1 /* $NetBSD: ip_carp.c,v 1.6 2006/10/12 01:32:38 christos 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 __unused) 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 /* 595 * Check if our own advertisement was duplicated 596 * from a non simplex interface. 597 * XXX If there is no address on our physical interface 598 * there is no way to distinguish our ads from the ones 599 * another carp host might have sent us. 600 */ 601 if ((sc->sc_carpdev->if_flags & IFF_SIMPLEX) == 0) { 602 struct sockaddr sa; 603 struct ifaddr *ifa; 604 605 bzero(&sa, sizeof(sa)); 606 sa.sa_family = af; 607 ifa = ifaof_ifpforaddr(&sa, sc->sc_carpdev); 608 609 if (ifa && af == AF_INET) { 610 struct ip *ip = mtod(m, struct ip *); 611 if (ip->ip_src.s_addr == 612 ifatoia(ifa)->ia_addr.sin_addr.s_addr) { 613 m_freem(m); 614 return; 615 } 616 } 617 #ifdef INET6 618 if (ifa && af == AF_INET6) { 619 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 620 struct in6_addr in6_src, in6_found; 621 622 in6_src = ip6->ip6_src; 623 in6_found = ifatoia6(ifa)->ia_addr.sin6_addr; 624 if (IN6_IS_ADDR_LINKLOCAL(&in6_src)) 625 in6_src.s6_addr16[1] = 0; 626 if (IN6_IS_ADDR_LINKLOCAL(&in6_found)) 627 in6_found.s6_addr16[1] = 0; 628 if (IN6_ARE_ADDR_EQUAL(&in6_src, &in6_found)) { 629 m_freem(m); 630 return; 631 } 632 } 633 #endif /* INET6 */ 634 } 635 636 microtime(&sc->sc_if.if_lastchange); 637 sc->sc_if.if_ipackets++; 638 sc->sc_if.if_ibytes += m->m_pkthdr.len; 639 640 /* verify the CARP version. */ 641 if (ch->carp_version != CARP_VERSION) { 642 carpstats.carps_badver++; 643 sc->sc_if.if_ierrors++; 644 CARP_LOG(sc, ("invalid version %d != %d", 645 ch->carp_version, CARP_VERSION)); 646 m_freem(m); 647 return; 648 } 649 650 /* verify the hash */ 651 if (carp_hmac_verify(sc, ch->carp_counter, ch->carp_md)) { 652 carpstats.carps_badauth++; 653 sc->sc_if.if_ierrors++; 654 CARP_LOG(sc, ("incorrect hash")); 655 m_freem(m); 656 return; 657 } 658 659 tmp_counter = ntohl(ch->carp_counter[0]); 660 tmp_counter = tmp_counter<<32; 661 tmp_counter += ntohl(ch->carp_counter[1]); 662 663 /* XXX Replay protection goes here */ 664 665 sc->sc_init_counter = 0; 666 sc->sc_counter = tmp_counter; 667 668 669 sc_tv.tv_sec = sc->sc_advbase; 670 if (carp_suppress_preempt && sc->sc_advskew < 240) 671 sc_tv.tv_usec = 240 * 1000000 / 256; 672 else 673 sc_tv.tv_usec = sc->sc_advskew * 1000000 / 256; 674 ch_tv.tv_sec = ch->carp_advbase; 675 ch_tv.tv_usec = ch->carp_advskew * 1000000 / 256; 676 677 switch (sc->sc_state) { 678 case INIT: 679 break; 680 case MASTER: 681 /* 682 * If we receive an advertisement from a backup who's going to 683 * be more frequent than us, go into BACKUP state. 684 */ 685 if (timercmp(&sc_tv, &ch_tv, >) || 686 timercmp(&sc_tv, &ch_tv, ==)) { 687 callout_stop(&sc->sc_ad_tmo); 688 carp_set_state(sc, BACKUP); 689 carp_setrun(sc, 0); 690 carp_setroute(sc, RTM_DELETE); 691 } 692 break; 693 case BACKUP: 694 /* 695 * If we're pre-empting masters who advertise slower than us, 696 * and this one claims to be slower, treat him as down. 697 */ 698 if (carp_opts[CARPCTL_PREEMPT] && timercmp(&sc_tv, &ch_tv, <)) { 699 carp_master_down(sc); 700 break; 701 } 702 703 /* 704 * If the master is going to advertise at such a low frequency 705 * that he's guaranteed to time out, we'd might as well just 706 * treat him as timed out now. 707 */ 708 sc_tv.tv_sec = sc->sc_advbase * 3; 709 if (timercmp(&sc_tv, &ch_tv, <)) { 710 carp_master_down(sc); 711 break; 712 } 713 714 /* 715 * Otherwise, we reset the counter and wait for the next 716 * advertisement. 717 */ 718 carp_setrun(sc, af); 719 break; 720 } 721 722 m_freem(m); 723 return; 724 } 725 726 /* 727 * Interface side of the CARP implementation. 