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(netmsg_t msg); 118 static void rtrequest1_msghandler(netmsg_t msg); 119 #endif 120 static void rtsearch_msghandler(netmsg_t msg); 121 static void rtmask_add_msghandler(netmsg_t msg); 122 123 static int rt_setshims(struct rtentry *, struct sockaddr **); 124 125 SYSCTL_NODE(_net, OID_AUTO, route, CTLFLAG_RW, 0, "Routing"); 126 127 #ifdef ROUTE_DEBUG 128 static int route_debug = 1; 129 SYSCTL_INT(_net_route, OID_AUTO, route_debug, CTLFLAG_RW, 130 &route_debug, 0, ""); 131 #endif 132 133 int route_assert_owner_access = 0; 134 SYSCTL_INT(_net_route, OID_AUTO, assert_owner_access, CTLFLAG_RW, 135 &route_assert_owner_access, 0, ""); 136 137 /* 138 * Initialize the route table(s) for protocol domains and 139 * create a helper thread which will be responsible for updating 140 * route table entries on each cpu. 141 */ 142 void 143 route_init(void) 144 { 145 int cpu; 146 thread_t rtd; 147 148 for (cpu = 0; cpu < ncpus; ++cpu) 149 bzero(&rtstatistics_percpu[cpu], sizeof(struct rtstatistics)); 150 rn_init(); /* initialize all zeroes, all ones, mask table */ 151 rtable_init(); /* call dom_rtattach() on each cpu */ 152 153 for (cpu = 0; cpu < ncpus; cpu++) { 154 lwkt_create(rtable_service_loop, NULL, &rtd, NULL, 155 0, cpu, "rtable_cpu %d", cpu); 156 rt_ports[cpu] = &rtd->td_msgport; 157 } 158 } 159 160 static void 161 rtable_init_oncpu(netmsg_t msg) 162 { 163 struct domain *dom; 164 int cpu = mycpuid; 165 166 SLIST_FOREACH(dom, &domains, dom_next) { 167 if (dom->dom_rtattach) { 168 dom->dom_rtattach( 169 (void **)&rt_tables[cpu][dom->dom_family], 170 dom->dom_rtoffset); 171 } 172 } 173 ifnet_forwardmsg(&msg->lmsg, cpu + 1); 174 } 175 176 static void 177 rtable_init(void) 178 { 179 struct netmsg_base msg; 180 181 netmsg_init(&msg, NULL, &curthread->td_msgport, 0, rtable_init_oncpu); 182 ifnet_domsg(&msg.lmsg, 0); 183 } 184 185 /* 186 * Our per-cpu table management protocol thread. All route table operations 187 * are sequentially chained through all cpus starting at cpu #0 in order to 188 * maintain duplicate route tables on each cpu. Having a spearate route 189 * table management thread allows the protocol and interrupt threads to 190 * issue route table changes. 191 */ 192 static void 193 rtable_service_loop(void *dummy __unused) 194 { 195 netmsg_base_t msg; 196 thread_t td = curthread; 197 198 while ((msg = lwkt_waitport(&td->td_msgport, 0)) != NULL) { 199 msg->nm_dispatch((netmsg_t)msg); 200 } 201 } 202 203 /* 204 * Routing statistics. 205 */ 206 #ifdef SMP 207 static int 208 sysctl_rtstatistics(SYSCTL_HANDLER_ARGS) 209 { 210 int cpu, error = 0; 211 212 for (cpu = 0; cpu < ncpus; ++cpu) { 213 if ((error = SYSCTL_OUT(req, &rtstatistics_percpu[cpu], 214 sizeof(struct rtstatistics)))) 215 break; 216 if ((error = SYSCTL_IN(req, &rtstatistics_percpu[cpu], 217 sizeof(struct rtstatistics)))) 218 break; 219 } 220 221 return (error); 222 } 223 SYSCTL_PROC(_net_route, OID_AUTO, stats, (CTLTYPE_OPAQUE|CTLFLAG_RW), 224 0, 0, sysctl_rtstatistics, "S,rtstatistics", "Routing statistics"); 225 #else 226 SYSCTL_STRUCT(_net_route, OID_AUTO, stats, CTLFLAG_RW, &rtstat, rtstatistics, 227 "Routing statistics"); 228 #endif 229 230 /* 231 * Packet routing routines. 232 */ 233 234 /* 235 * Look up and fill in the "ro_rt" rtentry field in a route structure given 236 * an address in the "ro_dst" field. Always send a report on a miss and 237 * always clone routes. 238 */ 239 void 240 rtalloc(struct route *ro) 241 { 242 rtalloc_ign(ro, 0UL); 243 } 244 245 /* 246 * Look up and fill in the "ro_rt" rtentry field in a route structure given 247 * an address in the "ro_dst" field. Always send a report on a miss and 248 * optionally clone routes when RTF_CLONING or RTF_PRCLONING are not being 249 * ignored. 250 */ 251 void 252 rtalloc_ign(struct route *ro, u_long ignoreflags) 253 { 254 if (ro->ro_rt != NULL) { 255 if (ro->ro_rt->rt_ifp != NULL && ro->ro_rt->rt_flags & RTF_UP) 256 return; 257 rtfree(ro->ro_rt); 258 ro->ro_rt = NULL; 259 } 260 ro->ro_rt = _rtlookup(&ro->ro_dst, RTL_REPORTMSG, ignoreflags); 261 } 262 263 /* 264 * Look up the route that matches the given "dst" address. 265 * 266 * Route lookup can have the side-effect of creating and returning 267 * a cloned route instead when "dst" matches a cloning route and the 268 * RTF_CLONING and RTF_PRCLONING flags are not being ignored. 269 * 270 * Any route returned has its reference count incremented. 