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. Neither the name of the University nor the names of its contributors 46 * may be used to endorse or promote products derived from this software 47 * without specific prior written permission. 48 * 49 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 50 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 51 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 52 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 53 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 54 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 55 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 56 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 57 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 58 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 59 * SUCH DAMAGE. 60 * 61 * @(#)route.c 8.3 (Berkeley) 1/9/95 62 * $FreeBSD: src/sys/net/route.c,v 1.59.2.10 2003/01/17 08:04:00 ru Exp $ 63 */ 64 65 #include "opt_inet.h" 66 #include "opt_mpls.h" 67 68 #include <sys/param.h> 69 #include <sys/systm.h> 70 #include <sys/malloc.h> 71 #include <sys/mbuf.h> 72 #include <sys/socket.h> 73 #include <sys/domain.h> 74 #include <sys/kernel.h> 75 #include <sys/sysctl.h> 76 #include <sys/globaldata.h> 77 #include <sys/thread.h> 78 79 #include <net/if.h> 80 #include <net/route.h> 81 #include <net/netisr.h> 82 83 #include <netinet/in.h> 84 #include <net/ip_mroute/ip_mroute.h> 85 86 #include <sys/thread2.h> 87 #include <sys/msgport2.h> 88 #include <net/netmsg2.h> 89 #include <net/netisr2.h> 90 91 #ifdef MPLS 92 #include <netproto/mpls/mpls.h> 93 #endif 94 95 static struct rtstatistics rtstatistics_percpu[MAXCPU]; 96 #define rtstat rtstatistics_percpu[mycpuid] 97 98 struct radix_node_head *rt_tables[MAXCPU][AF_MAX+1]; 99 100 static void rt_maskedcopy (struct sockaddr *, struct sockaddr *, 101 struct sockaddr *); 102 static void rtable_init(void); 103 static void rtinit_rtrequest_callback(int, int, struct rt_addrinfo *, 104 struct rtentry *, void *); 105 106 static void rtredirect_msghandler(netmsg_t msg); 107 static void rtrequest1_msghandler(netmsg_t msg); 108 static void rtsearch_msghandler(netmsg_t msg); 109 static void rtmask_add_msghandler(netmsg_t msg); 110 111 static int rt_setshims(struct rtentry *, struct sockaddr **); 112 113 SYSCTL_NODE(_net, OID_AUTO, route, CTLFLAG_RW, 0, "Routing"); 114 115 #ifdef ROUTE_DEBUG 116 static int route_debug = 1; 117 SYSCTL_INT(_net_route, OID_AUTO, route_debug, CTLFLAG_RW, 118 &route_debug, 0, ""); 119 #endif 120 121 int route_assert_owner_access = 1; 122 SYSCTL_INT(_net_route, OID_AUTO, assert_owner_access, CTLFLAG_RW, 123 &route_assert_owner_access, 0, ""); 124 125 u_long route_kmalloc_limit = 0; 126 TUNABLE_ULONG("net.route.kmalloc_limit", &route_kmalloc_limit); 127 128 /* 129 * Initialize the route table(s) for protocol domains and 130 * create a helper thread which will be responsible for updating 131 * route table entries on each cpu. 132 */ 133 void 134 route_init(void) 135 { 136 int cpu; 137 138 for (cpu = 0; cpu < ncpus; ++cpu) 139 bzero(&rtstatistics_percpu[cpu], sizeof(struct rtstatistics)); 140 rn_init(); /* initialize all zeroes, all ones, mask table */ 141 rtable_init(); /* call dom_rtattach() on each cpu */ 142 143 if (route_kmalloc_limit) 144 kmalloc_raise_limit(M_RTABLE, route_kmalloc_limit); 145 } 146 147 static void 148 rtable_init_oncpu(netmsg_t msg) 149 { 150 struct domain *dom; 151 int cpu = mycpuid; 152 153 SLIST_FOREACH(dom, &domains, dom_next) { 154 if (dom->dom_rtattach) { 155 dom->dom_rtattach( 156 (void **)&rt_tables[cpu][dom->dom_family], 157 dom->dom_rtoffset); 158 } 159 } 160 ifnet_forwardmsg(&msg->lmsg, cpu + 1); 161 } 162 163 static void 164 rtable_init(void) 165 { 166 struct netmsg_base msg; 167 168 netmsg_init(&msg, NULL, &curthread->td_msgport, 0, rtable_init_oncpu); 169 ifnet_domsg(&msg.lmsg, 0); 170 } 171 172 /* 173 * Routing statistics. 174 */ 175 static int 176 sysctl_rtstatistics(SYSCTL_HANDLER_ARGS) 177 { 178 int cpu, error = 0; 179 180 for (cpu = 0; cpu < ncpus; ++cpu) { 181 if ((error = SYSCTL_OUT(req, &rtstatistics_percpu[cpu], 182 sizeof(struct rtstatistics)))) 183 break; 184 if ((error = SYSCTL_IN(req, &rtstatistics_percpu[cpu], 185 sizeof(struct rtstatistics)))) 186 break; 187 } 188 189 return (error); 190 } 191 SYSCTL_PROC(_net_route, OID_AUTO, stats, (CTLTYPE_OPAQUE|CTLFLAG_RW), 192 0, 0, sysctl_rtstatistics, "S,rtstatistics", "Routing statistics"); 193 194 /* 195 * Packet routing routines. 196 */ 197 198 /* 199 * Look up and fill in the "ro_rt" rtentry field in a route structure given 200 * an address in the "ro_dst" field. Always send a report on a miss and 201 * always clone routes. 202 */ 203 void 204 rtalloc(struct route *ro) 205 { 206 rtalloc_ign(ro, 0UL); 207 } 208 209 /* 210 * Look up and fill in the "ro_rt" rtentry field in a route structure given 211 * an address in the "ro_dst" field. Always send a report on a miss and 212 * optionally clone routes when RTF_CLONING or RTF_PRCLONING are not being 213 * ignored. 214 */ 215 void 216 rtalloc_ign(struct route *ro, u_long ignoreflags) 217 { 218 if (ro->ro_rt != NULL) { 219 if (ro->ro_rt->rt_ifp != NULL && ro->ro_rt->rt_flags & RTF_UP) 220 return; 221 rtfree(ro->ro_rt); 222 ro->ro_rt = NULL; 223 } 224 ro->ro_rt = _rtlookup(&ro->ro_dst, RTL_REPORTMSG, ignoreflags); 225 } 226 227 /* 228 * Look up the route that matches the given "dst" address. 229 * 230 * Route lookup can have the side-effect of creating and returning 231 * a cloned route instead when "dst" matches a cloning route and the 232 * RTF_CLONING and RTF_PRCLONING flags are not being ignored. 233 * 234 * Any route returned has its reference count incremented. 235 */ 236 struct rtentry * 237 _rtlookup(struct sockaddr *dst, boolean_t generate_report, u_long ignore) 238 { 239 struct radix_node_head *rnh = rt_tables[mycpuid][dst->sa_family]; 240 struct rtentry *rt; 241 242 if (rnh == NULL) 243 goto unreach; 244 245 /* 246 * Look up route in the radix tree. 247 */ 248 rt = (struct rtentry *) rnh->rnh_matchaddr((char *)dst, rnh); 249 if (rt == NULL) 250 goto unreach; 251 252 /* 253 * Handle cloning routes. 