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/if_var.h> 81 #include <net/route.h> 82 #include <net/netisr.h> 83 84 #include <netinet/in.h> 85 #include <net/ip_mroute/ip_mroute.h> 86 87 #include <sys/thread2.h> 88 #include <sys/msgport2.h> 89 #include <net/netmsg2.h> 90 #include <net/netisr2.h> 91 92 #ifdef MPLS 93 #include <netproto/mpls/mpls.h> 94 #endif 95 96 static struct rtstatistics rtstatistics_percpu[MAXCPU] __cachealign; 97 #define rtstat rtstatistics_percpu[mycpuid] 98 99 struct radix_node_head *rt_tables[MAXCPU][AF_MAX+1]; 100 101 static void rt_maskedcopy (struct sockaddr *, struct sockaddr *, 102 struct sockaddr *); 103 static void rtable_init(void); 104 static void rtinit_rtrequest_callback(int, int, struct rt_addrinfo *, 105 struct rtentry *, void *); 106 107 static void rtredirect_msghandler(netmsg_t msg); 108 static void rtrequest1_msghandler(netmsg_t msg); 109 static void rtsearch_msghandler(netmsg_t msg); 110 static void rtmask_add_msghandler(netmsg_t msg); 111 112 static int rt_setshims(struct rtentry *, struct sockaddr **); 113 114 SYSCTL_NODE(_net, OID_AUTO, route, CTLFLAG_RW, 0, "Routing"); 115 116 #ifdef ROUTE_DEBUG 117 static int route_debug = 1; 118 SYSCTL_INT(_net_route, OID_AUTO, route_debug, CTLFLAG_RW, 119 &route_debug, 0, ""); 120 #endif 121 122 u_long route_kmalloc_limit = 0; 123 TUNABLE_ULONG("net.route.kmalloc_limit", &route_kmalloc_limit); 124 125 /* 126 * Initialize the route table(s) for protocol domains and 127 * create a helper thread which will be responsible for updating 128 * route table entries on each cpu. 129 */ 130 void 131 route_init(void) 132 { 133 int cpu; 134 135 if (route_kmalloc_limit) 136 kmalloc_raise_limit(M_RTABLE, route_kmalloc_limit); 137 138 for (cpu = 0; cpu < netisr_ncpus; ++cpu) 139 bzero(&rtstatistics_percpu[cpu], sizeof(struct rtstatistics)); 140 rn_init(); /* initialize all zeroes, all ones, mask table */ 141 rtable_init(); /* call dom_rtattach() on each cpu */ 142 } 143 144 static void 145 rtable_init_oncpu(netmsg_t msg) 146 { 147 struct domain *dom; 148 int cpu = mycpuid; 149 150 ASSERT_NETISR_NCPUS(cpu); 151 152 SLIST_FOREACH(dom, &domains, dom_next) { 153 if (dom->dom_rtattach) { 154 dom->dom_rtattach( 155 (void **)&rt_tables[cpu][dom->dom_family], 156 dom->dom_rtoffset); 157 } 158 } 159 netisr_forwardmsg(&msg->base, cpu + 1); 160 } 161 162 static void 163 rtable_init(void) 164 { 165 struct netmsg_base msg; 166 167 netmsg_init(&msg, NULL, &curthread->td_msgport, 0, rtable_init_oncpu); 168 netisr_domsg_global(&msg); 169 } 170 171 /* 172 * Routing statistics. 173 */ 174 static int 175 sysctl_rtstatistics(SYSCTL_HANDLER_ARGS) 176 { 177 int cpu, error = 0; 178 179 for (cpu = 0; cpu < netisr_ncpus; ++cpu) { 180 if ((error = SYSCTL_OUT(req, &rtstatistics_percpu[cpu], 181 sizeof(struct rtstatistics)))) 182 break; 183 if ((error = SYSCTL_IN(req, &rtstatistics_percpu[cpu], 184 sizeof(struct rtstatistics)))) 185 break; 186 } 187 188 return (error); 189 } 190 SYSCTL_PROC(_net_route, OID_AUTO, stats, (CTLTYPE_OPAQUE|CTLFLAG_RW), 191 0, 0, sysctl_rtstatistics, "S,rtstatistics", "Routing statistics"); 192 193 /* 194 * Packet routing routines. 195 */ 196 197 /* 198 * Look up and fill in the "ro_rt" rtentry field in a route structure given 199 * an address in the "ro_dst" field. Always send a report on a miss and 200 * always clone routes. 201 */ 202 void 203 rtalloc(struct route *ro) 204 { 205 rtalloc_ign(ro, 0UL); 206 } 207 208 /* 209 * Look up and fill in the "ro_rt" rtentry field in a route structure given 210 * an address in the "ro_dst" field. Always send a report on a miss and 211 * optionally clone routes when RTF_CLONING or RTF_PRCLONING are not being 212 * ignored. 213 */ 214 void 215 rtalloc_ign(struct route *ro, u_long ignoreflags) 216 { 217 if (ro->ro_rt != NULL) { 218 if (ro->ro_rt->rt_ifp != NULL && ro->ro_rt->rt_flags & RTF_UP) 219 return; 220 rtfree(ro->ro_rt); 221 ro->ro_rt = NULL; 222 } 223 ro->ro_rt = _rtlookup(&ro->ro_dst, RTL_REPORTMSG, ignoreflags); 224 } 225 226 /* 227 * Look up the route that matches the given "dst" address. 228 * 229 * Route lookup can have the side-effect of creating and returning 230 * a cloned route instead when "dst" matches a cloning route and the 231 * RTF_CLONING and RTF_PRCLONING flags are not being ignored. 232 * 233 * Any route returned has its reference count incremented. 234 */ 235 struct rtentry * 236 _rtlookup(struct sockaddr *dst, boolean_t generate_report, u_long ignore) 237 { 238 struct radix_node_head *rnh = rt_tables[mycpuid][dst->sa_family]; 239 struct rtentry *rt; 240 241 ASSERT_NETISR_NCPUS(mycpuid); 242 243 if (rnh == NULL) 244 goto unreach; 245 246 /* 247 * Look up route in the radix tree. 248 */ 249 rt = (struct rtentry *) rnh->rnh_matchaddr((char *)dst, rnh); 250 if (rt == NULL) 251 goto unreach; 252 253 /* 254 * Handle cloning routes. 255 */ 256 if ((rt->rt_flags & ~ignore & (RTF_CLONING | RTF_PRCLONING)) != 0) { 257 struct rtentry *clonedroute; 258 int error; 259 260 clonedroute = rt; /* copy in/copy out parameter */ 261 error = rtrequest(RTM_RESOLVE, dst, NULL, NULL, 0, 262 &clonedroute); /* clone the route */ 263 if (error != 0) { /* cloning failed */ 264 if (generate_report) 265 rt_dstmsg(RTM_MISS, dst, error); 266 rt->rt_refcnt++; 267 return (rt); /* return the uncloned route */ 268 } 269 if (generate_report) { 270 if (clonedroute->rt_flags & RTF_XRESOLVE) 271 rt_dstmsg(RTM_RESOLVE, dst, 0); 272 else 273 rt_rtmsg(RTM_ADD, clonedroute, 274 clonedroute->rt_ifp, 0); 275 } 276 return (clonedroute); /* return cloned route */ 277 } 278 279 /* 280 * Increment the reference count of the matched route and return. 281 */ 282 rt->rt_refcnt++; 283 return (rt); 284 285 unreach: 286 rtstat.rts_unreach++; 287 if (generate_report) 288 rt_dstmsg(RTM_MISS, dst, 0); 289 return (NULL); 290 } 291 292 void 293 rtfree(struct rtentry *rt) 294 { 295 296 ASSERT_NETISR_NCPUS(rt->rt_cpuid); 297 KASSERT(rt->rt_refcnt > 0, ("rtfree: rt_refcnt %ld", rt->rt_refcnt)); 298 299 --rt->rt_refcnt; 300 if (rt->rt_refcnt == 0) { 301 struct radix_node_head *rnh = 302 rt_tables[mycpuid][rt_key(rt)->sa_family]; 303 304 if (rnh->rnh_close) 305 rnh->rnh_close((struct radix_node *)rt, rnh); 306 if (!