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 u_long route_kmalloc_limit = 0; 138 TUNABLE_ULONG("net.route.kmalloc_limit", &route_kmalloc_limit); 139 140 /* 141 * Initialize the route table(s) for protocol domains and 142 * create a helper thread which will be responsible for updating 143 * route table entries on each cpu. 144 */ 145 void 146 route_init(void) 147 { 148 int cpu; 149 thread_t rtd; 150 151 for (cpu = 0; cpu < ncpus; ++cpu) 152 bzero(&rtstatistics_percpu[cpu], sizeof(struct rtstatistics)); 153 rn_init(); /* initialize all zeroes, all ones, mask table */ 154 rtable_init(); /* call dom_rtattach() on each cpu */ 155 156 for (cpu = 0; cpu < ncpus; cpu++) { 157 lwkt_create(rtable_service_loop, NULL, &rtd, NULL, 158 0, cpu, "rtable_cpu %d", cpu); 159 rt_ports[cpu] = &rtd->td_msgport; 160 } 161 162 if (route_kmalloc_limit) 163 kmalloc_raise_limit(M_RTABLE, route_kmalloc_limit); 164 } 165 166 static void 167 rtable_init_oncpu(netmsg_t msg) 168 { 169 struct domain *dom; 170 int cpu = mycpuid; 171 172 SLIST_FOREACH(dom, &domains, dom_next) { 173 if (dom->dom_rtattach) { 174 dom->dom_rtattach( 175 (void **)&rt_tables[cpu][dom->dom_family], 176 dom->dom_rtoffset); 177 } 178 } 179 ifnet_forwardmsg(&msg->lmsg, cpu + 1); 180 } 181 182 static void 183 rtable_init(void) 184 { 185 struct netmsg_base msg; 186 187 netmsg_init(&msg, NULL, &curthread->td_msgport, 0, rtable_init_oncpu); 188 ifnet_domsg(&msg.lmsg, 0); 189 } 190 191 /* 192 * Our per-cpu table management protocol thread. All route table operations 193 * are sequentially chained through all cpus starting at cpu #0 in order to 194 * maintain duplicate route tables on each cpu. Having a spearate route 195 * table management thread allows the protocol and interrupt threads to 196 * issue route table changes. 197 */ 198 static void 199 rtable_service_loop(void *dummy __unused) 200 { 201 netmsg_base_t msg; 202 thread_t td = curthread; 203 204 while ((msg = lwkt_waitport(&td->td_msgport, 0)) != NULL) { 205 msg->nm_dispatch((netmsg_t)msg); 206 } 207 } 208 209 /* 210 * Routing statistics. 211 */ 212 #ifdef SMP 213 static int 214 sysctl_rtstatistics(SYSCTL_HANDLER_ARGS) 215 { 216 int cpu, error = 0; 217 218 for (cpu = 0; cpu < ncpus; ++cpu) { 219 if ((error = SYSCTL_OUT(req, &rtstatistics_percpu[cpu], 220 sizeof(struct rtstatistics)))) 221 break; 222 if ((error = SYSCTL_IN(req, &rtstatistics_percpu[cpu], 223 sizeof(struct rtstatistics)))) 224 break; 225 } 226 227 return (error); 228 } 229 SYSCTL_PROC(_net_route, OID_AUTO, stats, (CTLTYPE_OPAQUE|CTLFLAG_RW), 230 0, 0, sysctl_rtstatistics, "S,rtstatistics", "Routing statistics"); 231 #else 232 SYSCTL_STRUCT(_net_route, OID_AUTO, stats, CTLFLAG_RW, &rtstat, rtstatistics, 233 "Routing statistics"); 234 #endif 235 236 /* 237 * Packet routing routines. 238 */ 239 240 /* 241 * Look up and fill in the "ro_rt" rtentry field in a route structure given 242 * an address in the "ro_dst" field. Always send a report on a miss and 243 * always clone routes. 244 */ 245 void 246 rtalloc(struct route *ro) 247 { 248 rtalloc_ign(ro, 0UL); 249 } 250 251 /* 252 * Look up and fill in the "ro_rt" rtentry field in a route structure given 253 * an address in the "ro_dst" field. Always send a report on a miss and 254 * optionally clone routes when RTF_CLONING or RTF_PRCLONING are not being 255 * ignored. 256 */ 257 void 258 rtalloc_ign(struct route *ro, u_long ignoreflags) 259 { 260 if (ro->ro_rt != NULL) { 261 if (ro->ro_rt->rt_ifp != NULL && ro->ro_rt->rt_flags & RTF_UP) 262 return; 263 rtfree(ro->ro_rt); 264 ro->ro_rt = NULL; 265 } 266 ro->ro_rt = _rtlookup(&ro->ro_dst, RTL_REPORTMSG, ignoreflags); 267 } 268 269 /* 270 * Look up the route that matches the given "dst" address. 271 * 272 * Route lookup can have the side-effect of creating and returning 273 * a cloned route instead when "dst" matches a cloning route and the 274 * RTF_CLONING and RTF_PRCLONING flags are not being ignored. 275 * 276 * Any route returned has its reference count incremented. 277 */ 278 struct rtentry * 279 _rtlookup(struct sockaddr *dst, boolean_t generate_report, u_long ignore) 280 { 281 struct radix_node_head *rnh = rt_tables[mycpuid][dst->sa_family]; 282 struct rtentry *rt; 283 284 if (rnh == NULL) 285 goto unreach; 286 287 /* 288 * Look up route in the radix tree. 289 */ 290 rt = (struct rtentry *) rnh->rnh_matchaddr((char *)dst, rnh); 291 if (rt == NULL) 292 goto unreach; 293 294 /* 295 * Handle cloning routes. 296 */ 297 if ((rt->rt_flags & ~ignore & (RTF_CLONING | RTF_PRCLONING)) != 0) { 298 struct rtentry *clonedroute; 299 int error; 300 301 clonedroute = rt; /* copy in/copy out parameter */ 302 error = rtrequest(RTM_RESOLVE, dst, NULL, NULL, 0, 303 &clonedroute); /* clone the route */ 304 if (error != 0) { /* cloning failed */ 305 if (generate_report) 306 rt_dstmsg(RTM_MISS, dst, error); 307 rt->rt_refcnt++; 308 return (rt); /* return the uncloned route */ 309 } 310 if (generate_report) { 311 if (clonedroute->rt_flags & RTF_XRESOLVE) 312 rt_dstmsg(RTM_RESOLVE, dst, 0); 313 else 314 rt_rtmsg(RTM_ADD, clonedroute, 315 clonedroute->rt_ifp, 0); 316 } 317 return (clonedroute); /* return cloned route */ 318 } 319 320 /* 321 * Increment the reference count of the matched route and return. 322 */ 323 rt->rt_refcnt++; 324 return (rt); 325 326 unreach: 327 rtstat.rts_unreach++; 328 if (generate_report) 329 rt_dstmsg(RTM_MISS, dst, 0); 330 return (NULL); 331 } 332 333 void 334 rtfree(struct rtentry *rt) 335 { 336 if (rt->rt_cpuid == mycpuid) 337 rtfree_oncpu(rt); 338 else 339 rtfree_remote(rt, 1); 340 } 341 342 void 343 rtfree_oncpu(struct rtentry *rt) 344 { 345 KKASSERT(rt->rt_cpuid == mycpuid); 346 KASSERT(rt->rt_refcnt > 0, ("rtfree: rt_refcnt %ld", rt->rt_refcnt)); 347 348 --rt->rt_refcnt; 349 if (rt->rt_refcnt == 0) { 350 struct radix_node_head *rnh = 351 rt_tables[mycpuid][rt_key(rt)->sa_family]; 352 353 if (rnh->rnh_close) 354 rnh->rnh_close((struct radix_node *)rt, rnh); 355 if (!