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 case RTM_GET: 1006 /* Get the item from the tree. */ 1007 rn = rnh->rnh_lookup((char *)rtinfo->rti_info[RTAX_DST], 1008 (char *)rtinfo->rti_info[RTAX_NETMASK], 1009 rnh); 1010 if (rn == NULL) 1011 gotoerr(ESRCH); 1012 if (ret_nrt != NULL) { 1013 rt = (struct rtentry *)rn; 1014 rt->rt_refcnt++; 1015 *ret_nrt = rt; 1016 } 1017 break; 1018 default: 1019 error = EOPNOTSUPP; 1020 } 1021 bad: 1022 #ifdef ROUTE_DEBUG 1023 if (route_debug) { 1024 if (error) 1025 kprintf("rti %p failed error %d\n", rtinfo, error); 1026 else 1027 kprintf("rti %p succeeded\n", rtinfo); 1028 } 1029 #endif 1030 crit_exit(); 1031 return (error); 1032 } 1033 1034 /* 1035 * Called from rtrequest(RTM_DELETE, ...) to fix up the route's ``family'' 1036 * (i.e., the routes related to it by the operation of cloning). This 1037 * routine is iterated over all potential former-child-routes by way of 1038 * rnh->rnh_walktree_from() above, and those that actually are children of 1039 * the late parent (passed in as VP here) are themselves deleted. 1040 */ 1041 static int 1042 rt_fixdelete(struct radix_node *rn, void *vp) 1043 { 1044 struct rtentry *rt = (struct rtentry *)rn; 1045 struct rtentry *rt0 = vp; 1046 1047 if (rt->rt_parent == rt0 && 1048 !(rt->rt_flags & (RTF_PINNED | RTF_CLONING | RTF_PRCLONING))) { 1049 return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 1050 rt->rt_flags, NULL); 1051 } 1052 return 0; 1053 } 1054 1055 /* 1056 * This routine is called from rt_setgate() to do the analogous thing for 1057 * adds and changes. There is the added complication in this case of a 1058 * middle insert; i.e., insertion of a new network route between an older 1059 * network route and (cloned) host routes. For this reason, a simple check 1060 * of rt->rt_parent is insufficient; each candidate route must be tested 1061 * against the (mask, value) of the new route (passed as before in vp) 1062 * to see if the new route matches it. 1063 * 1064 * XXX - it may be possible to do fixdelete() for changes and reserve this 1065 * routine just for adds. I'm not sure why I thought it was necessary to do 1066 * changes this way. 1067 */ 1068 #ifdef DEBUG 1069 static int rtfcdebug = 0; 1070 #endif 1071 1072 static int 1073 rt_fixchange(struct radix_node *rn, void *vp) 1074 { 1075 struct rtentry *rt = (struct rtentry *)rn; 1076 struct rtfc_arg *ap = vp; 1077 struct rtentry *rt0 = ap->rt0; 1078 struct radix_node_head *rnh = ap->rnh; 1079 u_char *xk1, *xm1, *xk2, *xmp; 1080 int i, len, mlen; 1081 1082 #ifdef DEBUG 1083 if (rtfcdebug) 1084 kprintf("rt_fixchange: rt %p, rt0 %p\n", rt, rt0); 1085 #endif 1086 1087 if (rt->rt_parent == NULL || 1088 (rt->rt_flags & (RTF_PINNED | RTF_CLONING | RTF_PRCLONING))) { 1089 #ifdef DEBUG 1090 if (rtfcdebug) kprintf("no parent, pinned or cloning\n"); 1091 #endif 1092 return 0; 1093 } 1094 1095 if (rt->rt_parent == rt0) { 1096 #ifdef DEBUG 1097 if (rtfcdebug) kprintf("parent match\n"); 1098 #endif 