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 nextcpu = mycpuid + 1; 152 153 SLIST_FOREACH(dom, &domains, dom_next) { 154 if (dom->dom_rtattach) { 155 dom->dom_rtattach( 156 (void **)&rt_tables[mycpuid][dom->dom_family], 157 dom->dom_rtoffset); 158 } 159 } 160 if (nextcpu < ncpus) 161 lwkt_forwardmsg(netisr_cpuport(nextcpu), &msg->lmsg); 162 else 163 lwkt_replymsg(&msg->lmsg, 0); 164 } 165 166 static void 167 rtable_init(void) 168 { 169 struct netmsg_base msg; 170 171 netmsg_init(&msg, NULL, &curthread->td_msgport, 0, rtable_init_oncpu); 172 rt_domsg_global(&msg); 173 } 174 175 /* 176 * Routing statistics. 177 */ 178 static int 179 sysctl_rtstatistics(SYSCTL_HANDLER_ARGS) 180 { 181 int cpu, error = 0; 182 183 for (cpu = 0; cpu < ncpus; ++cpu) { 184 if ((error = SYSCTL_OUT(req, &rtstatistics_percpu[cpu], 185 sizeof(struct rtstatistics)))) 186 break; 187 if ((error = SYSCTL_IN(req, &rtstatistics_percpu[cpu], 188 sizeof(struct rtstatistics)))) 189 break; 190 } 191 192 return (error); 193 } 194 SYSCTL_PROC(_net_route, OID_AUTO, stats, (CTLTYPE_OPAQUE|CTLFLAG_RW), 195 0, 0, sysctl_rtstatistics, "S,rtstatistics", "Routing statistics"); 196 197 /* 198 * Packet routing routines. 199 */ 200 201 /* 202 * Look up and fill in the "ro_rt" rtentry field in a route structure given 203 * an address in the "ro_dst" field. Always send a report on a miss and 204 * always clone routes. 205 */ 206 void 207 rtalloc(struct route *ro) 208 { 209 rtalloc_ign(ro, 0UL); 210 } 211 212 /* 213 * Look up and fill in the "ro_rt" rtentry field in a route structure given 214 * an address in the "ro_dst" field. Always send a report on a miss and 215 * optionally clone routes when RTF_CLONING or RTF_PRCLONING are not being 216 * ignored. 217 */ 218 void 219 rtalloc_ign(struct route *ro, u_long ignoreflags) 220 { 221 if (ro->ro_rt != NULL) { 222 if (ro->ro_rt->rt_ifp != NULL && ro->ro_rt->rt_flags & RTF_UP) 223 return; 224 rtfree(ro->ro_rt); 225 ro->ro_rt = NULL; 226 } 227 ro->ro_rt = _rtlookup(&ro->ro_dst, RTL_REPORTMSG, ignoreflags); 228 } 229 230 /* 231 * Look up the route that matches the given "dst" address. 232 * 233 * Route lookup can have the side-effect of creating and returning 234 * a cloned route instead when "dst" matches a cloning route and the 235 * RTF_CLONING and RTF_PRCLONING flags are not being ignored. 236 * 237 * Any route returned has its reference count incremented. 238 */ 239 struct rtentry * 240 _rtlookup(struct sockaddr *dst, boolean_t generate_report, u_long ignore) 241 { 242 struct radix_node_head *rnh = rt_tables[mycpuid][dst->sa_family]; 243 struct rtentry *rt; 244 245 if (rnh == NULL) 246 goto unreach; 247 248 /* 249 * Look up route in the radix tree. 250 */ 251 rt = (struct rtentry *) rnh->rnh_matchaddr((char *)dst, rnh); 252 if (rt == NULL) 253 goto unreach; 254 255 /* 256 * Handle cloning routes. 257 */ 258 if ((rt->rt_flags & ~ignore & (RTF_CLONING | RTF_PRCLONING)) != 0) { 259 struct rtentry *clonedroute; 260 int error; 261 262 clonedroute = rt; /* copy in/copy out parameter */ 263 error = rtrequest(RTM_RESOLVE, dst, NULL, NULL, 0, 264 &clonedroute); /* clone the route */ 265 if (error != 0) { /* cloning failed */ 266 if (generate_report) 267 rt_dstmsg(RTM_MISS, dst, error); 268 rt->rt_refcnt++; 269 return (rt); /* return the uncloned route */ 270 } 271 if (generate_report) { 272 if (clonedroute->rt_flags & RTF_XRESOLVE) 273 rt_dstmsg(RTM_RESOLVE, dst, 0); 274 else 275 rt_rtmsg(RTM_ADD, clonedroute, 276 clonedroute->rt_ifp, 0); 277 } 278 return (clonedroute); /* return cloned route */ 279 } 280 281 /* 282 * Increment the reference count of the matched route and return. 283 */ 284 rt->rt_refcnt++; 285 return (rt); 286 287 unreach: 288 rtstat.rts_unreach++; 289 if (generate_report) 290 rt_dstmsg(RTM_MISS, dst, 0); 291 return (NULL); 292 } 293 294 void 295 rtfree(struct rtentry *rt) 296 { 297 if (rt->rt_cpuid == mycpuid) 298 rtfree_oncpu(rt); 299 else 300 rtfree_remote(rt); 301 } 302 303 void 304 rtfree_oncpu(struct rtentry *rt) 305 { 306 KKASSERT(rt->rt_cpuid == mycpuid); 307 KASSERT(rt->rt_refcnt > 0, ("rtfree: rt_refcnt %ld", rt->rt_refcnt)); 308 309 --rt->rt_refcnt; 310 if (rt->rt_refcnt == 0) { 311 struct radix_node_head *rnh = 312 rt_tables[mycpuid][rt_key(rt)->sa_family]; 313 314 if (rnh->rnh_close) 315 rnh->rnh_close((struct radix_node *)rt, rnh); 316 if (!(rt->rt_flags & RTF_UP)) { 317 /* deallocate route */ 318 if (rt->rt_ifa != NULL) 319 IFAFREE(rt->rt_ifa); 320 if (rt->rt_parent != NULL) 321 RTFREE(rt->rt_parent); /* recursive call! */ 322 Free(rt_key(rt)); 323 Free(rt); 324 } 325 } 326 } 327 328 static void 329 rtfree_remote_dispatch(netmsg_t msg) 330 { 331 struct lwkt_msg *lmsg = &msg->lmsg; 332 struct rtentry *rt = lmsg->u.ms_resultp; 333 334 rtfree_oncpu(rt); 335 lwkt_replymsg(lmsg, 0); 336 } 337 338 void 339 rtfree_remote(struct rtentry *rt) 340 { 341 struct netmsg_base *msg; 342 struct lwkt_msg *lmsg; 343 344 KKASSERT(rt->rt_cpuid != mycpuid); 345 346 if (route_assert_owner_access) { 347 panic("rt remote free rt_cpuid %d, mycpuid %d", 348 rt->rt_cpuid, mycpuid); 349 } else { 350 kprintf("rt remote free rt_cpuid %d, mycpuid %d\n", 351 rt->rt_cpuid, mycpuid); 352 print_backtrace(-1); 353 } 354 355 msg = kmalloc(sizeof(*msg), M_LWKTMSG, M_INTWAIT); 356 netmsg_init(msg, NULL, &netisr_afree_rport, 0, rtfree_remote_dispatch); 357 lmsg = &msg->lmsg; 358 lmsg->u.ms_resultp = rt; 359 360 lwkt_sendmsg(netisr_cpuport(rt->rt_cpuid), lmsg); 361 } 362 363 int 364 rtredirect_oncpu(struct sockaddr *dst, struct sockaddr *gateway, 365 struct sockaddr *netmask, int flags, struct sockaddr *src) 366 { 367 struct rtentry *rt = NULL; 368 struct rt_addrinfo rtinfo; 369 struct ifaddr *ifa; 370 u_long *stat = NULL; 371 int error; 372 373 /* verify the gateway is directly reachable */ 374 if ((ifa = ifa_ifwithnet(gateway)) == NULL) { 375 error = ENETUNREACH; 376 goto out; 377 } 378 379 /* 380 * If the redirect isn't from our current router for this destination, 381 * it's either old or wrong. 