254 */ 255 if ((rt->rt_flags & ~ignore & (RTF_CLONING | RTF_PRCLONING)) != 0) { 256 struct rtentry *clonedroute; 257 int error; 258 259 clonedroute = rt; /* copy in/copy out parameter */ 260 error = rtrequest(RTM_RESOLVE, dst, NULL, NULL, 0, 261 &clonedroute); /* clone the route */ 262 if (error != 0) { /* cloning failed */ 263 if (generate_report) 264 rt_dstmsg(RTM_MISS, dst, error); 265 rt->rt_refcnt++; 266 return (rt); /* return the uncloned route */ 267 } 268 if (generate_report) { 269 if (clonedroute->rt_flags & RTF_XRESOLVE) 270 rt_dstmsg(RTM_RESOLVE, dst, 0); 271 else 272 rt_rtmsg(RTM_ADD, clonedroute, 273 clonedroute->rt_ifp, 0); 274 } 275 return (clonedroute); /* return cloned route */ 276 } 277 278 /* 279 * Increment the reference count of the matched route and return. 280 */ 281 rt->rt_refcnt++; 282 return (rt); 283 284 unreach: 285 rtstat.rts_unreach++; 286 if (generate_report) 287 rt_dstmsg(RTM_MISS, dst, 0); 288 return (NULL); 289 } 290 291 void 292 rtfree(struct rtentry *rt) 293 { 294 if (rt->rt_cpuid == mycpuid) 295 rtfree_oncpu(rt); 296 else 297 rtfree_remote(rt); 298 } 299 300 void 301 rtfree_oncpu(struct rtentry *rt) 302 { 303 KKASSERT(rt->rt_cpuid == mycpuid); 304 KASSERT(rt->rt_refcnt > 0, ("rtfree: rt_refcnt %ld", rt->rt_refcnt)); 305 306 --rt->rt_refcnt; 307 if (rt->rt_refcnt == 0) { 308 struct radix_node_head *rnh = 309 rt_tables[mycpuid][rt_key(rt)->sa_family]; 310 311 if (rnh->rnh_close) 312 rnh->rnh_close((struct radix_node *)rt, rnh); 313 if (!(rt->rt_flags & RTF_UP)) { 314 /* deallocate route */ 315 if (rt->rt_ifa != NULL) 316 IFAFREE(rt->rt_ifa); 317 if (rt->rt_parent != NULL) 318 RTFREE(rt->rt_parent); /* recursive call! */ 319 Free(rt_key(rt)); 320 Free(rt); 321 } 322 } 323 } 324 325 static void 326 rtfree_remote_dispatch(netmsg_t msg) 327 { 328 struct lwkt_msg *lmsg = &msg->lmsg; 329 struct rtentry *rt = lmsg->u.ms_resultp; 330 331 rtfree_oncpu(rt); 332 lwkt_replymsg(lmsg, 0); 333 } 334 335 void 336 rtfree_remote(struct rtentry *rt) 337 { 338 struct netmsg_base *msg; 339 struct lwkt_msg *lmsg; 340 341 KKASSERT(rt->rt_cpuid != mycpuid); 342 343 if (route_assert_owner_access) { 344 panic("rt remote free rt_cpuid %d, mycpuid %d", 345 rt->rt_cpuid, mycpuid); 346 } else { 347 kprintf("rt remote free rt_cpuid %d, mycpuid %d\n", 348 rt->rt_cpuid, mycpuid); 349 print_backtrace(-1); 350 } 351 352 msg = kmalloc(sizeof(*msg), M_LWKTMSG, M_INTWAIT); 353 netmsg_init(msg, NULL, &netisr_afree_rport, 0, rtfree_remote_dispatch); 354 lmsg = &msg->lmsg; 355 lmsg->u.ms_resultp = rt; 356 357 lwkt_sendmsg(netisr_cpuport(rt->rt_cpuid), lmsg); 358 } 359 360 static int 361 rtredirect_oncpu(struct sockaddr *dst, struct sockaddr *gateway, 362 struct sockaddr *netmask, int flags, struct sockaddr *src) 363 { 364 struct rtentry *rt = NULL; 365 struct rt_addrinfo rtinfo; 366 struct ifaddr *ifa; 367 u_long *stat = NULL; 368 int error; 369 370 /* verify the gateway is directly reachable */ 371 if ((ifa = ifa_ifwithnet(gateway)) == NULL) { 372 error = ENETUNREACH; 373 goto out; 374 } 375 376 /* 377 * If the redirect isn't from our current router for this destination, 378 * it's either old or wrong. 379 */ 380 if (!(flags & RTF_DONE) && /* XXX JH */ 381 (rt = rtpurelookup(dst)) != NULL && 382 (!sa_equal(src, rt->rt_gateway) || rt->rt_ifa != ifa)) { 383 error = EINVAL; 384 goto done; 385 } 386 387 /* 388 * If it redirects us to ourselves, we have a routing loop, 389 * perhaps as a result of an interface going down recently. 390 */ 391 if (ifa_ifwithaddr(gateway)) { 392 error = EHOSTUNREACH; 393 goto done; 394 } 395 396 /* 397 * Create a new entry if the lookup failed or if we got back 398 * a wildcard entry for the default route. This is necessary 399 * for hosts which use routing redirects generated by smart 400 * gateways to dynamically build the routing tables. 401 */ 402 if (rt == NULL) 403 goto create; 404 if ((rt_mask(rt) != NULL && rt_mask(rt)->sa_len < 2)) { 405 rtfree(rt); 406 goto create; 407 } 408 409 /* Ignore redirects for directly connected hosts. */ 410 if (!(rt->rt_flags & RTF_GATEWAY)) { 411 error = EHOSTUNREACH; 412 goto done; 413 } 414 415 if (!(rt->rt_flags & RTF_HOST) && (flags & RTF_HOST)) { 416 /* 417 * Changing from a network route to a host route. 418 * Create a new host route rather than smashing the 419 * network route. 420 */ 421 create: 422 flags |= RTF_GATEWAY | RTF_DYNAMIC; 423 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 424 rtinfo.rti_info[RTAX_DST] = dst; 425 rtinfo.rti_info[RTAX_GATEWAY] = gateway; 426 rtinfo.rti_info[RTAX_NETMASK] = netmask; 427 rtinfo.rti_flags = flags; 428 rtinfo.rti_ifa = ifa; 429 rt = NULL; /* copy-in/copy-out parameter */ 430 error = rtrequest1(RTM_ADD, &rtinfo, &rt); 431 if (rt != NULL) 432 flags = rt->rt_flags; 433 stat = &rtstat.rts_dynamic; 434 } else { 435 /* 436 * Smash the current notion of the gateway to this destination. 437 * Should check about netmask!!! 438 */ 439 rt->rt_flags |= RTF_MODIFIED; 440 flags |= RTF_MODIFIED; 441 442 /* We only need to report rtmsg on CPU0 */ 443 rt_setgate(rt, rt_key(rt), gateway, 444 mycpuid == 0 ? RTL_REPORTMSG : RTL_DONTREPORT); 445 error = 0; 446 stat = &rtstat.rts_newgateway; 447 } 448 449 done: 450 if (rt != NULL) 451 rtfree(rt); 452 out: 453 if (error != 0) 454 rtstat.rts_badredirect++; 455 else if (stat != NULL) 456 (*stat)++; 457 458 return error; 459 } 460 461 struct netmsg_rtredirect { 462 struct netmsg_base base; 463 struct sockaddr *dst; 464 struct sockaddr *gateway; 465 struct sockaddr *netmask; 466 int flags; 467 struct sockaddr *src; 468 }; 469 470 /* 471 * Force a routing table entry to the specified 472 * destination to go through the given gateway. 473 * Normally called as a result of a routing redirect 474 * message from the network layer. 475 * 476 * N.B.: must be called at splnet 477 */ 478 void 479 rtredirect(struct sockaddr *dst, struct sockaddr *gateway, 480 struct sockaddr *netmask, int flags, struct sockaddr *src) 481 { 482 struct rt_addrinfo rtinfo; 483 int error; 484 struct netmsg_rtredirect msg; 485 486 netmsg_init(&msg.base, NULL, &curthread->td_msgport, 487 0, rtredirect_msghandler); 488 msg.dst = dst; 489 msg.gateway = gateway; 490 msg.netmask = netmask; 491 msg.flags = flags; 492 msg.src = src; 493 error = rt_domsg_global(&msg.base); 494 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 495 rtinfo.rti_info[RTAX_DST] = dst; 496 rtinfo.rti_info[RTAX_GATEWAY] = gateway; 497 rtinfo.rti_info[RTAX_NETMASK] = netmask; 498 rtinfo.rti_info[RTAX_AUTHOR] = src; 499 rt_missmsg(RTM_REDIRECT, &rtinfo, flags, error); 500 } 501 502 static void 503 rtredirect_msghandler(netmsg_t msg) 504 { 505 struct netmsg_rtredirect *rmsg = (void *)msg; 506 int nextcpu; 507 508 rtredirect_oncpu(rmsg->dst, rmsg->gateway, rmsg->netmask, 509 rmsg->flags, rmsg->src); 510 nextcpu = mycpuid + 1; 511 if (nextcpu < ncpus) 512 lwkt_forwardmsg(netisr_cpuport(nextcpu), &msg->lmsg); 513 else 514 lwkt_replymsg(&msg->lmsg, 0); 515 } 516 517 /* 518 * Routing table ioctl interface. 