1099 return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 1100 rt->rt_flags, NULL); 1101 } 1102 1103 /* 1104 * There probably is a function somewhere which does this... 1105 * if not, there should be. 1106 */ 1107 len = imin(rt_key(rt0)->sa_len, rt_key(rt)->sa_len); 1108 1109 xk1 = (u_char *)rt_key(rt0); 1110 xm1 = (u_char *)rt_mask(rt0); 1111 xk2 = (u_char *)rt_key(rt); 1112 1113 /* avoid applying a less specific route */ 1114 xmp = (u_char *)rt_mask(rt->rt_parent); 1115 mlen = rt_key(rt->rt_parent)->sa_len; 1116 if (mlen > rt_key(rt0)->sa_len) { 1117 #ifdef DEBUG 1118 if (rtfcdebug) 1119 kprintf("rt_fixchange: inserting a less " 1120 "specific route\n"); 1121 #endif 1122 return 0; 1123 } 1124 for (i = rnh->rnh_treetop->rn_offset; i < mlen; i++) { 1125 if ((xmp[i] & ~(xmp[i] ^ xm1[i])) != xmp[i]) { 1126 #ifdef DEBUG 1127 if (rtfcdebug) 1128 kprintf("rt_fixchange: inserting a less " 1129 "specific route\n"); 1130 #endif 1131 return 0; 1132 } 1133 } 1134 1135 for (i = rnh->rnh_treetop->rn_offset; i < len; i++) { 1136 if ((xk2[i] & xm1[i]) != xk1[i]) { 1137 #ifdef DEBUG 1138 if (rtfcdebug) kprintf("no match\n"); 1139 #endif 1140 return 0; 1141 } 1142 } 1143 1144 /* 1145 * OK, this node is a clone, and matches the node currently being 1146 * changed/added under the node's mask. So, get rid of it. 1147 */ 1148 #ifdef DEBUG 1149 if (rtfcdebug) kprintf("deleting\n"); 1150 #endif 1151 return rtrequest(RTM_DELETE, rt_key(rt), NULL, rt_mask(rt), 1152 rt->rt_flags, NULL); 1153 } 1154 1155 int 1156 rt_setgate(struct rtentry *rt0, struct sockaddr *dst, struct sockaddr *gate, 1157 boolean_t generate_report) 1158 { 1159 char *space, *oldspace; 1160 int dlen = RT_ROUNDUP(dst->sa_len), glen = RT_ROUNDUP(gate->sa_len); 1161 struct rtentry *rt = rt0; 1162 struct radix_node_head *rnh = rt_tables[mycpuid][dst->sa_family]; 1163 1164 ASSERT_NETISR_NCPUS(mycpuid); 1165 1166 /* 1167 * A host route with the destination equal to the gateway 1168 * will interfere with keeping LLINFO in the routing 1169 * table, so disallow it. 1170 */ 1171 if (((rt0->rt_flags & (RTF_HOST | RTF_GATEWAY | RTF_LLINFO)) == 1172 (RTF_HOST | RTF_GATEWAY)) && 1173 dst->sa_len == gate->sa_len && 1174 sa_equal(dst, gate)) { 1175 /* 1176 * The route might already exist if this is an RTM_CHANGE 1177 * or a routing redirect, so try to delete it. 1178 */ 1179 if (rt_key(rt0) != NULL) 1180 rtrequest(RTM_DELETE, rt_key(rt0), rt0->rt_gateway, 1181 rt_mask(rt0), rt0->rt_flags, NULL); 1182 return EADDRNOTAVAIL; 1183 } 1184 1185 /* 1186 * Both dst and gateway are stored in the same malloc'ed chunk 1187 * (If I ever get my hands on....) 1188 * if we need to malloc a new chunk, then keep the old one around 1189 * till we don't need it any more. 1190 */ 1191 if (rt->rt_gateway == NULL || 1192 glen > RT_ROUNDUP(rt->rt_gateway->sa_len)) { 1193 oldspace = (char *)rt_key(rt); 1194 R_Malloc(space, char *, dlen + glen); 1195 if (space == NULL) 1196 return ENOBUFS; 1197 rt->rt_nodes->rn_key = space; 1198 } else { 1199 space = (char *)rt_key(rt); /* Just use the old space. */ 1200 oldspace = NULL; 1201 } 1202 1203 /* Set the gateway value. */ 1204 rt->rt_gateway = (struct sockaddr *)(space + dlen); 1205 bcopy(gate, rt->rt_gateway, glen); 1206 1207 if (oldspace != NULL) { 1208 /* 1209 * If we allocated a new chunk, preserve the original dst. 1210 * This way, rt_setgate() really just sets the gate 1211 * and leaves the dst field alone. 