382 */ 383 if (!(flags & RTF_DONE) && /* XXX JH */ 384 (rt = rtpurelookup(dst)) != NULL && 385 (!sa_equal(src, rt->rt_gateway) || rt->rt_ifa != ifa)) { 386 error = EINVAL; 387 goto done; 388 } 389 390 /* 391 * If it redirects us to ourselves, we have a routing loop, 392 * perhaps as a result of an interface going down recently. 393 */ 394 if (ifa_ifwithaddr(gateway)) { 395 error = EHOSTUNREACH; 396 goto done; 397 } 398 399 /* 400 * Create a new entry if the lookup failed or if we got back 401 * a wildcard entry for the default route. This is necessary 402 * for hosts which use routing redirects generated by smart 403 * gateways to dynamically build the routing tables. 404 */ 405 if (rt == NULL) 406 goto create; 407 if ((rt_mask(rt) != NULL && rt_mask(rt)->sa_len < 2)) { 408 rtfree(rt); 409 goto create; 410 } 411 412 /* Ignore redirects for directly connected hosts. */ 413 if (!(rt->rt_flags & RTF_GATEWAY)) { 414 error = EHOSTUNREACH; 415 goto done; 416 } 417 418 if (!(rt->rt_flags & RTF_HOST) && (flags & RTF_HOST)) { 419 /* 420 * Changing from a network route to a host route. 421 * Create a new host route rather than smashing the 422 * network route. 423 */ 424 create: 425 flags |= RTF_GATEWAY | RTF_DYNAMIC; 426 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 427 rtinfo.rti_info[RTAX_DST] = dst; 428 rtinfo.rti_info[RTAX_GATEWAY] = gateway; 429 rtinfo.rti_info[RTAX_NETMASK] = netmask; 430 rtinfo.rti_flags = flags; 431 rtinfo.rti_ifa = ifa; 432 rt = NULL; /* copy-in/copy-out parameter */ 433 error = rtrequest1(RTM_ADD, &rtinfo, &rt); 434 if (rt != NULL) 435 flags = rt->rt_flags; 436 stat = &rtstat.rts_dynamic; 437 } else { 438 /* 439 * Smash the current notion of the gateway to this destination. 440 * Should check about netmask!!! 441 */ 442 rt->rt_flags |= RTF_MODIFIED; 443 flags |= RTF_MODIFIED; 444 445 /* We only need to report rtmsg on CPU0 */ 446 rt_setgate(rt, rt_key(rt), gateway, 447 mycpuid == 0 ? RTL_REPORTMSG : RTL_DONTREPORT); 448 error = 0; 449 stat = &rtstat.rts_newgateway; 450 } 451 452 done: 453 if (rt != NULL) 454 rtfree(rt); 455 out: 456 if (error != 0) 457 rtstat.rts_badredirect++; 458 else if (stat != NULL) 459 (*stat)++; 460 461 return error; 462 } 463 464 struct netmsg_rtredirect { 465 struct netmsg_base base; 466 struct sockaddr *dst; 467 struct sockaddr *gateway; 468 struct sockaddr *netmask; 469 int flags; 470 struct sockaddr *src; 471 }; 472 473 /* 474 * Force a routing table entry to the specified 475 * destination to go through the given gateway. 476 * Normally called as a result of a routing redirect 477 * message from the network layer. 478 * 479 * N.B.: must be called at splnet 480 */ 481 void 482 rtredirect(struct sockaddr *dst, struct sockaddr *gateway, 483 struct sockaddr *netmask, int flags, struct sockaddr *src) 484 { 485 struct rt_addrinfo rtinfo; 486 int error; 487 struct netmsg_rtredirect msg; 488 489 netmsg_init(&msg.base, NULL, &curthread->td_msgport, 490 0, rtredirect_msghandler); 491 msg.dst = dst; 492 msg.gateway = gateway; 493 msg.netmask = netmask; 494 msg.flags = flags; 495 msg.src = src; 496 error = rt_domsg_global(&msg.base); 497 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 498 rtinfo.rti_info[RTAX_DST] = dst; 499 rtinfo.rti_info[RTAX_GATEWAY] = gateway; 500 rtinfo.rti_info[RTAX_NETMASK] = netmask; 501 rtinfo.rti_info[RTAX_AUTHOR] = src; 502 rt_missmsg(RTM_REDIRECT, &rtinfo, flags, error); 503 } 504 505 static void 506 rtredirect_msghandler(netmsg_t msg) 507 { 508 struct netmsg_rtredirect *rmsg = (void *)msg; 509 int nextcpu; 510 511 rtredirect_oncpu(rmsg->dst, rmsg->gateway, rmsg->netmask, 512 rmsg->flags, rmsg->src); 513 nextcpu = mycpuid + 1; 514 if (nextcpu < ncpus) 515 lwkt_forwardmsg(netisr_cpuport(nextcpu), &msg->lmsg); 516 else 517 lwkt_replymsg(&msg->lmsg, 0); 518 } 519 520 /* 521 * Routing table ioctl interface. 522 */ 523 int 524 rtioctl(u_long req, caddr_t data, struct ucred *cred) 525 { 526 #ifdef INET 527 /* Multicast goop, grrr... */ 528 return mrt_ioctl ? mrt_ioctl(req, data) : EOPNOTSUPP; 529 #else 530 return ENXIO; 531 #endif 532 } 533 534 struct ifaddr * 535 ifa_ifwithroute(int flags, struct sockaddr *dst, struct sockaddr *gateway) 536 { 537 struct ifaddr *ifa; 538 539 if (!(flags & RTF_GATEWAY)) { 540 /* 541 * If we are adding a route to an interface, 542 * and the interface is a point-to-point link, 543 * we should search for the destination 544 * as our clue to the interface. Otherwise 545 * we can use the local address. 546 */ 547 ifa = NULL; 548 if (flags & RTF_HOST) { 549 ifa = ifa_ifwithdstaddr(dst); 550 } 551 if (ifa == NULL) 552 ifa = ifa_ifwithaddr(gateway); 553 } else { 554 /* 555 * If we are adding a route to a remote net 556 * or host, the gateway may still be on the 557 * other end of a pt to pt link. 558 */ 559 ifa = ifa_ifwithdstaddr(gateway); 560 } 561 if (ifa == NULL) 562 ifa = ifa_ifwithnet(gateway); 563 if (ifa == NULL) { 564 struct rtentry *rt; 565 566 rt = rtpurelookup(gateway); 567 if (rt == NULL) 568 return (NULL); 569 rt->rt_refcnt--; 570 if ((ifa = rt->rt_ifa) == NULL) 571 return (NULL); 572 } 573 if (ifa->ifa_addr->sa_family != dst->sa_family) { 574 struct ifaddr *oldifa = ifa; 575 576 ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp); 577 if (ifa == NULL) 578 ifa = oldifa; 579 } 580 return (ifa); 581 } 582 583 static int rt_fixdelete (struct radix_node *, void *); 584 static int rt_fixchange (struct radix_node *, void *); 585 586 struct rtfc_arg { 587 struct rtentry *rt0; 588 struct radix_node_head *rnh; 589 }; 590 591 /* 592 * Set rtinfo->rti_ifa and rtinfo->rti_ifp. 593 */ 594 int 595 rt_getifa(struct rt_addrinfo *rtinfo) 596 { 597 struct sockaddr *gateway = rtinfo->rti_info[RTAX_GATEWAY]; 598 struct sockaddr *dst = rtinfo->rti_info[RTAX_DST]; 599 struct sockaddr *ifaaddr = rtinfo->rti_info[RTAX_IFA]; 600 int flags = rtinfo->rti_flags; 601 602 /* 603 * ifp may be specified by sockaddr_dl 604 * when protocol address is ambiguous. 