(rt->rt_flags & RTF_UP)) { 356 /* deallocate route */ 357 if (rt->rt_ifa != NULL) 358 IFAFREE(rt->rt_ifa); 359 if (rt->rt_parent != NULL) 360 RTFREE(rt->rt_parent); /* recursive call! */ 361 Free(rt_key(rt)); 362 Free(rt); 363 } 364 } 365 } 366 367 static void 368 rtfree_remote_dispatch(netmsg_t msg) 369 { 370 struct lwkt_msg *lmsg = &msg->lmsg; 371 struct rtentry *rt = lmsg->u.ms_resultp; 372 373 rtfree_oncpu(rt); 374 lwkt_replymsg(lmsg, 0); 375 } 376 377 void 378 rtfree_remote(struct rtentry *rt, int allow_panic) 379 { 380 struct netmsg_base msg; 381 struct lwkt_msg *lmsg; 382 383 KKASSERT(rt->rt_cpuid != mycpuid); 384 385 if (route_assert_owner_access && allow_panic) { 386 panic("rt remote free rt_cpuid %d, mycpuid %d\n", 387 rt->rt_cpuid, mycpuid); 388 } else { 389 kprintf("rt remote free rt_cpuid %d, mycpuid %d\n", 390 rt->rt_cpuid, mycpuid); 391 print_backtrace(-1); 392 } 393 394 netmsg_init(&msg, NULL, &curthread->td_msgport, 395 0, rtfree_remote_dispatch); 396 lmsg = &msg.lmsg; 397 lmsg->u.ms_resultp = rt; 398 399 lwkt_domsg(rtable_portfn(rt->rt_cpuid), lmsg, 0); 400 } 401 402 static int 403 rtredirect_oncpu(struct sockaddr *dst, struct sockaddr *gateway, 404 struct sockaddr *netmask, int flags, struct sockaddr *src) 405 { 406 struct rtentry *rt = NULL; 407 struct rt_addrinfo rtinfo; 408 struct ifaddr *ifa; 409 u_long *stat = NULL; 410 int error; 411 412 /* verify the gateway is directly reachable */ 413 if ((ifa = ifa_ifwithnet(gateway)) == NULL) { 414 error = ENETUNREACH; 415 goto out; 416 } 417 418 /* 419 * If the redirect isn't from our current router for this destination, 420 * it's either old or wrong. 421 */ 422 if (!(flags & RTF_DONE) && /* XXX JH */ 423 (rt = rtpurelookup(dst)) != NULL && 424 (!sa_equal(src, rt->rt_gateway) || rt->rt_ifa != ifa)) { 425 error = EINVAL; 426 goto done; 427 } 428 429 /* 430 * If it redirects us to ourselves, we have a routing loop, 431 * perhaps as a result of an interface going down recently. 432 */ 433 if (ifa_ifwithaddr(gateway)) { 434 error = EHOSTUNREACH; 435 goto done; 436 } 437 438 /* 439 * Create a new entry if the lookup failed or if we got back 440 * a wildcard entry for the default route. This is necessary 441 * for hosts which use routing redirects generated by smart 442 * gateways to dynamically build the routing tables. 443 */ 444 if (rt == NULL) 445 goto create; 446 if ((rt_mask(rt) != NULL && rt_mask(rt)->sa_len < 2)) { 447 rtfree(rt); 448 goto create; 449 } 450 451 /* Ignore redirects for directly connected hosts. */ 452 if (!(rt->rt_flags & RTF_GATEWAY)) { 453 error = EHOSTUNREACH; 454 goto done; 455 } 456 457 if (!(rt->rt_flags & RTF_HOST) && (flags & RTF_HOST)) { 458 /* 459 * Changing from a network route to a host route. 460 * Create a new host route rather than smashing the 461 * network route. 462 */ 463 create: 464 flags |= RTF_GATEWAY | RTF_DYNAMIC; 465 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 466 rtinfo.rti_info[RTAX_DST] = dst; 467 rtinfo.rti_info[RTAX_GATEWAY] = gateway; 468 rtinfo.rti_info[RTAX_NETMASK] = netmask; 469 rtinfo.rti_flags = flags; 470 rtinfo.rti_ifa = ifa; 471 rt = NULL; /* copy-in/copy-out parameter */ 472 error = rtrequest1(RTM_ADD, &rtinfo, &rt); 473 if (rt != NULL) 474 flags = rt->rt_flags; 475 stat = &rtstat.rts_dynamic; 476 } else { 477 /* 478 * Smash the current notion of the gateway to this destination. 479 * Should check about netmask!!! 480 */ 481 rt->rt_flags |= RTF_MODIFIED; 482 flags |= RTF_MODIFIED; 483 484 /* We only need to report rtmsg on CPU0 */ 485 rt_setgate(rt, rt_key(rt), gateway, 486 mycpuid == 0 ? RTL_REPORTMSG : RTL_DONTREPORT); 487 error = 0; 488 stat = &rtstat.rts_newgateway; 489 } 490 491 done: 492 if (rt != NULL) 493 rtfree(rt); 494 out: 495 if (error != 0) 496 rtstat.rts_badredirect++; 497 else if (stat != NULL) 498 (*stat)++; 499 500 return error; 501 } 502 503 #ifdef SMP 504 505 struct netmsg_rtredirect { 506 struct netmsg_base base; 507 struct sockaddr *dst; 508 struct sockaddr *gateway; 509 struct sockaddr *netmask; 510 int flags; 511 struct sockaddr *src; 512 }; 513 514 #endif 515 516 /* 517 * Force a routing table entry to the specified 518 * destination to go through the given gateway. 519 * Normally called as a result of a routing redirect 520 * message from the network layer. 521 * 522 * N.B.: must be called at splnet 523 */ 524 void 525 rtredirect(struct sockaddr *dst, struct sockaddr *gateway, 526 struct sockaddr *netmask, int flags, struct sockaddr *src) 527 { 528 struct rt_addrinfo rtinfo; 529 int error; 530 #ifdef SMP 531 struct netmsg_rtredirect msg; 532 533 netmsg_init(&msg.base, NULL, &curthread->td_msgport, 534 0, rtredirect_msghandler); 535 msg.dst = dst; 536 msg.gateway = gateway; 537 msg.netmask = netmask; 538 msg.flags = flags; 539 msg.src = src; 540 error = lwkt_domsg(rtable_portfn(0), &msg.base.lmsg, 0); 541 #else 542 error = rtredirect_oncpu(dst, gateway, netmask, flags, src); 543 #endif 544 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 545 rtinfo.rti_info[RTAX_DST] = dst; 546 rtinfo.rti_info[RTAX_GATEWAY] = gateway; 547 rtinfo.rti_info[RTAX_NETMASK] = netmask; 548 rtinfo.rti_info[RTAX_AUTHOR] = src; 549 rt_missmsg(RTM_REDIRECT, &rtinfo, flags, error); 550 } 551 552 #ifdef SMP 553 554 static void 555 rtredirect_msghandler(netmsg_t msg) 556 { 557 struct netmsg_rtredirect *rmsg = (void *)msg; 558 int nextcpu; 559 560 rtredirect_oncpu(rmsg->dst, rmsg->gateway, rmsg->netmask, 561 rmsg->flags, rmsg->src); 562 nextcpu = mycpuid + 1; 563 if (nextcpu < ncpus) 564 lwkt_forwardmsg(rtable_portfn(nextcpu), &msg->lmsg); 565 else 566 lwkt_replymsg(&msg->lmsg, 0); 567 } 568 569 #endif 570 571 /* 572 * Routing table ioctl interface. 