271 */ 272 struct rtentry * 273 _rtlookup(struct sockaddr *dst, boolean_t generate_report, u_long ignore) 274 { 275 struct radix_node_head *rnh = rt_tables[mycpuid][dst->sa_family]; 276 struct rtentry *rt; 277 278 if (rnh == NULL) 279 goto unreach; 280 281 /* 282 * Look up route in the radix tree. 283 */ 284 rt = (struct rtentry *) rnh->rnh_matchaddr((char *)dst, rnh); 285 if (rt == NULL) 286 goto unreach; 287 288 /* 289 * Handle cloning routes. 290 */ 291 if ((rt->rt_flags & ~ignore & (RTF_CLONING | RTF_PRCLONING)) != 0) { 292 struct rtentry *clonedroute; 293 int error; 294 295 clonedroute = rt; /* copy in/copy out parameter */ 296 error = rtrequest(RTM_RESOLVE, dst, NULL, NULL, 0, 297 &clonedroute); /* clone the route */ 298 if (error != 0) { /* cloning failed */ 299 if (generate_report) 300 rt_dstmsg(RTM_MISS, dst, error); 301 rt->rt_refcnt++; 302 return (rt); /* return the uncloned route */ 303 } 304 if (generate_report) { 305 if (clonedroute->rt_flags & RTF_XRESOLVE) 306 rt_dstmsg(RTM_RESOLVE, dst, 0); 307 else 308 rt_rtmsg(RTM_ADD, clonedroute, 309 clonedroute->rt_ifp, 0); 310 } 311 return (clonedroute); /* return cloned route */ 312 } 313 314 /* 315 * Increment the reference count of the matched route and return. 316 */ 317 rt->rt_refcnt++; 318 return (rt); 319 320 unreach: 321 rtstat.rts_unreach++; 322 if (generate_report) 323 rt_dstmsg(RTM_MISS, dst, 0); 324 return (NULL); 325 } 326 327 void 328 rtfree(struct rtentry *rt) 329 { 330 if (rt->rt_cpuid == mycpuid) 331 rtfree_oncpu(rt); 332 else 333 rtfree_remote(rt, 1); 334 } 335 336 void 337 rtfree_oncpu(struct rtentry *rt) 338 { 339 KKASSERT(rt->rt_cpuid == mycpuid); 340 KASSERT(rt->rt_refcnt > 0, ("rtfree: rt_refcnt %ld", rt->rt_refcnt)); 341 342 --rt->rt_refcnt; 343 if (rt->rt_refcnt == 0) { 344 struct radix_node_head *rnh = 345 rt_tables[mycpuid][rt_key(rt)->sa_family]; 346 347 if (rnh->rnh_close) 348 rnh->rnh_close((struct radix_node *)rt, rnh); 349 if (!(rt->rt_flags & RTF_UP)) { 350 /* deallocate route */ 351 if (rt->rt_ifa != NULL) 352 IFAFREE(rt->rt_ifa); 353 if (rt->rt_parent != NULL) 354 RTFREE(rt->rt_parent); /* recursive call! */ 355 Free(rt_key(rt)); 356 Free(rt); 357 } 358 } 359 } 360 361 static void 362 rtfree_remote_dispatch(netmsg_t msg) 363 { 364 struct lwkt_msg *lmsg = &msg->lmsg; 365 struct rtentry *rt = lmsg->u.ms_resultp; 366 367 rtfree_oncpu(rt); 368 lwkt_replymsg(lmsg, 0); 369 } 370 371 void 372 rtfree_remote(struct rtentry *rt, int allow_panic) 373 { 374 struct netmsg_base msg; 375 struct lwkt_msg *lmsg; 376 377 KKASSERT(rt->rt_cpuid != mycpuid); 378 379 if (route_assert_owner_access && allow_panic) { 380 panic("rt remote free rt_cpuid %d, mycpuid %d\n", 381 rt->rt_cpuid, mycpuid); 382 } else { 383 kprintf("rt remote free rt_cpuid %d, mycpuid %d\n", 384 rt->rt_cpuid, mycpuid); 385 print_backtrace(-1); 386 } 387 388 netmsg_init(&msg, NULL, &curthread->td_msgport, 389 0, rtfree_remote_dispatch); 390 lmsg = &msg.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_base base; 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.base, NULL, &curthread->td_msgport, 528 0, 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.base.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(netmsg_t msg) 550 { 551 struct netmsg_rtredirect *rmsg = (void *)msg; 552 int nextcpu; 553 554 rtredirect_oncpu(rmsg->dst, rmsg->gateway, rmsg->netmask, 555 rmsg->flags, rmsg->src); 556 nextcpu = mycpuid + 1; 557 if (nextcpu < ncpus) 558 lwkt_forwardmsg(rtable_portfn(nextcpu), &msg->lmsg); 559 else 560 lwkt_replymsg(&msg->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_base base; 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.base, NULL, &curthread->td_msgport, 748 0, rtrequest1_msghandler); 749 msg.base.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.base.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(netmsg_t msg) 776 { 777 struct netmsg_rtq *rmsg = (void *)msg; 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 = *rmsg->rtinfo; 790 791 error = rtrequest1(rmsg->req, &rtinfo, &rt); 792 if (rt) 793 --rt->rt_refcnt; 794 if (rmsg->callback) 795 rmsg->callback(rmsg->req, error, &rtinfo, rt, rmsg->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 (rmsg->base.lmsg.ms_error < 0 || error == 0) 804 rmsg->base.lmsg.ms_error = error; 805 806 nextcpu = mycpuid + 1; 807 if (error && rmsg->req != RTM_DELETE) { 808 if (mycpuid != 0) { 809 panic("rtrequest1_msghandler: rtrequest table " 810 "error was cpu%d, err %d\n", mycpuid, error); 811 } 812 lwkt_replymsg(&rmsg->base.lmsg, error); 813 } else if (nextcpu < ncpus) { 814 lwkt_forwardmsg(rtable_portfn(nextcpu), &rmsg->base.lmsg); 815 } else { 816 lwkt_replymsg(&rmsg->base.lmsg, rmsg->base.lmsg.ms_error); 817 } 818 } 819 820 #endif 821 822 int 823 rtrequest1(int req, struct rt_addrinfo *rtinfo, struct rtentry **ret_nrt) 824 { 825 struct sockaddr *dst = rtinfo->rti_info[RTAX_DST]; 826 struct rtentry *rt; 827 struct radix_node *rn; 828 struct radix_node_head *rnh; 829 struct ifaddr *ifa; 830 struct sockaddr *ndst; 831 boolean_t reportmsg; 832 int error = 0; 833 834 #define gotoerr(x) { error = x ; goto bad; } 835 836 #ifdef ROUTE_DEBUG 837 if (route_debug) 838 rt_addrinfo_print(req, rtinfo); 839 #endif 840 841 crit_enter(); 842 /* 843 * Find the correct routing tree to use for this Address Family 844 */ 845 if ((rnh = rt_tables[mycpuid][dst->sa_family]) == NULL) 846 gotoerr(EAFNOSUPPORT); 847 848 /* 849 * If we are adding a host route then we don't want to put 850 * a netmask in the tree, nor do we want to clone it. 851 */ 852 if (rtinfo->rti_flags & RTF_HOST) { 853 rtinfo->rti_info[RTAX_NETMASK] = NULL; 854 rtinfo->rti_flags &= ~(RTF_CLONING | RTF_PRCLONING); 855 } 856 857 switch (req) { 858 case RTM_DELETE: 859 /* Remove the item from the tree. */ 860 rn = rnh->rnh_deladdr((char *)rtinfo->rti_info[RTAX_DST], 861 (char *)rtinfo->rti_info[RTAX_NETMASK], 862 rnh); 863 if (rn == NULL) 864 gotoerr(ESRCH); 865 KASSERT(!(rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)), 866 ("rnh_deladdr returned flags 0x%x", rn->rn_flags)); 867 rt = (struct rtentry *)rn; 868 869 /* ref to prevent a deletion race */ 870 ++rt->rt_refcnt; 871 872 /* Free any routes cloned from this one. */ 873 if ((rt->rt_flags & (RTF_CLONING | RTF_PRCLONING)) && 874 rt_mask(rt) != NULL) { 875 rnh->rnh_walktree_from(rnh, (char *)rt_key(rt), 876 (char *)rt_mask(rt), 877 rt_fixdelete, rt); 878 } 879 880 if (rt->rt_gwroute != NULL) { 881 RTFREE(rt->rt_gwroute); 882 rt->rt_gwroute = NULL; 883 } 884 885 /* 886 * NB: RTF_UP must be set during the search above, 887 * because we might delete the last ref, causing 888 * rt to get freed prematurely. 889 */ 890 rt->rt_flags &= ~RTF_UP; 891 892 #ifdef ROUTE_DEBUG 893 if (route_debug) 894 rt_print(rtinfo, rt); 895 #endif 896 897 /* Give the protocol a chance to keep things in sync. */ 898 if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest) 899 ifa->ifa_rtrequest(RTM_DELETE, rt, rtinfo); 900 901 /* 902 * If the caller wants it, then it can have it, 903 * but it's up to it to free the rtentry as we won't be 904 * doing it. 905 */ 906 KASSERT(rt->rt_refcnt >= 0, 907 ("rtrequest1(DELETE): refcnt %ld", rt->rt_refcnt)); 908 if (ret_nrt != NULL) { 909 /* leave ref intact for return */ 910 *ret_nrt = rt; 911 } else { 912 /* deref / attempt to destroy */ 913 rtfree(rt); 914 } 915 break; 916 917 case RTM_RESOLVE: 918 if (ret_nrt == NULL || (rt = *ret_nrt) == NULL) 919 gotoerr(EINVAL); 920 ifa = rt->rt_ifa; 921 rtinfo->rti_flags = 922 rt->rt_flags & ~(RTF_CLONING | RTF_PRCLONING | RTF_STATIC); 923 rtinfo->rti_flags |= RTF_WASCLONED; 924 rtinfo->rti_info[RTAX_GATEWAY] = rt->rt_gateway; 925 if ((rtinfo->rti_info[RTAX_NETMASK] = rt->rt_genmask) == NULL) 926 rtinfo->rti_flags |= RTF_HOST; 927 rtinfo->rti_info[RTAX_MPLS1] = rt->rt_shim[0]; 928 rtinfo->rti_info[RTAX_MPLS2] = rt->rt_shim[1]; 929 rtinfo->rti_info[RTAX_MPLS3] = rt->rt_shim[2]; 930 goto makeroute; 931 932 case RTM_ADD: 933 KASSERT(!(rtinfo->rti_flags & RTF_GATEWAY) || 934 rtinfo->rti_info[RTAX_GATEWAY] != NULL, 935 ("rtrequest: GATEWAY but no gateway")); 936 937 if (rtinfo->rti_ifa == NULL && (error = rt_getifa(rtinfo))) 938 gotoerr(error); 939 ifa = rtinfo->rti_ifa; 940 makeroute: 941 R_Malloc(rt, struct rtentry *, sizeof(struct rtentry)); 942 if (rt == NULL) { 943 if (req == RTM_ADD) { 944 kprintf("rtrequest1: alloc rtentry failed on " 945 "cpu%d\n", mycpuid); 946 } 947 gotoerr(ENOBUFS); 948 } 949 bzero(rt, sizeof(struct rtentry)); 950 rt->rt_flags = RTF_UP | rtinfo->rti_flags; 951 rt->rt_cpuid = mycpuid; 952 953 if (mycpuid != 0 && req == RTM_ADD) { 954 /* For RTM_ADD, we have already sent rtmsg on CPU0. */ 955 reportmsg = RTL_DONTREPORT; 956 } else { 957 /* 958 * For RTM_ADD, we only send rtmsg on CPU0. 959 * For RTM_RESOLVE, we always send rtmsg. XXX 960 */ 961 reportmsg = RTL_REPORTMSG; 962 } 963 error = rt_setgate(rt, dst, rtinfo->rti_info[RTAX_GATEWAY], 964 reportmsg); 965 if (error != 0) { 966 Free(rt); 967 gotoerr(error); 968 } 969 970 ndst = rt_key(rt); 971 if (rtinfo->rti_info[RTAX_NETMASK] != NULL) 972 rt_maskedcopy(dst, ndst, 973 rtinfo->rti_info[RTAX_NETMASK]); 974 else 975 bcopy(dst, ndst, dst->sa_len); 976 977 if (rtinfo->rti_info[RTAX_MPLS1] != NULL) 978 rt_setshims(rt, rtinfo->rti_info); 979 980 /* 981 * Note that we now have a reference to the ifa. 982 * This moved from below so that rnh->rnh_addaddr() can 983 * examine the ifa and ifa->ifa_ifp if it so desires. 