ip_carp.c revision 1.1
1 /* $NetBSD: ip_carp.c,v 1.1 2006/05/18 09:05:51 liamjfoy Exp $ */ 2 /* $OpenBSD: ip_carp.c,v 1.113 2005/11/04 08:11:54 mcbride Exp $ */ 3 4 /* 5 * Copyright (c) 2002 Michael Shalayeff. All rights reserved. 6 * Copyright (c) 2003 Ryan McBride. All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 20 * IN NO EVENT SHALL THE AUTHOR OR HIS RELATIVES BE LIABLE FOR ANY DIRECT, 21 * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES 22 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR 23 * SERVICES; LOSS OF MIND, USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, 25 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING 26 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF 27 * THE POSSIBILITY OF SUCH DAMAGE. 28 */ 29 30 /* 31 * TODO: 32 * - iface reconfigure 33 * - support for hardware checksum calculations; 34 * 35 */ 36 37 #include <sys/param.h> 38 #include <sys/proc.h> 39 #include <sys/mbuf.h> 40 #include <sys/socket.h> 41 #include <sys/socketvar.h> 42 #include <sys/callout.h> 43 #include <sys/ioctl.h> 44 #include <sys/errno.h> 45 #include <sys/device.h> 46 #include <sys/time.h> 47 #include <sys/kernel.h> 48 #include <sys/kauth.h> 49 #include <sys/sysctl.h> 50 #include <sys/ucred.h> 51 #include <sys/syslog.h> 52 #include <sys/acct.h> 53 54 #include <machine/cpu.h> 55 56 #include <net/if.h> 57 #include <net/pfil.h> 58 #include <net/if_types.h> 59 #include <net/if_ether.h> 60 #include <net/route.h> 61 #include <net/netisr.h> 62 #include <netinet/if_inarp.h> 63 64 #include <machine/stdarg.h> 65 66 #if NFDDI > 0 67 #include <net/if_fddi.h> 68 #endif 69 #if NTOKEN > 0 70 #include <net/if_token.h> 71 #endif 72 73 #ifdef INET 74 #include <netinet/in.h> 75 #include <netinet/in_systm.h> 76 #include <netinet/in_var.h> 77 #include <netinet/ip.h> 78 #include <netinet/ip_var.h> 79 80 #include <net/if_dl.h> 81 #endif 82 83 #ifdef INET6 84 #include <netinet/icmp6.h> 85 #include <netinet/ip6.h> 86 #include <netinet6/ip6_var.h> 87 #include <netinet6/nd6.h> 88 #endif 89 90 #include "bpfilter.h" 91 #if NBPFILTER > 0 92 #include <net/bpf.h> 93 #endif 94 95 #include <sys/sha1.h> 96 97 #include <netinet/ip_carp.h> 98 99 struct carp_mc_entry { 100 LIST_ENTRY(carp_mc_entry) mc_entries; 101 union { 102 struct ether_multi *mcu_enm; 103 } mc_u; 104 struct sockaddr_storage mc_addr; 105 }; 106 #define mc_enm mc_u.mcu_enm 107 108 struct carp_softc { 109 struct ethercom sc_ac; 110 #define sc_if sc_ac.ec_if 111 #define sc_carpdev sc_ac.ec_if.if_carpdev 112 int ah_cookie; 113 int lh_cookie; 114 struct ip_moptions sc_imo; 115 #ifdef INET6 116 struct ip6_moptions sc_im6o; 117 #endif /* INET6 */ 118 TAILQ_ENTRY(carp_softc) sc_list; 119 120 enum { INIT = 0, BACKUP, MASTER } sc_state; 121 122 int sc_suppress; 123 int sc_bow_out; 124 125 int sc_sendad_errors; 126 #define CARP_SENDAD_MAX_ERRORS 3 127 int sc_sendad_success; 128 #define CARP_SENDAD_MIN_SUCCESS 3 129 130 int sc_vhid; 131 int sc_advskew; 132 int sc_naddrs; 133 int sc_naddrs6; 134 int sc_advbase; /* seconds */ 135 int sc_init_counter; 136 u_int64_t sc_counter; 137 138 /* authentication */ 139 #define CARP_HMAC_PAD 64 140 unsigned char sc_key[CARP_KEY_LEN]; 141 unsigned char sc_pad[CARP_HMAC_PAD]; 142 SHA1_CTX sc_sha1; 143 u_int32_t sc_hashkey[2]; 144 145 struct callout sc_ad_tmo; /* advertisement timeout */ 146 struct callout sc_md_tmo; /* master down timeout */ 147 struct callout sc_md6_tmo; /* master down timeout */ 148 149 LIST_HEAD(__carp_mchead, carp_mc_entry) carp_mc_listhead; 150 }; 151 152 int carp_suppress_preempt = 0; 153 int carp_opts[CARPCTL_MAXID] = { 0, 1, 0, 0, 0 }; /* XXX for now */ 154 struct carpstats carpstats; 155 156 struct carp_if { 157 TAILQ_HEAD(, carp_softc) vhif_vrs; 158 int vhif_nvrs; 159 160 struct ifnet *vhif_ifp; 161 }; 162 163 #define CARP_LOG(sc, s) \ 164 if (carp_opts[CARPCTL_LOG]) { \ 165 if (sc) \ 166 log(LOG_INFO, "%s: ", \ 167 (sc)->sc_if.if_xname); \ 168 else \ 169 log(LOG_INFO, "carp: "); \ 170 addlog s; \ 171 addlog("\n"); \ 172 } 173 174 void carp_hmac_prepare(struct carp_softc *); 175 void carp_hmac_generate(struct carp_softc *, u_int32_t *, 176 unsigned char *); 177 int carp_hmac_verify(struct carp_softc *, u_int32_t *, 178 unsigned char *); 179 void carp_setroute(struct carp_softc *, int); 180 void carp_proto_input_c(struct mbuf *, struct carp_header *, sa_family_t); 181 void carpattach(int); 182 void carpdetach(struct carp_softc *); 183 int carp_prepare_ad(struct mbuf *, struct carp_softc *, 184 struct carp_header *); 185 void carp_send_ad_all(void); 186 void carp_send_ad(void *); 187 void carp_send_arp(struct carp_softc *); 188 void carp_master_down(void *); 189 int carp_ioctl(struct ifnet *, u_long, caddr_t); 190 void carp_start(struct ifnet *); 191 void carp_setrun(struct carp_softc *, sa_family_t); 192 void carp_set_state(struct carp_softc *, int); 193 int carp_addrcount(struct carp_if *, struct in_ifaddr *, int); 194 enum { CARP_COUNT_MASTER, CARP_COUNT_RUNNING }; 195 196 void carp_multicast_cleanup(struct carp_softc *); 197 int carp_set_ifp(struct carp_softc *, struct ifnet *); 198 void carp_set_enaddr(struct carp_softc *); 199 void carp_addr_updated(void *); 200 u_int32_t carp_hash(struct carp_softc *, u_char *); 201 int carp_set_addr(struct carp_softc *, struct sockaddr_in *); 202 int carp_join_multicast(struct carp_softc *); 203 #ifdef INET6 204 void carp_send_na(struct carp_softc *); 205 int carp_set_addr6(struct carp_softc *, struct sockaddr_in6 *); 206 int carp_join_multicast6(struct carp_softc *); 207 #endif 208 int carp_clone_create(struct if_clone *, int); 209 int carp_clone_destroy(struct ifnet *); 210 int carp_ether_addmulti(struct carp_softc *, struct ifreq *); 211 int carp_ether_delmulti(struct carp_softc *, struct ifreq *); 212 void carp_ether_purgemulti(struct carp_softc *); 213 214 struct if_clone carp_cloner = 215 IF_CLONE_INITIALIZER("carp", carp_clone_create, carp_clone_destroy); 216 217 static __inline u_int16_t 218 carp_cksum(struct mbuf *m, int len) 219 { 220 return (in_cksum(m, len)); 221 } 222 223 void 224 carp_hmac_prepare(struct carp_softc *sc) 225 { 226 u_int8_t carp_version = CARP_VERSION, type = CARP_ADVERTISEMENT; 227 u_int8_t vhid = sc->sc_vhid & 0xff; 228 SHA1_CTX sha1ctx; 229 u_int32_t kmd[5]; 230 struct ifaddr *ifa; 231 int i, found; 232 struct in_addr last, cur, in; 233 #ifdef INET6 234 struct in6_addr last6, cur6, in6; 235 #endif /* INET6 */ 236 237 /* compute ipad from key */ 238 bzero(sc->sc_pad, sizeof(sc->sc_pad)); 239 bcopy(sc->sc_key, sc->sc_pad, sizeof(sc->sc_key)); 240 for (i = 0; i < sizeof(sc->sc_pad); i++) 241 sc->sc_pad[i] ^= 0x36; 242 243 /* precompute first part of inner hash */ 244 SHA1Init(&sc->sc_sha1); 245 SHA1Update(&sc->sc_sha1, sc->sc_pad, sizeof(sc->sc_pad)); 246 SHA1Update(&sc->sc_sha1, (void *)&carp_version, sizeof(carp_version)); 247 SHA1Update(&sc->sc_sha1, (void *)&type, sizeof(type)); 248 249 /* generate a key for the arpbalance hash, before the vhid is hashed */ 250 bcopy(&sc->sc_sha1, &sha1ctx, sizeof(sha1ctx)); 251 SHA1Final((unsigned char *)kmd, &sha1ctx); 252 sc->sc_hashkey[0] = kmd[0] ^ kmd[1]; 253 sc->sc_hashkey[1] = kmd[2] ^ kmd[3]; 254 255 /* the rest of the precomputation */ 256 SHA1Update(&sc->sc_sha1, (void *)&vhid, sizeof(vhid)); 257 258 /* Hash the addresses from smallest to largest, not interface order */ 259 #ifdef INET 260 cur.s_addr = 0; 261 do { 262 found = 0; 263 last = cur; 264 cur.s_addr = 0xffffffff; 265 TAILQ_FOREACH(ifa, &sc->sc_if.if_addrlist, ifa_list) { 266 in.s_addr = ifatoia(ifa)->ia_addr.sin_addr.s_addr; 267 if (ifa->ifa_addr->sa_family == AF_INET && 268 ntohl(in.s_addr) > ntohl(last.s_addr) && 269 ntohl(in.s_addr) < ntohl(cur.s_addr)) { 270 cur.s_addr = in.s_addr; 271 found++; 272 } 273 } 274 if (found) 275 SHA1Update(&sc->sc_sha1, (void *)&cur, sizeof(cur)); 276 } while (found); 277 #endif /* INET */ 278 279 #ifdef INET6 280 memset(&cur6, 0x00, sizeof(cur6)); 281 do { 282 found = 0; 283 last6 = cur6; 284 memset(&cur6, 0xff, sizeof(cur6)); 285 TAILQ_FOREACH(ifa, &sc->sc_if.if_addrlist, ifa_list) { 286 in6 = ifatoia6(ifa)->ia_addr.sin6_addr; 287 if (IN6_IS_ADDR_LINKLOCAL(&in6)) 288 in6.s6_addr16[1] = 0; 289 if (ifa->ifa_addr->sa_family == AF_INET6 && 290 memcmp(&in6, &last6, sizeof(in6)) > 0 && 291 memcmp(&in6, &cur6, sizeof(in6)) < 0) { 292 cur6 = in6; 293 found++; 294 } 295 } 296 if (found) 297 SHA1Update(&sc->sc_sha1, (void *)&cur6, sizeof(cur6)); 298 } while (found); 299 #endif /* INET6 */ 300 301 /* convert ipad to opad */ 302 for (i = 0; i < sizeof(sc->sc_pad); i++) 303 sc->sc_pad[i] ^= 0x36 ^ 0x5c; 304 } 305 306 void 307 carp_hmac_generate(struct carp_softc *sc, u_int32_t counter[2], 308 unsigned char md[20]) 309 { 310 SHA1_CTX sha1ctx; 311 312 /* fetch first half of inner hash */ 313 bcopy(&sc->sc_sha1, &sha1ctx, sizeof(sha1ctx)); 314 315 SHA1Update(&sha1ctx, (void *)counter, sizeof(sc->sc_counter)); 316 SHA1Final(md, &sha1ctx); 317 318 /* outer hash */ 319 SHA1Init(&sha1ctx); 320 SHA1Update(&sha1ctx, sc->sc_pad, sizeof(sc->sc_pad)); 321 SHA1Update(&sha1ctx, md, 20); 322 SHA1Final(md, &sha1ctx); 323 } 324 325 int 326 carp_hmac_verify(struct carp_softc *sc, u_int32_t counter[2], 327 unsigned char md[20]) 328 { 329 unsigned char md2[20]; 330 331 carp_hmac_generate(sc, counter, md2); 332 333 return (bcmp(md, md2, sizeof(md2))); 334 } 335 336 void 337 carp_setroute(struct carp_softc *sc, int cmd) 338 { 339 struct ifaddr *ifa; 340 int s; 341 342 s = splsoftnet(); 343 TAILQ_FOREACH(ifa, &sc->sc_if.if_addrlist, ifa_list) { 344 switch (ifa->ifa_addr->sa_family) { 345 case AF_INET: { 346 int count = 0; 347 struct sockaddr sa; 348 struct rtentry *rt; 349 struct radix_node_head *rnh = 350 rt_tables[ifa->ifa_addr->sa_family]; 351 struct radix_node *rn; 352 int hr_otherif, nr_ourif; 353 354 /* 355 * Avoid screwing with the routes if there are other 356 * carp interfaces which are master and have the same 357 * address. 358 */ 359 if (sc->sc_carpdev != NULL && 360 sc->sc_carpdev->if_carp != NULL) { 361 count = carp_addrcount( 362 (struct carp_if *)sc->sc_carpdev->if_carp, 363 ifatoia(ifa), CARP_COUNT_MASTER); 364 if ((cmd == RTM_ADD && count != 1) || 365 (cmd == RTM_DELETE && count != 0)) 366 continue; 367 } 368 369 /* Remove the existing host route, if any */ 370 rtrequest(RTM_DELETE, ifa->ifa_addr, 371 ifa->ifa_addr, ifa->ifa_netmask, 372 RTF_HOST, NULL); 373 374 /* Check for our address on another interface */ 375 rn = rnh->rnh_matchaddr(ifa->ifa_addr, rnh); 376 rt = (struct rtentry *)rn; 377 hr_otherif = (rt && rt->rt_ifp != &sc->sc_if && 378 rt->rt_flags & (RTF_CLONING|RTF_CLONED)); 379 380 /* Check for a network route on our interface */ 381 bcopy(ifa->ifa_addr, &sa, sizeof(sa)); 382 satosin(&sa)->sin_addr.s_addr = satosin(ifa->ifa_netmask 383 )->sin_addr.s_addr & satosin(&sa)->sin_addr.s_addr; 384 rn = rnh->rnh_lookup(&sa, ifa->ifa_netmask, rnh); 385 rt = (struct rtentry *)rn; 386 nr_ourif = (rt && rt->rt_ifp == &sc->sc_if); 387 388 switch (cmd) { 389 case RTM_ADD: 390 if (hr_otherif) { 391 ifa->ifa_rtrequest = NULL; 392 ifa->ifa_flags &= ~RTF_CLONING; 393 394 rtrequest(RTM_ADD, ifa->ifa_addr, 395 ifa->ifa_addr, ifa->ifa_netmask, 396 RTF_UP | RTF_HOST, NULL); 397 } 398 if (!hr_otherif || nr_ourif || !rt) { 399 if (nr_ourif && !(rt->rt_flags & 400 RTF_CLONING)) 401 rtrequest(RTM_DELETE, &sa, 402 ifa->ifa_addr, 403 ifa->ifa_netmask, 0, NULL); 404 405 ifa->ifa_rtrequest = arp_rtrequest; 406 ifa->ifa_flags |= RTF_CLONING; 407 408 if (rtrequest(RTM_ADD, ifa->ifa_addr, 409 ifa->ifa_addr, ifa->ifa_netmask, 0, 410 NULL) == 0) 411 ifa->ifa_flags |= IFA_ROUTE; 412 } 413 break; 414 case RTM_DELETE: 415 break; 416 default: 417 break; 418 } 419 break; 420 } 421 422 #ifdef INET6 423 case AF_INET6: 424 if (cmd == RTM_ADD) 425 in6_ifaddloop(ifa); 426 else 427 in6_ifremloop(ifa); 428 break; 429 #endif /* INET6 */ 430 default: 431 break; 432 } 433 } 434 splx(s); 435 } 436 437 /* 438 * process input packet. 439 * we have rearranged checks order compared to the rfc, 440 * but it seems more efficient this way or not possible otherwise. 441 */ 442 void 443 carp_proto_input(struct mbuf *m, ...) 444 { 445 struct ip *ip = mtod(m, struct ip *); 446 struct carp_softc *sc = NULL; 447 struct carp_header *ch; 448 int iplen, len, hlen; 449 va_list ap; 450 451 va_start(ap, m); 452 hlen = va_arg(ap, int); 453 va_end(ap); 454 455 carpstats.carps_ipackets++; 456 457 if (!carp_opts[CARPCTL_ALLOW]) { 458 m_freem(m); 459 return; 460 } 461 462 /* check if received on a valid carp interface */ 463 if (m->m_pkthdr.rcvif->if_type != IFT_CARP) { 464 carpstats.carps_badif++; 465 CARP_LOG(sc, ("packet received on non-carp interface: %s", 466 m->m_pkthdr.rcvif->if_xname)); 467 m_freem(m); 468 return; 469 } 470 471 /* verify that the IP TTL is 255. */ 472 if (ip->ip_ttl != CARP_DFLTTL) { 473 carpstats.carps_badttl++; 474 CARP_LOG(sc, ("received ttl %d != %d on %s", ip->ip_ttl, 475 CARP_DFLTTL, m->m_pkthdr.rcvif->if_xname)); 476 m_freem(m); 477 return; 478 } 479 480 /* 481 * verify that the received packet length is 482 * equal to the CARP header 483 */ 484 iplen = ip->ip_hl << 2; 485 len = iplen + sizeof(*ch); 486 if (len > m->m_pkthdr.len) { 487 carpstats.carps_badlen++; 488 CARP_LOG(sc, ("packet too short %d on %s", m->m_pkthdr.len, 489 m->m_pkthdr.rcvif->if_xname)); 490 m_freem(m); 491 return; 492 } 493 494 if ((m = m_pullup(m, len)) == NULL) { 495 carpstats.carps_hdrops++; 496 return; 497 } 498 ip = mtod(m, struct ip *); 499 ch = (struct carp_header *)((char *)ip + iplen); 500 /* verify the CARP checksum */ 501 m->m_data += iplen; 502 if (carp_cksum(m, len - iplen)) { 503 carpstats.carps_badsum++; 504 CARP_LOG(sc, ("checksum failed on %s", 505 m->m_pkthdr.rcvif->if_xname)); 506 m_freem(m); 507 return; 508 } 509 m->m_data -= iplen; 510 511 carp_proto_input_c(m, ch, AF_INET); 512 } 513 514 #ifdef INET6 515 int 516 carp6_proto_input(struct mbuf **mp, int *offp, int proto) 517 { 518 struct mbuf *m = *mp; 519 struct carp_softc *sc = NULL; 520 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 521 struct carp_header *ch; 522 u_int len; 523 524 carpstats.carps_ipackets6++; 525 526 if (!carp_opts[CARPCTL_ALLOW]) { 527 m_freem(m); 528 return (IPPROTO_DONE); 529 } 530 531 /* check if received on a valid carp interface */ 532 if (m->m_pkthdr.rcvif->if_type != IFT_CARP) { 533 carpstats.carps_badif++; 534 CARP_LOG(sc, ("packet received on non-carp interface: %s", 535 m->m_pkthdr.rcvif->if_xname)); 536 m_freem(m); 537 return (IPPROTO_DONE); 538 } 539 540 /* verify that the IP TTL is 255 */ 541 if (ip6->ip6_hlim != CARP_DFLTTL) { 542 carpstats.carps_badttl++; 543 CARP_LOG(sc, ("received ttl %d != %d on %s", ip6->ip6_hlim, 544 CARP_DFLTTL, m->m_pkthdr.rcvif->if_xname)); 545 m_freem(m); 546 return (IPPROTO_DONE); 547 } 548 549 /* verify that we have a complete carp packet */ 550 len = m->m_len; 551 IP6_EXTHDR_GET(ch, struct carp_header *, m, *offp, sizeof(*ch)); 552 if (ch == NULL) { 553 carpstats.carps_badlen++; 554 CARP_LOG(sc, ("packet size %u too small", len)); 555 return (IPPROTO_DONE); 556 } 557 558 559 /* verify the CARP checksum */ 560 m->m_data += *offp; 561 if (carp_cksum(m, sizeof(*ch))) { 562 carpstats.carps_badsum++; 563 CARP_LOG(sc, ("checksum failed, on %s", 564 m->m_pkthdr.rcvif->if_xname)); 565 m_freem(m); 566 return (IPPROTO_DONE); 567 } 568 m->m_data -= *offp; 569 570 carp_proto_input_c(m, ch, AF_INET6); 571 return (IPPROTO_DONE); 572 } 573 #endif /* INET6 */ 574 575 void 576 carp_proto_input_c(struct mbuf *m, struct carp_header *ch, sa_family_t af) 577 { 578 struct carp_softc *sc; 579 u_int64_t tmp_counter; 580 struct timeval sc_tv, ch_tv; 581 582 TAILQ_FOREACH(sc, &((struct carp_if *) 583 m->m_pkthdr.rcvif->if_carpdev->if_carp)->vhif_vrs, sc_list) 584 if (sc->sc_vhid == ch->carp_vhid) 585 break; 586 587 if (!sc || (sc->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) != 588 (IFF_UP|IFF_RUNNING)) { 589 carpstats.carps_badvhid++; 590 m_freem(m); 591 return; 592 } 593 594 microtime(&sc->sc_if.if_lastchange); 595 sc->sc_if.if_ipackets++; 596 sc->sc_if.if_ibytes += m->m_pkthdr.len; 597 598 /* verify the CARP version. */ 599 if (ch->carp_version != CARP_VERSION) { 600 carpstats.carps_badver++; 601 sc->sc_if.if_ierrors++; 602 CARP_LOG(sc, ("invalid version %d != %d", 603 ch->carp_version, CARP_VERSION)); 604 m_freem(m); 605 return; 606 } 607 608 /* verify the hash */ 609 if (carp_hmac_verify(sc, ch->carp_counter, ch->carp_md)) { 610 carpstats.carps_badauth++; 611 sc->sc_if.if_ierrors++; 612 CARP_LOG(sc, ("incorrect hash")); 613 m_freem(m); 614 return; 615 } 616 617 tmp_counter = ntohl(ch->carp_counter[0]); 618 tmp_counter = tmp_counter<<32; 619 tmp_counter += ntohl(ch->carp_counter[1]); 620 621 /* XXX Replay protection goes here */ 622 623 sc->sc_init_counter = 0; 624 sc->sc_counter = tmp_counter; 625 626 627 sc_tv.tv_sec = sc->sc_advbase; 628 if (carp_suppress_preempt && sc->sc_advskew < 240) 629 sc_tv.tv_usec = 240 * 1000000 / 256; 630 else 631 sc_tv.tv_usec = sc->sc_advskew * 1000000 / 256; 632 ch_tv.tv_sec = ch->carp_advbase; 633 ch_tv.tv_usec = ch->carp_advskew * 1000000 / 256; 634 635 switch (sc->sc_state) { 636 case INIT: 637 break; 638 case MASTER: 639 /* 640 * If we receive an advertisement from a backup who's going to 641 * be more frequent than us, go into BACKUP state. 642 */ 643 if (timercmp(&sc_tv, &ch_tv, >) || 644 timercmp(&sc_tv, &ch_tv, ==)) { 645 callout_stop(&sc->sc_ad_tmo); 646 carp_set_state(sc, BACKUP); 647 carp_setrun(sc, 0); 648 carp_setroute(sc, RTM_DELETE); 649 } 650 break; 651 case BACKUP: 652 /* 653 * If we're pre-empting masters who advertise slower than us, 654 * and this one claims to be slower, treat him as down. 655 */ 656 if (carp_opts[CARPCTL_PREEMPT] && timercmp(&sc_tv, &ch_tv, <)) { 657 carp_master_down(sc); 658 break; 659 } 660 661 /* 662 * If the master is going to advertise at such a low frequency 663 * that he's guaranteed to time out, we'd might as well just 664 * treat him as timed out now. 665 */ 666 sc_tv.tv_sec = sc->sc_advbase * 3; 667 if (timercmp(&sc_tv, &ch_tv, <)) { 668 carp_master_down(sc); 669 break; 670 } 671 672 /* 673 * Otherwise, we reset the counter and wait for the next 674 * advertisement. 675 */ 676 carp_setrun(sc, af); 677 break; 678 } 679 680 m_freem(m); 681 return; 682 } 683 684 /* 685 * Interface side of the CARP implementation. 686 */ 687 688 /* ARGSUSED */ 689 void 690 carpattach(int n) 691 { 692 if_clone_attach(&carp_cloner); 693 } 694 695 int 696 carp_clone_create(struct if_clone *ifc, int unit) 697 { 698 extern int ifqmaxlen; 699 struct carp_softc *sc; 700 struct ifnet *ifp; 701 702 sc = malloc(sizeof(*sc), M_DEVBUF, M_NOWAIT); 703 if (!sc) 704 return (ENOMEM); 705 bzero(sc, sizeof(*sc)); 706 707 sc->sc_suppress = 0; 708 sc->sc_advbase = CARP_DFLTINTV; 709 sc->sc_vhid = -1; /* required setting */ 710 sc->sc_advskew = 0; 711 sc->sc_init_counter = 1; 712 sc->sc_naddrs = sc->sc_naddrs6 = 0; 713 #ifdef INET6 714 sc->sc_im6o.im6o_multicast_hlim = CARP_DFLTTL; 715 #endif /* INET6 */ 716 717 callout_init(&sc->sc_ad_tmo); 718 callout_init(&sc->sc_md_tmo); 719 callout_init(&sc->sc_md6_tmo); 720 721 callout_setfunc(&sc->sc_ad_tmo, carp_send_ad, sc); 722 callout_setfunc(&sc->sc_md_tmo, carp_master_down, sc); 723 callout_setfunc(&sc->sc_md6_tmo, carp_master_down, sc); 724 725 LIST_INIT(&sc->carp_mc_listhead); 726 ifp = &sc->sc_if; 727 ifp->if_softc = sc; 728 snprintf(ifp->if_xname, sizeof ifp->if_xname, "%s%d", ifc->ifc_name, 729 unit); 730 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST; 731 ifp->if_ioctl = carp_ioctl; 732 ifp->if_start = carp_start; 733 ifp->if_output = carp_output; 734 ifp->if_type = IFT_CARP; 735 ifp->if_addrlen = ETHER_ADDR_LEN; 736 ifp->if_hdrlen = ETHER_HDR_LEN; 737 ifp->if_mtu = ETHERMTU; 738 IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen); 739 IFQ_SET_READY(&ifp->if_snd); 740 if_attach(ifp); 741 742 if_alloc_sadl(ifp); 743 ifp->if_broadcastaddr = etherbroadcastaddr; 744 carp_set_enaddr(sc); 745 LIST_INIT(&sc->sc_ac.ec_multiaddrs); 746 #if NBPFILTER > 0 747 bpfattach(ifp, DLT_EN10MB, ETHER_HDR_LEN); 748 #endif 749 return (0); 750 } 751 752 int 753 carp_clone_destroy(struct ifnet *ifp) 754 { 755 carpdetach(ifp->if_softc); 756 ether_ifdetach(ifp); 757 if_detach(ifp); 758 free(ifp->if_softc, M_DEVBUF); 759 760 return (0); 761 } 762 763 void 764 carpdetach(struct carp_softc *sc) 765 { 766 struct carp_if *cif; 767 int s; 768 769 callout_stop(&sc->sc_ad_tmo); 770 callout_stop(&sc->sc_md_tmo); 771 callout_stop(&sc->sc_md6_tmo); 772 773 if (sc->sc_suppress) 774 carp_suppress_preempt--; 775 sc->sc_suppress = 0; 776 777 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) 778 carp_suppress_preempt--; 779 sc->sc_sendad_errors = 0; 780 781 carp_set_state(sc, INIT); 782 sc->sc_if.if_flags &= ~IFF_UP; 783 carp_setrun(sc, 0); 784 carp_multicast_cleanup(sc); 785 786 s = splnet(); 787 if (sc->sc_carpdev != NULL) { 788 /* XXX linkstatehook removal */ 789 cif = (struct carp_if *)sc->sc_carpdev->if_carp; 790 TAILQ_REMOVE(&cif->vhif_vrs, sc, sc_list); 791 if (!--cif->vhif_nvrs) { 792 ifpromisc(sc->sc_carpdev, 0); 793 sc->sc_carpdev->if_carp = NULL; 794 FREE(cif, M_IFADDR); 795 } 796 } 797 sc->sc_carpdev = NULL; 798 splx(s); 799 } 800 801 /* Detach an interface from the carp. */ 802 void 803 carp_ifdetach(struct ifnet *ifp) 804 { 805 struct carp_softc *sc, *nextsc; 806 struct carp_if *cif = (struct carp_if *)ifp->if_carp; 807 808 for (sc = TAILQ_FIRST(&cif->vhif_vrs); sc; sc = nextsc) { 809 nextsc = TAILQ_NEXT(sc, sc_list); 810 carpdetach(sc); 811 } 812 } 813 814 int 815 carp_prepare_ad(struct mbuf *m, struct carp_softc *sc, struct carp_header *ch) 816 { 817 if (sc->sc_init_counter) { 818 /* this could also be seconds since unix epoch */ 819 sc->sc_counter = arc4random(); 820 sc->sc_counter = sc->sc_counter << 32; 821 sc->sc_counter += arc4random(); 822 } else 823 sc->sc_counter++; 824 825 ch->carp_counter[0] = htonl((sc->sc_counter>>32)&0xffffffff); 826 ch->carp_counter[1] = htonl(sc->sc_counter&0xffffffff); 827 828 carp_hmac_generate(sc, ch->carp_counter, ch->carp_md); 829 830 return (0); 831 } 832 833 void 834 carp_send_ad_all(void) 835 { 836 struct ifnet *ifp; 837 struct carp_if *cif; 838 struct carp_softc *vh; 839 840 TAILQ_FOREACH(ifp, &ifnet, if_list) { 841 if (ifp->if_carp == NULL || ifp->if_type == IFT_CARP) 842 continue; 843 844 cif = (struct carp_if *)ifp->if_carp; 845 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 846 if ((vh->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) == 847 (IFF_UP|IFF_RUNNING) && vh->sc_state == MASTER) 848 carp_send_ad(vh); 849 } 850 } 851 } 852 853 854 void 855 carp_send_ad(void *v) 856 { 857 struct carp_header ch; 858 struct timeval tv; 859 struct carp_softc *sc = v; 860 struct carp_header *ch_ptr; 861 struct mbuf *m; 862 int error, len, advbase, advskew, s; 863 struct ifaddr *ifa; 864 struct sockaddr sa; 865 866 s = splsoftnet(); 867 868 advbase = advskew = 0; /* Sssssh compiler */ 869 if (sc->sc_carpdev == NULL) { 870 sc->sc_if.if_oerrors++; 871 goto retry_later; 872 } 873 874 /* bow out if we've gone to backup (the carp interface is going down) */ 875 if (sc->sc_bow_out) { 876 sc->sc_bow_out = 0; 877 advbase = 255; 878 advskew = 255; 879 } else { 880 advbase = sc->sc_advbase; 881 if (!carp_suppress_preempt || sc->sc_advskew > 240) 882 advskew = sc->sc_advskew; 883 else 884 advskew = 240; 885 tv.tv_sec = advbase; 886 tv.tv_usec = advskew * 1000000 / 256; 887 } 888 889 ch.carp_version = CARP_VERSION; 890 ch.carp_type = CARP_ADVERTISEMENT; 891 ch.carp_vhid = sc->sc_vhid; 892 ch.carp_advbase = advbase; 893 ch.carp_advskew = advskew; 894 ch.carp_authlen = 7; /* XXX DEFINE */ 895 ch.carp_pad1 = 0; /* must be zero */ 896 ch.carp_cksum = 0; 897 898 899 #ifdef INET 900 if (sc->sc_naddrs) { 901 struct ip *ip; 902 903 MGETHDR(m, M_DONTWAIT, MT_HEADER); 904 if (m == NULL) { 905 sc->sc_if.if_oerrors++; 906 carpstats.carps_onomem++; 907 /* XXX maybe less ? */ 908 goto retry_later; 909 } 910 len = sizeof(*ip) + sizeof(ch); 911 m->m_pkthdr.len = len; 912 m->m_pkthdr.rcvif = NULL; 913 m->m_len = len; 914 MH_ALIGN(m, m->m_len); 915 m->m_flags |= M_MCAST; 916 ip = mtod(m, struct ip *); 917 ip->ip_v = IPVERSION; 918 ip->ip_hl = sizeof(*ip) >> 2; 919 ip->ip_tos = IPTOS_LOWDELAY; 920 ip->ip_len = htons(len); 921 ip->ip_id = htons(ip_randomid()); 922 ip->ip_off = htons(IP_DF); 923 ip->ip_ttl = CARP_DFLTTL; 924 ip->ip_p = IPPROTO_CARP; 925 ip->ip_sum = 0; 926 927 bzero(&sa, sizeof(sa)); 928 sa.sa_family = AF_INET; 929 ifa = ifaof_ifpforaddr(&sa, sc->sc_carpdev); 930 if (ifa == NULL) 931 ip->ip_src.s_addr = 0; 932 else 933 ip->ip_src.s_addr = 934 ifatoia(ifa)->ia_addr.sin_addr.s_addr; 935 ip->ip_dst.s_addr = INADDR_CARP_GROUP; 936 937 ch_ptr = (struct carp_header *)(&ip[1]); 938 bcopy(&ch, ch_ptr, sizeof(ch)); 939 if (carp_prepare_ad(m, sc, ch_ptr)) 940 goto retry_later; 941 942 m->m_data += sizeof(*ip); 943 ch_ptr->carp_cksum = carp_cksum(m, len - sizeof(*ip)); 944 m->m_data -= sizeof(*ip); 945 946 microtime(&sc->sc_if.if_lastchange); 947 sc->sc_if.if_opackets++; 948 sc->sc_if.if_obytes += len; 949 carpstats.carps_opackets++; 950 951 error = ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo, 952 NULL); 953 if (error) { 954 if (error == ENOBUFS) 955 carpstats.carps_onomem++; 956 else 957 CARP_LOG(sc, ("ip_output failed: %d", error)); 958 sc->sc_if.if_oerrors++; 959 if (sc->sc_sendad_errors < INT_MAX) 960 sc->sc_sendad_errors++; 961 if (sc->sc_sendad_errors == CARP_SENDAD_MAX_ERRORS) { 962 carp_suppress_preempt++; 963 if (carp_suppress_preempt == 1) 964 carp_send_ad_all(); 965 } 966 sc->sc_sendad_success = 0; 967 } else { 968 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) { 969 if (++sc->sc_sendad_success >= 970 CARP_SENDAD_MIN_SUCCESS) { 971 carp_suppress_preempt--; 972 sc->sc_sendad_errors = 0; 973 } 974 } else 975 sc->sc_sendad_errors = 0; 976 } 977 } 978 #endif /* INET */ 979 #ifdef INET6 980 if (sc->sc_naddrs6) { 981 struct ip6_hdr *ip6; 982 983 MGETHDR(m, M_DONTWAIT, MT_HEADER); 984 if (m == NULL) { 985 sc->sc_if.if_oerrors++; 986 carpstats.carps_onomem++; 987 /* XXX maybe less ? */ 988 goto retry_later; 989 } 990 len = sizeof(*ip6) + sizeof(ch); 991 m->m_pkthdr.len = len; 992 m->m_pkthdr.rcvif = NULL; 993 m->m_len = len; 994 MH_ALIGN(m, m->m_len); 995 m->m_flags |= M_MCAST; 996 ip6 = mtod(m, struct ip6_hdr *); 997 bzero(ip6, sizeof(*ip6)); 998 ip6->ip6_vfc |= IPV6_VERSION; 999 ip6->ip6_hlim = CARP_DFLTTL; 1000 ip6->ip6_nxt = IPPROTO_CARP; 1001 1002 /* set the source address */ 1003 bzero(&sa, sizeof(sa)); 1004 sa.sa_family = AF_INET6; 1005 ifa = ifaof_ifpforaddr(&sa, sc->sc_carpdev); 1006 if (ifa == NULL) /* This should never happen with IPv6 */ 1007 bzero(&ip6->ip6_src, sizeof(struct in6_addr)); 1008 else 1009 bcopy(ifatoia6(ifa)->ia_addr.sin6_addr.s6_addr, 1010 &ip6->ip6_src, sizeof(struct in6_addr)); 1011 /* set the multicast destination */ 1012 1013 ip6->ip6_dst.s6_addr8[0] = 0xff; 1014 ip6->ip6_dst.s6_addr8[1] = 0x02; 1015 ip6->ip6_dst.s6_addr8[15] = 0x12; 1016 1017 ch_ptr = (struct carp_header *)(&ip6[1]); 1018 bcopy(&ch, ch_ptr, sizeof(ch)); 1019 if (carp_prepare_ad(m, sc, ch_ptr)) 1020 goto retry_later; 1021 1022 m->m_data += sizeof(*ip6); 1023 ch_ptr->carp_cksum = carp_cksum(m, len - sizeof(*ip6)); 1024 m->m_data -= sizeof(*ip6); 1025 1026 microtime(&sc->sc_if.if_lastchange); 1027 sc->sc_if.if_opackets++; 1028 sc->sc_if.if_obytes += len; 1029 carpstats.carps_opackets6++; 1030 1031 error = ip6_output(m, NULL, NULL, 0, &sc->sc_im6o, NULL, NULL); 1032 if (error) { 1033 if (error == ENOBUFS) 1034 carpstats.carps_onomem++; 1035 else 1036 CARP_LOG(sc, ("ip6_output failed: %d", error)); 1037 sc->sc_if.if_oerrors++; 1038 if (sc->sc_sendad_errors < INT_MAX) 1039 sc->sc_sendad_errors++; 1040 if (sc->sc_sendad_errors == CARP_SENDAD_MAX_ERRORS) { 1041 carp_suppress_preempt++; 1042 if (carp_suppress_preempt == 1) 1043 carp_send_ad_all(); 1044 } 1045 sc->sc_sendad_success = 0; 1046 } else { 1047 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) { 1048 if (++sc->sc_sendad_success >= 1049 CARP_SENDAD_MIN_SUCCESS) { 1050 carp_suppress_preempt--; 1051 sc->sc_sendad_errors = 0; 1052 } 1053 } else 1054 sc->sc_sendad_errors = 0; 1055 } 1056 } 1057 #endif /* INET6 */ 1058 1059 retry_later: 1060 splx(s); 1061 if (advbase != 255 || advskew != 255) 1062 callout_schedule(&sc->sc_ad_tmo, tvtohz(&tv)); 1063 } 1064 1065 /* 1066 * Broadcast a gratuitous ARP request containing 1067 * the virtual router MAC address for each IP address 1068 * associated with the virtual router. 1069 */ 1070 void 1071 carp_send_arp(struct carp_softc *sc) 1072 { 1073 struct ifaddr *ifa; 1074 struct in_addr *in; 1075 int s = splsoftnet(); 1076 1077 TAILQ_FOREACH(ifa, &sc->sc_if.if_addrlist, ifa_list) { 1078 1079 if (ifa->ifa_addr->sa_family != AF_INET) 1080 continue; 1081 1082 in = &ifatoia(ifa)->ia_addr.sin_addr; 1083 arprequest(sc->sc_carpdev, in, in, LLADDR(sc->sc_if.if_sadl)); 1084 DELAY(1000); /* XXX */ 1085 } 1086 splx(s); 1087 } 1088 1089 #ifdef INET6 1090 void 1091 carp_send_na(struct carp_softc *sc) 1092 { 1093 struct ifaddr *ifa; 1094 struct in6_addr *in6; 1095 static struct in6_addr mcast = IN6ADDR_LINKLOCAL_ALLNODES_INIT; 1096 int s = splsoftnet(); 1097 1098 TAILQ_FOREACH(ifa, &sc->sc_if.if_addrlist, ifa_list) { 1099 1100 if (ifa->ifa_addr->sa_family != AF_INET6) 1101 continue; 1102 1103 in6 = &ifatoia6(ifa)->ia_addr.sin6_addr; 1104 nd6_na_output(sc->sc_carpdev, &mcast, in6, 1105 ND_NA_FLAG_OVERRIDE, 1, NULL); 1106 DELAY(1000); /* XXX */ 1107 } 1108 splx(s); 1109 } 1110 #endif /* INET6 */ 1111 1112 /* 1113 * Based on bridge_hash() in if_bridge.c 1114 */ 1115 #define mix(a,b,c) \ 1116 do { \ 1117 a -= b; a -= c; a ^= (c >> 13); \ 1118 b -= c; b -= a; b ^= (a << 8); \ 1119 c -= a; c -= b; c ^= (b >> 13); \ 1120 a -= b; a -= c; a ^= (c >> 12); \ 1121 b -= c; b -= a; b ^= (a << 16); \ 1122 c -= a; c -= b; c ^= (b >> 5); \ 1123 a -= b; a -= c; a ^= (c >> 3); \ 1124 b -= c; b -= a; b ^= (a << 10); \ 1125 c -= a; c -= b; c ^= (b >> 15); \ 1126 } while (0) 1127 1128 u_int32_t 1129 carp_hash(struct carp_softc *sc, u_char *src) 1130 { 1131 u_int32_t a = 0x9e3779b9, b = sc->sc_hashkey[0], c = sc->sc_hashkey[1]; 1132 1133 c += sc->sc_key[3] << 24; 1134 c += sc->sc_key[2] << 16; 1135 c += sc->sc_key[1] << 8; 1136 c += sc->sc_key[0]; 1137 b += src[5] << 8; 1138 b += src[4]; 1139 a += src[3] << 24; 1140 a += src[2] << 16; 1141 a += src[1] << 8; 1142 a += src[0]; 1143 1144 mix(a, b, c); 1145 return (c); 1146 } 1147 1148 int 1149 carp_addrcount(struct carp_if *cif, struct in_ifaddr *ia, int type) 1150 { 1151 struct carp_softc *vh; 1152 struct ifaddr *ifa; 1153 int count = 0; 1154 1155 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 1156 if ((type == CARP_COUNT_RUNNING && 1157 (vh->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) == 1158 (IFF_UP|IFF_RUNNING)) || 1159 (type == CARP_COUNT_MASTER && vh->sc_state == MASTER)) { 1160 TAILQ_FOREACH(ifa, &vh->sc_if.if_addrlist, ifa_list) { 1161 if (ifa->ifa_addr->sa_family == AF_INET && 1162 ia->ia_addr.sin_addr.s_addr == 1163 ifatoia(ifa)->ia_addr.sin_addr.s_addr) 1164 count++; 1165 } 1166 } 1167 } 1168 return (count); 1169 } 1170 1171 int 1172 carp_iamatch(struct in_ifaddr *ia, u_char *src, 1173 u_int32_t *count, u_int32_t index) 1174 { 1175 struct carp_softc *sc = ia->ia_ifp->if_softc; 1176 1177 if (carp_opts[CARPCTL_ARPBALANCE]) { 1178 /* 1179 * We use the source ip to decide which virtual host should 1180 * handle the request. If we're master of that virtual host, 1181 * then we respond, otherwise, just drop the arp packet on 1182 * the floor. 