519 */ 520 int 521 rtioctl(u_long req, caddr_t data, struct ucred *cred) 522 { 523 #ifdef INET 524 /* Multicast goop, grrr... */ 525 return mrt_ioctl ? mrt_ioctl(req, data) : EOPNOTSUPP; 526 #else 527 return ENXIO; 528 #endif 529 } 530 531 struct ifaddr * 532 ifa_ifwithroute(int flags, struct sockaddr *dst, struct sockaddr *gateway) 533 { 534 struct ifaddr *ifa; 535 536 if (!(flags & RTF_GATEWAY)) { 537 /* 538 * If we are adding a route to an interface, 539 * and the interface is a point-to-point link, 540 * we should search for the destination 541 * as our clue to the interface. Otherwise 542 * we can use the local address. 543 */ 544 ifa = NULL; 545 if (flags & RTF_HOST) { 546 ifa = ifa_ifwithdstaddr(dst); 547 } 548 if (ifa == NULL) 549 ifa = ifa_ifwithaddr(gateway); 550 } else { 551 /* 552 * If we are adding a route to a remote net 553 * or host, the gateway may still be on the 554 * other end of a pt to pt link. 555 */ 556 ifa = ifa_ifwithdstaddr(gateway); 557 } 558 if (ifa == NULL) 559 ifa = ifa_ifwithnet(gateway); 560 if (ifa == NULL) { 561 struct rtentry *rt; 562 563 rt = rtpurelookup(gateway); 564 if (rt == NULL) 565 return (NULL); 566 rt->rt_refcnt--; 567 if ((ifa = rt->rt_ifa) == NULL) 568 return (NULL); 569 } 570 if (ifa->ifa_addr->sa_family != dst->sa_family) { 571 struct ifaddr *oldifa = ifa; 572 573 ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp); 574 if (ifa == NULL) 575 ifa = oldifa; 576 } 577 return (ifa); 578 } 579 580 static int rt_fixdelete (struct radix_node *, void *); 581 static int rt_fixchange (struct radix_node *, void *); 582 583 struct rtfc_arg { 584 struct rtentry *rt0; 585 struct radix_node_head *rnh; 586 }; 587 588 /* 589 * Set rtinfo->rti_ifa and rtinfo->rti_ifp. 590 */ 591 int 592 rt_getifa(struct rt_addrinfo *rtinfo) 593 { 594 struct sockaddr *gateway = rtinfo->rti_info[RTAX_GATEWAY]; 595 struct sockaddr *dst = rtinfo->rti_info[RTAX_DST]; 596 struct sockaddr *ifaaddr = rtinfo->rti_info[RTAX_IFA]; 597 int flags = rtinfo->rti_flags; 598 599 /* 600 * ifp may be specified by sockaddr_dl 601 * when protocol address is ambiguous. 602 */ 603 if (rtinfo->rti_ifp == NULL) { 604 struct sockaddr *ifpaddr; 605 606 ifpaddr = rtinfo->rti_info[RTAX_IFP]; 607 if (ifpaddr != NULL && ifpaddr->sa_family == AF_LINK) { 608 struct ifaddr *ifa; 609 610 ifa = ifa_ifwithnet(ifpaddr); 611 if (ifa != NULL) 612 rtinfo->rti_ifp = ifa->ifa_ifp; 613 } 614 } 615 616 if (rtinfo->rti_ifa == NULL && ifaaddr != NULL) 617 rtinfo->rti_ifa = ifa_ifwithaddr(ifaaddr); 618 if (rtinfo->rti_ifa == NULL) { 619 struct sockaddr *sa; 620 621 sa = ifaaddr != NULL ? ifaaddr : 622 (gateway != NULL ? gateway : dst); 623 if (sa != NULL && rtinfo->rti_ifp != NULL) 624 rtinfo->rti_ifa = ifaof_ifpforaddr(sa, rtinfo->rti_ifp); 625 else if (dst != NULL && gateway != NULL) 626 rtinfo->rti_ifa = ifa_ifwithroute(flags, dst, gateway); 627 else if (sa != NULL) 628 rtinfo->rti_ifa = ifa_ifwithroute(flags, sa, sa); 629 } 630 if (rtinfo->rti_ifa == NULL) 631 return (ENETUNREACH); 632 633 if (rtinfo->rti_ifp == NULL) 634 rtinfo->rti_ifp = rtinfo->rti_ifa->ifa_ifp; 635 return (0); 636 } 637 638 /* 639 * Do appropriate manipulations of a routing tree given 640 * all the bits of info needed 641 */ 642 int 643 rtrequest( 644 int req, 645 struct sockaddr *dst, 646 struct sockaddr *gateway, 647 struct sockaddr *netmask, 648 int flags, 649 struct rtentry **ret_nrt) 650 { 651 struct rt_addrinfo rtinfo; 652 653 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 654 rtinfo.rti_info[RTAX_DST] = dst; 655 rtinfo.rti_info[RTAX_GATEWAY] = gateway; 656 rtinfo.rti_info[RTAX_NETMASK] = netmask; 657 rtinfo.rti_flags = flags; 658 return rtrequest1(req, &rtinfo, ret_nrt); 659 } 660 661 int 662 rtrequest_global( 663 int req, 664 struct sockaddr *dst, 665 struct sockaddr *gateway, 666 struct sockaddr *netmask, 667 int flags) 668 { 669 struct rt_addrinfo rtinfo; 670 671 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 672 rtinfo.rti_info[RTAX_DST] = dst; 673 rtinfo.rti_info[RTAX_GATEWAY] = gateway; 674 rtinfo.rti_info[RTAX_NETMASK] = netmask; 675 rtinfo.rti_flags = flags; 676 return rtrequest1_global(req, &rtinfo, NULL, NULL, RTREQ_PRIO_NORM); 677 } 678 679 struct netmsg_rtq { 680 struct netmsg_base base; 681 int req; 682 struct rt_addrinfo *rtinfo; 683 rtrequest1_callback_func_t callback; 684 void *arg; 685 }; 686 687 int 688 rtrequest1_global(int req, struct rt_addrinfo *rtinfo, 689 rtrequest1_callback_func_t callback, void *arg, boolean_t req_prio) 690 { 691 int error, flags = 0; 692 struct netmsg_rtq msg; 693 694 if (req_prio) 695 flags = MSGF_PRIORITY; 696 netmsg_init(&msg.base, NULL, &curthread->td_msgport, flags, 697 rtrequest1_msghandler); 698 msg.base.lmsg.ms_error = -1; 699 msg.req = req; 700 msg.rtinfo = rtinfo; 701 msg.callback = callback; 702 msg.arg = arg; 703 error = rt_domsg_global(&msg.base); 704 return (error); 705 } 706 707 /* 708 * Handle a route table request on the current cpu. Since the route table's 709 * are supposed to be identical on each cpu, an error occuring later in the 710 * message chain is considered system-fatal. 711 */ 712 static void 713 rtrequest1_msghandler(netmsg_t msg) 714 { 715 struct netmsg_rtq *rmsg = (void *)msg; 716 struct rt_addrinfo rtinfo; 717 struct rtentry *rt = NULL; 718 int nextcpu; 719 int error; 720 721 /* 722 * Copy the rtinfo. We need to make sure that the original 723 * rtinfo, which is setup by the caller, in the netmsg will 724 * _not_ be changed; else the next CPU on the netmsg forwarding 725 * path will see a different rtinfo than what this CPU has seen. 726 */ 727 rtinfo = *rmsg->rtinfo; 728 729 error = rtrequest1(rmsg->req, &rtinfo, &rt); 730 if (rt) 731 --rt->rt_refcnt; 732 if (rmsg->callback) 733 rmsg->callback(rmsg->req, error, &rtinfo, rt, rmsg->arg); 734 735 /* 736 * RTM_DELETE's are propogated even if an error occurs, since a 737 * cloned route might be undergoing deletion and cloned routes 738 * are not necessarily replicated. An overall error is returned 739 * only if no cpus have the route in question. 