1 /* 2 * Copyright (c) 2004, 2005 The DragonFly Project. All rights reserved. 3 * 4 * This code is derived from software contributed to The DragonFly Project 5 * by Jeffrey M. Hsu. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 3. Neither the name of The DragonFly Project nor the names of its 16 * contributors may be used to endorse or promote products derived 17 * from this software without specific, prior written permission. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 21 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS 22 * FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE 23 * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, 24 * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING, 25 * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; 26 * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED 27 * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, 28 * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT 29 * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 30 * SUCH DAMAGE. 31 */ 32 33 /* 34 * Copyright (c) 1980, 1986, 1991, 1993 35 * The Regents of the University of California. All rights reserved. 36 * 37 * Redistribution and use in source and binary forms, with or without 38 * modification, are permitted provided that the following conditions 39 * are met: 40 * 1. Redistributions of source code must retain the above copyright 41 * notice, this list of conditions and the following disclaimer. 42 * 2. Redistributions in binary form must reproduce the above copyright 43 * notice, this list of conditions and the following disclaimer in the 44 * documentation and/or other materials provided with the distribution. 45 * 3. All advertising materials mentioning features or use of this software 46 * must display the following acknowledgement: 47 * This product includes software developed by the University of 48 * California, Berkeley and its contributors. 49 * 4. Neither the name of the University nor the names of its contributors 50 * may be used to endorse or promote products derived from this software 51 * without specific prior written permission. 52 * 53 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 54 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 55 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 56 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 57 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 58 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 59 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 60 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 61 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 62 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 63 * SUCH DAMAGE. 64 * 65 * @(#)route.c 8.3 (Berkeley) 1/9/95 66 * $FreeBSD: src/sys/net/route.c,v 1.59.2.10 2003/01/17 08:04:00 ru Exp $ 67 * $DragonFly: src/sys/net/route.c,v 1.41 2008/11/09 10:50:15 sephe Exp $ 68 */ 69 70 #include "opt_inet.h" 71 #include "opt_mpls.h" 72 73 #include <sys/param.h> 74 #include <sys/systm.h> 75 #include <sys/malloc.h> 76 #include <sys/mbuf.h> 77 #include <sys/socket.h> 78 #include <sys/domain.h> 79 #include <sys/kernel.h> 80 #include <sys/sysctl.h> 81 #include <sys/globaldata.h> 82 #include <sys/thread.h> 83 84 #include <net/if.h> 85 #include <net/route.h> 86 #include <net/netisr.h> 87 88 #include <netinet/in.h> 89 #include <net/ip_mroute/ip_mroute.h> 90 91 #include <sys/thread2.h> 92 #include <sys/msgport2.h> 93 #include <sys/mplock2.h> 94 #include <net/netmsg2.h> 95 96 #ifdef MPLS 97 #include <netproto/mpls/mpls.h> 98 #endif 99 100 static struct rtstatistics rtstatistics_percpu[MAXCPU]; 101 #ifdef SMP 102 #define rtstat rtstatistics_percpu[mycpuid] 103 #else 104 #define rtstat rtstatistics_percpu[0] 105 #endif 106 107 struct radix_node_head *rt_tables[MAXCPU][AF_MAX+1]; 108 struct lwkt_port *rt_ports[MAXCPU]; 109 110 static void rt_maskedcopy (struct sockaddr *, struct sockaddr *, 111 struct sockaddr *); 112 static void rtable_init(void); 113 static void rtable_service_loop(void *dummy); 114 static void rtinit_rtrequest_callback(int, int, struct rt_addrinfo *, 115 struct rtentry *, void *); 116 117 #ifdef SMP 118 static void rtredirect_msghandler(netmsg_t msg); 119 static void rtrequest1_msghandler(netmsg_t msg); 120 #endif 121 static void rtsearch_msghandler(netmsg_t msg); 122 static void rtmask_add_msghandler(netmsg_t msg); 123 124 static int rt_setshims(struct rtentry *, struct sockaddr **); 125 126 SYSCTL_NODE(_net, OID_AUTO, route, CTLFLAG_RW, 0, "Routing"); 127 128 #ifdef ROUTE_DEBUG 129 static int route_debug = 1; 130 SYSCTL_INT(_net_route, OID_AUTO, route_debug, CTLFLAG_RW, 131 &route_debug, 0, ""); 132 #endif 133 134 int route_assert_owner_access = 0; 135 SYSCTL_INT(_net_route, OID_AUTO, assert_owner_access, CTLFLAG_RW, 136 &route_assert_owner_access, 0, ""); 137 138 /* 139 * Initialize the route table(s) for protocol domains and 140 * create a helper thread which will be responsible for updating 141 * route table entries on each cpu. 142 */ 143 void 144 route_init(void) 145 { 146 int cpu; 147 thread_t rtd; 148 149 for (cpu = 0; cpu < ncpus; ++cpu) 150 bzero(&rtstatistics_percpu[cpu], sizeof(struct rtstatistics)); 151 rn_init(); /* initialize all zeroes, all ones, mask table */ 152 rtable_init(); /* call dom_rtattach() on each cpu */ 153 154 for (cpu = 0; cpu < ncpus; cpu++) { 155 lwkt_create(rtable_service_loop, NULL, &rtd, NULL, 156 0, cpu, "rtable_cpu %d", cpu); 157 rt_ports[cpu] = &rtd->td_msgport; 158 } 159 } 160 161 static void 162 rtable_init_oncpu(netmsg_t msg) 163 { 164 struct domain *dom; 165 int cpu = mycpuid; 166 167 SLIST_FOREACH(dom, &domains, dom_next) { 168 if (dom->dom_rtattach) { 169 dom->dom_rtattach( 170 (void **)&rt_tables[cpu][dom->dom_family], 171 dom->dom_rtoffset); 172 } 173 } 174 ifnet_forwardmsg(&msg->lmsg, cpu + 1); 175 } 176 177 static void 178 rtable_init(void) 179 { 180 struct netmsg_base msg; 181 182 netmsg_init(&msg, NULL, &curthread->td_msgport, 0, rtable_init_oncpu); 183 ifnet_domsg(&msg.lmsg, 0); 184 } 185 186 /* 187 * Our per-cpu table management protocol thread. All route table operations 188 * are sequentially chained through all cpus starting at cpu #0 in order to 189 * maintain duplicate route tables on each cpu. Having a spearate route 190 * table management thread allows the protocol and interrupt threads to 191 * issue route table changes. 