1 /*- 2 * Copyright (c) 2002 Michael Shalayeff. 3 * Copyright (c) 2003 Ryan McBride. 4 * Copyright (c) 2011 Gleb Smirnoff <glebius@FreeBSD.org> 5 * All rights reserved. 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 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 17 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 18 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 19 * IN NO EVENT SHALL THE AUTHOR OR HIS RELATIVES BE LIABLE FOR ANY DIRECT, 20 * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES 21 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR 22 * SERVICES; LOSS OF MIND, USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, 24 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING 25 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF 26 * THE POSSIBILITY OF SUCH DAMAGE. 27 */ 28 29 #include <sys/cdefs.h> 30 __FBSDID("$FreeBSD$"); 31 32 #include "opt_bpf.h" 33 #include "opt_inet.h" 34 #include "opt_inet6.h" 35 36 #include <sys/param.h> 37 #include <sys/systm.h> 38 #include <sys/bus.h> 39 #include <sys/jail.h> 40 #include <sys/kernel.h> 41 #include <sys/limits.h> 42 #include <sys/malloc.h> 43 #include <sys/mbuf.h> 44 #include <sys/module.h> 45 #include <sys/priv.h> 46 #include <sys/proc.h> 47 #include <sys/protosw.h> 48 #include <sys/socket.h> 49 #include <sys/sockio.h> 50 #include <sys/sysctl.h> 51 #include <sys/syslog.h> 52 #include <sys/taskqueue.h> 53 54 #include <net/ethernet.h> 55 #include <net/fddi.h> 56 #include <net/if.h> 57 #include <net/if_dl.h> 58 #include <net/if_llatbl.h> 59 #include <net/if_types.h> 60 #include <net/iso88025.h> 61 #include <net/route.h> 62 #include <net/vnet.h> 63 64 #if defined(INET) || defined(INET6) 65 #include <netinet/in.h> 66 #include <netinet/in_var.h> 67 #include <netinet/ip_carp.h> 68 #include <netinet/ip.h> 69 #include <machine/in_cksum.h> 70 #endif 71 #ifdef INET 72 #include <netinet/ip_var.h> 73 #include <netinet/if_ether.h> 74 #endif 75 76 #ifdef INET6 77 #include <netinet/icmp6.h> 78 #include <netinet/ip6.h> 79 #include <netinet6/ip6protosw.h> 80 #include <netinet6/in6_var.h> 81 #include <netinet6/ip6_var.h> 82 #include <netinet6/scope6_var.h> 83 #include <netinet6/nd6.h> 84 #endif 85 86 #include <crypto/sha1.h> 87 88 static MALLOC_DEFINE(M_CARP, "CARP", "CARP addresses"); 89 90 struct carp_softc { 91 struct ifnet *sc_carpdev; /* Pointer to parent ifnet. */ 92 struct ifaddr **sc_ifas; /* Our ifaddrs. */ 93 struct sockaddr_dl sc_addr; /* Our link level address. */ 94 struct callout sc_ad_tmo; /* Advertising timeout. */ 95 #ifdef INET 96 struct callout sc_md_tmo; /* Master down timeout. */ 97 #endif 98 #ifdef INET6 99 struct callout sc_md6_tmo; /* XXX: Master down timeout. */ 100 #endif 101 struct mtx sc_mtx; 102 103 int sc_vhid; 104 int sc_advskew; 105 int sc_advbase; 106 107 int sc_naddrs; 108 int sc_naddrs6; 109 int sc_ifasiz; 110 enum { INIT = 0, BACKUP, MASTER } sc_state; 111 int sc_suppress; 112 int sc_sendad_errors; 113 #define CARP_SENDAD_MAX_ERRORS 3 114 int sc_sendad_success; 115 #define CARP_SENDAD_MIN_SUCCESS 3 116 117 int sc_init_counter; 118 uint64_t sc_counter; 119 120 /* authentication */ 121 #define CARP_HMAC_PAD 64 122 unsigned char sc_key[CARP_KEY_LEN]; 123 unsigned char sc_pad[CARP_HMAC_PAD]; 124 SHA1_CTX sc_sha1; 125 126 TAILQ_ENTRY(carp_softc) sc_list; /* On the carp_if list. */ 127 LIST_ENTRY(carp_softc) sc_next; /* On the global list. */ 128 }; 129 130 struct carp_if { 131 #ifdef INET 132 int cif_naddrs; 133 #endif 134 #ifdef INET6 135 int cif_naddrs6; 136 #endif 137 TAILQ_HEAD(, carp_softc) cif_vrs; 138 #ifdef INET 139 struct ip_moptions cif_imo; 140 #endif 141 #ifdef INET6 142 struct ip6_moptions cif_im6o; 143 #endif 144 struct ifnet *cif_ifp; 145 struct mtx cif_mtx; 146 }; 147 148 #define CARP_INET 0 149 #define CARP_INET6 1 150 static int proto_reg[] = {-1, -1}; 151 152 /* 153 * Brief design of carp(4). 154 * 155 * Any carp-capable ifnet may have a list of carp softcs hanging off 156 * its ifp->if_carp pointer. Each softc represents one unique virtual 157 * host id, or vhid. The softc has a back pointer to the ifnet. All 158 * softcs are joined in a global list, which has quite limited use. 159 * 160 * Any interface address that takes part in CARP negotiation has a 161 * pointer to the softc of its vhid, ifa->ifa_carp. That could be either 162 * AF_INET or AF_INET6 address. 163 * 164 * Although, one can get the softc's backpointer to ifnet and traverse 165 * through its ifp->if_addrhead queue to find all interface addresses 166 * involved in CARP, we keep a growable array of ifaddr pointers. This 167 * allows us to avoid grabbing the IF_ADDR_LOCK() in many traversals that 168 * do calls into the network stack, thus avoiding LORs. 169 * 170 * Locking: 171 * 172 * Each softc has a lock sc_mtx. It is used to synchronise carp_input_c(), 173 * callout-driven events and ioctl()s. 174 * 175 * To traverse the list of softcs on an ifnet we use CIF_LOCK(), to 176 * traverse the global list we use the mutex carp_mtx. 177 * 178 * Known issues with locking: 179 * 180 * - There is no protection for races between two ioctl() requests, 181 * neither SIOCSVH, nor SIOCAIFADDR & SIOCAIFADDR_IN6. I think that all 182 * interface ioctl()s should be serialized right in net/if.c. 183 * - Sending ad, we put the pointer to the softc in an mtag, and no reference 184 * counting is done on the softc. 185 * - On module unload we may race (?) with packet processing thread 186 * dereferencing our function pointers. 187 */ 188 189 static int carp_allow = 1; /* Accept incoming CARP packets. */ 190 static int carp_preempt = 0; /* Preempt slower nodes. */ 191 static int carp_log = 1; /* Log level. */ 192 static int carp_demotion = 0; /* Global advskew demotion. */ 193 static int carp_senderr_adj = CARP_MAXSKEW; /* Send error demotion factor */ 194 static int carp_ifdown_adj = CARP_MAXSKEW; /* Iface down demotion factor */ 195 static int carp_demote_adj_sysctl(SYSCTL_HANDLER_ARGS); 196 197 SYSCTL_NODE(_net_inet, IPPROTO_CARP, carp, CTLFLAG_RW, 0, "CARP"); 198 SYSCTL_INT(_net_inet_carp, OID_AUTO, allow, CTLFLAG_RW, &carp_allow, 0, 199 "Accept incoming CARP packets"); 200 SYSCTL_INT(_net_inet_carp, OID_AUTO, preempt, CTLFLAG_RW, &carp_preempt, 0, 201 "High-priority backup preemption mode"); 202 SYSCTL_INT(_net_inet_carp, OID_AUTO, log, CTLFLAG_RW, &carp_log, 0, 203 "CARP log level"); 204 SYSCTL_PROC(_net_inet_carp, OID_AUTO, demotion, CTLTYPE_INT|CTLFLAG_RW, 205 0, 0, carp_demote_adj_sysctl, "I", 206 "Adjust demotion factor (skew of advskew)"); 207 SYSCTL_INT(_net_inet_carp, OID_AUTO, senderr_demotion_factor, CTLFLAG_RW, 208 &carp_senderr_adj, 0, "Send error demotion factor adjustment"); 209 SYSCTL_INT(_net_inet_carp, OID_AUTO, ifdown_demotion_factor, CTLFLAG_RW, 210 &carp_ifdown_adj, 0, "Interface down demotion factor adjustment"); 211 212 static struct carpstats carpstats; 213 SYSCTL_STRUCT(_net_inet_carp, OID_AUTO, stats, CTLFLAG_RW, &carpstats, 214 carpstats, "CARP statistics (struct carpstats, netinet/ip_carp.h)"); 