728 */ 729 730 /* ARGSUSED */ 731 void 732 carpattach(int n __unused) 733 { 734 if_clone_attach(&carp_cloner); 735 } 736 737 int 738 carp_clone_create(struct if_clone *ifc, int unit) 739 { 740 extern int ifqmaxlen; 741 struct carp_softc *sc; 742 struct ifnet *ifp; 743 744 sc = malloc(sizeof(*sc), M_DEVBUF, M_NOWAIT); 745 if (!sc) 746 return (ENOMEM); 747 bzero(sc, sizeof(*sc)); 748 749 sc->sc_suppress = 0; 750 sc->sc_advbase = CARP_DFLTINTV; 751 sc->sc_vhid = -1; /* required setting */ 752 sc->sc_advskew = 0; 753 sc->sc_init_counter = 1; 754 sc->sc_naddrs = sc->sc_naddrs6 = 0; 755 #ifdef INET6 756 sc->sc_im6o.im6o_multicast_hlim = CARP_DFLTTL; 757 #endif /* INET6 */ 758 759 callout_init(&sc->sc_ad_tmo); 760 callout_init(&sc->sc_md_tmo); 761 callout_init(&sc->sc_md6_tmo); 762 763 callout_setfunc(&sc->sc_ad_tmo, carp_send_ad, sc); 764 callout_setfunc(&sc->sc_md_tmo, carp_master_down, sc); 765 callout_setfunc(&sc->sc_md6_tmo, carp_master_down, sc); 766 767 LIST_INIT(&sc->carp_mc_listhead); 768 ifp = &sc->sc_if; 769 ifp->if_softc = sc; 770 snprintf(ifp->if_xname, sizeof ifp->if_xname, "%s%d", ifc->ifc_name, 771 unit); 772 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST; 773 ifp->if_ioctl = carp_ioctl; 774 ifp->if_start = carp_start; 775 ifp->if_output = carp_output; 776 ifp->if_type = IFT_CARP; 777 ifp->if_addrlen = ETHER_ADDR_LEN; 778 ifp->if_hdrlen = ETHER_HDR_LEN; 779 ifp->if_mtu = ETHERMTU; 780 IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen); 781 IFQ_SET_READY(&ifp->if_snd); 782 if_attach(ifp); 783 784 if_alloc_sadl(ifp); 785 ifp->if_broadcastaddr = etherbroadcastaddr; 786 carp_set_enaddr(sc); 787 LIST_INIT(&sc->sc_ac.ec_multiaddrs); 788 #if NBPFILTER > 0 789 bpfattach(ifp, DLT_EN10MB, ETHER_HDR_LEN); 790 #endif 791 return (0); 792 } 793 794 int 795 carp_clone_destroy(struct ifnet *ifp) 796 { 797 carpdetach(ifp->if_softc); 798 ether_ifdetach(ifp); 799 if_detach(ifp); 800 free(ifp->if_softc, M_DEVBUF); 801 802 return (0); 803 } 804 805 void 806 carpdetach(struct carp_softc *sc) 807 { 808 struct carp_if *cif; 809 int s; 810 811 callout_stop(&sc->sc_ad_tmo); 812 callout_stop(&sc->sc_md_tmo); 813 callout_stop(&sc->sc_md6_tmo); 814 815 if (sc->sc_suppress) 816 carp_suppress_preempt--; 817 sc->sc_suppress = 0; 818 819 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) 820 carp_suppress_preempt--; 821 sc->sc_sendad_errors = 0; 822 823 carp_set_state(sc, INIT); 824 sc->sc_if.if_flags &= ~IFF_UP; 825 carp_setrun(sc, 0); 826 carp_multicast_cleanup(sc); 827 828 s = splnet(); 829 if (sc->sc_carpdev != NULL) { 830 /* XXX linkstatehook removal */ 831 cif = (struct carp_if *)sc->sc_carpdev->if_carp; 832 TAILQ_REMOVE(&cif->vhif_vrs, sc, sc_list); 833 if (!--cif->vhif_nvrs) { 834 ifpromisc(sc->sc_carpdev, 0); 835 sc->sc_carpdev->if_carp = NULL; 836 FREE(cif, M_IFADDR); 837 } 838 } 839 sc->sc_carpdev = NULL; 840 splx(s); 841 } 842 843 /* Detach an interface from the carp. */ 844 void 845 carp_ifdetach(struct ifnet *ifp) 846 { 847 struct carp_softc *sc, *nextsc; 848 struct carp_if *cif = (struct carp_if *)ifp->if_carp; 849 850 for (sc = TAILQ_FIRST(&cif->vhif_vrs); sc; sc = nextsc) { 851 nextsc = TAILQ_NEXT(sc, sc_list); 852 carpdetach(sc); 853 } 854 } 855 856 int 857 carp_prepare_ad(struct mbuf *m __unused, struct carp_softc *sc, 858 struct carp_header *ch) 859 { 860 if (sc->sc_init_counter) { 861 /* this could also be seconds since unix epoch */ 862 sc->sc_counter = arc4random(); 863 sc->sc_counter = sc->sc_counter << 32; 864 sc->sc_counter += arc4random(); 865 } else 866 sc->sc_counter++; 867 868 ch->carp_counter[0] = htonl((sc->sc_counter>>32)&0xffffffff); 869 ch->carp_counter[1] = htonl(sc->sc_counter&0xffffffff); 870 871 carp_hmac_generate(sc, ch->carp_counter, ch->carp_md); 872 873 return (0); 874 } 875 876 void 877 carp_send_ad_all(void) 878 { 879 struct ifnet *ifp; 880 struct carp_if *cif; 881 struct carp_softc *vh; 882 883 TAILQ_FOREACH(ifp, &ifnet, if_list) { 884 if (ifp->if_carp == NULL || ifp->if_type == IFT_CARP) 885 continue; 886 887 cif = (struct carp_if *)ifp->if_carp; 888 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 889 if ((vh->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) == 890 (IFF_UP|IFF_RUNNING) && vh->sc_state == MASTER) 891 carp_send_ad(vh); 892 } 893 } 894 } 895 896 897 void 898 carp_send_ad(void *v) 899 { 900 struct carp_header ch; 901 struct timeval tv; 902 struct carp_softc *sc = v; 903 struct carp_header *ch_ptr; 904 struct mbuf *m; 905 int error, len, advbase, advskew, s; 906 struct ifaddr *ifa; 907 struct sockaddr sa; 908 909 s = splsoftnet(); 910 911 advbase = advskew = 0; /* Sssssh compiler */ 912 if (sc->sc_carpdev == NULL) { 913 sc->sc_if.if_oerrors++; 914 goto retry_later; 915 } 916 917 /* bow out if we've gone to backup (the carp interface is going down) */ 918 