1212 */ 1213 bcopy(dst, space, dlen); 1214 Free(oldspace); 1215 } 1216 1217 /* 1218 * If there is already a gwroute, it's now almost definitely wrong 1219 * so drop it. 1220 */ 1221 if (rt->rt_gwroute != NULL) { 1222 RTFREE(rt->rt_gwroute); 1223 rt->rt_gwroute = NULL; 1224 } 1225 if (rt->rt_flags & RTF_GATEWAY) { 1226 /* 1227 * Cloning loop avoidance: In the presence of 1228 * protocol-cloning and bad configuration, it is 1229 * possible to get stuck in bottomless mutual recursion 1230 * (rtrequest rt_setgate rtlookup). We avoid this 1231 * by not allowing protocol-cloning to operate for 1232 * gateways (which is probably the correct choice 1233 * anyway), and avoid the resulting reference loops 1234 * by disallowing any route to run through itself as 1235 * a gateway. This is obviously mandatory when we 1236 * get rt->rt_output(). 1237 * 1238 * This breaks TTCP for hosts outside the gateway! XXX JH 1239 */ 1240 rt->rt_gwroute = _rtlookup(gate, generate_report, 1241 RTF_PRCLONING); 1242 if (rt->rt_gwroute == rt) { 1243 rt->rt_gwroute = NULL; 1244 --rt->rt_refcnt; 1245 return EDQUOT; /* failure */ 1246 } 1247 } 1248 1249 /* 1250 * This isn't going to do anything useful for host routes, so 1251 * don't bother. Also make sure we have a reasonable mask 1252 * (we don't yet have one during adds). 1253 */ 1254 if (!(rt->rt_flags & RTF_HOST) && rt_mask(rt) != NULL) { 1255 struct rtfc_arg arg = { rt, rnh }; 1256 1257 rnh->rnh_walktree_from(rnh, (char *)rt_key(rt), 1258 (char *)rt_mask(rt), 1259 rt_fixchange, &arg); 1260 } 1261 1262 return 0; 1263 } 1264 1265 static void 1266 rt_maskedcopy( 1267 struct sockaddr *src, 1268 struct sockaddr *dst, 1269 struct sockaddr *netmask) 1270 { 1271 u_char *cp1 = (u_char *)src; 1272 u_char *cp2 = (u_char *)dst; 1273 u_char *cp3 = (u_char *)netmask; 1274 u_char *cplim = cp2 + *cp3; 1275 u_char *cplim2 = cp2 + *cp1; 1276 1277 *cp2++ = *cp1++; *cp2++ = *cp1++; /* copies sa_len & sa_family */ 1278 cp3 += 2; 1279 if (cplim > cplim2) 1280 cplim = cplim2; 1281 while (cp2 < cplim) 1282 *cp2++ = *cp1++ & *cp3++; 1283 if (cp2 < cplim2) 1284 bzero(cp2, cplim2 - cp2); 1285 } 1286 1287 int 1288 rt_llroute(struct sockaddr *dst, struct rtentry *rt0, struct rtentry **drt) 1289 { 1290 struct rtentry *up_rt, *rt; 1291 1292 ASSERT_NETISR_NCPUS(mycpuid); 1293 1294 if (!(rt0->rt_flags & RTF_UP)) { 1295 up_rt = rtlookup(dst); 1296 if (up_rt == NULL) 1297 return (EHOSTUNREACH); 1298 up_rt->rt_refcnt--; 1299 } else 1300 up_rt = rt0; 1301 if (up_rt->rt_flags & RTF_GATEWAY) { 1302 if (up_rt->rt_gwroute == NULL) { 1303 up_rt->rt_gwroute = rtlookup(up_rt->rt_gateway); 1304 if (up_rt->rt_gwroute == NULL) 1305 return (EHOSTUNREACH); 1306 } else if (!