573 */ 574 int 575 rtioctl(u_long req, caddr_t data, struct ucred *cred) 576 { 577 #ifdef INET 578 /* Multicast goop, grrr... */ 579 return mrt_ioctl ? mrt_ioctl(req, data) : EOPNOTSUPP; 580 #else 581 return ENXIO; 582 #endif 583 } 584 585 struct ifaddr * 586 ifa_ifwithroute(int flags, struct sockaddr *dst, struct sockaddr *gateway) 587 { 588 struct ifaddr *ifa; 589 590 if (!(flags & RTF_GATEWAY)) { 591 /* 592 * If we are adding a route to an interface, 593 * and the interface is a point-to-point link, 594 * we should search for the destination 595 * as our clue to the interface. Otherwise 596 * we can use the local address. 597 */ 598 ifa = NULL; 599 if (flags & RTF_HOST) { 600 ifa = ifa_ifwithdstaddr(dst); 601 } 602 if (ifa == NULL) 603 ifa = ifa_ifwithaddr(gateway); 604 } else { 605 /* 606 * If we are adding a route to a remote net 607 * or host, the gateway may still be on the 608 * other end of a pt to pt link. 609 */ 610 ifa = ifa_ifwithdstaddr(gateway); 611 } 612 if (ifa == NULL) 613 ifa = ifa_ifwithnet(gateway); 614 if (ifa == NULL) { 615 struct rtentry *rt; 616 617 rt = rtpurelookup(gateway); 618 if (rt == NULL) 619 return (NULL); 620 rt->rt_refcnt--; 621 if ((ifa = rt->rt_ifa) == NULL) 622 return (NULL); 623 } 624 if (ifa->ifa_addr->sa_family != dst->sa_family) { 625 struct ifaddr *oldifa = ifa; 626 627 ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp); 628 if (ifa == NULL) 629 ifa = oldifa; 630 } 631 return (ifa); 632 } 633 634 static int rt_fixdelete (struct radix_node *, void *); 635 static int rt_fixchange (struct radix_node *, void *); 636 637 struct rtfc_arg { 638 struct rtentry *rt0; 639 struct radix_node_head *rnh; 640 }; 641 642 /* 643 * Set rtinfo->rti_ifa and rtinfo->rti_ifp. 644 */ 645 int 646 rt_getifa(struct rt_addrinfo *rtinfo) 647 { 648 struct sockaddr *gateway = rtinfo->rti_info[RTAX_GATEWAY]; 649 struct sockaddr *dst = rtinfo->rti_info[RTAX_DST]; 650 struct sockaddr *ifaaddr = rtinfo->rti_info[RTAX_IFA]; 651 int flags = rtinfo->rti_flags; 652 653 /* 654 * ifp may be specified by sockaddr_dl 655 * when protocol address is ambiguous. 656 */ 657 if (rtinfo->rti_ifp == NULL) { 658 struct sockaddr *ifpaddr; 659 660 ifpaddr = rtinfo->rti_info[RTAX_IFP]; 661 if (ifpaddr != NULL && ifpaddr->sa_family == AF_LINK) { 662 struct ifaddr *ifa; 663 664 ifa = ifa_ifwithnet(ifpaddr); 665 if (ifa != NULL) 666 rtinfo->rti_ifp = ifa->ifa_ifp; 667 } 668 } 669 670 if (rtinfo->rti_ifa == NULL && ifaaddr != NULL) 671 rtinfo->rti_ifa = ifa_ifwithaddr(ifaaddr); 672 if (rtinfo->rti_ifa == NULL) { 673 struct sockaddr *sa; 674 675 sa = ifaaddr != NULL ? ifaaddr : 676 (gateway != NULL ? gateway : dst); 677 if (sa != NULL && rtinfo->rti_ifp != NULL) 678 rtinfo->rti_ifa = ifaof_ifpforaddr(sa, rtinfo->rti_ifp); 679 else if (dst != NULL && gateway != NULL) 680 rtinfo->rti_ifa = ifa_ifwithroute(flags, dst, gateway); 681 else if (sa != NULL) 682 rtinfo->rti_ifa = ifa_ifwithroute(flags, sa, sa); 683 } 684 if (rtinfo->rti_ifa == NULL) 685 return (ENETUNREACH); 686 687 if (rtinfo->rti_ifp == NULL) 688 rtinfo->rti_ifp = rtinfo->rti_ifa->ifa_ifp; 689 return (0); 690 } 691 692 /* 693 * Do appropriate manipulations of a routing tree given 694 * all the bits of info needed 695 */ 696 int 697 rtrequest( 698 int req, 699 struct sockaddr *dst, 700 struct sockaddr *gateway, 701 struct sockaddr *netmask, 702 int flags, 703 struct rtentry **ret_nrt) 704 { 705 struct rt_addrinfo rtinfo; 706 707 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 708 rtinfo.rti_info[RTAX_DST] = dst; 709 rtinfo.rti_info[RTAX_GATEWAY] = gateway; 710 rtinfo.rti_info[RTAX_NETMASK] = netmask; 711 rtinfo.rti_flags = flags; 712 return rtrequest1(req, &rtinfo, ret_nrt); 713 } 714 715 int 716 rtrequest_global( 717 int req, 718 struct sockaddr *dst, 719 struct sockaddr *gateway, 720 struct sockaddr *netmask, 721 int flags) 722 { 723 struct rt_addrinfo rtinfo; 724 725 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 726 rtinfo.rti_info[RTAX_DST] = dst; 727 rtinfo.rti_info[RTAX_GATEWAY] = gateway; 728 rtinfo.rti_info[RTAX_NETMASK] = netmask; 729 rtinfo.rti_flags = flags; 730 return rtrequest1_global(req, &rtinfo, NULL, NULL); 731 } 732 733 #ifdef SMP 734 735 struct netmsg_rtq { 736 struct netmsg_base base; 737 int req; 738 struct rt_addrinfo *rtinfo; 739 rtrequest1_callback_func_t callback; 740 void *arg; 741 }; 742 743 #endif 744 745 int 746 rtrequest1_global(int req, struct rt_addrinfo *rtinfo, 747 rtrequest1_callback_func_t callback, void *arg) 748 { 749 int error; 750 #ifdef SMP 751 struct netmsg_rtq msg; 752 753 netmsg_init(&msg.base, NULL, &curthread->td_msgport, 754 0, rtrequest1_msghandler); 755 msg.base.lmsg.ms_error = -1; 756 msg.req = req; 757 msg.rtinfo = rtinfo; 758 msg.callback = callback; 759 msg.arg = arg; 760 error = lwkt_domsg(rtable_portfn(0), &msg.base.lmsg, 0); 761 #else 762 struct rtentry *rt = NULL; 763 764 error = rtrequest1(req, rtinfo, &rt); 765 if (rt) 766 --rt->rt_refcnt; 767 if (callback) 768 callback(req, error, rtinfo, rt, arg); 769 #endif 770 return (error); 771 } 772 773 /* 774 * Handle a route table request on the current cpu. Since the route table's 775 * are supposed to be identical on each cpu, an error occuring later in the 776 * message chain is considered system-fatal. 