1 /* 2 * Copyright (c) 2002 Michael Shalayeff. All rights reserved. 3 * Copyright (c) 2003 Ryan McBride. All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 15 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 16 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 17 * IN NO EVENT SHALL THE AUTHOR OR HIS RELATIVES BE LIABLE FOR ANY DIRECT, 18 * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES 19 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR 20 * SERVICES; LOSS OF MIND, USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, 22 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING 23 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF 24 * THE POSSIBILITY OF SUCH DAMAGE. 25 */ 26 27 #include <sys/cdefs.h> 28 __FBSDID("$FreeBSD$"); 29 30 #include "opt_carp.h" 31 #include "opt_bpf.h" 32 #include "opt_inet.h" 33 #include "opt_inet6.h" 34 35 #include <sys/types.h> 36 #include <sys/param.h> 37 #include <sys/systm.h> 38 #include <sys/conf.h> 39 #include <sys/kernel.h> 40 #include <sys/limits.h> 41 #include <sys/malloc.h> 42 #include <sys/mbuf.h> 43 #include <sys/module.h> 44 #include <sys/time.h> 45 #include <sys/priv.h> 46 #include <sys/proc.h> 47 #include <sys/sysctl.h> 48 #include <sys/syslog.h> 49 #include <sys/signalvar.h> 50 #include <sys/filio.h> 51 #include <sys/sockio.h> 52 53 #include <sys/socket.h> 54 #include <sys/vnode.h> 55 #include <sys/vimage.h> 56 57 #include <machine/stdarg.h> 58 59 #include <net/bpf.h> 60 #include <net/ethernet.h> 61 #include <net/fddi.h> 62 #include <net/iso88025.h> 63 #include <net/if.h> 64 #include <net/if_clone.h> 65 #include <net/if_dl.h> 66 #include <net/if_types.h> 67 #include <net/route.h> 68 69 #ifdef INET 70 #include <netinet/in.h> 71 #include <netinet/in_var.h> 72 #include <netinet/in_systm.h> 73 #include <netinet/ip.h> 74 #include <netinet/ip_var.h> 75 #include <netinet/if_ether.h> 76 #include <machine/in_cksum.h> 77 #include <netinet/vinet.h> 78 #endif 79 80 #ifdef INET6 81 #include <netinet/icmp6.h> 82 #include <netinet/ip6.h> 83 #include <netinet6/ip6_var.h> 84 #include <netinet6/scope6_var.h> 85 #include <netinet6/nd6.h> 86 #include <netinet6/vinet6.h> 87 #endif 88 89 #include <crypto/sha1.h> 90 #include <netinet/ip_carp.h> 91 92 #define CARP_IFNAME "carp" 93 static MALLOC_DEFINE(M_CARP, "CARP", "CARP interfaces"); 94 SYSCTL_DECL(_net_inet_carp); 95 96 struct carp_softc { 97 struct ifnet *sc_ifp; /* Interface clue */ 98 struct ifnet *sc_carpdev; /* Pointer to parent interface */ 99 struct in_ifaddr *sc_ia; /* primary iface address */ 100 struct ip_moptions sc_imo; 101 #ifdef INET6 102 struct in6_ifaddr *sc_ia6; /* primary iface address v6 */ 103 struct ip6_moptions sc_im6o; 104 #endif /* INET6 */ 105 TAILQ_ENTRY(carp_softc) sc_list; 106 107 enum { INIT = 0, BACKUP, MASTER } sc_state; 108 109 int sc_flags_backup; 110 int sc_suppress; 111 112 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_vhid; 118 int sc_advskew; 119 int sc_naddrs; 120 int sc_naddrs6; 121 int sc_advbase; /* seconds */ 122 int sc_init_counter; 123 u_int64_t sc_counter; 124 125 /* authentication */ 126 #define CARP_HMAC_PAD 64 127 unsigned char sc_key[CARP_KEY_LEN]; 128 unsigned char sc_pad[CARP_HMAC_PAD]; 129 SHA1_CTX sc_sha1; 130 131 struct callout sc_ad_tmo; /* advertisement timeout */ 132 struct callout sc_md_tmo; /* master down timeout */ 133 struct callout sc_md6_tmo; /* master down timeout */ 134 135 LIST_ENTRY(carp_softc) sc_next; /* Interface clue */ 136 }; 137 #define SC2IFP(sc) ((sc)->sc_ifp) 138 139 int carp_suppress_preempt = 0; 140 int carp_opts[CARPCTL_MAXID] = { 0, 1, 0, 1, 0, 0 }; /* XXX for now */ 141 SYSCTL_INT(_net_inet_carp, CARPCTL_ALLOW, allow, CTLFLAG_RW, 142 &carp_opts[CARPCTL_ALLOW], 0, "Accept incoming CARP packets"); 143 SYSCTL_INT(_net_inet_carp, CARPCTL_PREEMPT, preempt, CTLFLAG_RW, 144 &carp_opts[CARPCTL_PREEMPT], 0, "high-priority backup preemption mode"); 145 SYSCTL_INT(_net_inet_carp, CARPCTL_LOG, log, CTLFLAG_RW, 146 &carp_opts[CARPCTL_LOG], 0, "log bad carp packets"); 147 SYSCTL_INT(_net_inet_carp, CARPCTL_ARPBALANCE, arpbalance, CTLFLAG_RW, 148 &carp_opts[CARPCTL_ARPBALANCE], 0, "balance arp responses"); 149 SYSCTL_INT(_net_inet_carp, OID_AUTO, suppress_preempt, CTLFLAG_RD, 150 &carp_suppress_preempt, 0, "Preemption is suppressed"); 151 152 struct carpstats carpstats; 153 SYSCTL_STRUCT(_net_inet_carp, CARPCTL_STATS, stats, CTLFLAG_RW, 154 &carpstats, carpstats, 155 "CARP statistics (struct carpstats, netinet/ip_carp.h)"); 156 157 struct carp_if { 158 TAILQ_HEAD(, carp_softc) vhif_vrs; 159 int vhif_nvrs; 160 161 struct ifnet *vhif_ifp; 162 struct mtx vhif_mtx; 163 }; 164 165 /* Get carp_if from softc. Valid after carp_set_addr{,6}. */ 166 #define SC2CIF(sc) ((struct carp_if *)(sc)->sc_carpdev->if_carp) 167 168 /* lock per carp_if queue */ 169 #define CARP_LOCK_INIT(cif) mtx_init(&(cif)->vhif_mtx, "carp_if", \ 170 NULL, MTX_DEF) 171 #define CARP_LOCK_DESTROY(cif) mtx_destroy(&(cif)->vhif_mtx) 172 #define CARP_LOCK_ASSERT(cif) mtx_assert(&(cif)->vhif_mtx, MA_OWNED) 173 #define CARP_LOCK(cif) mtx_lock(&(cif)->vhif_mtx) 174 #define CARP_UNLOCK(cif) mtx_unlock(&(cif)->vhif_mtx) 175 176 #define CARP_SCLOCK(sc) mtx_lock(&SC2CIF(sc)->vhif_mtx) 177 #define CARP_SCUNLOCK(sc) mtx_unlock(&SC2CIF(sc)->vhif_mtx) 178 #define CARP_SCLOCK_ASSERT(sc) mtx_assert(&SC2CIF(sc)->vhif_mtx, MA_OWNED) 179 180 #define CARP_LOG(...) do { \ 181 if (carp_opts[CARPCTL_LOG] > 0) \ 182 log(LOG_INFO, __VA_ARGS__); \ 183 } while (0) 184 185 #define CARP_DEBUG(...) do { \ 186 if (carp_opts[CARPCTL_LOG] > 1) \ 187 log(LOG_DEBUG, __VA_ARGS__); \ 188 } while (0) 189 190 static void carp_hmac_prepare(struct carp_softc *); 191 static void carp_hmac_generate(struct carp_softc *, u_int32_t *, 192 unsigned char *); 193 static int carp_hmac_verify(struct carp_softc *, u_int32_t *, 194 unsigned char *); 195 static void carp_setroute(struct carp_softc *, int); 196 static void carp_input_c(struct mbuf *, struct carp_header *, sa_family_t); 197 static int carp_clone_create(struct if_clone *, int, caddr_t); 198 static void carp_clone_destroy(struct ifnet *); 199 static void carpdetach(struct carp_softc *, int); 200 static int carp_prepare_ad(struct mbuf *, struct carp_softc *, 201 struct carp_header *); 202 static void carp_send_ad_all(void); 203 static void carp_send_ad(void *); 204 static void carp_send_ad_locked(struct carp_softc *); 205 static void carp_send_arp(struct carp_softc *); 206 static void carp_master_down(void *); 207 static void carp_master_down_locked(struct carp_softc *); 208 static int carp_ioctl(struct ifnet *, u_long, caddr_t); 209 static int carp_looutput(struct ifnet *, struct mbuf *, struct sockaddr *, 210 struct rtentry *); 211 static void carp_start(struct ifnet *); 212 static void carp_setrun(struct carp_softc *, sa_family_t); 213 static void carp_set_state(struct carp_softc *, int); 214 static int carp_addrcount(struct carp_if *, struct in_ifaddr *, int); 215 enum { CARP_COUNT_MASTER, CARP_COUNT_RUNNING }; 216 217 static void carp_multicast_cleanup(struct carp_softc *); 218 static int carp_set_addr(struct carp_softc *, struct sockaddr_in *); 219 static int carp_del_addr(struct carp_softc *, struct sockaddr_in *); 220 static void carp_carpdev_state_locked(struct carp_if *); 221 static void carp_sc_state_locked(struct carp_softc *); 222 #ifdef INET6 223 static void carp_send_na(struct carp_softc *); 224 static int carp_set_addr6(struct carp_softc *, struct sockaddr_in6 *); 225 static int carp_del_addr6(struct carp_softc *, struct sockaddr_in6 *); 226 static void carp_multicast6_cleanup(struct carp_softc *); 227 #endif 228 229 static LIST_HEAD(, carp_softc) carpif_list; 230 static struct mtx carp_mtx; 231 IFC_SIMPLE_DECLARE(carp, 0); 232 233 static eventhandler_tag if_detach_event_tag; 234 235 static __inline u_int16_t 236 carp_cksum(struct mbuf *m, int len) 237 { 238 return (in_cksum(m, len)); 239 } 240 241 static void 242 carp_hmac_prepare(struct carp_softc *sc) 243 { 244 u_int8_t version = CARP_VERSION, type = CARP_ADVERTISEMENT; 245 u_int8_t vhid = sc->sc_vhid & 0xff; 246 struct ifaddr *ifa; 247 int i, found; 248 #ifdef INET 249 struct in_addr last, cur, in; 250 #endif 251 #ifdef INET6 252 struct in6_addr last6, cur6, in6; 253 #endif 254 255 if (sc->sc_carpdev) 256 CARP_SCLOCK(sc); 257 258 /* XXX: possible race here */ 259 260 /* compute ipad from key */ 261 bzero(sc->sc_pad, sizeof(sc->sc_pad)); 262 bcopy(sc->sc_key, sc->sc_pad, sizeof(sc->sc_key)); 263 for (i = 0; i < sizeof(sc->sc_pad); i++) 264 sc->sc_pad[i] ^= 0x36; 265 266 /* precompute first part of inner hash */ 267 SHA1Init(&sc->sc_sha1); 268 SHA1Update(&sc->sc_sha1, sc->sc_pad, sizeof(sc->sc_pad)); 269 SHA1Update(&sc->sc_sha1, (void *)&version, sizeof(version)); 270 SHA1Update(&sc->sc_sha1, (void *)&type, sizeof(type)); 271 SHA1Update(&sc->sc_sha1, (void *)&vhid, sizeof(vhid)); 272 #ifdef INET 273 cur.s_addr = 0; 274 do { 275 found = 0; 276 last = cur; 277 cur.s_addr = 0xffffffff; 278 TAILQ_FOREACH(ifa, &SC2IFP(sc)->if_addrlist, ifa_list) { 279 in.s_addr = ifatoia(ifa)->ia_addr.sin_addr.s_addr; 280 if (ifa->ifa_addr->sa_family == AF_INET && 281 ntohl(in.s_addr) > ntohl(last.s_addr) && 282 ntohl(in.s_addr) < ntohl(cur.s_addr)) { 283 cur.s_addr = in.s_addr; 284 found++; 285 } 286 } 287 if (found) 288 SHA1Update(&sc->sc_sha1, (void *)&cur, sizeof(cur)); 289 } while (found); 290 #endif /* INET */ 291 #ifdef INET6 292 memset(&cur6, 0, sizeof(cur6)); 293 do { 294 found = 0; 295 last6 = cur6; 296 memset(&cur6, 0xff, sizeof(cur6)); 297 TAILQ_FOREACH(ifa, &SC2IFP(sc)->if_addrlist, ifa_list) { 298 in6 = ifatoia6(ifa)->ia_addr.sin6_addr; 299 if (IN6_IS_SCOPE_EMBED(&in6)) 300 in6.s6_addr16[1] = 0; 301 if (ifa->ifa_addr->sa_family == AF_INET6 && 302 memcmp(&in6, &last6, sizeof(in6)) > 0 && 303 memcmp(&in6, &cur6, sizeof(in6)) < 0) { 304 cur6 = in6; 305 found++; 306 } 307 } 308 if (found) 309 SHA1Update(&sc->sc_sha1, (void *)&cur6, sizeof(cur6)); 310 } while (found); 311 #endif /* INET6 */ 312 313 /* convert ipad to opad */ 314 for (i = 0; i < sizeof(sc->sc_pad); i++) 315 sc->sc_pad[i] ^= 0x36 ^ 0x5c; 316 317 if (sc->sc_carpdev) 318 CARP_SCUNLOCK(sc); 319 } 320 321 static void 322 carp_hmac_generate(struct carp_softc *sc, u_int32_t counter[2], 323 unsigned char md[20]) 324 { 325 SHA1_CTX sha1ctx; 326 327 /* fetch first half of inner hash */ 328 bcopy(&sc->sc_sha1, &sha1ctx, sizeof(sha1ctx)); 329 330 SHA1Update(&sha1ctx, (void *)counter, sizeof(sc->sc_counter)); 331 SHA1Final(md, &sha1ctx); 332 333 /* outer hash */ 334 SHA1Init(&sha1ctx); 335 SHA1Update(&sha1ctx, sc->sc_pad, sizeof(sc->sc_pad)); 336 SHA1Update(&sha1ctx, md, 20); 337 SHA1Final(md, &sha1ctx); 338 } 339 340 static int 341 carp_hmac_verify(struct carp_softc *sc, u_int32_t counter[2], 342 unsigned char md[20]) 343 { 344 unsigned char md2[20]; 345 346 CARP_SCLOCK_ASSERT(sc); 347 348 carp_hmac_generate(sc, counter, md2); 349 350 return (bcmp(md, md2, sizeof(md2))); 351 } 352 353 static void 354 carp_setroute(struct carp_softc *sc, int cmd) 355 { 356 struct ifaddr *ifa; 357 int s; 358 359 if (sc->sc_carpdev) 360 CARP_SCLOCK_ASSERT(sc); 361 362 s = splnet(); 363 TAILQ_FOREACH(ifa, &SC2IFP(sc)->if_addrlist, ifa_list) { 364 if (ifa->ifa_addr->sa_family == AF_INET && 365 sc->sc_carpdev != NULL) { 366 int count = carp_addrcount( 367 (struct carp_if *)sc->sc_carpdev->if_carp, 368 ifatoia(ifa), CARP_COUNT_MASTER); 369 370 if ((cmd == RTM_ADD && count == 1) || 371 (cmd == RTM_DELETE && count == 0)) 372 rtinit(ifa, cmd, RTF_UP | RTF_HOST); 373 } 374 #ifdef INET6 375 if (ifa->ifa_addr->sa_family == AF_INET6) { 376 if (cmd == RTM_ADD) 377 in6_ifaddloop(ifa); 378 else 379 in6_ifremloop(ifa); 380 } 381 #endif /* INET6 */ 382 } 383 splx(s); 384 } 385 386 static int 387 carp_clone_create(struct if_clone *ifc, int unit, caddr_t params) 388 { 389 390 struct carp_softc *sc; 391 struct ifnet *ifp; 392 393 sc = malloc(sizeof(*sc), M_CARP, M_WAITOK|M_ZERO); 394 ifp = SC2IFP(sc) = if_alloc(IFT_ETHER); 395 if (ifp == NULL) { 396 free(sc, M_CARP); 397 return (ENOSPC); 398 } 399 400 sc->sc_flags_backup = 0; 401 sc->sc_suppress = 0; 402 sc->sc_advbase = CARP_DFLTINTV; 403 sc->sc_vhid = -1; /* required setting */ 404 sc->sc_advskew = 0; 405 sc->sc_init_counter = 1; 406 sc->sc_naddrs = sc->sc_naddrs6 = 0; /* M_ZERO? */ 407 #ifdef INET6 408 sc->sc_im6o.im6o_multicast_hlim = CARP_DFLTTL; 409 #endif 410 sc->sc_imo.imo_membership = (struct in_multi **)malloc( 411 (sizeof(struct in_multi *) * IP_MIN_MEMBERSHIPS), M_CARP, 412 M_WAITOK); 413 sc->sc_imo.imo_mfilters = NULL; 414 sc->sc_imo.imo_max_memberships = IP_MIN_MEMBERSHIPS; 415 sc->sc_imo.imo_multicast_vif = -1; 416 417 callout_init(&sc->sc_ad_tmo, CALLOUT_MPSAFE); 418 callout_init(&sc->sc_md_tmo, CALLOUT_MPSAFE); 419 callout_init(&sc->sc_md6_tmo, CALLOUT_MPSAFE); 420 421 ifp->if_softc = sc; 422 if_initname(ifp, CARP_IFNAME, unit); 423 ifp->if_mtu = ETHERMTU; 424 ifp->if_flags = IFF_LOOPBACK; 425 ifp->if_ioctl = carp_ioctl; 426 ifp->if_output = carp_looutput; 427 ifp->if_start = carp_start; 428 ifp->if_type = IFT_CARP; 429 ifp->if_snd.ifq_maxlen = ifqmaxlen; 430 ifp->if_hdrlen = 0; 431 if_attach(ifp); 432 bpfattach(SC2IFP(sc), DLT_NULL, sizeof(u_int32_t)); 433 mtx_lock(&carp_mtx); 434 LIST_INSERT_HEAD(&carpif_list, sc, sc_next); 435 mtx_unlock(&carp_mtx); 436 return (0); 437 } 438 439 static void 440 carp_clone_destroy(struct ifnet *ifp) 441 { 442 struct carp_softc *sc = ifp->if_softc; 443 444 if (sc->sc_carpdev) 445 CARP_SCLOCK(sc); 446 carpdetach(sc, 1); /* Returns unlocked. */ 447 448 mtx_lock(&carp_mtx); 449 LIST_REMOVE(sc, sc_next); 450 mtx_unlock(&carp_mtx); 451 bpfdetach(ifp); 452 if_detach(ifp); 453 if_free_type(ifp, IFT_ETHER); 454 free(sc->sc_imo.imo_membership, M_CARP); 455 free(sc, M_CARP); 456 } 457 458 /* 459 * This function can be called on CARP interface destroy path, 460 * and in case of the removal of the underlying interface as 461 * well. We differentiate these two cases. In the latter case 462 * we do not cleanup our multicast memberships, since they 463 * are already freed. Also, in the latter case we do not 464 * release the lock on return, because the function will be 465 * called once more, for another CARP instance on the same 466 * interface. 467 */ 468 static void 469 carpdetach(struct carp_softc *sc, int unlock) 470 { 471 struct carp_if *cif; 472 473 callout_stop(&sc->sc_ad_tmo); 474 callout_stop(&sc->sc_md_tmo); 475 callout_stop(&sc->sc_md6_tmo); 476 477 if (sc->sc_suppress) 478 carp_suppress_preempt--; 479 sc->sc_suppress = 0; 480 481 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) 482 carp_suppress_preempt--; 483 sc->sc_sendad_errors = 0; 484 485 carp_set_state(sc, INIT); 486 SC2IFP(sc)->if_flags &= ~IFF_UP; 487 carp_setrun(sc, 0); 488 if (unlock) 489 carp_multicast_cleanup(sc); 490 #ifdef INET6 491 carp_multicast6_cleanup(sc); 492 #endif 493 494 if (sc->sc_carpdev != NULL) { 495 cif = (struct carp_if *)sc->sc_carpdev->if_carp; 496 CARP_LOCK_ASSERT(cif); 497 TAILQ_REMOVE(&cif->vhif_vrs, sc, sc_list); 498 if (!