if (sc->sc_bow_out) { 919 sc->sc_bow_out = 0; 920 advbase = 255; 921 advskew = 255; 922 } else { 923 advbase = sc->sc_advbase; 924 if (!carp_suppress_preempt || sc->sc_advskew > 240) 925 advskew = sc->sc_advskew; 926 else 927 advskew = 240; 928 tv.tv_sec = advbase; 929 tv.tv_usec = advskew * 1000000 / 256; 930 } 931 932 ch.carp_version = CARP_VERSION; 933 ch.carp_type = CARP_ADVERTISEMENT; 934 ch.carp_vhid = sc->sc_vhid; 935 ch.carp_advbase = advbase; 936 ch.carp_advskew = advskew; 937 ch.carp_authlen = 7; /* XXX DEFINE */ 938 ch.carp_pad1 = 0; /* must be zero */ 939 ch.carp_cksum = 0; 940 941 942 #ifdef INET 943 if (sc->sc_naddrs) { 944 struct ip *ip; 945 946 MGETHDR(m, M_DONTWAIT, MT_HEADER); 947 if (m == NULL) { 948 sc->sc_if.if_oerrors++; 949 carpstats.carps_onomem++; 950 /* XXX maybe less ? */ 951 goto retry_later; 952 } 953 len = sizeof(*ip) + sizeof(ch); 954 m->m_pkthdr.len = len; 955 m->m_pkthdr.rcvif = NULL; 956 m->m_len = len; 957 MH_ALIGN(m, m->m_len); 958 m->m_flags |= M_MCAST; 959 ip = mtod(m, struct ip *); 960 ip->ip_v = IPVERSION; 961 ip->ip_hl = sizeof(*ip) >> 2; 962 ip->ip_tos = IPTOS_LOWDELAY; 963 ip->ip_len = htons(len); 964 ip->ip_id = htons(ip_randomid()); 965 ip->ip_off = htons(IP_DF); 966 ip->ip_ttl = CARP_DFLTTL; 967 ip->ip_p = IPPROTO_CARP; 968 ip->ip_sum = 0; 969 970 bzero(&sa, sizeof(sa)); 971 sa.sa_family = AF_INET; 972 ifa = ifaof_ifpforaddr(&sa, sc->sc_carpdev); 973 if (ifa == NULL) 974 ip->ip_src.s_addr = 0; 975 else 976 ip->ip_src.s_addr = 977 ifatoia(ifa)->ia_addr.sin_addr.s_addr; 978 ip->ip_dst.s_addr = INADDR_CARP_GROUP; 979 980 ch_ptr = (struct carp_header *)(&ip[1]); 981 bcopy(&ch, ch_ptr, sizeof(ch)); 982 if (carp_prepare_ad(m, sc, ch_ptr)) 983 goto retry_later; 984 985 m->m_data += sizeof(*ip); 986 ch_ptr->carp_cksum = carp_cksum(m, len - sizeof(*ip)); 987 m->m_data -= sizeof(*ip); 988 989 microtime(&sc->sc_if.if_lastchange); 990 sc->sc_if.if_opackets++; 991 sc->sc_if.if_obytes += len; 992 carpstats.carps_opackets++; 993 994 error = ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo, 995 NULL); 996 if (error) { 997 if (error == ENOBUFS) 998 carpstats.carps_onomem++; 999 else 1000 CARP_LOG(sc, ("ip_output failed: %d", error)); 1001 sc->sc_if.if_oerrors++; 1002 if (sc->sc_sendad_errors < INT_MAX) 1003 sc->sc_sendad_errors++; 1004 if (sc->sc_sendad_errors == CARP_SENDAD_MAX_ERRORS) { 1005 carp_suppress_preempt++; 1006 if (carp_suppress_preempt == 1) 1007 carp_send_ad_all(); 1008 } 1009 sc->sc_sendad_success = 0; 1010 } else { 1011 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) { 1012 if (++sc->sc_sendad_success >= 1013 CARP_SENDAD_MIN_SUCCESS) { 1014 carp_suppress_preempt--; 1015 sc->sc_sendad_errors = 0; 1016 } 1017 } else 1018 sc->sc_sendad_errors = 0; 1019 } 1020 } 1021 #endif /* INET */ 1022 #ifdef INET6 1023 if (sc->sc_naddrs6) { 1024 struct ip6_hdr *ip6; 1025 1026 MGETHDR(m, M_DONTWAIT, MT_HEADER); 1027 if (m == NULL) { 1028 sc->sc_if.if_oerrors++; 1029 carpstats.carps_onomem++; 1030 /* XXX maybe less ? */ 1031 goto retry_later; 1032 } 1033 len = sizeof(*ip6) + sizeof(ch); 1034 m->m_pkthdr.len = len; 1035 m->m_pkthdr.rcvif = NULL; 1036 m->m_len = len; 1037 MH_ALIGN(m, m->m_len); 1038 m->m_flags |= M_MCAST; 1039 ip6 = mtod(m, struct ip6_hdr *); 1040 bzero(ip6, sizeof(*ip6)); 1041 ip6->ip6_vfc |= IPV6_VERSION; 1042 ip6->ip6_hlim = CARP_DFLTTL; 1043 ip6->ip6_nxt = IPPROTO_CARP; 1044 1045 /* set the source address */ 1046 bzero(&sa, sizeof(sa)); 1047 sa.sa_family = AF_INET6; 1048 ifa = ifaof_ifpforaddr(&sa, sc->sc_carpdev); 1049 if (ifa == NULL) /* This should never happen with IPv6 */ 1050 bzero(&ip6->ip6_src, sizeof(struct in6_addr)); 1051 else 1052 bcopy(ifatoia6(ifa)->ia_addr.sin6_addr.s6_addr, 1053 &ip6->ip6_src, sizeof(struct in6_addr)); 1054 /* set the multicast destination */ 1055 1056 ip6->ip6_dst.s6_addr8[0] = 0xff; 1057 ip6->ip6_dst.s6_addr8[1] = 0x02; 1058 ip6->ip6_dst.s6_addr8[15] = 0x12; 1059 1060 ch_ptr = (struct carp_header *)(&ip6[1]); 1061 bcopy(&ch, ch_ptr, sizeof(ch)); 1062 if (carp_prepare_ad(m, sc, ch_ptr)) 1063 goto retry_later; 1064 1065 m->m_data += sizeof(*ip6); 1066 ch_ptr->carp_cksum = carp_cksum(m, len - sizeof(*ip6)); 1067 m->m_data -= sizeof(*ip6); 1068 1069 microtime(&sc->sc_if.if_lastchange); 1070 