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] __cachealign; 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 u_long route_kmalloc_limit = 0; 122 TUNABLE_ULONG("net.route.kmalloc_limit", &route_kmalloc_limit); 123 124 /* 125 * Initialize the route table(s) for protocol domains and 126 * create a helper thread which will be responsible for updating 127 * route table entries on each cpu. 128 */ 129 void 130 route_init(void) 131 { 132 int cpu; 133 134 if (route_kmalloc_limit) 135 kmalloc_raise_limit(M_RTABLE, route_kmalloc_limit); 136 137 for (cpu = 0; cpu < netisr_ncpus; ++cpu) 138 bzero(&rtstatistics_percpu[cpu], sizeof(struct rtstatistics)); 139 rn_init(); /* initialize all zeroes, all ones, mask table */ 140 rtable_init(); /* call dom_rtattach() on each cpu */ 141 } 142 143 static void 144 rtable_init_oncpu(netmsg_t msg) 145 { 146 struct domain *dom; 147 int cpu = mycpuid; 148 149 ASSERT_NETISR_NCPUS(cpu); 150 151 SLIST_FOREACH(dom, &domains, dom_next) { 152 if (dom->dom_rtattach) { 153 dom->dom_rtattach( 154 (void **)&rt_tables[cpu][dom->dom_family], 155 dom->dom_rtoffset); 156 } 157 } 158 netisr_forwardmsg(&msg->base, cpu + 1); 159 } 160 161 static void 162 rtable_init(void) 163 { 164 struct netmsg_base msg; 165 166 netmsg_init(&msg, NULL, &curthread->td_msgport, 0, rtable_init_oncpu); 167 netisr_domsg_global(&msg); 168 } 169 170 /* 171 * Routing statistics. 172 */ 173 static int 174 sysctl_rtstatistics(SYSCTL_HANDLER_ARGS) 175 { 176 int cpu, error = 0; 177 178 for (cpu = 0; cpu < netisr_ncpus; ++cpu) { 179 if ((error = SYSCTL_OUT(req, &rtstatistics_percpu[cpu], 180 sizeof(struct rtstatistics)))) 181 break; 182 if ((error = SYSCTL_IN(req, &rtstatistics_percpu[cpu], 183 sizeof(struct rtstatistics)))) 184 break; 185 } 186 187 return (error); 188 } 189 SYSCTL_PROC(_net_route, OID_AUTO, stats, (CTLTYPE_OPAQUE|CTLFLAG_RW), 190 0, 0, sysctl_rtstatistics, "S,rtstatistics", "Routing statistics"); 191 192 /* 193 * Packet routing routines. 194 */ 195 196 /* 197 * Look up and fill in the "ro_rt" rtentry field in a route structure given 198 * an address in the "ro_dst" field. Always send a report on a miss and 199 * always clone routes. 200 */ 201 void 202 rtalloc(struct route *ro) 203 { 204 rtalloc_ign(ro, 0UL); 205 } 206 207 /* 208 * Look up and fill in the "ro_rt" rtentry field in a route structure given 209 * an address in the "ro_dst" field. Always send a report on a miss and 210 * optionally clone routes when RTF_CLONING or RTF_PRCLONING are not being 211 * ignored. 212 */ 213 void 214 rtalloc_ign(struct route *ro, u_long ignoreflags) 215 { 216 if (ro->ro_rt != NULL) { 217 if (ro->ro_rt->rt_ifp != NULL && ro->ro_rt->rt_flags & RTF_UP) 218 return; 219 rtfree(ro->ro_rt); 220 ro->ro_rt = NULL; 221 } 222 ro->ro_rt = _rtlookup(&ro->ro_dst, RTL_REPORTMSG, ignoreflags); 223 } 224 225 /* 226 * Look up the route that matches the given "dst" address. 227 * 228 * Route lookup can have the side-effect of creating and returning 229 * a cloned route instead when "dst" matches a cloning route and the 230 * RTF_CLONING and RTF_PRCLONING flags are not being ignored. 231 * 232 * Any route returned has its reference count incremented. 233 */ 234 struct rtentry * 235 _rtlookup(struct sockaddr *dst, boolean_t generate_report, u_long ignore) 236 { 237 struct radix_node_head *rnh = rt_tables[mycpuid][dst->sa_family]; 238 struct rtentry *rt; 239 240 ASSERT_NETISR_NCPUS(mycpuid); 241 242 if (rnh == NULL) 243 goto unreach; 244 245 /* 246 * Look up route in the radix tree. 247 */ 248 rt = (struct rtentry *) rnh->rnh_matchaddr((char *)dst, rnh); 249 if (rt == NULL) 250 goto unreach; 251 252 /* 253 * Handle cloning routes. 254 */ 255 if ((rt->rt_flags & ~ignore & (RTF_CLONING | RTF_PRCLONING)) != 0) { 256 struct rtentry *clonedroute; 257 int error; 258 259 clonedroute = rt; /* copy in/copy out parameter */ 260 error = rtrequest(RTM_RESOLVE, dst, NULL, NULL, 0, 261 &clonedroute); /* clone the route */ 262 if (error != 0) { /* cloning failed */ 263 if (generate_report) 264 rt_dstmsg(RTM_MISS, dst, error); 265 rt->rt_refcnt++; 266 return (rt); /* return the uncloned route */ 267 } 268 if (generate_report) { 269 if (clonedroute->rt_flags & RTF_XRESOLVE) 270 rt_dstmsg(RTM_RESOLVE, dst, 0); 271 else 272 rt_rtmsg(RTM_ADD, clonedroute, 273 clonedroute->rt_ifp, 0); 274 } 275 return (clonedroute); /* return cloned route */ 276 } 277 278 /* 279 * Increment the reference count of the matched route and return. 280 */ 281 rt->rt_refcnt++; 282 return (rt); 283 284 unreach: 285 rtstat.rts_unreach++; 286 if (generate_report) 287 rt_dstmsg(RTM_MISS, dst, 0); 288 return (NULL); 289 } 290 291 void 292 rtfree(struct rtentry *rt) 293 { 294 295 ASSERT_NETISR_NCPUS(rt->rt_cpuid); 296 KASSERT(rt->rt_refcnt > 0, ("rtfree: rt_refcnt %ld", rt->rt_refcnt)); 297 298 --rt->rt_refcnt; 299 if (rt->rt_refcnt == 0) { 300 struct radix_node_head *rnh = 301 rt_tables[mycpuid][rt_key(rt)->sa_family]; 302 303 if (rnh->rnh_close) 304 rnh->rnh_close((struct radix_node *)rt, rnh); 305 if (!(rt->rt_flags & RTF_UP)) { 306 /* deallocate route */ 307 if (rt->rt_ifa != NULL) 308 IFAFREE(rt->rt_ifa); 309 if (rt->rt_parent != NULL) 310 RTFREE(rt->rt_parent); /* recursive call! */ 311 Free(rt_key(rt)); 312 Free(rt); 313 } 314 } 315 } 316 317 static void 318 rtfree_async_dispatch(netmsg_t msg) 319 { 320 struct rtentry *rt = msg->lmsg.u.ms_resultp; 321 322 rtfree(rt); 323 netisr_replymsg(&msg->base, 0); 324 } 325 326 void 327 rtfree_async(struct rtentry *rt) 328 { 329 struct netmsg_base *msg; 330 331 if (IN_NETISR_NCPUS(rt->rt_cpuid)) { 332 rtfree(rt); 333 return; 334 } 335 336 KASSERT(rt->rt_refcnt > 0, 337 ("rtfree_async: rt_refcnt %ld", rt->rt_refcnt)); 338 339 msg = kmalloc(sizeof(*msg), M_LWKTMSG, M_INTWAIT); 340 netmsg_init(msg, NULL, &netisr_afree_rport, 0, rtfree_async_dispatch); 341 msg->lmsg.u.ms_resultp = rt; 342 343 netisr_sendmsg(msg, rt->rt_cpuid); 344 } 345 346 int 347 rtredirect_oncpu(struct sockaddr *dst, struct sockaddr *gateway, 348 struct sockaddr *netmask, int flags, struct sockaddr *src) 349 { 350 struct rtentry *rt = NULL; 351 struct rt_addrinfo rtinfo; 352 struct ifaddr *ifa; 353 u_long *stat = NULL; 354 int error; 355 356 ASSERT_NETISR_NCPUS(mycpuid); 357 358 /* verify the gateway is directly reachable */ 359 if ((ifa = ifa_ifwithnet(gateway)) == NULL) { 360 error = ENETUNREACH; 361 goto out; 362 } 363 364 /* 365 * If the redirect isn't from our current router for this destination, 366 * it's either old or wrong. 