Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]
FreeBSD/Linux Kernel Cross Reference
sys/netinet/ip_carp.c
Version:
- FREEBSD - FREEBSD-13-STABLE - FREEBSD-13-0 - FREEBSD-12-STABLE - FREEBSD-12-0 - FREEBSD-11-STABLE - FREEBSD-11-0 - FREEBSD-10-STABLE - FREEBSD-10-0 - FREEBSD-9-STABLE - FREEBSD-9-0 - FREEBSD-8-STABLE - FREEBSD-8-0 - FREEBSD-7-STABLE - FREEBSD-7-0 - FREEBSD-6-STABLE - FREEBSD-6-0 - FREEBSD-5-STABLE - FREEBSD-5-0 - FREEBSD-4-STABLE - FREEBSD-3-STABLE - FREEBSD22 - l41 - OPENBSD - linux-2.6 - MK84 - PLAN9 - xnu-8792
SearchContext: - none - 3 - 10
SearchContext: - none - 3 - 10
1 /* $FreeBSD: releng/6.1/sys/netinet/ip_carp.c 153719 2005-12-25 21:59:20Z mlaier $ */ 2 3 /* 4 * Copyright (c) 2002 Michael Shalayeff. All rights reserved. 5 * Copyright (c) 2003 Ryan McBride. 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 "opt_carp.h" 30 #include "opt_bpf.h" 31 #include "opt_inet.h" 32 #include "opt_inet6.h" 33 34 #include <sys/types.h> 35 #include <sys/param.h> 36 #include <sys/systm.h> 37 #include <sys/conf.h> 38 #include <sys/kernel.h> 39 #include <sys/limits.h> 40 #include <sys/malloc.h> 41 #include <sys/mbuf.h> 42 #include <sys/module.h> 43 #include <sys/time.h> 44 #include <sys/proc.h> 45 #include <sys/sysctl.h> 46 #include <sys/syslog.h> 47 #include <sys/signalvar.h> 48 #include <sys/filio.h> 49 #include <sys/sockio.h> 50 51 #include <sys/socket.h> 52 #include <sys/vnode.h> 53 54 #include <machine/stdarg.h> 55 56 #include <net/bpf.h> 57 #include <net/ethernet.h> 58 #include <net/fddi.h> 59 #include <net/iso88025.h> 60 #include <net/if.h> 61 #include <net/if_clone.h> 62 #include <net/if_types.h> 63 #include <net/route.h> 64 65 #ifdef INET 66 #include <netinet/in.h> 67 #include <netinet/in_var.h> 68 #include <netinet/in_systm.h> 69 #include <netinet/ip.h> 70 #include <netinet/ip_var.h> 71 #include <netinet/if_ether.h> 72 #include <machine/in_cksum.h> 73 #endif 74 75 #ifdef INET6 76 #include <netinet/icmp6.h> 77 #include <netinet/ip6.h> 78 #include <netinet6/ip6_var.h> 79 #include <netinet6/scope6_var.h> 80 #include <netinet6/nd6.h> 81 #include <net/if_dl.h> 82 #endif 83 84 #include <crypto/sha1.h> 85 #include <netinet/ip_carp.h> 86 87 #define CARP_IFNAME "carp" 88 static MALLOC_DEFINE(M_CARP, "CARP", "CARP interfaces"); 89 SYSCTL_DECL(_net_inet_carp); 90 91 struct carp_softc { 92 struct ifnet *sc_ifp; /* Interface clue */ 93 struct ifnet *sc_carpdev; /* Pointer to parent interface */ 94 struct in_ifaddr *sc_ia; /* primary iface address */ 95 struct ip_moptions sc_imo; 96 #ifdef INET6 97 struct in6_ifaddr *sc_ia6; /* primary iface address v6 */ 98 struct ip6_moptions sc_im6o; 99 #endif /* INET6 */ 100 TAILQ_ENTRY(carp_softc) sc_list; 101 102 enum { INIT = 0, BACKUP, MASTER } sc_state; 103 104 int sc_flags_backup; 105 int sc_suppress; 106 107 int sc_sendad_errors; 108 #define CARP_SENDAD_MAX_ERRORS 3 109 int sc_sendad_success; 110 #define CARP_SENDAD_MIN_SUCCESS 3 111 112 int sc_vhid; 113 int sc_advskew; 114 int sc_naddrs; 115 int sc_naddrs6; 116 int sc_advbase; /* seconds */ 117 int sc_init_counter; 118 u_int64_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 struct callout sc_ad_tmo; /* advertisement timeout */ 127 struct callout sc_md_tmo; /* master down timeout */ 128 struct callout sc_md6_tmo; /* master down timeout */ 129 130 LIST_ENTRY(carp_softc) sc_next; /* Interface clue */ 131 }; 132 #define SC2IFP(sc) ((sc)->sc_ifp) 133 134 int carp_suppress_preempt = 0; 135 int carp_opts[CARPCTL_MAXID] = { 0, 1, 0, 1, 0, 0 }; /* XXX for now */ 136 SYSCTL_INT(_net_inet_carp, CARPCTL_ALLOW, allow, CTLFLAG_RW, 137 &carp_opts[CARPCTL_ALLOW], 0, "Accept incoming CARP packets"); 138 SYSCTL_INT(_net_inet_carp, CARPCTL_PREEMPT, preempt, CTLFLAG_RW, 139 &carp_opts[CARPCTL_PREEMPT], 0, "high-priority backup preemption mode"); 140 SYSCTL_INT(_net_inet_carp, CARPCTL_LOG, log, CTLFLAG_RW, 141 &carp_opts[CARPCTL_LOG], 0, "log bad carp packets"); 142 SYSCTL_INT(_net_inet_carp, CARPCTL_ARPBALANCE, arpbalance, CTLFLAG_RW, 143 &carp_opts[CARPCTL_ARPBALANCE], 0, "balance arp responses"); 144 SYSCTL_INT(_net_inet_carp, OID_AUTO, suppress_preempt, CTLFLAG_RD, 145 &carp_suppress_preempt, 0, "Preemption is suppressed"); 146 147 struct carpstats carpstats; 148 SYSCTL_STRUCT(_net_inet_carp, CARPCTL_STATS, stats, CTLFLAG_RW, 149 &carpstats, carpstats, 150 "CARP statistics (struct carpstats, netinet/ip_carp.h)"); 151 152 struct carp_if { 153 TAILQ_HEAD(, carp_softc) vhif_vrs; 154 int vhif_nvrs; 155 156 struct ifnet *vhif_ifp; 157 struct mtx vhif_mtx; 158 }; 159 160 /* Get carp_if from softc. Valid after carp_set_addr{,6}. */ 161 #define SC2CIF(sc) ((struct carp_if *)(sc)->sc_carpdev->if_carp) 162 163 /* lock per carp_if queue */ 164 #define CARP_LOCK_INIT(cif) mtx_init(&(cif)->vhif_mtx, "carp_if", \ 165 NULL, MTX_DEF) 166 #define CARP_LOCK_DESTROY(cif) mtx_destroy(&(cif)->vhif_mtx) 167 #define CARP_LOCK_ASSERT(cif) mtx_assert(&(cif)->vhif_mtx, MA_OWNED) 168 #define CARP_LOCK(cif) mtx_lock(&(cif)->vhif_mtx) 169 #define CARP_UNLOCK(cif) mtx_unlock(&(cif)->vhif_mtx) 170 171 #define CARP_SCLOCK(sc) mtx_lock(&SC2CIF(sc)->vhif_mtx) 172 #define CARP_SCUNLOCK(sc) mtx_unlock(&SC2CIF(sc)->vhif_mtx) 173 #define CARP_SCLOCK_ASSERT(sc) mtx_assert(&SC2CIF(sc)->vhif_mtx, MA_OWNED) 174 175 #define CARP_LOG(...) do { \ 176 if (carp_opts[CARPCTL_LOG] > 0) \ 177 log(LOG_INFO, __VA_ARGS__); \ 178 } while (0) 179 180 #define CARP_DEBUG(...) do { \ 181 if (carp_opts[CARPCTL_LOG] > 1) \ 182 log(LOG_DEBUG, __VA_ARGS__); \ 183 } while (0) 184 185 static void carp_hmac_prepare(struct carp_softc *); 186 static void carp_hmac_generate(struct carp_softc *, u_int32_t *, 187 unsigned char *); 188 static int carp_hmac_verify(struct carp_softc *, u_int32_t *, 189 unsigned char *); 190 static void carp_setroute(struct carp_softc *, int); 191 static void carp_input_c(struct mbuf *, struct carp_header *, sa_family_t); 192 static int carp_clone_create(struct if_clone *, int); 193 static void carp_clone_destroy(struct ifnet *); 194 static void carpdetach(struct carp_softc *); 195 static int carp_prepare_ad(struct mbuf *, struct carp_softc *, 196 struct carp_header *); 197 static void carp_send_ad_all(void); 198 static void carp_send_ad(void *); 199 static void carp_send_ad_locked(struct carp_softc *); 200 static void carp_send_arp(struct carp_softc *); 201 static void carp_master_down(void *); 202 static void carp_master_down_locked(struct carp_softc *); 203 static int carp_ioctl(struct ifnet *, u_long, caddr_t); 204 static int carp_looutput(struct ifnet *, struct mbuf *, struct sockaddr *, 205 struct rtentry *); 206 static void carp_start(struct ifnet *); 207 static void carp_setrun(struct carp_softc *, sa_family_t); 208 static void carp_set_state(struct carp_softc *, int); 209 static int carp_addrcount(struct carp_if *, struct in_ifaddr *, int); 210 enum { CARP_COUNT_MASTER, CARP_COUNT_RUNNING }; 211 212 static int carp_set_addr(struct carp_softc *, struct sockaddr_in *); 213 static int carp_del_addr(struct carp_softc *, struct sockaddr_in *); 214 static void carp_carpdev_state_locked(struct carp_if *); 215 static void carp_sc_state_locked(struct carp_softc *); 216 #ifdef INET6 217 static void carp_send_na(struct carp_softc *); 218 static int carp_set_addr6(struct carp_softc *, struct sockaddr_in6 *); 219 static int carp_del_addr6(struct carp_softc *, struct sockaddr_in6 *); 220 #endif 221 222 static LIST_HEAD(, carp_softc) carpif_list; 223 static struct mtx carp_mtx; 224 IFC_SIMPLE_DECLARE(carp, 0); 225 226 static __inline u_int16_t 227 carp_cksum(struct mbuf *m, int len) 228 { 229 return (in_cksum(m, len)); 230 } 231 232 static void 233 carp_hmac_prepare(struct carp_softc *sc) 234 { 235 u_int8_t version = CARP_VERSION, type = CARP_ADVERTISEMENT; 236 u_int8_t vhid = sc->sc_vhid & 0xff; 237 struct ifaddr *ifa; 238 int i; 239 #ifdef INET6 240 struct in6_addr in6; 241 #endif 242 243 if (sc->sc_carpdev) 244 CARP_SCLOCK(sc); 245 246 /* XXX: possible race here */ 247 248 /* compute ipad from key */ 249 bzero(sc->sc_pad, sizeof(sc->sc_pad)); 250 bcopy(sc->sc_key, sc->sc_pad, sizeof(sc->sc_key)); 251 for (i = 0; i < sizeof(sc->sc_pad); i++) 252 sc->sc_pad[i] ^= 0x36; 253 254 /* precompute first part of inner hash */ 255 SHA1Init(&sc->sc_sha1); 256 SHA1Update(&sc->sc_sha1, sc->sc_pad, sizeof(sc->sc_pad)); 257 SHA1Update(&sc->sc_sha1, (void *)&version, sizeof(version)); 258 SHA1Update(&sc->sc_sha1, (void *)&type, sizeof(type)); 259 SHA1Update(&sc->sc_sha1, (void *)&vhid, sizeof(vhid)); 260 #ifdef INET 261 TAILQ_FOREACH(ifa, &SC2IFP(sc)->if_addrlist, ifa_list) { 262 if (ifa->ifa_addr->sa_family == AF_INET) 263 SHA1Update(&sc->sc_sha1, 264 (void *)&ifatoia(ifa)->ia_addr.sin_addr.s_addr, 265 sizeof(struct in_addr)); 266 } 267 #endif /* INET */ 268 #ifdef INET6 269 TAILQ_FOREACH(ifa, &SC2IFP(sc)->if_addrlist, ifa_list) { 270 if (ifa->ifa_addr->sa_family == AF_INET6) { 271 in6 = ifatoia6(ifa)->ia_addr.sin6_addr; 272 in6_clearscope(&in6); 273 SHA1Update(&sc->sc_sha1, (void *)&in6, sizeof(in6)); 274 } 275 } 276 #endif /* INET6 */ 277 278 /* convert ipad to opad */ 279 for (i = 0; i < sizeof(sc->sc_pad); i++) 280 sc->sc_pad[i] ^= 0x36 ^ 0x5c; 281 282 if (sc->sc_carpdev) 283 CARP_SCUNLOCK(sc); 284 } 285 286 static void 287 carp_hmac_generate(struct carp_softc *sc, u_int32_t counter[2], 288 unsigned char md[20]) 289 { 290 SHA1_CTX sha1ctx; 291 292 /* fetch first half of inner hash */ 293 bcopy(&sc->sc_sha1, &sha1ctx, sizeof(sha1ctx)); 294 295 SHA1Update(&sha1ctx, (void *)counter, sizeof(sc->sc_counter)); 296 SHA1Final(md, &sha1ctx); 297 298 /* outer hash */ 299 SHA1Init(&sha1ctx); 300 SHA1Update(&sha1ctx, sc->sc_pad, sizeof(sc->sc_pad)); 301 SHA1Update(&sha1ctx, md, 20); 302 SHA1Final(md, &sha1ctx); 303 } 304 305 static int 306 carp_hmac_verify(struct carp_softc *sc, u_int32_t counter[2], 307 unsigned char md[20]) 308 { 309 unsigned char md2[20]; 310 311 CARP_SCLOCK_ASSERT(sc); 312 313 carp_hmac_generate(sc, counter, md2); 314 315 return (bcmp(md, md2, sizeof(md2))); 316 } 317 318 static void 319 carp_setroute(struct carp_softc *sc, int cmd) 320 { 321 struct ifaddr *ifa; 322 int s; 323 324 if (sc->sc_carpdev) 325 CARP_SCLOCK_ASSERT(sc); 326 327 s = splnet(); 328 TAILQ_FOREACH(ifa, &SC2IFP(sc)->if_addrlist, ifa_list) { 329 if (ifa->ifa_addr->sa_family == AF_INET && 330 sc->sc_carpdev != NULL) { 331 int count = carp_addrcount( 332 (struct carp_if *)sc->sc_carpdev->if_carp, 333 ifatoia(ifa), CARP_COUNT_MASTER); 334 335 if ((cmd == RTM_ADD && count == 1) || 336 (cmd == RTM_DELETE && count == 0)) 337 rtinit(ifa, cmd, RTF_UP | RTF_HOST); 338 } 339 #ifdef INET6 340 if (ifa->ifa_addr->sa_family == AF_INET6) { 341 if (cmd == RTM_ADD) 342 in6_ifaddloop(ifa); 343 else 344 in6_ifremloop(ifa); 345 } 346 #endif /* INET6 */ 347 } 348 splx(s); 349 } 350 351 static int 352 carp_clone_create(struct if_clone *ifc, int unit) 353 { 354 355 struct carp_softc *sc; 356 struct ifnet *ifp; 357 358 MALLOC(sc, struct carp_softc *, sizeof(*sc), M_CARP, M_WAITOK|M_ZERO); 359 ifp = SC2IFP(sc) = if_alloc(IFT_ETHER); 360 if (ifp == NULL) { 361 FREE(sc, M_CARP); 362 return (ENOSPC); 363 } 364 365 sc->sc_flags_backup = 0; 366 sc->sc_suppress = 0; 367 sc->sc_advbase = CARP_DFLTINTV; 368 sc->sc_vhid = -1; /* required setting */ 369 sc->sc_advskew = 0; 370 sc->sc_init_counter = 1; 371 sc->sc_naddrs = sc->sc_naddrs6 = 0; /* M_ZERO? */ 372 #ifdef INET6 373 sc->sc_im6o.im6o_multicast_hlim = CARP_DFLTTL; 374 #endif 375 376 callout_init(&sc->sc_ad_tmo, NET_CALLOUT_MPSAFE); 377 callout_init(&sc->sc_md_tmo, NET_CALLOUT_MPSAFE); 378 callout_init(&sc->sc_md6_tmo, NET_CALLOUT_MPSAFE); 379 380 ifp->if_softc = sc; 381 if_initname(ifp, CARP_IFNAME, unit); 382 ifp->if_mtu = ETHERMTU; 383 ifp->if_flags = IFF_LOOPBACK; 384 ifp->if_ioctl = carp_ioctl; 385 ifp->if_output = carp_looutput; 386 ifp->if_start = carp_start; 387 ifp->if_type = IFT_CARP; 388 ifp->if_snd.ifq_maxlen = ifqmaxlen; 389 ifp->if_hdrlen = 0; 390 if_attach(ifp); 391 bpfattach(SC2IFP(sc), DLT_NULL, sizeof(u_int32_t)); 392 mtx_lock(&carp_mtx); 393 LIST_INSERT_HEAD(&carpif_list, sc, sc_next); 394 mtx_unlock(&carp_mtx); 395 return (0); 396 } 397 398 static void 399 carp_clone_destroy(struct ifnet *ifp) 400 { 401 struct carp_softc *sc = ifp->if_softc; 402 struct carp_if *cif; 403 struct ip_moptions *imo = &sc->sc_imo; 404 #ifdef INET6 405 struct ip6_moptions *im6o = &sc->sc_im6o; 406 #endif 407 408 /* carpdetach(sc); */ 409 410 /* 411 * If an interface is destroyed which is suppressing the preemption, 412 * decrease the global counter, otherwise the host will never get 413 * out of the carp supressing state. 414 */ 415 if (sc->sc_suppress) 416 carp_suppress_preempt--; 417 sc->sc_suppress = 0; 418 419 callout_stop(&sc->sc_ad_tmo); 420 callout_stop(&sc->sc_md_tmo); 421 callout_stop(&sc->sc_md6_tmo); 422 423 if (imo->imo_num_memberships) { 424 in_delmulti(imo->imo_membership[--imo->imo_num_memberships]); 425 imo->imo_multicast_ifp = NULL; 426 } 427 #ifdef INET6 428 while (!LIST_EMPTY(&im6o->im6o_memberships)) { 429 struct in6_multi_mship *imm = 430 LIST_FIRST(&im6o->im6o_memberships); 431 LIST_REMOVE(imm, i6mm_chain); 432 in6_leavegroup(imm); 433 } 434 im6o->im6o_multicast_ifp = NULL; 435 #endif 436 437 /* Remove ourself from parents if_carp queue */ 438 if (sc->sc_carpdev && (cif = sc->sc_carpdev->if_carp)) { 439 CARP_LOCK(cif); 440 TAILQ_REMOVE(&cif->vhif_vrs, sc, sc_list); 441 if (!--cif->vhif_nvrs) { 442 sc->sc_carpdev->if_carp = NULL; 443 CARP_LOCK_DESTROY(cif); 444 FREE(cif, M_CARP); 445 ifpromisc(sc->sc_carpdev, 0); 446 sc->sc_carpdev = NULL; 447 } else { 448 CARP_UNLOCK(cif); 449 } 450 } 451 452 mtx_lock(&carp_mtx); 453 LIST_REMOVE(sc, sc_next); 454 mtx_unlock(&carp_mtx); 455 bpfdetach(ifp); 456 if_detach(ifp); 457 if_free_type(ifp, IFT_ETHER); 458 free(sc, M_CARP); 459 } 460 461 /* 462 * process input packet. 463 * we have rearranged checks order compared to the rfc, 464 * but it seems more efficient this way or not possible otherwise. 