192 */ 193 static void 194 rtable_service_loop(void *dummy __unused) 195 { 196 netmsg_base_t msg; 197 thread_t td = curthread; 198 199 get_mplock(); /* XXX is this mpsafe yet? */ 200 201 while ((msg = lwkt_waitport(&td->td_msgport, 0)) != NULL) { 202 msg->nm_dispatch((netmsg_t)msg); 203 } 204 } 205 206 /* 207 * Routing statistics. 208 */ 209 #ifdef SMP 210 static int 211 sysctl_rtstatistics(SYSCTL_HANDLER_ARGS) 212 { 213 int cpu, error = 0; 214 215 for (cpu = 0; cpu < ncpus; ++cpu) { 216 if ((error = SYSCTL_OUT(req, &rtstatistics_percpu[cpu], 217 sizeof(struct rtstatistics)))) 218 break; 219 if ((error = SYSCTL_IN(req, &rtstatistics_percpu[cpu], 220 sizeof(struct rtstatistics)))) 221 break; 222 } 223 224 return (error); 225 } 226 SYSCTL_PROC(_net_route, OID_AUTO, stats, (CTLTYPE_OPAQUE|CTLFLAG_RW), 227 0, 0, sysctl_rtstatistics, "S,rtstatistics", "Routing statistics"); 228 #else 229 SYSCTL_STRUCT(_net_route, OID_AUTO, stats, CTLFLAG_RW, &rtstat, rtstatistics, 230 "Routing statistics"); 231 #endif 232 233 /* 234 * Packet routing routines. 235 */ 236 237 /* 238 * Look up and fill in the "ro_rt" rtentry field in a route structure given 239 * an address in the "ro_dst" field. Always send a report on a miss and 240 * always clone routes. 241 */ 242 void 243 rtalloc(struct route *ro) 244 { 245 rtalloc_ign(ro, 0UL); 246 } 247 248 /* 249 * Look up and fill in the "ro_rt" rtentry field in a route structure given 250 * an address in the "ro_dst" field. Always send a report on a miss and 251 * optionally clone routes when RTF_CLONING or RTF_PRCLONING are not being 252 * ignored. 253 */ 254 void 255 rtalloc_ign(struct route *ro, u_long ignoreflags) 256 { 257 if (ro->ro_rt != NULL) { 258 if (ro->ro_rt->rt_ifp != NULL && ro->ro_rt->rt_flags & RTF_UP) 259 return; 260 rtfree(ro->ro_rt); 261 ro->ro_rt = NULL; 262 } 263 ro->ro_rt = _rtlookup(&ro->ro_dst, RTL_REPORTMSG, ignoreflags); 264 } 265 266 /* 267 * Look up the route that matches the given "dst" address. 268 * 269 * Route lookup can have the side-effect of creating and returning 270 * a cloned route instead when "dst" matches a cloning route and the 271 * RTF_CLONING and RTF_PRCLONING flags are not being ignored. 272 * 273 * Any route returned has its reference count incremented. 274 */ 275 struct rtentry * 276 _rtlookup(struct sockaddr *dst, boolean_t generate_report, u_long ignore) 277 { 278 struct radix_node_head *rnh = rt_tables[mycpuid][dst->sa_family]; 279 struct rtentry *rt; 280 281 if (rnh == NULL) 282 goto unreach; 283 284 /* 285 * Look up route in the radix tree. 286 */ 287 rt = (struct rtentry *) rnh->rnh_matchaddr((char *)dst, rnh); 288 if (rt == NULL) 289 goto unreach; 290 291 /* 292 * Handle cloning routes. 293 */ 294 if ((rt->rt_flags & ~ignore & (RTF_CLONING | RTF_PRCLONING)) != 0) { 295 struct rtentry *clonedroute; 296 int error; 297 298 clonedroute = rt; /* copy in/copy out parameter */ 299 error = rtrequest(RTM_RESOLVE, dst, NULL, NULL, 0, 300 &clonedroute); /* clone the route */ 301 if (error != 0) { /* cloning failed */ 302 if (generate_report) 303 rt_dstmsg(RTM_MISS, dst, error); 304 rt->rt_refcnt++; 305 return (rt); /* return the uncloned route */ 306 } 307 if (generate_report) { 308 if (clonedroute->rt_flags & RTF_XRESOLVE) 309 rt_dstmsg(RTM_RESOLVE, dst, 0); 310 else 311 rt_rtmsg(RTM_ADD, clonedroute, 312 clonedroute->rt_ifp, 0); 313 } 314 return (clonedroute); /* return cloned route */ 315 } 316 317 /* 318 * Increment the reference count of the matched route and return. 319 */ 320 rt->rt_refcnt++; 321 return (rt); 322 323 unreach: 324 rtstat.rts_unreach++; 325 if (generate_report) 326 rt_dstmsg(RTM_MISS, dst, 0); 327 return (NULL); 328 } 329 330 void 331 rtfree(struct rtentry *rt) 332 { 333 if (rt->rt_cpuid == mycpuid) 334 rtfree_oncpu(rt); 335 else 336 rtfree_remote(rt, 1); 337 } 338 339 void 340 rtfree_oncpu(struct rtentry *rt) 341 { 342 KKASSERT(rt->rt_cpuid == mycpuid); 343 KASSERT(rt->rt_refcnt > 0, ("rtfree: rt_refcnt %ld", rt->rt_refcnt)); 344 345 --rt->rt_refcnt; 346 if (rt->rt_refcnt == 0) { 347 struct radix_node_head *rnh = 348 rt_tables[mycpuid][rt_key(rt)->sa_family]; 349 350 if (rnh->rnh_close) 351 rnh->rnh_close((struct radix_node *)rt, rnh); 352 if (!(rt->rt_flags & RTF_UP)) { 353 /* deallocate route */ 354 if (rt->rt_ifa != NULL) 355 IFAFREE(rt->rt_ifa); 356 if (rt->rt_parent != NULL) 357 RTFREE(rt->rt_parent); /* recursive call! */ 358 Free(rt_key(rt)); 359 Free(rt); 360 } 361 } 362 } 363 364 static void 365 rtfree_remote_dispatch(netmsg_t msg) 366 { 367 struct lwkt_msg *lmsg = &msg->lmsg; 368 struct rtentry *rt = lmsg->u.ms_resultp; 369 370 rtfree_oncpu(rt); 371 lwkt_replymsg(lmsg, 0); 372 } 373 374 void 375 rtfree_remote(struct rtentry *rt, int allow_panic) 376 { 377 struct netmsg_base msg; 378 struct lwkt_msg *lmsg; 379 380 KKASSERT(rt->rt_cpuid != mycpuid); 381 382 if (route_assert_owner_access && allow_panic) { 383 panic("rt remote free rt_cpuid %d, mycpuid %d\n", 384 rt->rt_cpuid, mycpuid); 385 } else { 386 kprintf("rt remote free rt_cpuid %d, mycpuid %d\n", 387 rt->rt_cpuid, mycpuid); 388 print_backtrace(-1); 389 } 390 391 netmsg_init(&msg, NULL, &curthread->td_msgport, 392 0, rtfree_remote_dispatch); 393 lmsg = &msg.lmsg; 394 lmsg->u.ms_resultp = rt; 395 396 lwkt_domsg(rtable_portfn(rt->rt_cpuid), lmsg, 0); 397 } 398 399 static int 400 rtredirect_oncpu(struct sockaddr *dst, struct sockaddr *gateway, 401 struct sockaddr *netmask, int flags, struct sockaddr *src) 402 { 403 struct rtentry *rt = NULL; 404 struct rt_addrinfo rtinfo; 405 struct ifaddr *ifa; 406 u_long *stat = NULL; 407 int error; 408 409 /* verify the gateway is directly reachable */ 410 if ((ifa = ifa_ifwithnet(gateway)) == NULL) { 411 error = ENETUNREACH; 412 goto out; 413 } 414 415 /* 416 * If the redirect isn't from our current router for this destination, 417 * it's either old or wrong. 418 */ 419 if (!(flags & RTF_DONE) && /* XXX JH */ 420 (rt = rtpurelookup(dst)) != NULL && 421 (!sa_equal(src, rt->rt_gateway) || rt->rt_ifa != ifa)) { 422 error = EINVAL; 423 goto done; 424 } 425 426 /* 427 * If it redirects us to ourselves, we have a routing loop, 428 * perhaps as a result of an interface going down recently. 