777 */ 778 #ifdef SMP 779 780 static void 781 rtrequest1_msghandler(netmsg_t msg) 782 { 783 struct netmsg_rtq *rmsg = (void *)msg; 784 struct rt_addrinfo rtinfo; 785 struct rtentry *rt = NULL; 786 int nextcpu; 787 int error; 788 789 /* 790 * Copy the rtinfo. We need to make sure that the original 791 * rtinfo, which is setup by the caller, in the netmsg will 792 * _not_ be changed; else the next CPU on the netmsg forwarding 793 * path will see a different rtinfo than what this CPU has seen. 794 */ 795 rtinfo = *rmsg->rtinfo; 796 797 error = rtrequest1(rmsg->req, &rtinfo, &rt); 798 if (rt) 799 --rt->rt_refcnt; 800 if (rmsg->callback) 801 rmsg->callback(rmsg->req, error, &rtinfo, rt, rmsg->arg); 802 803 /* 804 * RTM_DELETE's are propogated even if an error occurs, since a 805 * cloned route might be undergoing deletion and cloned routes 806 * are not necessarily replicated. An overall error is returned 807 * only if no cpus have the route in question. 808 */ 809 if (rmsg->base.lmsg.ms_error < 0 || error == 0) 810 rmsg->base.lmsg.ms_error = error; 811 812 nextcpu = mycpuid + 1; 813 if (error && rmsg->req != RTM_DELETE) { 814 if (mycpuid != 0) { 815 panic("rtrequest1_msghandler: rtrequest table " 816 "error was cpu%d, err %d\n", mycpuid, error); 817 } 818 lwkt_replymsg(&rmsg->base.lmsg, error); 819 } else if (nextcpu < ncpus) { 820 lwkt_forwardmsg(rtable_portfn(nextcpu), &rmsg->base.lmsg); 821 } else { 822 lwkt_replymsg(&rmsg->base.lmsg, rmsg->base.lmsg.ms_error); 823 } 824 } 825 826 #endif 827 828 int 829 rtrequest1(int req, struct rt_addrinfo *rtinfo, struct rtentry **ret_nrt) 830 { 831 struct sockaddr *dst = rtinfo->rti_info[RTAX_DST]; 832 struct rtentry *rt; 833 struct radix_node *rn; 834 struct radix_node_head *rnh; 835 struct ifaddr *ifa; 836 struct sockaddr *ndst; 837 boolean_t reportmsg; 838 int error = 0; 839 840 #define gotoerr(x) { error = x ; goto bad; } 841 842 #ifdef ROUTE_DEBUG 843 if (route_debug) 844 rt_addrinfo_print(req, rtinfo); 845 #endif 846 847 crit_enter(); 848 /* 849 * Find the correct routing tree to use for this Address Family 850 */ 851 if ((rnh = rt_tables[mycpuid][dst->sa_family]) == NULL) 852 gotoerr(EAFNOSUPPORT); 853 854 /* 855 * If we are adding a host route then we don't want to put 856 * a netmask in the tree, nor do we want to clone it. 857 */ 858 if (rtinfo->rti_flags & RTF_HOST) { 859 rtinfo->rti_info[RTAX_NETMASK] = NULL; 860 rtinfo->rti_flags &= ~(RTF_CLONING | RTF_PRCLONING); 861 } 862 863 switch (req) { 864 case RTM_DELETE: 865 /* Remove the item from the tree. */ 866 rn = rnh->rnh_deladdr((char *)rtinfo->rti_info[RTAX_DST], 867 (char *)rtinfo->rti_info[RTAX_NETMASK], 868 rnh); 869 if (rn == NULL) 870 gotoerr(ESRCH); 871 KASSERT(!(rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)), 872 ("rnh_deladdr returned flags 0x%x", rn->rn_flags)); 873 rt = (struct rtentry *)rn; 874 875 /* ref to prevent a deletion race */ 876 ++rt->rt_refcnt; 877 878 /* Free any routes cloned from this one. */ 879 if ((rt->rt_flags & (RTF_CLONING | RTF_PRCLONING)) && 880 rt_mask(rt) != NULL) { 881 rnh->rnh_walktree_from(rnh, (char *)rt_key(rt), 882 (char *)rt_mask(rt), 883 rt_fixdelete, rt); 884 } 885 886 if (rt->rt_gwroute != NULL) { 887 RTFREE(rt->rt_gwroute); 888 rt->rt_gwroute = NULL; 889 } 890 891 /* 892 * NB: RTF_UP must be set during the search above, 893 * because we might delete the last ref, causing 894 * rt to get freed prematurely. 895 */ 896 rt->rt_flags &= ~RTF_UP; 897 898 #ifdef ROUTE_DEBUG 899 if (route_debug) 900 rt_print(rtinfo, rt); 901 #endif 902 903 /* Give the protocol a chance to keep things in sync. */ 904 if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest) 905 ifa->ifa_rtrequest(RTM_DELETE, rt, rtinfo); 906 907 /* 908 * If the caller wants it, then it can have it, 909 * but it's up to it to free the rtentry as we won't be 910 * doing it. 911 */ 912 KASSERT(rt->rt_refcnt >= 0, 913 ("rtrequest1(DELETE): refcnt %ld", rt->rt_refcnt)); 914 if (ret_nrt != NULL) { 915 /* leave ref intact for return */ 916 *ret_nrt = rt; 917 } else { 918 /* deref / attempt to destroy */ 919 rtfree(rt); 920 } 921 break; 922 923 case RTM_RESOLVE: 924 if (ret_nrt == NULL || (rt = *ret_nrt) == NULL) 925 gotoerr(EINVAL); 926 ifa = rt->rt_ifa; 927 rtinfo->rti_flags = 928 rt->rt_flags & ~(RTF_CLONING | RTF_PRCLONING | RTF_STATIC); 929 rtinfo->rti_flags |= RTF_WASCLONED; 930 rtinfo->rti_info[RTAX_GATEWAY] = rt->rt_gateway; 931 if ((rtinfo->rti_info[RTAX_NETMASK] = rt->rt_genmask) == NULL) 932 rtinfo->rti_flags |= RTF_HOST; 933 rtinfo->rti_info[RTAX_MPLS1] = rt->rt_shim[0]; 934 rtinfo->rti_info[RTAX_MPLS2] = rt->rt_shim[1]; 935 rtinfo->rti_info[RTAX_MPLS3] = rt->rt_shim[2]; 936 goto makeroute; 937 938 case RTM_ADD: 939 KASSERT(!(rtinfo->rti_flags & RTF_GATEWAY) || 940 rtinfo->rti_info[RTAX_GATEWAY] != NULL, 941 ("rtrequest: GATEWAY but no gateway")); 942 943 if (rtinfo->rti_ifa == NULL && (error = rt_getifa(rtinfo))) 944 gotoerr(error); 945 ifa = rtinfo->rti_ifa; 946 makeroute: 947 R_Malloc(rt, struct rtentry *, sizeof(struct rtentry)); 948 if (rt == NULL) { 949 if (req == RTM_ADD) { 950 kprintf("rtrequest1: alloc rtentry failed on " 951 "cpu%d\n", mycpuid); 952 } 953 gotoerr(ENOBUFS); 954 } 955 bzero(rt, sizeof(struct rtentry)); 956 rt->rt_flags = RTF_UP | rtinfo->rti_flags; 957 rt->rt_cpuid = mycpuid; 958 959 if (mycpuid != 0 && req == RTM_ADD) { 960 /* For RTM_ADD, we have already sent rtmsg on CPU0. */ 961 reportmsg = RTL_DONTREPORT; 962 } else { 963 /* 964 * For RTM_ADD, we only send rtmsg on CPU0. 965 * For RTM_RESOLVE, we always send rtmsg. XXX 966 */ 967 reportmsg = RTL_REPORTMSG; 968 } 969 error = rt_setgate(rt, dst, rtinfo->rti_info[RTAX_GATEWAY], 970 reportmsg); 971 if (error != 0) { 972 Free(rt); 973 gotoerr(error); 974 } 975 976 ndst = rt_key(rt); 977 if (rtinfo->rti_info[RTAX_NETMASK] != NULL) 978 rt_maskedcopy(dst, ndst, 979 rtinfo->rti_info[RTAX_NETMASK]); 980 else 981 bcopy(dst, ndst, dst->sa_len); 982 983 if (rtinfo->rti_info[RTAX_MPLS1] != NULL) 984 rt_setshims(rt, rtinfo->rti_info); 985 986 /* 987 * Note that we now have a reference to the ifa. 988 * This moved from below so that rnh->rnh_addaddr() can 989 * examine the ifa and ifa->ifa_ifp if it so desires. 