984 */ 985 IFAREF(ifa); 986 rt->rt_ifa = ifa; 987 rt->rt_ifp = ifa->ifa_ifp; 988 /* XXX mtu manipulation will be done in rnh_addaddr -- itojun */ 989 990 rn = rnh->rnh_addaddr((char *)ndst, 991 (char *)rtinfo->rti_info[RTAX_NETMASK], 992 rnh, rt->rt_nodes); 993 if (rn == NULL) { 994 struct rtentry *oldrt; 995 996 /* 997 * We already have one of these in the tree. 998 * We do a special hack: if the old route was 999 * cloned, then we blow it away and try 1000 * re-inserting the new one. 1001 */ 1002 oldrt = rtpurelookup(ndst); 1003 if (oldrt != NULL) { 1004 --oldrt->rt_refcnt; 1005 if (oldrt->rt_flags & RTF_WASCLONED) { 1006 rtrequest(RTM_DELETE, rt_key(oldrt), 1007 oldrt->rt_gateway, 1008 rt_mask(oldrt), 1009 oldrt->rt_flags, NULL); 1010 rn = rnh->rnh_addaddr((char *)ndst, 1011 (char *) 1012 rtinfo->rti_info[RTAX_NETMASK], 1013 rnh, rt->rt_nodes); 1014 } 1015 } 1016 } 1017 1018 /* 1019 * If it still failed to go into the tree, 1020 * then un-make it (this should be a function). 1021 */ 1022 if (rn == NULL) { 1023 if (rt->rt_gwroute != NULL) 1024 rtfree(rt->rt_gwroute); 1025 IFAFREE(ifa); 1026 Free(rt_key(rt)); 1027 Free(rt); 1028 gotoerr(EEXIST); 1029 } 1030 1031 /* 1032 * If we got here from RESOLVE, then we are cloning 1033 * so clone the rest, and note that we 1034 * are a clone (and increment the parent's references) 1035 */ 1036 if (req == RTM_RESOLVE) { 1037 rt->rt_rmx = (*ret_nrt)->rt_rmx; /* copy metrics */ 1038 rt->rt_rmx.rmx_pksent = 0; /* reset packet counter */ 1039 if ((*ret_nrt)->rt_flags & 1040 (RTF_CLONING | RTF_PRCLONING)) { 1041 rt->rt_parent = *ret_nrt; 1042 (*ret_nrt)->rt_refcnt++; 1043 } 1044 } 1045 1046 /* 1047 * if this protocol has something to add to this then 1048 * allow it to do that as well. 1049 */ 1050 if (ifa->ifa_rtrequest != NULL) 1051 ifa->ifa_rtrequest(req, rt, rtinfo); 1052 1053 /* 1054 * We repeat the same procedure from rt_setgate() here because 1055 * it doesn't fire when we call it there because the node 1056 * hasn't been added to the tree yet. 1057 */ 1058 if (req == RTM_ADD && !(rt->rt_flags & RTF_HOST) && 1059 rt_mask(rt) != NULL) { 1060 struct rtfc_arg arg = { rt, rnh }; 1061 1062 rnh->rnh_walktree_from(rnh, (char *)rt_key(rt), 1063 (char *)rt_mask(rt), 1064 rt_fixchange, &arg); 1065 } 1066 1067 #ifdef ROUTE_DEBUG 1068 if (route_debug) 1069 rt_print(rtinfo, rt); 1070 #endif 1071 /* 1072 * Return the resulting rtentry, 1073 * increasing the number of references by one. 1074 */ 1075 if (ret_nrt != NULL) { 1076 rt->rt_refcnt++; 1077 *ret_nrt = rt; 1078 } 1079 break; 1080 default: 1081 error = EOPNOTSUPP; 1082 } 1083 bad: 1084 #ifdef ROUTE_DEBUG 1085 if (route_debug) { 1086 if (error) 1087 kprintf("rti %p failed error %d\n", rtinfo, error); 1088 else 1089 kprintf("rti %p succeeded\n", rtinfo); 1090 } 1091 #endif 1092 crit_exit(); 1093 return (error); 1094 } 1095 1096 /* 1097 * Called from rtrequest(RTM_DELETE, ...) to fix up the route's ``family'' 1098 * (i.e., the routes related to it by the operation of cloning). This 1099 * routine is iterated over all potential former-child-routes by way of 1100 * rnh->rnh_walktree_from() above, and those that actually are children of 1101 * the late parent (passed in as VP here) are themselves deleted. 1102 */ 1103 static int 1104 rt_fixdelete(struct radix_node *rn, void *vp) 1105 { 1106 struct rtentry *rt = (struct rtentry *)rn; 1107 struct rtentry *rt0 = vp; 1108 1109 if (rt->rt_parent == rt0 && 1110 !(rt->rt_flags & (RTF_PINNED | RTF_CLONING | RTF_PRCLONING))) { 1111 return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 1112 rt->rt_flags, NULL); 1113 } 1114 return 0; 1115 } 1116 1117 /* 1118 * This routine is called from rt_setgate() to do the analogous thing for 1119 * adds and changes. There is the added complication in this case of a 1120 * middle insert; i.e., insertion of a new network route between an older 1121 * network route and (cloned) host routes. For this reason, a simple check 1122 * of rt->rt_parent is insufficient; each candidate route must be tested 1123 * against the (mask, value) of the new route (passed as before in vp) 1124 * to see if the new route matches it. 1125 * 1126 * XXX - it may be possible to do fixdelete() for changes and reserve this 1127 * routine just for adds. I'm not sure why I thought it was necessary to do 1128 * changes this way. 1129 */ 1130 #ifdef DEBUG 1131 static int rtfcdebug = 0; 1132 #endif 1133 1134 static int 1135 rt_fixchange(struct radix_node *rn, void *vp) 1136 { 1137 struct rtentry *rt = (struct rtentry *)rn; 1138 struct rtfc_arg *ap = vp; 1139 struct rtentry *rt0 = ap->rt0; 1140 struct radix_node_head *rnh = ap->rnh; 1141 u_char *xk1, *xm1, *xk2, *xmp; 1142 int i, len, mlen; 1143 1144 #ifdef DEBUG 1145 if (rtfcdebug) 1146 kprintf("rt_fixchange: rt %p, rt0 %p\n", rt, rt0); 1147 #endif 1148 1149 if (rt->rt_parent == NULL || 1150 (rt->rt_flags & (RTF_PINNED | RTF_CLONING | RTF_PRCLONING))) { 1151 #ifdef DEBUG 1152 if (rtfcdebug) kprintf("no parent, pinned or cloning\n"); 1153 #endif 1154 return 0; 1155 } 1156 1157 if (rt->rt_parent == rt0) { 1158 #ifdef DEBUG 1159 if (rtfcdebug) kprintf("parent match\n"); 1160 #endif 1161 return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 1162 rt->rt_flags, NULL); 1163 } 1164 1165 /* 1166 * There probably is a function somewhere which does this... 1167 * if not, there should be. 1168 */ 1169 len = imin(rt_key(rt0)->sa_len, rt_key(rt)->sa_len); 1170 1171 xk1 = (u_char *)rt_key(rt0); 1172 xm1 = (u_char *)rt_mask(rt0); 1173 xk2 = (u_char *)rt_key(rt); 1174 1175 /* avoid applying a less specific route */ 1176 xmp = (u_char *)rt_mask(rt->rt_parent); 1177 mlen = rt_key(rt->rt_parent)->sa_len; 1178 if (mlen > rt_key(rt0)->sa_len) { 1179 #ifdef DEBUG 1180 if (rtfcdebug) 1181 kprintf("rt_fixchange: inserting a less " 1182 "specific route\n"); 1183 #endif 1184 return 0; 1185 } 1186 for (i = rnh->rnh_treetop->rn_offset; i < mlen; i++) { 1187 if ((xmp[i] & ~(xmp[i] ^ xm1[i])) != xmp[i]) { 1188 #ifdef DEBUG 1189 if (rtfcdebug) 1190 kprintf("rt_fixchange: inserting a less " 1191 "specific route\n"); 1192 #endif 1193 return 0; 1194 } 1195 } 1196 1197 for (i = rnh->rnh_treetop->rn_offset; i < len; i++) { 1198 if ((xk2[i] & xm1[i]) != xk1[i]) { 1199 #ifdef DEBUG 1200 if (rtfcdebug) kprintf("no match\n"); 1201 #endif 1202 return 0; 1203 } 1204 } 1205 1206 /* 1207 * OK, this node is a clone, and matches the node currently being 1208 * changed/added under the node's mask. So, get rid of it. 1209 */ 1210 #ifdef DEBUG 1211 if (rtfcdebug) kprintf("deleting\n"); 1212 #endif 1213 return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 1214 rt->rt_flags, NULL); 1215 } 1216 1217 #define ROUNDUP(a) (a>0 ? (1 + (((a) - 1) | (sizeof(long) - 1))) : sizeof(long)) 1218 1219 int 1220 rt_setgate(struct rtentry *rt0, struct sockaddr *dst, struct sockaddr *gate, 1221 boolean_t generate_report) 1222 { 1223 char *space, *oldspace; 1224 int dlen = ROUNDUP(dst->sa_len), glen = ROUNDUP(gate->sa_len); 1225 struct rtentry *rt = rt0; 1226 struct radix_node_head *rnh = rt_tables[mycpuid][dst->sa_family]; 1227 1228 /* 1229 * A host route with the destination equal to the gateway 1230 * will interfere with keeping LLINFO in the routing 1231 * table, so disallow it. 1232 */ 1233 if (((rt0->rt_flags & (RTF_HOST | RTF_GATEWAY | RTF_LLINFO)) == 1234 (RTF_HOST | RTF_GATEWAY)) && 1235 dst->sa_len == gate->sa_len && 1236 sa_equal(dst, gate)) { 1237 /* 1238 * The route might already exist if this is an RTM_CHANGE 1239 * or a routing redirect, so try to delete it. 1240 */ 1241 if (rt_key(rt0) != NULL) 1242 rtrequest(RTM_DELETE, rt_key(rt0), rt0->rt_gateway, 1243 rt_mask(rt0), rt0->rt_flags, NULL); 1244 return EADDRNOTAVAIL; 1245 } 1246 1247 /* 1248 * Both dst and gateway are stored in the same malloc'ed chunk 1249 * (If I ever get my hands on....) 1250 * if we need to malloc a new chunk, then keep the old one around 1251 * till we don't need it any more. 1252 */ 1253 if (rt->rt_gateway == NULL || glen > ROUNDUP(rt->rt_gateway->sa_len)) { 1254 oldspace = (char *)rt_key(rt); 1255 R_Malloc(space, char *, dlen + glen); 1256 if (space == NULL) 1257 return ENOBUFS; 1258 rt->rt_nodes->rn_key = space; 1259 } else { 1260 space = (char *)rt_key(rt); /* Just use the old space. */ 1261 oldspace = NULL; 1262 } 1263 1264 /* Set the gateway value. */ 1265 rt->rt_gateway = (struct sockaddr *)(space + dlen); 1266 bcopy(gate, rt->rt_gateway, glen); 1267 1268 if (oldspace != NULL) { 1269 /* 1270 * If we allocated a new chunk, preserve the original dst. 1271 * This way, rt_setgate() really just sets the gate 1272 * and leaves the dst field alone. 1273 */ 1274 bcopy(dst, space, dlen); 1275 Free(oldspace); 1276 } 1277 1278 /* 1279 * If there is already a gwroute, it's now almost definitely wrong 1280 * so drop it. 1281 */ 1282 if (rt->rt_gwroute != NULL) { 1283 RTFREE(rt->rt_gwroute); 1284 rt->rt_gwroute = NULL; 1285 } 1286 if (rt->rt_flags & RTF_GATEWAY) { 1287 /* 1288 * Cloning loop avoidance: In the presence of 1289 * protocol-cloning and bad configuration, it is 1290 * possible to get stuck in bottomless mutual recursion 1291 * (rtrequest rt_setgate rtlookup). We avoid this 1292 * by not allowing protocol-cloning to operate for 1293 * gateways (which is probably the correct choice 1294 * anyway), and avoid the resulting reference loops 1295 * by disallowing any route to run through itself as 1296 * a gateway. This is obviously mandatory when we 1297 * get rt->rt_output(). 