367 */ 368 if (!(flags & RTF_DONE) && /* XXX JH */ 369 (rt = rtpurelookup(dst)) != NULL && 370 (!sa_equal(src, rt->rt_gateway) || rt->rt_ifa != ifa)) { 371 error = EINVAL; 372 goto done; 373 } 374 375 /* 376 * If it redirects us to ourselves, we have a routing loop, 377 * perhaps as a result of an interface going down recently. 378 */ 379 if (ifa_ifwithaddr(gateway)) { 380 error = EHOSTUNREACH; 381 goto done; 382 } 383 384 /* 385 * Create a new entry if the lookup failed or if we got back 386 * a wildcard entry for the default route. This is necessary 387 * for hosts which use routing redirects generated by smart 388 * gateways to dynamically build the routing tables. 389 */ 390 if (rt == NULL) 391 goto create; 392 if ((rt_mask(rt) != NULL && rt_mask(rt)->sa_len < 2)) { 393 rtfree(rt); 394 goto create; 395 } 396 397 /* Ignore redirects for directly connected hosts. */ 398 if (!(rt->rt_flags & RTF_GATEWAY)) { 399 error = EHOSTUNREACH; 400 goto done; 401 } 402 403 if (!(rt->rt_flags & RTF_HOST) && (flags & RTF_HOST)) { 404 /* 405 * Changing from a network route to a host route. 406 * Create a new host route rather than smashing the 407 * network route. 408 */ 409 create: 410 flags |= RTF_GATEWAY | RTF_DYNAMIC; 411 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 412 rtinfo.rti_info[RTAX_DST] = dst; 413 rtinfo.rti_info[RTAX_GATEWAY] = gateway; 414 rtinfo.rti_info[RTAX_NETMASK] = netmask; 415 rtinfo.rti_flags = flags; 416 rtinfo.rti_ifa = ifa; 417 rt = NULL; /* copy-in/copy-out parameter */ 418 error = rtrequest1(RTM_ADD, &rtinfo, &rt); 419 if (rt != NULL) 420 flags = rt->rt_flags; 421 stat = &rtstat.rts_dynamic; 422 } else { 423 /* 424 * Smash the current notion of the gateway to this destination. 425 * Should check about netmask!!! 426 */ 427 rt->rt_flags |= RTF_MODIFIED; 428 flags |= RTF_MODIFIED; 429 430 /* We only need to report rtmsg on CPU0 */ 431 rt_setgate(rt, rt_key(rt), gateway, 432 mycpuid == 0 ? RTL_REPORTMSG : RTL_DONTREPORT); 433 error = 0; 434 stat = &rtstat.rts_newgateway; 435 } 436 437 done: 438 if (rt != NULL) 439 rtfree(rt); 440 out: 441 if (error != 0) 442 rtstat.rts_badredirect++; 443 else if (stat != NULL) 444 (*stat)++; 445 446 return error; 447 } 448 449 struct netmsg_rtredirect { 450 struct netmsg_base base; 451 struct sockaddr *dst; 452 struct sockaddr *gateway; 453 struct sockaddr *netmask; 454 int flags; 455 struct sockaddr *src; 456 }; 457 458 /* 459 * Force a routing table entry to the specified 460 * destination to go through the given gateway. 461 * Normally called as a result of a routing redirect 462 * message from the network layer. 463 */ 464 void 465 rtredirect(struct sockaddr *dst, struct sockaddr *gateway, 466 struct sockaddr *netmask, int flags, struct sockaddr *src) 467 { 468 struct rt_addrinfo rtinfo; 469 int error; 470 struct netmsg_rtredirect msg; 471 472 netmsg_init(&msg.base, NULL, &curthread->td_msgport, 473 0, rtredirect_msghandler); 474 msg.dst = dst; 475 msg.gateway = gateway; 476 msg.netmask = netmask; 477 msg.flags = flags; 478 msg.src = src; 479 error = netisr_domsg_global(&msg.base); 480 481 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 482 rtinfo.rti_info[RTAX_DST] = dst; 483 rtinfo.rti_info[RTAX_GATEWAY] = gateway; 484 rtinfo.rti_info[RTAX_NETMASK] = netmask; 485 rtinfo.rti_info[RTAX_AUTHOR] = src; 486 rt_missmsg(RTM_REDIRECT, &rtinfo, flags, error); 487 } 488 489 static void 490 rtredirect_msghandler(netmsg_t msg) 491 { 492 struct netmsg_rtredirect *rmsg = (void *)msg; 493 494 rtredirect_oncpu(rmsg->dst, rmsg->gateway, rmsg->netmask, 495 rmsg->flags, rmsg->src); 496 netisr_forwardmsg(&msg->base, mycpuid + 1); 497 } 498 499 /* 500 * Routing table ioctl interface. 501 */ 502 int 503 rtioctl(u_long req, caddr_t data, struct ucred *cred) 504 { 505 #ifdef INET 506 /* Multicast goop, grrr... */ 507 return mrt_ioctl ? mrt_ioctl(req, data) : EOPNOTSUPP; 508 #else 509 return ENXIO; 510 #endif 511 } 512 513 struct ifaddr * 514 ifa_ifwithroute(int flags, struct sockaddr *dst, struct sockaddr *gateway) 515 { 516 struct ifaddr *ifa; 517 518 if (!(flags & RTF_GATEWAY)) { 519 /* 520 * If we are adding a route to an interface, 521 * and the interface is a point-to-point link, 522 * we should search for the destination 523 * as our clue to the interface. Otherwise 524 * we can use the local address. 525 */ 526 ifa = NULL; 527 if (flags & RTF_HOST) { 528 ifa = ifa_ifwithdstaddr(dst); 529 } 530 if (ifa == NULL) 531 ifa = ifa_ifwithaddr(gateway); 532 } else { 533 /* 534 * If we are adding a route to a remote net 535 * or host, the gateway may still be on the 536 * other end of a pt to pt link. 537 */ 538 ifa = ifa_ifwithdstaddr(gateway); 539 } 540 if (ifa == NULL) 541 ifa = ifa_ifwithnet(gateway); 542 if (ifa == NULL) { 543 struct rtentry *rt; 544 545 rt = rtpurelookup(gateway); 546 if (rt == NULL) 547 return (NULL); 548 rt->rt_refcnt--; 549 if ((ifa = rt->rt_ifa) == NULL) 550 return (NULL); 551 } 552 if (ifa->ifa_addr->sa_family != dst->sa_family) { 553 struct ifaddr *oldifa = ifa; 554 555 ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp); 556 if (ifa == NULL) 557 ifa = oldifa; 558 } 559 return (ifa); 560 } 561 562 static int rt_fixdelete (struct radix_node *, void *); 563 static int rt_fixchange (struct radix_node *, void *); 564 565 struct rtfc_arg { 566 struct rtentry *rt0; 567 struct radix_node_head *rnh; 568 }; 569 570 /* 571 * Set rtinfo->rti_ifa and rtinfo->rti_ifp. 572 */ 573 int 574 rt_getifa(struct rt_addrinfo *rtinfo) 575 { 576 struct sockaddr *gateway = rtinfo->rti_info[RTAX_GATEWAY]; 577 struct sockaddr *dst = rtinfo->rti_info[RTAX_DST]; 578 struct sockaddr *ifaaddr = rtinfo->rti_info[RTAX_IFA]; 579 int flags = rtinfo->rti_flags; 580 581 /* 582 * ifp may be specified by sockaddr_dl 583 * when protocol address is ambiguous. 