in the "ro_dst" field. Always send a report on a miss and 249 * optionally clone routes when RTF_CLONING or RTF_PRCLONING are not being 250 * ignored. 251 */ 252 void 253 rtalloc_ign(struct route *ro, u_long ignoreflags) 254 { 255 if (ro->ro_rt != NULL) { 256 if (ro->ro_rt->rt_ifp != NULL && ro->ro_rt->rt_flags & RTF_UP) 257 return; 258 rtfree(ro->ro_rt); 259 ro->ro_rt = NULL; 260 } 261 ro->ro_rt = _rtlookup(&ro->ro_dst, RTL_REPORTMSG, ignoreflags); 262 } 263 264 /* 265 * Look up the route that matches the given "dst" address. 266 * 267 * Route lookup can have the side-effect of creating and returning 268 * a cloned route instead when "dst" matches a cloning route and the 269 * RTF_CLONING and RTF_PRCLONING flags are not being ignored. 270 * 271 * Any route returned has its reference count incremented. 272 */ 273 struct rtentry * 274 _rtlookup(struct sockaddr *dst, boolean_t generate_report, u_long ignore) 275 { 276 struct radix_node_head *rnh = rt_tables[mycpuid][dst->sa_family]; 277 struct rtentry *rt; 278 279 if (rnh == NULL) 280 goto unreach; 281 282 /* 283 * Look up route in the radix tree. 284 */ 285 rt = (struct rtentry *) rnh->rnh_matchaddr((char *)dst, rnh); 286 if (rt == NULL) 287 goto unreach; 288 289 /* 290 * Handle cloning routes. 291 */ 292 if ((rt->rt_flags & ~ignore & (RTF_CLONING | RTF_PRCLONING)) != 0) { 293 struct rtentry *clonedroute; 294 int error; 295 296 clonedroute = rt; /* copy in/copy out parameter */ 297 error = rtrequest(RTM_RESOLVE, dst, NULL, NULL, 0, 298 &clonedroute); /* clone the route */ 299 if (error != 0) { /* cloning failed */ 300 if (generate_report) 301 rt_dstmsg(RTM_MISS, dst, error); 302 rt->rt_refcnt++; 303 return (rt); /* return the uncloned route */ 304 } 305 if (generate_report) { 306 if (clonedroute->rt_flags & RTF_XRESOLVE) 307 rt_dstmsg(RTM_RESOLVE, dst, 0); 308 else 309 rt_rtmsg(RTM_ADD, clonedroute, 310 clonedroute->rt_ifp, 0); 311 } 312 return (clonedroute); /* return cloned route */ 313 } 314 315 /* 316 * Increment the reference count of the matched route and return. 317 */ 318 rt->rt_refcnt++; 319 return (rt); 320 321 unreach: 322 rtstat.rts_unreach++; 323 if (generate_report) 324 rt_dstmsg(RTM_MISS, dst, 0); 325 return (NULL); 326 } 327 328 void 329 rtfree(struct rtentry *rt) 330 { 331 if (rt->rt_cpuid == mycpuid) 332 rtfree_oncpu(rt); 333 else 334 rtfree_remote(rt, 1); 335 } 336 337 void 338 rtfree_oncpu(struct rtentry *rt) 339 { 340 KKASSERT(rt->rt_cpuid == mycpuid); 341 KASSERT(rt->rt_refcnt > 0, ("rtfree: rt_refcnt %ld", rt->rt_refcnt)); 342 343 --rt->rt_refcnt; 344 if (rt->rt_refcnt == 0) { 345 struct radix_node_head *rnh = 346 rt_tables[mycpuid][rt_key(rt)->sa_family]; 347 348 if (rnh->rnh_close) 349 rnh->rnh_close((struct radix_node *)rt, rnh); 350 if (!(rt->rt_flags & RTF_UP)) { 351 /* deallocate route */ 352 if (rt->rt_ifa != NULL) 353 IFAFREE(rt->rt_ifa); 354 if (rt->rt_parent != NULL) 355 RTFREE(rt->rt_parent); /* recursive call! */ 356 Free(rt_key(rt)); 357 Free(rt); 358 } 359 } 360 } 361 362 static void 363 rtfree_remote_dispatch(struct netmsg *nmsg) 364 { 365 struct lwkt_msg *lmsg = &nmsg->nm_lmsg; 366 struct rtentry *rt = lmsg->u.ms_resultp; 367 368 rtfree_oncpu(rt); 369 lwkt_replymsg(lmsg, 0); 370 } 371 372 void 373 rtfree_remote(struct rtentry *rt, int allow_panic) 374 { 375 struct netmsg nmsg; 376 struct lwkt_msg *lmsg; 377 378 KKASSERT(rt->rt_cpuid != mycpuid); 379 380 if (route_assert_owner_access && allow_panic) { 381 panic("rt remote free rt_cpuid %d, mycpuid %d\n", 382 rt->rt_cpuid, mycpuid); 383 } else { 384 kprintf("rt remote free rt_cpuid %d, mycpuid %d\n", 385 rt->rt_cpuid, mycpuid); 386 print_backtrace(); 387 } 388 389 netmsg_init(&nmsg, &curthread->td_msgport, 0, rtfree_remote_dispatch); 390 lmsg = &nmsg.nm_lmsg; 391 lmsg->u.ms_resultp = rt; 392 393 lwkt_domsg(rtable_portfn(rt->rt_cpuid), lmsg, 0); 394 } 395 396 static int 397 rtredirect_oncpu(struct sockaddr *dst, struct sockaddr *gateway, 398 struct sockaddr *netmask, int flags, struct sockaddr *src) 399 { 400 struct rtentry *rt = NULL; 401 struct rt_addrinfo rtinfo; 402 struct ifaddr *ifa; 403 u_long *stat = NULL; 404 int error; 405 406 /* verify the gateway is directly reachable */ 407 if ((ifa = ifa_ifwithnet(gateway)) == NULL) { 408 error = ENETUNREACH; 409 goto out; 410 } 411 412 /* 413 * If the redirect isn't from our current router for this destination, 414 * it's either old or wrong. 415 */ 416 if (!(flags & RTF_DONE) && /* XXX JH */ 417 (rt = rtpurelookup(dst)) != NULL && 418 (!sa_equal(src, rt->rt_gateway) || rt->rt_ifa != ifa)) { 419 error = EINVAL; 420 goto done; 421 } 422 423 /* 424 * If it redirects us to ourselves, we have a routing loop, 425 * perhaps as a result of an interface going down recently. 426 */ 427 if (ifa_ifwithaddr(gateway)) { 428 error = EHOSTUNREACH; 429 goto done; 430 } 431 432 /* 433 * Create a new entry if the lookup failed or if we got back 434 * a wildcard entry for the default route. This is necessary 435 * for hosts which use routing redirects generated by smart 436 * gateways to dynamically build the routing tables. 437 */ 438 if (rt == NULL) 439 goto create; 440 if ((rt_mask(rt) != NULL && rt_mask(rt)->sa_len < 2)) { 441 rtfree(rt); 442 goto create; 443 } 444 445 /* Ignore redirects for directly connected hosts. */ 446 if (!(rt->rt_flags & RTF_GATEWAY)) { 447 error = EHOSTUNREACH; 448 goto done; 449 } 450 451 if (!