1298 * 1299 * This breaks TTCP for hosts outside the gateway! XXX JH 1300 */ 1301 rt->rt_gwroute = _rtlookup(gate, generate_report, 1302 RTF_PRCLONING); 1303 if (rt->rt_gwroute == rt) { 1304 rt->rt_gwroute = NULL; 1305 --rt->rt_refcnt; 1306 return EDQUOT; /* failure */ 1307 } 1308 } 1309 1310 /* 1311 * This isn't going to do anything useful for host routes, so 1312 * don't bother. Also make sure we have a reasonable mask 1313 * (we don't yet have one during adds). 1314 */ 1315 if (!(rt->rt_flags & RTF_HOST) && rt_mask(rt) != NULL) { 1316 struct rtfc_arg arg = { rt, rnh }; 1317 1318 rnh->rnh_walktree_from(rnh, (char *)rt_key(rt), 1319 (char *)rt_mask(rt), 1320 rt_fixchange, &arg); 1321 } 1322 1323 return 0; 1324 } 1325 1326 static void 1327 rt_maskedcopy( 1328 struct sockaddr *src, 1329 struct sockaddr *dst, 1330 struct sockaddr *netmask) 1331 { 1332 u_char *cp1 = (u_char *)src; 1333 u_char *cp2 = (u_char *)dst; 1334 u_char *cp3 = (u_char *)netmask; 1335 u_char *cplim = cp2 + *cp3; 1336 u_char *cplim2 = cp2 + *cp1; 1337 1338 *cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */ 1339 cp3 += 2; 1340 if (cplim > cplim2) 1341 cplim = cplim2; 1342 while (cp2 < cplim) 1343 *cp2++ = *cp1++ & *cp3++; 1344 if (cp2 < cplim2) 1345 bzero(cp2, cplim2 - cp2); 1346 } 1347 1348 int 1349 rt_llroute(struct sockaddr *dst, struct rtentry *rt0, struct rtentry **drt) 1350 { 1351 struct rtentry *up_rt, *rt; 1352 1353 if (!(rt0->rt_flags & RTF_UP)) { 1354 up_rt = rtlookup(dst); 1355 if (up_rt == NULL) 1356 return (EHOSTUNREACH); 1357 up_rt->rt_refcnt--; 1358 } else 1359 up_rt = rt0; 1360 if (up_rt->rt_flags & RTF_GATEWAY) { 1361 if (up_rt->rt_gwroute == NULL) { 1362 up_rt->rt_gwroute = rtlookup(up_rt->rt_gateway); 1363 if (up_rt->rt_gwroute == NULL) 1364 return (EHOSTUNREACH); 1365 } else if (!(up_rt->rt_gwroute->rt_flags & RTF_UP)) { 1366 rtfree(up_rt->rt_gwroute); 1367 up_rt->rt_gwroute = rtlookup(up_rt->rt_gateway); 1368 if (up_rt->rt_gwroute == NULL) 1369 return (EHOSTUNREACH); 1370 } 1371 rt = up_rt->rt_gwroute; 1372 } else 1373 rt = up_rt; 1374 if (rt->rt_flags & RTF_REJECT && 1375 (rt->rt_rmx.rmx_expire == 0 || /* rt doesn't expire */ 1376 time_second < rt->rt_rmx.rmx_expire)) /* rt not expired */ 1377 return (rt->rt_flags & RTF_HOST ? EHOSTDOWN : EHOSTUNREACH); 1378 *drt = rt; 1379 return 0; 1380 } 1381 1382 static int 1383 rt_setshims(struct rtentry *rt, struct sockaddr **rt_shim){ 1384 int i; 1385 1386 for (i=0; i<3; i++) { 1387 struct sockaddr *shim = rt_shim[RTAX_MPLS1 + i]; 1388 int shimlen; 1389 1390 if (shim == NULL) 1391 break; 1392 1393 shimlen = ROUNDUP(shim->sa_len); 1394 R_Malloc(rt->rt_shim[i], struct sockaddr *, shimlen); 1395 bcopy(shim, rt->rt_shim[i], shimlen); 1396 } 1397 1398 return 0; 1399 } 1400 1401 #ifdef ROUTE_DEBUG 1402 1403 /* 1404 * Print out a route table entry 1405 */ 1406 void 1407 rt_print(struct rt_addrinfo *rtinfo, struct rtentry *rn) 1408 { 1409 kprintf("rti %p cpu %d route %p flags %08lx: ", 1410 rtinfo, mycpuid, rn, rn->rt_flags); 1411 sockaddr_print(rt_key(rn)); 1412 kprintf(" mask "); 1413 sockaddr_print(rt_mask(rn)); 1414 kprintf(" gw "); 1415 sockaddr_print(rn->rt_gateway); 1416 kprintf(" ifc \"%s\"", rn->rt_ifp ? rn->rt_ifp->if_dname : "?"); 1417 kprintf(" ifa %p\n", rn->rt_ifa); 1418 } 1419 1420 void 1421 rt_addrinfo_print(int cmd, struct rt_addrinfo *rti) 1422 { 1423 int didit = 0; 1424 int i; 1425 1426 #ifdef ROUTE_DEBUG 1427 if (cmd == RTM_DELETE && route_debug > 1) 1428 print_backtrace(-1); 1429 #endif 1430 1431 switch(cmd) { 1432 case RTM_ADD: 1433 kprintf("ADD "); 1434 break; 1435 case RTM_RESOLVE: 1436 kprintf("RES "); 1437 break; 1438 case RTM_DELETE: 1439 kprintf("DEL "); 1440 break; 1441 default: 1442 kprintf("C%02d ", cmd); 1443 break; 1444 } 1445 kprintf("rti %p cpu %d ", rti, mycpuid); 1446 for (i = 0; i < rti->rti_addrs; ++i) { 1447 if (rti->rti_info[i] == NULL) 1448 continue; 1449 if (didit) 1450 kprintf(" ,"); 1451 switch(i) { 1452 case RTAX_DST: 1453 kprintf("(DST "); 1454 break; 1455 case RTAX_GATEWAY: 1456 kprintf("(GWY "); 1457 break; 1458 case RTAX_NETMASK: 1459 kprintf("(MSK "); 1460 break; 1461 case RTAX_GENMASK: 1462 kprintf("(GEN "); 1463 break; 1464 case RTAX_IFP: 1465 kprintf("(IFP "); 1466 break; 1467 case RTAX_IFA: 1468 kprintf("(IFA "); 1469 break; 1470 case RTAX_AUTHOR: 1471 kprintf("(AUT "); 1472 break; 1473 case RTAX_BRD: 1474 kprintf("(BRD "); 1475 break; 1476 default: 1477 kprintf("(?