465 */ 466 void 467 carp_input(struct mbuf *m, int hlen) 468 { 469 struct ip *ip = mtod(m, struct ip *); 470 struct carp_header *ch; 471 int iplen, len; 472 473 carpstats.carps_ipackets++; 474 475 if (!carp_opts[CARPCTL_ALLOW]) { 476 m_freem(m); 477 return; 478 } 479 480 /* check if received on a valid carp interface */ 481 if (m->m_pkthdr.rcvif->if_carp == NULL) { 482 carpstats.carps_badif++; 483 CARP_LOG("carp_input: packet received on non-carp " 484 "interface: %s\n", 485 m->m_pkthdr.rcvif->if_xname); 486 m_freem(m); 487 return; 488 } 489 490 /* verify that the IP TTL is 255. */ 491 if (ip->ip_ttl != CARP_DFLTTL) { 492 carpstats.carps_badttl++; 493 CARP_LOG("carp_input: received ttl %d != 255i on %s\n", 494 ip->ip_ttl, 495 m->m_pkthdr.rcvif->if_xname); 496 m_freem(m); 497 return; 498 } 499 500 iplen = ip->ip_hl << 2; 501 502 if (m->m_pkthdr.len < iplen + sizeof(*ch)) { 503 carpstats.carps_badlen++; 504 CARP_LOG("carp_input: received len %zd < " 505 "sizeof(struct carp_header)\n", 506 m->m_len - sizeof(struct ip)); 507 m_freem(m); 508 return; 509 } 510 511 if (iplen + sizeof(*ch) < m->m_len) { 512 if ((m = m_pullup(m, iplen + sizeof(*ch))) == NULL) { 513 carpstats.carps_hdrops++; 514 CARP_LOG("carp_input: pullup failed\n"); 515 return; 516 } 517 ip = mtod(m, struct ip *); 518 } 519 ch = (struct carp_header *)((char *)ip + iplen); 520 521 /* 522 * verify that the received packet length is 523 * equal to the CARP header 524 */ 525 len = iplen + sizeof(*ch); 526 if (len > m->m_pkthdr.len) { 527 carpstats.carps_badlen++; 528 CARP_LOG("carp_input: packet too short %d on %s\n", 529 m->m_pkthdr.len, 530 m->m_pkthdr.rcvif->if_xname); 531 m_freem(m); 532 return; 533 } 534 535 if ((m = m_pullup(m, len)) == NULL) { 536 carpstats.carps_hdrops++; 537 return; 538 } 539 ip = mtod(m, struct ip *); 540 ch = (struct carp_header *)((char *)ip + iplen); 541 542 /* verify the CARP checksum */ 543 m->m_data += iplen; 544 if (carp_cksum(m, len - iplen)) { 545 carpstats.carps_badsum++; 546 CARP_LOG("carp_input: checksum failed on %s\n", 547 m->m_pkthdr.rcvif->if_xname); 548 m_freem(m); 549 return; 550 } 551 m->m_data -= iplen; 552 553 carp_input_c(m, ch, AF_INET); 554 } 555 556 #ifdef INET6 557 int 558 carp6_input(struct mbuf **mp, int *offp, int proto) 559 { 560 struct mbuf *m = *mp; 561 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 562 struct carp_header *ch; 563 u_int len; 564 565 carpstats.carps_ipackets6++; 566 567 if (!carp_opts[CARPCTL_ALLOW]) { 568 m_freem(m); 569 return (IPPROTO_DONE); 570 } 571 572 /* check if received on a valid carp interface */ 573 if (m->m_pkthdr.rcvif->if_carp == NULL) { 574 carpstats.carps_badif++; 575 CARP_LOG("carp6_input: packet received on non-carp " 576 "interface: %s\n", 577 m->m_pkthdr.rcvif->if_xname); 578 m_freem(m); 579 return (IPPROTO_DONE); 580 } 581 582 /* verify that the IP TTL is 255 */ 583 if (ip6->ip6_hlim != CARP_DFLTTL) { 584 carpstats.carps_badttl++; 585 CARP_LOG("carp6_input: received ttl %d != 255 on %s\n", 586 ip6->ip6_hlim, 587 m->m_pkthdr.rcvif->if_xname); 588 m_freem(m); 589 return (IPPROTO_DONE); 590 } 591 592 /* verify that we have a complete carp packet */ 593 len = m->m_len; 594 IP6_EXTHDR_GET(ch, struct carp_header *, m, *offp, sizeof(*ch)); 595 if (ch == NULL) { 596 carpstats.carps_badlen++; 597 CARP_LOG("carp6_input: packet size %u too small\n", len); 598 return (IPPROTO_DONE); 599 } 600 601 602 /* verify the CARP checksum */ 603 m->m_data += *offp; 604 if (carp_cksum(m, sizeof(*ch))) { 605 carpstats.carps_badsum++; 606 CARP_LOG("carp6_input: checksum failed, on %s\n", 607 m->m_pkthdr.rcvif->if_xname); 608 m_freem(m); 609 return (IPPROTO_DONE); 610 } 611 m->m_data -= *offp; 612 613 carp_input_c(m, ch, AF_INET6); 614 return (IPPROTO_DONE); 615 } 616 #endif /* INET6 */ 617 618 static void 619 carp_input_c(struct mbuf *m, struct carp_header *ch, sa_family_t af) 620 { 621 struct ifnet *ifp = m->m_pkthdr.rcvif; 622 struct carp_softc *sc; 623 u_int64_t tmp_counter; 624 struct timeval sc_tv, ch_tv; 625 626 /* verify that the VHID is valid on the receiving interface */ 627 CARP_LOCK(ifp->if_carp); 628 TAILQ_FOREACH(sc, &((struct carp_if *)ifp->if_carp)->vhif_vrs, sc_list) 629 if (sc->sc_vhid == ch->carp_vhid) 630 break; 631 632 if (!sc || !((SC2IFP(sc)->if_flags & IFF_UP) && 633 (SC2IFP(sc)->if_drv_flags & IFF_DRV_RUNNING))) { 634 carpstats.carps_badvhid++; 635 CARP_UNLOCK(ifp->if_carp); 636 m_freem(m); 637 return; 638 } 639 640 getmicrotime(&SC2IFP(sc)->if_lastchange); 641 SC2IFP(sc)->if_ipackets++; 642 SC2IFP(sc)->if_ibytes += m->m_pkthdr.len; 643 644 if (SC2IFP(sc)->if_bpf) { 645 struct ip *ip = mtod(m, struct ip *); 646 uint32_t af1 = af; 647 648 /* BPF wants net byte order */ 649 ip->ip_len = htons(ip->ip_len + (ip->ip_hl << 2)); 650 ip->ip_off = htons(ip->ip_off); 651 bpf_mtap2(SC2IFP(sc)->if_bpf, &af1, sizeof(af1), m); 652 } 653 654 /* verify the CARP version. */ 655 if (ch->carp_version != CARP_VERSION) { 656 carpstats.carps_badver++; 657 SC2IFP(sc)->if_ierrors++; 658 CARP_UNLOCK(ifp->if_carp); 659 CARP_LOG("%s; invalid version %d\n", 660 SC2IFP(sc)->if_xname, 661 ch->carp_version); 662 m_freem(m); 663 return; 664 } 665 666 /* verify the hash */ 667 if (carp_hmac_verify(sc, ch->carp_counter, ch->carp_md)) { 668 carpstats.carps_badauth++; 669 SC2IFP(sc)->if_ierrors++; 670 CARP_UNLOCK(ifp->if_carp); 671 CARP_LOG("%s: incorrect hash\n", SC2IFP(sc)->if_xname); 672 m_freem(m); 673 return; 674 } 675 676 tmp_counter = ntohl(ch->carp_counter[0]); 677 tmp_counter = tmp_counter<<32; 678 tmp_counter += ntohl(ch->carp_counter[1]); 679 680 /* XXX Replay protection goes here */ 681 682 sc->sc_init_counter = 0; 683 sc->sc_counter = tmp_counter; 684 685 sc_tv.tv_sec = sc->sc_advbase; 686 if (carp_suppress_preempt && sc->sc_advskew < 240) 687 sc_tv.tv_usec = 240 * 1000000 / 256; 688 else 689 sc_tv.tv_usec = sc->sc_advskew * 1000000 / 256; 690 ch_tv.tv_sec = ch->carp_advbase; 691 ch_tv.tv_usec = ch->carp_advskew * 1000000 / 256; 692 693 switch (sc->sc_state) { 694 case INIT: 695 break; 696 case MASTER: 697 /* 698 * If we receive an advertisement from a master who's going to 699 * be more frequent than us, go into BACKUP state. 700 */ 701 if (timevalcmp(&sc_tv, &ch_tv, >) || 702 timevalcmp(&sc_tv, &ch_tv, ==)) { 703 callout_stop(&sc->sc_ad_tmo); 704 CARP_DEBUG("%s: MASTER -> BACKUP " 705 "(more frequent advertisement received)\n", 706 SC2IFP(sc)->if_xname); 707 carp_set_state(sc, BACKUP); 708 carp_setrun(sc, 0); 709 carp_setroute(sc, RTM_DELETE); 710 } 711 break; 712 case BACKUP: 713 /* 714 * If we're pre-empting masters who advertise slower than us, 715 * and this one claims to be slower, treat him as down. 716 */ 717 if (carp_opts[CARPCTL_PREEMPT] && 718 timevalcmp(&sc_tv, &ch_tv, <)) { 719 CARP_DEBUG("%s: BACKUP -> MASTER " 720 "(preempting a slower master)\n", 721 SC2IFP(sc)->if_xname); 722 carp_master_down_locked(sc); 723 break; 724 } 725 726 /* 727 * If the master is going to advertise at such a low frequency 728 * that he's guaranteed to time out, we'd might as well just 729 * treat him as timed out now. 730 */ 731 sc_tv.tv_sec = sc->sc_advbase * 3; 732 if (timevalcmp(&sc_tv, &ch_tv, <)) { 733 CARP_DEBUG("%s: BACKUP -> MASTER " 734 "(master timed out)\n", 735 SC2IFP(sc)->if_xname); 736 carp_master_down_locked(sc); 737 break; 738 } 739 740 /* 741 * Otherwise, we reset the counter and wait for the next 742 * advertisement. 