740 */ 741 if (rmsg->base.lmsg.ms_error < 0 || error == 0) 742 rmsg->base.lmsg.ms_error = error; 743 744 nextcpu = mycpuid + 1; 745 if (error && rmsg->req != RTM_DELETE) { 746 if (mycpuid != 0) { 747 panic("rtrequest1_msghandler: rtrequest table " 748 "error was cpu%d, err %d\n", mycpuid, error); 749 } 750 lwkt_replymsg(&rmsg->base.lmsg, error); 751 } else if (nextcpu < ncpus) { 752 lwkt_forwardmsg(netisr_cpuport(nextcpu), &rmsg->base.lmsg); 753 } else { 754 lwkt_replymsg(&rmsg->base.lmsg, rmsg->base.lmsg.ms_error); 755 } 756 } 757 758 int 759 rtrequest1(int req, struct rt_addrinfo *rtinfo, struct rtentry **ret_nrt) 760 { 761 struct sockaddr *dst = rtinfo->rti_info[RTAX_DST]; 762 struct rtentry *rt; 763 struct radix_node *rn; 764 struct radix_node_head *rnh; 765 struct ifaddr *ifa; 766 struct sockaddr *ndst; 767 boolean_t reportmsg; 768 int error = 0; 769 770 #define gotoerr(x) { error = x ; goto bad; } 771 772 #ifdef ROUTE_DEBUG 773 if (route_debug) 774 rt_addrinfo_print(req, rtinfo); 775 #endif 776 777 crit_enter(); 778 /* 779 * Find the correct routing tree to use for this Address Family 780 */ 781 if ((rnh = rt_tables[mycpuid][dst->sa_family]) == NULL) 782 gotoerr(EAFNOSUPPORT); 783 784 /* 785 * If we are adding a host route then we don't want to put 786 * a netmask in the tree, nor do we want to clone it. 787 */ 788 if (rtinfo->rti_flags & RTF_HOST) { 789 rtinfo->rti_info[RTAX_NETMASK] = NULL; 790 rtinfo->rti_flags &= ~(RTF_CLONING | RTF_PRCLONING); 791 } 792 793 switch (req) { 794 case RTM_DELETE: 795 /* Remove the item from the tree. */ 796 rn = rnh->rnh_deladdr((char *)rtinfo->rti_info[RTAX_DST], 797 (char *)rtinfo->rti_info[RTAX_NETMASK], 798 rnh); 799 if (rn == NULL) 800 gotoerr(ESRCH); 801 KASSERT(!(rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)), 802 ("rnh_deladdr returned flags 0x%x", rn->rn_flags)); 803 rt = (struct rtentry *)rn; 804 805 /* ref to prevent a deletion race */ 806 ++rt->rt_refcnt; 807 808 /* Free any routes cloned from this one. */ 809 if ((rt->rt_flags & (RTF_CLONING | RTF_PRCLONING)) && 810 rt_mask(rt) != NULL) { 811 rnh->rnh_walktree_from(rnh, (char *)rt_key(rt), 812 (char *)rt_mask(rt), 813 rt_fixdelete, rt); 814 } 815 816 if (rt->rt_gwroute != NULL) { 817 RTFREE(rt->rt_gwroute); 818 rt->rt_gwroute = NULL; 819 } 820 821 /* 822 * NB: RTF_UP must be set during the search above, 823 * because we might delete the last ref, causing 824 * rt to get freed prematurely. 825 */ 826 rt->rt_flags &= ~RTF_UP; 827 828 #ifdef ROUTE_DEBUG 829 if (route_debug) 830 rt_print(rtinfo, rt); 831 #endif 832 833 /* Give the protocol a chance to keep things in sync. */ 834 if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest) 835 ifa->ifa_rtrequest(RTM_DELETE, rt); 836 837 /* 838 * If the caller wants it, then it can have it, 839 * but it's up to it to free the rtentry as we won't be 840 * doing it. 841 */ 842 KASSERT(rt->rt_refcnt >= 0, 843 ("rtrequest1(DELETE): refcnt %ld", rt->rt_refcnt)); 844 if (ret_nrt != NULL) { 845 /* leave ref intact for return */ 846 *ret_nrt = rt; 847 } else { 848 /* deref / attempt to destroy */ 849 rtfree(rt); 850 } 851 break; 852 853 case RTM_RESOLVE: 854 if (ret_nrt == NULL || (rt = *ret_nrt) == NULL) 855 gotoerr(EINVAL); 856 857 KASSERT(rt->rt_cpuid == mycpuid, 858 ("rt resolve rt_cpuid %d, mycpuid %d", 859 rt->rt_cpuid, mycpuid)); 860 861 ifa = rt->rt_ifa; 862 rtinfo->rti_flags = 863 rt->rt_flags & ~(RTF_CLONING | RTF_PRCLONING | RTF_STATIC); 864 rtinfo->rti_flags |= RTF_WASCLONED; 865 rtinfo->rti_info[RTAX_GATEWAY] = rt->rt_gateway; 866 if ((rtinfo->rti_info[RTAX_NETMASK] = rt->rt_genmask) == NULL) 867 rtinfo->rti_flags |= RTF_HOST; 868 rtinfo->rti_info[RTAX_MPLS1] = rt->rt_shim[0]; 869 rtinfo->rti_info[RTAX_MPLS2] = rt->rt_shim[1]; 870 rtinfo->rti_info[RTAX_MPLS3] = rt->rt_shim[2]; 871 goto makeroute; 872 873 case RTM_ADD: 874 KASSERT(!(rtinfo->rti_flags & RTF_GATEWAY) || 875 rtinfo->rti_info[RTAX_GATEWAY] != NULL, 876 ("rtrequest: GATEWAY but no gateway")); 877 878 if (rtinfo->rti_ifa == NULL && (error = rt_getifa(rtinfo))) 879 gotoerr(error); 880 ifa = rtinfo->rti_ifa; 881 makeroute: 882 R_Malloc(rt, struct rtentry *, sizeof(struct rtentry)); 883 if (rt == NULL) { 884 if (req == RTM_ADD) { 885 kprintf("rtrequest1: alloc rtentry failed on " 886 "cpu%d\n", mycpuid); 887 } 888 gotoerr(ENOBUFS); 889 } 890 bzero(rt, sizeof(struct rtentry)); 891 rt->rt_flags = RTF_UP | rtinfo->rti_flags; 892 rt->rt_cpuid = mycpuid; 893 894 if (mycpuid != 0 && req == RTM_ADD) { 895 /* For RTM_ADD, we have already sent rtmsg on CPU0. */ 896 reportmsg = RTL_DONTREPORT; 897 } else { 898 /* 899 * For RTM_ADD, we only send rtmsg on CPU0. 900 * For RTM_RESOLVE, we always send rtmsg. XXX 901 */ 902 reportmsg = RTL_REPORTMSG; 903 } 904 error = rt_setgate(rt, dst, rtinfo->rti_info[RTAX_GATEWAY], 905 reportmsg); 906 if (error != 0) { 907 Free(rt); 908 gotoerr(error); 909 } 910 911 ndst = rt_key(rt); 912 if (rtinfo->rti_info[RTAX_NETMASK] != NULL) 913 rt_maskedcopy(dst, ndst, 914 rtinfo->rti_info[RTAX_NETMASK]); 915 else 916 bcopy(dst, ndst, dst->sa_len); 917 918 if (rtinfo->rti_info[RTAX_MPLS1] != NULL) 919 rt_setshims(rt, rtinfo->rti_info); 920 921 /* 922 * Note that we now have a reference to the ifa. 923 * This moved from below so that rnh->rnh_addaddr() can 924 * examine the ifa and ifa->ifa_ifp if it so desires. 925 */ 926 IFAREF(ifa); 927 rt->rt_ifa = ifa; 928 rt->rt_ifp = ifa->ifa_ifp; 929 /* XXX mtu manipulation will be done in rnh_addaddr -- itojun */ 930 931 rn = rnh->rnh_addaddr((char *)ndst, 932 (char *)rtinfo->rti_info[RTAX_NETMASK], 933 rnh, rt->rt_nodes); 934 if (rn == NULL) { 935 struct rtentry *oldrt; 936 937 /* 938 * We already have one of these in the tree. 939 * We do a special hack: if the old route was 940 * cloned, then we blow it away and try 941 * re-inserting the new one. 942 */ 943 oldrt = rtpurelookup(ndst); 944 if (oldrt != NULL) { 945 --oldrt->rt_refcnt; 946 if (oldrt->rt_flags & RTF_WASCLONED) { 947 rtrequest(RTM_DELETE, rt_key(oldrt), 948 oldrt->rt_gateway, 949 rt_mask(oldrt), 950 oldrt->rt_flags, NULL); 951 rn = rnh->rnh_addaddr((char *)ndst, 952 (char *) 953 rtinfo->rti_info[RTAX_NETMASK], 954 rnh, rt->rt_nodes); 955 } 956 } 957 } 958 959 /* 960 * If it still failed to go into the tree, 961 * then un-make it (this should be a function). 