584 */ 585 if (rtinfo->rti_ifp == NULL) { 586 struct sockaddr *ifpaddr; 587 588 ifpaddr = rtinfo->rti_info[RTAX_IFP]; 589 if (ifpaddr != NULL && ifpaddr->sa_family == AF_LINK) { 590 struct ifaddr *ifa; 591 592 ifa = ifa_ifwithnet(ifpaddr); 593 if (ifa != NULL) 594 rtinfo->rti_ifp = ifa->ifa_ifp; 595 } 596 } 597 598 if (rtinfo->rti_ifa == NULL && ifaaddr != NULL) 599 rtinfo->rti_ifa = ifa_ifwithaddr(ifaaddr); 600 if (rtinfo->rti_ifa == NULL) { 601 struct sockaddr *sa; 602 603 sa = ifaaddr != NULL ? ifaaddr : 604 (gateway != NULL ? gateway : dst); 605 if (sa != NULL && rtinfo->rti_ifp != NULL) 606 rtinfo->rti_ifa = ifaof_ifpforaddr(sa, rtinfo->rti_ifp); 607 else if (dst != NULL && gateway != NULL) 608 rtinfo->rti_ifa = ifa_ifwithroute(flags, dst, gateway); 609 else if (sa != NULL) 610 rtinfo->rti_ifa = ifa_ifwithroute(flags, sa, sa); 611 } 612 if (rtinfo->rti_ifa == NULL) 613 return (ENETUNREACH); 614 615 if (rtinfo->rti_ifp == NULL) 616 rtinfo->rti_ifp = rtinfo->rti_ifa->ifa_ifp; 617 return (0); 618 } 619 620 /* 621 * Do appropriate manipulations of a routing tree given 622 * all the bits of info needed 623 */ 624 int 625 rtrequest( 626 int req, 627 struct sockaddr *dst, 628 struct sockaddr *gateway, 629 struct sockaddr *netmask, 630 int flags, 631 struct rtentry **ret_nrt) 632 { 633 struct rt_addrinfo rtinfo; 634 635 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 636 rtinfo.rti_info[RTAX_DST] = dst; 637 rtinfo.rti_info[RTAX_GATEWAY] = gateway; 638 rtinfo.rti_info[RTAX_NETMASK] = netmask; 639 rtinfo.rti_flags = flags; 640 return rtrequest1(req, &rtinfo, ret_nrt); 641 } 642 643 int 644 rtrequest_global( 645 int req, 646 struct sockaddr *dst, 647 struct sockaddr *gateway, 648 struct sockaddr *netmask, 649 int flags) 650 { 651 struct rt_addrinfo rtinfo; 652 653 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 654 rtinfo.rti_info[RTAX_DST] = dst; 655 rtinfo.rti_info[RTAX_GATEWAY] = gateway; 656 rtinfo.rti_info[RTAX_NETMASK] = netmask; 657 rtinfo.rti_flags = flags; 658 return rtrequest1_global(req, &rtinfo, NULL, NULL, RTREQ_PRIO_NORM); 659 } 660 661 struct netmsg_rtq { 662 struct netmsg_base base; 663 int req; 664 struct rt_addrinfo *rtinfo; 665 rtrequest1_callback_func_t callback; 666 void *arg; 667 }; 668 669 int 670 rtrequest1_global(int req, struct rt_addrinfo *rtinfo, 671 rtrequest1_callback_func_t callback, void *arg, boolean_t req_prio) 672 { 673 struct netmsg_rtq msg; 674 int flags = 0; 675 676 if (req_prio) 677 flags = MSGF_PRIORITY; 678 netmsg_init(&msg.base, NULL, &curthread->td_msgport, flags, 679 rtrequest1_msghandler); 680 msg.base.lmsg.ms_error = -1; 681 msg.req = req; 682 msg.rtinfo = rtinfo; 683 msg.callback = callback; 684 msg.arg = arg; 685 return (netisr_domsg_global(&msg.base)); 686 } 687 688 /* 689 * Handle a route table request on the current cpu. Since the route table's 690 * are supposed to be identical on each cpu, an error occuring later in the 691 * message chain is considered system-fatal. 692 */ 693 static void 694 rtrequest1_msghandler(netmsg_t msg) 695 { 696 struct netmsg_rtq *rmsg = (void *)msg; 697 struct rt_addrinfo rtinfo; 698 struct rtentry *rt = NULL; 699 int error; 700 701 /* 702 * Copy the rtinfo. We need to make sure that the original 703 * rtinfo, which is setup by the caller, in the netmsg will 704 * _not_ be changed; else the next CPU on the netmsg forwarding 705 * path will see a different rtinfo than what this CPU has seen. 706 */ 707 rtinfo = *rmsg->rtinfo; 708 709 error = rtrequest1(rmsg->req, &rtinfo, &rt); 710 if (rt) 711 --rt->rt_refcnt; 712 if (rmsg->callback) 713 rmsg->callback(rmsg->req, error, &rtinfo, rt, rmsg->arg); 714 715 /* 716 * RTM_DELETE's are propogated even if an error occurs, since a 717 * cloned route might be undergoing deletion and cloned routes 718 * are not necessarily replicated. An overall error is returned 719 * only if no cpus have the route in question. 720 */ 721 if (rmsg->base.lmsg.ms_error < 0 || error == 0) 722 rmsg->base.lmsg.ms_error = error; 723 724 if (error && rmsg->req != RTM_DELETE) { 725 if (mycpuid != 0) { 726 panic("rtrequest1_msghandler: rtrequest table req %d, " 727 "failed on cpu%d, error %d\n", 728 rmsg->req, mycpuid, error); 729 } 730 netisr_replymsg(&rmsg->base, error); 731 } else { 732 netisr_forwardmsg_error(&rmsg->base, mycpuid + 1, 733 rmsg->base.lmsg.ms_error); 734 } 735 } 736 737 int 738 rtrequest1(int req, struct rt_addrinfo *rtinfo, struct rtentry **ret_nrt) 739 { 740 struct sockaddr *dst = rtinfo->rti_info[RTAX_DST]; 741 struct rtentry *rt; 742 struct radix_node *rn; 743 struct radix_node_head *rnh; 744 struct ifaddr *ifa; 745 struct sockaddr *ndst; 746 boolean_t reportmsg; 747 int error = 0; 748 749 ASSERT_NETISR_NCPUS(mycpuid); 750 751 #define gotoerr(x) { error = x ; goto bad; } 752 753 #ifdef ROUTE_DEBUG 754 if (route_debug) 755 rt_addrinfo_print(req, rtinfo); 756 #endif 757 758 crit_enter(); 759 /* 760 * Find the correct routing tree to use for this Address Family 761 */ 762 if ((rnh = rt_tables[mycpuid][dst->sa_family]) == NULL) 763 gotoerr(EAFNOSUPPORT); 764 765 /* 766 * If we are adding a host route then we don't want to put 767 * a netmask in the tree, nor do we want to clone it. 768 */ 769 if (rtinfo->rti_flags & RTF_HOST) { 770 rtinfo->rti_info[RTAX_NETMASK] = NULL; 771 rtinfo->rti_flags &= ~(RTF_CLONING | RTF_PRCLONING); 772 } 773 774 switch (req) { 775 case RTM_DELETE: 776 /* Remove the item from the tree. */ 777 rn = rnh->rnh_deladdr((char *)rtinfo->rti_info[RTAX_DST], 778 (char *)rtinfo->rti_info[RTAX_NETMASK], 779 rnh); 780 if (rn == NULL) 781 gotoerr(ESRCH); 782 KASSERT(!(rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)), 783 ("rnh_deladdr returned flags 0x%x", rn->rn_flags)); 784 rt = (struct rtentry *)rn; 785 786 /* ref to prevent a deletion race */ 787 ++rt->rt_refcnt; 788 789 /* Free any routes cloned from this one. */ 790 if ((rt->rt_flags & (RTF_CLONING | RTF_PRCLONING)) && 791 rt_mask(rt) != NULL) { 792 rnh->rnh_walktree_from(rnh, (char *)rt_key(rt), 793 (char *)rt_mask(rt), 794 rt_fixdelete, rt); 795 } 796 797 if (rt->rt_gwroute != NULL) { 798 RTFREE(rt->rt_gwroute); 799 rt->rt_gwroute = NULL; 800 } 801 802 /* 803 * NB: RTF_UP must be set during the search above, 804 * because we might delete the last ref, causing 805 * rt to get freed prematurely. 806 */ 807 rt->rt_flags &= ~RTF_UP; 808 809 #ifdef ROUTE_DEBUG 810 if (route_debug) 811 rt_print(rtinfo, rt); 812 #endif 813 814 /* Give the protocol a chance to keep things in sync. */ 815 if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest) 816 ifa->ifa_rtrequest(RTM_DELETE, rt); 817 818 /* 819 * If the caller wants it, then it can have it, 820 * but it's up to it to free the rtentry as we won't be 821 * doing it. 822 */ 823 KASSERT(rt->rt_refcnt >= 0, 824 ("rtrequest1(DELETE): refcnt %ld", rt->rt_refcnt)); 825 if (ret_nrt != NULL) { 826 /* leave ref intact for return */ 827 *ret_nrt = rt; 828 } else { 829 /* deref / attempt to destroy */ 830 rtfree(rt); 831 } 832 break; 833 834 case RTM_RESOLVE: 835 if (ret_nrt == NULL || (rt = *ret_nrt) == NULL) 836 gotoerr(EINVAL); 837 838 KASSERT(rt->rt_cpuid == mycpuid, 839 ("rt resolve rt_cpuid %d, mycpuid %d", 840 rt->rt_cpuid, mycpuid)); 841 842 ifa = rt->rt_ifa; 843 rtinfo->rti_flags = 844 rt->rt_flags & ~(RTF_CLONING | RTF_PRCLONING | RTF_STATIC); 845 rtinfo->rti_flags |= RTF_WASCLONED; 846 rtinfo->rti_info[RTAX_GATEWAY] = rt->rt_gateway; 847 if ((rtinfo->rti_info[RTAX_NETMASK] = rt->rt_genmask) == NULL) 848 rtinfo->rti_flags |= RTF_HOST; 849 rtinfo->rti_info[RTAX_MPLS1] = rt->rt_shim[0]; 850 rtinfo->rti_info[RTAX_MPLS2] = rt->rt_shim[1]; 851 rtinfo->rti_info[RTAX_MPLS3] = rt->rt_shim[2]; 852 goto makeroute; 853 854 case RTM_ADD: 855 KASSERT(!(rtinfo->rti_flags & RTF_GATEWAY) || 856 rtinfo->rti_info[RTAX_GATEWAY] != NULL, 857 ("rtrequest: GATEWAY but no gateway")); 858 859 if (rtinfo->rti_ifa == NULL && (error = rt_getifa(rtinfo))) 860 gotoerr(error); 861 ifa = rtinfo->rti_ifa; 862 makeroute: 863 R_Malloc(rt, struct rtentry *, sizeof(struct rtentry)); 864 if (rt == NULL) { 865 if (req == RTM_ADD) { 866 kprintf("rtrequest1: alloc rtentry failed on " 867 "cpu%d\n", mycpuid); 868 } 869 gotoerr(ENOBUFS); 870 } 871 bzero(rt, sizeof(struct rtentry)); 872 rt->rt_flags = RTF_UP | rtinfo->rti_flags; 873 rt->rt_cpuid = mycpuid; 874 875 if (mycpuid != 0 && req == RTM_ADD) { 876 /* For RTM_ADD, we have already sent rtmsg on CPU0. */ 877 reportmsg = RTL_DONTREPORT; 878 } else { 879 /* 880 * For RTM_ADD, we only send rtmsg on CPU0. 881 * For RTM_RESOLVE, we always send rtmsg. XXX 882 */ 883 reportmsg = RTL_REPORTMSG; 884 } 885 error = rt_setgate(rt, dst, rtinfo->rti_info[RTAX_GATEWAY], 886 reportmsg); 887 if (error != 0) { 888 Free(rt); 889 gotoerr(error); 890 } 891 892 ndst = rt_key(rt); 893 if (rtinfo->rti_info[RTAX_NETMASK] != NULL) 894 rt_maskedcopy(dst, ndst, 895 rtinfo->rti_info[RTAX_NETMASK]); 896 else 897 bcopy(dst, ndst, dst->sa_len); 898 899 if (rtinfo->rti_info[RTAX_MPLS1] != NULL) 900 rt_setshims(rt, rtinfo->rti_info); 901 902 /* 903 * Note that we now have a reference to the ifa. 904 * This moved from below so that rnh->rnh_addaddr() can 905 * examine the ifa and ifa->ifa_ifp if it so desires. 906 */ 907 IFAREF(ifa); 908 rt->rt_ifa = ifa; 909 rt->rt_ifp = ifa->ifa_ifp; 910 /* XXX mtu manipulation will be done in rnh_addaddr -- itojun */ 911 912 rn = rnh->rnh_addaddr((char *)ndst, 913 (char *)rtinfo->rti_info[RTAX_NETMASK], 914 rnh, rt->rt_nodes); 915 if (rn == NULL) { 916 struct rtentry *oldrt; 917 918 /* 919 * We already have one of these in the tree. 920 * We do a special hack: if the old route was 921 * cloned, then we blow it away and try 922 * re-inserting the new one. 923 */ 924 oldrt = rtpurelookup(ndst); 925 if (oldrt != NULL) { 926 --oldrt->rt_refcnt; 927 if (oldrt->rt_flags & RTF_WASCLONED) { 928 rtrequest(RTM_DELETE, rt_key(oldrt), 929 oldrt->rt_gateway, 930 rt_mask(oldrt), 931 oldrt->rt_flags, NULL); 932 rn = rnh->rnh_addaddr((char *)ndst, 933 (char *) 934 rtinfo->rti_info[RTAX_NETMASK], 935 rnh, rt->rt_nodes); 936 } 937 } 938 } 939 /* NOTE: rt_ifa may have been changed */ 940 ifa = rt->rt_ifa; 941 942 /* 943 * If it still failed to go into the tree, 944 * then un-make it (this should be a function). 945 */ 946 if (rn == NULL) { 947 if (rt->rt_gwroute != NULL) 948 rtfree(rt->rt_gwroute); 949 IFAFREE(ifa); 950 Free(rt_key(rt)); 951 Free(rt); 952 gotoerr(EEXIST); 953 } 954 955 /* 956 * If we got here from RESOLVE, then we are cloning 957 * so clone the rest, and note that we 958 * are a clone (and increment the parent's references) 959 */ 960 if (req == RTM_RESOLVE) { 961 rt->rt_rmx = (*ret_nrt)->rt_rmx; /* copy metrics */ 962 rt->rt_rmx.rmx_pksent = 0; /* reset packet counter */ 963 if ((*ret_nrt)->rt_flags & 964 (RTF_CLONING | RTF_PRCLONING)) { 965 rt->rt_parent = *ret_nrt; 966 (*ret_nrt)->rt_refcnt++; 967 } 968 } 969 970 /* 971 * if this protocol has something to add to this then 972 * allow it to do that as well. 973 */ 974 if (ifa->ifa_rtrequest != NULL) 975 ifa->ifa_rtrequest(req, rt); 976 977 /* 978 * We repeat the same procedure from rt_setgate() here because 979 * it doesn't fire when we call it there because the node 980 * hasn't been added to the tree yet. 981 */ 982 if (req == RTM_ADD && !