(rt->rt_flags & RTF_UP)) { 307 /* deallocate route */ 308 if (rt->rt_ifa != NULL) 309 IFAFREE(rt->rt_ifa); 310 if (rt->rt_parent != NULL) 311 RTFREE(rt->rt_parent); /* recursive call! */ 312 Free(rt_key(rt)); 313 Free(rt); 314 } 315 } 316 } 317 318 static void 319 rtfree_async_dispatch(netmsg_t msg) 320 { 321 struct rtentry *rt = msg->lmsg.u.ms_resultp; 322 323 rtfree(rt); 324 netisr_replymsg(&msg->base, 0); 325 } 326 327 void 328 rtfree_async(struct rtentry *rt) 329 { 330 struct netmsg_base *msg; 331 332 if (IN_NETISR_NCPUS(rt->rt_cpuid)) { 333 rtfree(rt); 334 return; 335 } 336 337 KASSERT(rt->rt_refcnt > 0, 338 ("rtfree_async: rt_refcnt %ld", rt->rt_refcnt)); 339 340 msg = kmalloc(sizeof(*msg), M_LWKTMSG, M_INTWAIT); 341 netmsg_init(msg, NULL, &netisr_afree_rport, 0, rtfree_async_dispatch); 342 msg->lmsg.u.ms_resultp = rt; 343 344 netisr_sendmsg(msg, rt->rt_cpuid); 345 } 346 347 int 348 rtredirect_oncpu(struct sockaddr *dst, struct sockaddr *gateway, 349 struct sockaddr *netmask, int flags, struct sockaddr *src) 350 { 351 struct rtentry *rt = NULL; 352 struct rt_addrinfo rtinfo; 353 struct ifaddr *ifa; 354 u_long *stat = NULL; 355 int error; 356 357 ASSERT_NETISR_NCPUS(mycpuid); 358 359 /* verify the gateway is directly reachable */ 360 if ((ifa = ifa_ifwithnet(gateway)) == NULL) { 361 error = ENETUNREACH; 362 goto out; 363 } 364 365 /* 366 * If the redirect isn't from our current router for this destination, 367 * it's either old or wrong. 368 */ 369 if (!(flags & RTF_DONE) && /* XXX JH */ 370 (rt = rtpurelookup(dst)) != NULL && 371 (!sa_equal(src, rt->rt_gateway) || rt->rt_ifa != ifa)) { 372 error = EINVAL; 373 goto done; 374 } 375 376 /* 377 * If it redirects us to ourselves, we have a routing loop, 378 * perhaps as a result of an interface going down recently. 379 */ 380 if (ifa_ifwithaddr(gateway)) { 381 error = EHOSTUNREACH; 382 goto done; 383 } 384 385 /* 386 * Create a new entry if the lookup failed or if we got back 387 * a wildcard entry for the default route. This is necessary 388 * for hosts which use routing redirects generated by smart 389 * gateways to dynamically build the routing tables. 390 */ 391 if (rt == NULL) 392 goto create; 393 if ((rt_mask(rt) != NULL && rt_mask(rt)->sa_len < 2)) { 394 rtfree(rt); 395 goto create; 396 } 397 398 /* Ignore redirects for directly connected hosts. */ 399 if (!(rt->rt_flags & RTF_GATEWAY)) { 400 error = EHOSTUNREACH; 401 goto done; 402 } 403 404 if (!(rt->rt_flags & RTF_HOST) && (flags & RTF_HOST)) { 405 /* 406 * Changing from a network route to a host route. 407 * Create a new host route rather than smashing the 408 * network route. 409 */ 410 create: 411 flags |= RTF_GATEWAY | RTF_DYNAMIC; 412 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 413 rtinfo.rti_info[RTAX_DST] = dst; 414 rtinfo.rti_info[RTAX_GATEWAY] = gateway; 415 rtinfo.rti_info[RTAX_NETMASK] = netmask; 416 rtinfo.rti_flags = flags; 417 rtinfo.rti_ifa = ifa; 418 rt = NULL; /* copy-in/copy-out parameter */ 419 error = rtrequest1(RTM_ADD, &rtinfo, &rt); 420 if (rt != NULL) 421 flags = rt->rt_flags; 422 stat = &rtstat.rts_dynamic; 423 } else { 424 /* 425 * Smash the current notion of the gateway to this destination. 426 * Should check about netmask!!! 427 */ 428 rt->rt_flags |= RTF_MODIFIED; 429 flags |= RTF_MODIFIED; 430 431 /* We only need to report rtmsg on CPU0 */ 432 rt_setgate(rt, rt_key(rt), gateway, 433 mycpuid == 0 ? RTL_REPORTMSG : RTL_DONTREPORT); 434 error = 0; 435 stat = &rtstat.rts_newgateway; 436 } 437 438 done: 439 if (rt != NULL) 440 rtfree(rt); 441 out: 442 if (error != 0) 443 rtstat.rts_badredirect++; 444 else if (stat != NULL) 445 (*stat)++; 446 447 return error; 448 } 449 450 struct netmsg_rtredirect { 451 struct netmsg_base base; 452 struct sockaddr *dst; 453 struct sockaddr *gateway; 454 struct sockaddr *netmask; 455 int flags; 456 struct sockaddr *src; 457 }; 458 459 /* 460 * Force a routing table entry to the specified 461 * destination to go through the given gateway. 462 * Normally called as a result of a routing redirect 463 * message from the network layer. 464 */ 465 void 466 rtredirect(struct sockaddr *dst, struct sockaddr *gateway, 467 struct sockaddr *netmask, int flags, struct sockaddr *src) 468 { 469 struct rt_addrinfo rtinfo; 470 int error; 471 struct netmsg_rtredirect msg; 472 473 netmsg_init(&msg.base, NULL, &curthread->td_msgport, 474 0, rtredirect_msghandler); 475 msg.dst = dst; 476 msg.gateway = gateway; 477 msg.netmask = netmask; 478 msg.flags = flags; 479 msg.src = src; 480 error = netisr_domsg_global(&msg.base); 481 482 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 483 rtinfo.rti_info[RTAX_DST] = dst; 484 rtinfo.rti_info[RTAX_GATEWAY] = gateway; 485 rtinfo.rti_info[RTAX_NETMASK] = netmask; 486 rtinfo.rti_info[RTAX_AUTHOR] = src; 487 rt_missmsg(RTM_REDIRECT, &rtinfo, flags, error); 488 } 489 490 static void 491 rtredirect_msghandler(netmsg_t msg) 492 { 493 struct netmsg_rtredirect *rmsg = (void *)msg; 494 495 rtredirect_oncpu(rmsg->dst, rmsg->gateway, rmsg->netmask, 496 rmsg->flags, rmsg->src); 497 netisr_forwardmsg(&msg->base, mycpuid + 1); 498 } 499 500 /* 501 * Routing table ioctl interface. 502 */ 503 int 504 rtioctl(u_long req, caddr_t data, struct ucred *cred) 505 { 506 #ifdef INET 507 /* Multicast goop, grrr... */ 508 return mrt_ioctl ? mrt_ioctl(req, data) : EOPNOTSUPP; 509 #else 510 return ENXIO; 511 #endif 512 } 513 514 struct ifaddr * 515 ifa_ifwithroute(int flags, struct sockaddr *dst, struct sockaddr *gateway) 516 { 517 struct ifaddr *ifa; 518 519 if (!(flags & RTF_GATEWAY)) { 520 /* 521 * If we are adding a route to an interface, 522 * and the interface is a point-to-point link, 523 * we should search for the destination 524 * as our clue to the interface. Otherwise 525 * we can use the local address. 