605 */ 606 if (rtinfo->rti_ifp == NULL) { 607 struct sockaddr *ifpaddr; 608 609 ifpaddr = rtinfo->rti_info[RTAX_IFP]; 610 if (ifpaddr != NULL && ifpaddr->sa_family == AF_LINK) { 611 struct ifaddr *ifa; 612 613 ifa = ifa_ifwithnet(ifpaddr); 614 if (ifa != NULL) 615 rtinfo->rti_ifp = ifa->ifa_ifp; 616 } 617 } 618 619 if (rtinfo->rti_ifa == NULL && ifaaddr != NULL) 620 rtinfo->rti_ifa = ifa_ifwithaddr(ifaaddr); 621 if (rtinfo->rti_ifa == NULL) { 622 struct sockaddr *sa; 623 624 sa = ifaaddr != NULL ? ifaaddr : 625 (gateway != NULL ? gateway : dst); 626 if (sa != NULL && rtinfo->rti_ifp != NULL) 627 rtinfo->rti_ifa = ifaof_ifpforaddr(sa, rtinfo->rti_ifp); 628 else if (dst != NULL && gateway != NULL) 629 rtinfo->rti_ifa = ifa_ifwithroute(flags, dst, gateway); 630 else if (sa != NULL) 631 rtinfo->rti_ifa = ifa_ifwithroute(flags, sa, sa); 632 } 633 if (rtinfo->rti_ifa == NULL) 634 return (ENETUNREACH); 635 636 if (rtinfo->rti_ifp == NULL) 637 rtinfo->rti_ifp = rtinfo->rti_ifa->ifa_ifp; 638 return (0); 639 } 640 641 /* 642 * Do appropriate manipulations of a routing tree given 643 * all the bits of info needed 644 */ 645 int 646 rtrequest( 647 int req, 648 struct sockaddr *dst, 649 struct sockaddr *gateway, 650 struct sockaddr *netmask, 651 int flags, 652 struct rtentry **ret_nrt) 653 { 654 struct rt_addrinfo rtinfo; 655 656 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 657 rtinfo.rti_info[RTAX_DST] = dst; 658 rtinfo.rti_info[RTAX_GATEWAY] = gateway; 659 rtinfo.rti_info[RTAX_NETMASK] = netmask; 660 rtinfo.rti_flags = flags; 661 return rtrequest1(req, &rtinfo, ret_nrt); 662 } 663 664 int 665 rtrequest_global( 666 int req, 667 struct sockaddr *dst, 668 struct sockaddr *gateway, 669 struct sockaddr *netmask, 670 int flags) 671 { 672 struct rt_addrinfo rtinfo; 673 674 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 675 rtinfo.rti_info[RTAX_DST] = dst; 676 rtinfo.rti_info[RTAX_GATEWAY] = gateway; 677 rtinfo.rti_info[RTAX_NETMASK] = netmask; 678 rtinfo.rti_flags = flags; 679 return rtrequest1_global(req, &rtinfo, NULL, NULL, RTREQ_PRIO_NORM); 680 } 681 682 struct netmsg_rtq { 683 struct netmsg_base base; 684 int req; 685 struct rt_addrinfo *rtinfo; 686 rtrequest1_callback_func_t callback; 687 void *arg; 688 }; 689 690 int 691 rtrequest1_global(int req, struct rt_addrinfo *rtinfo, 692 rtrequest1_callback_func_t callback, void *arg, boolean_t req_prio) 693 { 694 int error, flags = 0; 695 struct netmsg_rtq msg; 696 697 if (req_prio) 698 flags = MSGF_PRIORITY; 699 netmsg_init(&msg.base, NULL, &curthread->td_msgport, flags, 700 rtrequest1_msghandler); 701 msg.base.lmsg.ms_error = -1; 702 msg.req = req; 703 msg.rtinfo = rtinfo; 704 msg.callback = callback; 705 msg.arg = arg; 706 error = rt_domsg_global(&msg.base); 707 return (error); 708 } 709 710 /* 711 * Handle a route table request on the current cpu. Since the route table's 712 * are supposed to be identical on each cpu, an error occuring later in the 713 * message chain is considered system-fatal. 714 */ 715 static void 716 rtrequest1_msghandler(netmsg_t msg) 717 { 718 struct netmsg_rtq *rmsg = (void *)msg; 719 struct rt_addrinfo rtinfo; 720 struct rtentry *rt = NULL; 721 int nextcpu; 722 int error; 723 724 /* 725 * Copy the rtinfo. We need to make sure that the original 726 * rtinfo, which is setup by the caller, in the netmsg will 727 * _not_ be changed; else the next CPU on the netmsg forwarding 728 * path will see a different rtinfo than what this CPU has seen. 729 */ 730 rtinfo = *rmsg->rtinfo; 731 732 error = rtrequest1(rmsg->req, &rtinfo, &rt); 733 if (rt) 734 --rt->rt_refcnt; 735 if (rmsg->callback) 736 rmsg->callback(rmsg->req, error, &rtinfo, rt, rmsg->arg); 737 738 /* 739 * RTM_DELETE's are propogated even if an error occurs, since a 740 * cloned route might be undergoing deletion and cloned routes 741 * are not necessarily replicated. An overall error is returned 742 * only if no cpus have the route in question. 743 */ 744 if (rmsg->base.lmsg.ms_error < 0 || error == 0) 745 rmsg->base.lmsg.ms_error = error; 746 747 nextcpu = mycpuid + 1; 748 if (error && rmsg->req != RTM_DELETE) { 749 if (mycpuid != 0) { 750 panic("rtrequest1_msghandler: rtrequest table req %d, " 751 "failed on cpu%d, error %d\n", 752 rmsg->req, mycpuid, error); 753 } 754 lwkt_replymsg(&rmsg->base.lmsg, error); 755 } else if (nextcpu < ncpus) { 756 lwkt_forwardmsg(netisr_cpuport(nextcpu), &rmsg->base.lmsg); 757 } else { 758 lwkt_replymsg(&rmsg->base.lmsg, rmsg->base.lmsg.ms_error); 759 } 760 } 761 762 int 763 rtrequest1(int req, struct rt_addrinfo *rtinfo, struct rtentry **ret_nrt) 764 { 765 struct sockaddr *dst = rtinfo->rti_info[RTAX_DST]; 766 struct rtentry *rt; 767 struct radix_node *rn; 768 struct radix_node_head *rnh; 769 struct ifaddr *ifa; 770 struct sockaddr *ndst; 771 boolean_t reportmsg; 772 int error = 0; 773 774 #define gotoerr(x) { error = x ; goto bad; } 775 776 #ifdef ROUTE_DEBUG 777 if (route_debug) 778 rt_addrinfo_print(req, rtinfo); 779 #endif 780 781 crit_enter(); 782 /* 783 * Find the correct routing tree to use for this Address Family 784 */ 785 if ((rnh = rt_tables[mycpuid][dst->sa_family]) == NULL) 786 gotoerr(EAFNOSUPPORT); 787 788 /* 789 * If we are adding a host route then we don't want to put 790 * a netmask in the tree, nor do we want to clone it. 791 */ 792 if (rtinfo->rti_flags & RTF_HOST) { 793 rtinfo->rti_info[RTAX_NETMASK] = NULL; 794 rtinfo->rti_flags &= ~(RTF_CLONING | RTF_PRCLONING); 795 } 796 797 switch (req) { 798 case RTM_DELETE: 799 /* Remove the item from the tree. */ 800 rn = rnh->rnh_deladdr((char *)rtinfo->rti_info[RTAX_DST], 801 (char *)rtinfo->rti_info[RTAX_NETMASK], 802 rnh); 803 if (rn == NULL) 804 gotoerr(ESRCH); 805 KASSERT(!(rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)), 806 ("rnh_deladdr returned flags 0x%x", rn->rn_flags)); 807 rt = (struct rtentry *)rn; 808 809 /* ref to prevent a deletion race */ 810 ++rt->rt_refcnt; 811 812 /* Free any routes cloned from this one. */ 813 if ((rt->rt_flags & (RTF_CLONING | RTF_PRCLONING)) && 814 rt_mask(rt) != NULL) { 815 rnh->rnh_walktree_from(rnh, (char *)rt_key(rt), 816 (char *)rt_mask(rt), 817 rt_fixdelete, rt); 818 } 819 820 if (rt->rt_gwroute != NULL) { 821 RTFREE(rt->rt_gwroute); 822 rt->rt_gwroute = NULL; 823 } 824 825 /* 826 * NB: RTF_UP must be set during the search above, 827 * because we might delete the last ref, causing 828 * rt to get freed prematurely. 829 */ 830 rt->rt_flags &= ~RTF_UP; 831 832 #ifdef ROUTE_DEBUG 833 if (route_debug) 834 rt_print(rtinfo, rt); 835 #endif 836 837 /* Give the protocol a chance to keep things in sync. */ 838 if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest) 839 ifa->ifa_rtrequest(RTM_DELETE, rt); 840 841 /* 842 * If the caller wants it, then it can have it, 843 * but it's up to it to free the rtentry as we won't be 844 * doing it. 845 */ 846 KASSERT(rt->rt_refcnt >= 0, 847 ("rtrequest1(DELETE): refcnt %ld", rt->rt_refcnt)); 848 if (ret_nrt != NULL) { 849 /* leave ref intact for return */ 850 *ret_nrt = rt; 851 } else { 852 /* deref / attempt to destroy */ 853 rtfree(rt); 854 } 855 break; 856 857 case RTM_RESOLVE: 858 if (ret_nrt == NULL || (rt = *ret_nrt) == NULL) 859 gotoerr(EINVAL); 860 861 KASSERT(rt->rt_cpuid == mycpuid, 862 ("rt resolve rt_cpuid %d, mycpuid %d", 863 rt->rt_cpuid, mycpuid)); 864 865 ifa = rt->rt_ifa; 866 rtinfo->rti_flags = 867 rt->rt_flags & ~(RTF_CLONING | RTF_PRCLONING | RTF_STATIC); 868 rtinfo->rti_flags |= RTF_WASCLONED; 869 rtinfo->rti_info[RTAX_GATEWAY] = rt->rt_gateway; 870 if ((rtinfo->rti_info[RTAX_NETMASK] = rt->rt_genmask) == NULL) 871 rtinfo->rti_flags |= RTF_HOST; 872 rtinfo->rti_info[RTAX_MPLS1] = rt->rt_shim[0]; 873 rtinfo->rti_info[RTAX_MPLS2] = rt->rt_shim[1]; 874 rtinfo->rti_info[RTAX_MPLS3] = rt->rt_shim[2]; 875 goto makeroute; 876 877 case RTM_ADD: 878 KASSERT(!(rtinfo->rti_flags & RTF_GATEWAY) || 879 rtinfo->rti_info[RTAX_GATEWAY] != NULL, 880 ("rtrequest: GATEWAY but no gateway")); 881 882 if (rtinfo->rti_ifa == NULL && (error = rt_getifa(rtinfo))) 883 gotoerr(error); 884 ifa = rtinfo->rti_ifa; 885 makeroute: 886 R_Malloc(rt, struct rtentry *, sizeof(struct rtentry)); 887 if (rt == NULL) { 888 if (req == RTM_ADD) { 889 kprintf("rtrequest1: alloc rtentry failed on " 890 "cpu%d\n", mycpuid); 891 } 892 gotoerr(ENOBUFS); 893 } 894 bzero(rt, sizeof(struct rtentry)); 895 rt->rt_flags = RTF_UP | rtinfo->rti_flags; 896 rt->rt_cpuid = mycpuid; 897 898 if (mycpuid != 0 && req == RTM_ADD) { 899 /* For RTM_ADD, we have already sent rtmsg on CPU0. */ 900 reportmsg = RTL_DONTREPORT; 901 } else { 902 /* 903 * For RTM_ADD, we only send rtmsg on CPU0. 904 * For RTM_RESOLVE, we always send rtmsg. XXX 905 */ 906 reportmsg = RTL_REPORTMSG; 907 } 908 error = rt_setgate(rt, dst, rtinfo->rti_info[RTAX_GATEWAY], 909 reportmsg); 910 if (error != 0) { 911 Free(rt); 912 gotoerr(error); 913 } 914 915 ndst = rt_key(rt); 916 if (rtinfo->rti_info[RTAX_NETMASK] != NULL) 917 rt_maskedcopy(dst, ndst, 918 rtinfo->rti_info[RTAX_NETMASK]); 919 else 920 bcopy(dst, ndst, dst->sa_len); 921 922 if (rtinfo->rti_info[RTAX_MPLS1] != NULL) 923 rt_setshims(rt, rtinfo->rti_info); 924 925 /* 926 * Note that we now have a reference to the ifa. 927 * This moved from below so that rnh->rnh_addaddr() can 928 * examine the ifa and ifa->ifa_ifp if it so desires. 929 */ 930 IFAREF(ifa); 931 rt->rt_ifa = ifa; 932 rt->rt_ifp = ifa->ifa_ifp; 933 /* XXX mtu manipulation will be done in rnh_addaddr -- itojun */ 934 935 rn = rnh->rnh_addaddr((char *)ndst, 936 (char *)rtinfo->rti_info[RTAX_NETMASK], 937 rnh, rt->rt_nodes); 938 if (rn == NULL) { 939 struct rtentry *oldrt; 940 941 /* 942 * We already have one of these in the tree. 943 * We do a special hack: if the old route was 944 * cloned, then we blow it away and try 945 * re-inserting the new one. 946 */ 947 oldrt = rtpurelookup(ndst); 948 if (oldrt != NULL) { 949 --oldrt->rt_refcnt; 950 if (oldrt->rt_flags & RTF_WASCLONED) { 951 rtrequest(RTM_DELETE, rt_key(oldrt), 952 oldrt->rt_gateway, 953 rt_mask(oldrt), 954 oldrt->rt_flags, NULL); 955 rn = rnh->rnh_addaddr((char *)ndst, 956 (char *) 957 rtinfo->rti_info[RTAX_NETMASK], 958 rnh, rt->rt_nodes); 959 } 960 } 961 } 962 963 /* 964 * If it still failed to go into the tree, 965 * then un-make it (this should be a function). 966 */ 967 if (rn == NULL) { 968 if (rt->rt_gwroute != NULL) 969 rtfree(rt->rt_gwroute); 970 IFAFREE(ifa); 971 Free(rt_key(rt)); 972 Free(rt); 973 gotoerr(EEXIST); 974 } 975 976 /* 977 * If we got here from RESOLVE, then we are cloning 978 * so clone the rest, and note that we 979 * are a clone (and increment the parent's references) 980 */ 981 if (req == RTM_RESOLVE) { 982 rt->rt_rmx = (*ret_nrt)->rt_rmx; /* copy metrics */ 983 rt->rt_rmx.rmx_pksent = 0; /* reset packet counter */ 984 if ((*ret_nrt)->rt_flags & 985 (RTF_CLONING | RTF_PRCLONING)) { 986 rt->rt_parent = *ret_nrt; 987 (*ret_nrt)->rt_refcnt++; 988 } 989 } 990 991 /* 992 * if this protocol has something to add to this then 993 * allow it to do that as well. 994 */ 995 if (ifa->ifa_rtrequest != NULL) 996 ifa->ifa_rtrequest(req, rt); 997 998 /* 999 * We repeat the same procedure from rt_setgate() here because 1000 * it doesn't fire when we call it there because the node 1001 * hasn't been added to the tree yet. 1002 */ 1003 if (req == RTM_ADD && !