sc->sc_if.if_opackets++; 1071 sc->sc_if.if_obytes += len; 1072 carpstats.carps_opackets6++; 1073 1074 error = ip6_output(m, NULL, NULL, 0, &sc->sc_im6o, NULL, NULL); 1075 if (error) { 1076 if (error == ENOBUFS) 1077 carpstats.carps_onomem++; 1078 else 1079 CARP_LOG(sc, ("ip6_output failed: %d", error)); 1080 sc->sc_if.if_oerrors++; 1081 if (sc->sc_sendad_errors < INT_MAX) 1082 sc->sc_sendad_errors++; 1083 if (sc->sc_sendad_errors == CARP_SENDAD_MAX_ERRORS) { 1084 carp_suppress_preempt++; 1085 if (carp_suppress_preempt == 1) 1086 carp_send_ad_all(); 1087 } 1088 sc->sc_sendad_success = 0; 1089 } else { 1090 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) { 1091 if (++sc->sc_sendad_success >= 1092 CARP_SENDAD_MIN_SUCCESS) { 1093 carp_suppress_preempt--; 1094 sc->sc_sendad_errors = 0; 1095 } 1096 } else 1097 sc->sc_sendad_errors = 0; 1098 } 1099 } 1100 #endif /* INET6 */ 1101 1102 retry_later: 1103 splx(s); 1104 if (advbase != 255 || advskew != 255) 1105 callout_schedule(&sc->sc_ad_tmo, tvtohz(&tv)); 1106 } 1107 1108 /* 1109 * Broadcast a gratuitous ARP request containing 1110 * the virtual router MAC address for each IP address 1111 * associated with the virtual router. 1112 */ 1113 void 1114 carp_send_arp(struct carp_softc *sc) 1115 { 1116 struct ifaddr *ifa; 1117 struct in_addr *in; 1118 int s = splsoftnet(); 1119 1120 TAILQ_FOREACH(ifa, &sc->sc_if.if_addrlist, ifa_list) { 1121 1122 if (ifa->ifa_addr->sa_family != AF_INET) 1123 continue; 1124 1125 in = &ifatoia(ifa)->ia_addr.sin_addr; 1126 arprequest(sc->sc_carpdev, in, in, LLADDR(sc->sc_if.if_sadl)); 1127 DELAY(1000); /* XXX */ 1128 } 1129 splx(s); 1130 } 1131 1132 #ifdef INET6 1133 void 1134 carp_send_na(struct carp_softc *sc) 1135 { 1136 struct ifaddr *ifa; 1137 struct in6_addr *in6; 1138 static struct in6_addr mcast = IN6ADDR_LINKLOCAL_ALLNODES_INIT; 1139 int s = splsoftnet(); 1140 1141 TAILQ_FOREACH(ifa, &sc->sc_if.if_addrlist, ifa_list) { 1142 1143 if (ifa->ifa_addr->sa_family != AF_INET6) 1144 continue; 1145 1146 in6 = &ifatoia6(ifa)->ia_addr.sin6_addr; 1147 nd6_na_output(sc->sc_carpdev, &mcast, in6, 1148 ND_NA_FLAG_OVERRIDE, 1, NULL); 1149 DELAY(1000); /* XXX */ 1150 } 1151 splx(s); 1152 } 1153 #endif /* INET6 */ 1154 1155 /* 1156 * Based on bridge_hash() in if_bridge.c 1157 */ 1158 #define mix(a,b,c) \ 1159 do { \ 1160 a -= b; a -= c; a ^= (c >> 13); \ 1161 b -= c; b -= a; b ^= (a << 8); \ 1162 c -= a; c -= b; c ^= (b >> 13); \ 1163 a -= b; a -= c; a ^= (c >> 12); \ 1164 b -= c; b -= a; b ^= (a << 16); \ 1165 c -= a; c -= b; c ^= (b >> 5); \ 1166 a -= b; a -= c; a ^= (c >> 3); \ 1167 b -= c; b -= a; b ^= (a << 10); \ 1168 c -= a; c -= b; c ^= (b >> 15); \ 1169 } while (0) 1170 1171 u_int32_t 1172 carp_hash(struct carp_softc *sc, u_char *src) 1173 { 1174 u_int32_t a = 0x9e3779b9, b = sc->sc_hashkey[0], c = sc->sc_hashkey[1]; 1175 1176 c += sc->sc_key[3] << 24; 1177 c += sc->sc_key[2] << 16; 1178 c += sc->sc_key[1] << 8; 1179 c += sc->sc_key[0]; 1180 b += src[5] << 8; 1181 b += src[4]; 1182 a += src[3] << 24; 1183 a += src[2] << 16; 1184 a += src[1] << 8; 1185 a += src[0]; 1186 1187 mix(a, b, c); 1188 return (c); 1189 } 1190 1191 int 1192 carp_addrcount(struct carp_if *cif, struct in_ifaddr *ia, int type) 1193 { 1194 struct carp_softc *vh; 1195 struct ifaddr *ifa; 1196 int count = 0; 1197 1198 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 1199 if ((type == CARP_COUNT_RUNNING && 1200 (vh->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) == 1201 (IFF_UP|IFF_RUNNING)) || 1202 (type == CARP_COUNT_MASTER && vh->sc_state == MASTER)) { 1203 TAILQ_FOREACH(ifa, &vh->sc_if.if_addrlist, ifa_list) { 1204 if (ifa->ifa_addr->sa_family == AF_INET && 1205 ia->ia_addr.sin_addr.s_addr == 1206 ifatoia(ifa)->ia_addr.sin_addr.s_addr) 1207 count++; 1208 } 1209 } 1210 } 1211 return (count); 1212 } 1213 1214 int 1215 carp_iamatch(struct in_ifaddr *ia, u_char *src, 1216 u_int32_t *count, u_int32_t index) 1217 { 1218 struct carp_softc *sc = ia->ia_ifp->if_softc; 1219 1220 if (carp_opts[CARPCTL_ARPBALANCE]) { 1221 /* 1222 * We use the source ip to decide which virtual host should 1223 * handle the request. If we're master of that virtual host, 1224 * then we respond, otherwise, just drop the arp packet on 1225 * the floor. 