--cif->vhif_nvrs) { 499 ifpromisc(sc->sc_carpdev, 0); 500 sc->sc_carpdev->if_carp = NULL; 501 CARP_LOCK_DESTROY(cif); 502 free(cif, M_IFADDR); 503 } else if (unlock) 504 CARP_UNLOCK(cif); 505 sc->sc_carpdev = NULL; 506 } 507 } 508 509 /* Detach an interface from the carp. */ 510 static void 511 carp_ifdetach(void *arg __unused, struct ifnet *ifp) 512 { 513 struct carp_if *cif = (struct carp_if *)ifp->if_carp; 514 struct carp_softc *sc, *nextsc; 515 516 if (cif == NULL) 517 return; 518 519 /* 520 * XXX: At the end of for() cycle the lock will be destroyed. 521 */ 522 CARP_LOCK(cif); 523 for (sc = TAILQ_FIRST(&cif->vhif_vrs); sc; sc = nextsc) { 524 nextsc = TAILQ_NEXT(sc, sc_list); 525 carpdetach(sc, 0); 526 } 527 } 528 529 /* 530 * process input packet. 531 * we have rearranged checks order compared to the rfc, 532 * but it seems more efficient this way or not possible otherwise. 533 */ 534 void 535 carp_input(struct mbuf *m, int hlen) 536 { 537 struct ip *ip = mtod(m, struct ip *); 538 struct carp_header *ch; 539 int iplen, len; 540 541 carpstats.carps_ipackets++; 542 543 if (!carp_opts[CARPCTL_ALLOW]) { 544 m_freem(m); 545 return; 546 } 547 548 /* check if received on a valid carp interface */ 549 if (m->m_pkthdr.rcvif->if_carp == NULL) { 550 carpstats.carps_badif++; 551 CARP_LOG("carp_input: packet received on non-carp " 552 "interface: %s\n", 553 m->m_pkthdr.rcvif->if_xname); 554 m_freem(m); 555 return; 556 } 557 558 /* verify that the IP TTL is 255. */ 559 if (ip->ip_ttl != CARP_DFLTTL) { 560 carpstats.carps_badttl++; 561 CARP_LOG("carp_input: received ttl %d != 255i on %s\n", 562 ip->ip_ttl, 563 m->m_pkthdr.rcvif->if_xname); 564 m_freem(m); 565 return; 566 } 567 568 iplen = ip->ip_hl << 2; 569 570 if (m->m_pkthdr.len < iplen + sizeof(*ch)) { 571 carpstats.carps_badlen++; 572 CARP_LOG("carp_input: received len %zd < " 573 "sizeof(struct carp_header)\n", 574 m->m_len - sizeof(struct ip)); 575 m_freem(m); 576 return; 577 } 578 579 if (iplen + sizeof(*ch) < m->m_len) { 580 if ((m = m_pullup(m, iplen + sizeof(*ch))) == NULL) { 581 carpstats.carps_hdrops++; 582 CARP_LOG("carp_input: pullup failed\n"); 583 return; 584 } 585 ip = mtod(m, struct ip *); 586 } 587 ch = (struct carp_header *)((char *)ip + iplen); 588 589 /* 590 * verify that the received packet length is 591 * equal to the CARP header 592 */ 593 len = iplen + sizeof(*ch); 594 if (len > m->m_pkthdr.len) { 595 carpstats.carps_badlen++; 596 CARP_LOG("carp_input: packet too short %d on %s\n", 597 m->m_pkthdr.len, 598 m->m_pkthdr.rcvif->if_xname); 599 m_freem(m); 600 return; 601 } 602 603 if ((m = m_pullup(m, len)) == NULL) { 604 carpstats.carps_hdrops++; 605 return; 606 } 607 ip = mtod(m, struct ip *); 608 ch = (struct carp_header *)((char *)ip + iplen); 609 610 /* verify the CARP checksum */ 611 m->m_data += iplen; 612 if (carp_cksum(m, len - iplen)) { 613 carpstats.carps_badsum++; 614 CARP_LOG("carp_input: checksum failed on %s\n", 615 m->m_pkthdr.rcvif->if_xname); 616 m_freem(m); 617 return; 618 } 619 m->m_data -= iplen; 620 621 carp_input_c(m, ch, AF_INET); 622 } 623 624 #ifdef INET6 625 int 626 carp6_input(struct mbuf **mp, int *offp, int proto) 627 { 628 struct mbuf *m = *mp; 629 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 630 struct carp_header *ch; 631 u_int len; 632 633 carpstats.carps_ipackets6++; 634 635 if (!carp_opts[CARPCTL_ALLOW]) { 636 m_freem(m); 637 return (IPPROTO_DONE); 638 } 639 640 /* check if received on a valid carp interface */ 641 if (m->m_pkthdr.rcvif->if_carp == NULL) { 642 carpstats.carps_badif++; 643 CARP_LOG("carp6_input: packet received on non-carp " 644 "interface: %s\n", 645 m->m_pkthdr.rcvif->if_xname); 646 m_freem(m); 647 return (IPPROTO_DONE); 648 } 649 650 /* verify that the IP TTL is 255 */ 651 if (ip6->ip6_hlim != CARP_DFLTTL) { 652 carpstats.carps_badttl++; 653 CARP_LOG("carp6_input: received ttl %d != 255 on %s\n", 654 ip6->ip6_hlim, 655 m->m_pkthdr.rcvif->if_xname); 656 m_freem(m); 657 return (IPPROTO_DONE); 658 } 659 660 /* verify that we have a complete carp packet */ 661 len = m->m_len; 662 IP6_EXTHDR_GET(ch, struct carp_header *, m, *offp, sizeof(*ch)); 663 if (ch == NULL) { 664 carpstats.carps_badlen++; 665 CARP_LOG("carp6_input: packet size %u too small\n", len); 666 return (IPPROTO_DONE); 667 } 668 669 670 /* verify the CARP checksum */ 671 m->m_data += *offp; 672 if (carp_cksum(m, sizeof(*ch))) { 673 carpstats.carps_badsum++; 674 CARP_LOG("carp6_input: checksum failed, on %s\n", 675 m->m_pkthdr.rcvif->if_xname); 676 m_freem(m); 677 return (IPPROTO_DONE); 678 } 679 m->m_data -= *offp; 680 681 carp_input_c(m, ch, AF_INET6); 682 return (IPPROTO_DONE); 683 } 684 #endif /* INET6 */ 685 686 static void 687 carp_input_c(struct mbuf *m, struct carp_header *ch, sa_family_t af) 688 { 689 struct ifnet *ifp = m->m_pkthdr.rcvif; 690 struct carp_softc *sc; 691 u_int64_t tmp_counter; 692 struct timeval sc_tv, ch_tv; 693 694 /* verify that the VHID is valid on the receiving interface */ 695 CARP_LOCK(ifp->if_carp); 696 TAILQ_FOREACH(sc, &((struct carp_if *)ifp->if_carp)->vhif_vrs, sc_list) 697 if (sc->sc_vhid == ch->carp_vhid) 698 break; 699 700 if (!sc || !((SC2IFP(sc)->if_flags & IFF_UP) && 701 (SC2IFP(sc)->if_drv_flags & IFF_DRV_RUNNING))) { 702 carpstats.carps_badvhid++; 703 CARP_UNLOCK(ifp->if_carp); 704 m_freem(m); 705 return; 706 } 707 708 getmicrotime(&SC2IFP(sc)->if_lastchange); 709 SC2IFP(sc)->if_ipackets++; 710 SC2IFP(sc)->if_ibytes += m->m_pkthdr.len; 711 712 if (bpf_peers_present(SC2IFP(sc)->if_bpf)) { 713 struct ip *ip = mtod(m, struct ip *); 714 uint32_t af1 = af; 715 716 /* BPF wants net byte order */ 717 ip->ip_len = htons(ip->ip_len + (ip->ip_hl << 2)); 718 ip->ip_off = htons(ip->ip_off); 719 bpf_mtap2(SC2IFP(sc)->if_bpf, &af1, sizeof(af1), m); 720 } 721 722 /* verify the CARP version. */ 723 if (ch->carp_version != CARP_VERSION) { 724 carpstats.carps_badver++; 725 SC2IFP(sc)->if_ierrors++; 726 CARP_UNLOCK(ifp->if_carp); 727 CARP_LOG("%s; invalid version %d\n", 728 SC2IFP(sc)->if_xname, 729 ch->carp_version); 730 m_freem(m); 731 return; 732 } 733 734 /* verify the hash */ 735 if (carp_hmac_verify(sc, ch->carp_counter, ch->carp_md)) { 736 carpstats.carps_badauth++; 737 SC2IFP(sc)->if_ierrors++; 738 CARP_UNLOCK(ifp->if_carp); 739 CARP_LOG("%s: incorrect hash\n", SC2IFP(sc)->if_xname); 740 m_freem(m); 741 return; 742 } 743 744 tmp_counter = ntohl(ch->carp_counter[0]); 745 tmp_counter = tmp_counter<<32; 746 tmp_counter += ntohl(ch->carp_counter[1]); 747 748 /* XXX Replay protection goes here */ 749 750 sc->sc_init_counter = 0; 751 sc->sc_counter = tmp_counter; 752 753 sc_tv.tv_sec = sc->sc_advbase; 754 if (carp_suppress_preempt && sc->sc_advskew < 240) 755 sc_tv.tv_usec = 240 * 1000000 / 256; 756 else 757 sc_tv.tv_usec = sc->sc_advskew * 1000000 / 256; 758 ch_tv.tv_sec = ch->carp_advbase; 759 ch_tv.tv_usec = ch->carp_advskew * 1000000 / 256; 760 761 switch (sc->sc_state) { 762 case INIT: 763 break; 764 case MASTER: 765 /* 766 * If we receive an advertisement from a master who's going to 767 * be more frequent than us, go into BACKUP state. 768 */ 769 if (timevalcmp(&sc_tv, &ch_tv, >) || 770 timevalcmp(&sc_tv, &ch_tv, ==)) { 771 callout_stop(&sc->sc_ad_tmo); 772 CARP_DEBUG("%s: MASTER -> BACKUP " 773 "(more frequent advertisement received)\n", 774 SC2IFP(sc)->if_xname); 775 carp_set_state(sc, BACKUP); 776 carp_setrun(sc, 0); 777 carp_setroute(sc, RTM_DELETE); 778 } 779 break; 780 case BACKUP: 781 /* 782 * If we're pre-empting masters who advertise slower than us, 783 * and this one claims to be slower, treat him as down. 784 */ 785 if (carp_opts[CARPCTL_PREEMPT] && 786 timevalcmp(&sc_tv, &ch_tv, <)) { 787 CARP_DEBUG("%s: BACKUP -> MASTER " 788 "(preempting a slower master)\n", 789 SC2IFP(sc)->if_xname); 790 carp_master_down_locked(sc); 791 break; 792 } 793 794 /* 795 * If the master is going to advertise at such a low frequency 796 * that he's guaranteed to time out, we'd might as well just 797 * treat him as timed out now. 798 */ 799 sc_tv.tv_sec = sc->sc_advbase * 3; 800 if (timevalcmp(&sc_tv, &ch_tv, <)) { 801 CARP_DEBUG("%s: BACKUP -> MASTER " 802 "(master timed out)\n", 803 SC2IFP(sc)->if_xname); 804 carp_master_down_locked(sc); 805 break; 806 } 807 808 /* 809 * Otherwise, we reset the counter and wait for the next 810 * advertisement. 