(rt->rt_flags & RTF_HOST) && 1004 rt_mask(rt) != NULL) { 1005 struct rtfc_arg arg = { rt, rnh }; 1006 1007 rnh->rnh_walktree_from(rnh, (char *)rt_key(rt), 1008 (char *)rt_mask(rt), 1009 rt_fixchange, &arg); 1010 } 1011 1012 #ifdef ROUTE_DEBUG 1013 if (route_debug) 1014 rt_print(rtinfo, rt); 1015 #endif 1016 /* 1017 * Return the resulting rtentry, 1018 * increasing the number of references by one. 1019 */ 1020 if (ret_nrt != NULL) { 1021 rt->rt_refcnt++; 1022 *ret_nrt = rt; 1023 } 1024 break; 1025 default: 1026 error = EOPNOTSUPP; 1027 } 1028 bad: 1029 #ifdef ROUTE_DEBUG 1030 if (route_debug) { 1031 if (error) 1032 kprintf("rti %p failed error %d\n", rtinfo, error); 1033 else 1034 kprintf("rti %p succeeded\n", rtinfo); 1035 } 1036 #endif 1037 crit_exit(); 1038 return (error); 1039 } 1040 1041 /* 1042 * Called from rtrequest(RTM_DELETE, ...) to fix up the route's ``family'' 1043 * (i.e., the routes related to it by the operation of cloning). This 1044 * routine is iterated over all potential former-child-routes by way of 1045 * rnh->rnh_walktree_from() above, and those that actually are children of 1046 * the late parent (passed in as VP here) are themselves deleted. 1047 */ 1048 static int 1049 rt_fixdelete(struct radix_node *rn, void *vp) 1050 { 1051 struct rtentry *rt = (struct rtentry *)rn; 1052 struct rtentry *rt0 = vp; 1053 1054 if (rt->rt_parent == rt0 && 1055 !(rt->rt_flags & (RTF_PINNED | RTF_CLONING | RTF_PRCLONING))) { 1056 return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 1057 rt->rt_flags, NULL); 1058 } 1059 return 0; 1060 } 1061 1062 /* 1063 * This routine is called from rt_setgate() to do the analogous thing for 1064 * adds and changes. There is the added complication in this case of a 1065 * middle insert; i.e., insertion of a new network route between an older 1066 * network route and (cloned) host routes. For this reason, a simple check 1067 * of rt->rt_parent is insufficient; each candidate route must be tested 1068 * against the (mask, value) of the new route (passed as before in vp) 1069 * to see if the new route matches it. 1070 * 1071 * XXX - it may be possible to do fixdelete() for changes and reserve this 1072 * routine just for adds. I'm not sure why I thought it was necessary to do 1073 * changes this way. 1074 */ 1075 #ifdef DEBUG 1076 static int rtfcdebug = 0; 1077 #endif 1078 1079 static int 1080 rt_fixchange(struct radix_node *rn, void *vp) 1081 { 1082 struct rtentry *rt = (struct rtentry *)rn; 1083 struct rtfc_arg *ap = vp; 1084 struct rtentry *rt0 = ap->rt0; 1085 struct radix_node_head *rnh = ap->rnh; 1086 u_char *xk1, *xm1, *xk2, *xmp; 1087 int i, len, mlen; 1088 1089 #ifdef DEBUG 1090 if (rtfcdebug) 1091 kprintf("rt_fixchange: rt %p, rt0 %p\n", rt, rt0); 1092 #endif 1093 1094 if (rt->rt_parent == NULL || 1095 (rt->rt_flags & (RTF_PINNED | RTF_CLONING | RTF_PRCLONING))) { 1096 #ifdef DEBUG 1097 if (rtfcdebug) kprintf("no parent, pinned or cloning\n"); 1098 #endif 1099 return 0; 1100 } 1101 1102 if (rt->rt_parent == rt0) { 1103 #ifdef DEBUG 1104 if (rtfcdebug) kprintf("parent match\n"); 1105 #endif 1106 return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 1107 rt->rt_flags, NULL); 1108 } 1109 1110 /* 1111 * There probably is a function somewhere which does this... 1112 * if not, there should be. 1113 */ 1114 len = imin(rt_key(rt0)->sa_len, rt_key(rt)->sa_len); 1115 1116 xk1 = (u_char *)rt_key(rt0); 1117 xm1 = (u_char *)rt_mask(rt0); 1118 xk2 = (u_char *)rt_key(rt); 1119 1120 /* avoid applying a less specific route */ 1121 xmp = (u_char *)rt_mask(rt->rt_parent); 1122 mlen = rt_key(rt->rt_parent)->sa_len; 1123 if (mlen > rt_key(rt0)->sa_len) { 1124 #ifdef DEBUG 1125 if (rtfcdebug) 1126 kprintf("rt_fixchange: inserting a less " 1127 "specific route\n"); 1128 #endif 1129 return 0; 1130 } 1131 for (i = rnh->rnh_treetop->rn_offset; i < mlen; i++) { 1132 if ((xmp[i] & ~(xmp[i] ^ xm1[i])) != xmp[i]) { 1133 #ifdef DEBUG 1134 if (rtfcdebug) 1135 kprintf("rt_fixchange: inserting a less " 1136 "specific route\n"); 1137 #endif 1138 return 0; 1139 } 1140 } 1141 1142 for (i = rnh->rnh_treetop->rn_offset; i < len; i++) { 1143 if ((xk2[i] & xm1[i]) != xk1[i]) { 1144 #ifdef DEBUG 1145 if (rtfcdebug) kprintf("no match\n"); 1146 #endif 1147 return 0; 1148 } 1149 } 1150 1151 /* 1152 * OK, this node is a clone, and matches the node currently being 1153 * changed/added under the node's mask. So, get rid of it. 1154 */ 1155 #ifdef DEBUG 1156 if (rtfcdebug) kprintf("deleting\n"); 1157 #endif 1158 return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 1159 rt->rt_flags, NULL); 1160 } 1161 1162 int 1163 rt_setgate(struct rtentry *rt0, struct sockaddr *dst, struct sockaddr *gate, 1164 boolean_t generate_report) 1165 { 1166 char *space, *oldspace; 1167 int dlen = RT_ROUNDUP(dst->sa_len), glen = RT_ROUNDUP(gate->sa_len); 1168 struct rtentry *rt = rt0; 1169 struct radix_node_head *rnh = rt_tables[mycpuid][dst->sa_family]; 1170 1171 /* 1172 * A host route with the destination equal to the gateway 1173 * will interfere with keeping LLINFO in the routing 1174 * table, so disallow it. 1175 */ 1176 if (((rt0->rt_flags & (RTF_HOST | RTF_GATEWAY | RTF_LLINFO)) == 1177 (RTF_HOST | RTF_GATEWAY)) && 1178 dst->sa_len == gate->sa_len && 1179 sa_equal(dst, gate)) { 1180 /* 1181 * The route might already exist if this is an RTM_CHANGE 1182 * or a routing redirect, so try to delete it. 1183 */ 1184 if (rt_key(rt0) != NULL) 1185 rtrequest(RTM_DELETE, rt_key(rt0), rt0->rt_gateway, 1186 rt_mask(rt0), rt0->rt_flags, NULL); 1187 return EADDRNOTAVAIL; 1188 } 1189 1190 /* 1191 * Both dst and gateway are stored in the same malloc'ed chunk 1192 * (If I ever get my hands on....) 1193 * if we need to malloc a new chunk, then keep the old one around 1194 * till we don't need it any more. 1195 */ 1196 if (rt->rt_gateway == NULL || 1197 glen > RT_ROUNDUP(rt->rt_gateway->sa_len)) { 1198 oldspace = (char *)rt_key(rt); 1199 R_Malloc(space, char *, dlen + glen); 1200 if (space == NULL) 1201 return ENOBUFS; 1202 rt->rt_nodes->rn_key = space; 1203 } else { 1204 space = (char *)rt_key(rt); /* Just use the old space. */ 1205 oldspace = NULL; 1206 } 1207 1208 /* Set the gateway value. */ 1209 rt->rt_gateway = (struct sockaddr *)(space + dlen); 1210 bcopy(gate, rt->rt_gateway, glen); 1211 1212 if (oldspace != NULL) { 1213 /* 1214 * If we allocated a new chunk, preserve the original dst. 1215 * This way, rt_setgate() really just sets the gate 1216 * and leaves the dst field alone. 