990 */ 991 IFAREF(ifa); 992 rt->rt_ifa = ifa; 993 rt->rt_ifp = ifa->ifa_ifp; 994 /* XXX mtu manipulation will be done in rnh_addaddr -- itojun */ 995 996 rn = rnh->rnh_addaddr((char *)ndst, 997 (char *)rtinfo->rti_info[RTAX_NETMASK], 998 rnh, rt->rt_nodes); 999 if (rn == NULL) { 1000 struct rtentry *oldrt; 1001 1002 /* 1003 * We already have one of these in the tree. 1004 * We do a special hack: if the old route was 1005 * cloned, then we blow it away and try 1006 * re-inserting the new one. 1007 */ 1008 oldrt = rtpurelookup(ndst); 1009 if (oldrt != NULL) { 1010 --oldrt->rt_refcnt; 1011 if (oldrt->rt_flags & RTF_WASCLONED) { 1012 rtrequest(RTM_DELETE, rt_key(oldrt), 1013 oldrt->rt_gateway, 1014 rt_mask(oldrt), 1015 oldrt->rt_flags, NULL); 1016 rn = rnh->rnh_addaddr((char *)ndst, 1017 (char *) 1018 rtinfo->rti_info[RTAX_NETMASK], 1019 rnh, rt->rt_nodes); 1020 } 1021 } 1022 } 1023 1024 /* 1025 * If it still failed to go into the tree, 1026 * then un-make it (this should be a function). 1027 */ 1028 if (rn == NULL) { 1029 if (rt->rt_gwroute != NULL) 1030 rtfree(rt->rt_gwroute); 1031 IFAFREE(ifa); 1032 Free(rt_key(rt)); 1033 Free(rt); 1034 gotoerr(EEXIST); 1035 } 1036 1037 /* 1038 * If we got here from RESOLVE, then we are cloning 1039 * so clone the rest, and note that we 1040 * are a clone (and increment the parent's references) 1041 */ 1042 if (req == RTM_RESOLVE) { 1043 rt->rt_rmx = (*ret_nrt)->rt_rmx; /* copy metrics */ 1044 rt->rt_rmx.rmx_pksent = 0; /* reset packet counter */ 1045 if ((*ret_nrt)->rt_flags & 1046 (RTF_CLONING | RTF_PRCLONING)) { 1047 rt->rt_parent = *ret_nrt; 1048 (*ret_nrt)->rt_refcnt++; 1049 } 1050 } 1051 1052 /* 1053 * if this protocol has something to add to this then 1054 * allow it to do that as well. 1055 */ 1056 if (ifa->ifa_rtrequest != NULL) 1057 ifa->ifa_rtrequest(req, rt, rtinfo); 1058 1059 /* 1060 * We repeat the same procedure from rt_setgate() here because 1061 * it doesn't fire when we call it there because the node 1062 * hasn't been added to the tree yet. 1063 */ 1064 if (req == RTM_ADD && !(rt->rt_flags & RTF_HOST) && 1065 rt_mask(rt) != NULL) { 1066 struct rtfc_arg arg = { rt, rnh }; 1067 1068 rnh->rnh_walktree_from(rnh, (char *)rt_key(rt), 1069 (char *)rt_mask(rt), 1070 rt_fixchange, &arg); 1071 } 1072 1073 #ifdef ROUTE_DEBUG 1074 if (route_debug) 1075 rt_print(rtinfo, rt); 1076 #endif 1077 /* 1078 * Return the resulting rtentry, 1079 * increasing the number of references by one. 1080 */ 1081 if (ret_nrt != NULL) { 1082 rt->rt_refcnt++; 1083 *ret_nrt = rt; 1084 } 1085 break; 1086 default: 1087 error = EOPNOTSUPP; 1088 } 1089 bad: 1090 #ifdef ROUTE_DEBUG 1091 if (route_debug) { 1092 if (error) 1093 kprintf("rti %p failed error %d\n", rtinfo, error); 1094 else 1095 kprintf("rti %p succeeded\n", rtinfo); 1096 } 1097 #endif 1098 crit_exit(); 1099 return (error); 1100 } 1101 1102 /* 1103 * Called from rtrequest(RTM_DELETE, ...) to fix up the route's ``family'' 1104 * (i.e., the routes related to it by the operation of cloning). This 1105 * routine is iterated over all potential former-child-routes by way of 1106 * rnh->rnh_walktree_from() above, and those that actually are children of 1107 * the late parent (passed in as VP here) are themselves deleted. 1108 */ 1109 static int 1110 rt_fixdelete(struct radix_node *rn, void *vp) 1111 { 1112 struct rtentry *rt = (struct rtentry *)rn; 1113 struct rtentry *rt0 = vp; 1114 1115 if (rt->rt_parent == rt0 && 1116 !(rt->rt_flags & (RTF_PINNED | RTF_CLONING | RTF_PRCLONING))) { 1117 return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 1118 rt->rt_flags, NULL); 1119 } 1120 return 0; 1121 } 1122 1123 /* 1124 * This routine is called from rt_setgate() to do the analogous thing for 1125 * adds and changes. There is the added complication in this case of a 1126 * middle insert; i.e., insertion of a new network route between an older 1127 * network route and (cloned) host routes. For this reason, a simple check 1128 * of rt->rt_parent is insufficient; each candidate route must be tested 1129 * against the (mask, value) of the new route (passed as before in vp) 1130 * to see if the new route matches it. 1131 * 1132 * XXX - it may be possible to do fixdelete() for changes and reserve this 1133 * routine just for adds. I'm not sure why I thought it was necessary to do 1134 * changes this way. 1135 */ 1136 #ifdef DEBUG 1137 static int rtfcdebug = 0; 1138 #endif 1139 1140 static int 1141 rt_fixchange(struct radix_node *rn, void *vp) 1142 { 1143 struct rtentry *rt = (struct rtentry *)rn; 1144 struct rtfc_arg *ap = vp; 1145 struct rtentry *rt0 = ap->rt0; 1146 struct radix_node_head *rnh = ap->rnh; 1147 u_char *xk1, *xm1, *xk2, *xmp; 1148 int i, len, mlen; 1149 1150 #ifdef DEBUG 1151 if (rtfcdebug) 1152 kprintf("rt_fixchange: rt %p, rt0 %p\n", rt, rt0); 1153 #endif 1154 1155 if (rt->rt_parent == NULL || 1156 (rt->rt_flags & (RTF_PINNED | RTF_CLONING | RTF_PRCLONING))) { 1157 #ifdef DEBUG 1158 if (rtfcdebug) kprintf("no parent, pinned or cloning\n"); 1159 #endif 1160 return 0; 1161 } 1162 1163 if (rt->rt_parent == rt0) { 1164 #ifdef DEBUG 1165 if (rtfcdebug) kprintf("parent match\n"); 1166 #endif 1167 return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 1168 rt->rt_flags, NULL); 1169 } 1170 1171 /* 1172 * There probably is a function somewhere which does this... 1173 * if not, there should be. 1174 */ 1175 len = imin(rt_key(rt0)->sa_len, rt_key(rt)->sa_len); 1176 1177 xk1 = (u_char *)rt_key(rt0); 1178 xm1 = (u_char *)rt_mask(rt0); 1179 xk2 = (u_char *)rt_key(rt); 1180 1181 /* avoid applying a less specific route */ 1182 xmp = (u_char *)rt_mask(rt->rt_parent); 1183 mlen = rt_key(rt->rt_parent)->sa_len; 1184 if (mlen > rt_key(rt0)->sa_len) { 1185 #ifdef DEBUG 1186 if (rtfcdebug) 1187 kprintf("rt_fixchange: inserting a less " 1188 "specific route\n"); 1189 #endif 