(up_rt->rt_gwroute->rt_flags & RTF_UP)) { 1307 rtfree(up_rt->rt_gwroute); 1308 up_rt->rt_gwroute = rtlookup(up_rt->rt_gateway); 1309 if (up_rt->rt_gwroute == NULL) 1310 return (EHOSTUNREACH); 1311 } 1312 rt = up_rt->rt_gwroute; 1313 } else 1314 rt = up_rt; 1315 if (rt->rt_flags & RTF_REJECT && 1316 (rt->rt_rmx.rmx_expire == 0 || /* rt doesn't expire */ 1317 time_uptime < rt->rt_rmx.rmx_expire)) /* rt not expired */ 1318 return (rt->rt_flags & RTF_HOST ? EHOSTDOWN : EHOSTUNREACH); 1319 *drt = rt; 1320 return 0; 1321 } 1322 1323 static int 1324 rt_setshims(struct rtentry *rt, struct sockaddr **rt_shim){ 1325 int i; 1326 1327 for (i=0; i<3; i++) { 1328 struct sockaddr *shim = rt_shim[RTAX_MPLS1 + i]; 1329 int shimlen; 1330 1331 if (shim == NULL) 1332 break; 1333 1334 shimlen = RT_ROUNDUP(shim->sa_len); 1335 R_Malloc(rt->rt_shim[i], struct sockaddr *, shimlen); 1336 bcopy(shim, rt->rt_shim[i], shimlen); 1337 } 1338 1339 return 0; 1340 } 1341 1342 #ifdef ROUTE_DEBUG 1343 1344 /* 1345 * Print out a route table entry 1346 */ 1347 void 1348 rt_print(struct rt_addrinfo *rtinfo, struct rtentry *rn) 1349 { 1350 kprintf("rti %p cpu %d route %p flags %08lx: ", 1351 rtinfo, mycpuid, rn, rn->rt_flags); 1352 sockaddr_print(rt_key(rn)); 1353 kprintf(" mask "); 1354 sockaddr_print(rt_mask(rn)); 1355 kprintf(" gw "); 1356 sockaddr_print(rn->rt_gateway); 1357 kprintf(" ifc \"%s\"", rn->rt_ifp ? rn->rt_ifp->if_dname : "?"); 1358 kprintf(" ifa %p\n", rn->rt_ifa); 1359 } 1360 1361 void 1362 rt_addrinfo_print(int cmd, struct rt_addrinfo *rti) 1363 { 1364 int didit = 0; 1365 int i; 1366 1367 #ifdef ROUTE_DEBUG 1368 if (cmd == RTM_DELETE && route_debug > 1) 1369 print_backtrace(-1); 1370 #endif 1371 1372 switch(cmd) { 1373 case RTM_ADD: 1374 kprintf("ADD "); 1375 break; 1376 case RTM_RESOLVE: 1377 kprintf("RES "); 1378 break; 1379 case RTM_DELETE: 1380 kprintf("DEL "); 1381 break; 1382 default: 1383 kprintf("C%02d ", cmd); 1384 break; 1385 } 1386 kprintf("rti %p cpu %d ", rti, mycpuid); 1387 for (i = 0; i < rti->rti_addrs; ++i) { 1388 if (rti->rti_info[i] == NULL) 1389 continue; 1390 if (didit) 1391 kprintf(" ,"); 1392 switch(i) { 1393 case RTAX_DST: 1394 kprintf("(DST "); 1395 break; 1396 case RTAX_GATEWAY: 1397 kprintf("(GWY "); 1398 break; 1399 case RTAX_NETMASK: 1400 kprintf("(MSK "); 1401 break; 1402 case RTAX_GENMASK: 1403 kprintf("(GEN "); 1404 break; 1405 case RTAX_IFP: 1406 kprintf("(IFP "); 1407 break; 1408 case RTAX_IFA: 1409 kprintf("(IFA "); 1410 break; 1411 case RTAX_AUTHOR: 1412 kprintf("(AUT "); 1413 break; 1414 case RTAX_BRD: 1415 kprintf("(BRD "); 1416 break; 1417 default: 1418 kprintf("(?