215 216 #define CARP_LOCK_INIT(sc) mtx_init(&(sc)->sc_mtx, "carp_softc", \ 217 NULL, MTX_DEF) 218 #define CARP_LOCK_DESTROY(sc) mtx_destroy(&(sc)->sc_mtx) 219 #define CARP_LOCK_ASSERT(sc) mtx_assert(&(sc)->sc_mtx, MA_OWNED) 220 #define CARP_LOCK(sc) mtx_lock(&(sc)->sc_mtx) 221 #define CARP_UNLOCK(sc) mtx_unlock(&(sc)->sc_mtx) 222 #define CIF_LOCK_INIT(cif) mtx_init(&(cif)->cif_mtx, "carp_if", \ 223 NULL, MTX_DEF) 224 #define CIF_LOCK_DESTROY(cif) mtx_destroy(&(cif)->cif_mtx) 225 #define CIF_LOCK_ASSERT(cif) mtx_assert(&(cif)->cif_mtx, MA_OWNED) 226 #define CIF_LOCK(cif) mtx_lock(&(cif)->cif_mtx) 227 #define CIF_UNLOCK(cif) mtx_unlock(&(cif)->cif_mtx) 228 #define CIF_FREE(cif) do { \ 229 CIF_LOCK_ASSERT(cif); \ 230 if (TAILQ_EMPTY(&(cif)->cif_vrs)) \ 231 carp_free_if(cif); \ 232 else \ 233 CIF_UNLOCK(cif); \ 234 } while (0) 235 236 #define CARP_LOG(...) do { \ 237 if (carp_log > 0) \ 238 log(LOG_INFO, "carp: " __VA_ARGS__); \ 239 } while (0) 240 241 #define CARP_DEBUG(...) do { \ 242 if (carp_log > 1) \ 243 log(LOG_DEBUG, __VA_ARGS__); \ 244 } while (0) 245 246 #define IFNET_FOREACH_IFA(ifp, ifa) \ 247 IF_ADDR_LOCK_ASSERT(ifp); \ 248 TAILQ_FOREACH((ifa), &(ifp)->if_addrhead, ifa_link) \ 249 if ((ifa)->ifa_carp != NULL) 250 251 #define CARP_FOREACH_IFA(sc, ifa) \ 252 CARP_LOCK_ASSERT(sc); \ 253 for (int _i = 0; \ 254 _i < (sc)->sc_naddrs + (sc)->sc_naddrs6 && \ 255 ((ifa) = sc->sc_ifas[_i]) != NULL; \ 256 ++_i) 257 258 #define IFNET_FOREACH_CARP(ifp, sc) \ 259 CIF_LOCK_ASSERT(ifp->if_carp); \ 260 TAILQ_FOREACH((sc), &(ifp)->if_carp->cif_vrs, sc_list) 261 262 #define DEMOTE_ADVSKEW(sc) \ 263 (((sc)->sc_advskew + carp_demotion > CARP_MAXSKEW) ? \ 264 CARP_MAXSKEW : ((sc)->sc_advskew + carp_demotion)) 265 266 static void carp_input_c(struct mbuf *, struct carp_header *, sa_family_t); 267 static struct carp_softc 268 *carp_alloc(struct ifnet *); 269 static void carp_detach_locked(struct ifaddr *); 270 static void carp_destroy(struct carp_softc *); 271 static struct carp_if 272 *carp_alloc_if(struct ifnet *); 273 static void carp_free_if(struct carp_if *); 274 static void carp_set_state(struct carp_softc *, int); 275 static void carp_sc_state(struct carp_softc *); 276 static void carp_setrun(struct carp_softc *, sa_family_t); 277 static void carp_master_down(void *); 278 static void carp_master_down_locked(struct carp_softc *); 279 static void carp_send_ad(void *); 280 static void carp_send_ad_locked(struct carp_softc *); 281 static void carp_addroute(struct carp_softc *); 282 static void carp_ifa_addroute(struct ifaddr *); 283 static void carp_delroute(struct carp_softc *); 284 static void carp_ifa_delroute(struct ifaddr *); 285 static void carp_send_ad_all(void *, int); 286 static void carp_demote_adj(int, char *); 287 288 static LIST_HEAD(, carp_softc) carp_list; 289 static struct mtx carp_mtx; 290 static struct task carp_sendall_task = 291 TASK_INITIALIZER(0, carp_send_ad_all, NULL); 292 293 static void 294 carp_hmac_prepare(struct carp_softc *sc) 295 { 296 uint8_t version = CARP_VERSION, type = CARP_ADVERTISEMENT; 297 uint8_t vhid = sc->sc_vhid & 0xff; 298 struct ifaddr *ifa; 299 int i, found; 300 #ifdef INET 301 struct in_addr last, cur, in; 302 #endif 303 #ifdef INET6 304 struct in6_addr last6, cur6, in6; 305 #endif 306 307 CARP_LOCK_ASSERT(sc); 308 309 /* Compute ipad from key. */ 310 bzero(sc->sc_pad, sizeof(sc->sc_pad)); 311 bcopy(sc->sc_key, sc->sc_pad, sizeof(sc->sc_key)); 312 for (i = 0; i < sizeof(sc->sc_pad); i++) 313 sc->sc_pad[i] ^= 0x36; 314 315 /* Precompute first part of inner hash. */ 316 SHA1Init(&sc->sc_sha1); 317 SHA1Update(&sc->sc_sha1, sc->sc_pad, sizeof(sc->sc_pad)); 318 SHA1Update(&sc->sc_sha1, (void *)&version, sizeof(version)); 319 SHA1Update(&sc->sc_sha1, (void *)&type, sizeof(type)); 320 SHA1Update(&sc->sc_sha1, (void *)&vhid, sizeof(vhid)); 321 #ifdef INET 322 cur.s_addr = 0; 323 do { 324 found = 0; 325 last = cur; 326 cur.s_addr = 0xffffffff; 327 CARP_FOREACH_IFA(sc, ifa) { 328 in.s_addr = ifatoia(ifa)->ia_addr.sin_addr.s_addr; 329 if (ifa->ifa_addr->sa_family == AF_INET && 330 ntohl(in.s_addr) > ntohl(last.s_addr) && 331 ntohl(in.s_addr) < ntohl(cur.s_addr)) { 332 cur.s_addr = in.s_addr; 333 found++; 334 } 335 } 336 if (found) 337 SHA1Update(&sc->sc_sha1, (void *)&cur, sizeof(cur)); 338 } while (found); 339 #endif /* INET */ 340 #ifdef INET6 341 memset(&cur6, 0, sizeof(cur6)); 342 do { 343 found = 0; 344 last6 = cur6; 345 memset(&cur6, 0xff, sizeof(cur6)); 346 CARP_FOREACH_IFA(sc, ifa) { 347 in6 = ifatoia6(ifa)->ia_addr.sin6_addr; 348 if (IN6_IS_SCOPE_EMBED(&in6)) 349 in6.s6_addr16[1] = 0; 350 if (ifa->ifa_addr->sa_family == AF_INET6 && 351 memcmp(&in6, &last6, sizeof(in6)) > 0 && 352 memcmp(&in6, &cur6, sizeof(in6)) < 0) { 353 cur6 = in6; 354 found++; 355 } 356 } 357 if (found) 358 SHA1Update(&sc->sc_sha1, (void *)&cur6, sizeof(cur6)); 359 } while (found); 360 #endif /* INET6 */ 361 362 /* convert ipad to opad */ 363 for (i = 0; i < sizeof(sc->sc_pad); i++) 364 sc->sc_pad[i] ^= 0x36 ^ 0x5c; 365 } 366 367 static void 368 carp_hmac_generate(struct carp_softc *sc, uint32_t counter[2], 369 unsigned char md[20]) 370 { 371 SHA1_CTX sha1ctx; 372 373 CARP_LOCK_ASSERT(sc); 374 375 /* fetch first half of inner hash */ 376 bcopy(&sc->sc_sha1, &sha1ctx, sizeof(sha1ctx)); 377 378 SHA1Update(&sha1ctx, (void *)counter, sizeof(sc->sc_counter)); 379 SHA1Final(md, &sha1ctx); 380 381 /* outer hash */ 382 SHA1Init(&sha1ctx); 383 SHA1Update(&sha1ctx, sc->sc_pad, sizeof(sc->sc_pad)); 384 SHA1Update(&sha1ctx, md, 20); 385 SHA1Final(md, &sha1ctx); 386 } 387 388 static int 389 carp_hmac_verify(struct carp_softc *sc, uint32_t counter[2], 390 unsigned char md[20]) 391 { 392 unsigned char md2[20]; 393 394 CARP_LOCK_ASSERT(sc); 395 396 carp_hmac_generate(sc, counter, md2); 397 398 return (bcmp(md, md2, sizeof(md2))); 399 } 400 401 /* 402 * process input packet. 403 * we have rearranged checks order compared to the rfc, 404 * but it seems more efficient this way or not possible otherwise. 