811 */ 812 carp_setrun(sc, af); 813 break; 814 } 815 816 CARP_UNLOCK(ifp->if_carp); 817 818 m_freem(m); 819 return; 820 } 821 822 static int 823 carp_prepare_ad(struct mbuf *m, struct carp_softc *sc, struct carp_header *ch) 824 { 825 struct m_tag *mtag; 826 struct ifnet *ifp = SC2IFP(sc); 827 828 if (sc->sc_init_counter) { 829 /* this could also be seconds since unix epoch */ 830 sc->sc_counter = arc4random(); 831 sc->sc_counter = sc->sc_counter << 32; 832 sc->sc_counter += arc4random(); 833 } else 834 sc->sc_counter++; 835 836 ch->carp_counter[0] = htonl((sc->sc_counter>>32)&0xffffffff); 837 ch->carp_counter[1] = htonl(sc->sc_counter&0xffffffff); 838 839 carp_hmac_generate(sc, ch->carp_counter, ch->carp_md); 840 841 /* Tag packet for carp_output */ 842 mtag = m_tag_get(PACKET_TAG_CARP, sizeof(struct ifnet *), M_NOWAIT); 843 if (mtag == NULL) { 844 m_freem(m); 845 SC2IFP(sc)->if_oerrors++; 846 return (ENOMEM); 847 } 848 bcopy(&ifp, (caddr_t)(mtag + 1), sizeof(struct ifnet *)); 849 m_tag_prepend(m, mtag); 850 851 return (0); 852 } 853 854 static void 855 carp_send_ad_all(void) 856 { 857 struct carp_softc *sc; 858 859 mtx_lock(&carp_mtx); 860 LIST_FOREACH(sc, &carpif_list, sc_next) { 861 if (sc->sc_carpdev == NULL) 862 continue; 863 CARP_SCLOCK(sc); 864 if ((SC2IFP(sc)->if_flags & IFF_UP) && 865 (SC2IFP(sc)->if_drv_flags & IFF_DRV_RUNNING) && 866 sc->sc_state == MASTER) 867 carp_send_ad_locked(sc); 868 CARP_SCUNLOCK(sc); 869 } 870 mtx_unlock(&carp_mtx); 871 } 872 873 static void 874 carp_send_ad(void *v) 875 { 876 struct carp_softc *sc = v; 877 878 CARP_SCLOCK(sc); 879 carp_send_ad_locked(sc); 880 CARP_SCUNLOCK(sc); 881 } 882 883 static void 884 carp_send_ad_locked(struct carp_softc *sc) 885 { 886 struct carp_header ch; 887 struct timeval tv; 888 struct carp_header *ch_ptr; 889 struct mbuf *m; 890 int len, advbase, advskew; 891 892 CARP_SCLOCK_ASSERT(sc); 893 894 /* bow out if we've lost our UPness or RUNNINGuiness */ 895 if (!((SC2IFP(sc)->if_flags & IFF_UP) && 896 (SC2IFP(sc)->if_drv_flags & IFF_DRV_RUNNING))) { 897 advbase = 255; 898 advskew = 255; 899 } else { 900 advbase = sc->sc_advbase; 901 if (!carp_suppress_preempt || sc->sc_advskew > 240) 902 advskew = sc->sc_advskew; 903 else 904 advskew = 240; 905 tv.tv_sec = advbase; 906 tv.tv_usec = advskew * 1000000 / 256; 907 } 908 909 ch.carp_version = CARP_VERSION; 910 ch.carp_type = CARP_ADVERTISEMENT; 911 ch.carp_vhid = sc->sc_vhid; 912 ch.carp_advbase = advbase; 913 ch.carp_advskew = advskew; 914 ch.carp_authlen = 7; /* XXX DEFINE */ 915 ch.carp_pad1 = 0; /* must be zero */ 916 ch.carp_cksum = 0; 917 918 #ifdef INET 919 INIT_VNET_INET(curvnet); 920 if (sc->sc_ia) { 921 struct ip *ip; 922 923 MGETHDR(m, M_DONTWAIT, MT_HEADER); 924 if (m == NULL) { 925 SC2IFP(sc)->if_oerrors++; 926 carpstats.carps_onomem++; 927 /* XXX maybe less ? */ 928 if (advbase != 255 || advskew != 255) 929 callout_reset(&sc->sc_ad_tmo, tvtohz(&tv), 930 carp_send_ad, sc); 931 return; 932 } 933 len = sizeof(*ip) + sizeof(ch); 934 m->m_pkthdr.len = len; 935 m->m_pkthdr.rcvif = NULL; 936 m->m_len = len; 937 MH_ALIGN(m, m->m_len); 938 m->m_flags |= M_MCAST; 939 ip = mtod(m, struct ip *); 940 ip->ip_v = IPVERSION; 941 ip->ip_hl = sizeof(*ip) >> 2; 942 ip->ip_tos = IPTOS_LOWDELAY; 943 ip->ip_len = len; 944 ip->ip_id = ip_newid(); 945 ip->ip_off = IP_DF; 946 ip->ip_ttl = CARP_DFLTTL; 947 ip->ip_p = IPPROTO_CARP; 948 ip->ip_sum = 0; 949 ip->ip_src.s_addr = sc->sc_ia->ia_addr.sin_addr.s_addr; 950 ip->ip_dst.s_addr = htonl(INADDR_CARP_GROUP); 951 952 ch_ptr = (struct carp_header *)(&ip[1]); 953 bcopy(&ch, ch_ptr, sizeof(ch)); 954 if (carp_prepare_ad(m, sc, ch_ptr)) 955 return; 956 957 m->m_data += sizeof(*ip); 958 ch_ptr->carp_cksum = carp_cksum(m, len - sizeof(*ip)); 959 m->m_data -= sizeof(*ip); 960 961 getmicrotime(&SC2IFP(sc)->if_lastchange); 962 SC2IFP(sc)->if_opackets++; 963 SC2IFP(sc)->if_obytes += len; 964 carpstats.carps_opackets++; 965 966 if (ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo, NULL)) { 967 SC2IFP(sc)->if_oerrors++; 968 if (sc->sc_sendad_errors < INT_MAX) 969 sc->sc_sendad_errors++; 970 if (sc->sc_sendad_errors == CARP_SENDAD_MAX_ERRORS) { 971 carp_suppress_preempt++; 972 if (carp_suppress_preempt == 1) { 973 CARP_SCUNLOCK(sc); 974 carp_send_ad_all(); 975 CARP_SCLOCK(sc); 976 } 977 } 978 sc->sc_sendad_success = 0; 979 } else { 980 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) { 981 if (++sc->sc_sendad_success >= 982 CARP_SENDAD_MIN_SUCCESS) { 983 carp_suppress_preempt--; 984 sc->sc_sendad_errors = 0; 985 } 986 } else 987 sc->sc_sendad_errors = 0; 988 } 989 } 990 #endif /* INET */ 991 #ifdef INET6 992 if (sc->sc_ia6) { 993 struct ip6_hdr *ip6; 994 995 MGETHDR(m, M_DONTWAIT, MT_HEADER); 996 if (m == NULL) { 997 SC2IFP(sc)->if_oerrors++; 998 carpstats.carps_onomem++; 999 /* XXX maybe less ? */ 1000 if (advbase != 255 || advskew != 255) 1001 callout_reset(&sc->sc_ad_tmo, tvtohz(&tv), 1002 carp_send_ad, sc); 1003 return; 1004 } 1005 len = sizeof(*ip6) + sizeof(ch); 1006 m->m_pkthdr.len = len; 1007 m->m_pkthdr.rcvif = NULL; 1008 m->m_len = len; 1009 MH_ALIGN(m, m->m_len); 1010 m->m_flags |= M_MCAST; 1011 ip6 = mtod(m, struct ip6_hdr *); 1012 bzero(ip6, sizeof(*ip6)); 1013 ip6->ip6_vfc |= IPV6_VERSION; 1014 ip6->ip6_hlim = CARP_DFLTTL; 1015 ip6->ip6_nxt = IPPROTO_CARP; 1016 bcopy(&sc->sc_ia6->ia_addr.sin6_addr, &ip6->ip6_src, 1017 sizeof(struct in6_addr)); 1018 /* set the multicast destination */ 1019 1020 ip6->ip6_dst.s6_addr16[0] = htons(0xff02); 1021 ip6->ip6_dst.s6_addr8[15] = 0x12; 1022 if (in6_setscope(&ip6->ip6_dst, sc->sc_carpdev, NULL) != 0) { 1023 SC2IFP(sc)->if_oerrors++; 1024 m_freem(m); 1025 CARP_LOG("%s: in6_setscope failed\n", __func__); 1026 return; 1027 } 1028 1029 ch_ptr = (struct carp_header *)(&ip6[1]); 1030 bcopy(&ch, ch_ptr, sizeof(ch)); 1031 if (carp_prepare_ad(m, sc, ch_ptr)) 1032 return; 1033 1034 m->m_data += sizeof(*ip6); 1035 ch_ptr->carp_cksum = carp_cksum(m, len - sizeof(*ip6)); 1036 m->m_data -= sizeof(*ip6); 1037 1038 getmicrotime(&SC2IFP(sc)->if_lastchange); 1039 SC2IFP(sc)->if_opackets++; 1040 SC2IFP(sc)->if_obytes += len; 1041 carpstats.carps_opackets6++; 1042 1043 if (ip6_output(m, NULL, NULL, 0, &sc->sc_im6o, NULL, NULL)) { 1044 SC2IFP(sc)->if_oerrors++; 1045 if (sc->sc_sendad_errors < INT_MAX) 1046 sc->sc_sendad_errors++; 1047 if (sc->sc_sendad_errors == CARP_SENDAD_MAX_ERRORS) { 1048 carp_suppress_preempt++; 1049 if (carp_suppress_preempt == 1) { 1050 CARP_SCUNLOCK(sc); 1051 carp_send_ad_all(); 1052 CARP_SCLOCK(sc); 1053 } 1054 } 1055 sc->sc_sendad_success = 0; 1056 } else { 1057 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) { 1058 if (++sc->sc_sendad_success >= 1059 CARP_SENDAD_MIN_SUCCESS) { 1060 carp_suppress_preempt--; 1061 sc->sc_sendad_errors = 0; 1062 } 1063 } else 1064 sc->sc_sendad_errors = 0; 1065 } 1066 } 1067 #endif /* INET6 */ 1068 1069 if (advbase != 255 || advskew != 255) 1070 callout_reset(&sc->sc_ad_tmo, tvtohz(&tv), 1071 carp_send_ad, sc); 1072 1073 } 1074 1075 /* 1076 * Broadcast a gratuitous ARP request containing 1077 * the virtual router MAC address for each IP address 1078 * associated with the virtual router. 