%02d ", i); 1419 break; 1420 } 1421 sockaddr_print(rti->rti_info[i]); 1422 kprintf(")"); 1423 didit = 1; 1424 } 1425 kprintf("\n"); 1426 } 1427 1428 void 1429 sockaddr_print(struct sockaddr *sa) 1430 { 1431 struct sockaddr_in *sa4; 1432 struct sockaddr_in6 *sa6; 1433 int len; 1434 int i; 1435 1436 if (sa == NULL) { 1437 kprintf("NULL"); 1438 return; 1439 } 1440 1441 len = sa->sa_len - offsetof(struct sockaddr, sa_data[0]); 1442 1443 switch(sa->sa_family) { 1444 case AF_INET: 1445 case AF_INET6: 1446 default: 1447 switch(sa->sa_family) { 1448 case AF_INET: 1449 sa4 = (struct sockaddr_in *)sa; 1450 kprintf("INET %d %d.%d.%d.%d", 1451 ntohs(sa4->sin_port), 1452 (ntohl(sa4->sin_addr.s_addr) >> 24) & 255, 1453 (ntohl(sa4->sin_addr.s_addr) >> 16) & 255, 1454 (ntohl(sa4->sin_addr.s_addr) >> 8) & 255, 1455 (ntohl(sa4->sin_addr.s_addr) >> 0) & 255 1456 ); 1457 break; 1458 case AF_INET6: 1459 sa6 = (struct sockaddr_in6 *)sa; 1460 kprintf("INET6 %d %04x:%04x%04x:%04x:%04x:%04x:%04x:%04x", 1461 ntohs(sa6->sin6_port), 1462 sa6->sin6_addr.s6_addr16[0], 1463 sa6->sin6_addr.s6_addr16[1], 1464 sa6->sin6_addr.s6_addr16[2], 1465 sa6->sin6_addr.s6_addr16[3], 1466 sa6->sin6_addr.s6_addr16[4], 1467 sa6->sin6_addr.s6_addr16[5], 1468 sa6->sin6_addr.s6_addr16[6], 1469 sa6->sin6_addr.s6_addr16[7] 1470 ); 1471 break; 1472 default: 1473 kprintf("AF%d ", sa->sa_family); 1474 while (len > 0 && sa->sa_data[len-1] == 0) 1475 --len; 1476 1477 for (i = 0; i < len; ++i) { 1478 if (i) 1479 kprintf("."); 1480 kprintf("%d", (unsigned char)sa->sa_data[i]); 1481 } 1482 break; 1483 } 1484 } 1485 } 1486 1487 #endif 1488 1489 /* 1490 * Set up a routing table entry, normally for an interface. 1491 */ 1492 int 1493 rtinit(struct ifaddr *ifa, int cmd, int flags) 1494 { 1495 struct sockaddr *dst, *deldst, *netmask; 1496 struct mbuf *m = NULL; 1497 struct radix_node_head *rnh; 1498 struct radix_node *rn; 1499 struct rt_addrinfo rtinfo; 1500 int error; 1501 1502 ASSERT_NETISR0; 1503 1504 if (flags & RTF_HOST) { 1505 dst = ifa->ifa_dstaddr; 1506 netmask = NULL; 1507 } else { 1508 dst = ifa->ifa_addr; 1509 netmask = ifa->ifa_netmask; 1510 } 1511 /* 1512 * If it's a delete, check that if it exists, it's on the correct 1513 * interface or we might scrub a route to another ifa which would 1514 * be confusing at best and possibly worse. 1515 */ 1516 if (cmd == RTM_DELETE) { 1517 /* 1518 * It's a delete, so it should already exist.. 1519 * If it's a net, mask off the host bits 1520 * (Assuming we have a mask) 1521 */ 1522 if (netmask != NULL) { 1523 m = m_get(M_NOWAIT, MT_SONAME); 1524 if (m == NULL) 1525 return (ENOBUFS); 1526 mbuftrackid(m, 34); 1527 deldst = mtod(m, struct sockaddr *); 1528 rt_maskedcopy(dst, deldst, netmask); 1529 dst = deldst; 1530 } 1531 /* 1532 * Look up an rtentry that is in the routing tree and 1533 * contains the correct info. 1534 */ 1535 if ((rnh = rt_tables[mycpuid][dst->sa_family]) == NULL || 1536 (rn = rnh->rnh_lookup((char *)dst, 1537 (char *)netmask, rnh)) == NULL || 1538 ((struct rtentry *)rn)->rt_ifa != ifa || 1539 !sa_equal((struct sockaddr *)rn->rn_key, dst)) { 1540 if (m != NULL) 1541 m_free(m); 1542 return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH); 