526 */ 527 ifa = NULL; 528 if (flags & RTF_HOST) { 529 ifa = ifa_ifwithdstaddr(dst); 530 } 531 if (ifa == NULL) 532 ifa = ifa_ifwithaddr(gateway); 533 } else { 534 /* 535 * If we are adding a route to a remote net 536 * or host, the gateway may still be on the 537 * other end of a pt to pt link. 538 */ 539 ifa = ifa_ifwithdstaddr(gateway); 540 } 541 if (ifa == NULL) 542 ifa = ifa_ifwithnet(gateway); 543 if (ifa == NULL) { 544 struct rtentry *rt; 545 546 rt = rtpurelookup(gateway); 547 if (rt == NULL) 548 return (NULL); 549 rt->rt_refcnt--; 550 if ((ifa = rt->rt_ifa) == NULL) 551 return (NULL); 552 } 553 if (ifa->ifa_addr->sa_family != dst->sa_family) { 554 struct ifaddr *oldifa = ifa; 555 556 ifa = ifaof_ifpforaddr(dst, ifa->ifa_ifp); 557 if (ifa == NULL) 558 ifa = oldifa; 559 } 560 return (ifa); 561 } 562 563 static int rt_fixdelete (struct radix_node *, void *); 564 static int rt_fixchange (struct radix_node *, void *); 565 566 struct rtfc_arg { 567 struct rtentry *rt0; 568 struct radix_node_head *rnh; 569 }; 570 571 /* 572 * Set rtinfo->rti_ifa and rtinfo->rti_ifp. 573 */ 574 int 575 rt_getifa(struct rt_addrinfo *rtinfo) 576 { 577 struct sockaddr *gateway = rtinfo->rti_info[RTAX_GATEWAY]; 578 struct sockaddr *dst = rtinfo->rti_info[RTAX_DST]; 579 struct sockaddr *ifaaddr = rtinfo->rti_info[RTAX_IFA]; 580 int flags = rtinfo->rti_flags; 581 582 /* 583 * ifp may be specified by sockaddr_dl 584 * when protocol address is ambiguous. 585 */ 586 if (rtinfo->rti_ifp == NULL) { 587 struct sockaddr *ifpaddr; 588 589 ifpaddr = rtinfo->rti_info[RTAX_IFP]; 590 if (ifpaddr != NULL && ifpaddr->sa_family == AF_LINK) { 591 struct ifaddr *ifa; 592 593 ifa = ifa_ifwithnet(ifpaddr); 594 if (ifa != NULL) 595 rtinfo->rti_ifp = ifa->ifa_ifp; 596 } 597 } 598 599 if (rtinfo->rti_ifa == NULL && ifaaddr != NULL) 600 rtinfo->rti_ifa = ifa_ifwithaddr(ifaaddr); 601 if (rtinfo->rti_ifa == NULL) { 602 struct sockaddr *sa; 603 604 sa = ifaaddr != NULL ? ifaaddr : 605 (gateway != NULL ? gateway : dst); 606 if (sa != NULL && rtinfo->rti_ifp != NULL) 607 rtinfo->rti_ifa = ifaof_ifpforaddr(sa, rtinfo->rti_ifp); 608 else if (dst != NULL && gateway != NULL) 609 rtinfo->rti_ifa = ifa_ifwithroute(flags, dst, gateway); 610 else if (sa != NULL) 611 rtinfo->rti_ifa = ifa_ifwithroute(flags, sa, sa); 612 } 613 if (rtinfo->rti_ifa == NULL) 614 return (ENETUNREACH); 615 616 if (rtinfo->rti_ifp == NULL) 617 rtinfo->rti_ifp = rtinfo->rti_ifa->ifa_ifp; 618 return (0); 619 } 620 621 /* 622 * Do appropriate manipulations of a routing tree given 623 * all the bits of info needed 624 */ 625 int 626 rtrequest( 627 int req, 628 struct sockaddr *dst, 629 struct sockaddr *gateway, 630 struct sockaddr *netmask, 631 int flags, 632 struct rtentry **ret_nrt) 633 { 634 struct rt_addrinfo rtinfo; 635 636 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 637 rtinfo.rti_info[RTAX_DST] = dst; 638 rtinfo.rti_info[RTAX_GATEWAY] = gateway; 639 rtinfo.rti_info[RTAX_NETMASK] = netmask; 640 rtinfo.rti_flags = flags; 641 return rtrequest1(req, &rtinfo, ret_nrt); 642 } 643 644 int 645 rtrequest_global( 646 int req, 647 struct sockaddr *dst, 648 struct sockaddr *gateway, 649 struct sockaddr *netmask, 650 int flags) 651 { 652 struct rt_addrinfo rtinfo; 653 654 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 655 rtinfo.rti_info[RTAX_DST] = dst; 656 rtinfo.rti_info[RTAX_GATEWAY] = gateway; 657 rtinfo.rti_info[RTAX_NETMASK] = netmask; 658 rtinfo.rti_flags = flags; 659 return rtrequest1_global(req, &rtinfo, NULL, NULL, RTREQ_PRIO_NORM); 660 } 661 662 struct netmsg_rtq { 663 struct netmsg_base base; 664 int req; 665 struct rt_addrinfo *rtinfo; 666 rtrequest1_callback_func_t callback; 667 void *arg; 668 }; 669 670 int 671 rtrequest1_global(int req, struct rt_addrinfo *rtinfo, 672 rtrequest1_callback_func_t callback, void *arg, boolean_t req_prio) 673 { 674 struct netmsg_rtq msg; 675 int flags = 0; 676 677 if (req_prio) 678 flags = MSGF_PRIORITY; 679 netmsg_init(&msg.base, NULL, &curthread->td_msgport, flags, 680 rtrequest1_msghandler); 681 msg.base.lmsg.ms_error = -1; 682 msg.req = req; 683 msg.rtinfo = rtinfo; 684 msg.callback = callback; 685 msg.arg = arg; 686 return (netisr_domsg_global(&msg.base)); 687 } 688 689 /* 690 * Handle a route table request on the current cpu. Since the route table's 691 * are supposed to be identical on each cpu, an error occuring later in the 692 * message chain is considered system-fatal. 693 */ 694 static void 695 rtrequest1_msghandler(netmsg_t msg) 696 { 697 struct netmsg_rtq *rmsg = (void *)msg; 698 struct rt_addrinfo rtinfo; 699 struct rtentry *rt = NULL; 700 int error; 701 702 /* 703 * Copy the rtinfo. We need to make sure that the original 704 * rtinfo, which is setup by the caller, in the netmsg will 705 * _not_ be changed; else the next CPU on the netmsg forwarding 706 * path will see a different rtinfo than what this CPU has seen. 707 */ 708 rtinfo = *rmsg->rtinfo; 709 710 error = rtrequest1(rmsg->req, &rtinfo, &rt); 711 if (rt) 712 --rt->rt_refcnt; 713 if (rmsg->callback) 714 rmsg->callback(rmsg->req, error, &rtinfo, rt, rmsg->arg); 715 716 /* 717 * RTM_DELETE's are propogated even if an error occurs, since a 718 * cloned route might be undergoing deletion and cloned routes 719 * are not necessarily replicated. An overall error is returned 720 * only if no cpus have the route in question. 721 */ 722 if (rmsg->base.lmsg.ms_error < 0 || error == 0) 723 rmsg->base.lmsg.ms_error = error; 724 725 if (error && rmsg->req != RTM_DELETE) { 726 if (mycpuid != 0) { 727 panic("rtrequest1_msghandler: rtrequest table req %d, " 728 "failed on cpu%d, error %d\n", 729 rmsg->req, mycpuid, error); 730 } 731 netisr_replymsg(&rmsg->base, error); 732 } else { 733 netisr_forwardmsg_error(&rmsg->base, mycpuid + 1, 734 rmsg->base.lmsg.ms_error); 735 } 736 } 737 738 int 739 rtrequest1(int req, struct rt_addrinfo *rtinfo, struct rtentry **ret_nrt) 740 { 741 struct sockaddr *dst = rtinfo->rti_info[RTAX_DST]; 742 struct rtentry *rt; 743 struct radix_node *rn; 744 struct radix_node_head *rnh; 745 struct ifaddr *ifa; 746 struct sockaddr *ndst; 747 boolean_t reportmsg; 748 int error = 0; 749 750 ASSERT_NETISR_NCPUS(mycpuid); 751 752 #define gotoerr(x) { error = x ; goto bad; } 753 754 #ifdef ROUTE_DEBUG 755 if (route_debug) 756 rt_addrinfo_print(req, rtinfo); 757 #endif 758 759 crit_enter(); 760 /* 761 * Find the correct routing tree to use for this Address Family 762 */ 763 if ((rnh = rt_tables[mycpuid][dst->sa_family]) == NULL) 764 gotoerr(EAFNOSUPPORT); 765 766 /* 767 * If we are adding a host route then we don't want to put 768 * a netmask in the tree, nor do we want to clone it. 769 */ 770 if (rtinfo->rti_flags & RTF_HOST) { 771 rtinfo->rti_info[RTAX_NETMASK] = NULL; 772 rtinfo->rti_flags &= ~(RTF_CLONING | RTF_PRCLONING); 773 } 774 775 switch (req) { 776 case RTM_DELETE: 777 /* Remove the item from the tree. */ 778 rn = rnh->rnh_deladdr((char *)rtinfo->rti_info[RTAX_DST], 779 (char *)rtinfo->rti_info[RTAX_NETMASK], 780 rnh); 781 if (rn == NULL) 782 gotoerr(ESRCH); 783 KASSERT(!