1190 return 0; 1191 } 1192 for (i = rnh->rnh_treetop->rn_offset; i < mlen; i++) { 1193 if ((xmp[i] & ~(xmp[i] ^ xm1[i])) != xmp[i]) { 1194 #ifdef DEBUG 1195 if (rtfcdebug) 1196 kprintf("rt_fixchange: inserting a less " 1197 "specific route\n"); 1198 #endif 1199 return 0; 1200 } 1201 } 1202 1203 for (i = rnh->rnh_treetop->rn_offset; i < len; i++) { 1204 if ((xk2[i] & xm1[i]) != xk1[i]) { 1205 #ifdef DEBUG 1206 if (rtfcdebug) kprintf("no match\n"); 1207 #endif 1208 return 0; 1209 } 1210 } 1211 1212 /* 1213 * OK, this node is a clone, and matches the node currently being 1214 * changed/added under the node's mask. So, get rid of it. 1215 */ 1216 #ifdef DEBUG 1217 if (rtfcdebug) kprintf("deleting\n"); 1218 #endif 1219 return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 1220 rt->rt_flags, NULL); 1221 } 1222 1223 #define ROUNDUP(a) (a>0 ? (1 + (((a) - 1) | (sizeof(long) - 1))) : sizeof(long)) 1224 1225 int 1226 rt_setgate(struct rtentry *rt0, struct sockaddr *dst, struct sockaddr *gate, 1227 boolean_t generate_report) 1228 { 1229 char *space, *oldspace; 1230 int dlen = ROUNDUP(dst->sa_len), glen = ROUNDUP(gate->sa_len); 1231 struct rtentry *rt = rt0; 1232 struct radix_node_head *rnh = rt_tables[mycpuid][dst->sa_family]; 1233 1234 /* 1235 * A host route with the destination equal to the gateway 1236 * will interfere with keeping LLINFO in the routing 1237 * table, so disallow it. 1238 */ 1239 if (((rt0->rt_flags & (RTF_HOST | RTF_GATEWAY | RTF_LLINFO)) == 1240 (RTF_HOST | RTF_GATEWAY)) && 1241 dst->sa_len == gate->sa_len && 1242 sa_equal(dst, gate)) { 1243 /* 1244 * The route might already exist if this is an RTM_CHANGE 1245 * or a routing redirect, so try to delete it. 1246 */ 1247 if (rt_key(rt0) != NULL) 1248 rtrequest(RTM_DELETE, rt_key(rt0), rt0->rt_gateway, 1249 rt_mask(rt0), rt0->rt_flags, NULL); 1250 return EADDRNOTAVAIL; 1251 } 1252 1253 /* 1254 * Both dst and gateway are stored in the same malloc'ed chunk 1255 * (If I ever get my hands on....) 1256 * if we need to malloc a new chunk, then keep the old one around 1257 * till we don't need it any more. 1258 */ 1259 if (rt->rt_gateway == NULL || glen > ROUNDUP(rt->rt_gateway->sa_len)) { 1260 oldspace = (char *)rt_key(rt); 1261 R_Malloc(space, char *, dlen + glen); 1262 if (space == NULL) 1263 return ENOBUFS; 1264 rt->rt_nodes->rn_key = space; 1265 } else { 1266 space = (char *)rt_key(rt); /* Just use the old space. */ 1267 oldspace = NULL; 1268 } 1269 1270 /* Set the gateway value. */ 1271 rt->rt_gateway = (struct sockaddr *)(space + dlen); 1272 bcopy(gate, rt->rt_gateway, glen); 1273 1274 if (oldspace != NULL) { 1275 /* 1276 * If we allocated a new chunk, preserve the original dst. 1277 * This way, rt_setgate() really just sets the gate 1278 * and leaves the dst field alone. 1279 */ 1280 bcopy(dst, space, dlen); 1281 Free(oldspace); 1282 } 1283 1284 /* 1285 * If there is already a gwroute, it's now almost definitely wrong 1286 * so drop it. 1287 */ 1288 if (rt->rt_gwroute != NULL) { 1289 RTFREE(rt->rt_gwroute); 1290 rt->rt_gwroute = NULL; 1291 } 1292 if (rt->rt_flags & RTF_GATEWAY) { 1293 /* 1294 * Cloning loop avoidance: In the presence of 1295 * protocol-cloning and bad configuration, it is 1296 * possible to get stuck in bottomless mutual recursion 1297 * (rtrequest rt_setgate rtlookup). We avoid this 1298 * by not allowing protocol-cloning to operate for 1299 * gateways (which is probably the correct choice 1300 * anyway), and avoid the resulting reference loops 1301 * by disallowing any route to run through itself as 1302 * a gateway. This is obviously mandatory when we 1303 * get rt->rt_output(). 1304 * 1305 * This breaks TTCP for hosts outside the gateway! XXX JH 1306 */ 1307 rt->rt_gwroute = _rtlookup(gate, generate_report, 1308 RTF_PRCLONING); 1309 if (rt->rt_gwroute == rt) { 1310 rt->rt_gwroute = NULL; 1311 --rt->rt_refcnt; 1312 return EDQUOT; /* failure */ 1313 } 1314 } 1315 1316 /* 1317 * This isn't going to do anything useful for host routes, so 1318 * don't bother. Also make sure we have a reasonable mask 1319 * (we don't yet have one during adds). 1320 */ 1321 if (!(rt->rt_flags & RTF_HOST) && rt_mask(rt) != NULL) { 1322 struct rtfc_arg arg = { rt, rnh }; 1323 1324 rnh->rnh_walktree_from(rnh, (char *)rt_key(rt), 1325 (char *)rt_mask(rt), 1326 rt_fixchange, &arg); 1327 } 1328 1329 return 0; 1330 } 1331 1332 static void 1333 rt_maskedcopy( 1334 struct sockaddr *src, 1335 struct sockaddr *dst, 1336 struct sockaddr *netmask) 1337 { 1338 u_char *cp1 = (u_char *)src; 1339 u_char *cp2 = (u_char *)dst; 1340 u_char *cp3 = (u_char *)netmask; 1341 u_char *cplim = cp2 + *cp3; 1342 u_char *cplim2 = cp2 + *cp1; 1343 1344 *cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */ 1345 cp3 += 2; 1346 if (cplim > cplim2) 1347 cplim = cplim2; 1348 while (cp2 < cplim) 1349 *cp2++ = *cp1++ & *cp3++; 1350 if (cp2 < cplim2) 1351 bzero(cp2, cplim2 - cp2); 1352 } 1353 1354 int 1355 rt_llroute(struct sockaddr *dst, struct rtentry *rt0, struct rtentry **drt) 1356 { 1357 struct rtentry *up_rt, *rt; 1358 1359 if (!(rt0->rt_flags & RTF_UP)) { 1360 up_rt = rtlookup(dst); 1361 if (up_rt == NULL) 1362 return (EHOSTUNREACH); 1363 up_rt->rt_refcnt--; 1364 } else 1365 up_rt = rt0; 1366 if (up_rt->rt_flags & RTF_GATEWAY) { 1367 if (up_rt->rt_gwroute == NULL) { 1368 up_rt->rt_gwroute = rtlookup(up_rt->rt_gateway); 1369 if (up_rt->rt_gwroute == NULL) 1370 return (EHOSTUNREACH); 1371 } else if (!