405 */ 406 #ifdef INET 407 void 408 carp_input(struct mbuf *m, int hlen) 409 { 410 struct ip *ip = mtod(m, struct ip *); 411 struct carp_header *ch; 412 int iplen, len; 413 414 CARPSTATS_INC(carps_ipackets); 415 416 if (!carp_allow) { 417 m_freem(m); 418 return; 419 } 420 421 /* verify that the IP TTL is 255. */ 422 if (ip->ip_ttl != CARP_DFLTTL) { 423 CARPSTATS_INC(carps_badttl); 424 CARP_DEBUG("%s: received ttl %d != 255 on %s\n", __func__, 425 ip->ip_ttl, 426 m->m_pkthdr.rcvif->if_xname); 427 m_freem(m); 428 return; 429 } 430 431 iplen = ip->ip_hl << 2; 432 433 if (m->m_pkthdr.len < iplen + sizeof(*ch)) { 434 CARPSTATS_INC(carps_badlen); 435 CARP_DEBUG("%s: received len %zd < sizeof(struct carp_header) " 436 "on %s\n", __func__, m->m_len - sizeof(struct ip), 437 m->m_pkthdr.rcvif->if_xname); 438 m_freem(m); 439 return; 440 } 441 442 if (iplen + sizeof(*ch) < m->m_len) { 443 if ((m = m_pullup(m, iplen + sizeof(*ch))) == NULL) { 444 CARPSTATS_INC(carps_hdrops); 445 CARP_DEBUG("%s: pullup failed\n", __func__); 446 return; 447 } 448 ip = mtod(m, struct ip *); 449 } 450 ch = (struct carp_header *)((char *)ip + iplen); 451 452 /* 453 * verify that the received packet length is 454 * equal to the CARP header 455 */ 456 len = iplen + sizeof(*ch); 457 if (len > m->m_pkthdr.len) { 458 CARPSTATS_INC(carps_badlen); 459 CARP_DEBUG("%s: packet too short %d on %s\n", __func__, 460 m->m_pkthdr.len, 461 m->m_pkthdr.rcvif->if_xname); 462 m_freem(m); 463 return; 464 } 465 466 if ((m = m_pullup(m, len)) == NULL) { 467 CARPSTATS_INC(carps_hdrops); 468 return; 469 } 470 ip = mtod(m, struct ip *); 471 ch = (struct carp_header *)((char *)ip + iplen); 472 473 /* verify the CARP checksum */ 474 m->m_data += iplen; 475 if (in_cksum(m, len - iplen)) { 476 CARPSTATS_INC(carps_badsum); 477 CARP_DEBUG("%s: checksum failed on %s\n", __func__, 478 m->m_pkthdr.rcvif->if_xname); 479 m_freem(m); 480 return; 481 } 482 m->m_data -= iplen; 483 484 carp_input_c(m, ch, AF_INET); 485 } 486 #endif 487 488 #ifdef INET6 489 int 490 carp6_input(struct mbuf **mp, int *offp, int proto) 491 { 492 struct mbuf *m = *mp; 493 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 494 struct carp_header *ch; 495 u_int len; 496 497 CARPSTATS_INC(carps_ipackets6); 498 499 if (!carp_allow) { 500 m_freem(m); 501 return (IPPROTO_DONE); 502 } 503 504 /* check if received on a valid carp interface */ 505 if (m->m_pkthdr.rcvif->if_carp == NULL) { 506 CARPSTATS_INC(carps_badif); 507 CARP_DEBUG("%s: packet received on non-carp interface: %s\n", 508 __func__, m->m_pkthdr.rcvif->if_xname); 509 m_freem(m); 510 return (IPPROTO_DONE); 511 } 512 513 /* verify that the IP TTL is 255 */ 514 if (ip6->ip6_hlim != CARP_DFLTTL) { 515 CARPSTATS_INC(carps_badttl); 516 CARP_DEBUG("%s: received ttl %d != 255 on %s\n", __func__, 517 ip6->ip6_hlim, m->m_pkthdr.rcvif->if_xname); 518 m_freem(m); 519 return (IPPROTO_DONE); 520 } 521 522 /* verify that we have a complete carp packet */ 523 len = m->m_len; 524 IP6_EXTHDR_GET(ch, struct carp_header *, m, *offp, sizeof(*ch)); 525 if (ch == NULL) { 526 CARPSTATS_INC(carps_badlen); 527 CARP_DEBUG("%s: packet size %u too small\n", __func__, len); 528 return (IPPROTO_DONE); 529 } 530 531 532 /* verify the CARP checksum */ 533 m->m_data += *offp; 534 if (in_cksum(m, sizeof(*ch))) { 535 CARPSTATS_INC(carps_badsum); 536 CARP_DEBUG("%s: checksum failed, on %s\n", __func__, 537 m->m_pkthdr.rcvif->if_xname); 538 m_freem(m); 539 return (IPPROTO_DONE); 540 } 541 m->m_data -= *offp; 542 543 carp_input_c(m, ch, AF_INET6); 544 return (IPPROTO_DONE); 545 } 546 #endif /* INET6 */ 547 548 static void 549 carp_input_c(struct mbuf *m, struct carp_header *ch, sa_family_t af) 550 { 551 struct ifnet *ifp = m->m_pkthdr.rcvif; 552 struct ifaddr *ifa; 553 struct carp_softc *sc; 554 uint64_t tmp_counter; 555 struct timeval sc_tv, ch_tv; 556 557 /* verify that the VHID is valid on the receiving interface */ 558 IF_ADDR_RLOCK(ifp); 559 IFNET_FOREACH_IFA(ifp, ifa) 560 if (ifa->ifa_addr->sa_family == af && 561 ifa->ifa_carp->sc_vhid == ch->carp_vhid) { 562 ifa_ref(ifa); 563 break; 564 } 565 IF_ADDR_RUNLOCK(ifp); 566 567 if (ifa == NULL) { 568 CARPSTATS_INC(carps_badvhid); 569 m_freem(m); 570 return; 571 } 572 573 /* verify the CARP version. */ 574 if (ch->carp_version != CARP_VERSION) { 575 CARPSTATS_INC(carps_badver); 576 CARP_DEBUG("%s: invalid version %d\n", ifp->if_xname, 577 ch->carp_version); 578 ifa_free(ifa); 579 m_freem(m); 580 return; 581 } 582 583 sc = ifa->ifa_carp; 584 CARP_LOCK(sc); 585 ifa_free(ifa); 586 587 if (carp_hmac_verify(sc, ch->carp_counter, ch->carp_md)) { 588 CARPSTATS_INC(carps_badauth); 589 CARP_DEBUG("%s: incorrect hash for VHID %u@%s\n", __func__, 590 sc->sc_vhid, ifp->if_xname); 591 goto out; 592 } 593 594 tmp_counter = ntohl(ch->carp_counter[0]); 595 tmp_counter = tmp_counter<<32; 596 tmp_counter += ntohl(ch->carp_counter[1]); 597 598 /* XXX Replay protection goes here */ 599 600 sc->sc_init_counter = 0; 601 sc->sc_counter = tmp_counter; 602 603 sc_tv.tv_sec = sc->sc_advbase; 604 sc_tv.tv_usec = DEMOTE_ADVSKEW(sc) * 1000000 / 256; 605 ch_tv.tv_sec = ch->carp_advbase; 606 ch_tv.tv_usec = ch->carp_advskew * 1000000 / 256; 607 608 switch (sc->sc_state) { 609 case INIT: 610 break; 611 case MASTER: 612 /* 613 * If we receive an advertisement from a master who's going to 614 * be more frequent than us, go into BACKUP state. 615 */ 616 if (timevalcmp(&sc_tv, &ch_tv, >) || 617 timevalcmp(&sc_tv, &ch_tv, ==)) { 618 callout_stop(&sc->sc_ad_tmo); 619 CARP_LOG("VHID %u@%s: MASTER -> BACKUP " 620 "(more frequent advertisement received)\n", 621 sc->sc_vhid, 622 sc->sc_carpdev->if_xname); 623 carp_set_state(sc, BACKUP); 624 carp_setrun(sc, 0); 625 carp_delroute(sc); 626 } 627 break; 628 case BACKUP: 629 /* 630 * If we're pre-empting masters who advertise slower than us, 631 * and this one claims to be slower, treat him as down. 632 */ 633 if (carp_preempt && timevalcmp(&sc_tv, &ch_tv, <)) { 634 CARP_LOG("VHID %u@%s: BACKUP -> MASTER " 635 "(preempting a slower master)\n", 636 sc->sc_vhid, 637 sc->sc_carpdev->if_xname); 638 carp_master_down_locked(sc); 639 break; 640 } 641 642 /* 643 * If the master is going to advertise at such a low frequency 644 * that he's guaranteed to time out, we'd might as well just 645 * treat him as timed out now. 646 */ 647 sc_tv.tv_sec = sc->sc_advbase * 3; 648 if (timevalcmp(&sc_tv, &ch_tv, <)) { 649 CARP_LOG("VHID %u@%s: BACKUP -> MASTER " 650 "(master timed out)\n", 651 sc->sc_vhid, 652 sc->sc_carpdev->if_xname); 653 carp_master_down_locked(sc); 654 break; 655 } 656 657 /* 658 * Otherwise, we reset the counter and wait for the next 659 * advertisement. 660 */ 661 carp_setrun(sc, af); 662 break; 663 } 664 665 out: 666 CARP_UNLOCK(sc); 667 m_freem(m); 668 } 669 670 static int 671 carp_prepare_ad(struct mbuf *m, struct carp_softc *sc, struct carp_header *ch) 672 { 673 struct m_tag *mtag; 674 675 if (sc->sc_init_counter) { 676 /* this could also be seconds since unix epoch */ 677 sc->sc_counter = arc4random(); 678 sc->sc_counter = sc->sc_counter << 32; 679 sc->sc_counter += arc4random(); 680 } else 681 sc->sc_counter++; 682 683 ch->carp_counter[0] = htonl((sc->sc_counter>>32)&0xffffffff); 684 ch->carp_counter[1] = htonl(sc->sc_counter&0xffffffff); 685 686 carp_hmac_generate(sc, ch->carp_counter, ch->carp_md); 687 688 /* Tag packet for carp_output */ 689 if ((mtag = m_tag_get(PACKET_TAG_CARP, sizeof(struct carp_softc *), 690 M_NOWAIT)) == NULL) { 691 m_freem(m); 692 CARPSTATS_INC(carps_onomem); 693 return (ENOMEM); 694 } 695 bcopy(&sc, mtag + 1, sizeof(sc)); 696 m_tag_prepend(m, mtag); 697 698 return (0); 699 } 700 701 /* 702 * To avoid LORs and possible recursions this function shouldn't 703 * be called directly, but scheduled via taskqueue. 