(rn->rn_flags & (RNF_ACTIVE | RNF_ROOT)), 784 ("rnh_deladdr returned flags 0x%x", rn->rn_flags)); 785 rt = (struct rtentry *)rn; 786 787 /* ref to prevent a deletion race */ 788 ++rt->rt_refcnt; 789 790 /* Free any routes cloned from this one. */ 791 if ((rt->rt_flags & (RTF_CLONING | RTF_PRCLONING)) && 792 rt_mask(rt) != NULL) { 793 rnh->rnh_walktree_from(rnh, (char *)rt_key(rt), 794 (char *)rt_mask(rt), 795 rt_fixdelete, rt); 796 } 797 798 if (rt->rt_gwroute != NULL) { 799 RTFREE(rt->rt_gwroute); 800 rt->rt_gwroute = NULL; 801 } 802 803 /* 804 * NB: RTF_UP must be set during the search above, 805 * because we might delete the last ref, causing 806 * rt to get freed prematurely. 807 */ 808 rt->rt_flags &= ~RTF_UP; 809 810 #ifdef ROUTE_DEBUG 811 if (route_debug) 812 rt_print(rtinfo, rt); 813 #endif 814 815 /* Give the protocol a chance to keep things in sync. */ 816 if ((ifa = rt->rt_ifa) && ifa->ifa_rtrequest) 817 ifa->ifa_rtrequest(RTM_DELETE, rt); 818 819 /* 820 * If the caller wants it, then it can have it, 821 * but it's up to it to free the rtentry as we won't be 822 * doing it. 823 */ 824 KASSERT(rt->rt_refcnt >= 0, 825 ("rtrequest1(DELETE): refcnt %ld", rt->rt_refcnt)); 826 if (ret_nrt != NULL) { 827 /* leave ref intact for return */ 828 *ret_nrt = rt; 829 } else { 830 /* deref / attempt to destroy */ 831 rtfree(rt); 832 } 833 break; 834 835 case RTM_RESOLVE: 836 if (ret_nrt == NULL || (rt = *ret_nrt) == NULL) 837 gotoerr(EINVAL); 838 839 KASSERT(rt->rt_cpuid == mycpuid, 840 ("rt resolve rt_cpuid %d, mycpuid %d", 841 rt->rt_cpuid, mycpuid)); 842 843 ifa = rt->rt_ifa; 844 rtinfo->rti_flags = 845 rt->rt_flags & ~(RTF_CLONING | RTF_PRCLONING | RTF_STATIC); 846 rtinfo->rti_flags |= RTF_WASCLONED; 847 rtinfo->rti_info[RTAX_GATEWAY] = rt->rt_gateway; 848 if ((rtinfo->rti_info[RTAX_NETMASK] = rt->rt_genmask) == NULL) 849 rtinfo->rti_flags |= RTF_HOST; 850 rtinfo->rti_info[RTAX_MPLS1] = rt->rt_shim[0]; 851 rtinfo->rti_info[RTAX_MPLS2] = rt->rt_shim[1]; 852 rtinfo->rti_info[RTAX_MPLS3] = rt->rt_shim[2]; 853 goto makeroute; 854 855 case RTM_ADD: 856 KASSERT(!(rtinfo->rti_flags & RTF_GATEWAY) || 857 rtinfo->rti_info[RTAX_GATEWAY] != NULL, 858 ("rtrequest: GATEWAY but no gateway")); 859 860 if (rtinfo->rti_ifa == NULL && (error = rt_getifa(rtinfo))) 861 gotoerr(error); 862 ifa = rtinfo->rti_ifa; 863 makeroute: 864 R_Malloc(rt, struct rtentry *, sizeof(struct rtentry)); 865 if (rt == NULL) { 866 if (req == RTM_ADD) { 867 kprintf("rtrequest1: alloc rtentry failed on " 868 "cpu%d\n", mycpuid); 869 } 870 gotoerr(ENOBUFS); 871 } 872 bzero(rt, sizeof(struct rtentry)); 873 rt->rt_flags = RTF_UP | rtinfo->rti_flags; 874 rt->rt_cpuid = mycpuid; 875 876 if (mycpuid != 0 && req == RTM_ADD) { 877 /* For RTM_ADD, we have already sent rtmsg on CPU0. */ 878 reportmsg = RTL_DONTREPORT; 879 } else { 880 /* 881 * For RTM_ADD, we only send rtmsg on CPU0. 882 * For RTM_RESOLVE, we always send rtmsg. XXX 883 */ 884 reportmsg = RTL_REPORTMSG; 885 } 886 error = rt_setgate(rt, dst, rtinfo->rti_info[RTAX_GATEWAY], 887 reportmsg); 888 if (error != 0) { 889 Free(rt); 890 gotoerr(error); 891 } 892 893 ndst = rt_key(rt); 894 if (rtinfo->rti_info[RTAX_NETMASK] != NULL) 895 rt_maskedcopy(dst, ndst, 896 rtinfo->rti_info[RTAX_NETMASK]); 897 else 898 bcopy(dst, ndst, dst->sa_len); 899 900 if (rtinfo->rti_info[RTAX_MPLS1] != NULL) 901 rt_setshims(rt, rtinfo->rti_info); 902 903 /* 904 * Note that we now have a reference to the ifa. 905 * This moved from below so that rnh->rnh_addaddr() can 906 * examine the ifa and ifa->ifa_ifp if it so desires. 907 */ 908 IFAREF(ifa); 909 rt->rt_ifa = ifa; 910 rt->rt_ifp = ifa->ifa_ifp; 911 /* XXX mtu manipulation will be done in rnh_addaddr -- itojun */ 912 913 rn = rnh->rnh_addaddr((char *)ndst, 914 (char *)rtinfo->rti_info[RTAX_NETMASK], 915 rnh, rt->rt_nodes); 916 if (rn == NULL) { 917 struct rtentry *oldrt; 918 919 /* 920 * We already have one of these in the tree. 921 * We do a special hack: if the old route was 922 * cloned, then we blow it away and try 923 * re-inserting the new one. 924 */ 925 oldrt = rtpurelookup(ndst); 926 if (oldrt != NULL) { 927 --oldrt->rt_refcnt; 928 if (oldrt->rt_flags & RTF_WASCLONED) { 929 rtrequest(RTM_DELETE, rt_key(oldrt), 930 oldrt->rt_gateway, 931 rt_mask(oldrt), 932 oldrt->rt_flags, NULL); 933 rn = rnh->rnh_addaddr((char *)ndst, 934 (char *) 935 rtinfo->rti_info[RTAX_NETMASK], 936 rnh, rt->rt_nodes); 937 } 938 } 939 } 940 /* NOTE: rt_ifa may have been changed */ 941 ifa = rt->rt_ifa; 942 943 /* 944 * If it still failed to go into the tree, 945 * then un-make it (this should be a function). 946 */ 947 if (rn == NULL) { 948 if (rt->rt_gwroute != NULL) 949 rtfree(rt->rt_gwroute); 950 IFAFREE(ifa); 951 Free(rt_key(rt)); 952 Free(rt); 953 gotoerr(EEXIST); 954 } 955 956 /* 957 * If we got here from RESOLVE, then we are cloning 958 * so clone the rest, and note that we 959 * are a clone (and increment the parent's references) 960 */ 961 if (req == RTM_RESOLVE) { 962 rt->rt_rmx = (*ret_nrt)->rt_rmx; /* copy metrics */ 963 rt->rt_rmx.rmx_pksent = 0; /* reset packet counter */ 964 if ((*ret_nrt)->rt_flags & 965 (RTF_CLONING | RTF_PRCLONING)) { 966 rt->rt_parent = *ret_nrt; 967 (*ret_nrt)->rt_refcnt++; 968 } 969 } 970 971 /* 972 * if this protocol has something to add to this then 973 * allow it to do that as well. 974 */ 975 if (ifa->ifa_rtrequest != NULL) 976 ifa->ifa_rtrequest(req, rt); 977 978 /* 979 * We repeat the same procedure from rt_setgate() here because 980 * it doesn't fire when we call it there because the node 981 * hasn't been added to the tree yet. 982 */ 983 if (req == RTM_ADD && !