(rt->rt_flags & RTF_HOST) && 983 rt_mask(rt) != NULL) { 984 struct rtfc_arg arg = { rt, rnh }; 985 986 rnh->rnh_walktree_from(rnh, (char *)rt_key(rt), 987 (char *)rt_mask(rt), 988 rt_fixchange, &arg); 989 } 990 991 #ifdef ROUTE_DEBUG 992 if (route_debug) 993 rt_print(rtinfo, rt); 994 #endif 995 /* 996 * Return the resulting rtentry, 997 * increasing the number of references by one. 998 */ 999 if (ret_nrt != NULL) { 1000 rt->rt_refcnt++; 1001 *ret_nrt = rt; 1002 } 1003 break; 1004 default: 1005 error = EOPNOTSUPP; 1006 } 1007 bad: 1008 #ifdef ROUTE_DEBUG 1009 if (route_debug) { 1010 if (error) 1011 kprintf("rti %p failed error %d\n", rtinfo, error); 1012 else 1013 kprintf("rti %p succeeded\n", rtinfo); 1014 } 1015 #endif 1016 crit_exit(); 1017 return (error); 1018 } 1019 1020 /* 1021 * Called from rtrequest(RTM_DELETE, ...) to fix up the route's ``family'' 1022 * (i.e., the routes related to it by the operation of cloning). This 1023 * routine is iterated over all potential former-child-routes by way of 1024 * rnh->rnh_walktree_from() above, and those that actually are children of 1025 * the late parent (passed in as VP here) are themselves deleted. 1026 */ 1027 static int 1028 rt_fixdelete(struct radix_node *rn, void *vp) 1029 { 1030 struct rtentry *rt = (struct rtentry *)rn; 1031 struct rtentry *rt0 = vp; 1032 1033 if (rt->rt_parent == rt0 && 1034 !(rt->rt_flags & (RTF_PINNED | RTF_CLONING | RTF_PRCLONING))) { 1035 return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 1036 rt->rt_flags, NULL); 1037 } 1038 return 0; 1039 } 1040 1041 /* 1042 * This routine is called from rt_setgate() to do the analogous thing for 1043 * adds and changes. There is the added complication in this case of a 1044 * middle insert; i.e., insertion of a new network route between an older 1045 * network route and (cloned) host routes. For this reason, a simple check 1046 * of rt->rt_parent is insufficient; each candidate route must be tested 1047 * against the (mask, value) of the new route (passed as before in vp) 1048 * to see if the new route matches it. 1049 * 1050 * XXX - it may be possible to do fixdelete() for changes and reserve this 1051 * routine just for adds. I'm not sure why I thought it was necessary to do 1052 * changes this way. 1053 */ 1054 #ifdef DEBUG 1055 static int rtfcdebug = 0; 1056 #endif 1057 1058 static int 1059 rt_fixchange(struct radix_node *rn, void *vp) 1060 { 1061 struct rtentry *rt = (struct rtentry *)rn; 1062 struct rtfc_arg *ap = vp; 1063 struct rtentry *rt0 = ap->rt0; 1064 struct radix_node_head *rnh = ap->rnh; 1065 u_char *xk1, *xm1, *xk2, *xmp; 1066 int i, len, mlen; 1067 1068 #ifdef DEBUG 1069 if (rtfcdebug) 1070 kprintf("rt_fixchange: rt %p, rt0 %p\n", rt, rt0); 1071 #endif 1072 1073 if (rt->rt_parent == NULL || 1074 (rt->rt_flags & (RTF_PINNED | RTF_CLONING | RTF_PRCLONING))) { 1075 #ifdef DEBUG 1076 if (rtfcdebug) kprintf("no parent, pinned or cloning\n"); 1077 #endif 1078 return 0; 1079 } 1080 1081 if (rt->rt_parent == rt0) { 1082 #ifdef DEBUG 1083 if (rtfcdebug) kprintf("parent match\n"); 1084 #endif 1085 return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 1086 rt->rt_flags, NULL); 1087 } 1088 1089 /* 1090 * There probably is a function somewhere which does this... 1091 * if not, there should be. 1092 */ 1093 len = imin(rt_key(rt0)->sa_len, rt_key(rt)->sa_len); 1094 1095 xk1 = (u_char *)rt_key(rt0); 1096 xm1 = (u_char *)rt_mask(rt0); 1097 xk2 = (u_char *)rt_key(rt); 1098 1099 /* avoid applying a less specific route */ 1100 xmp = (u_char *)rt_mask(rt->rt_parent); 1101 mlen = rt_key(rt->rt_parent)->sa_len; 1102 if (mlen > rt_key(rt0)->sa_len) { 1103 #ifdef DEBUG 1104 if (rtfcdebug) 1105 kprintf("rt_fixchange: inserting a less " 1106 "specific route\n"); 1107 #endif 1108 return 0; 1109 } 1110 for (i = rnh->rnh_treetop->rn_offset; i < mlen; i++) { 1111 if ((xmp[i] & ~(xmp[i] ^ xm1[i])) != xmp[i]) { 1112 #ifdef DEBUG 1113 if (rtfcdebug) 1114 kprintf("rt_fixchange: inserting a less " 1115 "specific route\n"); 1116 #endif 1117 return 0; 1118 } 1119 } 1120 1121 for (i = rnh->rnh_treetop->rn_offset; i < len; i++) { 1122 if ((xk2[i] & xm1[i]) != xk1[i]) { 1123 #ifdef DEBUG 1124 if (rtfcdebug) kprintf("no match\n"); 1125 #endif 1126 return 0; 1127 } 1128 } 1129 1130 /* 1131 * OK, this node is a clone, and matches the node currently being 1132 * changed/added under the node's mask. So, get rid of it. 1133 */ 1134 #ifdef DEBUG 1135 if (rtfcdebug) kprintf("deleting\n"); 1136 #endif 1137 return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 1138 rt->rt_flags, NULL); 1139 } 1140 1141 int 1142 rt_setgate(struct rtentry *rt0, struct sockaddr *dst, struct sockaddr *gate, 1143 boolean_t generate_report) 1144 { 1145 char *space, *oldspace; 1146 int dlen = RT_ROUNDUP(dst->sa_len), glen = RT_ROUNDUP(gate->sa_len); 1147 struct rtentry *rt = rt0; 1148 struct radix_node_head *rnh = rt_tables[mycpuid][dst->sa_family]; 1149 1150 ASSERT_NETISR_NCPUS(mycpuid); 1151 1152 /* 1153 * A host route with the destination equal to the gateway 1154 * will interfere with keeping LLINFO in the routing 1155 * table, so disallow it. 1156 */ 1157 if (((rt0->rt_flags & (RTF_HOST | RTF_GATEWAY | RTF_LLINFO)) == 1158 (RTF_HOST | RTF_GATEWAY)) && 1159 dst->sa_len == gate->sa_len && 1160 sa_equal(dst, gate)) { 1161 /* 1162 * The route might already exist if this is an RTM_CHANGE 1163 * or a routing redirect, so try to delete it. 1164 */ 1165 if (rt_key(rt0) != NULL) 1166 rtrequest(RTM_DELETE, rt_key(rt0), rt0->rt_gateway, 1167 rt_mask(rt0), rt0->rt_flags, NULL); 1168 return EADDRNOTAVAIL; 1169 } 1170 1171 /* 1172 * Both dst and gateway are stored in the same malloc'ed chunk 1173 * (If I ever get my hands on....) 1174 * if we need to malloc a new chunk, then keep the old one around 1175 * till we don't need it any more. 1176 */ 1177 if (rt->rt_gateway == NULL || 1178 glen > RT_ROUNDUP(rt->rt_gateway->sa_len)) { 1179 oldspace = (char *)rt_key(rt); 1180 R_Malloc(space, char *, dlen + glen); 1181 if (space == NULL) 1182 return ENOBUFS; 1183 rt->rt_nodes->rn_key = space; 1184 } else { 1185 space = (char *)rt_key(rt); /* Just use the old space. */ 1186 oldspace = NULL; 1187 } 1188 1189 /* Set the gateway value. */ 1190 rt->rt_gateway = (struct sockaddr *)(space + dlen); 1191 bcopy(gate, rt->rt_gateway, glen); 1192 1193 if (oldspace != NULL) { 1194 /* 1195 * If we allocated a new chunk, preserve the original dst. 1196 * This way, rt_setgate() really just sets the gate 1197 * and leaves the dst field alone. 