(rt->rt_flags & RTF_HOST) && (flags & RTF_HOST)) { 452 /* 453 * Changing from a network route to a host route. 454 * Create a new host route rather than smashing the 455 * network route. 456 */ 457 create: 458 flags |= RTF_GATEWAY | RTF_DYNAMIC; 459 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 460 rtinfo.rti_info[RTAX_DST] = dst; 461 rtinfo.rti_info[RTAX_GATEWAY] = gateway; 462 rtinfo.rti_info[RTAX_NETMASK] = netmask; 463 rtinfo.rti_flags = flags; 464 rtinfo.rti_ifa = ifa; 465 rt = NULL; /* copy-in/copy-out parameter */ 466 error = rtrequest1(RTM_ADD, &rtinfo, &rt); 467 if (rt != NULL) 468 flags = rt->rt_flags; 469 stat = &rtstat.rts_dynamic; 470 } else { 471 /* 472 * Smash the current notion of the gateway to this destination. 473 * Should check about netmask!!! 474 */ 475 rt->rt_flags |= RTF_MODIFIED; 476 flags |= RTF_MODIFIED; 477 478 /* We only need to report rtmsg on CPU0 */ 479 rt_setgate(rt, rt_key(rt), gateway, 480 mycpuid == 0 ? RTL_REPORTMSG : RTL_DONTREPORT); 481 error = 0; 482 stat = &rtstat.rts_newgateway; 483 } 484 485 done: 486 if (rt != NULL) 487 rtfree(rt); 488 out: 489 if (error != 0) 490 rtstat.rts_badredirect++; 491 else if (stat != NULL) 492 (*stat)++; 493 494 return error; 495 } 496 497 #ifdef SMP 498 499 struct netmsg_rtredirect { 500 struct netmsg netmsg; 501 struct sockaddr *dst; 502 struct sockaddr *gateway; 503 struct sockaddr *netmask; 504 int flags; 505 struct sockaddr *src; 506 }; 507 508 #endif 509 510 /* 511 * Force a routing table entry to the specified 512 * destination to go through the given gateway. 513 * Normally called as a result of a routing redirect 514 * message from the network layer. 515 * 516 * N.B.: must be called at splnet 517 */ 518 void 519 rtredirect(struct sockaddr *dst, struct sockaddr *gateway, 520 struct sockaddr *netmask, int flags, struct sockaddr *src) 521 { 522 struct rt_addrinfo rtinfo; 523 int error; 524 #ifdef SMP 525 struct netmsg_rtredirect msg; 526 527 netmsg_init(&msg.netmsg, &curthread->td_msgport, 0, 528 rtredirect_msghandler); 529 msg.dst = dst; 530 msg.gateway = gateway; 531 msg.netmask = netmask; 532 msg.flags = flags; 533 msg.src = src; 534 error = lwkt_domsg(rtable_portfn(0), &msg.netmsg.nm_lmsg, 0); 535 #else 536 error = rtredirect_oncpu(dst, gateway, netmask, flags, src); 537 #endif 538 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 539 rtinfo.rti_info[RTAX_DST] = dst; 540 rtinfo.rti_info[RTAX_GATEWAY] = gateway; 541 rtinfo.rti_info[RTAX_NETMASK] = netmask; 542 rtinfo.rti_info[RTAX_AUTHOR] = src; 543 rt_missmsg(RTM_REDIRECT, &rtinfo, flags, error); 544 } 545 546 #ifdef SMP 547 548 static void 549 rtredirect_msghandler(struct netmsg *netmsg) 550 { 551 struct netmsg_rtredirect *msg = (void *)netmsg; 552 int nextcpu; 553 554 rtredirect_oncpu(msg->dst, msg->gateway, msg->netmask, 555 msg->flags, msg->src); 556 nextcpu = mycpuid + 1; 557 if (nextcpu < ncpus) 558 lwkt_forwardmsg(rtable_portfn(nextcpu), &netmsg->nm_lmsg); 559 else 560 lwkt_replymsg(&netmsg->nm_lmsg, 0); 561 } 562 563 #endif 564 565 /* 566 * Routing table ioctl interface. 567 */ 568 int 569 rtioctl(u_long req, caddr_t data, struct ucred *cred) 570 { 571 #ifdef INET 572 /* Multicast goop, grrr... */ 573 return mrt_ioctl ? mrt_ioctl(req, data) : EOPNOTSUPP; 574 #else 575 return ENXIO; 576 #endif 577 } 578 579 struct ifaddr * 580 ifa_ifwithroute(int flags, struct sockaddr *dst, struct sockaddr *gateway) 581 { 582 struct ifaddr *ifa; 583 584 if (!(flags & RTF_GATEWAY)) { 585 /* 586 * If we are adding a route to an interface, 587 * and the interface is a point-to-point link, 588 * we should search for the destination 589 * as our clue to the interface. Otherwise 590 * we can use the local address. 591 */ 592 ifa = NULL; 593 if (flags & RTF_HOST) { 594 ifa = ifa_ifwithdstaddr(dst); 595 } 596 if (ifa == NULL) 597 ifa = ifa_ifwithaddr(gateway); 598 } else { 599 /* 600 * If we are adding a route to a remote net 601 * or host, the gateway may still be on the 602 * other end of a pt to pt link. 603 */ 604 ifa = ifa_ifwithdstaddr(gateway); 605 } 606 if (ifa == NULL) 607 ifa = ifa_ifwithnet(gateway); 608 if (ifa == NULL) { 609 struct rtentry *rt; 610 611 rt = rtpurelookup(gateway); 612 if (rt == NULL) 613 return (NULL); 614 rt->rt_refcnt--; 615 if ((ifa = rt->rt_ifa) == NULL) 616 return (NULL); 617 } 618 if (ifa->ifa_addr->sa_family != dst->sa_family) { 619 struct ifaddr *oldifa = ifa; 620 621 ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp); 622 if (ifa == NULL) 623 ifa = oldifa; 624 } 625 return (ifa); 626 } 627 628 static int rt_fixdelete (struct radix_node *, void *); 629 static int rt_fixchange (struct radix_node *, void *); 630 631 struct rtfc_arg { 632 struct rtentry *rt0; 633 struct radix_node_head *rnh; 634 }; 635 636 /* 637 * Set rtinfo->rti_ifa and rtinfo->rti_ifp. 638 */ 639 int 640 rt_getifa(struct rt_addrinfo *rtinfo) 641 { 642 struct sockaddr *gateway = rtinfo->rti_info[RTAX_GATEWAY]; 643 struct sockaddr *dst = rtinfo->rti_info[RTAX_DST]; 644 struct sockaddr *ifaaddr = rtinfo->rti_info[RTAX_IFA]; 645 int flags = rtinfo->rti_flags; 646 647 /* 648 * ifp may be specified by sockaddr_dl 649 * when protocol address is ambiguous. 