743 */ 744 carp_setrun(sc, af); 745 break; 746 } 747 748 CARP_UNLOCK(ifp->if_carp); 749 750 m_freem(m); 751 return; 752 } 753 754 static void 755 carpdetach(struct carp_softc *sc) 756 { 757 struct ifaddr *ifa; 758 759 callout_stop(&sc->sc_ad_tmo); 760 callout_stop(&sc->sc_md_tmo); 761 callout_stop(&sc->sc_md6_tmo); 762 763 while ((ifa = TAILQ_FIRST(&SC2IFP(sc)->if_addrlist)) != NULL) 764 if (ifa->ifa_addr->sa_family == AF_INET) { 765 struct in_ifaddr *ia = ifatoia(ifa); 766 767 carp_del_addr(sc, &ia->ia_addr); 768 769 /* ripped screaming from in_control(SIOCDIFADDR) */ 770 in_ifscrub(SC2IFP(sc), ia); 771 TAILQ_REMOVE(&SC2IFP(sc)->if_addrlist, ifa, ifa_link); 772 TAILQ_REMOVE(&in_ifaddrhead, ia, ia_link); 773 IFAFREE((&ia->ia_ifa)); 774 } 775 } 776 777 /* Detach an interface from the carp. */ 778 void 779 carp_ifdetach(struct ifnet *ifp) 780 { 781 struct carp_softc *sc; 782 struct carp_if *cif = (struct carp_if *)ifp->if_carp; 783 784 CARP_LOCK(cif); 785 TAILQ_FOREACH(sc, &cif->vhif_vrs, sc_list) 786 carpdetach(sc); 787 CARP_UNLOCK(cif); 788 } 789 790 static int 791 carp_prepare_ad(struct mbuf *m, struct carp_softc *sc, struct carp_header *ch) 792 { 793 struct m_tag *mtag; 794 struct ifnet *ifp = SC2IFP(sc); 795 796 if (sc->sc_init_counter) { 797 /* this could also be seconds since unix epoch */ 798 sc->sc_counter = arc4random(); 799 sc->sc_counter = sc->sc_counter << 32; 800 sc->sc_counter += arc4random(); 801 } else 802 sc->sc_counter++; 803 804 ch->carp_counter[0] = htonl((sc->sc_counter>>32)&0xffffffff); 805 ch->carp_counter[1] = htonl(sc->sc_counter&0xffffffff); 806 807 carp_hmac_generate(sc, ch->carp_counter, ch->carp_md); 808 809 /* Tag packet for carp_output */ 810 mtag = m_tag_get(PACKET_TAG_CARP, sizeof(struct ifnet *), M_NOWAIT); 811 if (mtag == NULL) { 812 m_freem(m); 813 SC2IFP(sc)->if_oerrors++; 814 return (ENOMEM); 815 } 816 bcopy(&ifp, (caddr_t)(mtag + 1), sizeof(struct ifnet *)); 817 m_tag_prepend(m, mtag); 818 819 return (0); 820 } 821 822 static void 823 carp_send_ad_all(void) 824 { 825 struct carp_softc *sc; 826 827 mtx_lock(&carp_mtx); 828 LIST_FOREACH(sc, &carpif_list, sc_next) { 829 if (sc->sc_carpdev == NULL) 830 continue; 831 CARP_SCLOCK(sc); 832 if ((SC2IFP(sc)->if_flags & IFF_UP) && 833 (SC2IFP(sc)->if_drv_flags & IFF_DRV_RUNNING) && 834 sc->sc_state == MASTER) 835 carp_send_ad_locked(sc); 836 CARP_SCUNLOCK(sc); 837 } 838 mtx_unlock(&carp_mtx); 839 } 840 841 static void 842 carp_send_ad(void *v) 843 { 844 struct carp_softc *sc = v; 845 846 CARP_SCLOCK(sc); 847 carp_send_ad_locked(sc); 848 CARP_SCUNLOCK(sc); 849 } 850 851 static void 852 carp_send_ad_locked(struct carp_softc *sc) 853 { 854 struct carp_header ch; 855 struct timeval tv; 856 struct carp_header *ch_ptr; 857 struct mbuf *m; 858 int len, advbase, advskew; 859 860 CARP_SCLOCK_ASSERT(sc); 861 862 /* bow out if we've lost our UPness or RUNNINGuiness */ 863 if (!((SC2IFP(sc)->if_flags & IFF_UP) && 864 (SC2IFP(sc)->if_drv_flags & IFF_DRV_RUNNING))) { 865 advbase = 255; 866 advskew = 255; 867 } else { 868 advbase = sc->sc_advbase; 869 if (!carp_suppress_preempt || sc->sc_advskew > 240) 870 advskew = sc->sc_advskew; 871 else 872 advskew = 240; 873 tv.tv_sec = advbase; 874 tv.tv_usec = advskew * 1000000 / 256; 875 } 876 877 ch.carp_version = CARP_VERSION; 878 ch.carp_type = CARP_ADVERTISEMENT; 879 ch.carp_vhid = sc->sc_vhid; 880 ch.carp_advbase = advbase; 881 ch.carp_advskew = advskew; 882 ch.carp_authlen = 7; /* XXX DEFINE */ 883 ch.carp_pad1 = 0; /* must be zero */ 884 ch.carp_cksum = 0; 885 886 #ifdef INET 887 if (sc->sc_ia) { 888 struct ip *ip; 889 890 MGETHDR(m, M_DONTWAIT, MT_HEADER); 891 if (m == NULL) { 892 SC2IFP(sc)->if_oerrors++; 893 carpstats.carps_onomem++; 894 /* XXX maybe less ? */ 895 if (advbase != 255 || advskew != 255) 896 callout_reset(&sc->sc_ad_tmo, tvtohz(&tv), 897 carp_send_ad, sc); 898 return; 899 } 900 len = sizeof(*ip) + sizeof(ch); 901 m->m_pkthdr.len = len; 902 m->m_pkthdr.rcvif = NULL; 903 m->m_len = len; 904 MH_ALIGN(m, m->m_len); 905 m->m_flags |= M_MCAST; 906 ip = mtod(m, struct ip *); 907 ip->ip_v = IPVERSION; 908 ip->ip_hl = sizeof(*ip) >> 2; 909 ip->ip_tos = IPTOS_LOWDELAY; 910 ip->ip_len = len; 911 ip->ip_id = ip_newid(); 912 ip->ip_off = IP_DF; 913 ip->ip_ttl = CARP_DFLTTL; 914 ip->ip_p = IPPROTO_CARP; 915 ip->ip_sum = 0; 916 ip->ip_src.s_addr = sc->sc_ia->ia_addr.sin_addr.s_addr; 917 ip->ip_dst.s_addr = htonl(INADDR_CARP_GROUP); 918 919 ch_ptr = (struct carp_header *)(&ip[1]); 920 bcopy(&ch, ch_ptr, sizeof(ch)); 921 if (carp_prepare_ad(m, sc, ch_ptr)) 922 return; 923 924 m->m_data += sizeof(*ip); 925 ch_ptr->carp_cksum = carp_cksum(m, len - sizeof(*ip)); 926 m->m_data -= sizeof(*ip); 927 928 getmicrotime(&SC2IFP(sc)->if_lastchange); 929 SC2IFP(sc)->if_opackets++; 930 SC2IFP(sc)->if_obytes += len; 931 carpstats.carps_opackets++; 932 933 if (ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo, NULL)) { 934 SC2IFP(sc)->if_oerrors++; 935 if (sc->sc_sendad_errors < INT_MAX) 936 sc->sc_sendad_errors++; 937 if (sc->sc_sendad_errors == CARP_SENDAD_MAX_ERRORS) { 938 carp_suppress_preempt++; 939 if (carp_suppress_preempt == 1) { 940 CARP_SCUNLOCK(sc); 941 carp_send_ad_all(); 942 CARP_SCLOCK(sc); 943 } 944 } 945 sc->sc_sendad_success = 0; 946 } else { 947 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) { 948 if (++sc->sc_sendad_success >= 949 CARP_SENDAD_MIN_SUCCESS) { 950 carp_suppress_preempt--; 951 sc->sc_sendad_errors = 0; 952 } 953 } else 954 sc->sc_sendad_errors = 0; 955 } 956 } 957 #endif /* INET */ 958 #ifdef INET6 959 if (sc->sc_ia6) { 960 struct ip6_hdr *ip6; 961 962 MGETHDR(m, M_DONTWAIT, MT_HEADER); 963 if (m == NULL) { 964 SC2IFP(sc)->if_oerrors++; 965 carpstats.carps_onomem++; 966 /* XXX maybe less ? */ 967 if (advbase != 255 || advskew != 255) 968 callout_reset(&sc->sc_ad_tmo, tvtohz(&tv), 969 carp_send_ad, sc); 970 return; 971 } 972 len = sizeof(*ip6) + sizeof(ch); 973 m->m_pkthdr.len = len; 974 m->m_pkthdr.rcvif = NULL; 975 m->m_len = len; 976 MH_ALIGN(m, m->m_len); 977 m->m_flags |= M_MCAST; 978 ip6 = mtod(m, struct ip6_hdr *); 979 bzero(ip6, sizeof(*ip6)); 980 ip6->ip6_vfc |= IPV6_VERSION; 981 ip6->ip6_hlim = CARP_DFLTTL; 982 ip6->ip6_nxt = IPPROTO_CARP; 983 bcopy(&sc->sc_ia6->ia_addr.sin6_addr, &ip6->ip6_src, 984 sizeof(struct in6_addr)); 985 /* set the multicast destination */ 986 987 ip6->ip6_dst.s6_addr8[0] = 0xff; 988 ip6->ip6_dst.s6_addr8[1] = 0x02; 989 ip6->ip6_dst.s6_addr8[15] = 0x12; 990 991 ch_ptr = (struct carp_header *)(&ip6[1]); 992 bcopy(&ch, ch_ptr, sizeof(ch)); 993 if (carp_prepare_ad(m, sc, ch_ptr)) 994 return; 995 996 m->m_data += sizeof(*ip6); 997 ch_ptr->carp_cksum = carp_cksum(m, len - sizeof(*ip6)); 998 m->m_data -= sizeof(*ip6); 999 1000 getmicrotime(&SC2IFP(sc)->if_lastchange); 1001 SC2IFP(sc)->if_opackets++; 1002 SC2IFP(sc)->if_obytes += len; 1003 carpstats.carps_opackets6++; 1004 1005 if (ip6_output(m, NULL, NULL, 0, &sc->sc_im6o, NULL, NULL)) { 1006 SC2IFP(sc)->if_oerrors++; 1007 if (sc->sc_sendad_errors < INT_MAX) 1008 sc->sc_sendad_errors++; 1009 if (sc->sc_sendad_errors == CARP_SENDAD_MAX_ERRORS) { 1010 carp_suppress_preempt++; 1011 if (carp_suppress_preempt == 1) { 1012 CARP_SCUNLOCK(sc); 1013 carp_send_ad_all(); 1014 CARP_SCLOCK(sc); 1015 } 1016 } 1017 sc->sc_sendad_success = 0; 1018 } else { 1019 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) { 1020 if (++sc->sc_sendad_success >= 1021 CARP_SENDAD_MIN_SUCCESS) { 1022 carp_suppress_preempt--; 1023 sc->sc_sendad_errors = 0; 1024 } 1025 } else 1026 sc->sc_sendad_errors = 0; 1027 } 1028 } 1029 #endif /* INET6 */ 1030 1031 if (advbase != 255 || advskew != 255) 1032 callout_reset(&sc->sc_ad_tmo, tvtohz(&tv), 1033 carp_send_ad, sc); 1034 1035 } 1036 1037 /* 1038 * Broadcast a gratuitous ARP request containing 1039 * the virtual router MAC address for each IP address 1040 * associated with the virtual router. 