962 */ 963 if (rn == NULL) { 964 if (rt->rt_gwroute != NULL) 965 rtfree(rt->rt_gwroute); 966 IFAFREE(ifa); 967 Free(rt_key(rt)); 968 Free(rt); 969 gotoerr(EEXIST); 970 } 971 972 /* 973 * If we got here from RESOLVE, then we are cloning 974 * so clone the rest, and note that we 975 * are a clone (and increment the parent's references) 976 */ 977 if (req == RTM_RESOLVE) { 978 rt->rt_rmx = (*ret_nrt)->rt_rmx; /* copy metrics */ 979 rt->rt_rmx.rmx_pksent = 0; /* reset packet counter */ 980 if ((*ret_nrt)->rt_flags & 981 (RTF_CLONING | RTF_PRCLONING)) { 982 rt->rt_parent = *ret_nrt; 983 (*ret_nrt)->rt_refcnt++; 984 } 985 } 986 987 /* 988 * if this protocol has something to add to this then 989 * allow it to do that as well. 990 */ 991 if (ifa->ifa_rtrequest != NULL) 992 ifa->ifa_rtrequest(req, rt); 993 994 /* 995 * We repeat the same procedure from rt_setgate() here because 996 * it doesn't fire when we call it there because the node 997 * hasn't been added to the tree yet. 998 */ 999 if (req == RTM_ADD && !(rt->rt_flags & RTF_HOST) && 1000 rt_mask(rt) != NULL) { 1001 struct rtfc_arg arg = { rt, rnh }; 1002 1003 rnh->rnh_walktree_from(rnh, (char *)rt_key(rt), 1004 (char *)rt_mask(rt), 1005 rt_fixchange, &arg); 1006 } 1007 1008 #ifdef ROUTE_DEBUG 1009 if (route_debug) 1010 rt_print(rtinfo, rt); 1011 #endif 1012 /* 1013 * Return the resulting rtentry, 1014 * increasing the number of references by one. 1015 */ 1016 if (ret_nrt != NULL) { 1017 rt->rt_refcnt++; 1018 *ret_nrt = rt; 1019 } 1020 break; 1021 default: 1022 error = EOPNOTSUPP; 1023 } 1024 bad: 1025 #ifdef ROUTE_DEBUG 1026 if (route_debug) { 1027 if (error) 1028 kprintf("rti %p failed error %d\n", rtinfo, error); 1029 else 1030 kprintf("rti %p succeeded\n", rtinfo); 1031 } 1032 #endif 1033 crit_exit(); 1034 return (error); 1035 } 1036 1037 /* 1038 * Called from rtrequest(RTM_DELETE, ...) to fix up the route's ``family'' 1039 * (i.e., the routes related to it by the operation of cloning). This 1040 * routine is iterated over all potential former-child-routes by way of 1041 * rnh->rnh_walktree_from() above, and those that actually are children of 1042 * the late parent (passed in as VP here) are themselves deleted. 1043 */ 1044 static int 1045 rt_fixdelete(struct radix_node *rn, void *vp) 1046 { 1047 struct rtentry *rt = (struct rtentry *)rn; 1048 struct rtentry *rt0 = vp; 1049 1050 if (rt->rt_parent == rt0 && 1051 !(rt->rt_flags & (RTF_PINNED | RTF_CLONING | RTF_PRCLONING))) { 1052 return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 1053 rt->rt_flags, NULL); 1054 } 1055 return 0; 1056 } 1057 1058 /* 1059 * This routine is called from rt_setgate() to do the analogous thing for 1060 * adds and changes. There is the added complication in this case of a 1061 * middle insert; i.e., insertion of a new network route between an older 1062 * network route and (cloned) host routes. For this reason, a simple check 1063 * of rt->rt_parent is insufficient; each candidate route must be tested 1064 * against the (mask, value) of the new route (passed as before in vp) 1065 * to see if the new route matches it. 1066 * 1067 * XXX - it may be possible to do fixdelete() for changes and reserve this 1068 * routine just for adds. I'm not sure why I thought it was necessary to do 1069 * changes this way. 1070 */ 1071 #ifdef DEBUG 1072 static int rtfcdebug = 0; 1073 #endif 1074 1075 static int 1076 rt_fixchange(struct radix_node *rn, void *vp) 1077 { 1078 struct rtentry *rt = (struct rtentry *)rn; 1079 struct rtfc_arg *ap = vp; 1080 struct rtentry *rt0 = ap->rt0; 1081 struct radix_node_head *rnh = ap->rnh; 1082 u_char *xk1, *xm1, *xk2, *xmp; 1083 int i, len, mlen; 1084 1085 #ifdef DEBUG 1086 if (rtfcdebug) 1087 kprintf("rt_fixchange: rt %p, rt0 %p\n", rt, rt0); 1088 #endif 1089 1090 if (rt->rt_parent == NULL || 1091 (rt->rt_flags & (RTF_PINNED | RTF_CLONING | RTF_PRCLONING))) { 1092 #ifdef DEBUG 1093 if (rtfcdebug) kprintf("no parent, pinned or cloning\n"); 1094 #endif 1095 return 0; 1096 } 1097 1098 if (rt->rt_parent == rt0) { 1099 #ifdef DEBUG 1100 if (rtfcdebug) kprintf("parent match\n"); 1101 #endif 1102 return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 1103 rt->rt_flags, NULL); 1104 } 1105 1106 /* 1107 * There probably is a function somewhere which does this... 1108 * if not, there should be. 1109 */ 1110 len = imin(rt_key(rt0)->sa_len, rt_key(rt)->sa_len); 1111 1112 xk1 = (u_char *)rt_key(rt0); 1113 xm1 = (u_char *)rt_mask(rt0); 1114 xk2 = (u_char *)rt_key(rt); 1115 1116 /* avoid applying a less specific route */ 1117 xmp = (u_char *)rt_mask(rt->rt_parent); 1118 mlen = rt_key(rt->rt_parent)->sa_len; 1119 if (mlen > rt_key(rt0)->sa_len) { 1120 #ifdef DEBUG 1121 if (rtfcdebug) 1122 kprintf("rt_fixchange: inserting a less " 1123 "specific route\n"); 1124 #endif 1125 return 0; 1126 } 1127 for (i = rnh->rnh_treetop->rn_offset; i < mlen; i++) { 1128 if ((xmp[i] & ~(xmp[i] ^ xm1[i])) != xmp[i]) { 1129 #ifdef DEBUG 1130 if (rtfcdebug) 1131 kprintf("rt_fixchange: inserting a less " 1132 "specific route\n"); 1133 #endif 1134 return 0; 1135 } 1136 } 1137 1138 for (i = rnh->rnh_treetop->rn_offset; i < len; i++) { 1139 if ((xk2[i] & xm1[i]) != xk1[i]) { 1140 #ifdef DEBUG 1141 if (rtfcdebug) kprintf("no match\n"); 1142 #endif 1143 return 0; 1144 } 1145 } 1146 1147 /* 1148 * OK, this node is a clone, and matches the node currently being 1149 * changed/added under the node's mask. So, get rid of it. 1150 */ 1151 #ifdef DEBUG 1152 if (rtfcdebug) kprintf("deleting\n"); 1153 #endif 1154 return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 1155 rt->rt_flags, NULL); 1156 } 1157 1158 int 1159 rt_setgate(struct rtentry *rt0, struct sockaddr *dst, struct sockaddr *gate, 1160 boolean_t generate_report) 1161 { 1162 char *space, *oldspace; 1163 int dlen = RT_ROUNDUP(dst->sa_len), glen = RT_ROUNDUP(gate->sa_len); 1164 struct rtentry *rt = rt0; 1165 struct radix_node_head *rnh = rt_tables[mycpuid][dst->sa_family]; 1166 1167 /* 1168 * A host route with the destination equal to the gateway 1169 * will interfere with keeping LLINFO in the routing 1170 * table, so disallow it. 1171 */ 1172 if (((rt0->rt_flags & (RTF_HOST | RTF_GATEWAY | RTF_LLINFO)) == 1173 (RTF_HOST | RTF_GATEWAY)) && 1174 dst->sa_len == gate->sa_len && 1175 sa_equal(dst, gate)) { 1176 /* 1177 * The route might already exist if this is an RTM_CHANGE 1178 * or a routing redirect, so try to delete it. 1179 */ 1180 if (rt_key(rt0) != NULL) 1181 rtrequest(RTM_DELETE, rt_key(rt0), rt0->rt_gateway, 1182 rt_mask(rt0), rt0->rt_flags, NULL); 1183 return EADDRNOTAVAIL; 1184 } 1185 1186 /* 1187 * Both dst and gateway are stored in the same malloc'ed chunk 1188 * (If I ever get my hands on....) 1189 * if we need to malloc a new chunk, then keep the old one around 1190 * till we don't need it any more. 1191 */ 1192 if (rt->rt_gateway == NULL || 1193 glen > RT_ROUNDUP(rt->rt_gateway->sa_len)) { 1194 oldspace = (char *)rt_key(rt); 1195 R_Malloc(space, char *, dlen + glen); 1196 if (space == NULL) 1197 return ENOBUFS; 1198 rt->rt_nodes->rn_key = space; 1199 } else { 1200 space = (char *)rt_key(rt); /* Just use the old space. */ 1201 oldspace = NULL; 1202 } 1203 1204 /* Set the gateway value. */ 1205 rt->rt_gateway = (struct sockaddr *)(space + dlen); 1206 bcopy(gate, rt->rt_gateway, glen); 1207 1208 if (oldspace != NULL) { 1209 /* 1210 * If we allocated a new chunk, preserve the original dst. 1211 * This way, rt_setgate() really just sets the gate 1212 * and leaves the dst field alone. 1213 */ 1214 bcopy(dst, space, dlen); 1215 Free(oldspace); 1216 } 1217 1218 /* 1219 * If there is already a gwroute, it's now almost definitely wrong 1220 * so drop it. 1221 */ 1222 if (rt->rt_gwroute != NULL) { 1223 RTFREE(rt->rt_gwroute); 1224 rt->rt_gwroute = NULL; 1225 } 1226 if (rt->rt_flags & RTF_GATEWAY) { 1227 /* 1228 * Cloning loop avoidance: In the presence of 1229 * protocol-cloning and bad configuration, it is 1230 * possible to get stuck in bottomless mutual recursion 1231 * (rtrequest rt_setgate rtlookup). We avoid this 1232 * by not allowing protocol-cloning to operate for 1233 * gateways (which is probably the correct choice 1234 * anyway), and avoid the resulting reference loops 1235 * by disallowing any route to run through itself as 1236 * a gateway. This is obviously mandatory when we 1237 * get rt->rt_output(). 1238 * 1239 * This breaks TTCP for hosts outside the gateway! XXX JH 1240 */ 1241 rt->rt_gwroute = _rtlookup(gate, generate_report, 1242 RTF_PRCLONING); 1243 if (rt->rt_gwroute == rt) { 1244 rt->rt_gwroute = NULL; 1245 --rt->rt_refcnt; 1246 return EDQUOT; /* failure */ 1247 } 1248 } 1249 1250 /* 1251 * This isn't going to do anything useful for host routes, so 1252 * don't bother. Also make sure we have a reasonable mask 1253 * (we don't yet have one during adds). 1254 */ 1255 if (!(rt->rt_flags & RTF_HOST) && rt_mask(rt) != NULL) { 1256 struct rtfc_arg arg = { rt, rnh }; 1257 1258 rnh->rnh_walktree_from(rnh, (char *)rt_key(rt), 1259 (char *)rt_mask(rt), 1260 rt_fixchange, &arg); 1261 } 1262 1263 return 0; 1264 } 1265 1266 static void 1267 rt_maskedcopy( 1268 struct sockaddr *src, 1269 struct sockaddr *dst, 1270 struct sockaddr *netmask) 1271 { 1272 u_char *cp1 = (u_char *)src; 1273 u_char *cp2 = (u_char *)dst; 1274 u_char *cp3 = (u_char *)netmask; 1275 u_char *cplim = cp2 + *cp3; 1276 u_char *cplim2 = cp2 + *cp1; 1277 1278 *cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */ 1279 cp3 += 2; 1280 if (cplim > cplim2) 1281 cplim = cplim2; 1282 while (cp2 < cplim) 1283 *cp2++ = *cp1++ & *cp3++; 1284 if (cp2 < cplim2) 1285 bzero(cp2, cplim2 - cp2); 1286 } 1287 1288 int 1289 rt_llroute(struct sockaddr *dst, struct rtentry *rt0, struct rtentry **drt) 1290 { 1291 struct rtentry *up_rt, *rt; 1292 1293 if (!(rt0->rt_flags & RTF_UP)) { 1294 up_rt = rtlookup(dst); 1295 if (up_rt == NULL) 1296 return (EHOSTUNREACH); 1297 up_rt->rt_refcnt--; 1298 } else 1299 up_rt = rt0; 1300 if (up_rt->rt_flags & RTF_GATEWAY) { 1301 if (up_rt->rt_gwroute == NULL) { 1302 up_rt->rt_gwroute = rtlookup(up_rt->rt_gateway); 1303 if (up_rt->rt_gwroute == NULL) 1304 return (EHOSTUNREACH); 1305 } else if (!(up_rt->rt_gwroute->rt_flags & RTF_UP)) { 1306 rtfree(up_rt->rt_gwroute); 1307 up_rt->rt_gwroute = rtlookup(up_rt->rt_gateway); 1308 if (up_rt->rt_gwroute == NULL) 1309 return (EHOSTUNREACH); 1310 } 1311 rt = up_rt->rt_gwroute; 1312 } else 1313 rt = up_rt; 1314 if (rt->rt_flags & RTF_REJECT && 1315 (rt->rt_rmx.rmx_expire == 0 || /* rt doesn't expire */ 1316 time_uptime < rt->rt_rmx.rmx_expire)) /* rt not expired */ 1317 return (rt->rt_flags & RTF_HOST ? EHOSTDOWN : EHOSTUNREACH); 1318 *drt = rt; 1319 return 0; 1320 } 1321 1322 static int 1323 rt_setshims(struct rtentry *rt, struct sockaddr **rt_shim){ 1324 int i; 1325 1326 for (i=0; i<3; i++) { 1327 struct sockaddr *shim = rt_shim[RTAX_MPLS1 + i]; 1328 int shimlen; 1329 1330 if (shim == NULL) 1331 break; 1332 1333 shimlen = RT_ROUNDUP(shim->sa_len); 1334 R_Malloc(rt->rt_shim[i], struct sockaddr *, shimlen); 1335 bcopy(shim, rt->rt_shim[i], shimlen); 1336 } 1337 1338 return 0; 1339 } 1340 1341 #ifdef ROUTE_DEBUG 1342 1343 /* 1344 * Print out a route table entry 1345 */ 1346 void 1347 rt_print(struct rt_addrinfo *rtinfo, struct rtentry *rn) 1348 { 1349 kprintf("rti %p cpu %d route %p flags %08lx: ", 1350 rtinfo, mycpuid, rn, rn->rt_flags); 1351 sockaddr_print(rt_key(rn)); 1352 kprintf(" mask "); 1353 sockaddr_print(rt_mask(rn)); 1354 kprintf(" gw "); 1355 sockaddr_print(rn->rt_gateway); 1356 kprintf(" ifc \"%s\"", rn->rt_ifp ? rn->rt_ifp->if_dname : "?"); 1357 kprintf(" ifa %p\n", rn->rt_ifa); 1358 } 1359 1360 void 1361 rt_addrinfo_print(int cmd, struct rt_addrinfo *rti) 1362 { 1363 int didit = 0; 1364 int i; 1365 1366 #ifdef ROUTE_DEBUG 1367 if (cmd == RTM_DELETE && route_debug > 1) 1368 print_backtrace(-1); 1369 #endif 1370 1371 switch(cmd) { 1372 case RTM_ADD: 1373 kprintf("ADD "); 1374 break; 1375 case RTM_RESOLVE: 1376 kprintf("RES "); 1377 break; 1378 case RTM_DELETE: 1379 kprintf("DEL "); 1380 break; 1381 default: 1382 kprintf("C%02d ", cmd); 1383 break; 1384 } 1385 kprintf("rti %p cpu %d ", rti, mycpuid); 1386 for (i = 0; i < rti->rti_addrs; ++i) { 1387 if (rti->rti_info[i] == NULL) 1388 continue; 1389 if (didit) 1390 kprintf(" ,"); 1391 switch(i) { 1392 case RTAX_DST: 1393 kprintf("(DST "); 1394 break; 1395 case RTAX_GATEWAY: 1396 kprintf("(GWY "); 1397 break; 1398 case RTAX_NETMASK: 1399 kprintf("(MSK "); 1400 break; 1401 case RTAX_GENMASK: 1402 kprintf("(GEN "); 1403 break; 1404 case RTAX_IFP: 1405 kprintf("(IFP "); 1406 break; 1407 case RTAX_IFA: 1408 kprintf("(IFA "); 1409 break; 1410 case RTAX_AUTHOR: 1411 kprintf("(AUT "); 1412 break; 1413 case RTAX_BRD: 1414 kprintf("(BRD "); 1415 break; 1416 default: 1417 kprintf("(?