650 */ 651 if (rtinfo->rti_ifp == NULL) { 652 struct sockaddr *ifpaddr; 653 654 ifpaddr = rtinfo->rti_info[RTAX_IFP]; 655 if (ifpaddr != NULL && ifpaddr->sa_family == AF_LINK) { 656 struct ifaddr *ifa; 657 658 ifa = ifa_ifwithnet(ifpaddr); 659 if (ifa != NULL) 660 rtinfo->rti_ifp = ifa->ifa_ifp; 661 } 662 } 663 664 if (rtinfo->rti_ifa == NULL && ifaaddr != NULL) 665 rtinfo->rti_ifa = ifa_ifwithaddr(ifaaddr); 666 if (rtinfo->rti_ifa == NULL) { 667 struct sockaddr *sa; 668 669 sa = ifaaddr != NULL ? ifaaddr : 670 (gateway != NULL ? gateway : dst); 671 if (sa != NULL && rtinfo->rti_ifp != NULL) 672 rtinfo->rti_ifa = ifaof_ifpforaddr(sa, rtinfo->rti_ifp); 673 else if (dst != NULL && gateway != NULL) 674 rtinfo->rti_ifa = ifa_ifwithroute(flags, dst, gateway); 675 else if (sa != NULL) 676 rtinfo->rti_ifa = ifa_ifwithroute(flags, sa, sa); 677 } 678 if (rtinfo->rti_ifa == NULL) 679 return (ENETUNREACH); 680 681 if (rtinfo->rti_ifp == NULL) 682 rtinfo->rti_ifp = rtinfo->rti_ifa->ifa_ifp; 683 return (0); 684 } 685 686 /* 687 * Do appropriate manipulations of a routing tree given 688 * all the bits of info needed 689 */ 690 int 691 rtrequest( 692 int req, 693 struct sockaddr *dst, 694 struct sockaddr *gateway, 695 struct sockaddr *netmask, 696 int flags, 697 struct rtentry **ret_nrt) 698 { 699 struct rt_addrinfo rtinfo; 700 701 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 702 rtinfo.rti_info[RTAX_DST] = dst; 703 rtinfo.rti_info[RTAX_GATEWAY] = gateway; 704 rtinfo.rti_info[RTAX_NETMASK] = netmask; 705 rtinfo.rti_flags = flags; 706 return rtrequest1(req, &rtinfo, ret_nrt); 707 } 708 709 int 710 rtrequest_global( 711 int req, 712 struct sockaddr *dst, 713 struct sockaddr *gateway, 714 struct sockaddr *netmask, 715 int flags) 716 { 717 struct rt_addrinfo rtinfo; 718 719 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 720 rtinfo.rti_info[RTAX_DST] = dst; 721 rtinfo.rti_info[RTAX_GATEWAY] = gateway; 722 rtinfo.rti_info[RTAX_NETMASK] = netmask; 723 rtinfo.rti_flags = flags; 724 return rtrequest1_global(req, &rtinfo, NULL, NULL); 725 } 726 727 #ifdef SMP 728 729 struct netmsg_rtq { 730 struct netmsg netmsg; 731 int req; 732 struct rt_addrinfo *rtinfo; 733 rtrequest1_callback_func_t callback; 734 void *arg; 735 }; 736 737 #endif 738 739 int 740 rtrequest1_global(int req, struct rt_addrinfo *rtinfo, 741 rtrequest1_callback_func_t callback, void *arg) 742 { 743 int error; 744 #ifdef SMP 745 struct netmsg_rtq msg; 746 747 netmsg_init(&msg.netmsg, &curthread->td_msgport, 0, 748 rtrequest1_msghandler); 749 msg.netmsg.nm_lmsg.ms_error = -1; 750 msg.req = req; 751 msg.rtinfo = rtinfo; 752 msg.callback = callback; 753 msg.arg = arg; 754 error = lwkt_domsg(rtable_portfn(0), &msg.netmsg.nm_lmsg, 0); 755 #else 756 struct rtentry *rt = NULL; 757 758 error = rtrequest1(req, rtinfo, &rt); 759 if (rt) 760 --rt->rt_refcnt; 761 if (callback) 762 callback(req, error, rtinfo, rt, arg); 763 #endif 764 return (error); 765 } 766 767 /* 768 * Handle a route table request on the current cpu. Since the route table's 769 * are supposed to be identical on each cpu, an error occuring later in the 770 * message chain is considered system-fatal. 771 */ 772 #ifdef SMP 773 774 static void 775 rtrequest1_msghandler(struct netmsg *netmsg) 776 { 777 struct netmsg_rtq *msg = (void *)netmsg; 778 struct rt_addrinfo rtinfo; 779 struct rtentry *rt = NULL; 780 int nextcpu; 781 int error; 782 783 /* 784 * Copy the rtinfo. We need to make sure that the original 785 * rtinfo, which is setup by the caller, in the netmsg will 786 * _not_ be changed; else the next CPU on the netmsg forwarding 787 * path will see a different rtinfo than what this CPU has seen. 788 */ 789 rtinfo = *msg->rtinfo; 790 791 error = rtrequest1(msg->req, &rtinfo, &rt); 792 if (rt) 793 --rt->rt_refcnt; 794 if (msg->callback) 795 msg->callback(msg->req, error, &rtinfo, rt, msg->arg); 796 797 /* 798 * RTM_DELETE's are propogated even if an error occurs, since a 799 * cloned route might be undergoing deletion and cloned routes 800 * are not necessarily replicated. An overall error is returned 801 * only if no cpus have the route in question. 802 */ 803 if (msg->netmsg.nm_lmsg.ms_error < 0 || error == 0) 804 msg->netmsg.nm_lmsg.ms_error = error; 805 806 nextcpu = mycpuid + 1; 807 if (error && msg->req != RTM_DELETE) { 808 if (mycpuid != 0) { 809 panic("rtrequest1_msghandler: rtrequest table " 810 "error was not on cpu #0"); 811 } 812 lwkt_replymsg(&msg->netmsg.nm_lmsg, error); 813 } else if (nextcpu < ncpus) { 814 lwkt_forwardmsg(rtable_portfn(nextcpu), &msg->netmsg.nm_lmsg); 815 } else { 816 lwkt_replymsg(&msg->netmsg.nm_lmsg, 817 msg->netmsg.nm_lmsg.ms_error); 818 } 819 } 820 821 #endif 822 823 int 824 rtrequest1(int req, struct rt_addrinfo *rtinfo, struct rtentry **ret_nrt) 825 { 826 struct sockaddr *dst = rtinfo->rti_info[RTAX_DST]; 827 struct rtentry *rt; 828 struct radix_node *rn; 829 struct radix_node_head *rnh; 830 struct ifaddr *ifa; 831 struct sockaddr *ndst; 832 boolean_t reportmsg; 833 int error = 0; 834 835 #define gotoerr(x) { error = x ; goto bad; } 836 837 #ifdef ROUTE_DEBUG 838 if (route_debug) 839 rt_addrinfo_print(req, rtinfo); 840 #endif 841 842 crit_enter(); 843 /* 844 * Find the correct routing tree to use for this Address Family 845 */ 846 if ((rnh = rt_tables[mycpuid][dst->sa_family]) == NULL) 847 gotoerr(EAFNOSUPPORT); 848 849 /* 850 * If we are adding a host route then we don't want to put 851 * a netmask in the tree, nor do we want to clone it. 852 */ 853 if (rtinfo->rti_flags & RTF_HOST) { 854 rtinfo->rti_info[RTAX_NETMASK] = NULL; 855 rtinfo->rti_flags &= ~(RTF_CLONING | RTF_PRCLONING); 856 } 857 858 switch (req) { 859 case RTM_DELETE: 860 /* Remove the item from the tree. */ 861 rn = rnh->rnh_deladdr((char *)rtinfo->rti_info[RTAX_DST], 862 (char *)rtinfo->rti_info[RTAX_NETMASK], 863 rnh); 864 if (rn == NULL) 865 gotoerr(ESRCH); 866 KASSERT(!(rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)), 867 ("rnh_deladdr returned flags 0x%x", rn->rn_flags)); 868 rt = (struct rtentry *)rn; 869 870 /* ref to prevent a deletion race */ 871 ++rt->rt_refcnt; 872 873 /* Free any routes cloned from this one. */ 874 if ((rt->rt_flags & (RTF_CLONING | RTF_PRCLONING)) && 875 rt_mask(rt) != NULL) { 876 rnh->rnh_walktree_from(rnh, (char *)rt_key(rt), 877 (char *)rt_mask(rt), 878 rt_fixdelete, rt); 879 } 880 881 if (rt->rt_gwroute != NULL) { 882 RTFREE(rt->rt_gwroute); 883 rt->rt_gwroute = NULL; 884 } 885 886 /* 887 * NB: RTF_UP must be set during the search above, 888 * because we might delete the last ref, causing 889 * rt to get freed prematurely. 890 */ 891 rt->rt_flags &= ~RTF_UP; 892 893 #ifdef ROUTE_DEBUG 894 if (route_debug) 895 rt_print(rtinfo, rt); 896 #endif 897 898 /* Give the protocol a chance to keep things in sync. */ 899 if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest) 900 ifa->ifa_rtrequest(RTM_DELETE, rt, rtinfo); 901 902 /* 903 * If the caller wants it, then it can have it, 904 * but it's up to it to free the rtentry as we won't be 905 * doing it. 906 */ 907 KASSERT(rt->rt_refcnt >= 0, 908 ("rtrequest1(DELETE): refcnt %ld", rt->rt_refcnt)); 909 if (ret_nrt != NULL) { 910 /* leave ref intact for return */ 911 *ret_nrt = rt; 912 } else { 913 /* deref / attempt to destroy */ 914 rtfree(rt); 915 } 916 break; 917 918 case RTM_RESOLVE: 919 if (ret_nrt == NULL || (rt = *ret_nrt) == NULL) 920 gotoerr(EINVAL); 921 ifa = rt->rt_ifa; 922 rtinfo->rti_flags = 923 rt->rt_flags & ~(RTF_CLONING | RTF_PRCLONING | RTF_STATIC); 924 rtinfo->rti_flags |= RTF_WASCLONED; 925 rtinfo->rti_info[RTAX_GATEWAY] = rt->rt_gateway; 926 if ((rtinfo->rti_info[RTAX_NETMASK] = rt->rt_genmask) == NULL) 927 rtinfo->rti_flags |= RTF_HOST; 928 rtinfo->rti_info[RTAX_MPLS1] = rt->rt_shim[0]; 929 rtinfo->rti_info[RTAX_MPLS2] = rt->rt_shim[1]; 930 rtinfo->rti_info[RTAX_MPLS3] = rt->rt_shim[2]; 931 goto makeroute; 932 933 case RTM_ADD: 934 KASSERT(!(rtinfo->rti_flags & RTF_GATEWAY) || 935 rtinfo->rti_info[RTAX_GATEWAY] != NULL, 936 ("rtrequest: GATEWAY but no gateway")); 937 938 if (rtinfo->rti_ifa == NULL && (error = rt_getifa(rtinfo))) 939 gotoerr(error); 940 ifa = rtinfo->rti_ifa; 941 makeroute: 942 R_Malloc(rt, struct rtentry *, sizeof(struct rtentry)); 943 if (rt == NULL) 944 gotoerr(ENOBUFS); 945 bzero(rt, sizeof(struct rtentry)); 946 rt->rt_flags = RTF_UP | rtinfo->rti_flags; 947 rt->rt_cpuid = mycpuid; 948 949 if (mycpuid != 0 && req == RTM_ADD) { 950 /* For RTM_ADD, we have already sent rtmsg on CPU0. */ 951 reportmsg = RTL_DONTREPORT; 952 } else { 953 /* 954 * For RTM_ADD, we only send rtmsg on CPU0. 955 * For RTM_RESOLVE, we always send rtmsg. XXX 956 */ 957 reportmsg = RTL_REPORTMSG; 958 } 959 error = rt_setgate(rt, dst, rtinfo->rti_info[RTAX_GATEWAY], 960 reportmsg); 961 if (error != 0) { 962 Free(rt); 963 gotoerr(error); 964 } 965 966 ndst = rt_key(rt); 967 if (rtinfo->rti_info[RTAX_NETMASK] != NULL) 968 rt_maskedcopy(dst, ndst, 969 rtinfo->rti_info[RTAX_NETMASK]); 970 else 971 bcopy(dst, ndst, dst->sa_len); 972 973 if (rtinfo->rti_info[RTAX_MPLS1] != NULL) 974 rt_setshims(rt, rtinfo->rti_info); 975 976 /* 977 * Note that we now have a reference to the ifa. 978 * This moved from below so that rnh->rnh_addaddr() can 979 * examine the ifa and ifa->ifa_ifp if it so desires. 980 */ 981 IFAREF(ifa); 982 rt->rt_ifa = ifa; 983 rt->rt_ifp = ifa->ifa_ifp; 984 /* XXX mtu manipulation will be done in rnh_addaddr -- itojun */ 985 986 rn = rnh->rnh_addaddr((char *)ndst, 987 (char *)rtinfo->rti_info[RTAX_NETMASK], 988 rnh, rt->rt_nodes); 989 if (rn == NULL) { 990 struct rtentry *oldrt; 991 992 /* 993 * We already have one of these in the tree. 994 * We do a special hack: if the old route was 995 * cloned, then we blow it away and try 996 * re-inserting the new one. 997 */ 998 oldrt = rtpurelookup(ndst); 999 if (oldrt != NULL) { 1000 --oldrt->rt_refcnt; 1001 if (oldrt->rt_flags & RTF_WASCLONED) { 1002 rtrequest(RTM_DELETE, rt_key(oldrt), 1003 oldrt->rt_gateway, 1004 rt_mask(oldrt), 1005 oldrt->rt_flags, NULL); 1006 rn = rnh->rnh_addaddr((char *)ndst, 1007 (char *) 1008 rtinfo->rti_info[RTAX_NETMASK], 1009 rnh, rt->rt_nodes); 1010 } 1011 } 1012 } 1013 1014 /* 1015 * If it still failed to go into the tree, 1016 * then un-make it (this should be a function). 1017 */ 1018 if (rn == NULL) { 1019 if (rt->rt_gwroute != NULL) 1020 rtfree(rt->rt_gwroute); 1021 IFAFREE(ifa); 1022 Free(rt_key(rt)); 1023 Free(rt); 1024 gotoerr(EEXIST); 1025 } 1026 1027 /* 1028 * If we got here from RESOLVE, then we are cloning 1029 * so clone the rest, and note that we 1030 * are a clone (and increment the parent's references) 1031 */ 1032 if (req == RTM_RESOLVE) { 1033 rt->rt_rmx = (*ret_nrt)->rt_rmx; /* copy metrics */ 1034 rt->rt_rmx.rmx_pksent = 0; /* reset packet counter */ 1035 if ((*ret_nrt)->rt_flags & 1036 (RTF_CLONING | RTF_PRCLONING)) { 1037 rt->rt_parent = *ret_nrt; 1038 (*ret_nrt)->rt_refcnt++; 1039 } 1040 } 1041 1042 /* 1043 * if this protocol has something to add to this then 1044 * allow it to do that as well. 1045 */ 1046 if (ifa->ifa_rtrequest != NULL) 1047 ifa->ifa_rtrequest(req, rt, rtinfo); 1048 1049 /* 1050 * We repeat the same procedure from rt_setgate() here because 1051 * it doesn't fire when we call it there because the node 1052 * hasn't been added to the tree yet. 1053 */ 1054 if (req == RTM_ADD && !