429 */ 430 if (ifa_ifwithaddr(gateway)) { 431 error = EHOSTUNREACH; 432 goto done; 433 } 434 435 /* 436 * Create a new entry if the lookup failed or if we got back 437 * a wildcard entry for the default route. This is necessary 438 * for hosts which use routing redirects generated by smart 439 * gateways to dynamically build the routing tables. 440 */ 441 if (rt == NULL) 442 goto create; 443 if ((rt_mask(rt) != NULL && rt_mask(rt)->sa_len < 2)) { 444 rtfree(rt); 445 goto create; 446 } 447 448 /* Ignore redirects for directly connected hosts. */ 449 if (!(rt->rt_flags & RTF_GATEWAY)) { 450 error = EHOSTUNREACH; 451 goto done; 452 } 453 454 if (!(rt->rt_flags & RTF_HOST) && (flags & RTF_HOST)) { 455 /* 456 * Changing from a network route to a host route. 457 * Create a new host route rather than smashing the 458 * network route. 459 */ 460 create: 461 flags |= RTF_GATEWAY | RTF_DYNAMIC; 462 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 463 rtinfo.rti_info[RTAX_DST] = dst; 464 rtinfo.rti_info[RTAX_GATEWAY] = gateway; 465 rtinfo.rti_info[RTAX_NETMASK] = netmask; 466 rtinfo.rti_flags = flags; 467 rtinfo.rti_ifa = ifa; 468 rt = NULL; /* copy-in/copy-out parameter */ 469 error = rtrequest1(RTM_ADD, &rtinfo, &rt); 470 if (rt != NULL) 471 flags = rt->rt_flags; 472 stat = &rtstat.rts_dynamic; 473 } else { 474 /* 475 * Smash the current notion of the gateway to this destination. 476 * Should check about netmask!!! 477 */ 478 rt->rt_flags |= RTF_MODIFIED; 479 flags |= RTF_MODIFIED; 480 481 /* We only need to report rtmsg on CPU0 */ 482 rt_setgate(rt, rt_key(rt), gateway, 483 mycpuid == 0 ? RTL_REPORTMSG : RTL_DONTREPORT); 484 error = 0; 485 stat = &rtstat.rts_newgateway; 486 } 487 488 done: 489 if (rt != NULL) 490 rtfree(rt); 491 out: 492 if (error != 0) 493 rtstat.rts_badredirect++; 494 else if (stat != NULL) 495 (*stat)++; 496 497 return error; 498 } 499 500 #ifdef SMP 501 502 struct netmsg_rtredirect { 503 struct netmsg_base base; 504 struct sockaddr *dst; 505 struct sockaddr *gateway; 506 struct sockaddr *netmask; 507 int flags; 508 struct sockaddr *src; 509 }; 510 511 #endif 512 513 /* 514 * Force a routing table entry to the specified 515 * destination to go through the given gateway. 516 * Normally called as a result of a routing redirect 517 * message from the network layer. 518 * 519 * N.B.: must be called at splnet 520 */ 521 void 522 rtredirect(struct sockaddr *dst, struct sockaddr *gateway, 523 struct sockaddr *netmask, int flags, struct sockaddr *src) 524 { 525 struct rt_addrinfo rtinfo; 526 int error; 527 #ifdef SMP 528 struct netmsg_rtredirect msg; 529 530 netmsg_init(&msg.base, NULL, &curthread->td_msgport, 531 0, rtredirect_msghandler); 532 msg.dst = dst; 533 msg.gateway = gateway; 534 msg.netmask = netmask; 535 msg.flags = flags; 536 msg.src = src; 537 error = lwkt_domsg(rtable_portfn(0), &msg.base.lmsg, 0); 538 #else 539 error = rtredirect_oncpu(dst, gateway, netmask, flags, src); 540 #endif 541 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 542 rtinfo.rti_info[RTAX_DST] = dst; 543 rtinfo.rti_info[RTAX_GATEWAY] = gateway; 544 rtinfo.rti_info[RTAX_NETMASK] = netmask; 545 rtinfo.rti_info[RTAX_AUTHOR] = src; 546 rt_missmsg(RTM_REDIRECT, &rtinfo, flags, error); 547 } 548 549 #ifdef SMP 550 551 static void 552 rtredirect_msghandler(netmsg_t msg) 553 { 554 struct netmsg_rtredirect *rmsg = (void *)msg; 555 int nextcpu; 556 557 rtredirect_oncpu(rmsg->dst, rmsg->gateway, rmsg->netmask, 558 rmsg->flags, rmsg->src); 559 nextcpu = mycpuid + 1; 560 if (nextcpu < ncpus) 561 lwkt_forwardmsg(rtable_portfn(nextcpu), &msg->lmsg); 562 else 563 lwkt_replymsg(&msg->lmsg, 0); 564 } 565 566 #endif 567 568 /* 569 * Routing table ioctl interface. 570 */ 571 int 572 rtioctl(u_long req, caddr_t data, struct ucred *cred) 573 { 574 #ifdef INET 575 /* Multicast goop, grrr... */ 576 return mrt_ioctl ? mrt_ioctl(req, data) : EOPNOTSUPP; 577 #else 578 return ENXIO; 579 #endif 580 } 581 582 struct ifaddr * 583 ifa_ifwithroute(int flags, struct sockaddr *dst, struct sockaddr *gateway) 584 { 585 struct ifaddr *ifa; 586 587 if (!(flags & RTF_GATEWAY)) { 588 /* 589 * If we are adding a route to an interface, 590 * and the interface is a point-to-point link, 591 * we should search for the destination 592 * as our clue to the interface. Otherwise 593 * we can use the local address. 594 */ 595 ifa = NULL; 596 if (flags & RTF_HOST) { 597 ifa = ifa_ifwithdstaddr(dst); 598 } 599 if (ifa == NULL) 600 ifa = ifa_ifwithaddr(gateway); 601 } else { 602 /* 603 * If we are adding a route to a remote net 604 * or host, the gateway may still be on the 605 * other end of a pt to pt link. 606 */ 607 ifa = ifa_ifwithdstaddr(gateway); 608 } 609 if (ifa == NULL) 610 ifa = ifa_ifwithnet(gateway); 611 if (ifa == NULL) { 612 struct rtentry *rt; 613 614 rt = rtpurelookup(gateway); 615 if (rt == NULL) 616 return (NULL); 617 rt->rt_refcnt--; 618 if ((ifa = rt->rt_ifa) == NULL) 619 return (NULL); 620 } 621 if (ifa->ifa_addr->sa_family != dst->sa_family) { 622 struct ifaddr *oldifa = ifa; 623 624 ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp); 625 if (ifa == NULL) 626 ifa = oldifa; 627 } 628 return (ifa); 629 } 630 631 static int rt_fixdelete (struct radix_node *, void *); 632 static int rt_fixchange (struct radix_node *, void *); 633 634 struct rtfc_arg { 635 struct rtentry *rt0; 636 struct radix_node_head *rnh; 637 }; 638 639 /* 640 * Set rtinfo->rti_ifa and rtinfo->rti_ifp. 641 */ 642 int 643 rt_getifa(struct rt_addrinfo *rtinfo) 644 { 645 struct sockaddr *gateway = rtinfo->rti_info[RTAX_GATEWAY]; 646 struct sockaddr *dst = rtinfo->rti_info[RTAX_DST]; 647 struct sockaddr *ifaaddr = rtinfo->rti_info[RTAX_IFA]; 648 int flags = rtinfo->rti_flags; 649 650 /* 651 * ifp may be specified by sockaddr_dl 652 * when protocol address is ambiguous. 