(rt->rt_flags & RTF_HOST) && 984 rt_mask(rt) != NULL) { 985 struct rtfc_arg arg = { rt, rnh }; 986 987 rnh->rnh_walktree_from(rnh, (char *)rt_key(rt), 988 (char *)rt_mask(rt), 989 rt_fixchange, &arg); 990 } 991 992 #ifdef ROUTE_DEBUG 993 if (route_debug) 994 rt_print(rtinfo, rt); 995 #endif 996 /* 997 * Return the resulting rtentry, 998 * increasing the number of references by one. 999 */ 1000 if (ret_nrt != NULL) { 1001 rt->rt_refcnt++; 1002 *ret_nrt = rt; 1003 } 1004 break; 1005 default: 1006 error = EOPNOTSUPP; 1007 } 1008 bad: 1009 #ifdef ROUTE_DEBUG 1010 if (route_debug) { 1011 if (error) 1012 kprintf("rti %p failed error %d\n", rtinfo, error); 1013 else 1014 kprintf("rti %p succeeded\n", rtinfo); 1015 } 1016 #endif 1017 crit_exit(); 1018 return (error); 1019 } 1020 1021 /* 1022 * Called from rtrequest(RTM_DELETE, ...) to fix up the route's ``family'' 1023 * (i.e., the routes related to it by the operation of cloning). This 1024 * routine is iterated over all potential former-child-routes by way of 1025 * rnh->rnh_walktree_from() above, and those that actually are children of 1026 * the late parent (passed in as VP here) are themselves deleted. 1027 */ 1028 static int 1029 rt_fixdelete(struct radix_node *rn, void *vp) 1030 { 1031 struct rtentry *rt = (struct rtentry *)rn; 1032 struct rtentry *rt0 = vp; 1033 1034 if (rt->rt_parent == rt0 && 1035 !(rt->rt_flags & (RTF_PINNED | RTF_CLONING | RTF_PRCLONING))) { 1036 return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 1037 rt->rt_flags, NULL); 1038 } 1039 return 0; 1040 } 1041 1042 /* 1043 * This routine is called from rt_setgate() to do the analogous thing for 1044 * adds and changes. There is the added complication in this case of a 1045 * middle insert; i.e., insertion of a new network route between an older 1046 * network route and (cloned) host routes. For this reason, a simple check 1047 * of rt->rt_parent is insufficient; each candidate route must be tested 1048 * against the (mask, value) of the new route (passed as before in vp) 1049 * to see if the new route matches it. 1050 * 1051 * XXX - it may be possible to do fixdelete() for changes and reserve this 1052 * routine just for adds. I'm not sure why I thought it was necessary to do 1053 * changes this way. 1054 */ 1055 #ifdef DEBUG 1056 static int rtfcdebug = 0; 1057 #endif 1058 1059 static int 1060 rt_fixchange(struct radix_node *rn, void *vp) 1061 { 1062 struct rtentry *rt = (struct rtentry *)rn; 1063 struct rtfc_arg *ap = vp; 1064 struct rtentry *rt0 = ap->rt0; 1065 struct radix_node_head *rnh = ap->rnh; 1066 u_char *xk1, *xm1, *xk2, *xmp; 1067 int i, len, mlen; 1068 1069 #ifdef DEBUG 1070 if (rtfcdebug) 1071 kprintf("rt_fixchange: rt %p, rt0 %p\n", rt, rt0); 1072 #endif 1073 1074 if (rt->rt_parent == NULL || 1075 (rt->rt_flags & (RTF_PINNED | RTF_CLONING | RTF_PRCLONING))) { 1076 #ifdef DEBUG 1077 if (rtfcdebug) kprintf("no parent, pinned or cloning\n"); 1078 #endif 1079 return 0; 1080 } 1081 1082 if (rt->rt_parent == rt0) { 1083 #ifdef DEBUG 1084 if (rtfcdebug) kprintf("parent match\n"); 1085 #endif 1086 return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 1087 rt->rt_flags, NULL); 1088 } 1089 1090 /* 1091 * There probably is a function somewhere which does this... 1092 * if not, there should be. 1093 */ 1094 len = imin(rt_key(rt0)->sa_len, rt_key(rt)->sa_len); 1095 1096 xk1 = (u_char *)rt_key(rt0); 1097 xm1 = (u_char *)rt_mask(rt0); 1098 xk2 = (u_char *)rt_key(rt); 1099 1100 /* avoid applying a less specific route */ 1101 xmp = (u_char *)rt_mask(rt->rt_parent); 1102 mlen = rt_key(rt->rt_parent)->sa_len; 1103 if (mlen > rt_key(rt0)->sa_len) { 1104 #ifdef DEBUG 1105 if (rtfcdebug) 1106 kprintf("rt_fixchange: inserting a less " 1107 "specific route\n"); 1108 #endif 1109 return 0; 1110 } 1111 for (i = rnh->rnh_treetop->rn_offset; i < mlen; i++) { 1112 if ((xmp[i] & ~(xmp[i] ^ xm1[i])) != xmp[i]) { 1113 #ifdef DEBUG 1114 if (rtfcdebug) 1115 kprintf("rt_fixchange: inserting a less " 1116 "specific route\n"); 1117 #endif 1118 return 0; 1119 } 1120 } 1121 1122 for (i = rnh->rnh_treetop->rn_offset; i < len; i++) { 1123 if ((xk2[i] & xm1[i]) != xk1[i]) { 1124 #ifdef DEBUG 1125 if (rtfcdebug) kprintf("no match\n"); 1126 #endif 1127 return 0; 1128 } 1129 } 1130 1131 /* 1132 * OK, this node is a clone, and matches the node currently being 1133 * changed/added under the node's mask. So, get rid of it. 1134 */ 1135 #ifdef DEBUG 1136 if (rtfcdebug) kprintf("deleting\n"); 1137 #endif 1138 return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 1139 rt->rt_flags, NULL); 1140 } 1141 1142 int 1143 rt_setgate(struct rtentry *rt0, struct sockaddr *dst, struct sockaddr *gate, 1144 boolean_t generate_report) 1145 { 1146 char *space, *oldspace; 1147 int dlen = RT_ROUNDUP(dst->sa_len), glen = RT_ROUNDUP(gate->sa_len); 1148 struct rtentry *rt = rt0; 1149 struct radix_node_head *rnh = rt_tables[mycpuid][dst->sa_family]; 1150 1151 ASSERT_NETISR_NCPUS(mycpuid); 1152 1153 /* 1154 * A host route with the destination equal to the gateway 1155 * will interfere with keeping LLINFO in the routing 1156 * table, so disallow it. 1157 */ 1158 if (((rt0->rt_flags & (RTF_HOST | RTF_GATEWAY | RTF_LLINFO)) == 1159 (RTF_HOST | RTF_GATEWAY)) && 1160 dst->sa_len == gate->sa_len && 1161 sa_equal(dst, gate)) { 1162 /* 1163 * The route might already exist if this is an RTM_CHANGE 1164 * or a routing redirect, so try to delete it. 1165 */ 1166 if (rt_key(rt0) != NULL) 1167 rtrequest(RTM_DELETE, rt_key(rt0), rt0->rt_gateway, 1168 rt_mask(rt0), rt0->rt_flags, NULL); 1169 return EADDRNOTAVAIL; 1170 } 1171 1172 /* 1173 * Both dst and gateway are stored in the same malloc'ed chunk 1174 * (If I ever get my hands on....) 1175 * if we need to malloc a new chunk, then keep the old one around 1176 * till we don't need it any more. 1177 */ 1178 if (rt->rt_gateway == NULL || 1179 glen > RT_ROUNDUP(rt->rt_gateway->sa_len)) { 1180 oldspace = (char *)rt_key(rt); 1181 R_Malloc(space, char *, dlen + glen); 1182 if (space == NULL) 1183 return ENOBUFS; 1184 rt->rt_nodes->rn_key = space; 1185 } else { 1186 space = (char *)rt_key(rt); /* Just use the old space. */ 1187 oldspace = NULL; 1188 } 1189 1190 /* Set the gateway value. */ 1191 rt->rt_gateway = (struct sockaddr *)(space + dlen); 1192 bcopy(gate, rt->rt_gateway, glen); 1193 1194 if (oldspace != NULL) { 1195 /* 1196 * If we allocated a new chunk, preserve the original dst. 1197 * This way, rt_setgate() really just sets the gate 1198 * and leaves the dst field alone. 