(rt->rt_flags & RTF_HOST) && 1055 rt_mask(rt) != NULL) { 1056 struct rtfc_arg arg = { rt, rnh }; 1057 1058 rnh->rnh_walktree_from(rnh, (char *)rt_key(rt), 1059 (char *)rt_mask(rt), 1060 rt_fixchange, &arg); 1061 } 1062 1063 #ifdef ROUTE_DEBUG 1064 if (route_debug) 1065 rt_print(rtinfo, rt); 1066 #endif 1067 /* 1068 * Return the resulting rtentry, 1069 * increasing the number of references by one. 1070 */ 1071 if (ret_nrt != NULL) { 1072 rt->rt_refcnt++; 1073 *ret_nrt = rt; 1074 } 1075 break; 1076 default: 1077 error = EOPNOTSUPP; 1078 } 1079 bad: 1080 #ifdef ROUTE_DEBUG 1081 if (route_debug) { 1082 if (error) 1083 kprintf("rti %p failed error %d\n", rtinfo, error); 1084 else 1085 kprintf("rti %p succeeded\n", rtinfo); 1086 } 1087 #endif 1088 crit_exit(); 1089 return (error); 1090 } 1091 1092 /* 1093 * Called from rtrequest(RTM_DELETE, ...) to fix up the route's ``family'' 1094 * (i.e., the routes related to it by the operation of cloning). This 1095 * routine is iterated over all potential former-child-routes by way of 1096 * rnh->rnh_walktree_from() above, and those that actually are children of 1097 * the late parent (passed in as VP here) are themselves deleted. 1098 */ 1099 static int 1100 rt_fixdelete(struct radix_node *rn, void *vp) 1101 { 1102 struct rtentry *rt = (struct rtentry *)rn; 1103 struct rtentry *rt0 = vp; 1104 1105 if (rt->rt_parent == rt0 && 1106 !(rt->rt_flags & (RTF_PINNED | RTF_CLONING | RTF_PRCLONING))) { 1107 return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 1108 rt->rt_flags, NULL); 1109 } 1110 return 0; 1111 } 1112 1113 /* 1114 * This routine is called from rt_setgate() to do the analogous thing for 1115 * adds and changes. There is the added complication in this case of a 1116 * middle insert; i.e., insertion of a new network route between an older 1117 * network route and (cloned) host routes. For this reason, a simple check 1118 * of rt->rt_parent is insufficient; each candidate route must be tested 1119 * against the (mask, value) of the new route (passed as before in vp) 1120 * to see if the new route matches it. 1121 * 1122 * XXX - it may be possible to do fixdelete() for changes and reserve this 1123 * routine just for adds. I'm not sure why I thought it was necessary to do 1124 * changes this way. 1125 */ 1126 #ifdef DEBUG 1127 static int rtfcdebug = 0; 1128 #endif 1129 1130 static int 1131 rt_fixchange(struct radix_node *rn, void *vp) 1132 { 1133 struct rtentry *rt = (struct rtentry *)rn; 1134 struct rtfc_arg *ap = vp; 1135 struct rtentry *rt0 = ap->rt0; 1136 struct radix_node_head *rnh = ap->rnh; 1137 u_char *xk1, *xm1, *xk2, *xmp; 1138 int i, len, mlen; 1139 1140 #ifdef DEBUG 1141 if (rtfcdebug) 1142 kprintf("rt_fixchange: rt %p, rt0 %p\n", rt, rt0); 1143 #endif 1144 1145 if (rt->rt_parent == NULL || 1146 (rt->rt_flags & (RTF_PINNED | RTF_CLONING | RTF_PRCLONING))) { 1147 #ifdef DEBUG 1148 if (rtfcdebug) kprintf("no parent, pinned or cloning\n"); 1149 #endif 1150 return 0; 1151 } 1152 1153 if (rt->rt_parent == rt0) { 1154 #ifdef DEBUG 1155 if (rtfcdebug) kprintf("parent match\n"); 1156 #endif 1157 return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 1158 rt->rt_flags, NULL); 1159 } 1160 1161 /* 1162 * There probably is a function somewhere which does this... 1163 * if not, there should be. 1164 */ 1165 len = imin(rt_key(rt0)->sa_len, rt_key(rt)->sa_len); 1166 1167 xk1 = (u_char *)rt_key(rt0); 1168 xm1 = (u_char *)rt_mask(rt0); 1169 xk2 = (u_char *)rt_key(rt); 1170 1171 /* avoid applying a less specific route */ 1172 xmp = (u_char *)rt_mask(rt->rt_parent); 1173 mlen = rt_key(rt->rt_parent)->sa_len; 1174 if (mlen > rt_key(rt0)->sa_len) { 1175 #ifdef DEBUG 1176 if (rtfcdebug) 1177 kprintf("rt_fixchange: inserting a less " 1178 "specific route\n"); 1179 #endif 1180 return 0; 1181 } 1182 for (i = rnh->rnh_treetop->rn_offset; i < mlen; i++) { 1183 if ((xmp[i] & ~(xmp[i] ^ xm1[i])) != xmp[i]) { 1184 #ifdef DEBUG 1185 if (rtfcdebug) 1186 kprintf("rt_fixchange: inserting a less " 1187 "specific route\n"); 1188 #endif 1189 return 0; 1190 } 1191 } 1192 1193 for (i = rnh->rnh_treetop->rn_offset; i < len; i++) { 1194 if ((xk2[i] & xm1[i]) != xk1[i]) { 1195 #ifdef DEBUG 1196 if (rtfcdebug) kprintf("no match\n"); 1197 #endif 1198 return 0; 1199 } 1200 } 1201 1202 /* 1203 * OK, this node is a clone, and matches the node currently being 1204 * changed/added under the node's mask. So, get rid of it. 1205 */ 1206 #ifdef DEBUG 1207 if (rtfcdebug) kprintf("deleting\n"); 1208 #endif 1209 return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 1210 rt->rt_flags, NULL); 1211 } 1212 1213 #define ROUNDUP(a) (a>0 ? (1 + (((a) - 1) | (sizeof(long) - 1))) : sizeof(long)) 1214 1215 int 1216 rt_setgate(struct rtentry *rt0, struct sockaddr *dst, struct sockaddr *gate, 1217 boolean_t generate_report) 1218 { 1219 char *space, *oldspace; 1220 int dlen = ROUNDUP(dst->sa_len), glen = ROUNDUP(gate->sa_len); 1221 struct rtentry *rt = rt0; 1222 struct radix_node_head *rnh = rt_tables[mycpuid][dst->sa_family]; 