(up_rt->rt_gwroute->rt_flags & RTF_UP)) { 1372 rtfree(up_rt->rt_gwroute); 1373 up_rt->rt_gwroute = rtlookup(up_rt->rt_gateway); 1374 if (up_rt->rt_gwroute == NULL) 1375 return (EHOSTUNREACH); 1376 } 1377 rt = up_rt->rt_gwroute; 1378 } else 1379 rt = up_rt; 1380 if (rt->rt_flags & RTF_REJECT && 1381 (rt->rt_rmx.rmx_expire == 0 || /* rt doesn't expire */ 1382 time_second < rt->rt_rmx.rmx_expire)) /* rt not expired */ 1383 return (rt->rt_flags & RTF_HOST ? EHOSTDOWN : EHOSTUNREACH); 1384 *drt = rt; 1385 return 0; 1386 } 1387 1388 static int 1389 rt_setshims(struct rtentry *rt, struct sockaddr **rt_shim){ 1390 int i; 1391 1392 for (i=0; i<3; i++) { 1393 struct sockaddr *shim = rt_shim[RTAX_MPLS1 + i]; 1394 int shimlen; 1395 1396 if (shim == NULL) 1397 break; 1398 1399 shimlen = ROUNDUP(shim->sa_len); 1400 R_Malloc(rt->rt_shim[i], struct sockaddr *, shimlen); 1401 bcopy(shim, rt->rt_shim[i], shimlen); 1402 } 1403 1404 return 0; 1405 } 1406 1407 #ifdef ROUTE_DEBUG 1408 1409 /* 1410 * Print out a route table entry 1411 */ 1412 void 1413 rt_print(struct rt_addrinfo *rtinfo, struct rtentry *rn) 1414 { 1415 kprintf("rti %p cpu %d route %p flags %08lx: ", 1416 rtinfo, mycpuid, rn, rn->rt_flags); 1417 sockaddr_print(rt_key(rn)); 1418 kprintf(" mask "); 1419 sockaddr_print(rt_mask(rn)); 1420 kprintf(" gw "); 1421 sockaddr_print(rn->rt_gateway); 1422 kprintf(" ifc \"%s\"", rn->rt_ifp ? rn->rt_ifp->if_dname : "?"); 1423 kprintf(" ifa %p\n", rn->rt_ifa); 1424 } 1425 1426 void 1427 rt_addrinfo_print(int cmd, struct rt_addrinfo *rti) 1428 { 1429 int didit = 0; 1430 int i; 1431 1432 #ifdef ROUTE_DEBUG 1433 if (cmd == RTM_DELETE && route_debug > 1) 1434 print_backtrace(-1); 1435 #endif 1436 1437 switch(cmd) { 1438 case RTM_ADD: 1439 kprintf("ADD "); 1440 break; 1441 case RTM_RESOLVE: 1442 kprintf("RES "); 1443 break; 1444 case RTM_DELETE: 1445 kprintf("DEL "); 1446 break; 1447 default: 1448 kprintf("C%02d ", cmd); 1449 break; 1450 } 1451 kprintf("rti %p cpu %d ", rti, mycpuid); 1452 for (i = 0; i < rti->rti_addrs; ++i) { 1453 if (rti->rti_info[i] == NULL) 1454 continue; 1455 if (didit) 1456 kprintf(" ,"); 1457 switch(i) { 1458 case RTAX_DST: 1459 kprintf("(DST "); 1460 break; 1461 case RTAX_GATEWAY: 1462 kprintf("(GWY "); 1463 break; 1464 case RTAX_NETMASK: 1465 kprintf("(MSK "); 1466 break; 1467 case RTAX_GENMASK: 1468 kprintf("(GEN "); 1469 break; 1470 case RTAX_IFP: 1471 kprintf("(IFP "); 1472 break; 1473 case RTAX_IFA: 1474 kprintf("(IFA "); 1475 break; 1476 case RTAX_AUTHOR: 1477 kprintf("(AUT "); 1478 break; 1479 case RTAX_BRD: 1480 kprintf("(BRD "); 1481 break; 1482 default: 1483 kprintf("(?%02d ", i); 1484 break; 1485 } 1486 sockaddr_print(rti->rti_info[i]); 1487 kprintf(")"); 1488 didit = 1; 1489 } 1490 kprintf("\n"); 1491 } 1492 1493 void 1494 sockaddr_print(struct sockaddr *sa) 1495 { 1496 struct sockaddr_in *sa4; 1497 struct sockaddr_in6 *sa6; 1498 int len; 1499 int i; 1500 1501 if (sa == NULL) { 1502 kprintf("NULL"); 1503 return; 1504 } 1505 1506 len = sa->sa_len - offsetof(struct sockaddr, sa_data[0]); 1507 1508 switch(sa->sa_family) { 1509 case AF_INET: 1510 case AF_INET6: 1511 default: 1512 switch(sa->sa_family) { 1513 case AF_INET: 1514 sa4 = (struct sockaddr_in *)sa; 1515 kprintf("INET %d %d.%d.%d.%d", 1516 ntohs(sa4->sin_port), 1517 (ntohl(sa4->sin_addr.s_addr) >> 24) & 255, 1518 (ntohl(sa4->sin_addr.s_addr) >> 16) & 255, 1519 (ntohl(sa4->sin_addr.s_addr) >> 8) & 255, 1520 (ntohl(sa4->sin_addr.s_addr) >> 0) & 255 1521 ); 1522 break; 1523 case AF_INET6: 1524 sa6 = (struct sockaddr_in6 *)sa; 1525 kprintf("INET6 %d %04x:%04x%04x:%04x:%04x:%04x:%04x:%04x", 1526 ntohs(sa6->sin6_port), 1527 sa6->sin6_addr.s6_addr16[0], 1528 sa6->sin6_addr.s6_addr16[1], 1529 sa6->sin6_addr.s6_addr16[2], 1530 sa6->sin6_addr.s6_addr16[3], 1531 sa6->sin6_addr.s6_addr16[4], 1532 sa6->sin6_addr.s6_addr16[5], 1533 sa6->sin6_addr.s6_addr16[6], 1534 sa6->sin6_addr.s6_addr16[7] 1535 ); 1536 break; 1537 default: 1538 kprintf("AF%d ", sa->sa_family); 1539 while (len > 0 && sa->sa_data[len-1] == 0) 1540 --len; 1541 1542 for (i = 0; i < len; ++i) { 1543 if (i) 1544 kprintf("."); 1545 kprintf("%d", (unsigned char)sa->sa_data[i]); 1546 } 1547 break; 1548 } 1549 } 1550 } 1551 1552 #endif 1553 1554 /* 1555 * Set up a routing table entry, normally for an interface. 1556 */ 1557 int 1558 rtinit(struct ifaddr *ifa, int cmd, int flags) 1559 { 1560 struct sockaddr *dst, *deldst, *netmask; 1561 struct mbuf *m = NULL; 1562 struct radix_node_head *rnh; 1563 struct radix_node *rn; 1564 struct rt_addrinfo rtinfo; 1565 int error; 1566 1567 if (flags & RTF_HOST) { 1568 dst = ifa->ifa_dstaddr; 1569 netmask = NULL; 1570 } else { 1571 dst = ifa->ifa_addr; 1572 netmask = ifa->ifa_netmask; 1573 } 1574 /* 1575 * If it's a delete, check that if it exists, it's on the correct 1576 * interface or we might scrub a route to another ifa which would 1577 * be confusing at best and possibly worse. 1578 */ 1579 if (cmd == RTM_DELETE) { 1580 /* 1581 * It's a delete, so it should already exist.. 1582 * If it's a net, mask off the host bits 1583 * (Assuming we have a mask) 1584 */ 1585 if (netmask != NULL) { 1586 m = m_get(MB_DONTWAIT, MT_SONAME); 1587 if (m == NULL) 1588 return (ENOBUFS); 1589 mbuftrackid(m, 34); 1590 deldst = mtod(m, struct sockaddr *); 1591 rt_maskedcopy(dst, deldst, netmask); 1592 dst = deldst; 1593 } 1594 /* 1595 * Look up an rtentry that is in the routing tree and 1596 * contains the correct info. 1597 */ 1598 if ((rnh = rt_tables[mycpuid][dst->sa_family]) == NULL || 1599 (rn = rnh->rnh_lookup((char *)dst, 1600 (char *)netmask, rnh)) == NULL || 1601 ((struct rtentry *)rn)->rt_ifa != ifa || 1602 !sa_equal((struct sockaddr *)rn->rn_key, dst)) { 1603 if (m != NULL) 1604 m_free(m); 1605 return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH); 1606 } 1607 /* XXX */ 1608 #if 0 1609 else { 1610 /* 1611 * One would think that as we are deleting, and we know 1612 * it doesn't exist, we could just return at this point 1613 * with an "ELSE" clause, but apparently not.. 1614 */ 1615 return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH); 1616 } 1617 #endif 1618 } 1619 /* 1620 * Do