653 */ 654 if (rtinfo->rti_ifp == NULL) { 655 struct sockaddr *ifpaddr; 656 657 ifpaddr = rtinfo->rti_info[RTAX_IFP]; 658 if (ifpaddr != NULL && ifpaddr->sa_family == AF_LINK) { 659 struct ifaddr *ifa; 660 661 ifa = ifa_ifwithnet(ifpaddr); 662 if (ifa != NULL) 663 rtinfo->rti_ifp = ifa->ifa_ifp; 664 } 665 } 666 667 if (rtinfo->rti_ifa == NULL && ifaaddr != NULL) 668 rtinfo->rti_ifa = ifa_ifwithaddr(ifaaddr); 669 if (rtinfo->rti_ifa == NULL) { 670 struct sockaddr *sa; 671 672 sa = ifaaddr != NULL ? ifaaddr : 673 (gateway != NULL ? gateway : dst); 674 if (sa != NULL && rtinfo->rti_ifp != NULL) 675 rtinfo->rti_ifa = ifaof_ifpforaddr(sa, rtinfo->rti_ifp); 676 else if (dst != NULL && gateway != NULL) 677 rtinfo->rti_ifa = ifa_ifwithroute(flags, dst, gateway); 678 else if (sa != NULL) 679 rtinfo->rti_ifa = ifa_ifwithroute(flags, sa, sa); 680 } 681 if (rtinfo->rti_ifa == NULL) 682 return (ENETUNREACH); 683 684 if (rtinfo->rti_ifp == NULL) 685 rtinfo->rti_ifp = rtinfo->rti_ifa->ifa_ifp; 686 return (0); 687 } 688 689 /* 690 * Do appropriate manipulations of a routing tree given 691 * all the bits of info needed 692 */ 693 int 694 rtrequest( 695 int req, 696 struct sockaddr *dst, 697 struct sockaddr *gateway, 698 struct sockaddr *netmask, 699 int flags, 700 struct rtentry **ret_nrt) 701 { 702 struct rt_addrinfo rtinfo; 703 704 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 705 rtinfo.rti_info[RTAX_DST] = dst; 706 rtinfo.rti_info[RTAX_GATEWAY] = gateway; 707 rtinfo.rti_info[RTAX_NETMASK] = netmask; 708 rtinfo.rti_flags = flags; 709 return rtrequest1(req, &rtinfo, ret_nrt); 710 } 711 712 int 713 rtrequest_global( 714 int req, 715 struct sockaddr *dst, 716 struct sockaddr *gateway, 717 struct sockaddr *netmask, 718 int flags) 719 { 720 struct rt_addrinfo rtinfo; 721 722 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 723 rtinfo.rti_info[RTAX_DST] = dst; 724 rtinfo.rti_info[RTAX_GATEWAY] = gateway; 725 rtinfo.rti_info[RTAX_NETMASK] = netmask; 726 rtinfo.rti_flags = flags; 727 return rtrequest1_global(req, &rtinfo, NULL, NULL); 728 } 729 730 #ifdef SMP 731 732 struct netmsg_rtq { 733 struct netmsg_base base; 734 int req; 735 struct rt_addrinfo *rtinfo; 736 rtrequest1_callback_func_t callback; 737 void *arg; 738 }; 739 740 #endif 741 742 int 743 rtrequest1_global(int req, struct rt_addrinfo *rtinfo, 744 rtrequest1_callback_func_t callback, void *arg) 745 { 746 int error; 747 #ifdef SMP 748 struct netmsg_rtq msg; 749 750 netmsg_init(&msg.base, NULL, &curthread->td_msgport, 751 0, rtrequest1_msghandler); 752 msg.base.lmsg.ms_error = -1; 753 msg.req = req; 754 msg.rtinfo = rtinfo; 755 msg.callback = callback; 756 msg.arg = arg; 757 error = lwkt_domsg(rtable_portfn(0), &msg.base.lmsg, 0); 758 #else 759 struct rtentry *rt = NULL; 760 761 error = rtrequest1(req, rtinfo, &rt); 762 if (rt) 763 --rt->rt_refcnt; 764 if (callback) 765 callback(req, error, rtinfo, rt, arg); 766 #endif 767 return (error); 768 } 769 770 /* 771 * Handle a route table request on the current cpu. Since the route table's 772 * are supposed to be identical on each cpu, an error occuring later in the 773 * message chain is considered system-fatal. 774 */ 775 #ifdef SMP 776 777 static void 778 rtrequest1_msghandler(netmsg_t msg) 779 { 780 struct netmsg_rtq *rmsg = (void *)msg; 781 struct rt_addrinfo rtinfo; 782 struct rtentry *rt = NULL; 783 int nextcpu; 784 int error; 785 786 /* 787 * Copy the rtinfo. We need to make sure that the original 788 * rtinfo, which is setup by the caller, in the netmsg will 789 * _not_ be changed; else the next CPU on the netmsg forwarding 790 * path will see a different rtinfo than what this CPU has seen. 791 */ 792 rtinfo = *rmsg->rtinfo; 793 794 error = rtrequest1(rmsg->req, &rtinfo, &rt); 795 if (rt) 796 --rt->rt_refcnt; 797 if (rmsg->callback) 798 rmsg->callback(rmsg->req, error, &rtinfo, rt, rmsg->arg); 799 800 /* 801 * RTM_DELETE's are propogated even if an error occurs, since a 802 * cloned route might be undergoing deletion and cloned routes 803 * are not necessarily replicated. An overall error is returned 804 * only if no cpus have the route in question. 805 */ 806 if (rmsg->base.lmsg.ms_error < 0 || error == 0) 807 rmsg->base.lmsg.ms_error = error; 808 809 nextcpu = mycpuid + 1; 810 if (error && rmsg->req != RTM_DELETE) { 811 if (mycpuid != 0) { 812 panic("rtrequest1_msghandler: rtrequest table " 813 "error was not on cpu #0"); 814 } 815 lwkt_replymsg(&rmsg->base.lmsg, error); 816 } else if (nextcpu < ncpus) { 817 lwkt_forwardmsg(rtable_portfn(nextcpu), &rmsg->base.lmsg); 818 } else { 819 lwkt_replymsg(&rmsg->base.lmsg, rmsg->base.lmsg.ms_error); 820 } 821 } 822 823 #endif 824 825 int 826 rtrequest1(int req, struct rt_addrinfo *rtinfo, struct rtentry **ret_nrt) 827 { 828 struct sockaddr *dst = rtinfo->rti_info[RTAX_DST]; 829 struct rtentry *rt; 830 struct radix_node *rn; 831 struct radix_node_head *rnh; 832 struct ifaddr *ifa; 833 struct sockaddr *ndst; 834 boolean_t reportmsg; 835 int error = 0; 836 837 #define gotoerr(x) { error = x ; goto bad; } 838 839 #ifdef ROUTE_DEBUG 840 if (route_debug) 841 rt_addrinfo_print(req, rtinfo); 842 #endif 843 844 crit_enter(); 845 /* 846 * Find the correct routing tree to use for this Address Family 847 */ 848 if ((rnh = rt_tables[mycpuid][dst->sa_family]) == NULL) 849 gotoerr(EAFNOSUPPORT); 850 851 /* 852 * If we are adding a host route then we don't want to put 853 * a netmask in the tree, nor do we want to clone it. 854 */ 855 if (rtinfo->rti_flags & RTF_HOST) { 856 rtinfo->rti_info[RTAX_NETMASK] = NULL; 857 rtinfo->rti_flags &= ~(RTF_CLONING | RTF_PRCLONING); 858 } 859 860 switch (req) { 861 case RTM_DELETE: 862 /* Remove the item from the tree. */ 863 rn = rnh->rnh_deladdr((char *)rtinfo->rti_info[RTAX_DST], 864 (char *)rtinfo->rti_info[RTAX_NETMASK], 865 rnh); 866 if (rn == NULL) 867 gotoerr(ESRCH); 868 KASSERT(!(rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)), 869 ("rnh_deladdr returned flags 0x%x", rn->rn_flags)); 870 rt = (struct rtentry *)rn; 871 872 /* ref to prevent a deletion race */ 873 ++rt->rt_refcnt; 874 875 /* Free any routes cloned from this one. */ 876 if ((rt->rt_flags & (RTF_CLONING | RTF_PRCLONING)) && 877 rt_mask(rt) != NULL) { 878 rnh->rnh_walktree_from(rnh, (char *)rt_key(rt), 879 (char *)rt_mask(rt), 880 rt_fixdelete, rt); 881 } 882 883 if (rt->rt_gwroute != NULL) { 884 RTFREE(rt->rt_gwroute); 885 rt->rt_gwroute = NULL; 886 } 887 888 /* 889 * NB: RTF_UP must be set during the search above, 890 * because we might delete the last ref, causing 891 * rt to get freed prematurely. 