1041 */ 1042 static void 1043 carp_send_arp(struct carp_softc *sc) 1044 { 1045 struct ifaddr *ifa; 1046 1047 TAILQ_FOREACH(ifa, &SC2IFP(sc)->if_addrlist, ifa_list) { 1048 1049 if (ifa->ifa_addr->sa_family != AF_INET) 1050 continue; 1051 1052 /* arprequest(sc->sc_carpdev, &in, &in, IFP2ENADDR(sc->sc_ifp)); */ 1053 arp_ifinit2(sc->sc_carpdev, ifa, IFP2ENADDR(sc->sc_ifp)); 1054 1055 DELAY(1000); /* XXX */ 1056 } 1057 } 1058 1059 #ifdef INET6 1060 static void 1061 carp_send_na(struct carp_softc *sc) 1062 { 1063 struct ifaddr *ifa; 1064 struct in6_addr *in6; 1065 static struct in6_addr mcast = IN6ADDR_LINKLOCAL_ALLNODES_INIT; 1066 1067 TAILQ_FOREACH(ifa, &SC2IFP(sc)->if_addrlist, ifa_list) { 1068 1069 if (ifa->ifa_addr->sa_family != AF_INET6) 1070 continue; 1071 1072 in6 = &ifatoia6(ifa)->ia_addr.sin6_addr; 1073 nd6_na_output(sc->sc_carpdev, &mcast, in6, 1074 ND_NA_FLAG_OVERRIDE, 1, NULL); 1075 DELAY(1000); /* XXX */ 1076 } 1077 } 1078 #endif /* INET6 */ 1079 1080 static int 1081 carp_addrcount(struct carp_if *cif, struct in_ifaddr *ia, int type) 1082 { 1083 struct carp_softc *vh; 1084 struct ifaddr *ifa; 1085 int count = 0; 1086 1087 CARP_LOCK_ASSERT(cif); 1088 1089 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 1090 if ((type == CARP_COUNT_RUNNING && 1091 (SC2IFP(vh)->if_flags & IFF_UP) && 1092 (SC2IFP(vh)->if_drv_flags & IFF_DRV_RUNNING)) || 1093 (type == CARP_COUNT_MASTER && vh->sc_state == MASTER)) { 1094 TAILQ_FOREACH(ifa, &SC2IFP(vh)->if_addrlist, 1095 ifa_list) { 1096 if (ifa->ifa_addr->sa_family == AF_INET && 1097 ia->ia_addr.sin_addr.s_addr == 1098 ifatoia(ifa)->ia_addr.sin_addr.s_addr) 1099 count++; 1100 } 1101 } 1102 } 1103 return (count); 1104 } 1105 1106 int 1107 carp_iamatch(void *v, struct in_ifaddr *ia, 1108 struct in_addr *isaddr, u_int8_t **enaddr) 1109 { 1110 struct carp_if *cif = v; 1111 struct carp_softc *vh; 1112 int index, count = 0; 1113 struct ifaddr *ifa; 1114 1115 CARP_LOCK(cif); 1116 1117 if (carp_opts[CARPCTL_ARPBALANCE]) { 1118 /* 1119 * XXX proof of concept implementation. 1120 * We use the source ip to decide which virtual host should 1121 * handle the request. If we're master of that virtual host, 1122 * then we respond, otherwise, just drop the arp packet on 1123 * the floor. 1124 */ 1125 count = carp_addrcount(cif, ia, CARP_COUNT_RUNNING); 1126 if (count == 0) { 1127 /* should never reach this */ 1128 CARP_UNLOCK(cif); 1129 return (0); 1130 } 1131 1132 /* this should be a hash, like pf_hash() */ 1133 index = ntohl(isaddr->s_addr) % count; 1134 count = 0; 1135 1136 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 1137 if ((SC2IFP(vh)->if_flags & IFF_UP) && 1138 (SC2IFP(vh)->if_drv_flags & IFF_DRV_RUNNING)) { 1139 TAILQ_FOREACH(ifa, &SC2IFP(vh)->if_addrlist, 1140 ifa_list) { 1141 if (ifa->ifa_addr->sa_family == 1142 AF_INET && 1143 ia->ia_addr.sin_addr.s_addr == 1144 ifatoia(ifa)->ia_addr.sin_addr.s_addr) { 1145 if (count == index) { 1146 if (vh->sc_state == 1147 MASTER) { 1148 *enaddr = IFP2ENADDR(vh->sc_ifp); 1149 CARP_UNLOCK(cif); 1150 return (1); 1151 } else { 1152 CARP_UNLOCK(cif); 1153 return (0); 1154 } 1155 } 1156 count++; 1157 } 1158 } 1159 } 1160 } 1161 } else { 1162 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 1163 if ((SC2IFP(vh)->if_flags & IFF_UP) && 1164 (SC2IFP(vh)->if_drv_flags & IFF_DRV_RUNNING) && 1165 ia->ia_ifp == SC2IFP(vh) && 1166 vh->sc_state == MASTER) { 1167 *enaddr = IFP2ENADDR(vh->sc_ifp); 1168 CARP_UNLOCK(cif); 1169 return (1); 1170 } 1171 } 1172 } 1173 CARP_UNLOCK(cif); 1174 return (0); 1175 } 1176 1177 #ifdef INET6 1178 struct ifaddr * 1179 carp_iamatch6(void *v, struct in6_addr *taddr) 1180 { 1181 struct carp_if *cif = v; 1182 struct carp_softc *vh; 1183 struct ifaddr *ifa; 1184 1185 CARP_LOCK(cif); 1186 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 1187 TAILQ_FOREACH(ifa, &SC2IFP(vh)->if_addrlist, ifa_list) { 1188 if (IN6_ARE_ADDR_EQUAL(taddr, 1189 &ifatoia6(ifa)->ia_addr.sin6_addr) && 1190 (SC2IFP(vh)->if_flags & IFF_UP) && 1191 (SC2IFP(vh)->if_drv_flags & IFF_DRV_RUNNING) && 1192 vh->sc_state == MASTER) { 1193 CARP_UNLOCK(cif); 1194 return (ifa); 1195 } 1196 } 1197 } 1198 CARP_UNLOCK(cif); 1199 1200 return (NULL); 1201 } 1202 1203 void * 1204 carp_macmatch6(void *v, struct mbuf *m, const struct in6_addr *taddr) 1205 { 1206 struct m_tag *mtag; 1207 struct carp_if *cif = v; 1208 struct carp_softc *sc; 1209 struct ifaddr *ifa; 1210 1211 CARP_LOCK(cif); 1212 TAILQ_FOREACH(sc, &cif->vhif_vrs, sc_list) { 1213 TAILQ_FOREACH(ifa, &SC2IFP(sc)->if_addrlist, ifa_list) { 1214 if (IN6_ARE_ADDR_EQUAL(taddr, 1215 &ifatoia6(ifa)->ia_addr.sin6_addr) && 1216 (SC2IFP(sc)->if_flags & IFF_UP) && 1217 (SC2IFP(sc)->if_drv_flags & IFF_DRV_RUNNING)) { 1218 struct ifnet *ifp = SC2IFP(sc); 1219 mtag = m_tag_get(PACKET_TAG_CARP, 1220 sizeof(struct ifnet *), M_NOWAIT); 1221 if (mtag == NULL) { 1222 /* better a bit than nothing */ 1223 CARP_UNLOCK(cif); 1224 return (IFP2ENADDR(sc->sc_ifp)); 1225 } 1226 bcopy(&ifp, (caddr_t)(mtag + 1), 1227 sizeof(struct ifnet *)); 1228 m_tag_prepend(m, mtag); 1229 1230 CARP_UNLOCK(cif); 1231 return (IFP2ENADDR(sc->sc_ifp)); 1232 } 1233 } 1234 } 1235 CARP_UNLOCK(cif); 1236 1237 return (NULL); 1238 } 1239 #endif 1240 1241 struct ifnet * 1242 carp_forus(void *v, void *dhost) 1243 { 1244 struct carp_if *cif = v; 1245 struct carp_softc *vh; 1246 u_int8_t *ena = dhost; 1247 1248 if (ena[0] || ena[1] || ena[2] != 0x5e || ena[3] || ena[4] != 1) 1249 return (NULL); 1250 1251 CARP_LOCK(cif); 1252 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) 1253 if ((SC2IFP(vh)->if_flags & IFF_UP) && 1254 (SC2IFP(vh)->if_drv_flags & IFF_DRV_RUNNING) && 1255 vh->sc_state == MASTER && 1256 !bcmp(dhost, IFP2ENADDR(vh->sc_ifp), ETHER_ADDR_LEN)) { 1257 CARP_UNLOCK(cif); 1258 return (SC2IFP(vh)); 1259 } 1260 1261 CARP_UNLOCK(cif); 1262 return (NULL); 1263 } 1264 1265 static void 1266 carp_master_down(void *v) 1267 { 1268 struct carp_softc *sc = v; 1269 1270 CARP_SCLOCK(sc); 1271 carp_master_down_locked(sc); 1272 CARP_SCUNLOCK(sc); 1273 } 1274 1275 static void 1276 carp_master_down_locked(struct carp_softc *sc) 1277 { 1278 if (sc->sc_carpdev) 1279 CARP_SCLOCK_ASSERT(sc); 1280 1281 switch (sc->sc_state) { 1282 case INIT: 1283 printf("%s: master_down event in INIT state\n", 1284 SC2IFP(sc)->if_xname); 1285 break; 1286 case MASTER: 1287 break; 1288 case BACKUP: 1289 carp_set_state(sc, MASTER); 1290 carp_send_ad_locked(sc); 1291 carp_send_arp(sc); 1292 #ifdef INET6 1293 carp_send_na(sc); 1294 #endif /* INET6 */ 1295 carp_setrun(sc, 0); 1296 carp_setroute(sc, RTM_ADD); 1297 break; 1298 } 1299 } 1300 1301 /* 1302 * When in backup state, af indicates whether to reset the master down timer 1303 * for v4 or v6. If it's set to zero, reset the ones which are already pending. 