1223 1224 /* 1225 * A host route with the destination equal to the gateway 1226 * will interfere with keeping LLINFO in the routing 1227 * table, so disallow it. 1228 */ 1229 if (((rt0->rt_flags & (RTF_HOST | RTF_GATEWAY | RTF_LLINFO)) == 1230 (RTF_HOST | RTF_GATEWAY)) && 1231 dst->sa_len == gate->sa_len && 1232 sa_equal(dst, gate)) { 1233 /* 1234 * The route might already exist if this is an RTM_CHANGE 1235 * or a routing redirect, so try to delete it. 1236 */ 1237 if (rt_key(rt0) != NULL) 1238 rtrequest(RTM_DELETE, rt_key(rt0), rt0->rt_gateway, 1239 rt_mask(rt0), rt0->rt_flags, NULL); 1240 return EADDRNOTAVAIL; 1241 } 1242 1243 /* 1244 * Both dst and gateway are stored in the same malloc'ed chunk 1245 * (If I ever get my hands on....) 1246 * if we need to malloc a new chunk, then keep the old one around 1247 * till we don't need it any more. 1248 */ 1249 if (rt->rt_gateway == NULL || glen > ROUNDUP(rt->rt_gateway->sa_len)) { 1250 oldspace = (char *)rt_key(rt); 1251 R_Malloc(space, char *, dlen + glen); 1252 if (space == NULL) 1253 return ENOBUFS; 1254 rt->rt_nodes->rn_key = space; 1255 } else { 1256 space = (char *)rt_key(rt); /* Just use the old space. */ 1257 oldspace = NULL; 1258 } 1259 1260 /* Set the gateway value. */ 1261 rt->rt_gateway = (struct sockaddr *)(space + dlen); 1262 bcopy(gate, rt->rt_gateway, glen); 1263 1264 if (oldspace != NULL) { 1265 /* 1266 * If we allocated a new chunk, preserve the original dst. 1267 * This way, rt_setgate() really just sets the gate 1268 * and leaves the dst field alone. 1269 */ 1270 bcopy(dst, space, dlen); 1271 Free(oldspace); 1272 } 1273 1274 /* 1275 * If there is already a gwroute, it's now almost definitely wrong 1276 * so drop it. 1277 */ 1278 if (rt->rt_gwroute != NULL) { 1279 RTFREE(rt->rt_gwroute); 1280 rt->rt_gwroute = NULL; 1281 } 1282 if (rt->rt_flags & RTF_GATEWAY) { 1283 /* 1284 * Cloning loop avoidance: In the presence of 1285 * protocol-cloning and bad configuration, it is 1286 * possible to get stuck in bottomless mutual recursion 1287 * (rtrequest rt_setgate rtlookup). We avoid this 1288 * by not allowing protocol-cloning to operate for 1289 * gateways (which is probably the correct choice 1290 * anyway), and avoid the resulting reference loops 1291 * by disallowing any route to run through itself as 1292 * a gateway. This is obviously mandatory when we 1293 * get rt->rt_output(). 1294 * 1295 * This breaks TTCP for hosts outside the gateway! XXX JH 1296 */ 1297 rt->rt_gwroute = _rtlookup(gate, generate_report, 1298 RTF_PRCLONING); 1299 if (rt->rt_gwroute == rt) { 1300 rt->rt_gwroute = NULL; 1301 --rt->rt_refcnt; 1302 return EDQUOT; /* failure */ 1303 } 1304 } 1305 1306 /* 1307 * This isn't going to do anything useful for host routes, so 1308 * don't bother. Also make sure we have a reasonable mask 1309 * (we don't yet have one during adds). 1310 */ 1311 if (!(rt->rt_flags & RTF_HOST) && rt_mask(rt) != NULL) { 1312 struct rtfc_arg arg = { rt, rnh }; 1313 1314 rnh->rnh_walktree_from(rnh, (char *)rt_key(rt), 1315 (char *)rt_mask(rt), 1316 rt_fixchange, &arg); 1317 } 1318 1319 return 0; 1320 } 1321 1322 static void 1323 rt_maskedcopy( 1324 struct sockaddr *src, 1325 struct sockaddr *dst, 1326 struct sockaddr *netmask) 1327 { 1328 u_char *cp1 = (u_char *)src; 1329 u_char *cp2 = (u_char *)dst; 1330 u_char *cp3 = (u_char *)netmask; 1331 u_char *cplim = cp2 + *cp3; 1332 u_char *cplim2 = cp2 + *cp1; 1333 1334 *cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */ 1335 cp3 += 2; 1336 if (cplim > cplim2) 1337 cplim = cplim2; 1338 while (cp2 < cplim) 1339 *cp2++ = *cp1++ & *cp3++; 1340 if (cp2 < cplim2) 1341 bzero(cp2, cplim2 - cp2); 1342 } 1343 1344 int 1345 rt_llroute(struct sockaddr *dst, struct rtentry *rt0, struct rtentry **drt) 1346 { 1347 struct rtentry *up_rt, *rt; 1348 1349 if (!(rt0->rt_flags & RTF_UP)) { 1350 up_rt = rtlookup(dst); 1351 if (up_rt == NULL) 1352 return (EHOSTUNREACH); 1353 up_rt->rt_refcnt--; 1354 } else 1355 up_rt = rt0; 1356 if (up_rt->rt_flags & RTF_GATEWAY) { 1357 if (up_rt->rt_gwroute == NULL) { 1358 up_rt->rt_gwroute = rtlookup(up_rt->rt_gateway); 1359 if (up_rt->rt_gwroute == NULL) 1360 return (EHOSTUNREACH); 1361 } else if (!(up_rt->rt_gwroute->rt_flags & RTF_UP)) { 1362 rtfree(up_rt->rt_gwroute); 1363 up_rt->rt_gwroute = rtlookup(up_rt->rt_gateway); 1364 if (up_rt->rt_gwroute == NULL) 1365 return (EHOSTUNREACH); 1366 } 1367 rt = up_rt->rt_gwroute; 1368 } else 1369 rt = up_rt; 1370 if (rt->rt_flags & RTF_REJECT && 1371 (rt->rt_rmx.rmx_expire == 0 || /* rt doesn't expire */ 1372 time_second < rt->rt_rmx.rmx_expire)) /* rt not expired */ 1373 return (rt->rt_flags & RTF_HOST ? EHOSTDOWN : EHOSTUNREACH); 1374 *drt = rt; 1375 return 0; 1376 } 1377 1378 static int 1379 rt_setshims(struct rtentry *rt, struct sockaddr **rt_shim){ 1380 int i; 1381 1382 for (i=0; i<3; i++) { 1383 struct sockaddr *shim = rt_shim[RTAX_MPLS1 + i]; 1384 int shimlen; 1385 1386 if (shim == NULL) 1387 break; 1388 1389 shimlen = ROUNDUP(shim->sa_len); 1390 R_Malloc(rt->rt_shim[i], struct sockaddr *, shimlen); 1391 bcopy(shim, rt->rt_shim[i], shimlen); 1392 } 1393 1394 return 0; 1395 } 1396 1397 #ifdef ROUTE_DEBUG 1398 1399 /* 1400 * Print out a route table entry 1401 */ 1402 void 1403 rt_print(struct rt_addrinfo *rtinfo, struct rtentry *rn) 1404 { 1405 kprintf("rti %p cpu %d route %p flags %08lx: ", 1406 