1226 */ 1227 1228 /* Count the elegible carp interfaces with this address */ 1229 if (*count == 0) 1230 *count = carp_addrcount( 1231 (struct carp_if *)ia->ia_ifp->if_carpdev->if_carp, 1232 ia, CARP_COUNT_RUNNING); 1233 1234 /* This should never happen, but... */ 1235 if (*count == 0) 1236 return (0); 1237 1238 if (carp_hash(sc, src) % *count == index - 1 && 1239 sc->sc_state == MASTER) { 1240 return (1); 1241 } 1242 } else { 1243 if (sc->sc_state == MASTER) 1244 return (1); 1245 } 1246 1247 return (0); 1248 } 1249 1250 #ifdef INET6 1251 struct ifaddr * 1252 carp_iamatch6(void *v, struct in6_addr *taddr) 1253 { 1254 struct carp_if *cif = v; 1255 struct carp_softc *vh; 1256 struct ifaddr *ifa; 1257 1258 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 1259 TAILQ_FOREACH(ifa, &vh->sc_if.if_addrlist, ifa_list) { 1260 if (IN6_ARE_ADDR_EQUAL(taddr, 1261 &ifatoia6(ifa)->ia_addr.sin6_addr) && 1262 ((vh->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) == 1263 (IFF_UP|IFF_RUNNING)) && vh->sc_state == MASTER) 1264 return (ifa); 1265 } 1266 } 1267 1268 return (NULL); 1269 } 1270 #endif /* INET6 */ 1271 1272 struct ifnet * 1273 carp_ourether(void *v, struct ether_header *eh, u_char iftype, int src) 1274 { 1275 struct carp_if *cif = (struct carp_if *)v; 1276 struct carp_softc *vh; 1277 u_int8_t *ena; 1278 1279 if (src) 1280 ena = (u_int8_t *)&eh->ether_shost; 1281 else 1282 ena = (u_int8_t *)&eh->ether_dhost; 1283 1284 switch (iftype) { 1285 case IFT_ETHER: 1286 case IFT_FDDI: 1287 if (ena[0] || ena[1] || ena[2] != 0x5e || ena[3] || ena[4] != 1) 1288 return (NULL); 1289 break; 1290 case IFT_ISO88025: 1291 if (ena[0] != 3 || ena[1] || ena[4] || ena[5]) 1292 return (NULL); 1293 break; 1294 default: 1295 return (NULL); 1296 break; 1297 } 1298 1299 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) 1300 if ((vh->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) == 1301 (IFF_UP|IFF_RUNNING) && vh->sc_state == MASTER && 1302 !bcmp(ena, LLADDR(vh->sc_if.if_sadl), 1303 ETHER_ADDR_LEN)) { 1304 return (&vh->sc_if); 1305 } 1306 1307 return (NULL); 1308 } 1309 1310 int 1311 carp_input(struct mbuf *m, u_int8_t *shost, u_int8_t *dhost, u_int16_t etype) 1312 { 1313 struct ether_header eh; 1314 struct carp_if *cif = (struct carp_if *)m->m_pkthdr.rcvif->if_carp; 1315 struct ifnet *ifp; 1316 1317 bcopy(shost, &eh.ether_shost, sizeof(eh.ether_shost)); 1318 bcopy(dhost, &eh.ether_dhost, sizeof(eh.ether_dhost)); 1319 eh.ether_type = etype; 1320 1321 if (m->m_flags & (M_BCAST|M_MCAST)) { 1322 struct carp_softc *vh; 1323 struct mbuf *m0; 1324 1325 /* 1326 * XXX Should really check the list of multicast addresses 1327 * for each CARP interface _before_ copying. 1328 */ 1329 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 1330 m0 = m_copym(m, 0, M_COPYALL, M_DONTWAIT); 1331 if (m0 == NULL) 1332 continue; 1333 m0->m_pkthdr.rcvif = &vh->sc_if; 1334 ether_input(&vh->sc_if, m0); 1335 } 1336 return (1); 1337 } 1338 1339 ifp = carp_ourether(cif, &eh, m->m_pkthdr.rcvif->if_type, 0); 1340 if (ifp == NULL) { 1341 return (1); 1342 } 1343 1344 m->m_pkthdr.rcvif = ifp; 1345 1346 #if NBPFILTER > 0 1347 if (ifp->if_bpf) 1348 bpf_mtap(ifp->if_bpf, m); 1349 #endif 1350 ifp->if_ipackets++; 1351 ether_input(ifp, m); 1352 return (0); 1353 } 1354 1355 void 1356 carp_master_down(void *v) 1357 { 1358 struct carp_softc *sc = v; 1359 1360 switch (sc->sc_state) { 1361 case INIT: 1362 printf("%s: master_down event in INIT state\n", 1363 sc->sc_if.if_xname); 1364 break; 1365 case MASTER: 1366 break; 1367 case BACKUP: 1368 carp_set_state(sc, MASTER); 1369 carp_send_ad(sc); 1370 carp_send_arp(sc); 1371 #ifdef INET6 1372 carp_send_na(sc); 1373 #endif /* INET6 */ 1374 carp_setrun(sc, 0); 1375 carp_setroute(sc, RTM_ADD); 1376 break; 1377 } 1378 } 1379 1380 /* 1381 * When in backup state, af indicates whether to reset the master down timer 1382 * for v4 or v6. If it's set to zero, reset the ones which are already pending. 