1198 */ 1199 bcopy(dst, space, dlen); 1200 Free(oldspace); 1201 } 1202 1203 /* 1204 * If there is already a gwroute, it's now almost definitely wrong 1205 * so drop it. 1206 */ 1207 if (rt->rt_gwroute != NULL) { 1208 RTFREE(rt->rt_gwroute); 1209 rt->rt_gwroute = NULL; 1210 } 1211 if (rt->rt_flags & RTF_GATEWAY) { 1212 /* 1213 * Cloning loop avoidance: In the presence of 1214 * protocol-cloning and bad configuration, it is 1215 * possible to get stuck in bottomless mutual recursion 1216 * (rtrequest rt_setgate rtlookup). We avoid this 1217 * by not allowing protocol-cloning to operate for 1218 * gateways (which is probably the correct choice 1219 * anyway), and avoid the resulting reference loops 1220 * by disallowing any route to run through itself as 1221 * a gateway. This is obviously mandatory when we 1222 * get rt->rt_output(). 1223 * 1224 * This breaks TTCP for hosts outside the gateway! XXX JH 1225 */ 1226 rt->rt_gwroute = _rtlookup(gate, generate_report, 1227 RTF_PRCLONING); 1228 if (rt->rt_gwroute == rt) { 1229 rt->rt_gwroute = NULL; 1230 --rt->rt_refcnt; 1231 return EDQUOT; /* failure */ 1232 } 1233 } 1234 1235 /* 1236 * This isn't going to do anything useful for host routes, so 1237 * don't bother. Also make sure we have a reasonable mask 1238 * (we don't yet have one during adds). 1239 */ 1240 if (!(rt->rt_flags & RTF_HOST) && rt_mask(rt) != NULL) { 1241 struct rtfc_arg arg = { rt, rnh }; 1242 1243 rnh->rnh_walktree_from(rnh, (char *)rt_key(rt), 1244 (char *)rt_mask(rt), 1245 rt_fixchange, &arg); 1246 } 1247 1248 return 0; 1249 } 1250 1251 static void 1252 rt_maskedcopy( 1253 struct sockaddr *src, 1254 struct sockaddr *dst, 1255 struct sockaddr *netmask) 1256 { 1257 u_char *cp1 = (u_char *)src; 1258 u_char *cp2 = (u_char *)dst; 1259 u_char *cp3 = (u_char *)netmask; 1260 u_char *cplim = cp2 + *cp3; 1261 u_char *cplim2 = cp2 + *cp1; 1262 1263 *cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */ 1264 cp3 += 2; 1265 if (cplim > cplim2) 1266 cplim = cplim2; 1267 while (cp2 < cplim) 1268 *cp2++ = *cp1++ & *cp3++; 1269 if (cp2 < cplim2) 1270 bzero(cp2, cplim2 - cp2); 1271 } 1272 1273 int 1274 rt_llroute(struct sockaddr *dst, struct rtentry *rt0, struct rtentry **drt) 1275 { 1276 struct rtentry *up_rt, *rt; 1277 1278 ASSERT_NETISR_NCPUS(mycpuid); 1279 1280 if (!(rt0->rt_flags & RTF_UP)) { 1281 up_rt = rtlookup(dst); 1282 if (up_rt == NULL) 1283 return (EHOSTUNREACH); 1284 up_rt->rt_refcnt--; 1285 } else 1286 up_rt = rt0; 1287 if (up_rt->rt_flags & RTF_GATEWAY) { 1288 if (up_rt->rt_gwroute == NULL) { 1289 up_rt->rt_gwroute = rtlookup(up_rt->rt_gateway); 1290 if (up_rt->rt_gwroute == NULL) 1291 return (EHOSTUNREACH); 1292 } else if (!(up_rt->rt_gwroute->rt_flags & RTF_UP)) { 1293 rtfree(up_rt->rt_gwroute); 1294 up_rt->rt_gwroute = rtlookup(up_rt->rt_gateway); 1295 if (up_rt->rt_gwroute == NULL) 1296 return (EHOSTUNREACH); 1297 } 1298 rt = up_rt->rt_gwroute; 1299 } else 1300 rt = up_rt; 1301 if (rt->rt_flags & RTF_REJECT && 1302 (rt->rt_rmx.rmx_expire == 0 || /* rt doesn't expire */ 1303 time_uptime < rt->rt_rmx.rmx_expire)) /* rt not expired */ 1304 return (rt->rt_flags & RTF_HOST ? EHOSTDOWN : EHOSTUNREACH); 1305 *drt = rt; 1306 return 0; 1307 } 1308 1309 static int 1310 rt_setshims(struct rtentry *rt, struct sockaddr **rt_shim){ 1311 int i; 1312 1313 for (i=0; i<3; i++) { 1314 struct sockaddr *shim = rt_shim[RTAX_MPLS1 + i]; 1315 int shimlen; 1316 1317 if (shim == NULL) 1318 break; 1319 1320 shimlen = RT_ROUNDUP(shim->sa_len); 1321 R_Malloc(rt->rt_shim[i], struct sockaddr *, shimlen); 1322 bcopy(shim, rt->rt_shim[i], shimlen); 1323 } 1324 1325 return 0; 1326 } 1327 1328 #ifdef ROUTE_DEBUG 1329 1330 /* 1331 * Print out a route table entry 1332 */ 1333 void 1334 rt_print(struct rt_addrinfo *rtinfo, struct rtentry *rn) 1335 { 1336 kprintf("rti %p cpu %d route %p flags %08lx: ", 1337 rtinfo, mycpuid, rn, rn->rt_flags); 1338 sockaddr_print(rt_key(rn)); 1339 kprintf(" mask "); 1340 sockaddr_print(rt_mask(rn)); 1341 kprintf(" gw "); 1342 sockaddr_print(rn->rt_gateway); 1343 kprintf(" ifc \"%s\"", rn->rt_ifp ? rn->rt_ifp->if_dname : "?"); 1344 kprintf(" ifa %p\n", rn->rt_ifa); 1345 } 1346 1347 void 1348 rt_addrinfo_print(int cmd, struct rt_addrinfo *rti) 1349 { 1350 int didit = 0; 1351 int i; 1352 1353 #ifdef ROUTE_DEBUG 1354 if (cmd == RTM_DELETE && route_debug > 1) 1355 print_backtrace(-1); 1356 #endif 1357 1358 switch(cmd) { 1359 case RTM_ADD: 1360 kprintf("ADD "); 1361 break; 1362 case RTM_RESOLVE: 1363 kprintf("RES "); 1364 break; 1365 case RTM_DELETE: 1366 kprintf("DEL "); 1367 break; 1368 default: 1369 kprintf("C%02d ", cmd); 1370 break; 1371 } 1372 kprintf("rti %p cpu %d ", rti, mycpuid); 1373 for (i = 0; i < rti->rti_addrs; ++i) { 1374 if (rti->rti_info[i] == NULL) 1375 continue; 1376 if (didit) 1377 kprintf(" ,"); 1378 switch(i) { 1379 case RTAX_DST: 1380 kprintf("(DST "); 1381 break; 1382 case RTAX_GATEWAY: 1383 kprintf("(GWY "); 1384 break; 1385 case RTAX_NETMASK: 1386 kprintf("(MSK "); 1387 break; 1388 case RTAX_GENMASK: 1389 kprintf("(GEN "); 1390 break; 1391 case RTAX_IFP: 1392 kprintf("(IFP "); 1393 break; 1394 case RTAX_IFA: 1395 kprintf("(IFA "); 1396 break; 1397 case RTAX_AUTHOR: 1398 kprintf("(AUT "); 1399 break; 1400 case RTAX_BRD: 1401 kprintf("(BRD "); 1402 break; 1403 default: 1404 kprintf("(?