704 */ 705 static void 706 carp_send_ad_all(void *ctx __unused, int pending __unused) 707 { 708 struct carp_softc *sc; 709 710 mtx_lock(&carp_mtx); 711 LIST_FOREACH(sc, &carp_list, sc_next) 712 if (sc->sc_state == MASTER) { 713 CARP_LOCK(sc); 714 CURVNET_SET(sc->sc_carpdev->if_vnet); 715 carp_send_ad_locked(sc); 716 CURVNET_RESTORE(); 717 CARP_UNLOCK(sc); 718 } 719 mtx_unlock(&carp_mtx); 720 } 721 722 /* Send a periodic advertisement, executed in callout context. */ 723 static void 724 carp_send_ad(void *v) 725 { 726 struct carp_softc *sc = v; 727 728 CARP_LOCK_ASSERT(sc); 729 CURVNET_SET(sc->sc_carpdev->if_vnet); 730 carp_send_ad_locked(sc); 731 CURVNET_RESTORE(); 732 CARP_UNLOCK(sc); 733 } 734 735 static void 736 carp_send_ad_locked(struct carp_softc *sc) 737 { 738 struct carp_header ch; 739 struct timeval tv; 740 struct sockaddr sa; 741 struct ifaddr *ifa; 742 struct carp_header *ch_ptr; 743 struct mbuf *m; 744 int len, advskew; 745 746 CARP_LOCK_ASSERT(sc); 747 748 advskew = DEMOTE_ADVSKEW(sc); 749 tv.tv_sec = sc->sc_advbase; 750 tv.tv_usec = advskew * 1000000 / 256; 751 752 ch.carp_version = CARP_VERSION; 753 ch.carp_type = CARP_ADVERTISEMENT; 754 ch.carp_vhid = sc->sc_vhid; 755 ch.carp_advbase = sc->sc_advbase; 756 ch.carp_advskew = advskew; 757 ch.carp_authlen = 7; /* XXX DEFINE */ 758 ch.carp_pad1 = 0; /* must be zero */ 759 ch.carp_cksum = 0; 760 761 /* XXXGL: OpenBSD picks first ifaddr with needed family. */ 762 763 #ifdef INET 764 if (sc->sc_naddrs) { 765 struct ip *ip; 766 767 MGETHDR(m, M_NOWAIT, MT_HEADER); 768 if (m == NULL) { 769 CARPSTATS_INC(carps_onomem); 770 goto resched; 771 } 772 len = sizeof(*ip) + sizeof(ch); 773 m->m_pkthdr.len = len; 774 m->m_pkthdr.rcvif = NULL; 775 m->m_len = len; 776 MH_ALIGN(m, m->m_len); 777 m->m_flags |= M_MCAST; 778 ip = mtod(m, struct ip *); 779 ip->ip_v = IPVERSION; 780 ip->ip_hl = sizeof(*ip) >> 2; 781 ip->ip_tos = IPTOS_LOWDELAY; 782 ip->ip_len = htons(len); 783 ip->ip_id = ip_newid(); 784 ip->ip_off = htons(IP_DF); 785 ip->ip_ttl = CARP_DFLTTL; 786 ip->ip_p = IPPROTO_CARP; 787 ip->ip_sum = 0; 788 789 bzero(&sa, sizeof(sa)); 790 sa.sa_family = AF_INET; 791 ifa = ifaof_ifpforaddr(&sa, sc->sc_carpdev); 792 if (ifa != NULL) { 793 ip->ip_src.s_addr = 794 ifatoia(ifa)->ia_addr.sin_addr.s_addr; 795 ifa_free(ifa); 796 } else 797 ip->ip_src.s_addr = 0; 798 ip->ip_dst.s_addr = htonl(INADDR_CARP_GROUP); 799 800 ch_ptr = (struct carp_header *)(&ip[1]); 801 bcopy(&ch, ch_ptr, sizeof(ch)); 802 if (carp_prepare_ad(m, sc, ch_ptr)) 803 goto resched; 804 805 m->m_data += sizeof(*ip); 806 ch_ptr->carp_cksum = in_cksum(m, len - sizeof(*ip)); 807 m->m_data -= sizeof(*ip); 808 809 CARPSTATS_INC(carps_opackets); 810 811 if (ip_output(m, NULL, NULL, IP_RAWOUTPUT, 812 &sc->sc_carpdev->if_carp->cif_imo, NULL)) { 813 if (sc->sc_sendad_errors < INT_MAX) 814 sc->sc_sendad_errors++; 815 if (sc->sc_sendad_errors == CARP_SENDAD_MAX_ERRORS) 816 carp_demote_adj(carp_senderr_adj, "send error"); 817 sc->sc_sendad_success = 0; 818 } else { 819 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) { 820 if (++sc->sc_sendad_success >= 821 CARP_SENDAD_MIN_SUCCESS) { 822 carp_demote_adj(-carp_senderr_adj, 823 "send ok"); 824 sc->sc_sendad_errors = 0; 825 } 826 } else 827 sc->sc_sendad_errors = 0; 828 } 829 } 830 #endif /* INET */ 831 #ifdef INET6 832 if (sc->sc_naddrs6) { 833 struct ip6_hdr *ip6; 834 835 MGETHDR(m, M_NOWAIT, MT_HEADER); 836 if (m == NULL) { 837 CARPSTATS_INC(carps_onomem); 838 goto resched; 839 } 840 len = sizeof(*ip6) + sizeof(ch); 841 m->m_pkthdr.len = len; 842 m->m_pkthdr.rcvif = NULL; 843 m->m_len = len; 844 MH_ALIGN(m, m->m_len); 845 m->m_flags |= M_MCAST; 846 ip6 = mtod(m, struct ip6_hdr *); 847 bzero(ip6, sizeof(*ip6)); 848 ip6->ip6_vfc |= IPV6_VERSION; 849 ip6->ip6_hlim = CARP_DFLTTL; 850 ip6->ip6_nxt = IPPROTO_CARP; 851 bzero(&sa, sizeof(sa)); 852 853 /* set the source address */ 854 sa.sa_family = AF_INET6; 855 ifa = ifaof_ifpforaddr(&sa, sc->sc_carpdev); 856 if (ifa != NULL) { 857 bcopy(IFA_IN6(ifa), &ip6->ip6_src, 858 sizeof(struct in6_addr)); 859 ifa_free(ifa); 860 } else 861 /* This should never happen with IPv6. */ 862 bzero(&ip6->ip6_src, sizeof(struct in6_addr)); 863 864 /* Set the multicast destination. */ 865 ip6->ip6_dst.s6_addr16[0] = htons(0xff02); 866 ip6->ip6_dst.s6_addr8[15] = 0x12; 867 if (in6_setscope(&ip6->ip6_dst, sc->sc_carpdev, NULL) != 0) { 868 m_freem(m); 869 CARP_DEBUG("%s: in6_setscope failed\n", __func__); 870 goto resched; 871 } 872 873 ch_ptr = (struct carp_header *)(&ip6[1]); 874 bcopy(&ch, ch_ptr, sizeof(ch)); 875 if (carp_prepare_ad(m, sc, ch_ptr)) 876 goto resched; 877 878 m->m_data += sizeof(*ip6); 879 ch_ptr->carp_cksum = in_cksum(m, len - sizeof(*ip6)); 880 m->m_data -= sizeof(*ip6); 881 882 CARPSTATS_INC(carps_opackets6); 883 884 if (ip6_output(m, NULL, NULL, 0, 885 &sc->sc_carpdev->if_carp->cif_im6o, NULL, NULL)) { 886 if (sc->sc_sendad_errors < INT_MAX) 887 sc->sc_sendad_errors++; 888 if (sc->sc_sendad_errors == CARP_SENDAD_MAX_ERRORS) 889 carp_demote_adj(carp_senderr_adj, 890 "send6 error"); 891 sc->sc_sendad_success = 0; 892 } else { 893 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) { 894 if (++sc->sc_sendad_success >= 895 CARP_SENDAD_MIN_SUCCESS) { 896 carp_demote_adj(-carp_senderr_adj, 897 "send6 ok"); 898 sc->sc_sendad_errors = 0; 899 } 900 } else 901 sc->sc_sendad_errors = 0; 902 } 903 } 904 #endif /* INET6 */ 905 906 resched: 907 callout_reset(&sc->sc_ad_tmo, tvtohz(&tv), carp_send_ad, sc); 908 } 909 910 static void 911 carp_addroute(struct carp_softc *sc) 912 { 913 struct ifaddr *ifa; 914 915 CARP_FOREACH_IFA(sc, ifa) 916 carp_ifa_addroute(ifa); 917 } 918 919 static void 920 carp_ifa_addroute(struct ifaddr *ifa) 921 { 922 923 switch (ifa->ifa_addr->sa_family) { 924 #ifdef INET 925 case AF_INET: 926 in_addprefix(ifatoia(ifa), RTF_UP); 927 ifa_add_loopback_route(ifa, 928 (struct sockaddr *)&ifatoia(ifa)->ia_addr); 929 break; 930 #endif 931 #ifdef INET6 932 case AF_INET6: 933 ifa_add_loopback_route(ifa, 934 (struct sockaddr *)&ifatoia6(ifa)->ia_addr); 935 in6_ifaddloop(ifa); 936 break; 937 #endif 938 } 939 } 940 941 static void 942 carp_delroute(struct carp_softc *sc) 943 { 944 struct ifaddr *ifa; 945 946 CARP_FOREACH_IFA(sc, ifa) 947 carp_ifa_delroute(ifa); 948 } 949 950 static void 951 carp_ifa_delroute(struct ifaddr *ifa) 952 { 953 954 switch (ifa->ifa_addr->sa_family) { 955 #ifdef INET 956 case AF_INET: 957 ifa_del_loopback_route(ifa, 958 (struct sockaddr *)&ifatoia(ifa)->ia_addr); 959 in_scrubprefix(ifatoia(ifa), LLE_STATIC); 960 break; 961 #endif 962 #ifdef INET6 963 case AF_INET6: 964 ifa_del_loopback_route(ifa, 965 (struct sockaddr *)&ifatoia6(ifa)->ia_addr); 966 in6_ifremloop(ifa); 967 break; 968 #endif 969 } 970 } 971 972 #ifdef INET 973 /* 974 * Broadcast a gratuitous ARP request containing 975 * the virtual router MAC address for each IP address 976 * associated with the virtual router. 