rtinfo, mycpuid, rn, rn->rt_flags); 1407 sockaddr_print(rt_key(rn)); 1408 kprintf(" mask "); 1409 sockaddr_print(rt_mask(rn)); 1410 kprintf(" gw "); 1411 sockaddr_print(rn->rt_gateway); 1412 kprintf(" ifc \"%s\"", rn->rt_ifp ? rn->rt_ifp->if_dname : "?"); 1413 kprintf(" ifa %p\n", rn->rt_ifa); 1414 } 1415 1416 void 1417 rt_addrinfo_print(int cmd, struct rt_addrinfo *rti) 1418 { 1419 int didit = 0; 1420 int i; 1421 1422 #ifdef ROUTE_DEBUG 1423 if (cmd == RTM_DELETE && route_debug > 1) 1424 print_backtrace(); 1425 #endif 1426 1427 switch(cmd) { 1428 case RTM_ADD: 1429 kprintf("ADD "); 1430 break; 1431 case RTM_RESOLVE: 1432 kprintf("RES "); 1433 break; 1434 case RTM_DELETE: 1435 kprintf("DEL "); 1436 break; 1437 default: 1438 kprintf("C%02d ", cmd); 1439 break; 1440 } 1441 kprintf("rti %p cpu %d ", rti, mycpuid); 1442 for (i = 0; i < rti->rti_addrs; ++i) { 1443 if (rti->rti_info[i] == NULL) 1444 continue; 1445 if (didit) 1446 kprintf(" ,"); 1447 switch(i) { 1448 case RTAX_DST: 1449 kprintf("(DST "); 1450 break; 1451 case RTAX_GATEWAY: 1452 kprintf("(GWY "); 1453 break; 1454 case RTAX_NETMASK: 1455 kprintf("(MSK "); 1456 break; 1457 case RTAX_GENMASK: 1458 kprintf("(GEN "); 1459 break; 1460 case RTAX_IFP: 1461 kprintf("(IFP "); 1462 break; 1463 case RTAX_IFA: 1464 kprintf("(IFA "); 1465 break; 1466 case RTAX_AUTHOR: 1467 kprintf("(AUT "); 1468 break; 1469 case RTAX_BRD: 1470 kprintf("(BRD "); 1471 break; 1472 default: 1473 kprintf("(?%02d ", i); 1474 break; 1475 } 1476 sockaddr_print(rti->rti_info[i]); 1477 kprintf(")"); 1478 didit = 1; 1479 } 1480 kprintf("\n"); 1481 } 1482 1483 void 1484 sockaddr_print(struct sockaddr *sa) 1485 { 1486 struct sockaddr_in *sa4; 1487 struct sockaddr_in6 *sa6; 1488 int len; 1489 int i; 1490 1491 if (sa == NULL) { 1492 kprintf("NULL"); 1493 return; 1494 } 1495 1496 len = sa->sa_len - offsetof(struct sockaddr, sa_data[0]); 1497 1498 switch(sa->sa_family) { 1499 case AF_INET: 1500 case AF_INET6: 1501 default: 1502 switch(sa->sa_family) { 1503 case AF_INET: 1504 sa4 = (struct sockaddr_in *)sa; 1505 kprintf("INET %d %d.%d.%d.%d", 1506 ntohs(sa4->sin_port), 1507 (ntohl(sa4->sin_addr.s_addr) >> 24) & 255, 1508 (ntohl(sa4->sin_addr.s_addr) >> 16) & 255, 1509 (ntohl(sa4->sin_addr.s_addr) >> 8) & 255, 1510 (ntohl(sa4->sin_addr.s_addr) >> 0) & 255 1511 ); 1512 break; 1513 case AF_INET6: 1514 sa6 = (struct sockaddr_in6 *)sa; 1515 kprintf("INET6 %d %04x:%04x%04x:%04x:%04x:%04x:%04x:%04x", 1516 ntohs(sa6->sin6_port), 1517 sa6->sin6_addr.s6_addr16[0], 1518 sa6->sin6_addr.s6_addr16[1], 1519 sa6->sin6_addr.s6_addr16[2], 1520 sa6->sin6_addr.s6_addr16[3], 1521 sa6->sin6_addr.s6_addr16[4], 1522 sa6->sin6_addr.s6_addr16[5], 1523 sa6->sin6_addr.s6_addr16[6], 1524 sa6->sin6_addr.s6_addr16[7] 1525 ); 1526 break; 1527 default: 1528 kprintf("AF%d ", sa->sa_family); 1529 while (len > 0 && sa->sa_data[len-1] == 0) 1530 --len; 1531 1532 for (i = 0; i < len; ++i) { 1533 if (i) 1534 kprintf("."); 1535 kprintf("%d", (unsigned char)sa->sa_data[i]); 1536 } 1537 break; 1538 } 1539 } 1540 } 1541 1542 #endif 1543 1544 /* 1545 * Set up a routing table entry, normally for an interface. 1546 */ 1547 int 1548 rtinit(struct ifaddr *ifa, int cmd, int flags) 1549 { 1550 struct sockaddr *dst, *deldst, *netmask; 1551 struct mbuf *m = NULL; 1552 struct radix_node_head *rnh; 1553 struct radix_node *rn; 1554 struct rt_addrinfo rtinfo; 1555 int error; 1556 1557 if (flags & RTF_HOST) { 1558 dst = ifa->ifa_dstaddr; 1559 netmask = NULL; 1560 } else { 1561 dst = ifa->ifa_addr; 1562 netmask = ifa->ifa_netmask; 1563 } 1564 /* 1565 * If it's a delete, check that if it exists, it's on the correct 1566 * interface or we might scrub a route to another ifa which would 1567 * be confusing at best and possibly worse. 1568 */ 1569 if (cmd == RTM_DELETE) { 1570 /* 1571 * It's a delete, so it should already exist.. 1572 * If it's a net, mask off the host bits 1573 * (Assuming we have a mask) 1574 */ 1575 if (netmask != NULL) { 1576 m = m_get(MB_DONTWAIT, MT_SONAME); 1577 if (m == NULL) 1578 return (ENOBUFS); 1579 mbuftrackid(m, 34); 1580 deldst = mtod(m, struct sockaddr *); 1581 rt_maskedcopy(dst, deldst, netmask); 1582 dst = deldst; 1583 } 1584 /* 1585 * Look up an rtentry that is in the routing tree and 1586 * contains the correct info. 1587 */ 1588 if ((rnh = rt_tables[mycpuid][dst->sa_family]) == NULL || 1589 (rn = rnh->rnh_lookup((char *)dst, 1590 (char *)netmask, rnh)) == NULL || 1591 ((struct rtentry *)rn)->rt_ifa != ifa || 1592 !sa_equal((struct sockaddr *)rn->rn_key, dst)) { 1593 if (m != NULL) 1594 m_free(m); 1595 return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH); 1596 } 1597 /* XXX */ 1598 #if 0 1599 else { 1600 /* 1601 * One would think that as we are deleting, and we know 1602 * it doesn't exist, we could just return at this point 1603 * with an "ELSE" clause, but apparently not.. 1604 */ 1605 return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH); 1606 } 1607 #endif 1608 } 1609 /* 1610 * Do the actual request 1611 */ 1612 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 1613 rtinfo.rti_info[RTAX_DST] = dst; 1614 rtinfo.rti_info[RTAX_GATEWAY] = ifa->ifa_addr; 1615 rtinfo.rti_info[RTAX_NETMASK] = netmask; 1616 rtinfo.rti_flags = flags | ifa->ifa_flags; 1617 rtinfo.rti_ifa = ifa; 1618 error = rtrequest1_global(cmd, &rtinfo, rtinit_rtrequest_callback, ifa); 1619 if (m != NULL) 1620 m_free(m); 1621 