1199 */ 1200 bcopy(dst, space, dlen); 1201 Free(oldspace); 1202 } 1203 1204 /* 1205 * If there is already a gwroute, it's now almost definitely wrong 1206 * so drop it. 1207 */ 1208 if (rt->rt_gwroute != NULL) { 1209 RTFREE(rt->rt_gwroute); 1210 rt->rt_gwroute = NULL; 1211 } 1212 if (rt->rt_flags & RTF_GATEWAY) { 1213 /* 1214 * Cloning loop avoidance: In the presence of 1215 * protocol-cloning and bad configuration, it is 1216 * possible to get stuck in bottomless mutual recursion 1217 * (rtrequest rt_setgate rtlookup). We avoid this 1218 * by not allowing protocol-cloning to operate for 1219 * gateways (which is probably the correct choice 1220 * anyway), and avoid the resulting reference loops 1221 * by disallowing any route to run through itself as 1222 * a gateway. This is obviously mandatory when we 1223 * get rt->rt_output(). 1224 * 1225 * This breaks TTCP for hosts outside the gateway! XXX JH 1226 */ 1227 rt->rt_gwroute = _rtlookup(gate, generate_report, 1228 RTF_PRCLONING); 1229 if (rt->rt_gwroute == rt) { 1230 rt->rt_gwroute = NULL; 1231 --rt->rt_refcnt; 1232 return EDQUOT; /* failure */ 1233 } 1234 } 1235 1236 /* 1237 * This isn't going to do anything useful for host routes, so 1238 * don't bother. Also make sure we have a reasonable mask 1239 * (we don't yet have one during adds). 1240 */ 1241 if (!(rt->rt_flags & RTF_HOST) && rt_mask(rt) != NULL) { 1242 struct rtfc_arg arg = { rt, rnh }; 1243 1244 rnh->rnh_walktree_from(rnh, (char *)rt_key(rt), 1245 (char *)rt_mask(rt), 1246 rt_fixchange, &arg); 1247 } 1248 1249 return 0; 1250 } 1251 1252 static void 1253 rt_maskedcopy( 1254 struct sockaddr *src, 1255 struct sockaddr *dst, 1256 struct sockaddr *netmask) 1257 { 1258 u_char *cp1 = (u_char *)src; 1259 u_char *cp2 = (u_char *)dst; 1260 u_char *cp3 = (u_char *)netmask; 1261 u_char *cplim = cp2 + *cp3; 1262 u_char *cplim2 = cp2 + *cp1; 1263 1264 *cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */ 1265 cp3 += 2; 1266 if (cplim > cplim2) 1267 cplim = cplim2; 1268 while (cp2 < cplim) 1269 *cp2++ = *cp1++ & *cp3++; 1270 if (cp2 < cplim2) 1271 bzero(cp2, cplim2 - cp2); 1272 } 1273 1274 int 1275 rt_llroute(struct sockaddr *dst, struct rtentry *rt0, struct rtentry **drt) 1276 { 1277 struct rtentry *up_rt, *rt; 1278 1279 ASSERT_NETISR_NCPUS(mycpuid); 1280 1281 if (!(rt0->rt_flags & RTF_UP)) { 1282 up_rt = rtlookup(dst); 1283 if (up_rt == NULL) 1284 return (EHOSTUNREACH); 1285 up_rt->rt_refcnt--; 1286 } else 1287 up_rt = rt0; 1288 if (up_rt->rt_flags & RTF_GATEWAY) { 1289 if (up_rt->rt_gwroute == NULL) { 1290 up_rt->rt_gwroute = rtlookup(up_rt->rt_gateway); 1291 if (up_rt->rt_gwroute == NULL) 1292 return (EHOSTUNREACH); 1293 } else if (!(up_rt->rt_gwroute->rt_flags & RTF_UP)) { 1294 rtfree(up_rt->rt_gwroute); 1295 up_rt->rt_gwroute = rtlookup(up_rt->rt_gateway); 1296 if (up_rt->rt_gwroute == NULL) 1297 return (EHOSTUNREACH); 1298 } 1299 rt = up_rt->rt_gwroute; 1300 } else 1301 rt = up_rt; 1302 if (rt->rt_flags & RTF_REJECT && 1303 (rt->rt_rmx.rmx_expire == 0 || /* rt doesn't expire */ 1304 time_uptime < rt->rt_rmx.rmx_expire)) /* rt not expired */ 1305 return (rt->rt_flags & RTF_HOST ? EHOSTDOWN : EHOSTUNREACH); 1306 *drt = rt; 1307 return 0; 1308 } 1309 1310 static int 1311 rt_setshims(struct rtentry *rt, struct sockaddr **rt_shim){ 1312 int i; 1313 1314 for (i=0; i<3; i++) { 1315 struct sockaddr *shim = rt_shim[RTAX_MPLS1 + i]; 1316 int shimlen; 1317 1318 if (shim == NULL) 1319 break; 1320 1321 shimlen = RT_ROUNDUP(shim->sa_len); 1322 R_Malloc(rt->rt_shim[i], struct sockaddr *, shimlen); 1323 bcopy(shim, rt->rt_shim[i], shimlen); 1324 } 1325 1326 return 0; 1327 } 1328 1329 #ifdef ROUTE_DEBUG 1330 1331 /* 1332 * Print out a route table entry 1333 */ 1334 void 1335 rt_print(struct rt_addrinfo *rtinfo, struct rtentry *rn) 1336 { 1337 kprintf("rti %p cpu %d route %p flags %08lx: ", 1338 rtinfo, mycpuid, rn, rn->rt_flags); 1339 sockaddr_print(rt_key(rn)); 1340 kprintf(" mask "); 1341 sockaddr_print(rt_mask(rn)); 1342 kprintf(" gw "); 1343 sockaddr_print(rn->rt_gateway); 1344 kprintf(" ifc \"%s\"", rn->rt_ifp ? rn->rt_ifp->if_dname : "?"); 1345 kprintf(" ifa %p\n", rn->rt_ifa); 1346 } 1347 1348 void 1349 rt_addrinfo_print(int cmd, struct rt_addrinfo *rti) 1350 { 1351 int didit = 0; 1352 int i; 1353 1354 #ifdef ROUTE_DEBUG 1355 if (cmd == RTM_DELETE && route_debug > 1) 1356 print_backtrace(-1); 1357 #endif 1358 1359 switch(cmd) { 1360 case RTM_ADD: 1361 kprintf("ADD "); 1362 break; 1363 case RTM_RESOLVE: 1364 kprintf("RES "); 1365 break; 1366 case RTM_DELETE: 1367 kprintf("DEL "); 1368 break; 1369 default: 1370 kprintf("C%02d ", cmd); 1371 break; 1372 } 1373 kprintf("rti %p cpu %d ", rti, mycpuid); 1374 for (i = 0; i < rti->rti_addrs; ++i) { 1375 if (rti->rti_info[i] == NULL) 1376 continue; 1377 if (didit) 1378 kprintf(" ,"); 1379 switch(i) { 1380 case RTAX_DST: 1381 kprintf("(DST "); 1382 break; 1383 case RTAX_GATEWAY: 1384 kprintf("(GWY "); 1385 break; 1386 case RTAX_NETMASK: 1387 kprintf("(MSK "); 1388 break; 1389 case RTAX_GENMASK: 1390 kprintf("(GEN "); 1391 break; 1392 case RTAX_IFP: 1393 kprintf("(IFP "); 1394 break; 1395 case RTAX_IFA: 1396 kprintf("(IFA "); 1397 break; 1398 case RTAX_AUTHOR: 1399 kprintf("(AUT "); 1400 break; 1401 case RTAX_BRD: 1402 kprintf("(BRD "); 1403 break; 1404 default: 1405 kprintf("(?