%02d ", i); 1418 break; 1419 } 1420 sockaddr_print(rti->rti_info[i]); 1421 kprintf(")"); 1422 didit = 1; 1423 } 1424 kprintf("\n"); 1425 } 1426 1427 void 1428 sockaddr_print(struct sockaddr *sa) 1429 { 1430 struct sockaddr_in *sa4; 1431 struct sockaddr_in6 *sa6; 1432 int len; 1433 int i; 1434 1435 if (sa == NULL) { 1436 kprintf("NULL"); 1437 return; 1438 } 1439 1440 len = sa->sa_len - offsetof(struct sockaddr, sa_data[0]); 1441 1442 switch(sa->sa_family) { 1443 case AF_INET: 1444 case AF_INET6: 1445 default: 1446 switch(sa->sa_family) { 1447 case AF_INET: 1448 sa4 = (struct sockaddr_in *)sa; 1449 kprintf("INET %d %d.%d.%d.%d", 1450 ntohs(sa4->sin_port), 1451 (ntohl(sa4->sin_addr.s_addr) >> 24) & 255, 1452 (ntohl(sa4->sin_addr.s_addr) >> 16) & 255, 1453 (ntohl(sa4->sin_addr.s_addr) >> 8) & 255, 1454 (ntohl(sa4->sin_addr.s_addr) >> 0) & 255 1455 ); 1456 break; 1457 case AF_INET6: 1458 sa6 = (struct sockaddr_in6 *)sa; 1459 kprintf("INET6 %d %04x:%04x%04x:%04x:%04x:%04x:%04x:%04x", 1460 ntohs(sa6->sin6_port), 1461 sa6->sin6_addr.s6_addr16[0], 1462 sa6->sin6_addr.s6_addr16[1], 1463 sa6->sin6_addr.s6_addr16[2], 1464 sa6->sin6_addr.s6_addr16[3], 1465 sa6->sin6_addr.s6_addr16[4], 1466 sa6->sin6_addr.s6_addr16[5], 1467 sa6->sin6_addr.s6_addr16[6], 1468 sa6->sin6_addr.s6_addr16[7] 1469 ); 1470 break; 1471 default: 1472 kprintf("AF%d ", sa->sa_family); 1473 while (len > 0 && sa->sa_data[len-1] == 0) 1474 --len; 1475 1476 for (i = 0; i < len; ++i) { 1477 if (i) 1478 kprintf("."); 1479 kprintf("%d", (unsigned char)sa->sa_data[i]); 1480 } 1481 break; 1482 } 1483 } 1484 } 1485 1486 #endif 1487 1488 /* 1489 * Set up a routing table entry, normally for an interface. 1490 */ 1491 int 1492 rtinit(struct ifaddr *ifa, int cmd, int flags) 1493 { 1494 struct sockaddr *dst, *deldst, *netmask; 1495 struct mbuf *m = NULL; 1496 struct radix_node_head *rnh; 1497 struct radix_node *rn; 1498 struct rt_addrinfo rtinfo; 1499 int error; 1500 1501 if (flags & RTF_HOST) { 1502 dst = ifa->ifa_dstaddr; 1503 netmask = NULL; 1504 } else { 1505 dst = ifa->ifa_addr; 1506 netmask = ifa->ifa_netmask; 1507 } 1508 /* 1509 * If it's a delete, check that if it exists, it's on the correct 1510 * interface or we might scrub a route to another ifa which would 1511 * be confusing at best and possibly worse. 1512 */ 1513 if (cmd == RTM_DELETE) { 1514 /* 1515 * It's a delete, so it should already exist.. 1516 * If it's a net, mask off the host bits 1517 * (Assuming we have a mask) 1518 */ 1519 if (netmask != NULL) { 1520 m = m_get(MB_DONTWAIT, MT_SONAME); 1521 if (m == NULL) 1522 return (ENOBUFS); 1523 mbuftrackid(m, 34); 1524 deldst = mtod(m, struct sockaddr *); 1525 rt_maskedcopy(dst, deldst, netmask); 1526 dst = deldst; 1527 } 1528 /* 1529 * Look up an rtentry that is in the routing tree and 1530 * contains the correct info. 1531 */ 1532 if ((rnh = rt_tables[mycpuid][dst->sa_family]) == NULL || 1533 (rn = rnh->rnh_lookup((char *)dst, 1534 (char *)netmask, rnh)) == NULL || 1535 ((struct rtentry *)rn)->rt_ifa != ifa || 1536 !sa_equal((struct sockaddr *)rn->rn_key, dst)) { 1537 if (m != NULL) 1538 m_free(m); 1539 return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH); 1540 } 1541 /* XXX */ 1542 #if 0 1543 else { 1544 /* 1545 * One would think that as we are deleting, and we know 1546 * it doesn't exist, we could just return at this point 1547 * with an "ELSE" clause, but apparently not.. 1548 */ 1549 return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH); 1550 } 1551 #endif 1552 } 1553 /* 1554 * Do the actual request 1555 */ 1556 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 1557 rtinfo.rti_info[RTAX_DST] = dst; 1558 rtinfo.rti_info[RTAX_GATEWAY] = ifa->ifa_addr; 1559 rtinfo.rti_info[RTAX_NETMASK] = netmask; 1560 rtinfo.rti_flags = flags | ifa->ifa_flags; 1561 rtinfo.rti_ifa = ifa; 1562 error = rtrequest1_global(cmd, &rtinfo, rtinit_rtrequest_callback, ifa, 1563 RTREQ_PRIO_HIGH); 1564 if (m != NULL) 1565 m_free(m); 1566 return (error); 1567 } 1568 1569 static void 1570 rtinit_rtrequest_callback(int cmd, int error, 1571 struct rt_addrinfo *rtinfo, struct rtentry *rt, 1572 void *arg) 1573 { 1574 struct ifaddr *ifa = arg; 1575 1576 if (error == 0 && rt) { 1577 if (mycpuid == 0) { 1578 ++rt->rt_refcnt; 1579 rt_newaddrmsg(cmd, ifa, error, rt); 1580 --rt->rt_refcnt; 1581 } 1582 if (cmd == RTM_DELETE) { 1583 if (rt->rt_refcnt == 0) { 1584 ++rt->rt_refcnt; 1585 rtfree(rt); 1586 } 1587 } 1588 } 1589 } 1590 1591 struct netmsg_rts { 1592 struct netmsg_base base; 1593 int req; 1594 struct rt_addrinfo *rtinfo; 1595 rtsearch_callback_func_t callback; 1596 void *arg; 1597 boolean_t exact_match; 1598 int found_cnt; 1599 }; 1600 1601 int 1602 rtsearch_global(int req, struct rt_addrinfo *rtinfo, 1603 rtsearch_callback_func_t callback, void *arg, boolean_t exact_match, 1604 boolean_t req_prio) 1605 { 1606 struct netmsg_rts msg; 1607 int flags = 0; 1608 1609 if (req_prio) 1610 flags = MSGF_PRIORITY; 1611 netmsg_init(&msg.base, NULL, &curthread->td_msgport, flags, 1612 rtsearch_msghandler); 1613 msg.req = req; 1614 msg.rtinfo = rtinfo; 1615 msg.callback = callback; 1616 msg.arg = arg; 1617 msg.exact_match = exact_match; 1618 msg.found_cnt = 0; 1619 return rt_domsg_global(&msg.base); 1620 } 1621 1622 static void 1623 rtsearch_msghandler(netmsg_t msg) 1624 { 1625 struct netmsg_rts *rmsg = (void *)msg; 1626 struct rt_addrinfo rtinfo; 1627 struct radix_node_head *rnh; 1628 struct rtentry *rt; 1629 int nextcpu, error; 1630 1631 /* 1632 * Copy the rtinfo. We need to make sure that the original 1633 * rtinfo, which is setup by the caller, in the netmsg will 1634 * _not_ be changed; else the next CPU on the netmsg forwarding 1635 * path will see a different rtinfo than what this CPU has seen. 1636 */ 1637 rtinfo = *rmsg->rtinfo; 1638 1639 /* 1640 * Find the correct routing tree to use for this Address Family 1641 */ 1642 if ((rnh = rt_tables[mycpuid][rtinfo.rti_dst->sa_family]) == NULL) { 1643 if (mycpuid != 0) 1644 panic("partially initialized routing tables"); 1645 lwkt_replymsg(&rmsg->base.lmsg, EAFNOSUPPORT); 1646 return; 1647 } 1648 1649 /* 1650 * Correct rtinfo for the host route searching. 