892 */ 893 rt->rt_flags &= ~RTF_UP; 894 895 #ifdef ROUTE_DEBUG 896 if (route_debug) 897 rt_print(rtinfo, rt); 898 #endif 899 900 /* Give the protocol a chance to keep things in sync. */ 901 if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest) 902 ifa->ifa_rtrequest(RTM_DELETE, rt, rtinfo); 903 904 /* 905 * If the caller wants it, then it can have it, 906 * but it's up to it to free the rtentry as we won't be 907 * doing it. 908 */ 909 KASSERT(rt->rt_refcnt >= 0, 910 ("rtrequest1(DELETE): refcnt %ld", rt->rt_refcnt)); 911 if (ret_nrt != NULL) { 912 /* leave ref intact for return */ 913 *ret_nrt = rt; 914 } else { 915 /* deref / attempt to destroy */ 916 rtfree(rt); 917 } 918 break; 919 920 case RTM_RESOLVE: 921 if (ret_nrt == NULL || (rt = *ret_nrt) == NULL) 922 gotoerr(EINVAL); 923 ifa = rt->rt_ifa; 924 rtinfo->rti_flags = 925 rt->rt_flags & ~(RTF_CLONING | RTF_PRCLONING | RTF_STATIC); 926 rtinfo->rti_flags |= RTF_WASCLONED; 927 rtinfo->rti_info[RTAX_GATEWAY] = rt->rt_gateway; 928 if ((rtinfo->rti_info[RTAX_NETMASK] = rt->rt_genmask) == NULL) 929 rtinfo->rti_flags |= RTF_HOST; 930 rtinfo->rti_info[RTAX_MPLS1] = rt->rt_shim[0]; 931 rtinfo->rti_info[RTAX_MPLS2] = rt->rt_shim[1]; 932 rtinfo->rti_info[RTAX_MPLS3] = rt->rt_shim[2]; 933 goto makeroute; 934 935 case RTM_ADD: 936 KASSERT(!(rtinfo->rti_flags & RTF_GATEWAY) || 937 rtinfo->rti_info[RTAX_GATEWAY] != NULL, 938 ("rtrequest: GATEWAY but no gateway")); 939 940 if (rtinfo->rti_ifa == NULL && (error = rt_getifa(rtinfo))) 941 gotoerr(error); 942 ifa = rtinfo->rti_ifa; 943 makeroute: 944 R_Malloc(rt, struct rtentry *, sizeof(struct rtentry)); 945 if (rt == NULL) 946 gotoerr(ENOBUFS); 947 bzero(rt, sizeof(struct rtentry)); 948 rt->rt_flags = RTF_UP | rtinfo->rti_flags; 949 rt->rt_cpuid = mycpuid; 950 951 if (mycpuid != 0 && req == RTM_ADD) { 952 /* For RTM_ADD, we have already sent rtmsg on CPU0. */ 953 reportmsg = RTL_DONTREPORT; 954 } else { 955 /* 956 * For RTM_ADD, we only send rtmsg on CPU0. 957 * For RTM_RESOLVE, we always send rtmsg. XXX 958 */ 959 reportmsg = RTL_REPORTMSG; 960 } 961 error = rt_setgate(rt, dst, rtinfo->rti_info[RTAX_GATEWAY], 962 reportmsg); 963 if (error != 0) { 964 Free(rt); 965 gotoerr(error); 966 } 967 968 ndst = rt_key(rt); 969 if (rtinfo->rti_info[RTAX_NETMASK] != NULL) 970 rt_maskedcopy(dst, ndst, 971 rtinfo->rti_info[RTAX_NETMASK]); 972 else 973 bcopy(dst, ndst, dst->sa_len); 974 975 if (rtinfo->rti_info[RTAX_MPLS1] != NULL) 976 rt_setshims(rt, rtinfo->rti_info); 977 978 /* 979 * Note that we now have a reference to the ifa. 980 * This moved from below so that rnh->rnh_addaddr() can 981 * examine the ifa and ifa->ifa_ifp if it so desires. 982 */ 983 IFAREF(ifa); 984 rt->rt_ifa = ifa; 985 rt->rt_ifp = ifa->ifa_ifp; 986 /* XXX mtu manipulation will be done in rnh_addaddr -- itojun */ 987 988 rn = rnh->rnh_addaddr((char *)ndst, 989 (char *)rtinfo->rti_info[RTAX_NETMASK], 990 rnh, rt->rt_nodes); 991 if (rn == NULL) { 992 struct rtentry *oldrt; 993 994 /* 995 * We already have one of these in the tree. 996 * We do a special hack: if the old route was 997 * cloned, then we blow it away and try 998 * re-inserting the new one. 999 */ 1000 oldrt = rtpurelookup(ndst); 1001 if (oldrt != NULL) { 1002 --oldrt->rt_refcnt; 1003 if (oldrt->rt_flags & RTF_WASCLONED) { 1004 rtrequest(RTM_DELETE, rt_key(oldrt), 1005 oldrt->rt_gateway, 1006 rt_mask(oldrt), 1007 oldrt->rt_flags, NULL); 1008 rn = rnh->rnh_addaddr((char *)ndst, 1009 (char *) 1010 rtinfo->rti_info[RTAX_NETMASK], 1011 rnh, rt->rt_nodes); 1012 } 1013 } 1014 } 1015 1016 /* 1017 * If it still failed to go into the tree, 1018 * then un-make it (this should be a function). 1019 */ 1020 if (rn == NULL) { 1021 if (rt->rt_gwroute != NULL) 1022 rtfree(rt->rt_gwroute); 1023 IFAFREE(ifa); 1024 Free(rt_key(rt)); 1025 Free(rt); 1026 gotoerr(EEXIST); 1027 } 1028 1029 /* 1030 * If we got here from RESOLVE, then we are cloning 1031 * so clone the rest, and note that we 1032 * are a clone (and increment the parent's references) 1033 */ 1034 if (req == RTM_RESOLVE) { 1035 rt->rt_rmx = (*ret_nrt)->rt_rmx; /* copy metrics */ 1036 rt->rt_rmx.rmx_pksent = 0; /* reset packet counter */ 1037 if ((*ret_nrt)->rt_flags & 1038 (RTF_CLONING | RTF_PRCLONING)) { 1039 rt->rt_parent = *ret_nrt; 1040 (*ret_nrt)->rt_refcnt++; 1041 } 1042 } 1043 1044 /* 1045 * if this protocol has something to add to this then 1046 * allow it to do that as well. 1047 */ 1048 if (ifa->ifa_rtrequest != NULL) 1049 ifa->ifa_rtrequest(req, rt, rtinfo); 1050 1051 /* 1052 * We repeat the same procedure from rt_setgate() here because 1053 * it doesn't fire when we call it there because the node 1054 * hasn't been added to the tree yet. 1055 */ 1056 if (req == RTM_ADD && !(rt->rt_flags & RTF_HOST) && 1057 rt_mask(rt) != NULL) { 1058 struct rtfc_arg arg = { rt, rnh }; 1059 1060 rnh->rnh_walktree_from(rnh, (char *)rt_key(rt), 1061 (char *)rt_mask(rt), 1062 rt_fixchange, &arg); 1063 } 1064 1065 #ifdef ROUTE_DEBUG 1066 if (route_debug) 1067 rt_print(rtinfo, rt); 1068 #endif 1069 /* 1070 * Return the resulting rtentry, 1071 * increasing the number of references by one. 1072 */ 1073 if (ret_nrt != NULL) { 1074 rt->rt_refcnt++; 1075 *ret_nrt = rt; 1076 } 1077 break; 1078 default: 1079 error = EOPNOTSUPP; 1080 } 1081 bad: 1082 #ifdef ROUTE_DEBUG 1083 if (route_debug) { 1084 if (error) 1085 kprintf("rti %p failed error %d\n", rtinfo, error); 1086 else 1087 kprintf("rti %p succeeded\n", rtinfo); 1088 } 1089 #endif 1090 crit_exit(); 1091 return (error); 1092 } 1093 1094 /* 1095 * Called from rtrequest(RTM_DELETE, ...) to fix up the route's ``family'' 1096 * (i.e., the routes related to it by the operation of cloning). This 1097 * routine is iterated over all potential former-child-routes by way of 1098 * rnh->rnh_walktree_from() above, and those that actually are children of 1099 * the late parent (passed in as VP here) are themselves deleted. 1100 */ 1101 static int 1102 rt_fixdelete(struct radix_node *rn, void *vp) 1103 { 1104 struct rtentry *rt = (struct rtentry *)rn; 1105 struct rtentry *rt0 = vp; 1106 1107 if (rt->rt_parent == rt0 && 1108 !