return (error); 1622 } 1623 1624 static void 1625 rtinit_rtrequest_callback(int cmd, int error, 1626 struct rt_addrinfo *rtinfo, struct rtentry *rt, 1627 void *arg) 1628 { 1629 struct ifaddr *ifa = arg; 1630 1631 if (error == 0 && rt) { 1632 if (mycpuid == 0) { 1633 ++rt->rt_refcnt; 1634 rt_newaddrmsg(cmd, ifa, error, rt); 1635 --rt->rt_refcnt; 1636 } 1637 if (cmd == RTM_DELETE) { 1638 if (rt->rt_refcnt == 0) { 1639 ++rt->rt_refcnt; 1640 rtfree(rt); 1641 } 1642 } 1643 } 1644 } 1645 1646 struct netmsg_rts { 1647 struct netmsg netmsg; 1648 int req; 1649 struct rt_addrinfo *rtinfo; 1650 rtsearch_callback_func_t callback; 1651 void *arg; 1652 boolean_t exact_match; 1653 int found_cnt; 1654 }; 1655 1656 int 1657 rtsearch_global(int req, struct rt_addrinfo *rtinfo, 1658 rtsearch_callback_func_t callback, void *arg, 1659 boolean_t exact_match) 1660 { 1661 struct netmsg_rts msg; 1662 1663 netmsg_init(&msg.netmsg, &curthread->td_msgport, 0, 1664 rtsearch_msghandler); 1665 msg.req = req; 1666 msg.rtinfo = rtinfo; 1667 msg.callback = callback; 1668 msg.arg = arg; 1669 msg.exact_match = exact_match; 1670 msg.found_cnt = 0; 1671 return lwkt_domsg(rtable_portfn(0), &msg.netmsg.nm_lmsg, 0); 1672 } 1673 1674 static void 1675 rtsearch_msghandler(struct netmsg *netmsg) 1676 { 1677 struct netmsg_rts *msg = (void *)netmsg; 1678 struct rt_addrinfo rtinfo; 1679 struct radix_node_head *rnh; 1680 struct rtentry *rt; 1681 int nextcpu, error; 1682 1683 /* 1684 * Copy the rtinfo. We need to make sure that the original 1685 * rtinfo, which is setup by the caller, in the netmsg will 1686 * _not_ be changed; else the next CPU on the netmsg forwarding 1687 * path will see a different rtinfo than what this CPU has seen. 1688 */ 1689 rtinfo = *msg->rtinfo; 1690 1691 /* 1692 * Find the correct routing tree to use for this Address Family 1693 */ 1694 if ((rnh = rt_tables[mycpuid][rtinfo.rti_dst->sa_family]) == NULL) { 1695 if (mycpuid != 0) 1696 panic("partially initialized routing tables\n"); 1697 lwkt_replymsg(&msg->netmsg.nm_lmsg, EAFNOSUPPORT); 1698 return; 1699 } 1700 1701 /* 1702 * Correct rtinfo for the host route searching. 1703 */ 1704 if (rtinfo.rti_flags & RTF_HOST) { 1705 rtinfo.rti_netmask = NULL; 1706 rtinfo.rti_flags &= ~(RTF_CLONING | RTF_PRCLONING); 1707 } 1708 1709 rt = (struct rtentry *) 1710 rnh->rnh_lookup((char *)rtinfo.rti_dst, 1711 (char *)rtinfo.rti_netmask, rnh); 1712 1713 /* 1714 * If we are asked to do the "exact match", we need to make sure 1715 * that host route searching got a host route while a network 1716 * route searching got a network route. 1717 */ 1718 if (rt != NULL && msg->exact_match && 1719 ((rt->rt_flags ^ rtinfo.rti_flags) & RTF_HOST)) 1720 rt = NULL; 1721 1722 if (rt == NULL) { 1723 /* 1724 * No matching routes have been found, don't count this 1725 * as a critical error (here, we set 'error' to 0), just 1726 * keep moving on, since at least prcloned routes are not 1727 * duplicated onto each CPU. 1728 */ 1729 error = 0; 1730 } else { 1731 msg->found_cnt++; 1732 1733 rt->rt_refcnt++; 1734 error = msg->callback(msg->req, &rtinfo, rt, msg->arg, 1735 msg->found_cnt); 1736 rt->rt_refcnt--; 1737 1738 if (error == EJUSTRETURN) { 1739 lwkt_replymsg(&msg->netmsg.nm_lmsg, 0); 1740 return; 1741 } 1742 } 1743 1744 nextcpu = mycpuid + 1; 1745 if (error) { 1746 KKASSERT(msg->found_cnt > 0); 1747 1748 /* 1749 * Under following cases, unrecoverable error has 1750 * not occured: 1751 * o Request is RTM_GET 1752 * o The first time that we find the route, but the 1753 * modification fails. 1754 */ 1755 if (msg->req != RTM_GET && msg->found_cnt > 1) { 1756 panic("rtsearch_msghandler: unrecoverable error " 1757 "cpu %d", mycpuid); 1758 } 1759 lwkt_replymsg(&msg->netmsg.nm_lmsg, error); 1760 } else if (nextcpu < ncpus) { 1761 lwkt_forwardmsg(rtable_portfn(nextcpu), &msg->netmsg.nm_lmsg); 1762 } else { 1763 if (msg->found_cnt == 0) { 1764 /* The requested route was never seen ... */ 1765 error = ESRCH; 1766 } 1767 lwkt_replymsg(&msg->netmsg.nm_lmsg, error); 1768 } 1769 } 1770 1771 int 1772 rtmask_add_global(struct sockaddr *mask) 1773 { 1774 struct netmsg nmsg; 1775 1776 netmsg_init(&nmsg, &curthread->td_msgport, 0, 1777 rtmask_add_msghandler); 1778 nmsg.nm_lmsg.u.ms_resultp = mask; 1779 1780 return lwkt_domsg(rtable_portfn(0), &nmsg.nm_lmsg, 0); 1781 } 1782 1783 struct sockaddr * 1784 _rtmask_lookup(struct sockaddr *mask, boolean_t search) 1785 { 1786 struct radix_node *n; 1787 1788 #define clen(s) (*(u_char *)(s)) 1789 n = rn_addmask((char *)mask, search, 1); 1790 if (n != NULL && 1791 mask->sa_len >= clen(n->rn_key) && 1792 bcmp((char *)mask + 1, 1793 (char *)n->rn_key + 1, clen(n->rn_key) - 1) == 0) { 1794 return (struct sockaddr *)n->rn_key; 1795 } else { 1796 return NULL; 1797 } 1798 #undef clen 1799 } 1800 1801 static void 1802 rtmask_add_msghandler(struct netmsg *nmsg) 1803 { 1804 struct lwkt_msg *lmsg = &nmsg->nm_lmsg; 1805 struct sockaddr *mask = lmsg->u.ms_resultp; 1806 int error = 0, nextcpu; 1807 1808 if (rtmask_lookup(mask) == NULL) 1809 error = ENOBUFS; 1810 1811 nextcpu = mycpuid + 1; 1812 if (!error && nextcpu < ncpus) 1813 lwkt_forwardmsg(rtable_portfn(nextcpu), lmsg); 1814 else 1815 lwkt_replymsg(lmsg, error); 1816 } 1817 1818 /* This must be before ip6_init2(), which is now SI_ORDER_MIDDLE */ 1819 SYSINIT(route, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0); 1820