1543 } 1544 /* XXX */ 1545 #if 0 1546 else { 1547 /* 1548 * One would think that as we are deleting, and we know 1549 * it doesn't exist, we could just return at this point 1550 * with an "ELSE" clause, but apparently not.. 1551 */ 1552 return (flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH); 1553 } 1554 #endif 1555 } 1556 /* 1557 * Do the actual request 1558 */ 1559 bzero(&rtinfo, sizeof(struct rt_addrinfo)); 1560 rtinfo.rti_info[RTAX_DST] = dst; 1561 rtinfo.rti_info[RTAX_GATEWAY] = ifa->ifa_addr; 1562 rtinfo.rti_info[RTAX_NETMASK] = netmask; 1563 rtinfo.rti_flags = flags | ifa->ifa_flags; 1564 rtinfo.rti_ifa = ifa; 1565 error = rtrequest1_global(cmd, &rtinfo, rtinit_rtrequest_callback, ifa, 1566 RTREQ_PRIO_HIGH); 1567 if (m != NULL) 1568 m_free(m); 1569 return (error); 1570 } 1571 1572 static void 1573 rtinit_rtrequest_callback(int cmd, int error, 1574 struct rt_addrinfo *rtinfo, struct rtentry *rt, 1575 void *arg) 1576 { 1577 struct ifaddr *ifa = arg; 1578 1579 if (error == 0 && rt) { 1580 if (mycpuid == 0) 1581 rt_newaddrmsg(cmd, ifa, error, rt); 1582 if (cmd == RTM_DELETE) { 1583 if (rt->rt_refcnt == 0) { 1584 ++rt->rt_refcnt; 1585 rtfree(rt); 1586 } 1587 } 1588 } 1589 } 1590 1591 struct netmsg_rts { 1592 struct netmsg_base base; 1593 int req; 1594 struct rt_addrinfo *rtinfo; 1595 rtsearch_callback_func_t callback; 1596 void *arg; 1597 boolean_t exact_match; 1598 int found_cnt; 1599 }; 1600 1601 int 1602 rtsearch_global(int req, struct rt_addrinfo *rtinfo, 1603 rtsearch_callback_func_t callback, void *arg, boolean_t exact_match, 1604 boolean_t req_prio) 1605 { 1606 struct netmsg_rts msg; 1607 int flags = 0; 1608 1609 if (req_prio) 1610 flags = MSGF_PRIORITY; 1611 netmsg_init(&msg.base, NULL, &curthread->td_msgport, flags, 1612 rtsearch_msghandler); 1613 msg.req = req; 1614 msg.rtinfo = rtinfo; 1615 msg.callback = callback; 1616 msg.arg = arg; 1617 msg.exact_match = exact_match; 1618 msg.found_cnt = 0; 1619 return (netisr_domsg_global(&msg.base)); 1620 } 1621 1622 static void 1623 rtsearch_msghandler(netmsg_t msg) 1624 { 1625 struct netmsg_rts *rmsg = (void *)msg; 1626 struct rt_addrinfo rtinfo; 1627 struct radix_node_head *rnh; 1628 struct rtentry *rt; 1629 int error; 1630 1631 ASSERT_NETISR_NCPUS(mycpuid); 1632 1633 /* 1634 * Copy the rtinfo. We need to make sure that the original 1635 * rtinfo, which is setup by the caller, in the netmsg will 1636 * _not_ be changed; else the next CPU on the netmsg forwarding 1637 * path will see a different rtinfo than what this CPU has seen. 1638 */ 1639 rtinfo = *rmsg->rtinfo; 1640 1641 /* 1642 * Find the correct routing tree to use for this Address Family 1643 */ 1644 if ((rnh = rt_tables[mycpuid][rtinfo.rti_dst->sa_family]) == NULL) { 1645 if (mycpuid != 0) 1646 panic("partially initialized routing tables"); 1647 netisr_replymsg(&rmsg->base, EAFNOSUPPORT); 1648 return; 1649 } 1650 1651 /* 1652 * Correct rtinfo for the host route searching. 