(rt->rt_flags & (RTF_PINNED | RTF_CLONING | RTF_PRCLONING))) { 1109 return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 1110 rt->rt_flags, NULL); 1111 } 1112 return 0; 1113 } 1114 1115 /* 1116 * This routine is called from rt_setgate() to do the analogous thing for 1117 * adds and changes. There is the added complication in this case of a 1118 * middle insert; i.e., insertion of a new network route between an older 1119 * network route and (cloned) host routes. For this reason, a simple check 1120 * of rt->rt_parent is insufficient; each candidate route must be tested 1121 * against the (mask, value) of the new route (passed as before in vp) 1122 * to see if the new route matches it. 1123 * 1124 * XXX - it may be possible to do fixdelete() for changes and reserve this 1125 * routine just for adds. I'm not sure why I thought it was necessary to do 1126 * changes this way. 1127 */ 1128 #ifdef DEBUG 1129 static int rtfcdebug = 0; 1130 #endif 1131 1132 static int 1133 rt_fixchange(struct radix_node *rn, void *vp) 1134 { 1135 struct rtentry *rt = (struct rtentry *)rn; 1136 struct rtfc_arg *ap = vp; 1137 struct rtentry *rt0 = ap->rt0; 1138 struct radix_node_head *rnh = ap->rnh; 1139 u_char *xk1, *xm1, *xk2, *xmp; 1140 int i, len, mlen; 1141 1142 #ifdef DEBUG 1143 if (rtfcdebug) 1144 kprintf("rt_fixchange: rt %p, rt0 %p\n", rt, rt0); 1145 #endif 1146 1147 if (rt->rt_parent == NULL || 1148 (rt->rt_flags & (RTF_PINNED | RTF_CLONING | RTF_PRCLONING))) { 1149 #ifdef DEBUG 1150 if (rtfcdebug) kprintf("no parent, pinned or cloning\n"); 1151 #endif 1152 return 0; 1153 } 1154 1155 if (rt->rt_parent == rt0) { 1156 #ifdef DEBUG 1157 if (rtfcdebug) kprintf("parent match\n"); 1158 #endif 1159 return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 1160 rt->rt_flags, NULL); 1161 } 1162 1163 /* 1164 * There probably is a function somewhere which does this... 1165 * if not, there should be. 1166 */ 1167 len = imin(rt_key(rt0)->sa_len, rt_key(rt)->sa_len); 1168 1169 xk1 = (u_char *)rt_key(rt0); 1170 xm1 = (u_char *)rt_mask(rt0); 1171 xk2 = (u_char *)rt_key(rt); 1172 1173 /* avoid applying a less specific route */ 1174 xmp = (u_char *)rt_mask(rt->rt_parent); 1175 mlen = rt_key(rt->rt_parent)->sa_len; 1176 if (mlen > rt_key(rt0)->sa_len) { 1177 #ifdef DEBUG 1178 if (rtfcdebug) 1179 kprintf("rt_fixchange: inserting a less " 1180 "specific route\n"); 1181 #endif 1182 return 0; 1183 } 1184 for (i = rnh->rnh_treetop->rn_offset; i < mlen; i++) { 1185 if ((xmp[i] & ~(xmp[i] ^ xm1[i])) != xmp[i]) { 1186 #ifdef DEBUG 1187 if (rtfcdebug) 1188 kprintf("rt_fixchange: inserting a less " 1189 "specific route\n"); 1190 #endif 1191 return 0; 1192 } 1193 } 1194 1195 for (i = rnh->rnh_treetop->rn_offset; i < len; i++) { 1196 if ((xk2[i] & xm1[i]) != xk1[i]) { 1197 #ifdef DEBUG 1198 if (rtfcdebug) kprintf("no match\n"); 1199 #endif 1200 return 0; 1201 } 1202 } 1203 1204 /* 1205 * OK, this node is a clone, and matches the node currently being 1206 * changed/added under the node's mask. So, get rid of it. 1207 */ 1208 #ifdef DEBUG 1209 if (rtfcdebug) kprintf("deleting\n"); 1210 #endif 1211 return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 1212 rt->rt_flags, NULL); 1213 } 1214 1215 #define ROUNDUP(a) (a>0 ? (1 + (((a) - 1) | (sizeof(long) - 1))) : sizeof(long)) 1216 1217 int 1218 rt_setgate(struct rtentry *rt0, struct sockaddr *dst, struct sockaddr *gate, 1219 boolean_t generate_report) 1220 { 1221 char *space, *oldspace; 1222 int dlen = ROUNDUP(dst->sa_len), glen = ROUNDUP(gate->sa_len); 1223 struct rtentry *rt = rt0; 1224 struct radix_node_head *rnh = rt_tables[mycpuid][dst->sa_family]; 1225 1226 /* 1227 * A host route with the destination equal to the gateway 1228 * will interfere with keeping LLINFO in the routing 1229 * table, so disallow it. 1230 */ 1231 if (((rt0->rt_flags & (RTF_HOST | RTF_GATEWAY | RTF_LLINFO)) == 1232 (RTF_HOST | RTF_GATEWAY)) && 1233 dst->sa_len == gate->sa_len && 1234 sa_equal(dst, gate)) { 1235 /* 1236 * The route might already exist if this is an RTM_CHANGE 1237 * or a routing redirect, so try to delete it. 1238 */ 1239 if (rt_key(rt0) != NULL) 1240 rtrequest(RTM_DELETE, rt_key(rt0), rt0->rt_gateway, 1241 rt_mask(rt0), rt0->rt_flags, NULL); 1242 return EADDRNOTAVAIL; 1243 } 1244 1245 /* 1246 * Both dst and gateway are stored in the same malloc'ed chunk 1247 * (If I ever get my hands on....) 1248 * if we need to malloc a new chunk, then keep the old one around 1249 * till we don't need it any more. 1250 */ 1251 if (rt->rt_gateway == NULL || glen > ROUNDUP(rt->rt_gateway->sa_len)) { 1252 oldspace = (char *)rt_key(rt); 1253 R_Malloc(space, char *, dlen + glen); 1254 if (space == NULL) 1255 return ENOBUFS; 1256 rt->rt_nodes->rn_key = space; 1257 } else { 1258 space = (char *)rt_key(rt); /* Just use the old space. */ 1259 oldspace = NULL; 1260 } 1261 1262 /* Set the gateway value. */ 1263 rt->rt_gateway = (struct sockaddr *)(space + dlen); 1264 bcopy(gate, rt->rt_gateway, glen); 1265 1266 if (oldspace != NULL) { 1267 /* 1268 * If we allocated a new chunk, preserve the original dst. 1269 * This way, rt_setgate() really just sets the gate 1270 * and leaves the dst field alone. 1271 */ 1272 bcopy(dst, space, dlen); 1273 Free(oldspace); 1274 } 1275 1276 /* 1277 * If there is already a gwroute, it's now almost definitely wrong 1278 * so drop it. 1279 */ 1280 if (rt->rt_gwroute != NULL) { 1281 RTFREE(rt->rt_gwroute); 1282 rt->rt_gwroute = NULL; 1283 } 1284 if (rt->rt_flags & RTF_GATEWAY) { 1285 /* 1286 * Cloning loop avoidance: In the presence of 1287 * protocol-cloning and bad configuration, it is 1288 * possible to get stuck in bottomless mutual recursion 1289 * (rtrequest rt_setgate rtlookup). We avoid this 1290 * by not allowing protocol-cloning to operate for 1291 * gateways (which is probably the correct choice 1292 * anyway), and avoid the resulting reference loops 1293 * by disallowing any route to run through itself as 1294 * a gateway. This is obviously mandatory when we 1295 * get rt->rt_output(). 1296 * 1297 * This breaks TTCP for hosts outside the gateway! XXX JH 1298 */ 1299 rt->rt_gwroute = _rtlookup(gate, generate_report, 1300 RTF_PRCLONING); 1301 if (rt->rt_gwroute == rt) { 1302 rt->rt_gwroute = NULL; 1303 --rt->rt_refcnt; 1304 return EDQUOT; /* failure */ 1305 } 1306 } 1307 1308 /* 1309 * This isn't going to do anything useful for host routes, so 1310 * don't bother. Also make sure we have a reasonable mask 1311 * (we don't yet have one during adds). 1312 */ 1313 if (!(rt->rt_flags & RTF_HOST) && rt_mask(rt) != NULL) { 1314 struct rtfc_arg arg = { rt, rnh }; 1315 1316 rnh->rnh_walktree_from(rnh, (char *)rt_key(rt), 1317 (char *)rt_mask(rt), 1318 rt_fixchange, &arg); 1319 } 1320 1321 return 0; 1322 } 1323 1324 static void 1325 rt_maskedcopy( 1326 struct sockaddr *src, 1327 struct sockaddr *dst, 1328 struct sockaddr *netmask) 1329 { 1330 u_char *cp1 = (u_char *)src; 1331 u_char *cp2 = (u_char *)dst; 1332 u_char *cp3 = (u_char *)netmask; 1333 u_char *cplim = cp2 + *cp3; 1334 u_char *cplim2 = cp2 + *cp1; 1335 1336 *cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */ 1337 cp3 += 2; 1338 if (cplim > cplim2) 1339 cplim = cplim2; 1340 while (cp2 < cplim) 1341 *cp2++ = *cp1++ & *cp3++; 1342 if (cp2 < cplim2) 1343 bzero(cp2, cplim2 - cp2); 1344 } 1345 1346 int 1347 rt_llroute(struct sockaddr *dst, struct rtentry *rt0, struct rtentry **drt) 1348 { 1349 struct rtentry *up_rt, *rt; 1350 1351 if (!