1183 */ 1184 1185 /* Count the elegible carp interfaces with this address */ 1186 if (*count == 0) 1187 *count = carp_addrcount( 1188 (struct carp_if *)ia->ia_ifp->if_carpdev->if_carp, 1189 ia, CARP_COUNT_RUNNING); 1190 1191 /* This should never happen, but... */ 1192 if (*count == 0) 1193 return (0); 1194 1195 if (carp_hash(sc, src) % *count == index - 1 && 1196 sc->sc_state == MASTER) { 1197 return (1); 1198 } 1199 } else { 1200 if (sc->sc_state == MASTER) 1201 return (1); 1202 } 1203 1204 return (0); 1205 } 1206 1207 #ifdef INET6 1208 struct ifaddr * 1209 carp_iamatch6(void *v, struct in6_addr *taddr) 1210 { 1211 struct carp_if *cif = v; 1212 struct carp_softc *vh; 1213 struct ifaddr *ifa; 1214 1215 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 1216 TAILQ_FOREACH(ifa, &vh->sc_if.if_addrlist, ifa_list) { 1217 if (IN6_ARE_ADDR_EQUAL(taddr, 1218 &ifatoia6(ifa)->ia_addr.sin6_addr) && 1219 ((vh->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) == 1220 (IFF_UP|IFF_RUNNING)) && vh->sc_state == MASTER) 1221 return (ifa); 1222 } 1223 } 1224 1225 return (NULL); 1226 } 1227 #endif /* INET6 */ 1228 1229 struct ifnet * 1230 carp_ourether(void *v, struct ether_header *eh, u_char iftype, int src) 1231 { 1232 struct carp_if *cif = (struct carp_if *)v; 1233 struct carp_softc *vh; 1234 u_int8_t *ena; 1235 1236 if (src) 1237 ena = (u_int8_t *)&eh->ether_shost; 1238 else 1239 ena = (u_int8_t *)&eh->ether_dhost; 1240 1241 switch (iftype) { 1242 case IFT_ETHER: 1243 case IFT_FDDI: 1244 if (ena[0] || ena[1] || ena[2] != 0x5e || ena[3] || ena[4] != 1) 1245 return (NULL); 1246 break; 1247 case IFT_ISO88025: 1248 if (ena[0] != 3 || ena[1] || ena[4] || ena[5]) 1249 return (NULL); 1250 break; 1251 default: 1252 return (NULL); 1253 break; 1254 } 1255 1256 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) 1257 if ((vh->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) == 1258 (IFF_UP|IFF_RUNNING) && vh->sc_state == MASTER && 1259 !bcmp(ena, LLADDR(vh->sc_if.if_sadl), 1260 ETHER_ADDR_LEN)) { 1261 return (&vh->sc_if); 1262 } 1263 1264 return (NULL); 1265 } 1266 1267 int 1268 carp_input(struct mbuf *m, u_int8_t *shost, u_int8_t *dhost, u_int16_t etype) 1269 { 1270 struct ether_header eh; 1271 struct carp_if *cif = (struct carp_if *)m->m_pkthdr.rcvif->if_carp; 1272 struct ifnet *ifp; 1273 1274 bcopy(shost, &eh.ether_shost, sizeof(eh.ether_shost)); 1275 bcopy(dhost, &eh.ether_dhost, sizeof(eh.ether_dhost)); 1276 eh.ether_type = etype; 1277 1278 if (m->m_flags & (M_BCAST|M_MCAST)) { 1279 struct carp_softc *vh; 1280 struct mbuf *m0; 1281 1282 /* 1283 * XXX Should really check the list of multicast addresses 1284 * for each CARP interface _before_ copying. 1285 */ 1286 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 1287 m0 = m_copym(m, 0, M_COPYALL, M_DONTWAIT); 1288 if (m0 == NULL) 1289 continue; 1290 m0->m_pkthdr.rcvif = &vh->sc_if; 1291 ether_input(&vh->sc_if, m0); 1292 } 1293 return (1); 1294 } 1295 1296 ifp = carp_ourether(cif, &eh, m->m_pkthdr.rcvif->if_type, 0); 1297 if (ifp == NULL) { 1298 return (1); 1299 } 1300 1301 m->m_pkthdr.rcvif = ifp; 1302 1303 #if NBPFILTER > 0 1304 if (ifp->if_bpf) 1305 bpf_mtap(ifp->if_bpf, m); 1306 #endif 1307 ifp->if_ipackets++; 1308 ether_input(ifp, m); 1309 return (0); 1310 } 1311 1312 void 1313 carp_master_down(void *v) 1314 { 1315 struct carp_softc *sc = v; 1316 1317 switch (sc->sc_state) { 1318 case INIT: 1319 printf("%s: master_down event in INIT state\n", 1320 sc->sc_if.if_xname); 1321 break; 1322 case MASTER: 1323 break; 1324 case BACKUP: 1325 carp_set_state(sc, MASTER); 1326 carp_send_ad(sc); 1327 carp_send_arp(sc); 1328 #ifdef INET6 1329 carp_send_na(sc); 1330 #endif /* INET6 */ 1331 carp_setrun(sc, 0); 1332 carp_setroute(sc, RTM_ADD); 1333 break; 1334 } 1335 } 1336 1337 /* 1338 * When in backup state, af indicates whether to reset the master down timer 1339 * for v4 or v6. If it's set to zero, reset the ones which are already pending. 1340 */ 1341 void 1342 carp_setrun(struct carp_softc *sc, sa_family_t af) 1343 { 1344 struct timeval tv; 1345 1346 if (sc->sc_carpdev == NULL) { 1347 sc->sc_if.if_flags &= ~IFF_RUNNING; 1348 carp_set_state(sc, INIT); 1349 return; 1350 } 1351 1352 if (sc->sc_if.if_flags & IFF_UP && sc->sc_vhid > 0 && 1353 (sc->sc_naddrs || sc->sc_naddrs6) && !sc->sc_suppress) { 1354 sc->sc_if.if_flags |= IFF_RUNNING; 1355 } else { 1356 sc->sc_if.if_flags &= ~IFF_RUNNING; 1357 carp_setroute(sc, RTM_DELETE); 1358 return; 1359 } 1360 1361 switch (sc->sc_state) { 1362 case INIT: 1363 carp_set_state(sc, BACKUP); 1364 carp_setroute(sc, RTM_DELETE); 1365 carp_setrun(sc, 0); 1366 break; 1367 case BACKUP: 1368 callout_stop(&sc->sc_ad_tmo); 1369 tv.tv_sec = 3 * sc->sc_advbase; 1370 tv.tv_usec = sc->sc_advskew * 1000000 / 256; 1371 switch (af) { 1372 #ifdef INET 1373 case AF_INET: 1374 callout_schedule(&sc->sc_md_tmo, tvtohz(&tv)); 1375 break; 1376 #endif /* INET */ 1377 #ifdef INET6 1378 case AF_INET6: 1379 callout_schedule(&sc->sc_md6_tmo, tvtohz(&tv)); 1380 break; 1381 #endif /* INET6 */ 1382 default: 1383 if (sc->sc_naddrs) 1384 callout_schedule(&sc->sc_md_tmo, tvtohz(&tv)); 1385 if (sc->sc_naddrs6) 1386 callout_schedule(&sc->sc_md6_tmo, tvtohz(&tv)); 1387 break; 1388 } 1389 break; 1390 case MASTER: 1391 tv.tv_sec = sc->sc_advbase; 1392 tv.tv_usec = sc->sc_advskew * 1000000 / 256; 1393 callout_schedule(&sc->sc_ad_tmo, tvtohz(&tv)); 1394 break; 1395 } 1396 } 1397 1398 void 1399 carp_multicast_cleanup(struct carp_softc *sc) 1400 { 1401 struct ip_moptions *imo = &sc->sc_imo; 1402 #ifdef INET6 1403 struct ip6_moptions *im6o = &sc->sc_im6o; 1404 #endif 1405 u_int16_t n = imo->imo_num_memberships; 1406 1407 /* Clean up our own multicast memberships */ 1408 while (n-- > 0) { 1409 if (imo->imo_membership[n] != NULL) { 1410 in_delmulti(imo->imo_membership[n]); 1411 imo->imo_membership[n] = NULL; 1412 } 1413 } 1414 imo->imo_num_memberships = 0; 1415 imo->imo_multicast_ifp = NULL; 1416 1417 #ifdef INET6 1418 while (!LIST_EMPTY(&im6o->im6o_memberships)) { 1419 struct in6_multi_mship *imm = 1420 LIST_FIRST(&im6o->im6o_memberships); 1421 1422 LIST_REMOVE(imm, i6mm_chain); 1423 in6_leavegroup(imm); 1424 } 1425 im6o->im6o_multicast_ifp = NULL; 1426 #endif 1427 1428 /* And any other multicast memberships */ 1429 carp_ether_purgemulti(sc); 1430 } 1431 1432 int 1433 carp_set_ifp(struct carp_softc *sc, struct ifnet *ifp) 1434 { 1435 struct carp_if *cif, *ncif = NULL; 1436 struct carp_softc *vr, *after = NULL; 1437 int myself = 0, error = 0; 1438 int s; 1439 1440 if (ifp == sc->sc_carpdev) 1441 return (0); 1442 1443 if (ifp != NULL) { 1444 if ((ifp->if_flags & IFF_MULTICAST) == 0) 1445 return (EADDRNOTAVAIL); 1446 1447 if (ifp->if_type == IFT_CARP) 1448 return (EINVAL); 1449 1450 if (ifp->if_carp == NULL) { 1451 MALLOC(ncif, struct carp_if *, sizeof(*cif), 1452 M_IFADDR, M_NOWAIT); 1453 if (ncif == NULL) 1454 return (ENOBUFS); 1455 if ((error = ifpromisc(ifp, 1))) { 1456 FREE(ncif, M_IFADDR); 