1079 */ 1080 static void 1081 carp_send_arp(struct carp_softc *sc) 1082 { 1083 struct ifaddr *ifa; 1084 1085 TAILQ_FOREACH(ifa, &SC2IFP(sc)->if_addrlist, ifa_list) { 1086 1087 if (ifa->ifa_addr->sa_family != AF_INET) 1088 continue; 1089 1090 /* arprequest(sc->sc_carpdev, &in, &in, IF_LLADDR(sc->sc_ifp)); */ 1091 arp_ifinit2(sc->sc_carpdev, ifa, IF_LLADDR(sc->sc_ifp)); 1092 1093 DELAY(1000); /* XXX */ 1094 } 1095 } 1096 1097 #ifdef INET6 1098 static void 1099 carp_send_na(struct carp_softc *sc) 1100 { 1101 struct ifaddr *ifa; 1102 struct in6_addr *in6; 1103 static struct in6_addr mcast = IN6ADDR_LINKLOCAL_ALLNODES_INIT; 1104 1105 TAILQ_FOREACH(ifa, &SC2IFP(sc)->if_addrlist, ifa_list) { 1106 1107 if (ifa->ifa_addr->sa_family != AF_INET6) 1108 continue; 1109 1110 in6 = &ifatoia6(ifa)->ia_addr.sin6_addr; 1111 nd6_na_output(sc->sc_carpdev, &mcast, in6, 1112 ND_NA_FLAG_OVERRIDE, 1, NULL); 1113 DELAY(1000); /* XXX */ 1114 } 1115 } 1116 #endif /* INET6 */ 1117 1118 static int 1119 carp_addrcount(struct carp_if *cif, struct in_ifaddr *ia, int type) 1120 { 1121 struct carp_softc *vh; 1122 struct ifaddr *ifa; 1123 int count = 0; 1124 1125 CARP_LOCK_ASSERT(cif); 1126 1127 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 1128 if ((type == CARP_COUNT_RUNNING && 1129 (SC2IFP(vh)->if_flags & IFF_UP) && 1130 (SC2IFP(vh)->if_drv_flags & IFF_DRV_RUNNING)) || 1131 (type == CARP_COUNT_MASTER && vh->sc_state == MASTER)) { 1132 TAILQ_FOREACH(ifa, &SC2IFP(vh)->if_addrlist, 1133 ifa_list) { 1134 if (ifa->ifa_addr->sa_family == AF_INET && 1135 ia->ia_addr.sin_addr.s_addr == 1136 ifatoia(ifa)->ia_addr.sin_addr.s_addr) 1137 count++; 1138 } 1139 } 1140 } 1141 return (count); 1142 } 1143 1144 int 1145 carp_iamatch(void *v, struct in_ifaddr *ia, 1146 struct in_addr *isaddr, u_int8_t **enaddr) 1147 { 1148 struct carp_if *cif = v; 1149 struct carp_softc *vh; 1150 int index, count = 0; 1151 struct ifaddr *ifa; 1152 1153 CARP_LOCK(cif); 1154 1155 if (carp_opts[CARPCTL_ARPBALANCE]) { 1156 /* 1157 * XXX proof of concept implementation. 1158 * We use the source ip to decide which virtual host should 1159 * handle the request. If we're master of that virtual host, 1160 * then we respond, otherwise, just drop the arp packet on 1161 * the floor. 1162 */ 1163 count = carp_addrcount(cif, ia, CARP_COUNT_RUNNING); 1164 if (count == 0) { 1165 /* should never reach this */ 1166 CARP_UNLOCK(cif); 1167 return (0); 1168 } 1169 1170 /* this should be a hash, like pf_hash() */ 1171 index = ntohl(isaddr->s_addr) % count; 1172 count = 0; 1173 1174 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 1175 if ((SC2IFP(vh)->if_flags & IFF_UP) && 1176 (SC2IFP(vh)->if_drv_flags & IFF_DRV_RUNNING)) { 1177 TAILQ_FOREACH(ifa, &SC2IFP(vh)->if_addrlist, 1178 ifa_list) { 1179 if (ifa->ifa_addr->sa_family == 1180 AF_INET && 1181 ia->ia_addr.sin_addr.s_addr == 1182 ifatoia(ifa)->ia_addr.sin_addr.s_addr) { 1183 if (count == index) { 1184 if (vh->sc_state == 1185 MASTER) { 1186 *enaddr = IF_LLADDR(vh->sc_ifp); 1187 CARP_UNLOCK(cif); 1188 return (1); 1189 } else { 1190 CARP_UNLOCK(cif); 1191 return (0); 1192 } 1193 } 1194 count++; 1195 } 1196 } 1197 } 1198 } 1199 } else { 1200 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 1201 if ((SC2IFP(vh)->if_flags & IFF_UP) && 1202 (SC2IFP(vh)->if_drv_flags & IFF_DRV_RUNNING) && 1203 ia->ia_ifp == SC2IFP(vh) && 1204 vh->sc_state == MASTER) { 1205 *enaddr = IF_LLADDR(vh->sc_ifp); 1206 CARP_UNLOCK(cif); 1207 return (1); 1208 } 1209 } 1210 } 1211 CARP_UNLOCK(cif); 1212 return (0); 1213 } 1214 1215 #ifdef INET6 1216 struct ifaddr * 1217 carp_iamatch6(void *v, struct in6_addr *taddr) 1218 { 1219 struct carp_if *cif = v; 1220 struct carp_softc *vh; 1221 struct ifaddr *ifa; 1222 1223 CARP_LOCK(cif); 1224 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 1225 TAILQ_FOREACH(ifa, &SC2IFP(vh)->if_addrlist, ifa_list) { 1226 if (IN6_ARE_ADDR_EQUAL(taddr, 1227 &ifatoia6(ifa)->ia_addr.sin6_addr) && 1228 (SC2IFP(vh)->if_flags & IFF_UP) && 1229 (SC2IFP(vh)->if_drv_flags & IFF_DRV_RUNNING) && 1230 vh->sc_state == MASTER) { 1231 CARP_UNLOCK(cif); 1232 return (ifa); 1233 } 1234 } 1235 } 1236 CARP_UNLOCK(cif); 1237 1238 return (NULL); 1239 } 1240 1241 void * 1242 carp_macmatch6(void *v, struct mbuf *m, const struct in6_addr *taddr) 1243 { 1244 struct m_tag *mtag; 1245 struct carp_if *cif = v; 1246 struct carp_softc *sc; 1247 struct ifaddr *ifa; 1248 1249 CARP_LOCK(cif); 1250 TAILQ_FOREACH(sc, &cif->vhif_vrs, sc_list) { 1251 TAILQ_FOREACH(ifa, &SC2IFP(sc)->if_addrlist, ifa_list) { 1252 if (IN6_ARE_ADDR_EQUAL(taddr, 1253 &ifatoia6(ifa)->ia_addr.sin6_addr) && 1254 (SC2IFP(sc)->if_flags & IFF_UP) && 1255 (SC2IFP(sc)->if_drv_flags & IFF_DRV_RUNNING)) { 1256 struct ifnet *ifp = SC2IFP(sc); 1257 mtag = m_tag_get(PACKET_TAG_CARP, 1258 sizeof(struct ifnet *), M_NOWAIT); 1259 if (mtag == NULL) { 1260 /* better a bit than nothing */ 1261 CARP_UNLOCK(cif); 1262 return (IF_LLADDR(sc->sc_ifp)); 1263 } 1264 bcopy(&ifp, (caddr_t)(mtag + 1), 1265 sizeof(struct ifnet *)); 1266 m_tag_prepend(m, mtag); 1267 1268 CARP_UNLOCK(cif); 1269 return (IF_LLADDR(sc->sc_ifp)); 1270 } 1271 } 1272 } 1273 CARP_UNLOCK(cif); 1274 1275 return (NULL); 1276 } 1277 #endif 1278 1279 struct ifnet * 1280 carp_forus(void *v, void *dhost) 1281 { 1282 struct carp_if *cif = v; 1283 struct carp_softc *vh; 1284 u_int8_t *ena = dhost; 1285 1286 if (ena[0] || ena[1] || ena[2] != 0x5e || ena[3] || ena[4] != 1) 1287 return (NULL); 1288 1289 CARP_LOCK(cif); 1290 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) 1291 if ((SC2IFP(vh)->if_flags & IFF_UP) && 1292 (SC2IFP(vh)->if_drv_flags & IFF_DRV_RUNNING) && 1293 vh->sc_state == MASTER && 1294 !bcmp(dhost, IF_LLADDR(vh->sc_ifp), ETHER_ADDR_LEN)) { 1295 CARP_UNLOCK(cif); 1296 return (SC2IFP(vh)); 1297 } 1298 1299 CARP_UNLOCK(cif); 1300 return (NULL); 1301 } 1302 1303 static void 1304 carp_master_down(void *v) 1305 { 1306 struct carp_softc *sc = v; 1307 1308 CARP_SCLOCK(sc); 1309 carp_master_down_locked(sc); 1310 CARP_SCUNLOCK(sc); 1311 } 1312 1313 static void 1314 carp_master_down_locked(struct carp_softc *sc) 1315 { 1316 if (sc->sc_carpdev) 1317 CARP_SCLOCK_ASSERT(sc); 1318 1319 switch (sc->sc_state) { 1320 case INIT: 1321 printf("%s: master_down event in INIT state\n", 1322 SC2IFP(sc)->if_xname); 1323 break; 1324 case MASTER: 1325 break; 1326 case BACKUP: 1327 carp_set_state(sc, MASTER); 1328 carp_send_ad_locked(sc); 1329 carp_send_arp(sc); 1330 #ifdef INET6 1331 carp_send_na(sc); 1332 #endif /* INET6 */ 1333 carp_setrun(sc, 0); 1334 carp_setroute(sc, RTM_ADD); 1335 break; 1336 } 1337 } 1338 1339 /* 1340 * When in backup state, af indicates whether to reset the master down timer 1341 * for v4 or v6. If it's set to zero, reset the ones which are already pending. 