977 */ 978 static void 979 carp_send_arp(struct carp_softc *sc) 980 { 981 struct ifaddr *ifa; 982 983 CARP_FOREACH_IFA(sc, ifa) 984 if (ifa->ifa_addr->sa_family == AF_INET) 985 arp_ifinit2(sc->sc_carpdev, ifa, LLADDR(&sc->sc_addr)); 986 } 987 988 int 989 carp_iamatch(struct ifaddr *ifa, uint8_t **enaddr) 990 { 991 struct carp_softc *sc = ifa->ifa_carp; 992 993 if (sc->sc_state == MASTER) { 994 *enaddr = LLADDR(&sc->sc_addr); 995 return (1); 996 } 997 998 return (0); 999 } 1000 #endif 1001 1002 #ifdef INET6 1003 static void 1004 carp_send_na(struct carp_softc *sc) 1005 { 1006 static struct in6_addr mcast = IN6ADDR_LINKLOCAL_ALLNODES_INIT; 1007 struct ifaddr *ifa; 1008 struct in6_addr *in6; 1009 1010 CARP_FOREACH_IFA(sc, ifa) { 1011 if (ifa->ifa_addr->sa_family != AF_INET6) 1012 continue; 1013 1014 in6 = IFA_IN6(ifa); 1015 nd6_na_output(sc->sc_carpdev, &mcast, in6, 1016 ND_NA_FLAG_OVERRIDE, 1, NULL); 1017 DELAY(1000); /* XXX */ 1018 } 1019 } 1020 1021 /* 1022 * Returns ifa in case it's a carp address and it is MASTER, or if the address 1023 * matches and is not a carp address. Returns NULL otherwise. 1024 */ 1025 struct ifaddr * 1026 carp_iamatch6(struct ifnet *ifp, struct in6_addr *taddr) 1027 { 1028 struct ifaddr *ifa; 1029 1030 ifa = NULL; 1031 IF_ADDR_RLOCK(ifp); 1032 TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) { 1033 if (ifa->ifa_addr->sa_family != AF_INET6) 1034 continue; 1035 if (!IN6_ARE_ADDR_EQUAL(taddr, IFA_IN6(ifa))) 1036 continue; 1037 if (ifa->ifa_carp && ifa->ifa_carp->sc_state != MASTER) 1038 ifa = NULL; 1039 else 1040 ifa_ref(ifa); 1041 break; 1042 } 1043 IF_ADDR_RUNLOCK(ifp); 1044 1045 return (ifa); 1046 } 1047 1048 caddr_t 1049 carp_macmatch6(struct ifnet *ifp, struct mbuf *m, const struct in6_addr *taddr) 1050 { 1051 struct ifaddr *ifa; 1052 1053 IF_ADDR_RLOCK(ifp); 1054 IFNET_FOREACH_IFA(ifp, ifa) 1055 if (ifa->ifa_addr->sa_family == AF_INET6 && 1056 IN6_ARE_ADDR_EQUAL(taddr, IFA_IN6(ifa))) { 1057 struct carp_softc *sc = ifa->ifa_carp; 1058 struct m_tag *mtag; 1059 1060 IF_ADDR_RUNLOCK(ifp); 1061 1062 mtag = m_tag_get(PACKET_TAG_CARP, 1063 sizeof(struct carp_softc *), M_NOWAIT); 1064 if (mtag == NULL) 1065 /* Better a bit than nothing. */ 1066 return (LLADDR(&sc->sc_addr)); 1067 1068 bcopy(&sc, mtag + 1, sizeof(sc)); 1069 m_tag_prepend(m, mtag); 1070 1071 return (LLADDR(&sc->sc_addr)); 1072 } 1073 IF_ADDR_RUNLOCK(ifp); 1074 1075 return (NULL); 1076 } 1077 #endif /* INET6 */ 1078 1079 int 1080 carp_forus(struct ifnet *ifp, u_char *dhost) 1081 { 1082 struct carp_softc *sc; 1083 uint8_t *ena = dhost; 1084 1085 if (ena[0] || ena[1] || ena[2] != 0x5e || ena[3] || ena[4] != 1) 1086 return (0); 1087 1088 CIF_LOCK(ifp->if_carp); 1089 IFNET_FOREACH_CARP(ifp, sc) { 1090 CARP_LOCK(sc); 1091 if (sc->sc_state == MASTER && !bcmp(dhost, LLADDR(&sc->sc_addr), 1092 ETHER_ADDR_LEN)) { 1093 CARP_UNLOCK(sc); 1094 CIF_UNLOCK(ifp->if_carp); 1095 return (1); 1096 } 1097 CARP_UNLOCK(sc); 1098 } 1099 CIF_UNLOCK(ifp->if_carp); 1100 1101 return (0); 1102 } 1103 1104 /* Master down timeout event, executed in callout context. */ 1105 static void 1106 carp_master_down(void *v) 1107 { 1108 struct carp_softc *sc = v; 1109 1110 CARP_LOCK_ASSERT(sc); 1111 1112 CURVNET_SET(sc->sc_carpdev->if_vnet); 1113 if (sc->sc_state == BACKUP) { 1114 CARP_LOG("VHID %u@%s: BACKUP -> MASTER (master down)\n", 1115 sc->sc_vhid, 1116 sc->sc_carpdev->if_xname); 1117 carp_master_down_locked(sc); 1118 } 1119 CURVNET_RESTORE(); 1120 1121 CARP_UNLOCK(sc); 1122 } 1123 1124 static void 1125 carp_master_down_locked(struct carp_softc *sc) 1126 { 1127 1128 CARP_LOCK_ASSERT(sc); 1129 1130 switch (sc->sc_state) { 1131 case BACKUP: 1132 carp_set_state(sc, MASTER); 1133 carp_send_ad_locked(sc); 1134 #ifdef INET 1135 carp_send_arp(sc); 1136 #endif 1137 #ifdef INET6 1138 carp_send_na(sc); 1139 #endif 1140 carp_setrun(sc, 0); 1141 carp_addroute(sc); 1142 break; 1143 case INIT: 1144 case MASTER: 1145 #ifdef INVARIANTS 1146 panic("carp: VHID %u@%s: master_down event in %s state\n", 1147 sc->sc_vhid, 1148 sc->sc_carpdev->if_xname, 1149 sc->sc_state ? "MASTER" : "INIT"); 1150 #endif 1151 break; 1152 } 1153 } 1154 1155 /* 1156 * When in backup state, af indicates whether to reset the master down timer 1157 * for v4 or v6. If it's set to zero, reset the ones which are already pending. 1158 */ 1159 static void 1160 carp_setrun(struct carp_softc *sc, sa_family_t af) 1161 { 1162 struct timeval tv; 1163 1164 CARP_LOCK_ASSERT(sc); 1165 1166 if ((sc->sc_carpdev->if_flags & IFF_UP) == 0 || 1167 sc->sc_carpdev->if_link_state != LINK_STATE_UP || 1168 (sc->sc_naddrs == 0 && sc->sc_naddrs6 == 0)) 1169 return; 1170 1171 switch (sc->sc_state) { 1172 case INIT: 1173 CARP_LOG("VHID %u@%s: INIT -> BACKUP\n", 1174 sc->sc_vhid, 1175 sc->sc_carpdev->if_xname); 1176 carp_set_state(sc, BACKUP); 1177 carp_setrun(sc, 0); 1178 break; 1179 case BACKUP: 1180 callout_stop(&sc->sc_ad_tmo); 1181 tv.tv_sec = 3 * sc->sc_advbase; 1182 tv.tv_usec = sc->sc_advskew * 1000000 / 256; 1183 switch (af) { 1184 #ifdef INET 1185 case AF_INET: 1186 callout_reset(&sc->sc_md_tmo, tvtohz(&tv), 1187 carp_master_down, sc); 1188 break; 1189 #endif 1190 #ifdef INET6 1191 case AF_INET6: 1192 callout_reset(&sc->sc_md6_tmo, tvtohz(&tv), 1193 carp_master_down, sc); 1194 break; 1195 #endif 1196 default: 1197 #ifdef INET 1198 if (sc->sc_naddrs) 1199 callout_reset(&sc->sc_md_tmo, tvtohz(&tv), 1200 carp_master_down, sc); 1201 #endif 1202 #ifdef INET6 1203 if (sc->sc_naddrs6) 1204 callout_reset(&sc->sc_md6_tmo, tvtohz(&tv), 1205 carp_master_down, sc); 1206 #endif 1207 break; 1208 } 1209 break; 1210 case MASTER: 1211 tv.tv_sec = sc->sc_advbase; 1212 tv.tv_usec = sc->sc_advskew * 1000000 / 256; 1213 callout_reset(&sc->sc_ad_tmo, tvtohz(&tv), 1214 carp_send_ad, sc); 1215 break; 1216 } 1217 } 1218 1219 /* 1220 * Setup multicast structures. 1221 */ 1222 static int 1223 carp_multicast_setup(struct carp_if *cif, sa_family_t sa) 1224 { 1225 struct ifnet *ifp = cif->cif_ifp; 1226 int error = 0; 1227 1228 CIF_LOCK_ASSERT(cif); 1229 1230 switch (sa) { 1231 #ifdef INET 1232 case AF_INET: 1233 { 1234 struct ip_moptions *imo = &cif->cif_imo; 1235 struct in_addr addr; 1236 1237 if (imo->imo_membership) 1238 return (0); 1239 1240 imo->imo_membership = (struct in_multi **)malloc( 1241 (sizeof(struct in_multi *) * IP_MIN_MEMBERSHIPS), M_CARP, 1242 M_NOWAIT); 1243 if (imo->imo_membership == NULL) 1244 return (ENOMEM); 1245 imo->imo_mfilters = NULL; 1246 imo->imo_max_memberships = IP_MIN_MEMBERSHIPS; 1247 imo->imo_multicast_vif = -1; 1248 1249 addr.s_addr = htonl(INADDR_CARP_GROUP); 1250 if ((error = in_joingroup(ifp, &addr, NULL, 1251 &imo->imo_membership[0])) != 0) { 1252 free(imo->imo_membership, M_CARP); 1253 break; 1254 } 1255 imo->imo_num_memberships++; 1256 imo->imo_multicast_ifp = ifp; 1257 imo->imo_multicast_ttl = CARP_DFLTTL; 1258 imo->imo_multicast_loop = 0; 1259 break; 1260 } 1261 #endif 1262 #ifdef INET6 1263 case AF_INET6: 1264 { 1265 struct ip6_moptions *im6o = &cif->cif_im6o; 1266 struct in6_addr in6; 1267 struct in6_multi *in6m; 1268 1269 if (im6o->im6o_membership) 1270 return (0); 1271 1272 im6o->im6o_membership = (struct in6_multi **)malloc( 1273 (sizeof(struct in6_multi *) * IPV6_MIN_MEMBERSHIPS), M_CARP, 1274 M_ZERO | M_NOWAIT); 1275 if (im6o->im6o_membership == NULL) 1276 return (ENOMEM); 1277 im6o->im6o_mfilters = NULL; 1278 im6o->im6o_max_memberships = IPV6_MIN_MEMBERSHIPS; 