1457 return (error); 1458 } 1459 1460 ncif->vhif_ifp = ifp; 1461 TAILQ_INIT(&ncif->vhif_vrs); 1462 } else { 1463 cif = (struct carp_if *)ifp->if_carp; 1464 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) 1465 if (vr != sc && vr->sc_vhid == sc->sc_vhid) 1466 return (EINVAL); 1467 } 1468 1469 /* detach from old interface */ 1470 if (sc->sc_carpdev != NULL) 1471 carpdetach(sc); 1472 1473 /* join multicast groups */ 1474 if (sc->sc_naddrs < 0 && 1475 (error = carp_join_multicast(sc)) != 0) { 1476 if (ncif != NULL) 1477 FREE(ncif, M_IFADDR); 1478 return (error); 1479 } 1480 1481 #ifdef INET6 1482 if (sc->sc_naddrs6 < 0 && 1483 (error = carp_join_multicast6(sc)) != 0) { 1484 if (ncif != NULL) 1485 FREE(ncif, M_IFADDR); 1486 carp_multicast_cleanup(sc); 1487 return (error); 1488 } 1489 #endif 1490 1491 /* attach carp interface to physical interface */ 1492 if (ncif != NULL) 1493 ifp->if_carp = (caddr_t)ncif; 1494 sc->sc_carpdev = ifp; 1495 cif = (struct carp_if *)ifp->if_carp; 1496 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) { 1497 if (vr == sc) 1498 myself = 1; 1499 if (vr->sc_vhid < sc->sc_vhid) 1500 after = vr; 1501 } 1502 1503 if (!myself) { 1504 /* We're trying to keep things in order */ 1505 if (after == NULL) { 1506 TAILQ_INSERT_TAIL(&cif->vhif_vrs, sc, sc_list); 1507 } else { 1508 TAILQ_INSERT_AFTER(&cif->vhif_vrs, after, 1509 sc, sc_list); 1510 } 1511 cif->vhif_nvrs++; 1512 } 1513 if (sc->sc_naddrs || sc->sc_naddrs6) 1514 sc->sc_if.if_flags |= IFF_UP; 1515 carp_set_enaddr(sc); 1516 s = splnet(); 1517 /* XXX linkstatehooks establish */ 1518 carp_carpdev_state(ifp); 1519 splx(s); 1520 } else { 1521 carpdetach(sc); 1522 sc->sc_if.if_flags &= ~(IFF_UP|IFF_RUNNING); 1523 } 1524 return (0); 1525 } 1526 1527 void 1528 carp_set_enaddr(struct carp_softc *sc) 1529 { 1530 if (sc->sc_carpdev && sc->sc_carpdev->if_type == IFT_ISO88025) { 1531 LLADDR(sc->sc_if.if_sadl)[0] = 3; 1532 LLADDR(sc->sc_if.if_sadl)[1] = 0; 1533 LLADDR(sc->sc_if.if_sadl)[2] = 0x40 >> (sc->sc_vhid - 1); 1534 LLADDR(sc->sc_if.if_sadl)[3] = 0x40000 >> (sc->sc_vhid - 1); 1535 LLADDR(sc->sc_if.if_sadl)[4] = 0; 1536 LLADDR(sc->sc_if.if_sadl)[5] = 0; 1537 } else { 1538 LLADDR(sc->sc_if.if_sadl)[0] = 0; 1539 LLADDR(sc->sc_if.if_sadl)[1] = 0; 1540 LLADDR(sc->sc_if.if_sadl)[2] = 0x5e; 1541 LLADDR(sc->sc_if.if_sadl)[3] = 0; 1542 LLADDR(sc->sc_if.if_sadl)[4] = 1; 1543 LLADDR(sc->sc_if.if_sadl)[5] = sc->sc_vhid; 1544 } 1545 } 1546 1547 void 1548 carp_addr_updated(void *v) 1549 { 1550 struct carp_softc *sc = (struct carp_softc *) v; 1551 struct ifaddr *ifa; 1552 int new_naddrs = 0, new_naddrs6 = 0; 1553 1554 TAILQ_FOREACH(ifa, &sc->sc_if.if_addrlist, ifa_list) { 1555 if (ifa->ifa_addr->sa_family == AF_INET) 1556 new_naddrs++; 1557 else if (ifa->ifa_addr->sa_family == AF_INET6) 1558 new_naddrs6++; 1559 } 1560 1561 /* Handle a callback after SIOCDIFADDR */ 1562 if (new_naddrs < sc->sc_naddrs || new_naddrs6 < sc->sc_naddrs6) { 1563 struct in_addr mc_addr; 1564 struct in_multi *inm; 1565 1566 sc->sc_naddrs = new_naddrs; 1567 sc->sc_naddrs6 = new_naddrs6; 1568 1569 /* Re-establish multicast membership removed by in_control */ 1570 mc_addr.s_addr = INADDR_CARP_GROUP; 1571 IN_LOOKUP_MULTI(mc_addr, &sc->sc_if, inm); 1572 if (inm == NULL) { 1573 bzero(&sc->sc_imo, sizeof(sc->sc_imo)); 1574 1575 if (sc->sc_carpdev != NULL && sc->sc_naddrs > 0) 1576 carp_join_multicast(sc); 1577 } 1578 1579 if (sc->sc_naddrs == 0 && sc->sc_naddrs6 == 0) { 1580 sc->sc_if.if_flags &= ~IFF_UP; 1581 carp_set_state(sc, INIT); 1582 } else 1583 carp_hmac_prepare(sc); 1584 } 1585 1586 carp_setrun(sc, 0); 1587 } 1588 1589 int 1590 carp_set_addr(struct carp_softc *sc, struct sockaddr_in *sin) 1591 { 1592 struct ifnet *ifp = sc->sc_carpdev; 1593 struct in_ifaddr *ia, *ia_if; 1594 int error = 0; 1595 1596 if (sin->sin_addr.s_addr == 0) { 1597 if (!(sc->sc_if.if_flags & IFF_UP)) 1598 carp_set_state(sc, INIT); 1599 if (sc->sc_naddrs) 1600 sc->sc_if.if_flags |= IFF_UP; 1601 carp_setrun(sc, 0); 1602 return (0); 1603 } 1604 1605 /* we have to do this by hand to ensure we don't match on ourselves */ 1606 ia_if = NULL; 1607 for (ia = TAILQ_FIRST(&in_ifaddrhead); ia; 1608 ia = TAILQ_NEXT(ia, ia_list)) { 1609 1610 /* and, yeah, we need a multicast-capable iface too */ 1611 if (ia->ia_ifp != &sc->sc_if && 1612 ia->ia_ifp->if_type != IFT_CARP && 1613 (ia->ia_ifp->if_flags & IFF_MULTICAST) && 1614 (sin->sin_addr.s_addr & ia->ia_subnetmask) == 1615 ia->ia_subnet) { 1616 if (!ia_if) 1617 ia_if = ia; 1618 } 1619 } 1620 1621 if (ia_if) { 1622 ia = ia_if; 1623 if (ifp) { 1624 if (ifp != ia->ia_ifp) 1625 return (EADDRNOTAVAIL); 1626 } else { 1627 ifp = ia->ia_ifp; 1628 } 1629 } 1630 1631 if ((error = carp_set_ifp(sc, ifp))) 1632 return (error); 1633 1634 if (sc->sc_carpdev == NULL) 1635 return (EADDRNOTAVAIL); 1636 1637 if (sc->sc_naddrs == 0 && (error = carp_join_multicast(sc)) != 0) 1638 return (error); 1639 1640 sc->sc_naddrs++; 1641 if (sc->sc_carpdev != NULL) 1642 sc->sc_if.if_flags |= IFF_UP; 1643 1644 carp_set_state(sc, INIT); 1645 carp_setrun(sc, 0); 1646 1647 /* 1648 * Hook if_addrhooks so that we get a callback after in_ifinit has run, 1649 * to correct any inappropriate routes that it inserted. 1650 */ 1651 if (sc->ah_cookie == 0) { 1652 /* XXX link address hook */ 1653 } 1654 1655 return (0); 1656 } 1657 1658 int 1659 carp_join_multicast(struct carp_softc *sc) 1660 { 1661 struct ip_moptions *imo = &sc->sc_imo, tmpimo; 1662 struct in_addr addr; 1663 1664 bzero(&tmpimo, sizeof(tmpimo)); 1665 addr.s_addr = INADDR_CARP_GROUP; 1666 if ((tmpimo.imo_membership[0] = 1667 in_addmulti(&addr, &sc->sc_if)) == NULL) { 1668 return (ENOBUFS); 1669 } 1670 1671 imo->imo_membership[0] = tmpimo.imo_membership[0]; 1672 imo->imo_num_memberships = 1; 1673 imo->imo_multicast_ifp = &sc->sc_if; 1674 imo->imo_multicast_ttl = CARP_DFLTTL; 1675 imo->imo_multicast_loop = 0; 1676 return (0); 1677 } 1678 1679 1680 #ifdef INET6 1681 int 1682 carp_set_addr6(struct carp_softc *sc, struct sockaddr_in6 *sin6) 1683 { 1684 struct ifnet *ifp = sc->sc_carpdev; 1685 struct in6_ifaddr *ia, *ia_if; 1686 int error = 0; 1687 1688 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) { 1689 if (!(sc->sc_if.if_flags & IFF_UP)) 1690 carp_set_state(sc, INIT); 1691 if (sc->sc_naddrs6) 1692 sc->sc_if.if_flags |= IFF_UP; 1693 carp_setrun(sc, 0); 1694 return (0); 1695 } 1696 1697 /* we have to do this by hand to ensure we don't match on ourselves */ 1698 ia_if = NULL; 1699 for (ia = in6_ifaddr; ia; ia = ia->ia_next) { 1700 int i; 1701 1702 for (i = 0; i < 4; i++) { 1703 if ((sin6->sin6_addr.s6_addr32[i] & 1704 ia->ia_prefixmask.sin6_addr.s6_addr32[i]) != 1705 (ia->ia_addr.sin6_addr.s6_addr32[i] & 1706 ia->ia_prefixmask.sin6_addr.s6_addr32[i])) 1707 break; 1708 } 1709 /* and, yeah, we need a multicast-capable iface too */ 1710 if (ia->ia_ifp != &sc->sc_if && 1711 ia->ia_ifp->if_type != IFT_CARP && 1712 (ia->ia_ifp->if_flags & IFF_MULTICAST) && 1713 (i == 4)) { 1714 if (!ia_if) 1715 ia_if = ia; 1716 } 1717 } 1718 1719 if (ia_if) { 1720 ia = ia_if; 1721 if (sc->sc_carpdev) { 1722 if (sc->sc_carpdev != ia->ia_ifp) 1723 return (EADDRNOTAVAIL); 1724 } else { 1725 ifp = ia->ia_ifp; 1726 } 1727 } 1728 1729 if ((error = carp_set_ifp(sc, ifp))) 1730 return (error); 1731 1732 if (sc->sc_carpdev == NULL) 1733 return (EADDRNOTAVAIL); 1734 1735 if (sc->sc_naddrs6 == 0 && (error = carp_join_multicast6(sc)) != 0) 1736 return (error); 1737 1738 sc->sc_naddrs6++; 1739 if (sc->sc_carpdev != NULL) 1740 sc->sc_if.if_flags |= IFF_UP; 1741 carp_set_state(sc, INIT); 1742 carp_setrun(sc, 0); 1743 1744 return (0); 1745 } 1746 1747 int 1748 carp_join_multicast6(struct carp_softc *sc) 1749 { 1750 struct in6_multi_mship *imm, *imm2; 1751 struct ip6_moptions *im6o = &sc->sc_im6o; 1752 struct sockaddr_in6 addr6; 1753 int error; 1754 1755 /* Join IPv6 CARP multicast group */ 1756 bzero(&addr6, sizeof(addr6)); 1757 addr6.sin6_family = AF_INET6; 1758 addr6.sin6_len = sizeof(addr6); 1759 addr6.sin6_addr.s6_addr16[0] = htons(0xff02); 1760 addr6.sin6_addr.s6_addr16[1] = htons(sc->sc_if.if_index); 1761 addr6.sin6_addr.s6_addr8[15] = 0x12; 1762 if ((imm = in6_joingroup(&sc->sc_if, 1763 &addr6.sin6_addr, &error, 0)) == NULL) { 1764 return (error); 1765 } 1766 /* join solicited multicast address */ 1767 bzero(&addr6.sin6_addr, sizeof(addr6.sin6_addr)); 1768 addr6.sin6_addr.s6_addr16[0] = htons(0xff02); 1769 addr6.sin6_addr.s6_addr16[1] = htons(sc->sc_if.if_index); 1770 addr6.sin6_addr.s6_addr32[1] = 0; 1771 addr6.sin6_addr.s6_addr32[2] = htonl(1); 1772 addr6.sin6_addr.s6_addr32[3] = 0; 1773 addr6.sin6_addr.s6_addr8[12] = 0xff; 1774 if ((imm2 = in6_joingroup(&sc->sc_if, 1775 &addr6.sin6_addr, &error, 0)) == NULL) { 1776 in6_leavegroup(imm); 1777 return (error); 1778 } 1779 1780 /* apply v6 multicast membership */ 1781 im6o->im6o_multicast_ifp = &sc->sc_if; 1782 if (imm) 1783 LIST_INSERT_HEAD(&im6o->im6o_memberships, imm, 1784 i6mm_chain); 1785 if (imm2) 1786 LIST_INSERT_HEAD(&im6o->im6o_memberships, imm2, 1787 i6mm_chain); 1788 1789 return (0); 1790 } 1791 1792 #endif /* INET6 */ 1793 1794 int 1795 carp_ioctl(struct ifnet *ifp, u_long cmd, caddr_t addr) 1796 { 1797 struct proc *p = curlwp->l_proc; /* XXX */ 1798 struct carp_softc *sc = ifp->if_softc, *vr; 1799 struct carpreq carpr; 1800 struct ifaddr *ifa; 1801 struct ifreq *ifr; 1802 struct ifaliasreq *ifra; 1803 struct ifnet *cdev = NULL; 1804 int error = 0; 1805 1806 ifa = (struct ifaddr *)addr; 1807 ifra = (struct ifaliasreq *)addr; 1808 ifr = (struct ifreq *)addr; 1809 1810 switch (cmd) { 1811 case SIOCSIFADDR: 1812 switch (ifa->ifa_addr->sa_family) { 1813 #ifdef INET 1814 case AF_INET: 1815 sc->sc_if.if_flags |= IFF_UP; 1816 bcopy(ifa->ifa_addr, ifa->ifa_dstaddr, 1817 sizeof(struct sockaddr)); 1818 error = carp_set_addr(sc, satosin(ifa->ifa_addr)); 1819 break; 1820 #endif /* INET */ 1821 #ifdef INET6 1822 case AF_INET6: 1823 sc->sc_if.if_flags|= IFF_UP; 1824 error = carp_set_addr6(sc, satosin6(ifa->ifa_addr)); 1825 break; 1826 #endif /* INET6 */ 1827 default: 1828 error = EAFNOSUPPORT; 1829 break; 1830 } 1831 break; 1832 1833 case SIOCAIFADDR: 1834 switch (ifa->ifa_addr->sa_family) { 1835 #ifdef INET 1836 case AF_INET: 1837 sc->sc_if.if_flags |= IFF_UP; 1838 bcopy(ifa->ifa_addr, ifa->ifa_dstaddr, 1839 sizeof(struct sockaddr)); 1840 error = carp_set_addr(sc, satosin(&ifra->ifra_addr)); 1841 break; 1842 #endif /* INET */ 1843 #ifdef INET6 1844 case AF_INET6: 1845 sc->sc_if.if_flags |= IFF_UP; 1846 error = carp_set_addr6(sc, satosin6(&ifra->ifra_addr)); 1847 break; 1848 #endif /* INET6 */ 1849 default: 1850 error = EAFNOSUPPORT; 1851 break; 1852 } 1853 break; 1854 1855 case SIOCSIFFLAGS: 1856 if (sc->sc_state != INIT && !(ifr->ifr_flags & IFF_UP)) { 1857 callout_stop(&sc->sc_ad_tmo); 1858 callout_stop(&sc->sc_md_tmo); 1859 callout_stop(&sc->sc_md6_tmo); 1860 if (sc->sc_state == MASTER) { 1861 /* we need the interface up to bow out */ 1862 sc->sc_if.if_flags |= IFF_UP; 1863 sc->sc_bow_out = 1; 1864 carp_send_ad(sc); 1865 } 1866 sc->sc_if.if_flags &= ~IFF_UP; 1867 carp_set_state(sc, INIT); 1868 carp_setrun(sc, 0); 1869 } else if (sc->sc_state == INIT && (ifr->ifr_flags & IFF_UP)) { 1870 sc->sc_if.if_flags |= IFF_UP; 1871 carp_setrun(sc, 0); 1872 } 1873 break; 1874 1875 case SIOCSVH: 1876 if (p == 0 || (error = kauth_authorize_generic(p->p_cred, 1877 KAUTH_GENERIC_ISSUSER, &p->p_acflag))) 1878 break; 1879 if ((error = copyin(ifr->ifr_data, &carpr, sizeof carpr))) 1880 break; 1881 error = 1; 1882 if (carpr.carpr_carpdev[0] != '\0' && 1883 (cdev = ifunit(carpr.carpr_carpdev)) == NULL) 1884 return (EINVAL); 1885 if ((error = carp_set_ifp(sc, cdev))) 1886 return (error); 1887 if (sc->sc_state != INIT && carpr.carpr_state != sc->sc_state) { 1888 switch (carpr.carpr_state) { 1889 case BACKUP: 1890 callout_stop(&sc->sc_ad_tmo); 1891 carp_set_state(sc, BACKUP); 1892 carp_setrun(sc, 0); 1893 carp_setroute(sc, RTM_DELETE); 1894 break; 1895 case MASTER: 1896 carp_master_down(sc); 1897 break; 1898 default: 1899 break; 1900 } 1901 } 1902 if (carpr.carpr_vhid > 0) { 1903 if (carpr.carpr_vhid > 255) { 1904 error = EINVAL; 1905 break; 1906 } 1907 if (sc->sc_carpdev) { 1908 struct carp_if *cif; 1909 cif = (struct carp_if *)sc->sc_carpdev->if_carp; 1910 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) 1911 if (vr != sc && 1912 vr->sc_vhid == carpr.carpr_vhid) 1913 return (EINVAL); 1914 } 1915 sc->sc_vhid = carpr.carpr_vhid; 1916 carp_set_enaddr(sc); 1917 carp_set_state(sc, INIT); 1918 error--; 1919 } 1920 if (carpr.carpr_advbase > 0 || carpr.carpr_advskew > 0) { 1921 if (carpr.carpr_advskew > 254) { 1922 error = EINVAL; 1923 break; 1924 } 1925 if (carpr.carpr_advbase > 255) { 1926 error = EINVAL; 1927 break; 1928 } 1929 sc->sc_advbase = carpr.carpr_advbase; 1930 sc->sc_advskew = carpr.carpr_advskew; 1931 error--; 1932 } 1933 bcopy(carpr.carpr_key, sc->sc_key, sizeof(sc->sc_key)); 1934 if (error > 0) 1935 error = EINVAL; 1936 else { 1937 error = 0; 1938 carp_setrun(sc, 0); 1939 } 1940 break; 1941 1942 case SIOCGVH: 1943 bzero(&carpr, sizeof(carpr)); 1944 if (sc->sc_carpdev != NULL) 1945 strlcpy(carpr.carpr_carpdev, sc->sc_carpdev->if_xname, 1946 IFNAMSIZ); 1947 carpr.carpr_state = sc->sc_state; 1948 carpr.carpr_vhid = sc->sc_vhid; 1949 carpr.carpr_advbase = sc->sc_advbase; 1950 carpr.carpr_advskew = sc->sc_advskew; 1951 1952 if (p != 0 || !(error = kauth_authorize_generic(p->p_cred, 1953 KAUTH_GENERIC_ISSUSER, &p->p_acflag))) 1954 bcopy(sc->sc_key, carpr.carpr_key, 1955 sizeof(carpr.carpr_key)); 1956 error = copyout(&carpr, ifr->ifr_data, sizeof(carpr)); 1957 break; 1958 1959 case SIOCADDMULTI: 1960 error = carp_ether_addmulti(sc, ifr); 1961 break; 1962 1963 case SIOCDELMULTI: 1964 error = carp_ether_delmulti(sc, ifr); 1965 break; 1966 1967 default: 1968 error = EINVAL; 1969 } 1970 1971 carp_hmac_prepare(sc); 1972 return (error); 1973 } 1974 1975 1976 /* 1977 * Start output on carp interface. This function should never be called. 1978 */ 1979 void 1980 carp_start(struct ifnet *ifp) 1981 { 1982 #ifdef DEBUG 1983 printf("%s: start called\n", ifp->if_xname); 1984 #endif 1985 } 1986 1987 int 1988 carp_output(struct ifnet *ifp, struct mbuf *m, struct sockaddr *sa, 1989 struct rtentry *rt) 1990 { 1991 struct carp_softc *sc = ((struct carp_softc *)ifp->if_softc); 1992 1993 if (sc->sc_carpdev != NULL && sc->sc_state == MASTER) { 1994 return (sc->sc_carpdev->if_output(ifp, m, sa, rt)); 1995 } else { 1996 m_freem(m); 1997 return (ENETUNREACH); 1998 } 1999 } 2000 2001 void 2002 carp_set_state(struct carp_softc *sc, int state) 2003 { 2004 if (sc->sc_state == state) 2005 return; 2006 2007 sc->sc_state = state; 2008 switch (state) { 2009 case BACKUP: 2010 sc->sc_if.if_link_state = LINK_STATE_DOWN; 2011 break; 2012 case MASTER: 2013 sc->sc_if.if_link_state = LINK_STATE_UP; 2014 break; 2015 default: 2016 sc->sc_if.if_link_state = LINK_STATE_UNKNOWN; 2017 break; 2018 } 2019 rt_ifmsg(&sc->sc_if); 2020 } 2021 2022 void 2023 carp_carpdev_state(void *v) 2024 { 2025 struct carp_if *cif; 2026 struct carp_softc *sc; 2027 struct ifnet *ifp = v; 2028 2029 if (ifp->if_type == IFT_CARP) 2030 return; 2031 2032 cif = (struct carp_if *)ifp->if_carp; 2033 2034 TAILQ_FOREACH(sc, &cif->vhif_vrs, sc_list) { 2035 int suppressed = sc->sc_suppress; 2036 2037 if (sc->sc_carpdev->if_link_state == LINK_STATE_DOWN || 2038 !