1217 */ 1218 bcopy(dst, space, dlen); 1219 Free(oldspace); 1220 } 1221 1222 /* 1223 * If there is already a gwroute, it's now almost definitely wrong 1224 * so drop it. 1225 */ 1226 if (rt->rt_gwroute != NULL) { 1227 RTFREE(rt->rt_gwroute); 1228 rt->rt_gwroute = NULL; 1229 } 1230 if (rt->rt_flags & RTF_GATEWAY) { 1231 /* 1232 * Cloning loop avoidance: In the presence of 1233 * protocol-cloning and bad configuration, it is 1234 * possible to get stuck in bottomless mutual recursion 1235 * (rtrequest rt_setgate rtlookup). We avoid this 1236 * by not allowing protocol-cloning to operate for 1237 * gateways (which is probably the correct choice 1238 * anyway), and avoid the resulting reference loops 1239 * by disallowing any route to run through itself as 1240 * a gateway. This is obviously mandatory when we 1241 * get rt->rt_output(). 1242 * 1243 * This breaks TTCP for hosts outside the gateway! XXX JH 1244 */ 1245 rt->rt_gwroute = _rtlookup(gate, generate_report, 1246 RTF_PRCLONING); 1247 if (rt->rt_gwroute == rt) { 1248 rt->rt_gwroute = NULL; 1249 --rt->rt_refcnt; 1250 return EDQUOT; /* failure */ 1251 } 1252 } 1253 1254 /* 1255 * This isn't going to do anything useful for host routes, so 1256 * don't bother. Also make sure we have a reasonable mask 1257 * (we don't yet have one during adds). 1258 */ 1259 if (!(rt->rt_flags & RTF_HOST) && rt_mask(rt) != NULL) { 1260 struct rtfc_arg arg = { rt, rnh }; 1261 1262 rnh->rnh_walktree_from(rnh, (char *)rt_key(rt), 1263 (char *)rt_mask(rt), 1264 rt_fixchange, &arg); 1265 } 1266 1267 return 0; 1268 } 1269 1270 static void 1271 rt_maskedcopy( 1272 struct sockaddr *src, 1273 struct sockaddr *dst, 1274 struct sockaddr *netmask) 1275 { 1276 u_char *cp1 = (u_char *)src; 1277 u_char *cp2 = (u_char *)dst; 1278 u_char *cp3 = (u_char *)netmask; 1279 u_char *cplim = cp2 + *cp3; 1280 u_char *cplim2 = cp2 + *cp1; 1281 1282 *cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */ 1283 cp3 += 2; 1284 if (cplim > cplim2) 1285 cplim = cplim2; 1286 while (cp2 < cplim) 1287 *cp2++ = *cp1++ & *cp3++; 1288 if (cp2 < cplim2) 1289 bzero(cp2, cplim2 - cp2); 1290 } 1291 1292 int 1293 rt_llroute(struct sockaddr *dst, struct rtentry *rt0, struct rtentry **drt) 1294 { 1295 struct rtentry *up_rt, *rt; 1296 1297 if (!(rt0->rt_flags & RTF_UP)) { 1298 up_rt = rtlookup(dst); 1299 if (up_rt == NULL) 1300 return (EHOSTUNREACH); 1301 up_rt->rt_refcnt--; 1302 } else 1303 up_rt = rt0; 1304 if (up_rt->rt_flags & RTF_GATEWAY) { 1305 if (up_rt->rt_gwroute == NULL) { 1306 up_rt->rt_gwroute = rtlookup(up_rt->rt_gateway); 1307 if (up_rt->rt_gwroute == NULL) 1308 return (EHOSTUNREACH); 1309 } else if (!(up_rt->rt_gwroute->rt_flags & RTF_UP)) { 1310 rtfree(up_rt->rt_gwroute); 1311 up_rt->rt_gwroute = rtlookup(up_rt->rt_gateway); 1312 if (up_rt->rt_gwroute == NULL) 1313 return (EHOSTUNREACH); 1314 } 1315 rt = up_rt->rt_gwroute; 1316 } else 1317 rt = up_rt; 1318 if (rt->rt_flags & RTF_REJECT && 1319 (rt->rt_rmx.rmx_expire == 0 || /* rt doesn't expire */ 1320 time_uptime < rt->rt_rmx.rmx_expire)) /* rt not expired */ 1321 return (rt->rt_flags & RTF_HOST ? EHOSTDOWN : EHOSTUNREACH); 1322 *drt = rt; 1323 return 0; 1324 } 1325 1326 static int 1327 rt_setshims(struct rtentry *rt, struct sockaddr **rt_shim){ 1328 int i; 1329 1330 for (i=0; i<3; i++) { 1331 struct sockaddr *shim = rt_shim[RTAX_MPLS1 + i]; 1332 int shimlen; 1333 1334 if (shim == NULL) 1335 break; 1336 1337 shimlen = RT_ROUNDUP(shim->sa_len); 1338 R_Malloc(rt->rt_shim[i], struct sockaddr *, shimlen); 1339 bcopy(shim, rt->rt_shim[i], shimlen); 1340 } 1341 1342 return 0; 1343 } 1344 1345 #ifdef ROUTE_DEBUG 1346 1347 /* 1348 * Print out a route table entry 1349 */ 1350 void 1351 rt_print(struct rt_addrinfo *rtinfo, struct rtentry *rn) 1352 { 1353 kprintf("rti %p cpu %d route %p flags %08lx: ", 1354 rtinfo, mycpuid, rn, rn->rt_flags); 1355 sockaddr_print(rt_key(rn)); 1356 kprintf(" mask "); 1357 sockaddr_print(rt_mask(rn)); 1358 kprintf(" gw "); 1359 sockaddr_print(rn->rt_gateway); 1360 kprintf(" ifc \"%s\"", rn->rt_ifp ? rn->rt_ifp->if_dname : "?"); 1361 kprintf(" ifa %p\n", rn->rt_ifa); 1362 } 1363 1364 void 1365 rt_addrinfo_print(int cmd, struct rt_addrinfo *rti) 1366 { 1367 int didit = 0; 1368 int i; 1369 1370 #ifdef ROUTE_DEBUG 1371 if (cmd == RTM_DELETE && route_debug > 1) 1372 print_backtrace(-1); 1373 #endif 1374 1375 switch(cmd) { 1376 case RTM_ADD: 1377 kprintf("ADD "); 1378 break; 1379 case RTM_RESOLVE: 1380 kprintf("RES "); 1381 break; 1382 case RTM_DELETE: 1383 kprintf("DEL "); 1384 break; 1385 default: 1386 kprintf("C%02d ", cmd); 1387 break; 1388 } 1389 kprintf("rti %p cpu %d ", rti, mycpuid); 1390 for (i = 0; i < rti->rti_addrs; ++i) { 1391 if (rti->rti_info[i] == NULL) 1392 continue; 1393 if (didit) 1394 kprintf(" ,"); 1395 switch(i) { 1396 case RTAX_DST: 1397 kprintf("(DST "); 1398 break; 1399 case RTAX_GATEWAY: 1400 kprintf("(GWY "); 1401 break; 1402 case RTAX_NETMASK: 1403 kprintf("(MSK "); 1404 break; 1405 case RTAX_GENMASK: 1406 kprintf("(GEN "); 1407 break; 1408 case RTAX_IFP: 1409 kprintf("(IFP "); 1410 break; 1411 case RTAX_IFA: 1412 kprintf("(IFA "); 1413 break; 1414 case RTAX_AUTHOR: 1415 kprintf("(AUT "); 1416 break; 1417 case RTAX_BRD: 1418 kprintf("(BRD "); 1419 break; 1420 default: 1421 kprintf("(?%02d ", i); 1422 break; 1423 } 1424 sockaddr_print(rti->rti_info[i]); 1425 kprintf(")"); 1426 didit = 1; 1427 } 1428 kprintf("\n"); 1429 } 1430 1431 void 1432 sockaddr_print(struct sockaddr *sa) 1433 { 1434 struct sockaddr_in *sa4; 1435 struct sockaddr_in6 *sa6; 1436 int len; 1437 int i; 1438 1439 if (sa == NULL) { 1440 kprintf("NULL"); 1441 return; 1442 } 1443 1444 len = sa->sa_len - offsetof(struct sockaddr, sa_data[0]); 1445 1446 switch(sa->sa_family) { 1447 case AF_INET: 1448 case AF_INET6: 1449 default: 1450 switch(sa->sa_family) { 1451 case AF_INET: 1452 sa4 = (struct sockaddr_in *)sa; 1453 kprintf("INET %d %d.%d.%d.