%02d ", i); 1478 break; 1479 } 1480 sockaddr_print(rti->rti_info[i]); 1481 kprintf(")"); 1482 didit = 1; 1483 } 1484 kprintf("\n"); 1485 } 1486 1487 void 1488 sockaddr_print(struct sockaddr *sa) 1489 { 1490 struct sockaddr_in *sa4; 1491 struct sockaddr_in6 *sa6; 1492 int len; 1493 int i; 1494 1495 if (sa == NULL) { 1496 kprintf("NULL"); 1497 return; 1498 } 1499 1500 len = sa->sa_len - offsetof(struct sockaddr, sa_data[0]); 1501 1502 switch(sa->sa_family) { 1503 case AF_INET: 1504 case AF_INET6: 1505 default: 1506 switch(sa->sa_family) { 1507 case AF_INET: 1508 sa4 = (struct sockaddr_in *)sa; 1509 kprintf("INET %d %d.%d.%d.%d", 1510 ntohs(sa4->sin_port), 1511 (ntohl(sa4->sin_addr.s_addr) >> 24) & 255, 1512 (ntohl(sa4->sin_addr.s_addr) >> 16) & 255, 1513 (ntohl(sa4->sin_addr.s_addr) >> 8) & 255, 1514 (ntohl(sa4->sin_addr.s_addr) >> 0) & 255 1515 ); 1516 break; 1517 case AF_INET6: 1518 sa6 = (struct sockaddr_in6 *)sa; 1519 kprintf("INET6 %d %04x:%04x%04x:%04x:%04x:%04x:%04x:%04x", 1520 ntohs(sa6->sin6_port), 1521 sa6->sin6_addr.s6_addr16[0], 1522 sa6->sin6_addr.s6_addr16[1], 1523 sa6->sin6_addr.s6_addr16[2], 1524 sa6->sin6_addr.s6_addr16[3], 1525 sa6->sin6_addr.s6_addr16[4], 1526 sa6->sin6_addr.s6_addr16[5], 1527 sa6->sin6_addr.s6_addr16[6], 1528 sa6->sin6_addr.s6_addr16[7] 1529 ); 1530 break; 1531 default: 1532 kprintf("AF%d ", sa->sa_family); 1533 while (len > 0 && sa->sa_data[len-1] == 0) 1534 --len; 1535 1536 for (i = 0; i < len; ++i) { 1537 if (i) 1538 kprintf("."); 1539 kprintf("%d", (unsigned char)sa->sa_data[i]); 1540 } 1541 break; 1542 } 1543 } 1544 } 1545 1546 #endif 1547 1548 /* 1549 * Set up a routing table entry, normally for an interface. 1550 */ 1551 int 1552 rtinit(struct ifaddr *ifa, int cmd, int flags) 1553 { 1554 struct sockaddr *dst, *deldst, *netmask; 1555 struct mbuf *m = NULL; 1556 struct radix_node_head *rnh; 1557 struct radix_node *rn; 1558 struct rt_addrinfo rtinfo; 1559 int error; 1560 1561 if (flags & RTF_HOST) { 1562 dst = ifa->ifa_dstaddr; 1563 netmask = NULL; 1564 } else { 1565 dst = ifa->ifa_addr; 1566 netmask = ifa->ifa_netmask; 1567 } 1568 /* 1569 * If it's a delete, check that if it exists, it's on the correct 1570 * interface or we might scrub a route to another ifa which would 1571 * be confusing at best and possibly worse. 1572 */ 1573 if (cmd == RTM_DELETE) { 1574 /* 1575 * It's a delete, so it should already exist.. 1576 * If it's a net, mask off the host bits 1577 * (Assuming we have a mask) 1578 */ 1579 if (netmask != NULL) { 1580 m = m_get(MB_DONTWAIT, MT_SONAME); 1581 if (m == NULL) 1582 return (ENOBUFS); 1583 mbuftrackid(m, 34); 1584 deldst = mtod(m, struct sockaddr *); 1585 rt_maskedcopy(dst, deldst, netmask); 1586 dst = deldst; 1587 } 1588 /* 1589 * Look up an rtentry that is in the routing tree and 1590 * contains the correct info. 1591 */ 1592 if ((rnh = rt_tables[mycpuid][dst->sa_family]) == NULL || 1593 (rn = rnh->rnh_lookup((char *)dst, 1594 (char *)netmask, rnh)) == NULL || 1595 ((struct rtentry *)rn)->rt_ifa != ifa || 1596 !sa_equal((struct sockaddr *)rn->rn_key, dst)) { 1597 if (m != NULL) 1598 m_free(m); 1599 return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH); 1600 } 1601 /* XXX */ 1602 #if 0 1603 else { 1604 /* 1605 * One would think that as we are deleting, and we know 1606 * it doesn't exist, we could just return at this point 1607 * with an "ELSE" clause, but apparently not.. 1608 */ 1609 return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH); 1610 } 1611 #endif 1612 } 1613 /* 1614 * Do the actual request 1615 */ 1616 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 1617 rtinfo.rti_info[RTAX_DST] = dst; 1618 rtinfo.rti_info[RTAX_GATEWAY] = ifa->ifa_addr; 1619 rtinfo.rti_info[RTAX_NETMASK] = netmask; 1620 rtinfo.rti_flags = flags | ifa->ifa_flags; 1621 rtinfo.rti_ifa = ifa; 1622 error = rtrequest1_global(cmd, &rtinfo, rtinit_rtrequest_callback, ifa); 1623 if (m != NULL) 1624 m_free(m); 1625 return (error); 1626 } 1627 1628 static void 1629 rtinit_rtrequest_callback(int cmd, int error, 1630 struct rt_addrinfo *rtinfo, struct rtentry *rt, 1631 void *arg) 1632 { 1633 struct ifaddr *ifa = arg; 1634 1635 if (error == 0 && rt) { 1636 if (mycpuid == 0) { 1637 ++rt->rt_refcnt; 1638 rt_newaddrmsg(cmd, ifa, error, rt); 1639 --rt->rt_refcnt; 1640 } 1641 if (cmd == RTM_DELETE) { 1642 if (rt->rt_refcnt == 0) { 1643 ++rt->rt_refcnt; 1644 rtfree(rt); 1645 } 1646 } 1647 } 1648 } 1649 1650 struct netmsg_rts { 1651 struct netmsg_base base; 1652 int req; 1653 struct rt_addrinfo *rtinfo; 1654 rtsearch_callback_func_t callback; 1655 void *arg; 1656 boolean_t exact_match; 1657 int found_cnt; 1658 }; 1659 1660 int 1661 rtsearch_global(int