1342 */ 1343 static void 1344 carp_setrun(struct carp_softc *sc, sa_family_t af) 1345 { 1346 struct timeval tv; 1347 1348 if (sc->sc_carpdev == NULL) { 1349 SC2IFP(sc)->if_drv_flags &= ~IFF_DRV_RUNNING; 1350 carp_set_state(sc, INIT); 1351 return; 1352 } else 1353 CARP_SCLOCK_ASSERT(sc); 1354 1355 if (SC2IFP(sc)->if_flags & IFF_UP && 1356 sc->sc_vhid > 0 && (sc->sc_naddrs || sc->sc_naddrs6)) 1357 SC2IFP(sc)->if_drv_flags |= IFF_DRV_RUNNING; 1358 else { 1359 SC2IFP(sc)->if_drv_flags &= ~IFF_DRV_RUNNING; 1360 carp_setroute(sc, RTM_DELETE); 1361 return; 1362 } 1363 1364 switch (sc->sc_state) { 1365 case INIT: 1366 if (carp_opts[CARPCTL_PREEMPT] && !carp_suppress_preempt) { 1367 carp_send_ad_locked(sc); 1368 carp_send_arp(sc); 1369 #ifdef INET6 1370 carp_send_na(sc); 1371 #endif /* INET6 */ 1372 CARP_DEBUG("%s: INIT -> MASTER (preempting)\n", 1373 SC2IFP(sc)->if_xname); 1374 carp_set_state(sc, MASTER); 1375 carp_setroute(sc, RTM_ADD); 1376 } else { 1377 CARP_DEBUG("%s: INIT -> BACKUP\n", SC2IFP(sc)->if_xname); 1378 carp_set_state(sc, BACKUP); 1379 carp_setroute(sc, RTM_DELETE); 1380 carp_setrun(sc, 0); 1381 } 1382 break; 1383 case BACKUP: 1384 callout_stop(&sc->sc_ad_tmo); 1385 tv.tv_sec = 3 * sc->sc_advbase; 1386 tv.tv_usec = sc->sc_advskew * 1000000 / 256; 1387 switch (af) { 1388 #ifdef INET 1389 case AF_INET: 1390 callout_reset(&sc->sc_md_tmo, tvtohz(&tv), 1391 carp_master_down, sc); 1392 break; 1393 #endif /* INET */ 1394 #ifdef INET6 1395 case AF_INET6: 1396 callout_reset(&sc->sc_md6_tmo, tvtohz(&tv), 1397 carp_master_down, sc); 1398 break; 1399 #endif /* INET6 */ 1400 default: 1401 if (sc->sc_naddrs) 1402 callout_reset(&sc->sc_md_tmo, tvtohz(&tv), 1403 carp_master_down, sc); 1404 if (sc->sc_naddrs6) 1405 callout_reset(&sc->sc_md6_tmo, tvtohz(&tv), 1406 carp_master_down, sc); 1407 break; 1408 } 1409 break; 1410 case MASTER: 1411 tv.tv_sec = sc->sc_advbase; 1412 tv.tv_usec = sc->sc_advskew * 1000000 / 256; 1413 callout_reset(&sc->sc_ad_tmo, tvtohz(&tv), 1414 carp_send_ad, sc); 1415 break; 1416 } 1417 } 1418 1419 static void 1420 carp_multicast_cleanup(struct carp_softc *sc) 1421 { 1422 struct ip_moptions *imo = &sc->sc_imo; 1423 u_int16_t n = imo->imo_num_memberships; 1424 1425 /* Clean up our own multicast memberships */ 1426 while (n-- > 0) { 1427 if (imo->imo_membership[n] != NULL) { 1428 in_delmulti(imo->imo_membership[n]); 1429 imo->imo_membership[n] = NULL; 1430 } 1431 } 1432 KASSERT(imo->imo_mfilters == NULL, 1433 ("%s: imo_mfilters != NULL", __func__)); 1434 imo->imo_num_memberships = 0; 1435 imo->imo_multicast_ifp = NULL; 1436 } 1437 1438 #ifdef INET6 1439 static void 1440 carp_multicast6_cleanup(struct carp_softc *sc) 1441 { 1442 struct ip6_moptions *im6o = &sc->sc_im6o; 1443 1444 while (!LIST_EMPTY(&im6o->im6o_memberships)) { 1445 struct in6_multi_mship *imm = 1446 LIST_FIRST(&im6o->im6o_memberships); 1447 1448 LIST_REMOVE(imm, i6mm_chain); 1449 in6_leavegroup(imm); 1450 } 1451 im6o->im6o_multicast_ifp = NULL; 1452 } 1453 #endif 1454 1455 static int 1456 carp_set_addr(struct carp_softc *sc, struct sockaddr_in *sin) 1457 { 1458 INIT_VNET_INET(curvnet); 1459 struct ifnet *ifp; 1460 struct carp_if *cif; 1461 struct in_ifaddr *ia, *ia_if; 1462 struct ip_moptions *imo = &sc->sc_imo; 1463 struct in_addr addr; 1464 u_long iaddr = htonl(sin->sin_addr.s_addr); 1465 int own, error; 1466 1467 if (sin->sin_addr.s_addr == 0) { 1468 if (!(SC2IFP(sc)->if_flags & IFF_UP)) 1469 carp_set_state(sc, INIT); 1470 if (sc->sc_naddrs) 1471 SC2IFP(sc)->if_flags |= IFF_UP; 1472 if (sc->sc_carpdev) 1473 CARP_SCLOCK(sc); 1474 carp_setrun(sc, 0); 1475 if (sc->sc_carpdev) 1476 CARP_SCUNLOCK(sc); 1477 return (0); 1478 } 1479 1480 /* we have to do it by hands to check we won't match on us */ 1481 ia_if = NULL; own = 0; 1482 TAILQ_FOREACH(ia, &V_in_ifaddrhead, ia_link) { 1483 /* and, yeah, we need a multicast-capable iface too */ 1484 if (ia->ia_ifp != SC2IFP(sc) && 1485 (ia->ia_ifp->if_flags & IFF_MULTICAST) && 1486 (iaddr & ia->ia_subnetmask) == ia->ia_subnet) { 1487 if (!ia_if) 1488 ia_if = ia; 1489 if (sin->sin_addr.s_addr == 1490 ia->ia_addr.sin_addr.s_addr) 1491 own++; 1492 } 1493 } 1494 1495 if (!ia_if) 1496 return (EADDRNOTAVAIL); 1497 1498 ia = ia_if; 1499 ifp = ia->ia_ifp; 1500 1501 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0 || 1502 (imo->imo_multicast_ifp && imo->imo_multicast_ifp != ifp)) 1503 return (EADDRNOTAVAIL); 1504 1505 if (imo->imo_num_memberships == 0) { 1506 addr.s_addr = htonl(INADDR_CARP_GROUP); 1507 if ((imo->imo_membership[0] = in_addmulti(&addr, ifp)) == NULL) 1508 return (ENOBUFS); 1509 imo->imo_num_memberships++; 1510 imo->imo_multicast_ifp = ifp; 1511 imo->imo_multicast_ttl = CARP_DFLTTL; 1512 imo->imo_multicast_loop = 0; 1513 } 1514 1515 if (!ifp->if_carp) { 1516 1517 cif = malloc(sizeof(*cif), M_CARP, 1518 M_WAITOK|M_ZERO); 1519 if (!cif) { 1520 error = ENOBUFS; 1521 goto cleanup; 1522 } 1523 if ((error = ifpromisc(ifp, 1))) { 1524 free(cif, M_CARP); 1525 goto cleanup; 1526 } 1527 1528 CARP_LOCK_INIT(cif); 1529 CARP_LOCK(cif); 1530 cif->vhif_ifp = ifp; 1531 TAILQ_INIT(&cif->vhif_vrs); 1532 ifp->if_carp = cif; 1533 1534 } else { 1535 struct carp_softc *vr; 1536 1537 cif = (struct carp_if *)ifp->if_carp; 1538 CARP_LOCK(cif); 1539 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) 1540 if (vr != sc && vr->sc_vhid == sc->sc_vhid) { 1541 CARP_UNLOCK(cif); 1542 error = EEXIST; 1543 goto cleanup; 1544 } 1545 } 1546 sc->sc_ia = ia; 1547 sc->sc_carpdev = ifp; 1548 1549 { /* XXX prevent endless loop if already in queue */ 1550 struct carp_softc *vr, *after = NULL; 1551 int myself = 0; 1552 cif = (struct carp_if *)ifp->if_carp; 1553 1554 /* XXX: cif should not change, right? So we still hold the lock */ 1555 CARP_LOCK_ASSERT(cif); 1556 1557 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) { 1558 if (vr == sc) 1559 myself = 1; 1560 if (vr->sc_vhid < sc->sc_vhid) 1561 after = vr; 1562 } 1563 1564 if (!myself) { 1565 /* We're trying to keep things in order */ 1566 if (after == NULL) { 1567 TAILQ_INSERT_TAIL(&cif->vhif_vrs, sc, sc_list); 1568 } else { 1569 TAILQ_INSERT_AFTER(&cif->vhif_vrs, after, sc, sc_list); 1570 } 1571 cif->vhif_nvrs++; 1572 } 1573 } 1574 1575 sc->sc_naddrs++; 1576 SC2IFP(sc)->if_flags |= IFF_UP; 1577 if (own) 1578 sc->sc_advskew = 0; 1579 carp_sc_state_locked(sc); 1580 carp_setrun(sc, 0); 1581 1582 CARP_UNLOCK(cif); 1583 1584 return (0); 1585 1586 cleanup: 1587 in_delmulti(imo->imo_membership[--imo->imo_num_memberships]); 1588 return (error); 1589 } 1590 1591 static int 1592 carp_del_addr(struct carp_softc *sc, struct sockaddr_in *sin) 1593 { 1594 int error = 0; 1595 1596 if (!--sc->sc_naddrs) { 1597 struct carp_if *cif = (struct carp_if *)sc->sc_carpdev->if_carp; 1598 struct ip_moptions *imo = &sc->sc_imo; 1599 1600 CARP_LOCK(cif); 1601 callout_stop(&sc->sc_ad_tmo); 1602 SC2IFP(sc)->if_flags &= ~IFF_UP; 1603 SC2IFP(sc)->if_drv_flags &= ~IFF_DRV_RUNNING; 1604 sc->sc_vhid = -1; 1605 in_delmulti(imo->imo_membership[--imo->imo_num_memberships]); 1606 imo->imo_multicast_ifp = NULL; 1607 TAILQ_REMOVE(&cif->vhif_vrs, sc, sc_list); 1608 if (!--cif->vhif_nvrs) { 1609 sc->sc_carpdev->if_carp = NULL; 1610 CARP_LOCK_DESTROY(cif); 1611 free(cif, M_IFADDR); 1612 } else { 1613 CARP_UNLOCK(cif); 1614 } 1615 } 1616 1617 return (error); 1618 } 1619 1620 #ifdef INET6 1621 static int 1622 carp_set_addr6(struct carp_softc *sc, struct sockaddr_in6 *sin6) 1623 { 1624 INIT_VNET_INET6(curvnet); 1625 struct ifnet *ifp; 1626 struct carp_if *cif; 1627 struct in6_ifaddr *ia, *ia_if; 1628 struct ip6_moptions *im6o = &sc->sc_im6o; 1629 struct in6_multi_mship *imm; 1630 struct in6_addr in6; 1631 int own, error; 1632 1633 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) { 1634 if (!