%d", 1454 ntohs(sa4->sin_port), 1455 (ntohl(sa4->sin_addr.s_addr) >> 24) & 255, 1456 (ntohl(sa4->sin_addr.s_addr) >> 16) & 255, 1457 (ntohl(sa4->sin_addr.s_addr) >> 8) & 255, 1458 (ntohl(sa4->sin_addr.s_addr) >> 0) & 255 1459 ); 1460 break; 1461 case AF_INET6: 1462 sa6 = (struct sockaddr_in6 *)sa; 1463 kprintf("INET6 %d %04x:%04x%04x:%04x:%04x:%04x:%04x:%04x", 1464 ntohs(sa6->sin6_port), 1465 sa6->sin6_addr.s6_addr16[0], 1466 sa6->sin6_addr.s6_addr16[1], 1467 sa6->sin6_addr.s6_addr16[2], 1468 sa6->sin6_addr.s6_addr16[3], 1469 sa6->sin6_addr.s6_addr16[4], 1470 sa6->sin6_addr.s6_addr16[5], 1471 sa6->sin6_addr.s6_addr16[6], 1472 sa6->sin6_addr.s6_addr16[7] 1473 ); 1474 break; 1475 default: 1476 kprintf("AF%d ", sa->sa_family); 1477 while (len > 0 && sa->sa_data[len-1] == 0) 1478 --len; 1479 1480 for (i = 0; i < len; ++i) { 1481 if (i) 1482 kprintf("."); 1483 kprintf("%d", (unsigned char)sa->sa_data[i]); 1484 } 1485 break; 1486 } 1487 } 1488 } 1489 1490 #endif 1491 1492 /* 1493 * Set up a routing table entry, normally for an interface. 1494 */ 1495 int 1496 rtinit(struct ifaddr *ifa, int cmd, int flags) 1497 { 1498 struct sockaddr *dst, *deldst, *netmask; 1499 struct mbuf *m = NULL; 1500 struct radix_node_head *rnh; 1501 struct radix_node *rn; 1502 struct rt_addrinfo rtinfo; 1503 int error; 1504 1505 if (flags & RTF_HOST) { 1506 dst = ifa->ifa_dstaddr; 1507 netmask = NULL; 1508 } else { 1509 dst = ifa->ifa_addr; 1510 netmask = ifa->ifa_netmask; 1511 } 1512 /* 1513 * If it's a delete, check that if it exists, it's on the correct 1514 * interface or we might scrub a route to another ifa which would 1515 * be confusing at best and possibly worse. 1516 */ 1517 if (cmd == RTM_DELETE) { 1518 /* 1519 * It's a delete, so it should already exist.. 1520 * If it's a net, mask off the host bits 1521 * (Assuming we have a mask) 1522 */ 1523 if (netmask != NULL) { 1524 m = m_get(M_NOWAIT, MT_SONAME); 1525 if (m == NULL) 1526 return (ENOBUFS); 1527 mbuftrackid(m, 34); 1528 deldst = mtod(m, struct sockaddr *); 1529 rt_maskedcopy(dst, deldst, netmask); 1530 dst = deldst; 1531 } 1532 /* 1533 * Look up an rtentry that is in the routing tree and 1534 * contains the correct info. 1535 */ 1536 if ((rnh = rt_tables[mycpuid][dst->sa_family]) == NULL || 1537 (rn = rnh->rnh_lookup((char *)dst, 1538 (char *)netmask, rnh)) == NULL || 1539 ((struct rtentry *)rn)->rt_ifa != ifa || 1540 !sa_equal((struct sockaddr *)rn->rn_key, dst)) { 1541 if (m != NULL) 1542 m_free(m); 1543 return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH); 1544 } 1545 /* XXX */ 1546 #if 0 1547 else { 1548 /* 1549 * One would think that as we are deleting, and we know 1550 * it doesn't exist, we could just return at this point 1551 * with an "ELSE" clause, but apparently not.. 1552 */ 1553 return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH); 1554 } 1555 #endif 1556 } 1557 /* 1558 * Do the actual request 1559 */ 1560 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 1561 rtinfo.rti_info[RTAX_DST] = dst; 1562 rtinfo.rti_info[RTAX_GATEWAY] = ifa->ifa_addr; 1563 rtinfo.rti_info[RTAX_NETMASK] = netmask; 1564 rtinfo.rti_flags = flags | ifa->ifa_flags; 1565 rtinfo.rti_ifa = ifa; 1566 error = rtrequest1_global(cmd, &rtinfo, rtinit_rtrequest_callback, ifa, 1567 RTREQ_PRIO_HIGH); 1568 if (m != NULL) 1569 m_free(m); 1570 return (error); 1571 } 1572 1573 static void 1574 rtinit_rtrequest_callback(int cmd, int error, 1575 struct rt_addrinfo *rtinfo, struct rtentry *rt, 1576 void *arg) 1577 { 1578 struct ifaddr *ifa = arg; 1579 1580 if (error == 0 && rt) { 1581 if (mycpuid == 0) { 1582 ++rt->rt_refcnt; 1583 rt_newaddrmsg(cmd, ifa, error, rt); 1584 --rt->rt_refcnt; 1585 } 1586 if (cmd == RTM_DELETE) { 1587 if (rt->rt_refcnt == 0) { 1588 ++rt->rt_refcnt; 1589 rtfree(rt); 1590 } 1591 } 1592 } 1593 } 1594 1595 struct netmsg_rts { 1596 struct netmsg_base base; 1597 int req; 1598 struct rt_addrinfo *rtinfo; 1599 rtsearch_callback_func_t callback; 1600 void *arg; 1601 boolean_t exact_match; 1602 int found_cnt; 1603 }; 1604 1605 int 1606 rtsearch_global(int req, struct rt_addrinfo *rtinfo, 1607 rtsearch_callback_func_t callback, void *arg, boolean_t exact_match, 1608 boolean_t req_prio) 1609 { 1610 struct netmsg_rts msg; 1611 int flags = 0; 1612 1613 if (req_prio) 1614 flags = MSGF_PRIORITY; 1615 netmsg_init(&msg.base, NULL, &curthread->td_msgport, flags, 1616 rtsearch_msghandler); 1617 msg.req = req; 1618 msg.rtinfo = rtinfo; 1619 msg.callback = callback; 1620 msg.arg = arg; 1621 msg.exact_match = exact_match; 1622 msg.found_cnt = 0; 1623 return rt_domsg_global(&msg.base); 1624 } 1625 1626 static void 1627 rtsearch_msghandler(netmsg_t msg) 1628 { 1629 struct netmsg_rts *rmsg = (void *)msg; 1630 struct rt_addrinfo rtinfo; 1631 struct radix_node_head *rnh; 1632 struct rtentry *rt; 1633 int nextcpu, error; 1634 1635 /* 1636 * Copy the rtinfo. We need to make sure that the original 1637 * rtinfo, which is setup by the caller, in the netmsg will 1638 * _not_ be changed; else the next CPU on the netmsg forwarding 1639 * path will see a different rtinfo than what this CPU has seen. 1640 */ 1641 rtinfo = *rmsg->rtinfo; 1642 1643 /* 1644 * Find the correct routing tree to use for this Address Family 1645 */ 1646 if ((rnh = rt_tables[mycpuid][rtinfo.rti_dst->sa_family]) == NULL) { 1647 if (mycpuid != 0) 1648 panic("partially initialized routing tables"); 1649 lwkt_replymsg(&rmsg->base.lmsg, EAFNOSUPPORT); 1650 return; 1651 } 1652 1653 /* 1654 * Correct rtinfo for the host route searching. 1655 */ 1656 if (rtinfo.rti_flags & RTF_HOST) { 1657 rtinfo.rti_netmask = NULL; 1658 rtinfo.rti_flags &= ~(RTF_CLONING | RTF_PRCLONING); 1659 } 1660 1661 rt = (struct rtentry *) 1662 rnh->rnh_lookup((char *)rtinfo.rti_dst, 1663 (char *)rtinfo.rti_netmask, rnh); 1664 1665 /* 1666 * If we are asked to do the "exact match", we need to make sure 1667 * that host route searching got a host route while a network 1668 * route searching got a network route. 