(rt0->rt_flags & RTF_UP)) { 1352 up_rt = rtlookup(dst); 1353 if (up_rt == NULL) 1354 return (EHOSTUNREACH); 1355 up_rt->rt_refcnt--; 1356 } else 1357 up_rt = rt0; 1358 if (up_rt->rt_flags & RTF_GATEWAY) { 1359 if (up_rt->rt_gwroute == NULL) { 1360 up_rt->rt_gwroute = rtlookup(up_rt->rt_gateway); 1361 if (up_rt->rt_gwroute == NULL) 1362 return (EHOSTUNREACH); 1363 } else if (!(up_rt->rt_gwroute->rt_flags & RTF_UP)) { 1364 rtfree(up_rt->rt_gwroute); 1365 up_rt->rt_gwroute = rtlookup(up_rt->rt_gateway); 1366 if (up_rt->rt_gwroute == NULL) 1367 return (EHOSTUNREACH); 1368 } 1369 rt = up_rt->rt_gwroute; 1370 } else 1371 rt = up_rt; 1372 if (rt->rt_flags & RTF_REJECT && 1373 (rt->rt_rmx.rmx_expire == 0 || /* rt doesn't expire */ 1374 time_second < rt->rt_rmx.rmx_expire)) /* rt not expired */ 1375 return (rt->rt_flags & RTF_HOST ? EHOSTDOWN : EHOSTUNREACH); 1376 *drt = rt; 1377 return 0; 1378 } 1379 1380 static int 1381 rt_setshims(struct rtentry *rt, struct sockaddr **rt_shim){ 1382 int i; 1383 1384 for (i=0; i<3; i++) { 1385 struct sockaddr *shim = rt_shim[RTAX_MPLS1 + i]; 1386 int shimlen; 1387 1388 if (shim == NULL) 1389 break; 1390 1391 shimlen = ROUNDUP(shim->sa_len); 1392 R_Malloc(rt->rt_shim[i], struct sockaddr *, shimlen); 1393 bcopy(shim, rt->rt_shim[i], shimlen); 1394 } 1395 1396 return 0; 1397 } 1398 1399 #ifdef ROUTE_DEBUG 1400 1401 /* 1402 * Print out a route table entry 1403 */ 1404 void 1405 rt_print(struct rt_addrinfo *rtinfo, struct rtentry *rn) 1406 { 1407 kprintf("rti %p cpu %d route %p flags %08lx: ", 1408 rtinfo, mycpuid, rn, rn->rt_flags); 1409 sockaddr_print(rt_key(rn)); 1410 kprintf(" mask "); 1411 sockaddr_print(rt_mask(rn)); 1412 kprintf(" gw "); 1413 sockaddr_print(rn->rt_gateway); 1414 kprintf(" ifc \"%s\"", rn->rt_ifp ? rn->rt_ifp->if_dname : "?"); 1415 kprintf(" ifa %p\n", rn->rt_ifa); 1416 } 1417 1418 void 1419 rt_addrinfo_print(int cmd, struct rt_addrinfo *rti) 1420 { 1421 int didit = 0; 1422 int i; 1423 1424 #ifdef ROUTE_DEBUG 1425 if (cmd == RTM_DELETE && route_debug > 1) 1426 print_backtrace(-1); 1427 #endif 1428 1429 switch(cmd) { 1430 case RTM_ADD: 1431 kprintf("ADD "); 1432 break; 1433 case RTM_RESOLVE: 1434 kprintf("RES "); 1435 break; 1436 case RTM_DELETE: 1437 kprintf("DEL "); 1438 break; 1439 default: 1440 kprintf("C%02d ", cmd); 1441 break; 1442 } 1443 kprintf("rti %p cpu %d ", rti, mycpuid); 1444 for (i = 0; i < rti->rti_addrs; ++i) { 1445 if (rti->rti_info[i] == NULL) 1446 continue; 1447 if (didit) 1448 kprintf(" ,"); 1449 switch(i) { 1450 case RTAX_DST: 1451 kprintf("(DST "); 1452 break; 1453 case RTAX_GATEWAY: 1454 kprintf("(GWY "); 1455 break; 1456 case RTAX_NETMASK: 1457 kprintf("(MSK "); 1458 break; 1459 case RTAX_GENMASK: 1460 kprintf("(GEN "); 1461 break; 1462 case RTAX_IFP: 1463 kprintf("(IFP "); 1464 break; 1465 case RTAX_IFA: 1466 kprintf("(IFA "); 1467 break; 1468 case RTAX_AUTHOR: 1469 kprintf("(AUT "); 1470 break; 1471 case RTAX_BRD: 1472 kprintf("(BRD "); 1473 break; 1474 default: 1475 kprintf("(?%02d ", i); 1476 break; 1477 } 1478 sockaddr_print(rti->rti_info[i]); 1479 kprintf(")"); 1480 didit = 1; 1481 } 1482 kprintf("\n"); 1483 } 1484 1485 void 1486 sockaddr_print(struct sockaddr *sa) 1487 { 1488 struct sockaddr_in *sa4; 1489 struct sockaddr_in6 *sa6; 1490 int len; 1491 int i; 1492 1493 if (sa == NULL) { 1494 kprintf("NULL"); 1495 return; 1496 } 1497 1498 len = sa->sa_len - offsetof(struct sockaddr, sa_data[0]); 1499 1500 switch(sa->sa_family) { 1501 case AF_INET: 1502 case AF_INET6: 1503 default: 1504 switch(sa->sa_family) { 1505 case AF_INET: 1506 sa4 = (struct sockaddr_in *)sa; 1507 kprintf("INET %d %d.%d.%d.%d", 1508 ntohs(sa4->sin_port), 1509 (ntohl(sa4->sin_addr.s_addr) >> 24) & 255, 1510 (ntohl(sa4->sin_addr.s_addr) >> 16) & 255, 1511 (ntohl(sa4->sin_addr.s_addr) >> 8) & 255, 1512 (ntohl(sa4->sin_addr.s_addr) >> 0) & 255 1513 ); 1514 break; 1515 case AF_INET6: 1516 sa6 = (struct sockaddr_in6 *)sa; 1517 kprintf("INET6 %d %04x:%04x%04x:%04x:%04x:%04x:%04x:%04x", 1518 ntohs(sa6->sin6_port), 1519 sa6->sin6_addr.s6_addr16[0], 1520 sa6->sin6_addr.s6_addr16[1], 1521 sa6->sin6_addr.s6_addr16[2], 1522 sa6->sin6_addr.s6_addr16[3], 1523 sa6->sin6_addr.s6_addr16[4], 1524 sa6->sin6_addr.s6_addr16[5], 1525 sa6->sin6_addr.s6_addr16[6], 1526 sa6->sin6_addr.s6_addr16[7] 1527 ); 1528 break; 1529 default: 1530 kprintf("AF%d ", sa->sa_family); 1531 while (len > 0 && sa->sa_data[len-1] == 0) 1532 --len; 1533 1534 for (i = 0; i < len; ++i) { 1535 if (i) 1536 kprintf("."); 1537 kprintf("%d", (unsigned char)sa->sa_data[i]); 1538 } 1539 break; 1540 } 1541 } 1542 } 1543 1544 #endif 1545 1546 /* 1547 * Set up a routing table entry, normally for an interface. 1548 */ 1549 int 1550 rtinit(struct ifaddr *ifa, int cmd, int flags) 1551 { 1552 struct sockaddr *dst, *deldst, *netmask; 1553 struct mbuf *m = NULL; 1554 struct radix_node_head *rnh; 1555 struct radix_node *rn; 1556 struct rt_addrinfo rtinfo; 1557 int error; 1558 1559 if (flags & RTF_HOST) { 1560 dst = ifa->ifa_dstaddr; 1561 netmask = NULL; 1562 } else { 1563 dst = ifa->ifa_addr; 1564 netmask = ifa->ifa_netmask; 1565 } 1566 /* 1567 * If it's a delete, check that if it exists, it's on the correct 1568 * interface or we might scrub a route to another ifa which would 1569 * be confusing at best and possibly worse. 1570 */ 1571 if (cmd == RTM_DELETE) { 1572 /* 1573 * It's a delete, so it should already exist.. 