1653 */ 1654 if (rtinfo.rti_flags & RTF_HOST) { 1655 rtinfo.rti_netmask = NULL; 1656 rtinfo.rti_flags &= ~(RTF_CLONING | RTF_PRCLONING); 1657 } 1658 1659 rt = (struct rtentry *) 1660 rnh->rnh_lookup((char *)rtinfo.rti_dst, 1661 (char *)rtinfo.rti_netmask, rnh); 1662 1663 /* 1664 * If we are asked to do the "exact match", we need to make sure 1665 * that host route searching got a host route while a network 1666 * route searching got a network route. 1667 */ 1668 if (rt != NULL && rmsg->exact_match && 1669 ((rt->rt_flags ^ rtinfo.rti_flags) & RTF_HOST)) 1670 rt = NULL; 1671 1672 if (rt == NULL) { 1673 /* 1674 * No matching routes have been found, don't count this 1675 * as a critical error (here, we set 'error' to 0), just 1676 * keep moving on, since at least prcloned routes are not 1677 * duplicated onto each CPU. 1678 */ 1679 error = 0; 1680 } else { 1681 rmsg->found_cnt++; 1682 1683 rt->rt_refcnt++; 1684 error = rmsg->callback(rmsg->req, &rtinfo, rt, rmsg->arg, 1685 rmsg->found_cnt); 1686 rt->rt_refcnt--; 1687 1688 if (error == EJUSTRETURN) { 1689 netisr_replymsg(&rmsg->base, 0); 1690 return; 1691 } 1692 } 1693 1694 if (error) { 1695 KKASSERT(rmsg->found_cnt > 0); 1696 1697 /* 1698 * Under following cases, unrecoverable error has 1699 * not occured: 1700 * o Request is RTM_GET 1701 * o The first time that we find the route, but the 1702 * modification fails. 1703 */ 1704 if (rmsg->req != RTM_GET && rmsg->found_cnt > 1) { 1705 panic("rtsearch_msghandler: unrecoverable error " 1706 "cpu %d", mycpuid); 1707 } 1708 netisr_replymsg(&rmsg->base, error); 1709 } else { 1710 if (rmsg->found_cnt == 0) { 1711 /* The requested route has not been seen ... */ 1712 error = ESRCH; 1713 } 1714 netisr_forwardmsg_error(&rmsg->base, mycpuid + 1, error); 1715 } 1716 } 1717 1718 int 1719 rtmask_add_global(struct sockaddr *mask, boolean_t req_prio) 1720 { 1721 struct netmsg_base msg; 1722 int flags = 0; 1723 1724 if (req_prio) 1725 flags = MSGF_PRIORITY; 1726 netmsg_init(&msg, NULL, &curthread->td_msgport, flags, 1727 rtmask_add_msghandler); 1728 msg.lmsg.u.ms_resultp = mask; 1729 1730 return (netisr_domsg_global(&msg)); 1731 } 1732 1733 struct sockaddr * 1734 _rtmask_lookup(struct sockaddr *mask, boolean_t search) 1735 { 1736 struct radix_node *n; 1737 1738 #define clen(s) (*(u_char *)(s)) 1739 n = rn_addmask((char *)mask, search, 1, rn_cpumaskhead(mycpuid)); 1740 if (n != NULL && 1741 mask->sa_len >= clen(n->rn_key) && 1742 bcmp((char *)mask + 1, 1743 (char *)n->rn_key + 1, clen(n->rn_key) - 1) == 0) { 1744 return (struct sockaddr *)n->rn_key; 1745 } else { 1746 return NULL; 1747 } 1748 #undef clen 1749 } 1750 1751 static void 1752 rtmask_add_msghandler(netmsg_t msg) 1753 { 1754 struct sockaddr *mask = msg->lmsg.u.ms_resultp; 1755 1756 ASSERT_NETISR_NCPUS(mycpuid); 1757 1758 if (rtmask_lookup(mask) == NULL) { 1759 netisr_replymsg(&msg->base, ENOBUFS); 1760 return; 1761 } 1762 netisr_forwardmsg(&msg->base, mycpuid + 1); 1763 } 1764 1765 /* This must be before ip6_init2(), which