(sc->sc_carpdev->if_flags & IFF_UP)) { 2039 sc->sc_if.if_flags &= ~IFF_RUNNING; 2040 callout_stop(&sc->sc_ad_tmo); 2041 callout_stop(&sc->sc_md_tmo); 2042 callout_stop(&sc->sc_md6_tmo); 2043 carp_set_state(sc, INIT); 2044 sc->sc_suppress = 1; 2045 carp_setrun(sc, 0); 2046 if (!suppressed) { 2047 carp_suppress_preempt++; 2048 if (carp_suppress_preempt == 1) 2049 carp_send_ad_all(); 2050 } 2051 } else { 2052 carp_set_state(sc, INIT); 2053 sc->sc_suppress = 0; 2054 carp_setrun(sc, 0); 2055 if (suppressed) 2056 carp_suppress_preempt--; 2057 } 2058 } 2059 } 2060 2061 int 2062 carp_ether_addmulti(struct carp_softc *sc, struct ifreq *ifr) 2063 { 2064 struct ifnet *ifp; 2065 struct carp_mc_entry *mc; 2066 u_int8_t addrlo[ETHER_ADDR_LEN], addrhi[ETHER_ADDR_LEN]; 2067 int error; 2068 2069 ifp = sc->sc_carpdev; 2070 if (ifp == NULL) 2071 return (EINVAL); 2072 2073 error = ether_addmulti(ifr, &sc->sc_ac); 2074 if (error != ENETRESET) 2075 return (error); 2076 2077 /* 2078 * This is new multicast address. We have to tell parent 2079 * about it. Also, remember this multicast address so that 2080 * we can delete them on unconfigure. 2081 */ 2082 MALLOC(mc, struct carp_mc_entry *, sizeof(struct carp_mc_entry), 2083 M_DEVBUF, M_NOWAIT); 2084 if (mc == NULL) { 2085 error = ENOMEM; 2086 goto alloc_failed; 2087 } 2088 2089 /* 2090 * As ether_addmulti() returns ENETRESET, following two 2091 * statement shouldn't fail. 2092 */ 2093 (void)ether_multiaddr(&ifr->ifr_addr, addrlo, addrhi); 2094 ETHER_LOOKUP_MULTI(addrlo, addrhi, &sc->sc_ac, mc->mc_enm); 2095 memcpy(&mc->mc_addr, &ifr->ifr_addr, ifr->ifr_addr.sa_len); 2096 LIST_INSERT_HEAD(&sc->carp_mc_listhead, mc, mc_entries); 2097 2098 error = (*ifp->if_ioctl)(ifp, SIOCADDMULTI, (caddr_t)ifr); 2099 if (error != 0) 2100 goto ioctl_failed; 2101 2102 return (error); 2103 2104 ioctl_failed: 2105 LIST_REMOVE(mc, mc_entries); 2106 FREE(mc, M_DEVBUF); 2107 alloc_failed: 2108 (void)ether_delmulti(ifr, &sc->sc_ac); 2109 2110 return (error); 2111 } 2112 2113 int 2114 carp_ether_delmulti(struct carp_softc *sc, struct ifreq *ifr) 2115 { 2116 struct ifnet *ifp; 2117 struct ether_multi *enm; 2118 struct carp_mc_entry *mc; 2119 u_int8_t addrlo[ETHER_ADDR_LEN], addrhi[ETHER_ADDR_LEN]; 2120 int error; 2121 2122 ifp = sc->sc_carpdev; 2123 if (ifp == NULL) 2124 return (EINVAL); 2125 2126 /* 2127 * Find a key to lookup carp_mc_entry. We have to do this 2128 * before calling ether_delmulti for obvious reason. 2129 */ 2130 if ((error = ether_multiaddr(&ifr->ifr_addr, addrlo, addrhi)) != 0) 2131 return (error); 2132 ETHER_LOOKUP_MULTI(addrlo, addrhi, &sc->sc_ac, enm); 2133 if (enm == NULL) 2134 return (EINVAL); 2135 2136 LIST_FOREACH(mc, &sc->carp_mc_listhead, mc_entries) 2137 if (mc->mc_enm == enm) 2138 break; 2139 2140 /* We won't delete entries we didn't add */ 2141 if (mc == NULL) 2142 return (EINVAL); 2143 2144 error = ether_delmulti(ifr, &sc->sc_ac); 2145 if (error != ENETRESET) 2146 return (error); 2147 2148 /* We no longer use this multicast address. Tell parent so. */ 2149 error = (*ifp->if_ioctl)(ifp, SIOCDELMULTI, (caddr_t)ifr); 2150 if (error == 0) { 2151 /* And forget about this address. */ 2152 LIST_REMOVE(mc, mc_entries); 2153 FREE(mc, M_DEVBUF); 2154 } else 2155 (void)ether_addmulti(ifr, &sc->sc_ac); 2156 return (error); 2157 } 2158 2159 /* 2160 * Delete any multicast address we have asked to add from parent 2161 * interface. Called when the carp is being unconfigured. 2162 */ 2163 void 2164 carp_ether_purgemulti(struct carp_softc *sc) 2165 { 2166 struct ifnet *ifp = sc->sc_carpdev; /* Parent. */ 2167 struct carp_mc_entry *mc; 2168 union { 2169 struct ifreq ifreq; 2170 struct { 2171 char ifr_name[IFNAMSIZ]; 2172 struct sockaddr_storage ifr_ss; 2173 } ifreq_storage; 2174 } u; 2175 struct ifreq *ifr = &u.ifreq; 2176 2177 if (ifp == NULL) 2178 return; 2179 2180 memcpy(ifr->ifr_name, ifp->if_xname, IFNAMSIZ); 2181 while ((mc = LIST_FIRST(&sc->carp_mc_listhead)) != NULL) { 2182 memcpy(&ifr->ifr_addr, &mc->mc_addr, mc->mc_addr.ss_len); 2183 (void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, (caddr_t)ifr); 2184 LIST_REMOVE(mc, mc_entries); 2185 FREE(mc, M_DEVBUF); 2186 } 2187 } 2188 2189 /* 2190 * Compute number of hz in the specified amount of time. This 2191 * function is from OpenBSD. 2192 */ 2193 int 2194 tvtohz(struct timeval *tv) 2195 { 2196 unsigned long ticks; 2197 long sec, usec; 2198 2199 /* 2200 * If the number of usecs in the whole seconds part of the time 2201 * fits in a long, then the total number of usecs will 2202 * fit in an unsigned long. Compute the total and convert it to 2203 * ticks, rounding up and adding 1 to allow for the current tick 2204 * to expire. Rounding also depends on unsigned long arithmetic 2205 * to avoid overflow. 2206 * 2207 * Otherwise, if the number of ticks in the whole seconds part of 2208 * the time fits in a long, then convert the parts to 2209 * ticks separately and add, using similar rounding methods and 2210 * overflow avoidance. This method would work in the previous 2211 * case but it is slightly slower and assumes that hz is integral. 2212 * 2213 * Otherwise, round the time down to the maximum 2214 * representable value. 2215 * 2216 * If ints have 32 bits, then the maximum value for any timeout in 2217 * 10ms ticks is 248 days. 2218 */ 2219 sec = tv->tv_sec; 2220 usec = tv->tv_usec; 2221 if (sec < 0 || (sec == 0 && usec <= 0)) 2222 ticks = 0; 2223 else if (sec <= LONG_MAX / 1000000) 2224 ticks = (sec * 1000000 + (unsigned long)usec + (tick - 1)) 2225 / tick + 1; 2226 else if (sec <= LONG_MAX / hz) 2227 ticks = sec * hz 2228 + ((unsigned long)usec + (tick - 1)) / tick + 1; 2229 else 2230 ticks = LONG_MAX; 2231 if (ticks > INT_MAX) 2232 ticks = INT_MAX; 2233 return ((int)ticks); 2234 } 2235 2236 SYSCTL_SETUP(sysctl_net_inet_carp_setup, "sysctl net.inet.carp subtree setup") 2237 { 2238 2239 sysctl_createv(clog, 0, NULL, NULL, 2240 CTLFLAG_PERMANENT, 2241 CTLTYPE_NODE, "net", NULL, 2242 NULL, 0, NULL, 0, 2243 CTL_NET, CTL_EOL); 2244 sysctl_createv(clog, 0, NULL, NULL, 2245 CTLFLAG_PERMANENT, 2246 CTLTYPE_NODE, "inet", NULL, 2247 NULL, 0, NULL, 0, 2248 CTL_NET, PF_INET, CTL_EOL); 2249 sysctl_createv(clog, 0, NULL, NULL, 2250 CTLFLAG_PERMANENT, 2251 CTLTYPE_NODE, "carp", 2252 SYSCTL_DESCR("CARP related settings"), 2253 NULL, 0, NULL, 0, 2254 CTL_NET, PF_INET, IPPROTO_CARP, CTL_EOL); 2255 2256 sysctl_createv(clog, 0, NULL, NULL, 2257 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2258 CTLTYPE_INT, "preempt", 2259 SYSCTL_DESCR("Enable CARP Preempt"), 2260 NULL, 0, &carp_opts[CARPCTL_PREEMPT], 0, 2261 CTL_NET, PF_INET, IPPROTO_CARP, 2262 CTL_CREATE, CTL_EOL); 2263 sysctl_createv(clog, 0, NULL, NULL, 2264 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2265 CTLTYPE_INT, "arpbalance", 2266 SYSCTL_DESCR("Enable ARP balancing"), 2267 NULL, 0, &carp_opts[CARPCTL_ARPBALANCE], 0, 2268 CTL_NET, PF_INET, IPPROTO_CARP, 2269 CTL_CREATE, CTL_EOL); 2270 sysctl_createv(clog, 0, NULL, NULL, 2271 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2272 CTLTYPE_INT, "allow", 2273 SYSCTL_DESCR("Enable CARP"), 2274 NULL, 0, &carp_opts[CARPCTL_ALLOW], 0, 2275 CTL_NET, PF_INET, IPPROTO_CARP, 2276 CTL_CREATE, CTL_EOL); 2277 sysctl_createv(clog, 0, NULL, NULL, 2278 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2279 CTLTYPE_INT, "log", 2280 SYSCTL_DESCR("CARP logging"), 2281 NULL, 0, &carp_opts[CARPCTL_LOG], 0, 2282 CTL_NET, PF_INET, IPPROTO_CARP, 2283 CTL_CREATE, CTL_EOL); 2284 sysctl_createv(clog, 0, NULL, NULL, 2285 CTLFLAG_PERMANENT, 2286 CTLTYPE_STRUCT, "stats", 2287 SYSCTL_DESCR("CARP statistics"), 2288 NULL, 0, &carpstats, sizeof(carpstats), 2289 CTL_NET, PF_INET, IPPROTO_CARP, CARPCTL_STATS, 2290 CTL_EOL); 2291 } 2292
Indexes created Mon Sep 22 21:09:42 GMT 2025