%02d ", i); 1406 break; 1407 } 1408 sockaddr_print(rti->rti_info[i]); 1409 kprintf(")"); 1410 didit = 1; 1411 } 1412 kprintf("\n"); 1413 } 1414 1415 void 1416 sockaddr_print(struct sockaddr *sa) 1417 { 1418 struct sockaddr_in *sa4; 1419 struct sockaddr_in6 *sa6; 1420 int len; 1421 int i; 1422 1423 if (sa == NULL) { 1424 kprintf("NULL"); 1425 return; 1426 } 1427 1428 len = sa->sa_len - offsetof(struct sockaddr, sa_data[0]); 1429 1430 switch(sa->sa_family) { 1431 case AF_INET: 1432 case AF_INET6: 1433 default: 1434 switch(sa->sa_family) { 1435 case AF_INET: 1436 sa4 = (struct sockaddr_in *)sa; 1437 kprintf("INET %d %d.%d.%d.%d", 1438 ntohs(sa4->sin_port), 1439 (ntohl(sa4->sin_addr.s_addr) >> 24) & 255, 1440 (ntohl(sa4->sin_addr.s_addr) >> 16) & 255, 1441 (ntohl(sa4->sin_addr.s_addr) >> 8) & 255, 1442 (ntohl(sa4->sin_addr.s_addr) >> 0) & 255 1443 ); 1444 break; 1445 case AF_INET6: 1446 sa6 = (struct sockaddr_in6 *)sa; 1447 kprintf("INET6 %d %04x:%04x%04x:%04x:%04x:%04x:%04x:%04x", 1448 ntohs(sa6->sin6_port), 1449 sa6->sin6_addr.s6_addr16[0], 1450 sa6->sin6_addr.s6_addr16[1], 1451 sa6->sin6_addr.s6_addr16[2], 1452 sa6->sin6_addr.s6_addr16[3], 1453 sa6->sin6_addr.s6_addr16[4], 1454 sa6->sin6_addr.s6_addr16[5], 1455 sa6->sin6_addr.s6_addr16[6], 1456 sa6->sin6_addr.s6_addr16[7] 1457 ); 1458 break; 1459 default: 1460 kprintf("AF%d ", sa->sa_family); 1461 while (len > 0 && sa->sa_data[len-1] == 0) 1462 --len; 1463 1464 for (i = 0; i < len; ++i) { 1465 if (i) 1466 kprintf("."); 1467 kprintf("%d", (unsigned char)sa->sa_data[i]); 1468 } 1469 break; 1470 } 1471 } 1472 } 1473 1474 #endif 1475 1476 /* 1477 * Set up a routing table entry, normally for an interface. 1478 */ 1479 int 1480 rtinit(struct ifaddr *ifa, int cmd, int flags) 1481 { 1482 struct sockaddr *dst, *deldst, *netmask; 1483 struct mbuf *m = NULL; 1484 struct radix_node_head *rnh; 1485 struct radix_node *rn; 1486 struct rt_addrinfo rtinfo; 1487 int error; 1488 1489 ASSERT_NETISR0; 1490 1491 if (flags & RTF_HOST) { 1492 dst = ifa->ifa_dstaddr; 1493 netmask = NULL; 1494 } else { 1495 dst = ifa->ifa_addr; 1496 netmask = ifa->ifa_netmask; 1497 } 1498 /* 1499 * If it's a delete, check that if it exists, it's on the correct 1500 * interface or we might scrub a route to another ifa which would 1501 * be confusing at best and possibly worse. 1502 */ 1503 if (cmd == RTM_DELETE) { 1504 /* 1505 * It's a delete, so it should already exist.. 1506 * If it's a net, mask off the host bits 1507 * (Assuming we have a mask) 1508 */ 1509 if (netmask != NULL) { 1510 m = m_get(M_NOWAIT, MT_SONAME); 1511 if (m == NULL) 1512 return (ENOBUFS); 1513 mbuftrackid(m, 34); 1514 deldst = mtod(m, struct sockaddr *); 1515 rt_maskedcopy(dst, deldst, netmask); 1516 dst = deldst; 1517 } 1518 /* 1519 * Look up an rtentry that is in the routing tree and 1520 * contains the correct info. 1521 */ 1522 if ((rnh = rt_tables[mycpuid][dst->sa_family]) == NULL || 1523 (rn = rnh->rnh_lookup((char *)dst, 1524 (char *)netmask, rnh)) == NULL || 1525 ((struct rtentry *)rn)->rt_ifa != ifa || 1526 !sa_equal((struct sockaddr *)rn->rn_key, dst)) { 1527 if (m != NULL) 1528 m_free(m); 1529 return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH); 1530 } 1531 /* XXX */ 1532 #if 0 1533 else { 1534 /* 1535 * One would think that as we are deleting, and we know 1536 * it doesn't exist, we could just return at this point 1537 * with an "ELSE" clause, but apparently not.. 1538 */ 1539 return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH); 1540 } 1541 #endif 1542 } 1543 /* 1544 * Do the actual request 1545 */ 1546 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 1547 rtinfo.rti_info[RTAX_DST] = dst; 1548 rtinfo.rti_info[RTAX_GATEWAY] = ifa->ifa_addr; 1549 rtinfo.rti_info[RTAX_NETMASK] = netmask; 1550 rtinfo.rti_flags = flags | ifa->ifa_flags; 1551 rtinfo.rti_ifa = ifa; 1552 error = rtrequest1_global(cmd, &rtinfo, rtinit_rtrequest_callback, ifa, 1553 RTREQ_PRIO_HIGH); 1554 if (m != NULL) 1555 m_free(m); 1556 return (error); 1557 } 1558 1559 static void 1560 rtinit_rtrequest_callback(int cmd, int error, 1561 struct rt_addrinfo *rtinfo, struct rtentry *rt, 1562 void *arg) 1563 { 1564 struct ifaddr *ifa = arg; 1565 1566 if (error == 0 && rt) { 1567 if (mycpuid == 0) { 1568 ++rt->rt_refcnt; 1569 rt_newaddrmsg(cmd, ifa, error, rt); 1570 --rt->rt_refcnt; 1571 } 1572 if (cmd == RTM_DELETE) { 1573 if (rt->rt_refcnt == 0) { 1574 ++rt->rt_refcnt; 1575 rtfree(rt); 1576 } 1577 } 1578 } 1579 } 1580 1581 struct netmsg_rts { 1582 struct netmsg_base base; 1583 int req; 1584 struct rt_addrinfo *rtinfo; 1585 rtsearch_callback_func_t callback; 1586 void *arg; 1587 boolean_t exact_match; 1588 int found_cnt; 1589 }; 1590 1591 int 1592 rtsearch_global(int req, struct rt_addrinfo *rtinfo, 1593 rtsearch_callback_func_t callback, void *arg, boolean_t exact_match, 1594 boolean_t req_prio) 1595 { 1596 struct netmsg_rts msg; 1597 int flags = 0; 1598 1599 if (req_prio) 1600 flags = MSGF_PRIORITY; 1601 netmsg_init(&msg.base, NULL, &curthread->td_msgport, flags, 1602 rtsearch_msghandler); 1603 msg.req = req; 1604 msg.rtinfo = rtinfo; 1605 msg.callback = callback; 1606 msg.arg = arg; 1607 msg.exact_match = exact_match; 1608 msg.found_cnt = 0; 1609 return (netisr_domsg_global(&msg.base)); 1610 } 1611 1612 static void 1613 rtsearch_msghandler(netmsg_t msg) 1614 { 1615 struct netmsg_rts *rmsg = (void *)msg; 1616 struct rt_addrinfo rtinfo; 1617 struct radix_node_head *rnh; 1618 struct rtentry *rt; 1619 int error; 1620 1621 ASSERT_NETISR_NCPUS(mycpuid); 1622 1623 /* 1624 * Copy the rtinfo. We need to make sure that the original 1625 * rtinfo, which is setup by the caller, in the netmsg will 1626 * _not_ be changed; else the next CPU on the netmsg forwarding 1627 * path will see a different rtinfo than what this CPU has seen. 1628 */ 1629 rtinfo = *rmsg->rtinfo; 1630 1631 /* 1632 * Find the correct routing tree to use for this Address Family 1633 */ 1634 if ((rnh = rt_tables[mycpuid][rtinfo.rti_dst->sa_family]) == NULL) { 1635 if (mycpuid != 0) 1636 panic("partially initialized routing tables"); 1637 netisr_replymsg(&rmsg->base, EAFNOSUPPORT); 1638 return; 1639 } 1640 1641 /* 1642 * Correct rtinfo for the host route searching. 