1574 * If it's a net, mask off the host bits 1575 * (Assuming we have a mask) 1576 */ 1577 if (netmask != NULL) { 1578 m = m_get(MB_DONTWAIT, MT_SONAME); 1579 if (m == NULL) 1580 return (ENOBUFS); 1581 mbuftrackid(m, 34); 1582 deldst = mtod(m, struct sockaddr *); 1583 rt_maskedcopy(dst, deldst, netmask); 1584 dst = deldst; 1585 } 1586 /* 1587 * Look up an rtentry that is in the routing tree and 1588 * contains the correct info. 1589 */ 1590 if ((rnh = rt_tables[mycpuid][dst->sa_family]) == NULL || 1591 (rn = rnh->rnh_lookup((char *)dst, 1592 (char *)netmask, rnh)) == NULL || 1593 ((struct rtentry *)rn)->rt_ifa != ifa || 1594 !sa_equal((struct sockaddr *)rn->rn_key, dst)) { 1595 if (m != NULL) 1596 m_free(m); 1597 return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH); 1598 } 1599 /* XXX */ 1600 #if 0 1601 else { 1602 /* 1603 * One would think that as we are deleting, and we know 1604 * it doesn't exist, we could just return at this point 1605 * with an "ELSE" clause, but apparently not.. 1606 */ 1607 return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH); 1608 } 1609 #endif 1610 } 1611 /* 1612 * Do the actual request 1613 */ 1614 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 1615 rtinfo.rti_info[RTAX_DST] = dst; 1616 rtinfo.rti_info[RTAX_GATEWAY] = ifa->ifa_addr; 1617 rtinfo.rti_info[RTAX_NETMASK] = netmask; 1618 rtinfo.rti_flags = flags | ifa->ifa_flags; 1619 rtinfo.rti_ifa = ifa; 1620 error = rtrequest1_global(cmd, &rtinfo, rtinit_rtrequest_callback, ifa); 1621 if (m != NULL) 1622 m_free(m); 1623 return (error); 1624 } 1625 1626 static void 1627 rtinit_rtrequest_callback(int cmd, int error, 1628 struct rt_addrinfo *rtinfo, struct rtentry *rt, 1629 void *arg) 1630 { 1631 struct ifaddr *ifa = arg; 1632 1633 if (error == 0 && rt) { 1634 if (mycpuid == 0) { 1635 ++rt->rt_refcnt; 1636 rt_newaddrmsg(cmd, ifa, error, rt); 1637 --rt->rt_refcnt; 1638 } 1639 if (cmd == RTM_DELETE) { 1640 if (rt->rt_refcnt == 0) { 1641 ++rt->rt_refcnt; 1642 rtfree(rt); 1643 } 1644 } 1645 } 1646 } 1647 1648 struct netmsg_rts { 1649 struct netmsg_base base; 1650 int req; 1651 struct rt_addrinfo *rtinfo; 1652 rtsearch_callback_func_t callback; 1653 void *arg; 1654 boolean_t exact_match; 1655 int found_cnt; 1656 }; 1657 1658 int 1659 rtsearch_global(int req, struct rt_addrinfo *rtinfo, 1660 rtsearch_callback_func_t callback, void *arg, 1661 boolean_t exact_match) 1662 { 1663 struct netmsg_rts msg; 1664 1665 netmsg_init(&msg.base, NULL, &curthread->td_msgport, 1666 0, rtsearch_msghandler); 1667 msg.req = req; 1668 msg.rtinfo = rtinfo; 1669 msg.callback = callback; 1670 msg.arg = arg; 1671 msg.exact_match = exact_match; 1672 msg.found_cnt = 0; 1673 return lwkt_domsg(rtable_portfn(0), &msg.base.lmsg, 0); 1674 } 1675 1676 static void 1677 rtsearch_msghandler(netmsg_t msg) 1678 { 1679 struct netmsg_rts *rmsg = (void *)msg; 1680 struct rt_addrinfo rtinfo; 1681 struct radix_node_head *rnh; 1682 struct rtentry *rt; 1683 int nextcpu, error; 1684 1685 /* 1686 * Copy the rtinfo. We need to make sure that the original 1687 * rtinfo, which is setup by the caller, in the netmsg will 1688 * _not_ be changed; else the next CPU on the netmsg forwarding 1689 * path will see a different rtinfo than what this CPU has seen. 1690 */ 1691 rtinfo = *rmsg->rtinfo; 1692 1693 /* 1694 * Find the correct routing tree to use for this Address Family 1695 */ 1696 if ((rnh = rt_tables[mycpuid][rtinfo.rti_dst->sa_family]) == NULL) { 1697 if (mycpuid != 0) 1698 panic("partially initialized routing tables\n"); 1699 lwkt_replymsg(&rmsg->base.lmsg, EAFNOSUPPORT); 1700 return; 1701 } 1702 1703 /* 1704 * Correct rtinfo for the host route searching. 1705 */ 1706 if (rtinfo.rti_flags & RTF_HOST) { 1707 rtinfo.rti_netmask = NULL; 1708 rtinfo.rti_flags &= ~(RTF_CLONING | RTF_PRCLONING); 1709 } 1710 1711 rt = (struct rtentry *) 1712 rnh->rnh_lookup((char *)rtinfo.rti_dst, 1713 (char *)rtinfo.rti_netmask, rnh); 1714 1715 /* 1716 * If we are asked to do the "exact match", we need to make sure 1717 * that host route searching got a host route while a network 1718 * route searching got a network route. 1719 */ 1720 if (rt != NULL && rmsg->exact_match && 1721 ((rt->rt_flags ^ rtinfo.rti_flags) & RTF_HOST)) 1722 rt = NULL; 1723 1724 if (rt == NULL) { 1725 /* 1726 * No matching routes have been found, don't count this 1727 * as a critical error (here, we set 'error' to 0), just 1728 * keep moving on, since at least prcloned routes are not 1729 * duplicated onto each CPU. 1730 */ 1731 error = 0; 1732 } else { 1733 rmsg->found_cnt++; 1734 1735 rt->rt_refcnt++; 1736 error = rmsg->callback(rmsg->req, &rtinfo, rt, rmsg->arg, 1737 rmsg->found_cnt); 1738 rt->rt_refcnt--; 1739 1740 if (error == EJUSTRETURN) { 1741 lwkt_replymsg(&rmsg->base.lmsg, 0); 1742 return; 1743 } 1744 } 1745 1746 nextcpu = mycpuid + 1; 1747 if (error) { 1748 KKASSERT(rmsg->found_cnt > 0); 1749 1750 /* 1751 * Under following cases, unrecoverable error has 1752 * not occured: 1753 * o Request is RTM_GET 1754 * o The first time that we find the route, but the 1755 * modification fails. 1756 */ 1757 if (rmsg->req != RTM_GET && rmsg->found_cnt > 1) { 1758 panic("rtsearch_msghandler: unrecoverable error " 1759 "cpu %d", mycpuid); 1760 } 1761 lwkt_replymsg(&rmsg->base.lmsg, error); 1762 } else if (nextcpu < ncpus) { 1763 lwkt_forwardmsg(rtable_portfn(nextcpu), &rmsg->base.lmsg); 1764 } else { 1765 if (rmsg->found_cnt == 0) { 1766 /* The requested route was never seen ... */ 1767 error = ESRCH; 1768 } 1769 lwkt_replymsg(&rmsg->base.lmsg, error); 1770 } 1771 } 1772 1773 int 1774 rtmask_add_global(struct sockaddr *mask) 1775 { 1776 struct netmsg_base msg; 1777 1778 netmsg_init(&msg, NULL, &curthread->td_msgport, 1779 0, rtmask_add_msghandler); 1780 msg.lmsg.u.ms_resultp = mask; 1781 1782 return lwkt_domsg(rtable_portfn(0), &msg.lmsg, 0); 1783 } 1784 1785 struct sockaddr * 1786 _rtmask_lookup(struct sockaddr *mask, boolean_t search) 1787 { 1788 struct radix_node *n; 1789 1790 #define clen(s) (*(u_char *)(s)) 1791 n = rn_addmask((char *)mask, search, 1); 1792 if (n != NULL && 1793 mask->sa_len >= clen(n->rn_key) && 1794 bcmp((char *)mask + 1, 1795 (char *)n->rn_key + 1, clen(n->rn_key) - 1) == 0) { 1796 return (struct sockaddr *)n->rn_key; 1797 } else { 1798 return NULL; 1799 } 1800 #undef clen 1801 } 1802 1803 static void 1804 rtmask_add_msghandler(netmsg_t msg) 1805 { 1806 struct lwkt_msg *lmsg = &msg->lmsg; 1807 struct sockaddr *mask = lmsg->u.ms_resultp; 1808 int error = 0, nextcpu; 1809 1810 if (rtmask_lookup(mask) == NULL) 1811 error = ENOBUFS; 1812 1813 nextcpu = mycpuid + 1; 1814 if (!error && nextcpu < ncpus) 1815 lwkt_forwardmsg(rtable_portfn(nextcpu), lmsg); 1816 else 1817 lwkt_replymsg(lmsg, error); 1818 } 1819 1820 /* This must be before ip6_init2(), which is now SI_ORDER_MIDDLE */ 1821 SYSINIT(route, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0); 1822