1304 */ 1305 static void 1306 carp_setrun(struct carp_softc *sc, sa_family_t af) 1307 { 1308 struct timeval tv; 1309 1310 if (sc->sc_carpdev) 1311 CARP_SCLOCK_ASSERT(sc); 1312 1313 if (SC2IFP(sc)->if_flags & IFF_UP && 1314 sc->sc_vhid > 0 && (sc->sc_naddrs || sc->sc_naddrs6)) 1315 SC2IFP(sc)->if_drv_flags |= IFF_DRV_RUNNING; 1316 else { 1317 SC2IFP(sc)->if_drv_flags &= ~IFF_DRV_RUNNING; 1318 carp_setroute(sc, RTM_DELETE); 1319 return; 1320 } 1321 1322 switch (sc->sc_state) { 1323 case INIT: 1324 if (carp_opts[CARPCTL_PREEMPT] && !carp_suppress_preempt) { 1325 carp_send_ad_locked(sc); 1326 carp_send_arp(sc); 1327 #ifdef INET6 1328 carp_send_na(sc); 1329 #endif /* INET6 */ 1330 CARP_DEBUG("%s: INIT -> MASTER (preempting)\n", 1331 SC2IFP(sc)->if_xname); 1332 carp_set_state(sc, MASTER); 1333 carp_setroute(sc, RTM_ADD); 1334 } else { 1335 CARP_DEBUG("%s: INIT -> BACKUP\n", SC2IFP(sc)->if_xname); 1336 carp_set_state(sc, BACKUP); 1337 carp_setroute(sc, RTM_DELETE); 1338 carp_setrun(sc, 0); 1339 } 1340 break; 1341 case BACKUP: 1342 callout_stop(&sc->sc_ad_tmo); 1343 tv.tv_sec = 3 * sc->sc_advbase; 1344 tv.tv_usec = sc->sc_advskew * 1000000 / 256; 1345 switch (af) { 1346 #ifdef INET 1347 case AF_INET: 1348 callout_reset(&sc->sc_md_tmo, tvtohz(&tv), 1349 carp_master_down, sc); 1350 break; 1351 #endif /* INET */ 1352 #ifdef INET6 1353 case AF_INET6: 1354 callout_reset(&sc->sc_md6_tmo, tvtohz(&tv), 1355 carp_master_down, sc); 1356 break; 1357 #endif /* INET6 */ 1358 default: 1359 if (sc->sc_naddrs) 1360 callout_reset(&sc->sc_md_tmo, tvtohz(&tv), 1361 carp_master_down, sc); 1362 if (sc->sc_naddrs6) 1363 callout_reset(&sc->sc_md6_tmo, tvtohz(&tv), 1364 carp_master_down, sc); 1365 break; 1366 } 1367 break; 1368 case MASTER: 1369 tv.tv_sec = sc->sc_advbase; 1370 tv.tv_usec = sc->sc_advskew * 1000000 / 256; 1371 callout_reset(&sc->sc_ad_tmo, tvtohz(&tv), 1372 carp_send_ad, sc); 1373 break; 1374 } 1375 } 1376 1377 static int 1378 carp_set_addr(struct carp_softc *sc, struct sockaddr_in *sin) 1379 { 1380 struct ifnet *ifp; 1381 struct carp_if *cif; 1382 struct in_ifaddr *ia, *ia_if; 1383 struct ip_moptions *imo = &sc->sc_imo; 1384 struct in_addr addr; 1385 u_long iaddr = htonl(sin->sin_addr.s_addr); 1386 int own, error; 1387 1388 if (sin->sin_addr.s_addr == 0) { 1389 if (!(SC2IFP(sc)->if_flags & IFF_UP)) 1390 carp_set_state(sc, INIT); 1391 if (sc->sc_naddrs) 1392 SC2IFP(sc)->if_flags |= IFF_UP; 1393 carp_setrun(sc, 0); 1394 return (0); 1395 } 1396 1397 /* we have to do it by hands to check we won't match on us */ 1398 ia_if = NULL; own = 0; 1399 TAILQ_FOREACH(ia, &in_ifaddrhead, ia_link) { 1400 /* and, yeah, we need a multicast-capable iface too */ 1401 if (ia->ia_ifp != SC2IFP(sc) && 1402 (ia->ia_ifp->if_flags & IFF_MULTICAST) && 1403 (iaddr & ia->ia_subnetmask) == ia->ia_subnet) { 1404 if (!ia_if) 1405 ia_if = ia; 1406 if (sin->sin_addr.s_addr == 1407 ia->ia_addr.sin_addr.s_addr) 1408 own++; 1409 } 1410 } 1411 1412 if (!ia_if) 1413 return (EADDRNOTAVAIL); 1414 1415 ia = ia_if; 1416 ifp = ia->ia_ifp; 1417 1418 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0 || 1419 (imo->imo_multicast_ifp && imo->imo_multicast_ifp != ifp)) 1420 return (EADDRNOTAVAIL); 1421 1422 if (imo->imo_num_memberships == 0) { 1423 addr.s_addr = htonl(INADDR_CARP_GROUP); 1424 if ((imo->imo_membership[0] = in_addmulti(&addr, ifp)) == NULL) 1425 return (ENOBUFS); 1426 imo->imo_num_memberships++; 1427 imo->imo_multicast_ifp = ifp; 1428 imo->imo_multicast_ttl = CARP_DFLTTL; 1429 imo->imo_multicast_loop = 0; 1430 } 1431 1432 if (!ifp->if_carp) { 1433 1434 MALLOC(cif, struct carp_if *, sizeof(*cif), M_CARP, 1435 M_WAITOK|M_ZERO); 1436 if (!cif) { 1437 error = ENOBUFS; 1438 goto cleanup; 1439 } 1440 if ((error = ifpromisc(ifp, 1))) { 1441 FREE(cif, M_CARP); 1442 goto cleanup; 1443 } 1444 1445 CARP_LOCK_INIT(cif); 1446 CARP_LOCK(cif); 1447 cif->vhif_ifp = ifp; 1448 TAILQ_INIT(&cif->vhif_vrs); 1449 ifp->if_carp = cif; 1450 1451 } else { 1452 struct carp_softc *vr; 1453 1454 cif = (struct carp_if *)ifp->if_carp; 1455 CARP_LOCK(cif); 1456 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) 1457 if (vr != sc && vr->sc_vhid == sc->sc_vhid) { 1458 CARP_UNLOCK(cif); 1459 error = EINVAL; 1460 goto cleanup; 1461 } 1462 } 1463 sc->sc_ia = ia; 1464 sc->sc_carpdev = ifp; 1465 1466 { /* XXX prevent endless loop if already in queue */ 1467 struct carp_softc *vr, *after = NULL; 1468 int myself = 0; 1469 cif = (struct carp_if *)ifp->if_carp; 1470 1471 /* XXX: cif should not change, right? So we still hold the lock */ 1472 CARP_LOCK_ASSERT(cif); 1473 1474 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) { 1475 if (vr == sc) 1476 myself = 1; 1477 if (vr->sc_vhid < sc->sc_vhid) 1478 after = vr; 1479 } 1480 1481 if (!myself) { 1482 /* We're trying to keep things in order */ 1483 if (after == NULL) { 1484 TAILQ_INSERT_TAIL(&cif->vhif_vrs, sc, sc_list); 1485 } else { 1486 TAILQ_INSERT_AFTER(&cif->vhif_vrs, after, sc, sc_list); 1487 } 1488 cif->vhif_nvrs++; 1489 } 1490 } 1491 1492 sc->sc_naddrs++; 1493 SC2IFP(sc)->if_flags |= IFF_UP; 1494 if (own) 1495 sc->sc_advskew = 0; 1496 carp_sc_state_locked(sc); 1497 carp_setrun(sc, 0); 1498 1499 CARP_UNLOCK(cif); 1500 1501 return (0); 1502 1503 cleanup: 1504 in_delmulti(imo->imo_membership[--imo->imo_num_memberships]); 1505 return (error); 1506 } 1507 1508 static int 1509 carp_del_addr(struct carp_softc *sc, struct sockaddr_in *sin) 1510 { 1511 int error = 0; 1512 1513 if (!--sc->sc_naddrs) { 1514 struct carp_if *cif = (struct carp_if *)sc->sc_carpdev->if_carp; 1515 struct ip_moptions *imo = &sc->sc_imo; 1516 1517 CARP_LOCK(cif); 1518 callout_stop(&sc->sc_ad_tmo); 1519 SC2IFP(sc)->if_flags &= ~IFF_UP; 1520 SC2IFP(sc)->if_drv_flags &= ~IFF_DRV_RUNNING; 1521 sc->sc_vhid = -1; 1522 in_delmulti(imo->imo_membership[--imo->imo_num_memberships]); 1523 imo->imo_multicast_ifp = NULL; 1524 TAILQ_REMOVE(&cif->vhif_vrs, sc, sc_list); 1525 if (!--cif->vhif_nvrs) { 1526 sc->sc_carpdev->if_carp = NULL; 1527 CARP_LOCK_DESTROY(cif); 1528 FREE(cif, M_IFADDR); 1529 } else { 1530 CARP_UNLOCK(cif); 1531 } 1532 } 1533 1534 return (error); 1535 } 1536 1537 #ifdef INET6 1538 static int 1539 carp_set_addr6(struct carp_softc *sc, struct sockaddr_in6 *sin6) 1540 { 1541 struct ifnet *ifp; 1542 struct carp_if *cif; 1543 struct in6_ifaddr *ia, *ia_if; 1544 struct ip6_moptions *im6o = &sc->sc_im6o; 1545 struct in6_multi_mship *imm; 1546 struct in6_addr in6; 1547 int own, error; 1548 1549 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) { 1550 if (!(SC2IFP(sc)->if_flags & IFF_UP)) 1551 carp_set_state(sc, INIT); 1552 if (sc->sc_naddrs6) 1553 SC2IFP(sc)->if_flags |= IFF_UP; 1554 carp_setrun(sc, 0); 1555 return (0); 1556 } 1557 1558 /* we have to do it by hands to check we won't match on us */ 1559 ia_if = NULL; own = 0; 1560 for (ia = in6_ifaddr; ia; ia = ia->ia_next) { 1561 int i; 1562 1563 for (i = 0; i < 4; i++) { 1564 if ((sin6->sin6_addr.s6_addr32[i] & 1565 ia->ia_prefixmask.sin6_addr.s6_addr32[i]) != 1566 (ia->ia_addr.sin6_addr.s6_addr32[i] & 1567 ia->ia_prefixmask.sin6_addr.s6_addr32[i])) 1568 break; 1569 } 1570 /* and, yeah, we need a multicast-capable iface too */ 1571 if (ia->ia_ifp != SC2IFP(sc) && 1572 (ia->ia_ifp->if_flags & IFF_MULTICAST) && 1573 (i == 4)) { 1574 if (!ia_if) 1575 ia_if = ia; 1576 if (IN6_ARE_ADDR_EQUAL(&sin6->sin6_addr, 1577 &ia->ia_addr.sin6_addr)) 1578 own++; 1579 } 1580 } 1581 1582 if (!ia_if) 1583 return (EADDRNOTAVAIL); 1584 ia = ia_if; 1585 ifp = ia->ia_ifp; 1586 1587 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0 || 1588 (im6o->im6o_multicast_ifp && im6o->im6o_multicast_ifp != ifp)) 