1383 */ 1384 void 1385 carp_setrun(struct carp_softc *sc, sa_family_t af) 1386 { 1387 struct timeval tv; 1388 1389 if (sc->sc_carpdev == NULL) { 1390 sc->sc_if.if_flags &= ~IFF_RUNNING; 1391 carp_set_state(sc, INIT); 1392 return; 1393 } 1394 1395 if (sc->sc_if.if_flags & IFF_UP && sc->sc_vhid > 0 && 1396 (sc->sc_naddrs || sc->sc_naddrs6) && !sc->sc_suppress) { 1397 sc->sc_if.if_flags |= IFF_RUNNING; 1398 } else { 1399 sc->sc_if.if_flags &= ~IFF_RUNNING; 1400 carp_setroute(sc, RTM_DELETE); 1401 return; 1402 } 1403 1404 switch (sc->sc_state) { 1405 case INIT: 1406 carp_set_state(sc, BACKUP); 1407 carp_setroute(sc, RTM_DELETE); 1408 carp_setrun(sc, 0); 1409 break; 1410 case BACKUP: 1411 callout_stop(&sc->sc_ad_tmo); 1412 tv.tv_sec = 3 * sc->sc_advbase; 1413 tv.tv_usec = sc->sc_advskew * 1000000 / 256; 1414 switch (af) { 1415 #ifdef INET 1416 case AF_INET: 1417 callout_schedule(&sc->sc_md_tmo, tvtohz(&tv)); 1418 break; 1419 #endif /* INET */ 1420 #ifdef INET6 1421 case AF_INET6: 1422 callout_schedule(&sc->sc_md6_tmo, tvtohz(&tv)); 1423 break; 1424 #endif /* INET6 */ 1425 default: 1426 if (sc->sc_naddrs) 1427 callout_schedule(&sc->sc_md_tmo, tvtohz(&tv)); 1428 if (sc->sc_naddrs6) 1429 callout_schedule(&sc->sc_md6_tmo, tvtohz(&tv)); 1430 break; 1431 } 1432 break; 1433 case MASTER: 1434 tv.tv_sec = sc->sc_advbase; 1435 tv.tv_usec = sc->sc_advskew * 1000000 / 256; 1436 callout_schedule(&sc->sc_ad_tmo, tvtohz(&tv)); 1437 break; 1438 } 1439 } 1440 1441 void 1442 carp_multicast_cleanup(struct carp_softc *sc) 1443 { 1444 struct ip_moptions *imo = &sc->sc_imo; 1445 #ifdef INET6 1446 struct ip6_moptions *im6o = &sc->sc_im6o; 1447 #endif 1448 u_int16_t n = imo->imo_num_memberships; 1449 1450 /* Clean up our own multicast memberships */ 1451 while (n-- > 0) { 1452 if (imo->imo_membership[n] != NULL) { 1453 in_delmulti(imo->imo_membership[n]); 1454 imo->imo_membership[n] = NULL; 1455 } 1456 } 1457 imo->imo_num_memberships = 0; 1458 imo->imo_multicast_ifp = NULL; 1459 1460 #ifdef INET6 1461 while (!LIST_EMPTY(&im6o->im6o_memberships)) { 1462 struct in6_multi_mship *imm = 1463 LIST_FIRST(&im6o->im6o_memberships); 1464 1465 LIST_REMOVE(imm, i6mm_chain); 1466 in6_leavegroup(imm); 1467 } 1468 im6o->im6o_multicast_ifp = NULL; 1469 #endif 1470 1471 /* And any other multicast memberships */ 1472 carp_ether_purgemulti(sc); 1473 } 1474 1475 int 1476 carp_set_ifp(struct carp_softc *sc, struct ifnet *ifp) 1477 { 1478 struct carp_if *cif, *ncif = NULL; 1479 struct carp_softc *vr, *after = NULL; 1480 int myself = 0, error = 0; 1481 int s; 1482 1483 if (ifp == sc->sc_carpdev) 1484 return (0); 1485 1486 if (ifp != NULL) { 1487 if ((ifp->if_flags & IFF_MULTICAST) == 0) 1488 return (EADDRNOTAVAIL); 1489 1490 if (ifp->if_type == IFT_CARP) 1491 return (EINVAL); 1492 1493 if (ifp->if_carp == NULL) { 1494 MALLOC(ncif, struct carp_if *, sizeof(*cif), 1495 M_IFADDR, M_NOWAIT); 1496 if (ncif == NULL) 1497 return (ENOBUFS); 1498 if ((error = ifpromisc(ifp, 1))) { 1499 FREE(ncif, M_IFADDR); 1500 return (error); 1501 } 1502 1503 ncif->vhif_ifp = ifp; 1504 TAILQ_INIT(&ncif->vhif_vrs); 1505 } else { 1506 cif = (struct carp_if *)ifp->if_carp; 1507 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) 1508 if (vr != sc && vr->sc_vhid == sc->sc_vhid) 1509 return (EINVAL); 1510 } 1511 1512 /* detach from old interface */ 1513 if (sc->sc_carpdev != NULL) 1514 carpdetach(sc); 1515 1516 /* join multicast groups */ 1517 if (sc->sc_naddrs < 0 && 1518 (error = carp_join_multicast(sc)) != 0) { 1519 if (ncif != NULL) 1520 FREE(ncif, M_IFADDR); 1521 return (error); 1522 } 1523 1524 #ifdef INET6 1525 if (sc->sc_naddrs6 < 0 && 1526 (error = carp_join_multicast6(sc)) != 0) { 1527 if (ncif != NULL) 1528 FREE(ncif, M_IFADDR); 1529 carp_multicast_cleanup(sc); 1530 return (error); 1531 } 1532 #endif 1533 1534 /* attach carp interface to physical interface */ 1535 if (ncif != NULL) 1536 ifp->if_carp = (caddr_t)ncif; 1537 sc->sc_carpdev = ifp; 1538 cif = (struct carp_if *)ifp->if_carp; 1539 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) { 1540 if (vr == sc) 1541 myself = 1; 1542 if (vr->sc_vhid < sc->sc_vhid) 1543 after = vr; 1544 } 1545 1546 if (!myself) { 1547 /* We're trying to keep things in order */ 1548 if (after == NULL) { 1549 TAILQ_INSERT_TAIL(&cif->vhif_vrs, sc, sc_list); 1550 } else { 1551 TAILQ_INSERT_AFTER(&cif->vhif_vrs, after, 1552 sc, sc_list); 1553 } 1554 cif->vhif_nvrs++; 1555 } 1556 if (sc->sc_naddrs || sc->sc_naddrs6) 1557 sc->sc_if.if_flags |= IFF_UP; 1558 carp_set_enaddr(sc); 1559 s = splnet(); 1560 /* XXX linkstatehooks establish */ 1561 carp_carpdev_state(ifp); 1562 splx(s); 1563 } else { 1564 carpdetach(sc); 1565 sc->sc_if.if_flags &= ~(IFF_UP|IFF_RUNNING); 1566 } 1567 return (0); 1568 } 1569 1570 void 1571 carp_set_enaddr(struct carp_softc *sc) 1572 { 1573 if (sc->sc_carpdev && sc->sc_carpdev->if_type == IFT_ISO88025) { 1574 LLADDR(sc->sc_if.if_sadl)[0] = 3; 1575 LLADDR(sc->sc_if.if_sadl)[1] = 0; 1576 LLADDR(sc->sc_if.if_sadl)[2] = 0x40 >> (sc->sc_vhid - 1); 1577 LLADDR(sc->sc_if.if_sadl)[3] = 0x40000 >> (sc->sc_vhid - 1); 1578 LLADDR(sc->sc_if.if_sadl)[4] = 0; 1579 LLADDR(sc->sc_if.if_sadl)[5] = 0; 1580 } else { 1581 LLADDR(sc->sc_if.if_sadl)[0] = 0; 1582 LLADDR(sc->sc_if.if_sadl)[1] = 0; 1583 LLADDR(sc->sc_if.if_sadl)[2] = 0x5e; 1584 LLADDR(sc->sc_if.if_sadl)[3] = 0; 1585 LLADDR(sc->sc_if.if_sadl)[4] = 1; 1586 LLADDR(sc->sc_if.if_sadl)[5] = sc->sc_vhid; 1587 } 1588 } 1589 1590 void 1591 carp_addr_updated(void *v) 1592 { 1593 struct carp_softc *sc = (struct carp_softc *) v; 1594 struct ifaddr *ifa; 1595 int new_naddrs = 0, new_naddrs6 = 0; 1596 1597 TAILQ_FOREACH(ifa, &sc->sc_if.if_addrlist, ifa_list) { 1598 if (ifa->ifa_addr->sa_family == AF_INET) 1599 new_naddrs++; 1600 else if (ifa->ifa_addr->sa_family == AF_INET6) 1601 new_naddrs6++; 1602 } 1603 1604 /* Handle a callback after SIOCDIFADDR */ 1605 if (new_naddrs < sc->sc_naddrs || new_naddrs6 < sc->sc_naddrs6) { 1606 struct in_addr mc_addr; 1607 struct in_multi *inm; 1608 1609 sc->sc_naddrs = new_naddrs; 1610 sc->sc_naddrs6 = new_naddrs6; 1611 1612 /* Re-establish multicast membership removed by in_control */ 1613 mc_addr.s_addr = INADDR_CARP_GROUP; 1614 IN_LOOKUP_MULTI(mc_addr, &sc->sc_if, inm); 1615 if (inm == NULL) { 1616 bzero(&sc->sc_imo, sizeof(sc->sc_imo)); 1617 1618 if (sc->sc_carpdev != NULL && sc->sc_naddrs > 0) 1619 carp_join_multicast(sc); 1620 } 1621 1622 if (sc->sc_naddrs == 0 && sc->sc_naddrs6 == 0) { 1623 sc->sc_if.if_flags &= ~IFF_UP; 1624 carp_set_state(sc, INIT); 1625 } else 1626 carp_hmac_prepare(sc); 1627 } 1628 1629 carp_setrun(sc, 0); 1630 } 1631 1632 int 1633 carp_set_addr(struct carp_softc *sc, struct sockaddr_in *sin) 1634 { 1635 struct ifnet *ifp = sc->sc_carpdev; 1636 struct in_ifaddr *ia, *ia_if; 1637 int error = 0; 1638 1639 if (sin->sin_addr.s_addr == 0) { 1640 if (!(sc->sc_if.if_flags & IFF_UP)) 1641 carp_set_state(sc, INIT); 1642 if (sc->sc_naddrs) 1643 sc->sc_if.if_flags |= IFF_UP; 1644 carp_setrun(sc, 0); 1645 return (0); 1646 } 1647 1648 /* we have to do this by hand to ensure we don't match on ourselves */ 1649 ia_if = NULL; 1650 for (ia = TAILQ_FIRST(&in_ifaddrhead); ia; 1651 ia = TAILQ_NEXT(ia, ia_list)) { 1652 1653 /* and, yeah, we need a multicast-capable iface too */ 1654 if (ia->ia_ifp != &sc->sc_if && 1655 ia->ia_ifp->if_type != IFT_CARP && 1656 (ia->ia_ifp->if_flags & IFF_MULTICAST) && 1657 (sin->sin_addr.s_addr & ia->ia_subnetmask) == 1658 ia->ia_subnet) { 1659 if (!ia_if) 1660 ia_if = ia; 1661 } 1662 } 1663 1664 if (ia_if) { 1665 ia = ia_if; 1666 if (ifp) { 1667 if (ifp != ia->ia_ifp) 1668 return (EADDRNOTAVAIL); 1669 } else { 1670 ifp = ia->ia_ifp; 1671 } 1672 } 1673 1674 if ((error = carp_set_ifp(sc, ifp))) 1675 return (error); 1676 1677 if (sc->sc_carpdev == NULL) 1678 return (EADDRNOTAVAIL); 1679 1680 if (sc->sc_naddrs == 0 && (error = carp_join_multicast(sc)) != 0) 1681 return (error); 1682 1683 sc->sc_naddrs++; 1684 if (sc->sc_carpdev != NULL) 1685 sc->sc_if.if_flags |= IFF_UP; 1686 1687 carp_set_state(sc, INIT); 1688 carp_setrun(sc, 0); 1689 1690 /* 1691 * Hook if_addrhooks so that we get a callback after in_ifinit has run, 1692 * to correct any inappropriate routes that it inserted. 1693 */ 1694 if (sc->ah_cookie == 0) { 1695 /* XXX link address hook */ 1696 } 1697 1698 return (0); 1699 } 1700 1701 int 1702 carp_join_multicast(struct carp_softc *sc) 1703 { 1704 struct ip_moptions *imo = &sc->sc_imo, tmpimo; 1705 struct in_addr addr; 1706 1707 bzero(&tmpimo, sizeof(tmpimo)); 1708 addr.s_addr = INADDR_CARP_GROUP; 1709 if ((tmpimo.imo_membership[0] = 1710 in_addmulti(&addr, &sc->sc_if)) == NULL) { 1711 return (ENOBUFS); 1712 } 1713 1714 imo->imo_membership[0] = tmpimo.imo_membership[0]; 1715 imo->imo_num_memberships = 1; 1716 imo->imo_multicast_ifp = &sc->sc_if; 1717 imo->imo_multicast_ttl = CARP_DFLTTL; 1718 imo->imo_multicast_loop = 0; 1719 return (0); 1720 } 1721 1722 1723 #ifdef INET6 1724 int 1725 carp_set_addr6(struct carp_softc *sc, struct sockaddr_in6 *sin6) 1726 { 1727 struct ifnet *ifp = sc->sc_carpdev; 1728 struct in6_ifaddr *ia, *ia_if; 1729 int error = 0; 1730 1731 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) { 1732 if (!