1651 */ 1652 if (rtinfo.rti_flags & RTF_HOST) { 1653 rtinfo.rti_netmask = NULL; 1654 rtinfo.rti_flags &= ~(RTF_CLONING | RTF_PRCLONING); 1655 } 1656 1657 rt = (struct rtentry *) 1658 rnh->rnh_lookup((char *)rtinfo.rti_dst, 1659 (char *)rtinfo.rti_netmask, rnh); 1660 1661 /* 1662 * If we are asked to do the "exact match", we need to make sure 1663 * that host route searching got a host route while a network 1664 * route searching got a network route. 1665 */ 1666 if (rt != NULL && rmsg->exact_match && 1667 ((rt->rt_flags ^ rtinfo.rti_flags) & RTF_HOST)) 1668 rt = NULL; 1669 1670 if (rt == NULL) { 1671 /* 1672 * No matching routes have been found, don't count this 1673 * as a critical error (here, we set 'error' to 0), just 1674 * keep moving on, since at least prcloned routes are not 1675 * duplicated onto each CPU. 1676 */ 1677 error = 0; 1678 } else { 1679 rmsg->found_cnt++; 1680 1681 rt->rt_refcnt++; 1682 error = rmsg->callback(rmsg->req, &rtinfo, rt, rmsg->arg, 1683 rmsg->found_cnt); 1684 rt->rt_refcnt--; 1685 1686 if (error == EJUSTRETURN) { 1687 lwkt_replymsg(&rmsg->base.lmsg, 0); 1688 return; 1689 } 1690 } 1691 1692 nextcpu = mycpuid + 1; 1693 if (error) { 1694 KKASSERT(rmsg->found_cnt > 0); 1695 1696 /* 1697 * Under following cases, unrecoverable error has 1698 * not occured: 1699 * o Request is RTM_GET 1700 * o The first time that we find the route, but the 1701 * modification fails. 1702 */ 1703 if (rmsg->req != RTM_GET && rmsg->found_cnt > 1) { 1704 panic("rtsearch_msghandler: unrecoverable error " 1705 "cpu %d", mycpuid); 1706 } 1707 lwkt_replymsg(&rmsg->base.lmsg, error); 1708 } else if (nextcpu < ncpus) { 1709 lwkt_forwardmsg(netisr_cpuport(nextcpu), &rmsg->base.lmsg); 1710 } else { 1711 if (rmsg->found_cnt == 0) { 1712 /* The requested route was never seen ... */ 1713 error = ESRCH; 1714 } 1715 lwkt_replymsg(&rmsg->base.lmsg, error); 1716 } 1717 } 1718 1719 int 1720 rtmask_add_global(struct sockaddr *mask, boolean_t req_prio) 1721 { 1722 struct netmsg_base msg; 1723 int flags = 0; 1724 1725 if (req_prio) 1726 flags = MSGF_PRIORITY; 1727 netmsg_init(&msg, NULL, &curthread->td_msgport, flags, 1728 rtmask_add_msghandler); 1729 msg.lmsg.u.ms_resultp = mask; 1730 1731 return rt_domsg_global(&msg); 1732 } 1733 1734 struct sockaddr * 1735 _rtmask_lookup(struct sockaddr *mask, boolean_t search) 1736 { 1737 struct radix_node *n; 1738 1739 #define clen(s) (*(u_char *)(s)) 1740 n = rn_addmask((char *)mask, search, 1, rn_cpumaskhead(mycpuid)); 1741 if (n != NULL && 1742 mask->sa_len >= clen(n->rn_key) && 1743 bcmp((char *)mask + 1, 1744 (char *)n->rn_key + 1, clen(n->rn_key) - 1) == 0) { 1745 return (struct sockaddr *)n->rn_key; 1746 } else { 1747 return NULL; 1748 } 1749 #undef clen 1750 } 1751 1752 static void 1753 rtmask_add_msghandler(netmsg_t msg) 1754 { 1755 struct lwkt_msg *lmsg = &msg->lmsg; 1756 struct sockaddr *mask = lmsg->u.ms_resultp; 1757 int error = 0, nextcpu; 1758 1759 if (rtmask_lookup(mask) == NULL) 1760 error = ENOBUFS; 1761 1762 nextcpu = mycpuid + 1; 1763 if (!error && nextcpu < ncpus) 1764 lwkt_forwardmsg(netisr_cpuport(nextcpu), lmsg); 1765 else 1766 lwkt_replymsg(lmsg, error); 1767 } 1768 1769 /* This must be before ip6_init2(), which is now SI_ORDER_MIDDLE */ 1770 SYSINIT(route, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0); 1771 1772 struct rtchange_arg { 1773 struct ifaddr *old_ifa; 1774 struct ifaddr *new_ifa; 1775 struct rtentry *rt; 1776 int changed; 1777 }; 1778 1779 static void 1780 rtchange_ifa(struct rtentry *rt, struct rtchange_arg *ap) 1781 { 1782 if (rt->rt_ifa->ifa_rtrequest != NULL) 1783 rt->rt_ifa->ifa_rtrequest(RTM_DELETE, rt); 1784 IFAFREE(rt->rt_ifa); 1785 1786 IFAREF(ap->new_ifa); 1787 rt->rt_ifa = ap->new_ifa; 1788 rt->rt_ifp = ap->new_ifa->ifa_ifp; 1789 if (rt->rt_ifa->ifa_rtrequest != NULL) 1790 rt->rt_ifa->ifa_rtrequest(RTM_ADD, rt); 1791 1792 ap->changed = 1; 1793 } 1794 1795 static int 1796 rtchange_callback(struct radix_node *rn, void *xap) 1797 { 1798 struct rtchange_arg *ap = xap; 1799 struct rtentry *rt = (struct rtentry *)rn; 1800 1801 if (rt->rt_ifa == ap->old_ifa) { 1802 if (rt->rt_flags & (RTF_CLONING | RTF_PRCLONING)) { 1803 /* 1804 * We could saw the branch off when we are 1805 * still sitting on it, if the ifa_rtrequest 1806 * DEL/ADD are called directly from here. 1807 */ 1808 ap->rt = rt; 1809 return EJUSTRETURN; 1810 } 1811 rtchange_ifa(rt, ap); 1812 } 1813 return 0; 1814 } 1815 1816 struct netmsg_rtchange { 1817 struct netmsg_base base; 1818 struct ifaddr *old_ifa; 1819 struct ifaddr *new_ifa; 1820 int changed; 1821 }; 1822 1823 static void 1824 rtchange_dispatch(netmsg_t msg) 1825 { 1826 struct netmsg_rtchange *rmsg = (void *)msg; 1827 struct radix_node_head *rnh; 1828 struct rtchange_arg arg; 1829 int nextcpu, cpu; 1830 1831 cpu = mycpuid; 1832 1833 memset(&arg, 0, sizeof(arg)); 1834 arg.old_ifa = rmsg->old_ifa; 1835 arg.new_ifa = rmsg->new_ifa; 1836 1837 rnh = rt_tables[cpu][AF_INET]; 1838 for (;;) { 1839 int error; 1840 1841 KKASSERT(arg.rt == NULL); 1842 error = rnh->rnh_walktree(rnh, rtchange_callback, &arg); 1843 if (arg.rt != NULL) { 1844 struct rtentry *rt; 1845 1846 rt = arg.rt; 1847 arg.rt = NULL; 1848 rtchange_ifa(rt, &arg); 1849 } else { 1850 break; 1851 } 1852 } 1853 if (arg.changed) 1854 rmsg->changed = 1; 1855 1856 nextcpu = cpu + 1; 1857 if (nextcpu < ncpus) 1858 lwkt_forwardmsg(netisr_cpuport(nextcpu), &rmsg->base.lmsg); 1859 else 1860 lwkt_replymsg(&rmsg->base.lmsg, 0); 1861 } 1862 1863 int 1864 rtchange(struct ifaddr *old_ifa, struct ifaddr *new_ifa) 1865 { 1866 struct netmsg_rtchange msg; 1867 1868 /* 1869 * XXX individual requests are not independantly chained, 1870 * which means that the per-cpu route tables will not be 1871 * consistent in the middle of the operation. If routes 1872 * related to the interface are manipulated while we are 1873 * doing this the inconsistancy could trigger a panic. 1874 */ 1875 netmsg_init(&msg.base, NULL, &curthread->td_msgport, MSGF_PRIORITY, 1876 rtchange_dispatch); 1877 msg.old_ifa = old_ifa; 1878 msg.new_ifa = new_ifa; 1879 msg.changed = 0; 1880 rt_domsg_global(&msg.base); 1881 1882 if (msg.changed) { 1883 old_ifa->ifa_flags &= ~IFA_ROUTE; 1884 new_ifa->ifa_flags |= IFA_ROUTE; 1885 return 0; 1886 } else { 1887 return ENOENT; 1888 } 1889 } 1890 1891 int 1892 rt_domsg_global(struct netmsg_base *nmsg) 1893 { 1894 ASSERT_CANDOMSG_NETISR0(curthread); 1895 return lwkt_domsg(netisr_cpuport(0), &nmsg->lmsg, 0); 1896 } 1897