1279 im6o->im6o_multicast_hlim = CARP_DFLTTL; 1280 im6o->im6o_multicast_ifp = ifp; 1281 1282 /* Join IPv6 CARP multicast group. */ 1283 bzero(&in6, sizeof(in6)); 1284 in6.s6_addr16[0] = htons(0xff02); 1285 in6.s6_addr8[15] = 0x12; 1286 if ((error = in6_setscope(&in6, ifp, NULL)) != 0) { 1287 free(im6o->im6o_membership, M_CARP); 1288 break; 1289 } 1290 in6m = NULL; 1291 if ((error = in6_mc_join(ifp, &in6, NULL, &in6m, 0)) != 0) { 1292 free(im6o->im6o_membership, M_CARP); 1293 break; 1294 } 1295 im6o->im6o_membership[0] = in6m; 1296 im6o->im6o_num_memberships++; 1297 1298 /* Join solicited multicast address. */ 1299 bzero(&in6, sizeof(in6)); 1300 in6.s6_addr16[0] = htons(0xff02); 1301 in6.s6_addr32[1] = 0; 1302 in6.s6_addr32[2] = htonl(1); 1303 in6.s6_addr32[3] = 0; 1304 in6.s6_addr8[12] = 0xff; 1305 if ((error = in6_setscope(&in6, ifp, NULL)) != 0) { 1306 in6_mc_leave(im6o->im6o_membership[0], NULL); 1307 free(im6o->im6o_membership, M_CARP); 1308 break; 1309 } 1310 in6m = NULL; 1311 if ((error = in6_mc_join(ifp, &in6, NULL, &in6m, 0)) != 0) { 1312 in6_mc_leave(im6o->im6o_membership[0], NULL); 1313 free(im6o->im6o_membership, M_CARP); 1314 break; 1315 } 1316 im6o->im6o_membership[1] = in6m; 1317 im6o->im6o_num_memberships++; 1318 break; 1319 } 1320 #endif 1321 } 1322 1323 return (error); 1324 } 1325 1326 /* 1327 * Free multicast structures. 1328 */ 1329 static void 1330 carp_multicast_cleanup(struct carp_if *cif, sa_family_t sa) 1331 { 1332 1333 CIF_LOCK_ASSERT(cif); 1334 switch (sa) { 1335 #ifdef INET 1336 case AF_INET: 1337 if (cif->cif_naddrs == 0) { 1338 struct ip_moptions *imo = &cif->cif_imo; 1339 1340 in_leavegroup(imo->imo_membership[0], NULL); 1341 KASSERT(imo->imo_mfilters == NULL, 1342 ("%s: imo_mfilters != NULL", __func__)); 1343 free(imo->imo_membership, M_CARP); 1344 imo->imo_membership = NULL; 1345 1346 } 1347 break; 1348 #endif 1349 #ifdef INET6 1350 case AF_INET6: 1351 if (cif->cif_naddrs6 == 0) { 1352 struct ip6_moptions *im6o = &cif->cif_im6o; 1353 1354 in6_mc_leave(im6o->im6o_membership[0], NULL); 1355 in6_mc_leave(im6o->im6o_membership[1], NULL); 1356 KASSERT(im6o->im6o_mfilters == NULL, 1357 ("%s: im6o_mfilters != NULL", __func__)); 1358 free(im6o->im6o_membership, M_CARP); 1359 im6o->im6o_membership = NULL; 1360 } 1361 break; 1362 #endif 1363 } 1364 } 1365 1366 int 1367 carp_output(struct ifnet *ifp, struct mbuf *m, struct sockaddr *sa) 1368 { 1369 struct m_tag *mtag; 1370 struct carp_softc *sc; 1371 1372 if (!sa) 1373 return (0); 1374 1375 switch (sa->sa_family) { 1376 #ifdef INET 1377 case AF_INET: 1378 break; 1379 #endif 1380 #ifdef INET6 1381 case AF_INET6: 1382 break; 1383 #endif 1384 default: 1385 return (0); 1386 } 1387 1388 mtag = m_tag_find(m, PACKET_TAG_CARP, NULL); 1389 if (mtag == NULL) 1390 return (0); 1391 1392 bcopy(mtag + 1, &sc, sizeof(sc)); 1393 1394 /* Set the source MAC address to the Virtual Router MAC Address. */ 1395 switch (ifp->if_type) { 1396 case IFT_ETHER: 1397 case IFT_BRIDGE: 1398 case IFT_L2VLAN: { 1399 struct ether_header *eh; 1400 1401 eh = mtod(m, struct ether_header *); 1402 eh->ether_shost[0] = 0; 1403 eh->ether_shost[1] = 0; 1404 eh->ether_shost[2] = 0x5e; 1405 eh->ether_shost[3] = 0; 1406 eh->ether_shost[4] = 1; 1407 eh->ether_shost[5] = sc->sc_vhid; 1408 } 1409 break; 1410 case IFT_FDDI: { 1411 struct fddi_header *fh; 1412 1413 fh = mtod(m, struct fddi_header *); 1414 fh->fddi_shost[0] = 0; 1415 fh->fddi_shost[1] = 0; 1416 fh->fddi_shost[2] = 0x5e; 1417 fh->fddi_shost[3] = 0; 1418 fh->fddi_shost[4] = 1; 1419 fh->fddi_shost[5] = sc->sc_vhid; 1420 } 1421 break; 1422 case IFT_ISO88025: { 1423 struct iso88025_header *th; 1424 th = mtod(m, struct iso88025_header *); 1425 th->iso88025_shost[0] = 3; 1426 th->iso88025_shost[1] = 0; 1427 th->iso88025_shost[2] = 0x40 >> (sc->sc_vhid - 1); 1428 th->iso88025_shost[3] = 0x40000 >> (sc->sc_vhid - 1); 1429 th->iso88025_shost[4] = 0; 1430 th->iso88025_shost[5] = 0; 1431 } 1432 break; 1433 default: 1434 printf("%s: carp is not supported for the %d interface type\n", 1435 ifp->if_xname, ifp->if_type); 1436 return (EOPNOTSUPP); 1437 } 1438 1439 return (0); 1440 } 1441 1442 static struct carp_softc* 1443 carp_alloc(struct ifnet *ifp) 1444 { 1445 struct carp_softc *sc; 1446 struct carp_if *cif; 1447 1448 if ((cif = ifp->if_carp) == NULL) { 1449 cif = carp_alloc_if(ifp); 1450 if (cif == NULL) 1451 return (NULL); 1452 } 1453 1454 sc = malloc(sizeof(*sc), M_CARP, M_WAITOK|M_ZERO); 1455 1456 sc->sc_advbase = CARP_DFLTINTV; 1457 sc->sc_vhid = -1; /* required setting */ 1458 sc->sc_init_counter = 1; 1459 sc->sc_state = INIT; 1460 1461 sc->sc_ifasiz = sizeof(struct ifaddr *); 1462 sc->sc_ifas = malloc(sc->sc_ifasiz, M_CARP, M_WAITOK|M_ZERO); 1463 sc->sc_carpdev = ifp; 1464 1465 CARP_LOCK_INIT(sc); 1466 #ifdef INET 1467 callout_init_mtx(&sc->sc_md_tmo, &sc->sc_mtx, CALLOUT_RETURNUNLOCKED); 1468 #endif 1469 #ifdef INET6 1470 callout_init_mtx(&sc->sc_md6_tmo, &sc->sc_mtx, CALLOUT_RETURNUNLOCKED); 1471 #endif 1472 callout_init_mtx(&sc->sc_ad_tmo, &sc->sc_mtx, CALLOUT_RETURNUNLOCKED); 1473 1474 CIF_LOCK(cif); 1475 TAILQ_INSERT_TAIL(&cif->cif_vrs, sc, sc_list); 1476 CIF_UNLOCK(cif); 1477 1478 mtx_lock(&carp_mtx); 1479 LIST_INSERT_HEAD(&carp_list, sc, sc_next); 1480 mtx_unlock(&carp_mtx); 1481 1482 return (sc); 1483 } 1484 1485 static int 1486 carp_grow_ifas(struct carp_softc *sc) 1487 { 1488 struct ifaddr **new; 1489 1490 CARP_LOCK_ASSERT(sc); 1491 1492 new = malloc(sc->sc_ifasiz * 2, M_CARP, M_NOWAIT|M_ZERO); 1493 if (new == NULL) 1494 return (ENOMEM); 1495 bcopy(sc->sc_ifas, new, sc->sc_ifasiz); 1496 free(sc->sc_ifas, M_CARP); 1497 sc->sc_ifas = new; 1498 sc->sc_ifasiz *= 2; 1499 1500 return (0); 1501 } 1502 1503 static void 1504 carp_destroy(struct carp_softc *sc) 1505 { 1506 struct ifnet *ifp = sc->sc_carpdev; 1507 struct carp_if *cif = ifp->if_carp; 1508 1509 CIF_LOCK_ASSERT(cif); 1510 1511 TAILQ_REMOVE(&cif->cif_vrs, sc, sc_list); 1512 1513 mtx_lock(&carp_mtx); 1514 LIST_REMOVE(sc, sc_next); 1515 mtx_unlock(&carp_mtx); 1516 1517 CARP_LOCK(sc); 1518 if (sc->sc_suppress) 1519 carp_demote_adj(-carp_ifdown_adj, "vhid removed"); 1520 callout_drain(&sc->sc_ad_tmo); 1521 #ifdef INET 1522 callout_drain(&sc->sc_md_tmo); 1523 #endif 1524 #ifdef INET6 1525 callout_drain(&sc->sc_md6_tmo); 1526 #endif 1527 CARP_LOCK_DESTROY(sc); 1528 1529 free(sc->sc_ifas, M_CARP); 1530 free(sc, M_CARP); 1531 } 1532 1533 static struct carp_if* 1534 carp_alloc_if(struct ifnet *ifp) 1535 { 1536 struct carp_if *cif; 1537 1538 cif = malloc(sizeof(*cif), M_CARP, M_WAITOK|M_ZERO); 1539 1540 if (ifpromisc(ifp, 1) != 0) 1541 goto cleanup; 1542 1543 CIF_LOCK_INIT(cif); 1544 cif->cif_ifp = ifp; 1545 TAILQ_INIT(&cif->cif_vrs); 1546 1547 IF_ADDR_WLOCK(ifp); 1548 ifp->if_carp = cif; 1549 if_ref(ifp); 1550 IF_ADDR_WUNLOCK(ifp); 1551 1552 return (cif); 1553 1554 cleanup: 1555 free(cif, M_CARP); 1556 1557 return (NULL); 1558 } 1559 1560 static void 1561 carp_free_if(struct carp_if *cif) 1562 { 1563 struct ifnet *ifp = cif->cif_ifp; 1564 1565 