the actual request 1621 */ 1622 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 1623 rtinfo.rti_info[RTAX_DST] = dst; 1624 rtinfo.rti_info[RTAX_GATEWAY] = ifa->ifa_addr; 1625 rtinfo.rti_info[RTAX_NETMASK] = netmask; 1626 rtinfo.rti_flags = flags | ifa->ifa_flags; 1627 rtinfo.rti_ifa = ifa; 1628 error = rtrequest1_global(cmd, &rtinfo, rtinit_rtrequest_callback, ifa); 1629 if (m != NULL) 1630 m_free(m); 1631 return (error); 1632 } 1633 1634 static void 1635 rtinit_rtrequest_callback(int cmd, int error, 1636 struct rt_addrinfo *rtinfo, struct rtentry *rt, 1637 void *arg) 1638 { 1639 struct ifaddr *ifa = arg; 1640 1641 if (error == 0 && rt) { 1642 if (mycpuid == 0) { 1643 ++rt->rt_refcnt; 1644 rt_newaddrmsg(cmd, ifa, error, rt); 1645 --rt->rt_refcnt; 1646 } 1647 if (cmd == RTM_DELETE) { 1648 if (rt->rt_refcnt == 0) { 1649 ++rt->rt_refcnt; 1650 rtfree(rt); 1651 } 1652 } 1653 } 1654 } 1655 1656 struct netmsg_rts { 1657 struct netmsg_base base; 1658 int req; 1659 struct rt_addrinfo *rtinfo; 1660 rtsearch_callback_func_t callback; 1661 void *arg; 1662 boolean_t exact_match; 1663 int found_cnt; 1664 }; 1665 1666 int 1667 rtsearch_global(int req, struct rt_addrinfo *rtinfo, 1668 rtsearch_callback_func_t callback, void *arg, 1669 boolean_t exact_match) 1670 { 1671 struct netmsg_rts msg; 1672 1673 netmsg_init(&msg.base, NULL, &curthread->td_msgport, 1674 0, rtsearch_msghandler); 1675 msg.req = req; 1676 msg.rtinfo = rtinfo; 1677 msg.callback = callback; 1678 msg.arg = arg; 1679 msg.exact_match = exact_match; 1680 msg.found_cnt = 0; 1681 return lwkt_domsg(rtable_portfn(0), &msg.base.lmsg, 0); 1682 } 1683 1684 static void 1685 rtsearch_msghandler(netmsg_t msg) 1686 { 1687 struct netmsg_rts *rmsg = (void *)msg; 1688 struct rt_addrinfo rtinfo; 1689 struct radix_node_head *rnh; 1690 struct rtentry *rt; 1691 int nextcpu, error; 1692 1693 /* 1694 * Copy the rtinfo. We need to make sure that the original 1695 * rtinfo, which is setup by the caller, in the netmsg will 1696 * _not_ be changed; else the next CPU on the netmsg forwarding 1697 * path will see a different rtinfo than what this CPU has seen. 1698 */ 1699 rtinfo = *rmsg->rtinfo; 1700 1701 /* 1702 * Find the correct routing tree to use for this Address Family 1703 */ 1704 if ((rnh = rt_tables[mycpuid][rtinfo.rti_dst->sa_family]) == NULL) { 1705 if (mycpuid != 0) 1706 panic("partially initialized routing tables\n"); 1707 lwkt_replymsg(&rmsg->base.lmsg, EAFNOSUPPORT); 1708 return; 1709 } 1710 1711 /* 1712 * Correct rtinfo for the host route searching. 1713 */ 1714 if (rtinfo.rti_flags & RTF_HOST) { 1715 rtinfo.rti_netmask = NULL; 1716 rtinfo.rti_flags &= ~(RTF_CLONING | RTF_PRCLONING); 1717 } 1718 1719 rt = (struct rtentry *) 1720 rnh->rnh_lookup((char *)rtinfo.rti_dst, 1721 (char *)rtinfo.rti_netmask, rnh); 1722 1723 /* 1724 * If we are asked to do the "exact match", we need to make sure 1725 * that host route searching got a host route while a network 1726 * route searching got a network route. 1727 */ 1728 if (rt != NULL && rmsg->exact_match && 1729 ((rt->rt_flags ^ rtinfo.rti_flags) & RTF_HOST)) 1730 rt = NULL; 1731 1732 if (rt == NULL) { 1733 /* 1734 * No matching routes have been found, don't count this 1735 * as a critical error (here, we set 'error' to 0), just 1736 * keep moving on, since at least prcloned routes are not 1737 * duplicated onto each CPU. 1738 */ 1739 error = 0; 1740 } else { 1741 rmsg->found_cnt++; 1742 1743 rt->rt_refcnt++; 1744 error = rmsg->callback(rmsg->req, &rtinfo, rt, rmsg->arg, 1745 rmsg->found_cnt); 1746 rt->rt_refcnt--; 1747 1748 if (error == EJUSTRETURN) { 1749 lwkt_replymsg(&rmsg->base.lmsg, 0); 1750 return; 1751 } 1752 } 1753 1754 nextcpu = mycpuid + 1; 1755 if (error) { 1756 KKASSERT(rmsg->found_cnt > 0); 1757 1758 /* 1759 * Under following cases, unrecoverable error has 1760 * not occured: 1761 * o Request is RTM_GET 1762 * o The first time that we find the route, but the 1763 * modification fails. 1764 */ 1765 if (rmsg->req != RTM_GET && rmsg->found_cnt > 1) { 1766 panic("rtsearch_msghandler: unrecoverable error " 1767 "cpu %d", mycpuid); 1768 } 1769 lwkt_replymsg(&rmsg->base.lmsg, error); 1770 } else if (nextcpu < ncpus) { 1771 lwkt_forwardmsg(rtable_portfn(nextcpu), &rmsg->base.lmsg); 1772 } else { 1773 if (rmsg->found_cnt == 0) { 1774 /* The requested route was never seen ... */ 1775 error = ESRCH; 1776 } 1777 lwkt_replymsg(&rmsg->base.lmsg, error); 1778 } 1779 } 1780 1781 int 1782 rtmask_add_global(struct sockaddr *mask) 1783 { 1784 struct netmsg_base msg; 1785 1786 netmsg_init(&msg, NULL, &curthread->td_msgport, 1787 0, rtmask_add_msghandler); 1788 msg.lmsg.u.ms_resultp = mask; 1789 1790 return lwkt_domsg(rtable_portfn(0), &msg.lmsg, 0); 1791 } 1792 1793 struct sockaddr * 1794 _rtmask_lookup(struct sockaddr *mask, boolean_t search) 1795 { 1796 struct radix_node *n; 1797 1798 #define clen(s) (*(u_char *)(s)) 1799 n = rn_addmask((char *)mask, search, 1, rn_cpumaskhead(mycpuid)); 1800 if (n != NULL && 1801 mask->sa_len >= clen(n->rn_key) && 1802 bcmp((char *)mask + 1, 1803 (char *)n->rn_key + 1, clen(n->rn_key) - 1) == 0) { 1804 return (struct sockaddr *)n->rn_key; 1805 } else { 1806 return NULL; 1807 } 1808 #undef clen 1809 } 1810 1811 static void 1812 rtmask_add_msghandler(netmsg_t msg) 1813 { 1814 struct lwkt_msg *lmsg = &msg->lmsg; 1815 struct sockaddr *mask = lmsg->u.ms_resultp; 1816 int error = 0, nextcpu; 1817 1818 if (rtmask_lookup(mask) == NULL) 1819 error = ENOBUFS; 1820 1821 nextcpu = mycpuid + 1; 1822 if (!error && nextcpu < ncpus) 1823 lwkt_forwardmsg(rtable_portfn(nextcpu), lmsg); 1824 else 1825 lwkt_replymsg(lmsg, error); 1826 } 1827 1828 /* This must be before ip6_init2(), which is now SI_ORDER_MIDDLE */ 1829 SYSINIT(route, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0); 1830