1589 return (EADDRNOTAVAIL); 1590 1591 if (!sc->sc_naddrs6) { 1592 im6o->im6o_multicast_ifp = ifp; 1593 1594 /* join CARP multicast address */ 1595 bzero(&in6, sizeof(in6)); 1596 in6.s6_addr16[0] = htons(0xff02); 1597 in6.s6_addr8[15] = 0x12; 1598 if (in6_setscope(&in6, ifp, NULL) != 0) 1599 goto cleanup; 1600 if ((imm = in6_joingroup(ifp, &in6, &error, 0)) == NULL) 1601 goto cleanup; 1602 LIST_INSERT_HEAD(&im6o->im6o_memberships, imm, i6mm_chain); 1603 1604 /* join solicited multicast address */ 1605 bzero(&in6, sizeof(in6)); 1606 in6.s6_addr16[0] = htons(0xff02); 1607 in6.s6_addr32[1] = 0; 1608 in6.s6_addr32[2] = htonl(1); 1609 in6.s6_addr32[3] = sin6->sin6_addr.s6_addr32[3]; 1610 in6.s6_addr8[12] = 0xff; 1611 if (in6_setscope(&in6, ifp, NULL) != 0) 1612 goto cleanup; 1613 if ((imm = in6_joingroup(ifp, &in6, &error, 0)) == NULL) 1614 goto cleanup; 1615 LIST_INSERT_HEAD(&im6o->im6o_memberships, imm, i6mm_chain); 1616 } 1617 1618 if (!ifp->if_carp) { 1619 MALLOC(cif, struct carp_if *, sizeof(*cif), M_CARP, 1620 M_WAITOK|M_ZERO); 1621 if (!cif) { 1622 error = ENOBUFS; 1623 goto cleanup; 1624 } 1625 if ((error = ifpromisc(ifp, 1))) { 1626 FREE(cif, M_CARP); 1627 goto cleanup; 1628 } 1629 1630 CARP_LOCK_INIT(cif); 1631 CARP_LOCK(cif); 1632 cif->vhif_ifp = ifp; 1633 TAILQ_INIT(&cif->vhif_vrs); 1634 ifp->if_carp = cif; 1635 1636 } else { 1637 struct carp_softc *vr; 1638 1639 cif = (struct carp_if *)ifp->if_carp; 1640 CARP_LOCK(cif); 1641 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) 1642 if (vr != sc && vr->sc_vhid == sc->sc_vhid) { 1643 CARP_UNLOCK(cif); 1644 error = EINVAL; 1645 goto cleanup; 1646 } 1647 } 1648 sc->sc_ia6 = ia; 1649 sc->sc_carpdev = ifp; 1650 1651 { /* XXX prevent endless loop if already in queue */ 1652 struct carp_softc *vr, *after = NULL; 1653 int myself = 0; 1654 cif = (struct carp_if *)ifp->if_carp; 1655 CARP_LOCK_ASSERT(cif); 1656 1657 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) { 1658 if (vr == sc) 1659 myself = 1; 1660 if (vr->sc_vhid < sc->sc_vhid) 1661 after = vr; 1662 } 1663 1664 if (!myself) { 1665 /* We're trying to keep things in order */ 1666 if (after == NULL) { 1667 TAILQ_INSERT_TAIL(&cif->vhif_vrs, sc, sc_list); 1668 } else { 1669 TAILQ_INSERT_AFTER(&cif->vhif_vrs, after, sc, sc_list); 1670 } 1671 cif->vhif_nvrs++; 1672 } 1673 } 1674 1675 sc->sc_naddrs6++; 1676 SC2IFP(sc)->if_flags |= IFF_UP; 1677 if (own) 1678 sc->sc_advskew = 0; 1679 carp_sc_state_locked(sc); 1680 carp_setrun(sc, 0); 1681 1682 CARP_UNLOCK(cif); 1683 1684 return (0); 1685 1686 cleanup: 1687 /* clean up multicast memberships */ 1688 if (!sc->sc_naddrs6) { 1689 while (!LIST_EMPTY(&im6o->im6o_memberships)) { 1690 imm = LIST_FIRST(&im6o->im6o_memberships); 1691 LIST_REMOVE(imm, i6mm_chain); 1692 in6_leavegroup(imm); 1693 } 1694 } 1695 return (error); 1696 } 1697 1698 static int 1699 carp_del_addr6(struct carp_softc *sc, struct sockaddr_in6 *sin6) 1700 { 1701 int error = 0; 1702 1703 if (!--sc->sc_naddrs6) { 1704 struct carp_if *cif = (struct carp_if *)sc->sc_carpdev->if_carp; 1705 struct ip6_moptions *im6o = &sc->sc_im6o; 1706 1707 CARP_LOCK(cif); 1708 callout_stop(&sc->sc_ad_tmo); 1709 SC2IFP(sc)->if_flags &= ~IFF_UP; 1710 SC2IFP(sc)->if_drv_flags &= ~IFF_DRV_RUNNING; 1711 sc->sc_vhid = -1; 1712 while (!LIST_EMPTY(&im6o->im6o_memberships)) { 1713 struct in6_multi_mship *imm = 1714 LIST_FIRST(&im6o->im6o_memberships); 1715 1716 LIST_REMOVE(imm, i6mm_chain); 1717 in6_leavegroup(imm); 1718 } 1719 im6o->im6o_multicast_ifp = NULL; 1720 TAILQ_REMOVE(&cif->vhif_vrs, sc, sc_list); 1721 if (!--cif->vhif_nvrs) { 1722 CARP_LOCK_DESTROY(cif); 1723 sc->sc_carpdev->if_carp = NULL; 1724 FREE(cif, M_IFADDR); 1725 } else 1726 CARP_UNLOCK(cif); 1727 } 1728 1729 return (error); 1730 } 1731 #endif /* INET6 */ 1732 1733 static int 1734 carp_ioctl(struct ifnet *ifp, u_long cmd, caddr_t addr) 1735 { 1736 struct carp_softc *sc = ifp->if_softc, *vr; 1737 struct carpreq carpr; 1738 struct ifaddr *ifa; 1739 struct ifreq *ifr; 1740 struct ifaliasreq *ifra; 1741 int locked = 0, error = 0; 1742 1743 ifa = (struct ifaddr *)addr; 1744 ifra = (struct ifaliasreq *)addr; 1745 ifr = (struct ifreq *)addr; 1746 1747 switch (cmd) { 1748 case SIOCSIFADDR: 1749 switch (ifa->ifa_addr->sa_family) { 1750 #ifdef INET 1751 case AF_INET: 1752 SC2IFP(sc)->if_flags |= IFF_UP; 1753 bcopy(ifa->ifa_addr, ifa->ifa_dstaddr, 1754 sizeof(struct sockaddr)); 1755 error = carp_set_addr(sc, satosin(ifa->ifa_addr)); 1756 break; 1757 #endif /* INET */ 1758 #ifdef INET6 1759 case AF_INET6: 1760 SC2IFP(sc)->if_flags |= IFF_UP; 1761 error = carp_set_addr6(sc, satosin6(ifa->ifa_addr)); 1762 break; 1763 #endif /* INET6 */ 1764 default: 1765 error = EAFNOSUPPORT; 1766 break; 1767 } 1768 break; 1769 1770 case SIOCAIFADDR: 1771 switch (ifa->ifa_addr->sa_family) { 1772 #ifdef INET 1773 case AF_INET: 1774 SC2IFP(sc)->if_flags |= IFF_UP; 1775 bcopy(ifa->ifa_addr, ifa->ifa_dstaddr, 1776 sizeof(struct sockaddr)); 1777 error = carp_set_addr(sc, satosin(&ifra->ifra_addr)); 1778 break; 1779 #endif /* INET */ 1780 #ifdef INET6 1781 case AF_INET6: 1782 SC2IFP(sc)->if_flags |= IFF_UP; 1783 error = carp_set_addr6(sc, satosin6(&ifra->ifra_addr)); 1784 break; 1785 #endif /* INET6 */ 1786 default: 1787 error = EAFNOSUPPORT; 1788 break; 1789 } 1790 break; 1791 1792 case SIOCDIFADDR: 1793 switch (ifa->ifa_addr->sa_family) { 1794 #ifdef INET 1795 case AF_INET: 1796 error = carp_del_addr(sc, satosin(&ifra->ifra_addr)); 1797 break; 1798 #endif /* INET */ 1799 #ifdef INET6 1800 case AF_INET6: 1801 error = carp_del_addr6(sc, satosin6(&ifra->ifra_addr)); 1802 break; 1803 #endif /* INET6 */ 1804 default: 1805 error = EAFNOSUPPORT; 1806 break; 1807 } 1808 break; 1809 1810 case SIOCSIFFLAGS: 1811 if (sc->sc_carpdev) { 1812 locked = 1; 1813 CARP_SCLOCK(sc); 1814 } 1815 if (sc->sc_state != INIT && !(ifr->ifr_flags & IFF_UP)) { 1816 callout_stop(&sc->sc_ad_tmo); 1817 callout_stop(&sc->sc_md_tmo); 1818 callout_stop(&sc->sc_md6_tmo); 1819 if (sc->sc_state == MASTER) 1820 carp_send_ad_locked(sc); 1821 carp_set_state(sc, INIT); 1822 carp_setrun(sc, 0); 1823 } else if (sc->sc_state == INIT && (ifr->ifr_flags & IFF_UP)) { 1824 SC2IFP(sc)->if_flags |= IFF_UP; 1825 carp_setrun(sc, 0); 1826 } 1827 break; 1828 1829 case SIOCSVH: 1830 if ((error = suser(curthread)) != 0) 1831 break; 1832 if ((error = copyin(ifr->ifr_data, &carpr, sizeof carpr))) 1833 break; 1834 error = 1; 1835 if (sc->sc_carpdev) { 1836 locked = 1; 1837 CARP_SCLOCK(sc); 1838 } 1839 if (sc->sc_state != INIT && carpr.carpr_state != sc->sc_state) { 1840 switch (carpr.carpr_state) { 1841 case BACKUP: 1842 callout_stop(&sc->sc_ad_tmo); 1843 carp_set_state(sc, BACKUP); 1844 carp_setrun(sc, 0); 1845 carp_setroute(sc, RTM_DELETE); 1846 break; 1847 case MASTER: 1848 carp_master_down_locked(sc); 1849 break; 1850 default: 1851 break; 1852 } 1853 } 1854 if (carpr.carpr_vhid > 0) { 1855 if (carpr.carpr_vhid > 255) { 1856 error = EINVAL; 1857 break; 1858 } 1859 if (sc->sc_carpdev) { 1860 struct carp_if *cif; 1861 cif = (struct carp_if *)sc->sc_carpdev->if_carp; 1862 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) 1863 if (vr != sc && 1864 vr->sc_vhid == carpr.carpr_vhid) 1865 return EEXIST; 1866 } 1867 sc->sc_vhid = carpr.carpr_vhid; 1868 IFP2ENADDR(sc->sc_ifp)[0] = 