CIF_LOCK_ASSERT(cif); 1566 KASSERT(TAILQ_EMPTY(&cif->cif_vrs), ("%s: softc list not empty", 1567 __func__)); 1568 1569 IF_ADDR_WLOCK(ifp); 1570 ifp->if_carp = NULL; 1571 if_rele(ifp); 1572 IF_ADDR_WUNLOCK(ifp); 1573 1574 CIF_LOCK_DESTROY(cif); 1575 1576 ifpromisc(ifp, 0); 1577 1578 free(cif, M_CARP); 1579 } 1580 1581 static void 1582 carp_carprcp(struct carpreq *carpr, struct carp_softc *sc, int priv) 1583 { 1584 1585 CARP_LOCK(sc); 1586 carpr->carpr_state = sc->sc_state; 1587 carpr->carpr_vhid = sc->sc_vhid; 1588 carpr->carpr_advbase = sc->sc_advbase; 1589 carpr->carpr_advskew = sc->sc_advskew; 1590 if (priv) 1591 bcopy(sc->sc_key, carpr->carpr_key, sizeof(carpr->carpr_key)); 1592 else 1593 bzero(carpr->carpr_key, sizeof(carpr->carpr_key)); 1594 CARP_UNLOCK(sc); 1595 } 1596 1597 int 1598 carp_ioctl(struct ifreq *ifr, u_long cmd, struct thread *td) 1599 { 1600 struct carpreq carpr; 1601 struct ifnet *ifp; 1602 struct carp_softc *sc = NULL; 1603 int error = 0, locked = 0; 1604 1605 if ((error = copyin(ifr->ifr_data, &carpr, sizeof carpr))) 1606 return (error); 1607 1608 ifp = ifunit_ref(ifr->ifr_name); 1609 if (ifp == NULL) 1610 return (ENXIO); 1611 1612 switch (ifp->if_type) { 1613 case IFT_ETHER: 1614 case IFT_L2VLAN: 1615 case IFT_BRIDGE: 1616 case IFT_FDDI: 1617 case IFT_ISO88025: 1618 break; 1619 default: 1620 error = EOPNOTSUPP; 1621 goto out; 1622 } 1623 1624 if ((ifp->if_flags & IFF_MULTICAST) == 0) { 1625 error = EADDRNOTAVAIL; 1626 goto out; 1627 } 1628 1629 switch (cmd) { 1630 case SIOCSVH: 1631 if ((error = priv_check(td, PRIV_NETINET_CARP))) 1632 break; 1633 if (carpr.carpr_vhid <= 0 || carpr.carpr_vhid > CARP_MAXVHID || 1634 carpr.carpr_advbase < 0 || carpr.carpr_advskew < 0) { 1635 error = EINVAL; 1636 break; 1637 } 1638 1639 if (ifp->if_carp) { 1640 CIF_LOCK(ifp->if_carp); 1641 IFNET_FOREACH_CARP(ifp, sc) 1642 if (sc->sc_vhid == carpr.carpr_vhid) 1643 break; 1644 CIF_UNLOCK(ifp->if_carp); 1645 } 1646 if (sc == NULL) { 1647 sc = carp_alloc(ifp); 1648 if (sc == NULL) { 1649 error = EINVAL; /* XXX: ifpromisc failed */ 1650 break; 1651 } 1652 1653 CARP_LOCK(sc); 1654 sc->sc_vhid = carpr.carpr_vhid; 1655 LLADDR(&sc->sc_addr)[0] = 0; 1656 LLADDR(&sc->sc_addr)[1] = 0; 1657 LLADDR(&sc->sc_addr)[2] = 0x5e; 1658 LLADDR(&sc->sc_addr)[3] = 0; 1659 LLADDR(&sc->sc_addr)[4] = 1; 1660 LLADDR(&sc->sc_addr)[5] = sc->sc_vhid; 1661 } else 1662 CARP_LOCK(sc); 1663 locked = 1; 1664 if (carpr.carpr_advbase > 0) { 1665 if (carpr.carpr_advbase > 255 || 1666 carpr.carpr_advbase < CARP_DFLTINTV) { 1667 error = EINVAL; 1668 break; 1669 } 1670 sc->sc_advbase = carpr.carpr_advbase; 1671 } 1672 if (carpr.carpr_advskew > 0) { 1673 if (carpr.carpr_advskew >= 255) { 1674 error = EINVAL; 1675 break; 1676 } 1677 sc->sc_advskew = carpr.carpr_advskew; 1678 } 1679 if (carpr.carpr_key[0] != '\0') { 1680 bcopy(carpr.carpr_key, sc->sc_key, sizeof(sc->sc_key)); 1681 carp_hmac_prepare(sc); 1682 } 1683 if (sc->sc_state != INIT && 1684 carpr.carpr_state != sc->sc_state) { 1685 switch (carpr.carpr_state) { 1686 case BACKUP: 1687 callout_stop(&sc->sc_ad_tmo); 1688 carp_set_state(sc, BACKUP); 1689 carp_setrun(sc, 0); 1690 carp_delroute(sc); 1691 break; 1692 case MASTER: 1693 carp_master_down_locked(sc); 1694 break; 1695 default: 1696 break; 1697 } 1698 } 1699 break; 1700 1701 case SIOCGVH: 1702 { 1703 int priveleged; 1704 1705 if (carpr.carpr_vhid < 0 || carpr.carpr_vhid > CARP_MAXVHID) { 1706 error = EINVAL; 1707 break; 1708 } 1709 if (carpr.carpr_count < 1) { 1710 error = EMSGSIZE; 1711 break; 1712 } 1713 if (ifp->if_carp == NULL) { 1714 error = ENOENT; 1715 break; 1716 } 1717 1718 priveleged = (priv_check(td, PRIV_NETINET_CARP) == 0); 1719 if (carpr.carpr_vhid != 0) { 1720 CIF_LOCK(ifp->if_carp); 1721 IFNET_FOREACH_CARP(ifp, sc) 1722 if (sc->sc_vhid == carpr.carpr_vhid) 1723 break; 1724 CIF_UNLOCK(ifp->if_carp); 1725 if (sc == NULL) { 1726 error = ENOENT; 1727 break; 1728 } 1729 carp_carprcp(&carpr, sc, priveleged); 1730 error = copyout(&carpr, ifr->ifr_data, sizeof(carpr)); 1731 } else { 1732 int i, count; 1733 1734 count = 0; 1735 CIF_LOCK(ifp->if_carp); 1736 IFNET_FOREACH_CARP(ifp, sc) 1737 count++; 1738 1739 if (count > carpr.carpr_count) { 1740 CIF_UNLOCK(ifp->if_carp); 1741 error = EMSGSIZE; 1742 break; 1743 } 1744 1745 i = 0; 1746 IFNET_FOREACH_CARP(ifp, sc) { 1747 carp_carprcp(&carpr, sc, priveleged); 1748 carpr.carpr_count = count; 1749 error = copyout(&carpr, ifr->ifr_data + 1750 (i * sizeof(carpr)), sizeof(carpr)); 1751 if (error) { 1752 CIF_UNLOCK(ifp->if_carp); 1753 break; 1754 } 1755 i++; 1756 } 1757 CIF_UNLOCK(ifp->if_carp); 1758 } 1759 break; 1760 } 1761 default: 1762 error = EINVAL; 1763 } 1764 1765 out: 1766 if (locked) 1767 CARP_UNLOCK(sc); 1768 if_rele(ifp); 1769 1770 return (error); 1771 } 1772 1773 static int 1774 carp_get_vhid(struct ifaddr *ifa) 1775 { 1776 1777 if (ifa == NULL || ifa->ifa_carp == NULL) 1778 return (0); 1779 1780 return (ifa->ifa_carp->sc_vhid); 1781 } 1782 1783 int 1784 carp_attach(struct ifaddr *ifa, int vhid) 1785 { 1786 struct ifnet *ifp = ifa->ifa_ifp; 1787 struct carp_if *cif = ifp->if_carp; 1788 struct carp_softc *sc; 1789 int index, error; 1790 1791 if (ifp->if_carp == NULL) 1792 return (ENOPROTOOPT); 1793 1794 switch (ifa->ifa_addr->sa_family) { 1795 #ifdef INET 1796 case AF_INET: 1797 #endif 1798 #ifdef INET6 1799 case AF_INET6: 1800 #endif 1801 break; 1802 default: 1803 return (EPROTOTYPE); 1804 } 1805 1806 CIF_LOCK(cif); 1807 IFNET_FOREACH_CARP(ifp, sc) 1808 if (sc->sc_vhid == vhid) 1809 break; 1810 if (sc == NULL) { 1811 CIF_UNLOCK(cif); 1812 return (ENOENT); 1813 } 1814 1815 if (ifa->ifa_carp) { 1816 if (ifa->ifa_carp->sc_vhid != vhid) 1817 carp_detach_locked(ifa); 1818 else { 1819 CIF_UNLOCK(cif); 1820 return (0); 1821 } 1822 } 1823 1824 error = carp_multicast_setup(cif, ifa->ifa_addr->sa_family); 1825 if (error) { 1826 CIF_FREE(cif); 1827 return (error); 1828 } 1829 1830 CARP_LOCK(sc); 1831 index = sc->sc_naddrs + sc->sc_naddrs6 + 1; 1832 if (index > sc->sc_ifasiz / sizeof(struct ifaddr *)) 1833 if ((error = carp_grow_ifas(sc)) != 0) { 1834 carp_multicast_cleanup(cif, 1835 ifa->ifa_addr->sa_family); 1836 CARP_UNLOCK(sc); 1837 CIF_FREE(cif); 1838 return (error); 1839 } 1840 1841 switch (ifa->ifa_addr->sa_family) { 1842 #ifdef INET 1843 case AF_INET: 1844 cif->cif_naddrs++; 1845 sc->sc_naddrs++; 1846 break; 1847 #endif 1848 #ifdef INET6 1849 case AF_INET6: 1850 cif->cif_naddrs6++; 1851 sc->sc_naddrs6++; 1852 break; 1853 #endif 1854 } 1855 1856 ifa_ref(ifa); 1857 sc->sc_ifas[index - 1] = ifa; 1858 ifa->ifa_carp = sc; 1859 1860 carp_hmac_prepare(sc); 1861 carp_sc_state(sc); 1862 1863 CARP_UNLOCK(sc); 1864 CIF_UNLOCK(cif); 1865 1866 return (0); 1867 } 1868 1869 void 1870 carp_detach(struct ifaddr *ifa) 1871 { 1872 struct ifnet *ifp = ifa->ifa_ifp; 1873 struct carp_if *cif = ifp->if_carp; 1874 1875 CIF_LOCK(cif); 1876 carp_detach_locked(ifa); 1877 CIF_FREE(cif); 1878 } 1879 1880 static void 1881 carp_detach_locked(struct ifaddr *ifa) 1882 { 1883 struct ifnet *ifp = ifa->ifa_ifp; 