1643 */ 1644 if (rtinfo.rti_flags & RTF_HOST) { 1645 rtinfo.rti_netmask = NULL; 1646 rtinfo.rti_flags &= ~(RTF_CLONING | RTF_PRCLONING); 1647 } 1648 1649 rt = (struct rtentry *) 1650 rnh->rnh_lookup((char *)rtinfo.rti_dst, 1651 (char *)rtinfo.rti_netmask, rnh); 1652 1653 /* 1654 * If we are asked to do the "exact match", we need to make sure 1655 * that host route searching got a host route while a network 1656 * route searching got a network route. 1657 */ 1658 if (rt != NULL && rmsg->exact_match && 1659 ((rt->rt_flags ^ rtinfo.rti_flags) & RTF_HOST)) 1660 rt = NULL; 1661 1662 if (rt == NULL) { 1663 /* 1664 * No matching routes have been found, don't count this 1665 * as a critical error (here, we set 'error' to 0), just 1666 * keep moving on, since at least prcloned routes are not 1667 * duplicated onto each CPU. 1668 */ 1669 error = 0; 1670 } else { 1671 rmsg->found_cnt++; 1672 1673 rt->rt_refcnt++; 1674 error = rmsg->callback(rmsg->req, &rtinfo, rt, rmsg->arg, 1675 rmsg->found_cnt); 1676 rt->rt_refcnt--; 1677 1678 if (error == EJUSTRETURN) { 1679 netisr_replymsg(&rmsg->base, 0); 1680 return; 1681 } 1682 } 1683 1684 if (error) { 1685 KKASSERT(rmsg->found_cnt > 0); 1686 1687 /* 1688 * Under following cases, unrecoverable error has 1689 * not occured: 1690 * o Request is RTM_GET 1691 * o The first time that we find the route, but the 1692 * modification fails. 1693 */ 1694 if (rmsg->req != RTM_GET && rmsg->found_cnt > 1) { 1695 panic("rtsearch_msghandler: unrecoverable error " 1696 "cpu %d", mycpuid); 1697 } 1698 netisr_replymsg(&rmsg->base, error); 1699 } else { 1700 if (rmsg->found_cnt == 0) { 1701 /* The requested route has not been seen ... */ 1702 error = ESRCH; 1703 } 1704 netisr_forwardmsg_error(&rmsg->base, mycpuid + 1, error); 1705 } 1706 } 1707 1708 int 1709 rtmask_add_global(struct sockaddr *mask, boolean_t req_prio) 1710 { 1711 struct netmsg_base msg; 1712 int flags = 0; 1713 1714 if (req_prio) 1715 flags = MSGF_PRIORITY; 1716 netmsg_init(&msg, NULL, &curthread->td_msgport, flags, 1717 rtmask_add_msghandler); 1718 msg.lmsg.u.ms_resultp = mask; 1719 1720 return (netisr_domsg_global(&msg)); 1721 } 1722 1723 struct sockaddr * 1724 _rtmask_lookup(struct sockaddr *mask, boolean_t search) 1725 { 1726 struct radix_node *n; 1727 1728 #define clen(s) (*(u_char *)(s)) 1729 n = rn_addmask((char *)mask, search, 1, rn_cpumaskhead(mycpuid)); 1730 if (n != NULL && 1731 mask->sa_len >= clen(n->rn_key) && 1732 bcmp((char *)mask + 1, 1733 (char *)n->rn_key + 1, clen(n->rn_key) - 1) == 0) { 1734 return (struct sockaddr *)n->rn_key; 1735 } else { 1736 return NULL; 1737 } 1738 #undef clen 1739 } 1740 1741 static void 1742 rtmask_add_msghandler(netmsg_t msg) 1743 { 1744 struct sockaddr *mask = msg->lmsg.u.ms_resultp; 1745 1746 ASSERT_NETISR_NCPUS(mycpuid); 1747 1748 if (rtmask_lookup(mask) == NULL) { 1749 netisr_replymsg(&msg->base, ENOBUFS); 1750 return; 1751 } 1752 netisr_forwardmsg(&msg->base, mycpuid + 1); 1753 } 1754 1755 /* This must be before ip6_init2(), which is now SI_ORDER_MIDDLE */ 1756 SYSINIT(route, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0); 1757 1758 struct rtchange_arg { 1759 struct ifaddr *old_ifa; 1760 struct ifaddr *new_ifa; 1761 struct rtentry *rt; 1762 int changed; 1763 }; 1764 1765 static void 1766 rtchange_ifa(struct rtentry *rt, struct rtchange_arg *ap) 1767 { 1768 if (rt->rt_ifa->ifa_rtrequest != NULL) 1769 rt->rt_ifa->ifa_rtrequest(RTM_DELETE, rt); 1770 IFAFREE(rt->rt_ifa); 1771 1772 IFAREF(ap->new_ifa); 1773 rt->rt_ifa = ap->new_ifa; 1774 rt->rt_ifp = ap->new_ifa->ifa_ifp; 1775 if (rt->rt_ifa->ifa_rtrequest != NULL) 1776 rt->rt_ifa->ifa_rtrequest(RTM_ADD, rt); 1777 1778 ap->changed = 1; 1779 } 1780 1781 static int 1782 rtchange_callback(struct radix_node *rn, void *xap) 1783 { 1784 struct rtchange_arg *ap = xap; 1785 struct rtentry *rt = (struct rtentry *)rn; 1786 1787 if (rt->rt_ifa == ap->old_ifa) { 1788 if (rt->rt_flags & (RTF_CLONING | RTF_PRCLONING)) { 1789 /* 1790 * We could saw the branch off when we are 1791 * still sitting on it, if the ifa_rtrequest 1792 * DEL/ADD are called directly from here. 1793 */ 1794 ap->rt = rt; 1795 return EJUSTRETURN; 1796 } 1797 rtchange_ifa(rt, ap); 1798 } 1799 return 0; 1800 } 1801 1802 struct netmsg_rtchange { 1803 struct netmsg_base base; 1804 struct ifaddr *old_ifa; 1805 struct ifaddr *new_ifa; 1806 int changed; 1807 }; 1808 1809 static void 1810 rtchange_dispatch(netmsg_t msg) 1811 { 1812 struct netmsg_rtchange *rmsg = (void *)msg; 1813 struct radix_node_head *rnh; 1814 struct rtchange_arg arg; 1815 int cpu; 1816 1817 cpu = mycpuid; 1818 ASSERT_NETISR_NCPUS(cpu); 1819 1820 memset(&arg, 0, sizeof(arg)); 1821 arg.old_ifa = rmsg->old_ifa; 1822 arg.new_ifa = rmsg->new_ifa; 1823 1824 rnh = rt_tables[cpu][AF_INET]; 1825 for (;;) { 1826 int error; 1827 1828 KKASSERT(arg.rt == NULL); 1829 error = rnh->rnh_walktree(rnh, rtchange_callback, &arg); 1830 if (arg.rt != NULL) { 1831 struct rtentry *rt; 1832 1833 rt = arg.rt; 1834 arg.rt = NULL; 1835 rtchange_ifa(rt, &arg); 1836 } else { 1837 break; 1838 } 1839 } 1840 if (arg.changed) 1841 rmsg->changed = 1; 1842 1843 netisr_forwardmsg(&rmsg->base, cpu + 1); 1844 } 1845 1846 int 1847 rtchange(struct ifaddr *old_ifa, struct ifaddr *new_ifa) 1848 { 1849 struct netmsg_rtchange msg; 1850 1851 /* 1852 * XXX individual requests are not independantly chained, 1853 * which means that the per-cpu route tables will not be 1854 * consistent in the middle of the operation. If routes 1855 * related to the interface are manipulated while we are 1856 * doing this the inconsistancy could trigger a panic. 1857 */ 1858 netmsg_init(&msg.base, NULL, &curthread->td_msgport, MSGF_PRIORITY, 1859 rtchange_dispatch); 1860 msg.old_ifa = old_ifa; 1861 msg.new_ifa = new_ifa; 1862 msg.changed = 0; 1863 netisr_domsg_global(&msg.base); 1864 1865 if (msg.changed) { 1866 old_ifa->ifa_flags &= ~IFA_ROUTE; 1867 new_ifa->ifa_flags |= IFA_ROUTE; 1868 return 0; 1869 } else { 1870 return ENOENT; 1871 } 1872 } 1873