0; 1869 IFP2ENADDR(sc->sc_ifp)[1] = 0; 1870 IFP2ENADDR(sc->sc_ifp)[2] = 0x5e; 1871 IFP2ENADDR(sc->sc_ifp)[3] = 0; 1872 IFP2ENADDR(sc->sc_ifp)[4] = 1; 1873 IFP2ENADDR(sc->sc_ifp)[5] = sc->sc_vhid; 1874 error--; 1875 } 1876 if (carpr.carpr_advbase > 0 || carpr.carpr_advskew > 0) { 1877 if (carpr.carpr_advskew >= 255) { 1878 error = EINVAL; 1879 break; 1880 } 1881 if (carpr.carpr_advbase > 255) { 1882 error = EINVAL; 1883 break; 1884 } 1885 sc->sc_advbase = carpr.carpr_advbase; 1886 sc->sc_advskew = carpr.carpr_advskew; 1887 error--; 1888 } 1889 bcopy(carpr.carpr_key, sc->sc_key, sizeof(sc->sc_key)); 1890 if (error > 0) 1891 error = EINVAL; 1892 else { 1893 error = 0; 1894 carp_setrun(sc, 0); 1895 } 1896 break; 1897 1898 case SIOCGVH: 1899 /* XXX: lockless read */ 1900 bzero(&carpr, sizeof(carpr)); 1901 carpr.carpr_state = sc->sc_state; 1902 carpr.carpr_vhid = sc->sc_vhid; 1903 carpr.carpr_advbase = sc->sc_advbase; 1904 carpr.carpr_advskew = sc->sc_advskew; 1905 if (suser(curthread) == 0) 1906 bcopy(sc->sc_key, carpr.carpr_key, 1907 sizeof(carpr.carpr_key)); 1908 error = copyout(&carpr, ifr->ifr_data, sizeof(carpr)); 1909 break; 1910 1911 default: 1912 error = EINVAL; 1913 } 1914 1915 if (locked) 1916 CARP_SCUNLOCK(sc); 1917 1918 carp_hmac_prepare(sc); 1919 1920 return (error); 1921 } 1922 1923 /* 1924 * XXX: this is looutput. We should eventually use it from there. 1925 */ 1926 static int 1927 carp_looutput(struct ifnet *ifp, struct mbuf *m, struct sockaddr *dst, 1928 struct rtentry *rt) 1929 { 1930 u_int32_t af; 1931 1932 M_ASSERTPKTHDR(m); /* check if we have the packet header */ 1933 1934 if (rt && rt->rt_flags & (RTF_REJECT|RTF_BLACKHOLE)) { 1935 m_freem(m); 1936 return (rt->rt_flags & RTF_BLACKHOLE ? 0 : 1937 rt->rt_flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH); 1938 } 1939 1940 ifp->if_opackets++; 1941 ifp->if_obytes += m->m_pkthdr.len; 1942 1943 /* BPF writes need to be handled specially. */ 1944 if (dst->sa_family == AF_UNSPEC) { 1945 bcopy(dst->sa_data, &af, sizeof(af)); 1946 dst->sa_family = af; 1947 } 1948 1949 #if 1 /* XXX */ 1950 switch (dst->sa_family) { 1951 case AF_INET: 1952 case AF_INET6: 1953 case AF_IPX: 1954 case AF_APPLETALK: 1955 break; 1956 default: 1957 printf("carp_looutput: af=%d unexpected\n", dst->sa_family); 1958 m_freem(m); 1959 return (EAFNOSUPPORT); 1960 } 1961 #endif 1962 return(if_simloop(ifp, m, dst->sa_family, 0)); 1963 } 1964 1965 /* 1966 * Start output on carp interface. This function should never be called. 1967 */ 1968 static void 1969 carp_start(struct ifnet *ifp) 1970 { 1971 #ifdef DEBUG 1972 printf("%s: start called\n", ifp->if_xname); 1973 #endif 1974 } 1975 1976 int 1977 carp_output(struct ifnet *ifp, struct mbuf *m, struct sockaddr *sa, 1978 struct rtentry *rt) 1979 { 1980 struct m_tag *mtag; 1981 struct carp_softc *sc; 1982 struct ifnet *carp_ifp; 1983 1984 if (!sa) 1985 return (0); 1986 1987 switch (sa->sa_family) { 1988 #ifdef INET 1989 case AF_INET: 1990 break; 1991 #endif /* INET */ 1992 #ifdef INET6 1993 case AF_INET6: 1994 break; 1995 #endif /* INET6 */ 1996 default: 1997 return (0); 1998 } 1999 2000 mtag = m_tag_find(m, PACKET_TAG_CARP, NULL); 2001 if (mtag == NULL) 2002 return (0); 2003 2004 bcopy(mtag + 1, &carp_ifp, sizeof(struct ifnet *)); 2005 sc = carp_ifp->if_softc; 2006 2007 /* Set the source MAC address to Virtual Router MAC Address */ 2008 switch (ifp->if_type) { 2009 case IFT_ETHER: 2010 case IFT_L2VLAN: { 2011 struct ether_header *eh; 2012 2013 eh = mtod(m, struct ether_header *); 2014 eh->ether_shost[0] = 0; 2015 eh->ether_shost[1] = 0; 2016 eh->ether_shost[2] = 0x5e; 2017 eh->ether_shost[3] = 0; 2018 eh->ether_shost[4] = 1; 2019 eh->ether_shost[5] = sc->sc_vhid; 2020 } 2021 break; 2022 case IFT_FDDI: { 2023 struct fddi_header *fh; 2024 2025 fh = mtod(m, struct fddi_header *); 2026 fh->fddi_shost[0] = 0; 2027 fh->fddi_shost[1] = 0; 2028 fh->fddi_shost[2] = 0x5e; 2029 fh->fddi_shost[3] = 0; 2030 fh->fddi_shost[4] = 1; 2031 fh->fddi_shost[5] = sc->sc_vhid; 2032 } 2033 break; 2034 case IFT_ISO88025: { 2035 struct iso88025_header *th; 2036 th = mtod(m, struct iso88025_header *); 2037 th->iso88025_shost[0] = 3; 2038 th->iso88025_shost[1] = 0; 2039 th->iso88025_shost[2] = 0x40 >> (sc->sc_vhid - 1); 2040 th->iso88025_shost[3] = 0x40000 >> (sc->sc_vhid - 1); 2041 th->iso88025_shost[4] = 0; 2042 th->iso88025_shost[5] = 0; 2043 } 2044 break; 2045 default: 2046 printf("%s: carp is not supported for this interface type\n", 2047 ifp->if_xname); 2048 return (EOPNOTSUPP); 2049 } 2050 2051 return (0); 2052 } 2053 2054 static void 2055 carp_set_state(struct carp_softc *sc, int state) 2056 { 2057 2058 if (sc->sc_carpdev) 2059 CARP_SCLOCK_ASSERT(sc); 2060 2061 if (sc->sc_state == state) 2062 return; 2063 2064 sc->sc_state = state; 2065 switch (state) { 2066 case BACKUP: 2067 SC2IFP(sc)->if_link_state = LINK_STATE_DOWN; 2068 break; 2069 case MASTER: 2070 SC2IFP(sc)->if_link_state = LINK_STATE_UP; 2071 break; 2072 default: 2073 SC2IFP(sc)->if_link_state = LINK_STATE_UNKNOWN; 2074 break; 2075 } 2076 rt_ifmsg(SC2IFP(sc)); 2077 } 2078 2079 void 2080 carp_carpdev_state(void *v) 2081 { 2082 struct carp_if *cif = v; 2083 2084 CARP_LOCK(cif); 2085 carp_carpdev_state_locked(cif); 2086 CARP_UNLOCK(cif); 2087 } 2088 2089 static void 2090 carp_carpdev_state_locked(struct carp_if *cif) 2091 { 2092 struct carp_softc *sc; 2093 2094 TAILQ_FOREACH(sc, &cif->vhif_vrs, sc_list) 2095 carp_sc_state_locked(sc); 2096 } 2097 2098 static void 2099 carp_sc_state_locked(struct carp_softc *sc) 2100 { 2101 CARP_SCLOCK_ASSERT(sc); 2102 2103 if (sc->sc_carpdev->if_link_state != LINK_STATE_UP || 2104 !(sc->sc_carpdev->if_flags & IFF_UP)) { 2105 sc->sc_flags_backup = SC2IFP(sc)->if_flags; 2106 SC2IFP(sc)->if_flags &= ~IFF_UP; 2107 SC2IFP(sc)->if_drv_flags &= ~IFF_DRV_RUNNING; 2108 callout_stop(&sc->sc_ad_tmo); 2109 callout_stop(&sc->sc_md_tmo); 2110 callout_stop(&sc->sc_md6_tmo); 2111 carp_set_state(sc, INIT); 2112 carp_setrun(sc, 0); 2113 if (!sc->sc_suppress) { 2114 carp_suppress_preempt++; 2115 if (carp_suppress_preempt == 1) { 2116 CARP_SCUNLOCK(sc); 2117 carp_send_ad_all(); 2118 CARP_SCLOCK(sc); 2119 } 2120 } 2121 sc->sc_suppress = 1; 2122 } else { 2123 SC2IFP(sc)->if_flags |= sc->sc_flags_backup; 2124 carp_set_state(sc, INIT); 2125 carp_setrun(sc, 0); 2126 if (sc->sc_suppress) 2127 carp_suppress_preempt--; 2128 sc->sc_suppress = 0; 2129 } 2130 2131 return; 2132 } 2133 2134 static int 2135 carp_modevent(module_t mod, int type, void *data) 2136 { 2137 int error = 0; 2138 2139 switch (type) { 2140 case MOD_LOAD: 2141 mtx_init(&carp_mtx, "carp_mtx", NULL, MTX_DEF); 2142 LIST_INIT(&carpif_list); 2143 if_clone_attach(&carp_cloner); 2144 break; 2145 2146 case MOD_UNLOAD: 2147 if_clone_detach(&carp_cloner); 2148 while (!LIST_EMPTY(&carpif_list)) 2149 carp_clone_destroy(SC2IFP(LIST_FIRST(&carpif_list))); 2150 mtx_destroy(&carp_mtx); 2151 break; 2152 2153 default: 2154 error = EINVAL; 2155 break; 2156 } 2157 2158 return error; 2159 } 2160 2161 static moduledata_t carp_mod = { 2162 "carp", 2163 carp_modevent, 2164 0 2165 }; 2166 2167 DECLARE_MODULE(carp, carp_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
Cache object: e7fb4965aea4e849e860005badf27fef
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]