1669 */ 1670 if (rt != NULL && rmsg->exact_match && 1671 ((rt->rt_flags ^ rtinfo.rti_flags) & RTF_HOST)) 1672 rt = NULL; 1673 1674 if (rt == NULL) { 1675 /* 1676 * No matching routes have been found, don't count this 1677 * as a critical error (here, we set 'error' to 0), just 1678 * keep moving on, since at least prcloned routes are not 1679 * duplicated onto each CPU. 1680 */ 1681 error = 0; 1682 } else { 1683 rmsg->found_cnt++; 1684 1685 rt->rt_refcnt++; 1686 error = rmsg->callback(rmsg->req, &rtinfo, rt, rmsg->arg, 1687 rmsg->found_cnt); 1688 rt->rt_refcnt--; 1689 1690 if (error == EJUSTRETURN) { 1691 lwkt_replymsg(&rmsg->base.lmsg, 0); 1692 return; 1693 } 1694 } 1695 1696 nextcpu = mycpuid + 1; 1697 if (error) { 1698 KKASSERT(rmsg->found_cnt > 0); 1699 1700 /* 1701 * Under following cases, unrecoverable error has 1702 * not occured: 1703 * o Request is RTM_GET 1704 * o The first time that we find the route, but the 1705 * modification fails. 1706 */ 1707 if (rmsg->req != RTM_GET && rmsg->found_cnt > 1) { 1708 panic("rtsearch_msghandler: unrecoverable error " 1709 "cpu %d", mycpuid); 1710 } 1711 lwkt_replymsg(&rmsg->base.lmsg, error); 1712 } else if (nextcpu < ncpus) { 1713 lwkt_forwardmsg(netisr_cpuport(nextcpu), &rmsg->base.lmsg); 1714 } else { 1715 if (rmsg->found_cnt == 0) { 1716 /* The requested route was never seen ... */ 1717 error = ESRCH; 1718 } 1719 lwkt_replymsg(&rmsg->base.lmsg, error); 1720 } 1721 } 1722 1723 int 1724 rtmask_add_global(struct sockaddr *mask, boolean_t req_prio) 1725 { 1726 struct netmsg_base msg; 1727 int flags = 0; 1728 1729 if (req_prio) 1730 flags = MSGF_PRIORITY; 1731 netmsg_init(&msg, NULL, &curthread->td_msgport, flags, 1732 rtmask_add_msghandler); 1733 msg.lmsg.u.ms_resultp = mask; 1734 1735 return rt_domsg_global(&msg); 1736 } 1737 1738 struct sockaddr * 1739 _rtmask_lookup(struct sockaddr *mask, boolean_t search) 1740 { 1741 struct radix_node *n; 1742 1743 #define clen(s) (*(u_char *)(s)) 1744 n = rn_addmask((char *)mask, search, 1, rn_cpumaskhead(mycpuid)); 1745 if (n != NULL && 1746 mask->sa_len >= clen(n->rn_key) && 1747 bcmp((char *)mask + 1, 1748 (char *)n->rn_key + 1, clen(n->rn_key) - 1) == 0) { 1749 return (struct sockaddr *)n->rn_key; 1750 } else { 1751 return NULL; 1752 } 1753 #undef clen 1754 } 1755 1756 static void 1757 rtmask_add_msghandler(netmsg_t msg) 1758 { 1759 struct lwkt_msg *lmsg = &msg->lmsg; 1760 struct sockaddr *mask = lmsg->u.ms_resultp; 1761 int error = 0, nextcpu; 1762 1763 if (rtmask_lookup(mask) == NULL) 1764 error = ENOBUFS; 1765 1766 nextcpu = mycpuid + 1; 1767 if (!error && nextcpu < ncpus) 1768 lwkt_forwardmsg(netisr_cpuport(nextcpu), lmsg); 1769 else 1770 lwkt_replymsg(lmsg, error); 1771 } 1772 1773 /* This must be before ip6_init2(), which is now SI_ORDER_MIDDLE */ 1774 SYSINIT(route, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0); 1775 1776 struct rtchange_arg { 1777 struct ifaddr *old_ifa; 1778 struct ifaddr *new_ifa; 1779 struct rtentry *rt; 1780 int changed; 1781 }; 1782 1783 static void 1784 rtchange_ifa(struct rtentry *rt, struct rtchange_arg *ap) 1785 { 1786 if (rt->rt_ifa->ifa_rtrequest != NULL) 1787 rt->rt_ifa->ifa_rtrequest(RTM_DELETE, rt); 1788 IFAFREE(rt->rt_ifa); 1789 1790 IFAREF(ap->new_ifa); 1791 rt->rt_ifa = ap->new_ifa; 1792 rt->rt_ifp = ap->new_ifa->ifa_ifp; 1793 if (rt->rt_ifa->ifa_rtrequest != NULL) 1794 rt->rt_ifa->ifa_rtrequest(RTM_ADD, rt); 1795 1796 ap->changed = 1; 1797 } 1798 1799 static int 1800 rtchange_callback(struct radix_node *rn, void *xap) 1801 { 1802 struct rtchange_arg *ap = xap; 1803 struct rtentry *rt = (struct rtentry *)rn; 1804 1805 if (rt->rt_ifa == ap->old_ifa) { 1806 if (rt->rt_flags & (RTF_CLONING | RTF_PRCLONING)) { 1807 /* 1808 * We could saw the branch off when we are 1809 * still sitting on it, if the ifa_rtrequest 1810 * DEL/ADD are called directly from here. 1811 */ 1812 ap->rt = rt; 1813 return EJUSTRETURN; 1814 } 1815 rtchange_ifa(rt, ap); 1816 } 1817 return 0; 1818 } 1819 1820 struct netmsg_rtchange { 1821 struct netmsg_base base; 1822 struct ifaddr *old_ifa; 1823 struct ifaddr *new_ifa; 1824 int changed; 1825 }; 1826 1827 static void 1828 rtchange_dispatch(netmsg_t msg) 1829 { 1830 struct netmsg_rtchange *rmsg = (void *)msg; 1831 struct radix_node_head *rnh; 1832 struct rtchange_arg arg; 1833 int nextcpu, cpu; 1834 1835 cpu = mycpuid; 1836 1837 memset(&arg, 0, sizeof(arg)); 1838 arg.old_ifa = rmsg->old_ifa; 1839 arg.new_ifa = rmsg->new_ifa; 1840 1841 rnh = rt_tables[cpu][AF_INET]; 1842 for (;;) { 1843 int error; 1844 1845 KKASSERT(arg.rt == NULL); 1846 error = rnh->rnh_walktree(rnh, rtchange_callback, &arg); 1847 if (arg.rt != NULL) { 1848 struct rtentry *rt; 1849 1850 rt = arg.rt; 1851 arg.rt = NULL; 1852 rtchange_ifa(rt, &arg); 1853 } else { 1854 break; 1855 } 1856 } 1857 if (arg.changed) 1858 rmsg->changed = 1; 1859 1860 nextcpu = cpu + 1; 1861 if (nextcpu < ncpus) 1862 lwkt_forwardmsg(netisr_cpuport(nextcpu), &rmsg->base.lmsg); 1863 else 1864 lwkt_replymsg(&rmsg->base.lmsg, 0); 1865 } 1866 1867 int 1868 rtchange(struct ifaddr *old_ifa, struct ifaddr *new_ifa) 1869 { 1870 struct netmsg_rtchange msg; 1871 1872 /* 1873 * XXX individual requests are not independantly chained, 1874 * which means that the per-cpu route tables will not be 1875 * consistent in the middle of the operation. If routes 1876 * related to the interface are manipulated while we are 1877 * doing this the inconsistancy could trigger a panic. 1878 */ 1879 netmsg_init(&msg.base, NULL, &curthread->td_msgport, MSGF_PRIORITY, 1880 rtchange_dispatch); 1881 msg.old_ifa = old_ifa; 1882 msg.new_ifa = new_ifa; 1883 msg.changed = 0; 1884 rt_domsg_global(&msg.base); 1885 1886 if (msg.changed) { 1887 old_ifa->ifa_flags &= ~IFA_ROUTE; 1888 new_ifa->ifa_flags |= IFA_ROUTE; 1889 return 0; 1890 } else { 1891 return ENOENT; 1892 } 1893 } 1894 1895 int 1896 rt_domsg_global(struct netmsg_base *nmsg) 1897 { 1898 ASSERT_CANDOMSG_NETISR0(curthread); 1899 return lwkt_domsg(netisr_cpuport(0), &nmsg->lmsg, 0); 1900 } 1901