1884 struct carp_if *cif = ifp->if_carp; 1885 struct carp_softc *sc = ifa->ifa_carp; 1886 int i, index; 1887 1888 KASSERT(sc != NULL, ("%s: %p not attached", __func__, ifa)); 1889 1890 CIF_LOCK_ASSERT(cif); 1891 CARP_LOCK(sc); 1892 1893 /* Shift array. */ 1894 index = sc->sc_naddrs + sc->sc_naddrs6; 1895 for (i = 0; i < index; i++) 1896 if (sc->sc_ifas[i] == ifa) 1897 break; 1898 KASSERT(i < index, ("%s: %p no backref", __func__, ifa)); 1899 for (; i < index - 1; i++) 1900 sc->sc_ifas[i] = sc->sc_ifas[i+1]; 1901 sc->sc_ifas[index - 1] = NULL; 1902 1903 switch (ifa->ifa_addr->sa_family) { 1904 #ifdef INET 1905 case AF_INET: 1906 cif->cif_naddrs--; 1907 sc->sc_naddrs--; 1908 break; 1909 #endif 1910 #ifdef INET6 1911 case AF_INET6: 1912 cif->cif_naddrs6--; 1913 sc->sc_naddrs6--; 1914 break; 1915 #endif 1916 } 1917 1918 carp_ifa_delroute(ifa); 1919 carp_multicast_cleanup(cif, ifa->ifa_addr->sa_family); 1920 1921 ifa->ifa_carp = NULL; 1922 ifa_free(ifa); 1923 1924 carp_hmac_prepare(sc); 1925 carp_sc_state(sc); 1926 1927 if (sc->sc_naddrs == 0 && sc->sc_naddrs6 == 0) { 1928 CARP_UNLOCK(sc); 1929 carp_destroy(sc); 1930 } else 1931 CARP_UNLOCK(sc); 1932 } 1933 1934 static void 1935 carp_set_state(struct carp_softc *sc, int state) 1936 { 1937 1938 CARP_LOCK_ASSERT(sc); 1939 1940 if (sc->sc_state != state) { 1941 const char *carp_states[] = { CARP_STATES }; 1942 char subsys[IFNAMSIZ+5]; 1943 1944 sc->sc_state = state; 1945 1946 snprintf(subsys, IFNAMSIZ+5, "%u@%s", sc->sc_vhid, 1947 sc->sc_carpdev->if_xname); 1948 devctl_notify("CARP", subsys, carp_states[state], NULL); 1949 } 1950 } 1951 1952 static void 1953 carp_linkstate(struct ifnet *ifp) 1954 { 1955 struct carp_softc *sc; 1956 1957 CIF_LOCK(ifp->if_carp); 1958 IFNET_FOREACH_CARP(ifp, sc) { 1959 CARP_LOCK(sc); 1960 carp_sc_state(sc); 1961 CARP_UNLOCK(sc); 1962 } 1963 CIF_UNLOCK(ifp->if_carp); 1964 } 1965 1966 static void 1967 carp_sc_state(struct carp_softc *sc) 1968 { 1969 1970 CARP_LOCK_ASSERT(sc); 1971 1972 if (sc->sc_carpdev->if_link_state != LINK_STATE_UP || 1973 !(sc->sc_carpdev->if_flags & IFF_UP)) { 1974 callout_stop(&sc->sc_ad_tmo); 1975 #ifdef INET 1976 callout_stop(&sc->sc_md_tmo); 1977 #endif 1978 #ifdef INET6 1979 callout_stop(&sc->sc_md6_tmo); 1980 #endif 1981 carp_set_state(sc, INIT); 1982 carp_setrun(sc, 0); 1983 if (!sc->sc_suppress) 1984 carp_demote_adj(carp_ifdown_adj, "interface down"); 1985 sc->sc_suppress = 1; 1986 } else { 1987 carp_set_state(sc, INIT); 1988 carp_setrun(sc, 0); 1989 if (sc->sc_suppress) 1990 carp_demote_adj(-carp_ifdown_adj, "interface up"); 1991 sc->sc_suppress = 0; 1992 } 1993 } 1994 1995 static void 1996 carp_demote_adj(int adj, char *reason) 1997 { 1998 atomic_add_int(&carp_demotion, adj); 1999 CARP_LOG("demoted by %d to %d (%s)\n", adj, carp_demotion, reason); 2000 taskqueue_enqueue(taskqueue_swi, &carp_sendall_task); 2001 } 2002 2003 static int 2004 carp_demote_adj_sysctl(SYSCTL_HANDLER_ARGS) 2005 { 2006 int new, error; 2007 2008 new = carp_demotion; 2009 error = sysctl_handle_int(oidp, &new, 0, req); 2010 if (error || !req->newptr) 2011 return (error); 2012 2013 carp_demote_adj(new, "sysctl"); 2014 2015 return (0); 2016 } 2017 2018 #ifdef INET 2019 extern struct domain inetdomain; 2020 static struct protosw in_carp_protosw = { 2021 .pr_type = SOCK_RAW, 2022 .pr_domain = &inetdomain, 2023 .pr_protocol = IPPROTO_CARP, 2024 .pr_flags = PR_ATOMIC|PR_ADDR, 2025 .pr_input = carp_input, 2026 .pr_output = (pr_output_t *)rip_output, 2027 .pr_ctloutput = rip_ctloutput, 2028 .pr_usrreqs = &rip_usrreqs 2029 }; 2030 #endif 2031 2032 #ifdef INET6 2033 extern struct domain inet6domain; 2034 static struct ip6protosw in6_carp_protosw = { 2035 .pr_type = SOCK_RAW, 2036 .pr_domain = &inet6domain, 2037 .pr_protocol = IPPROTO_CARP, 2038 .pr_flags = PR_ATOMIC|PR_ADDR, 2039 .pr_input = carp6_input, 2040 .pr_output = rip6_output, 2041 .pr_ctloutput = rip6_ctloutput, 2042 .pr_usrreqs = &rip6_usrreqs 2043 }; 2044 #endif 2045 2046 static void 2047 carp_mod_cleanup(void) 2048 { 2049 2050 #ifdef INET 2051 if (proto_reg[CARP_INET] == 0) { 2052 (void)ipproto_unregister(IPPROTO_CARP); 2053 pf_proto_unregister(PF_INET, IPPROTO_CARP, SOCK_RAW); 2054 proto_reg[CARP_INET] = -1; 2055 } 2056 carp_iamatch_p = NULL; 2057 #endif 2058 #ifdef INET6 2059 if (proto_reg[CARP_INET6] == 0) { 2060 (void)ip6proto_unregister(IPPROTO_CARP); 2061 pf_proto_unregister(PF_INET6, IPPROTO_CARP, SOCK_RAW); 2062 proto_reg[CARP_INET6] = -1; 2063 } 2064 carp_iamatch6_p = NULL; 2065 carp_macmatch6_p = NULL; 2066 #endif 2067 carp_ioctl_p = NULL; 2068 carp_attach_p = NULL; 2069 carp_detach_p = NULL; 2070 carp_get_vhid_p = NULL; 2071 carp_linkstate_p = NULL; 2072 carp_forus_p = NULL; 2073 carp_output_p = NULL; 2074 carp_demote_adj_p = NULL; 2075 mtx_unlock(&carp_mtx); 2076 taskqueue_drain(taskqueue_swi, &carp_sendall_task); 2077 mtx_destroy(&carp_mtx); 2078 } 2079 2080 static int 2081 carp_mod_load(void) 2082 { 2083 int err; 2084 2085 mtx_init(&carp_mtx, "carp_mtx", NULL, MTX_DEF); 2086 LIST_INIT(&carp_list); 2087 carp_get_vhid_p = carp_get_vhid; 2088 carp_forus_p = carp_forus; 2089 carp_output_p = carp_output; 2090 carp_linkstate_p = carp_linkstate; 2091 carp_ioctl_p = carp_ioctl; 2092 carp_attach_p = carp_attach; 2093 carp_detach_p = carp_detach; 2094 carp_demote_adj_p = carp_demote_adj; 2095 #ifdef INET6 2096 carp_iamatch6_p = carp_iamatch6; 2097 carp_macmatch6_p = carp_macmatch6; 2098 proto_reg[CARP_INET6] = pf_proto_register(PF_INET6, 2099 (struct protosw *)&in6_carp_protosw); 2100 if (proto_reg[CARP_INET6]) { 2101 printf("carp: error %d attaching to PF_INET6\n", 2102 proto_reg[CARP_INET6]); 2103 carp_mod_cleanup(); 2104 return (proto_reg[CARP_INET6]); 2105 } 2106 err = ip6proto_register(IPPROTO_CARP); 2107 if (err) { 2108 printf("carp: error %d registering with INET6\n", err); 2109 carp_mod_cleanup(); 2110 return (err); 2111 } 2112 #endif 2113 #ifdef INET 2114 carp_iamatch_p = carp_iamatch; 2115 proto_reg[CARP_INET] = pf_proto_register(PF_INET, &in_carp_protosw); 2116 if (proto_reg[CARP_INET]) { 2117 printf("carp: error %d attaching to PF_INET\n", 2118 proto_reg[CARP_INET]); 2119 carp_mod_cleanup(); 2120 return (proto_reg[CARP_INET]); 2121 } 2122 err = ipproto_register(IPPROTO_CARP); 2123 if (err) { 2124 printf("carp: error %d registering with INET\n", err); 2125 carp_mod_cleanup(); 2126 return (err); 2127 } 2128 #endif 2129 return (0); 2130 } 2131 2132 static int 2133 carp_modevent(module_t mod, int type, void *data) 2134 { 2135 switch (type) { 2136 case MOD_LOAD: 2137 return carp_mod_load(); 2138 /* NOTREACHED */ 2139 case MOD_UNLOAD: 2140 mtx_lock(&carp_mtx); 2141 if (LIST_EMPTY(&carp_list)) 2142 carp_mod_cleanup(); 2143 else { 2144 mtx_unlock(&carp_mtx); 2145 return (EBUSY); 2146 } 2147 break; 2148 2149 default: 2150 return (EINVAL); 2151 } 2152 2153 return (0); 2154 } 2155 2156 static moduledata_t carp_mod = { 2157 "carp", 2158 carp_modevent, 2159 0 2160 }; 2161 2162 DECLARE_MODULE(carp, carp_mod, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY); 2163