(SC2IFP(sc)->if_flags & IFF_UP)) 1635 carp_set_state(sc, INIT); 1636 if (sc->sc_naddrs6) 1637 SC2IFP(sc)->if_flags |= IFF_UP; 1638 if (sc->sc_carpdev) 1639 CARP_SCLOCK(sc); 1640 carp_setrun(sc, 0); 1641 if (sc->sc_carpdev) 1642 CARP_SCUNLOCK(sc); 1643 return (0); 1644 } 1645 1646 /* we have to do it by hands to check we won't match on us */ 1647 ia_if = NULL; own = 0; 1648 for (ia = V_in6_ifaddr; ia; ia = ia->ia_next) { 1649 int i; 1650 1651 for (i = 0; i < 4; i++) { 1652 if ((sin6->sin6_addr.s6_addr32[i] & 1653 ia->ia_prefixmask.sin6_addr.s6_addr32[i]) != 1654 (ia->ia_addr.sin6_addr.s6_addr32[i] & 1655 ia->ia_prefixmask.sin6_addr.s6_addr32[i])) 1656 break; 1657 } 1658 /* and, yeah, we need a multicast-capable iface too */ 1659 if (ia->ia_ifp != SC2IFP(sc) && 1660 (ia->ia_ifp->if_flags & IFF_MULTICAST) && 1661 (i == 4)) { 1662 if (!ia_if) 1663 ia_if = ia; 1664 if (IN6_ARE_ADDR_EQUAL(&sin6->sin6_addr, 1665 &ia->ia_addr.sin6_addr)) 1666 own++; 1667 } 1668 } 1669 1670 if (!ia_if) 1671 return (EADDRNOTAVAIL); 1672 ia = ia_if; 1673 ifp = ia->ia_ifp; 1674 1675 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0 || 1676 (im6o->im6o_multicast_ifp && im6o->im6o_multicast_ifp != ifp)) 1677 return (EADDRNOTAVAIL); 1678 1679 if (!sc->sc_naddrs6) { 1680 im6o->im6o_multicast_ifp = ifp; 1681 1682 /* join CARP multicast address */ 1683 bzero(&in6, sizeof(in6)); 1684 in6.s6_addr16[0] = htons(0xff02); 1685 in6.s6_addr8[15] = 0x12; 1686 if (in6_setscope(&in6, ifp, NULL) != 0) 1687 goto cleanup; 1688 if ((imm = in6_joingroup(ifp, &in6, &error, 0)) == NULL) 1689 goto cleanup; 1690 LIST_INSERT_HEAD(&im6o->im6o_memberships, imm, i6mm_chain); 1691 1692 /* join solicited multicast address */ 1693 bzero(&in6, sizeof(in6)); 1694 in6.s6_addr16[0] = htons(0xff02); 1695 in6.s6_addr32[1] = 0; 1696 in6.s6_addr32[2] = htonl(1); 1697 in6.s6_addr32[3] = sin6->sin6_addr.s6_addr32[3]; 1698 in6.s6_addr8[12] = 0xff; 1699 if (in6_setscope(&in6, ifp, NULL) != 0) 1700 goto cleanup; 1701 if ((imm = in6_joingroup(ifp, &in6, &error, 0)) == NULL) 1702 goto cleanup; 1703 LIST_INSERT_HEAD(&im6o->im6o_memberships, imm, i6mm_chain); 1704 } 1705 1706 if (!ifp->if_carp) { 1707 cif = malloc(sizeof(*cif), M_CARP, 1708 M_WAITOK|M_ZERO); 1709 if (!cif) { 1710 error = ENOBUFS; 1711 goto cleanup; 1712 } 1713 if ((error = ifpromisc(ifp, 1))) { 1714 free(cif, M_CARP); 1715 goto cleanup; 1716 } 1717 1718 CARP_LOCK_INIT(cif); 1719 CARP_LOCK(cif); 1720 cif->vhif_ifp = ifp; 1721 TAILQ_INIT(&cif->vhif_vrs); 1722 ifp->if_carp = cif; 1723 1724 } else { 1725 struct carp_softc *vr; 1726 1727 cif = (struct carp_if *)ifp->if_carp; 1728 CARP_LOCK(cif); 1729 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) 1730 if (vr != sc && vr->sc_vhid == sc->sc_vhid) { 1731 CARP_UNLOCK(cif); 1732 error = EINVAL; 1733 goto cleanup; 1734 } 1735 } 1736 sc->sc_ia6 = ia; 1737 sc->sc_carpdev = ifp; 1738 1739 { /* XXX prevent endless loop if already in queue */ 1740 struct carp_softc *vr, *after = NULL; 1741 int myself = 0; 1742 cif = (struct carp_if *)ifp->if_carp; 1743 CARP_LOCK_ASSERT(cif); 1744 1745 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) { 1746 if (vr == sc) 1747 myself = 1; 1748 if (vr->sc_vhid < sc->sc_vhid) 1749 after = vr; 1750 } 1751 1752 if (!myself) { 1753 /* We're trying to keep things in order */ 1754 if (after == NULL) { 1755 TAILQ_INSERT_TAIL(&cif->vhif_vrs, sc, sc_list); 1756 } else { 1757 TAILQ_INSERT_AFTER(&cif->vhif_vrs, after, sc, sc_list); 1758 } 1759 cif->vhif_nvrs++; 1760 } 1761 } 1762 1763 sc->sc_naddrs6++; 1764 SC2IFP(sc)->if_flags |= IFF_UP; 1765 if (own) 1766 sc->sc_advskew = 0; 1767 carp_sc_state_locked(sc); 1768 carp_setrun(sc, 0); 1769 1770 CARP_UNLOCK(cif); 1771 1772 return (0); 1773 1774 cleanup: 1775 /* clean up multicast memberships */ 1776 if (!sc->sc_naddrs6) { 1777 while (!LIST_EMPTY(&im6o->im6o_memberships)) { 1778 imm = LIST_FIRST(&im6o->im6o_memberships); 1779 LIST_REMOVE(imm, i6mm_chain); 1780 in6_leavegroup(imm); 1781 } 1782 } 1783 return (error); 1784 } 1785 1786 static int 1787 carp_del_addr6(struct carp_softc *sc, struct sockaddr_in6 *sin6) 1788 { 1789 int error = 0; 1790 1791 if (!--sc->sc_naddrs6) { 1792 struct carp_if *cif = (struct carp_if *)sc->sc_carpdev->if_carp; 1793 struct ip6_moptions *im6o = &sc->sc_im6o; 1794 1795 CARP_LOCK(cif); 1796 callout_stop(&sc->sc_ad_tmo); 1797 SC2IFP(sc)->if_flags &= ~IFF_UP; 1798 SC2IFP(sc)->if_drv_flags &= ~IFF_DRV_RUNNING; 1799 sc->sc_vhid = -1; 1800 while (!LIST_EMPTY(&im6o->im6o_memberships)) { 1801 struct in6_multi_mship *imm = 1802 LIST_FIRST(&im6o->im6o_memberships); 1803 1804 LIST_REMOVE(imm, i6mm_chain); 1805 in6_leavegroup(imm); 1806 } 1807 im6o->im6o_multicast_ifp = NULL; 1808 TAILQ_REMOVE(&cif->vhif_vrs, sc, sc_list); 1809 if (!--cif->vhif_nvrs) { 1810 CARP_LOCK_DESTROY(cif); 1811 sc->sc_carpdev->if_carp = NULL; 1812 free(cif, M_IFADDR); 1813 } else 1814 CARP_UNLOCK(cif); 1815 } 1816 1817 return (error); 1818 } 1819 #endif /* INET6 */ 1820 1821 static int 1822 carp_ioctl(struct ifnet *ifp, u_long cmd, caddr_t addr) 1823 { 1824 struct carp_softc *sc = ifp->if_softc, *vr; 1825 struct carpreq carpr; 1826 struct ifaddr *ifa; 1827 struct ifreq *ifr; 1828 struct ifaliasreq *ifra; 1829 int locked = 0, error = 0; 1830 1831 ifa = (struct ifaddr *)addr; 1832 ifra = (struct ifaliasreq *)addr; 1833 ifr = (struct ifreq *)addr; 1834 1835 switch (cmd) { 1836 case SIOCSIFADDR: 1837 switch (ifa->ifa_addr->sa_family) { 1838 #ifdef INET 1839 case AF_INET: 1840 SC2IFP(sc)->if_flags |= IFF_UP; 1841 bcopy(ifa->ifa_addr, ifa->ifa_dstaddr, 1842 sizeof(struct sockaddr)); 1843 error = carp_set_addr(sc, satosin(ifa->ifa_addr)); 1844 break; 1845 #endif /* INET */ 1846 #ifdef INET6 1847 case AF_INET6: 1848 SC2IFP(sc)->if_flags |= IFF_UP; 1849 error = carp_set_addr6(sc, satosin6(ifa->ifa_addr)); 1850 break; 1851 #endif /* INET6 */ 1852 default: 1853 error = EAFNOSUPPORT; 1854 break; 1855 } 1856 break; 1857 1858 case SIOCAIFADDR: 1859 switch (ifa->ifa_addr->sa_family) { 1860 #ifdef INET 1861 case AF_INET: 1862 SC2IFP(sc)->if_flags |= IFF_UP; 1863 bcopy(ifa->ifa_addr, ifa->ifa_dstaddr, 1864 sizeof(struct sockaddr)); 1865 error = carp_set_addr(sc, satosin(&ifra->ifra_addr)); 1866 break; 1867 #endif /* INET */ 1868 #ifdef INET6 1869 case AF_INET6: 1870 SC2IFP(sc)->if_flags |= IFF_UP; 1871 error = carp_set_addr6(sc, satosin6(&ifra->ifra_addr)); 1872 break; 1873 #endif /* INET6 */ 1874 default: 1875 error = EAFNOSUPPORT; 1876 break; 1877 } 1878 break; 1879 1880 case SIOCDIFADDR: 1881 switch (ifa->ifa_addr->sa_family) { 1882 #ifdef INET 1883 case AF_INET: 1884 error = carp_del_addr(sc, satosin(&ifra->ifra_addr)); 1885 break; 1886 #endif /* INET */ 1887 #ifdef INET6 1888 case AF_INET6: 1889 error = carp_del_addr6(sc, satosin6(&ifra->ifra_addr)); 1890 break; 1891 #endif /* INET6 */ 1892 default: 1893 error = EAFNOSUPPORT; 1894 break; 1895 } 1896 break; 1897 1898 case SIOCSIFFLAGS: 1899 if (sc->sc_carpdev) { 1900 locked = 1; 1901 CARP_SCLOCK(sc); 1902 } 1903 if (sc->sc_state != INIT && !(ifr->ifr_flags & IFF_UP)) { 1904 callout_stop(&sc->sc_ad_tmo); 1905 callout_stop(&sc->sc_md_tmo); 1906 callout_stop(&sc->sc_md6_tmo); 1907 if (sc->sc_state == MASTER) 1908 carp_send_ad_locked(sc); 1909 carp_set_state(sc, INIT); 1910 carp_setrun(sc, 0); 1911 } else if (sc->sc_state == INIT && (ifr->ifr_flags & IFF_UP)) { 1912 SC2IFP(sc)->if_flags |= IFF_UP; 1913 carp_setrun(sc, 0); 1914 } 1915 break; 1916 1917 case SIOCSVH: 1918 error = priv_check(curthread, PRIV_NETINET_CARP); 1919 if (error) 1920 break; 1921 if ((error = copyin(ifr->ifr_data, &carpr, sizeof carpr))) 1922 break; 1923 error = 1; 1924 if (sc->sc_carpdev) { 1925 locked = 1; 1926 CARP_SCLOCK(sc); 1927 } 1928 if (sc->sc_state != INIT && carpr.carpr_state != sc->sc_state) { 1929 switch (carpr.carpr_state) { 1930 case BACKUP: 1931 callout_stop(&sc->sc_ad_tmo); 1932 carp_set_state(sc, BACKUP); 1933 carp_setrun(sc, 0); 1934 carp_setroute(sc, RTM_DELETE); 1935 break; 1936 case MASTER: 1937 carp_master_down_locked(sc); 1938 break; 1939 default: 1940 break; 1941 } 1942 } 1943 if (carpr.carpr_vhid > 0) { 1944 if (carpr.carpr_vhid > 255) { 1945 error = EINVAL; 1946 break; 1947 } 1948 if (sc->sc_carpdev) { 1949 struct carp_if *cif; 1950 cif = (struct carp_if *)sc->sc_carpdev->if_carp; 1951 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) 1952 if (vr != sc && 1953 vr->sc_vhid == carpr.carpr_vhid) { 1954 error = EEXIST; 1955 break; 1956 } 1957 if (error == EEXIST) 1958 break; 1959 } 1960 sc->sc_vhid = carpr.carpr_vhid; 1961 IF_LLADDR(sc->sc_ifp)[0] = 0; 1962 IF_LLADDR(sc->sc_ifp)[1] = 0; 1963 IF_LLADDR(sc->sc_ifp)[2] = 0x5e; 1964 IF_LLADDR(sc->sc_ifp)[3] = 0; 1965 IF_LLADDR(sc->sc_ifp)[4] = 1; 1966 IF_LLADDR(sc->sc_ifp)[5] = sc->sc_vhid; 1967 error--; 1968 } 1969 if (carpr.carpr_advbase > 0 || carpr.carpr_advskew > 0) { 1970 if (carpr.carpr_advskew >= 255) { 1971 error = EINVAL; 1972 break; 1973 } 1974 if (carpr.carpr_advbase > 255) { 1975 error = EINVAL; 1976 break; 1977 } 1978 sc->sc_advbase = carpr.carpr_advbase; 1979 sc->sc_advskew = carpr.carpr_advskew; 1980 error--; 1981 } 1982 bcopy(carpr.carpr_key, sc->sc_key, sizeof(sc->sc_key)); 1983 if (error > 0) 1984 error = EINVAL; 1985 else { 1986 error = 0; 1987 carp_setrun(sc, 0); 1988 } 1989 break; 1990 1991 case SIOCGVH: 1992 /* XXX: lockless read */ 1993 bzero(&carpr, sizeof(carpr)); 1994 carpr.carpr_state = sc->sc_state; 1995 carpr.carpr_vhid = sc->sc_vhid; 1996 carpr.carpr_advbase = sc->sc_advbase; 1997 carpr.carpr_advskew = sc->sc_advskew; 1998 error = priv_check(curthread, PRIV_NETINET_CARP); 1999 if (error == 0) 2000 bcopy(sc->sc_key, carpr.carpr_key, 2001 sizeof(carpr.carpr_key)); 2002 error = copyout(&carpr, ifr->ifr_data, sizeof(carpr)); 2003 break; 2004 2005 default: 2006 error = EINVAL; 2007 } 2008 2009 if (locked) 2010 CARP_SCUNLOCK(sc); 2011 2012 carp_hmac_prepare(sc); 2013 2014 return (error); 2015 } 2016 2017 /* 2018 * XXX: this is looutput. We should eventually use it from there. 2019 */ 2020 static int 2021 carp_looutput(struct ifnet *ifp, struct mbuf *m, struct sockaddr *dst, 2022 struct rtentry *rt) 2023 { 2024 u_int32_t af; 2025 2026 M_ASSERTPKTHDR(m); /* check if we have the packet header */ 2027 2028 if (rt && rt->rt_flags & (RTF_REJECT|RTF_BLACKHOLE)) { 2029 m_freem(m); 2030 return (rt->rt_flags & RTF_BLACKHOLE ? 0 : 2031 rt->rt_flags & RTF_HOST ? EHOSTUNREACH : ENETUNREACH); 2032 } 2033 2034 ifp->if_opackets++; 2035 ifp->if_obytes += m->m_pkthdr.len; 2036 2037 /* BPF writes need to be handled specially. */ 2038 if (dst->sa_family == AF_UNSPEC) { 2039 bcopy(dst->sa_data, &af, sizeof(af)); 2040 dst->sa_family = af; 2041 } 2042 2043 #if 1 /* XXX */ 2044 switch (dst->sa_family) { 2045 case AF_INET: 2046 case AF_INET6: 2047 case AF_IPX: 2048 case AF_APPLETALK: 2049 break; 2050 default: 2051 printf("carp_looutput: af=%d unexpected\n", dst->sa_family); 2052 m_freem(m); 2053 return (EAFNOSUPPORT); 2054 } 2055 #endif 2056 return(if_simloop(ifp, m, dst->sa_family, 0)); 2057 } 2058 2059 /* 2060 * Start output on carp interface. This function should never be called. 2061 */ 2062 static void 2063 carp_start(struct ifnet *ifp) 2064 { 2065 #ifdef DEBUG 2066 printf("%s: start called\n", ifp->if_xname); 2067 #endif 2068 } 2069 2070 int 2071 carp_output(struct ifnet *ifp, struct mbuf *m, struct sockaddr *sa, 2072 struct rtentry *rt) 2073 { 2074 struct m_tag *mtag; 2075 struct carp_softc *sc; 2076 struct ifnet *carp_ifp; 2077 2078 if (!sa) 2079 return (0); 2080 2081 switch (sa->sa_family) { 2082 #ifdef INET 2083 case AF_INET: 2084 break; 2085 #endif /* INET */ 2086 #ifdef INET6 2087 case AF_INET6: 2088 break; 2089 #endif /* INET6 */ 2090 default: 2091 return (0); 2092 } 2093 2094 mtag = m_tag_find(m, PACKET_TAG_CARP, NULL); 2095 if (mtag == NULL) 2096 return (0); 2097 2098 bcopy(mtag + 1, &carp_ifp, sizeof(struct ifnet *)); 2099 sc = carp_ifp->if_softc; 2100 2101 /* Set the source MAC address to Virtual Router MAC Address */ 2102 switch (ifp->if_type) { 2103 case IFT_ETHER: 2104 case IFT_L2VLAN: { 2105 struct ether_header *eh; 2106 2107 eh = mtod(m, struct ether_header *); 2108 eh->ether_shost[0] = 0; 2109 eh->ether_shost[1] = 0; 2110 eh->ether_shost[2] = 0x5e; 2111 eh->ether_shost[3] = 0; 2112 eh->ether_shost[4] = 1; 2113 eh->ether_shost[5] = sc->sc_vhid; 2114 } 2115 break; 2116 case IFT_FDDI: { 2117 struct fddi_header *fh; 2118 2119 fh = mtod(m, struct fddi_header *); 2120 fh->fddi_shost[0] = 0; 2121 fh->fddi_shost[1] = 0; 2122 fh->fddi_shost[2] = 0x5e; 2123 fh->fddi_shost[3] = 0; 2124 fh->fddi_shost[4] = 1; 2125 fh->fddi_shost[5] = sc->sc_vhid; 2126 } 2127 break; 2128 case IFT_ISO88025: { 2129 struct iso88025_header *th; 2130 th = mtod(m, struct iso88025_header *); 2131 th->iso88025_shost[0] = 3; 2132 th->iso88025_shost[1] = 0; 2133 th->iso88025_shost[2] = 0x40 >> (sc->sc_vhid - 1); 2134 th->iso88025_shost[3] = 0x40000 >> (sc->sc_vhid - 1); 2135 th->iso88025_shost[4] = 0; 2136 th->iso88025_shost[5] = 0; 2137 } 2138 break; 2139 default: 2140 printf("%s: carp is not supported for this interface type\n", 2141 ifp->if_xname); 2142 return (EOPNOTSUPP); 2143 } 2144 2145 return (0); 2146 } 2147 2148 static void 2149 carp_set_state(struct carp_softc *sc, int state) 2150 { 2151 2152 if (sc->sc_carpdev) 2153 CARP_SCLOCK_ASSERT(sc); 2154 2155 if (sc->sc_state == state) 2156 return; 2157 2158 sc->sc_state = state; 2159 switch (state) { 2160 case BACKUP: 2161 SC2IFP(sc)->if_link_state = LINK_STATE_DOWN; 2162 break; 2163 case MASTER: 2164 SC2IFP(sc)->if_link_state = LINK_STATE_UP; 2165 break; 2166 default: 2167 SC2IFP(sc)->if_link_state = LINK_STATE_UNKNOWN; 2168 break; 2169 } 2170 rt_ifmsg(SC2IFP(sc)); 2171 } 2172 2173 void 2174 carp_carpdev_state(void *v) 2175 { 2176 struct carp_if *cif = v; 2177 2178 CARP_LOCK(cif); 2179 carp_carpdev_state_locked(cif); 2180 CARP_UNLOCK(cif); 2181 } 2182 2183 static void 2184 carp_carpdev_state_locked(struct carp_if *cif) 2185 { 2186 struct carp_softc *sc; 2187 2188 TAILQ_FOREACH(sc, &cif->vhif_vrs, sc_list) 2189 carp_sc_state_locked(sc); 2190 } 2191 2192 static void 2193 carp_sc_state_locked(struct carp_softc *sc) 2194 { 2195 CARP_SCLOCK_ASSERT(sc); 2196 2197 if (sc->sc_carpdev->if_link_state != LINK_STATE_UP || 2198 !(sc->sc_carpdev->if_flags & IFF_UP)) { 2199 sc->sc_flags_backup = SC2IFP(sc)->if_flags; 2200 SC2IFP(sc)->if_flags &= ~IFF_UP; 2201 SC2IFP(sc)->if_drv_flags &= ~IFF_DRV_RUNNING; 2202 callout_stop(&sc->sc_ad_tmo); 2203 callout_stop(&sc->sc_md_tmo); 2204 callout_stop(&sc->sc_md6_tmo); 2205 carp_set_state(sc, INIT); 2206 carp_setrun(sc, 0); 2207 if (!sc->sc_suppress) { 2208 carp_suppress_preempt++; 2209 if (carp_suppress_preempt == 1) { 2210 CARP_SCUNLOCK(sc); 2211 carp_send_ad_all(); 2212 CARP_SCLOCK(sc); 2213 } 2214 } 2215 sc->sc_suppress = 1; 2216 } else { 2217 SC2IFP(sc)->if_flags |= sc->sc_flags_backup; 2218 carp_set_state(sc, INIT); 2219 carp_setrun(sc, 0); 2220 if (sc->sc_suppress) 2221 carp_suppress_preempt--; 2222 sc->sc_suppress = 0; 2223 } 2224 2225 return; 2226 } 2227 2228 static int 2229 carp_modevent(module_t mod, int type, void *data) 2230 { 2231 switch (type) { 2232 case MOD_LOAD: 2233 if_detach_event_tag = EVENTHANDLER_REGISTER(ifnet_departure_event, 2234 carp_ifdetach, NULL, EVENTHANDLER_PRI_ANY); 2235 if (if_detach_event_tag == NULL) 2236 return (ENOMEM); 2237 mtx_init(&carp_mtx, "carp_mtx", NULL, MTX_DEF); 2238 LIST_INIT(&carpif_list); 2239 if_clone_attach(&carp_cloner); 2240 break; 2241 2242 case MOD_UNLOAD: 2243 EVENTHANDLER_DEREGISTER(ifnet_departure_event, if_detach_event_tag); 2244 if_clone_detach(&carp_cloner); 2245 mtx_destroy(&carp_mtx); 2246 break; 2247 2248 default: 2249 return (EINVAL); 2250 } 2251 2252 return (0); 2253 } 2254 2255 static moduledata_t carp_mod = { 2256 "carp", 2257 carp_modevent, 2258 0 2259 }; 2260 2261 DECLARE_MODULE(carp, carp_mod, SI_SUB_PSEUDO, SI_ORDER_ANY); 2262