%02d ", i); 1405 break; 1406 } 1407 sockaddr_print(rti->rti_info[i]); 1408 kprintf(")"); 1409 didit = 1; 1410 } 1411 kprintf("\n"); 1412 } 1413 1414 void 1415 sockaddr_print(struct sockaddr *sa) 1416 { 1417 struct sockaddr_in *sa4; 1418 struct sockaddr_in6 *sa6; 1419 int len; 1420 int i; 1421 1422 if (sa == NULL) { 1423 kprintf("NULL"); 1424 return; 1425 } 1426 1427 len = sa->sa_len - offsetof(struct sockaddr, sa_data[0]); 1428 1429 switch(sa->sa_family) { 1430 case AF_INET: 1431 case AF_INET6: 1432 default: 1433 switch(sa->sa_family) { 1434 case AF_INET: 1435 sa4 = (struct sockaddr_in *)sa; 1436 kprintf("INET %d %d.%d.%d.%d", 1437 ntohs(sa4->sin_port), 1438 (ntohl(sa4->sin_addr.s_addr) >> 24) & 255, 1439 (ntohl(sa4->sin_addr.s_addr) >> 16) & 255, 1440 (ntohl(sa4->sin_addr.s_addr) >> 8) & 255, 1441 (ntohl(sa4->sin_addr.s_addr) >> 0) & 255 1442 ); 1443 break; 1444 case AF_INET6: 1445 sa6 = (struct sockaddr_in6 *)sa; 1446 kprintf("INET6 %d %04x:%04x%04x:%04x:%04x:%04x:%04x:%04x", 1447 ntohs(sa6->sin6_port), 1448 sa6->sin6_addr.s6_addr16[0], 1449 sa6->sin6_addr.s6_addr16[1], 1450 sa6->sin6_addr.s6_addr16[2], 1451 sa6->sin6_addr.s6_addr16[3], 1452 sa6->sin6_addr.s6_addr16[4], 1453 sa6->sin6_addr.s6_addr16[5], 1454 sa6->sin6_addr.s6_addr16[6], 1455 sa6->sin6_addr.s6_addr16[7] 1456 ); 1457 break; 1458 default: 1459 kprintf("AF%d ", sa->sa_family); 1460 while (len > 0 && sa->sa_data[len-1] == 0) 1461 --len; 1462 1463 for (i = 0; i < len; ++i) { 1464 if (i) 1465 kprintf("."); 1466 kprintf("%d", (unsigned char)sa->sa_data[i]); 1467 } 1468 break; 1469 } 1470 } 1471 } 1472 1473 #endif 1474 1475 /* 1476 * Set up a routing table entry, normally for an interface. 1477 */ 1478 int 1479 rtinit(struct ifaddr *ifa, int cmd, int flags) 1480 { 1481 struct sockaddr *dst, *deldst, *netmask; 1482 struct mbuf *m = NULL; 1483 struct radix_node_head *rnh; 1484 struct radix_node *rn; 1485 struct rt_addrinfo rtinfo; 1486 int error; 1487 1488 ASSERT_NETISR0; 1489 1490 if (flags & RTF_HOST) { 1491 dst = ifa->ifa_dstaddr; 1492 netmask = NULL; 1493 } else { 1494 dst = ifa->ifa_addr; 1495 netmask = ifa->ifa_netmask; 1496 } 1497 /* 1498 * If it's a delete, check that if it exists, it's on the correct 1499 * interface or we might scrub a route to another ifa which would 1500 * be confusing at best and possibly worse. 1501 */ 1502 if (cmd == RTM_DELETE) { 1503 /* 1504 * It's a delete, so it should already exist.. 1505 * If it's a net, mask off the host bits 1506 * (Assuming we have a mask) 1507 */ 1508 if (netmask != NULL) { 1509 m = m_get(M_NOWAIT, MT_SONAME); 1510 if (m == NULL) 1511 return (ENOBUFS); 1512 mbuftrackid(m, 34); 1513 deldst = mtod(m, struct sockaddr *); 1514 rt_maskedcopy(dst, deldst, netmask); 1515 dst = deldst; 1516 } 1517 /* 1518 * Look up an rtentry that is in the routing tree and 1519 * contains the correct info. 1520 */ 1521 if ((rnh = rt_tables[mycpuid][dst->sa_family]) == NULL || 1522 (rn = rnh->rnh_lookup((char *)dst, 1523 (char *)netmask, rnh)) == NULL || 1524 ((struct rtentry *)rn)->rt_ifa != ifa || 1525 !sa_equal((struct sockaddr *)rn->rn_key, dst)) { 1526 if (m != NULL) 1527 m_free(m); 1528 return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH); 1529 } 1530 /* XXX */ 1531 #if 0 1532 else { 1533 /* 1534 * One would think that as we are deleting, and we know 1535 * it doesn't exist, we could just return at this point 1536 * with an "ELSE" clause, but apparently not.. 1537 */ 1538 return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH); 1539 } 1540 #endif 1541 } 1542 /* 1543 * Do the actual request 1544 */ 1545 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 1546 rtinfo.rti_info[RTAX_DST] = dst; 1547 rtinfo.rti_info[RTAX_GATEWAY] = ifa->ifa_addr; 1548 rtinfo.rti_info[RTAX_NETMASK] = netmask; 1549 rtinfo.rti_flags = flags | ifa->ifa_flags; 1550 rtinfo.rti_ifa = ifa; 1551 error = rtrequest1_global(cmd, &rtinfo, rtinit_rtrequest_callback, ifa, 1552 RTREQ_PRIO_HIGH); 1553 if (m != NULL) 1554 m_free(m); 1555 return (error); 1556 } 1557 1558 static void 1559 rtinit_rtrequest_callback(int cmd, int error, 1560 struct rt_addrinfo *rtinfo, struct rtentry *rt, 1561 void *arg) 1562 { 1563 struct ifaddr *ifa = arg; 1564 1565 if (error == 0 && rt) { 1566 if (mycpuid == 0) { 1567 ++rt->rt_refcnt; 1568 rt_newaddrmsg(cmd, ifa, error, rt); 1569 --rt->rt_refcnt; 1570 } 1571 if (cmd == RTM_DELETE) { 1572 if (rt->rt_refcnt == 0) { 1573 ++rt->rt_refcnt; 1574 rtfree(rt); 1575 } 1576 } 1577 } 1578 } 1579 1580 struct netmsg_rts { 1581 struct netmsg_base base; 1582 int req; 1583 struct rt_addrinfo *rtinfo; 1584 rtsearch_callback_func_t callback; 1585 void *arg; 1586 boolean_t exact_match; 1587 int found_cnt; 1588 }; 1589 1590 int 1591 rtsearch_global(int req, struct rt_addrinfo *rtinfo, 1592 rtsearch_callback_func_t callback, void *arg, boolean_t exact_match, 1593 boolean_t req_prio) 1594 { 1595 struct netmsg_rts msg; 1596 int flags = 0; 1597 1598 if (req_prio) 1599 flags = MSGF_PRIORITY; 1600 netmsg_init(&msg.base, NULL, &curthread->td_msgport, flags, 1601 rtsearch_msghandler); 1602 msg.req = req; 1603 msg.rtinfo = rtinfo; 1604 msg.callback = callback; 1605 msg.arg = arg; 1606 msg.exact_match = exact_match; 1607 msg.found_cnt = 0; 1608 return (netisr_domsg_global(&msg.base)); 1609 } 1610 1611 static void 1612 rtsearch_msghandler(netmsg_t msg) 1613 { 1614 struct netmsg_rts *rmsg = (void *)msg; 1615 struct rt_addrinfo rtinfo; 1616 struct radix_node_head *rnh; 1617 struct rtentry *rt; 1618 int error; 1619 1620 ASSERT_NETISR_NCPUS(mycpuid); 1621 1622 /* 1623 * Copy the rtinfo. We need to make sure that the original 1624 * rtinfo, which is setup by the caller, in the netmsg will 1625 * _not_ be changed; else the next CPU on the netmsg forwarding 1626 * path will see a different rtinfo than what this CPU has seen. 1627 */ 1628 rtinfo = *rmsg->rtinfo; 1629 1630 /* 1631 * Find the correct routing tree to use for this Address Family 1632 */ 1633 if ((rnh = rt_tables[mycpuid][rtinfo.rti_dst->sa_family]) == NULL) { 1634 if (mycpuid != 0) 1635 panic("partially initialized routing tables"); 1636 netisr_replymsg(&rmsg->base, EAFNOSUPPORT); 1637 return; 1638 } 1639 1640 /* 1641 * Correct rtinfo for the host route searching. 