is now SI_ORDER_MIDDLE */ 1766 SYSINIT(route, SI_SUB_PROTO_DOMAIN, SI_ORDER_THIRD, route_init, 0); 1767 1768 struct rtchange_arg { 1769 struct ifaddr *old_ifa; 1770 struct ifaddr *new_ifa; 1771 struct rtentry *rt; 1772 int changed; 1773 }; 1774 1775 static void 1776 rtchange_ifa(struct rtentry *rt, struct rtchange_arg *ap) 1777 { 1778 if (rt->rt_ifa->ifa_rtrequest != NULL) 1779 rt->rt_ifa->ifa_rtrequest(RTM_DELETE, rt); 1780 IFAFREE(rt->rt_ifa); 1781 1782 IFAREF(ap->new_ifa); 1783 rt->rt_ifa = ap->new_ifa; 1784 rt->rt_ifp = ap->new_ifa->ifa_ifp; 1785 if (rt->rt_ifa->ifa_rtrequest != NULL) 1786 rt->rt_ifa->ifa_rtrequest(RTM_ADD, rt); 1787 1788 ap->changed = 1; 1789 } 1790 1791 static int 1792 rtchange_callback(struct radix_node *rn, void *xap) 1793 { 1794 struct rtchange_arg *ap = xap; 1795 struct rtentry *rt = (struct rtentry *)rn; 1796 1797 if (rt->rt_ifa == ap->old_ifa) { 1798 if (rt->rt_flags & (RTF_CLONING | RTF_PRCLONING)) { 1799 /* 1800 * We could saw the branch off when we are 1801 * still sitting on it, if the ifa_rtrequest 1802 * DEL/ADD are called directly from here. 1803 */ 1804 ap->rt = rt; 1805 return EJUSTRETURN; 1806 } 1807 rtchange_ifa(rt, ap); 1808 } 1809 return 0; 1810 } 1811 1812 struct netmsg_rtchange { 1813 struct netmsg_base base; 1814 struct ifaddr *old_ifa; 1815 struct ifaddr *new_ifa; 1816 int changed; 1817 }; 1818 1819 static void 1820 rtchange_dispatch(netmsg_t msg) 1821 { 1822 struct netmsg_rtchange *rmsg = (void *)msg; 1823 struct radix_node_head *rnh; 1824 struct rtchange_arg arg; 1825 int cpu; 1826 1827 cpu = mycpuid; 1828 ASSERT_NETISR_NCPUS(cpu); 1829 1830 memset(&arg, 0, sizeof(arg)); 1831 arg.old_ifa = rmsg->old_ifa; 1832 arg.new_ifa = rmsg->new_ifa; 1833 1834 rnh = rt_tables[cpu][AF_INET]; 1835 for (;;) { 1836 int error; 1837 1838 KKASSERT(arg.rt == NULL); 1839 error = rnh->rnh_walktree(rnh, rtchange_callback, &arg); 1840 if (arg.rt != NULL) { 1841 struct rtentry *rt; 1842 1843 rt = arg.rt; 1844 arg.rt = NULL; 1845 rtchange_ifa(rt, &arg); 1846 } else { 1847 break; 1848 } 1849 } 1850 if (arg.changed) 1851 rmsg->changed = 1; 1852 1853 netisr_forwardmsg(&rmsg->base, cpu + 1); 1854 } 1855 1856 int 1857 rtchange(struct ifaddr *old_ifa, struct ifaddr *new_ifa) 1858 { 1859 struct netmsg_rtchange msg; 1860 1861 /* 1862 * XXX individual requests are not independantly chained, 1863 * which means that the per-cpu route tables will not be 1864 * consistent in the middle of the operation. If routes 1865 * related to the interface are manipulated while we are 1866 * doing this the inconsistancy could trigger a panic. 1867 */ 1868 netmsg_init(&msg.base, NULL, &curthread->td_msgport, MSGF_PRIORITY, 1869 rtchange_dispatch); 1870 msg.old_ifa = old_ifa; 1871 msg.new_ifa = new_ifa; 1872 msg.changed = 0; 1873 netisr_domsg_global(&msg.base); 1874 1875 if (msg.changed) { 1876 old_ifa->ifa_flags &= ~IFA_ROUTE; 1877 new_ifa->ifa_flags |= IFA_ROUTE; 1878 return 0; 1879 } else { 1880 return ENOENT; 1881 } 1882 } 1883