req, struct rt_addrinfo *rtinfo, 1662 rtsearch_callback_func_t callback, void *arg, 1663 boolean_t exact_match) 1664 { 1665 struct netmsg_rts msg; 1666 1667 netmsg_init(&msg.base, NULL, &curthread->td_msgport, 1668 0, rtsearch_msghandler); 1669 msg.req = req; 1670 msg.rtinfo = rtinfo; 1671 msg.callback = callback; 1672 msg.arg = arg; 1673 msg.exact_match = exact_match; 1674 msg.found_cnt = 0; 1675 return lwkt_domsg(rtable_portfn(0), &msg.base.lmsg, 0); 1676 } 1677 1678 static void 1679 rtsearch_msghandler(netmsg_t msg) 1680 { 1681 struct netmsg_rts *rmsg = (void *)msg; 1682 struct rt_addrinfo rtinfo; 1683 struct radix_node_head *rnh; 1684 struct rtentry *rt; 1685 int nextcpu, error; 1686 1687 /* 1688 * Copy the rtinfo. We need to make sure that the original 1689 * rtinfo, which is setup by the caller, in the netmsg will 1690 * _not_ be changed; else the next CPU on the netmsg forwarding 1691 * path will see a different rtinfo than what this CPU has seen. 1692 */ 1693 rtinfo = *rmsg->rtinfo; 1694 1695 /* 1696 * Find the correct routing tree to use for this Address Family 1697 */ 1698 if ((rnh = rt_tables[mycpuid][rtinfo.rti_dst->sa_family]) == NULL) { 1699 if (mycpuid != 0) 1700 panic("partially initialized routing tables\n"); 1701 lwkt_replymsg(&rmsg->base.lmsg, EAFNOSUPPORT); 1702 return; 1703 } 1704 1705 /* 1706 * Correct rtinfo for the host route searching. 1707 */ 1708 if (rtinfo.rti_flags & RTF_HOST) { 1709 rtinfo.rti_netmask = NULL; 1710 rtinfo.rti_flags &= ~(RTF_CLONING | RTF_PRCLONING); 1711 } 1712 1713 rt = (struct rtentry *) 1714 rnh->rnh_lookup((char *)rtinfo.rti_dst, 1715 (char *)rtinfo.rti_netmask, rnh); 1716 1717 /* 1718 * If we are asked to do the "exact match", we need to make sure 1719 * that host route searching got a host route while a network 1720 * route searching got a network route. 1721 */ 1722 if (rt != NULL && rmsg->exact_match && 1723 ((rt->rt_flags ^ rtinfo.rti_flags) & RTF_HOST)) 1724 rt = NULL; 1725 1726 if (rt == NULL) { 1727 /* 1728 * No matching routes have been found, don't count this 1729 * as a critical error (here, we set 'error' to 0), just 1730 * keep moving on, since at least prcloned routes are not 1731 * duplicated onto each CPU. 1732 */ 1733 error = 0; 1734 } else { 1735 rmsg->found_cnt++; 1736 1737 rt->rt_refcnt++; 1738 error = rmsg->callback(rmsg->req, &rtinfo, rt, rmsg->arg, 1739 rmsg->found_cnt); 1740 rt->rt_refcnt--; 1741 1742 if (error == EJUSTRETURN) { 1743 lwkt_replymsg(&rmsg->base.lmsg, 0); 1744 return; 1745 } 1746 } 1747 1748 nextcpu = mycpuid + 1; 1749 if (error) { 1750 KKASSERT(rmsg->found_cnt > 0); 1751 1752 /* 1753 * Under following cases, unrecoverable error has 1754 * not occured: 1755 * o Request is RTM_GET 1756 * o The first time that we find the route, but the 1757 * modification fails. 1758 */ 1759 if (rmsg->req != RTM_GET && rmsg->found_cnt > 1) { 1760 panic("rtsearch_msghandler: unrecoverable error " 1761 "cpu %d", mycpuid); 1762 } 1763 lwkt_replymsg(&rmsg->base.lmsg, error); 1764 } else if (nextcpu < ncpus) { 1765 lwkt_forwardmsg(rtable_portfn(nextcpu), &rmsg->base.lmsg); 1766 } else { 1767 if (rmsg->found_cnt == 0) { 1768 /* The requested route was never seen ... */ 1769 error = ESRCH; 1770 } 1771 lwkt_replymsg(&rmsg->base.lmsg, error); 1772 } 1773 } 1774 1775 int 1776 rtmask_add_global(struct sockaddr *mask) 1777 { 1778 struct netmsg_base msg; 1779 1780 netmsg_init(&msg, NULL, &curthread->td_msgport, 1781 0, rtmask_add_msghandler); 1782 msg.lmsg.u.ms_resultp = mask; 1783 1784 return lwkt_domsg(rtable_portfn(0), &msg.lmsg, 0); 1785 } 1786 1787 struct sockaddr * 1788 _rtmask_lookup(struct sockaddr *mask, boolean_t search) 1789 { 1790 struct radix_node *n; 1791 1792 #define clen(s) (*(u_char *)(s)) 1793 n = rn_addmask((char *)mask, search, 1, rn_cpumaskhead(mycpuid)); 1794 if (n != NULL && 1795 mask->sa_len >= clen(n->rn_key) && 1796 bcmp((char *)mask + 1, 1797 (char *)n->rn_key + 1, clen(n->rn_key) - 1) == 0) { 1798 return (struct sockaddr *)n->rn_key; 1799 } else { 1800 return NULL; 1801 } 1802 #undef clen 1803 } 1804 1805 static void 1806 rtmask_add_msghandler(netmsg_t msg) 1807 { 1808 struct lwkt_msg *lmsg = &msg->lmsg; 1809 struct sockaddr *mask = lmsg->u.ms_resultp; 1810 int error = 0, nextcpu; 1811 1812 if (rtmask_lookup(mask) == NULL) 1813 error = ENOBUFS; 1814 1815 nextcpu = mycpuid + 1; 1816 if (!error && nextcpu < ncpus) 1817 lwkt_forwardmsg(rtable_portfn(nextcpu), lmsg); 1818 else 1819 lwkt_replymsg(lmsg, error); 1820 } 1821 1822 /* This must be before ip6_init2(), which is now SI_ORDER_MIDDLE */ 1823 SYSINIT(route, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0); 1824