1642 */ 1643 if (rtinfo.rti_flags & RTF_HOST) { 1644 rtinfo.rti_netmask = NULL; 1645 rtinfo.rti_flags &= ~(RTF_CLONING | RTF_PRCLONING); 1646 } 1647 1648 rt = (struct rtentry *) 1649 rnh->rnh_lookup((char *)rtinfo.rti_dst, 1650 (char *)rtinfo.rti_netmask, rnh); 1651 1652 /* 1653 * If we are asked to do the "exact match", we need to make sure 1654 * that host route searching got a host route while a network 1655 * route searching got a network route. 1656 */ 1657 if (rt != NULL && rmsg->exact_match && 1658 ((rt->rt_flags ^ rtinfo.rti_flags) & RTF_HOST)) 1659 rt = NULL; 1660 1661 if (rt == NULL) { 1662 /* 1663 * No matching routes have been found, don't count this 1664 * as a critical error (here, we set 'error' to 0), just 1665 * keep moving on, since at least prcloned routes are not 1666 * duplicated onto each CPU. 1667 */ 1668 error = 0; 1669 } else { 1670 rmsg->found_cnt++; 1671 1672 rt->rt_refcnt++; 1673 error = rmsg->callback(rmsg->req, &rtinfo, rt, rmsg->arg, 1674 rmsg->found_cnt); 1675 rt->rt_refcnt--; 1676 1677 if (error == EJUSTRETURN) { 1678 netisr_replymsg(&rmsg->base, 0); 1679 return; 1680 } 1681 } 1682 1683 if (error) { 1684 KKASSERT(rmsg->found_cnt > 0); 1685 1686 /* 1687 * Under following cases, unrecoverable error has 1688 * not occured: 1689 * o Request is RTM_GET 1690 * o The first time that we find the route, but the 1691 * modification fails. 1692 */ 1693 if (rmsg->req != RTM_GET && rmsg->found_cnt > 1) { 1694 panic("rtsearch_msghandler: unrecoverable error " 1695 "cpu %d", mycpuid); 1696 } 1697 netisr_replymsg(&rmsg->base, error); 1698 } else { 1699 if (rmsg->found_cnt == 0) { 1700 /* The requested route has not been seen ... */ 1701 error = ESRCH; 1702 } 1703 netisr_forwardmsg_error(&rmsg->base, mycpuid + 1, error); 1704 } 1705 } 1706 1707 int 1708 rtmask_add_global(struct sockaddr *mask, boolean_t req_prio) 1709 { 1710 struct netmsg_base msg; 1711 int flags = 0; 1712 1713 if (req_prio) 1714 flags = MSGF_PRIORITY; 1715 netmsg_init(&msg, NULL, &curthread->td_msgport, flags, 1716 rtmask_add_msghandler); 1717 msg.lmsg.u.ms_resultp = mask; 1718 1719 return (netisr_domsg_global(&msg)); 1720 } 1721 1722 struct sockaddr * 1723 _rtmask_lookup(struct sockaddr *mask, boolean_t search) 1724 { 1725 struct radix_node *n; 1726 1727 #define clen(s) (*(u_char *)(s)) 1728 n = rn_addmask((char *)mask, search, 1, rn_cpumaskhead(mycpuid)); 1729 if (n != NULL && 1730 mask->sa_len >= clen(n->rn_key) && 1731 bcmp((char *)mask + 1, 1732 (char *)n->rn_key + 1, clen(n->rn_key) - 1) == 0) { 1733 return (struct sockaddr *)n->rn_key; 1734 } else { 1735 return NULL; 1736 } 1737 #undef clen 1738 } 1739 1740 static void 1741 rtmask_add_msghandler(netmsg_t msg) 1742 { 1743 struct sockaddr *mask = msg->lmsg.u.ms_resultp; 1744 1745 ASSERT_NETISR_NCPUS(mycpuid); 1746 1747 if (rtmask_lookup(mask) == NULL) { 1748 netisr_replymsg(&msg->base, ENOBUFS); 1749 return; 1750 } 1751 netisr_forwardmsg(&msg->base, mycpuid + 1); 1752 } 1753 1754 /* This must be before ip6_init2(), which is now SI_ORDER_MIDDLE */ 1755 SYSINIT(route, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0); 1756 1757 struct rtchange_arg { 1758 struct ifaddr *old_ifa; 1759 struct ifaddr *new_ifa; 1760 struct rtentry *rt; 1761 int changed; 1762 }; 1763 1764 static void 1765 rtchange_ifa(struct rtentry *rt, struct rtchange_arg *ap) 1766 { 1767 if (rt->rt_ifa->ifa_rtrequest != NULL) 1768 rt->rt_ifa->ifa_rtrequest(RTM_DELETE, rt); 1769 IFAFREE(rt->rt_ifa); 1770 1771 IFAREF(ap->new_ifa); 1772 rt->rt_ifa = ap->new_ifa; 1773 rt->rt_ifp = ap->new_ifa->ifa_ifp; 1774 if (rt->rt_ifa->ifa_rtrequest != NULL) 1775 rt->rt_ifa->ifa_rtrequest(RTM_ADD, rt); 1776 1777 ap->changed = 1; 1778 } 1779 1780 static int 1781 rtchange_callback(struct radix_node *rn, void *xap) 1782 { 1783 struct rtchange_arg *ap = xap; 1784 struct rtentry *rt = (struct rtentry *)rn; 1785 1786 if (rt->rt_ifa == ap->old_ifa) { 1787 if (rt->rt_flags & (RTF_CLONING | RTF_PRCLONING)) { 1788 /* 1789 * We could saw the branch off when we are 1790 * still sitting on it, if the ifa_rtrequest 1791 * DEL/ADD are called directly from here. 1792 */ 1793 ap->rt = rt; 1794 return EJUSTRETURN; 1795 } 1796 rtchange_ifa(rt, ap); 1797 } 1798 return 0; 1799 } 1800 1801 struct netmsg_rtchange { 1802 struct netmsg_base base; 1803 struct ifaddr *old_ifa; 1804 struct ifaddr *new_ifa; 1805 int changed; 1806 }; 1807 1808 static void 1809 rtchange_dispatch(netmsg_t msg) 1810 { 1811 struct netmsg_rtchange *rmsg = (void *)msg; 1812 struct radix_node_head *rnh; 1813 struct rtchange_arg arg; 1814 int cpu; 1815 1816 cpu = mycpuid; 1817 ASSERT_NETISR_NCPUS(cpu); 1818 1819 memset(&arg, 0, sizeof(arg)); 1820 arg.old_ifa = rmsg->old_ifa; 1821 arg.new_ifa = rmsg->new_ifa; 1822 1823 rnh = rt_tables[cpu][AF_INET]; 1824 for (;;) { 1825 int error; 1826 1827 KKASSERT(arg.rt == NULL); 1828 error = rnh->rnh_walktree(rnh, rtchange_callback, &arg); 1829 if (arg.rt != NULL) { 1830 struct rtentry *rt; 1831 1832 rt = arg.rt; 1833 arg.rt = NULL; 1834 rtchange_ifa(rt, &arg); 1835 } else { 1836 break; 1837 } 1838 } 1839 if (arg.changed) 1840 rmsg->changed = 1; 1841 1842 netisr_forwardmsg(&rmsg->base, cpu + 1); 1843 } 1844 1845 int 1846 rtchange(struct ifaddr *old_ifa, struct ifaddr *new_ifa) 1847 { 1848 struct netmsg_rtchange msg; 1849 1850 /* 1851 * XXX individual requests are not independantly chained, 1852 * which means that the per-cpu route tables will not be 1853 * consistent in the middle of the operation. If routes 1854 * related to the interface are manipulated while we are 1855 * doing this the inconsistancy could trigger a panic. 1856 */ 1857 netmsg_init(&msg.base, NULL, &curthread->td_msgport, MSGF_PRIORITY, 1858 rtchange_dispatch); 1859 msg.old_ifa = old_ifa; 1860 msg.new_ifa = new_ifa; 1861 msg.changed = 0; 1862 netisr_domsg_global(&msg.base); 1863 1864 if (msg.changed) { 1865 old_ifa->ifa_flags &= ~IFA_ROUTE; 1866 new_ifa->ifa_flags |= IFA_ROUTE; 1867 return 0; 1868 } else { 1869 return ENOENT; 1870 } 1871 } 1872