(sc->sc_if.if_flags & IFF_UP)) 1733 carp_set_state(sc, INIT); 1734 if (sc->sc_naddrs6) 1735 sc->sc_if.if_flags |= IFF_UP; 1736 carp_setrun(sc, 0); 1737 return (0); 1738 } 1739 1740 /* we have to do this by hand to ensure we don't match on ourselves */ 1741 ia_if = NULL; 1742 for (ia = in6_ifaddr; ia; ia = ia->ia_next) { 1743 int i; 1744 1745 for (i = 0; i < 4; i++) { 1746 if ((sin6->sin6_addr.s6_addr32[i] & 1747 ia->ia_prefixmask.sin6_addr.s6_addr32[i]) != 1748 (ia->ia_addr.sin6_addr.s6_addr32[i] & 1749 ia->ia_prefixmask.sin6_addr.s6_addr32[i])) 1750 break; 1751 } 1752 /* and, yeah, we need a multicast-capable iface too */ 1753 if (ia->ia_ifp != &sc->sc_if && 1754 ia->ia_ifp->if_type != IFT_CARP && 1755 (ia->ia_ifp->if_flags & IFF_MULTICAST) && 1756 (i == 4)) { 1757 if (!ia_if) 1758 ia_if = ia; 1759 } 1760 } 1761 1762 if (ia_if) { 1763 ia = ia_if; 1764 if (sc->sc_carpdev) { 1765 if (sc->sc_carpdev != ia->ia_ifp) 1766 return (EADDRNOTAVAIL); 1767 } else { 1768 ifp = ia->ia_ifp; 1769 } 1770 } 1771 1772 if ((error = carp_set_ifp(sc, ifp))) 1773 return (error); 1774 1775 if (sc->sc_carpdev == NULL) 1776 return (EADDRNOTAVAIL); 1777 1778 if (sc->sc_naddrs6 == 0 && (error = carp_join_multicast6(sc)) != 0) 1779 return (error); 1780 1781 sc->sc_naddrs6++; 1782 if (sc->sc_carpdev != NULL) 1783 sc->sc_if.if_flags |= IFF_UP; 1784 carp_set_state(sc, INIT); 1785 carp_setrun(sc, 0); 1786 1787 return (0); 1788 } 1789 1790 int 1791 carp_join_multicast6(struct carp_softc *sc) 1792 { 1793 struct in6_multi_mship *imm, *imm2; 1794 struct ip6_moptions *im6o = &sc->sc_im6o; 1795 struct sockaddr_in6 addr6; 1796 int error; 1797 1798 /* Join IPv6 CARP multicast group */ 1799 bzero(&addr6, sizeof(addr6)); 1800 addr6.sin6_family = AF_INET6; 1801 addr6.sin6_len = sizeof(addr6); 1802 addr6.sin6_addr.s6_addr16[0] = htons(0xff02); 1803 addr6.sin6_addr.s6_addr16[1] = htons(sc->sc_if.if_index); 1804 addr6.sin6_addr.s6_addr8[15] = 0x12; 1805 if ((imm = in6_joingroup(&sc->sc_if, 1806 &addr6.sin6_addr, &error, 0)) == NULL) { 1807 return (error); 1808 } 1809 /* join solicited multicast address */ 1810 bzero(&addr6.sin6_addr, sizeof(addr6.sin6_addr)); 1811 addr6.sin6_addr.s6_addr16[0] = htons(0xff02); 1812 addr6.sin6_addr.s6_addr16[1] = htons(sc->sc_if.if_index); 1813 addr6.sin6_addr.s6_addr32[1] = 0; 1814 addr6.sin6_addr.s6_addr32[2] = htonl(1); 1815 addr6.sin6_addr.s6_addr32[3] = 0; 1816 addr6.sin6_addr.s6_addr8[12] = 0xff; 1817 if ((imm2 = in6_joingroup(&sc->sc_if, 1818 &addr6.sin6_addr, &error, 0)) == NULL) { 1819 in6_leavegroup(imm); 1820 return (error); 1821 } 1822 1823 /* apply v6 multicast membership */ 1824 im6o->im6o_multicast_ifp = &sc->sc_if; 1825 if (imm) 1826 LIST_INSERT_HEAD(&im6o->im6o_memberships, imm, 1827 i6mm_chain); 1828 if (imm2) 1829 LIST_INSERT_HEAD(&im6o->im6o_memberships, imm2, 1830 i6mm_chain); 1831 1832 return (0); 1833 } 1834 1835 #endif /* INET6 */ 1836 1837 int 1838 carp_ioctl(struct ifnet *ifp, u_long cmd, caddr_t addr) 1839 { 1840 struct lwp *l = curlwp; /* XXX */ 1841 struct carp_softc *sc = ifp->if_softc, *vr; 1842 struct carpreq carpr; 1843 struct ifaddr *ifa; 1844 struct ifreq *ifr; 1845 struct ifaliasreq *ifra; 1846 struct ifnet *cdev = NULL; 1847 int error = 0; 1848 1849 ifa = (struct ifaddr *)addr; 1850 ifra = (struct ifaliasreq *)addr; 1851 ifr = (struct ifreq *)addr; 1852 1853 switch (cmd) { 1854 case SIOCSIFADDR: 1855 switch (ifa->ifa_addr->sa_family) { 1856 #ifdef INET 1857 case AF_INET: 1858 sc->sc_if.if_flags |= IFF_UP; 1859 bcopy(ifa->ifa_addr, ifa->ifa_dstaddr, 1860 sizeof(struct sockaddr)); 1861 error = carp_set_addr(sc, satosin(ifa->ifa_addr)); 1862 break; 1863 #endif /* INET */ 1864 #ifdef INET6 1865 case AF_INET6: 1866 sc->sc_if.if_flags|= IFF_UP; 1867 error = carp_set_addr6(sc, satosin6(ifa->ifa_addr)); 1868 break; 1869 #endif /* INET6 */ 1870 default: 1871 error = EAFNOSUPPORT; 1872 break; 1873 } 1874 break; 1875 1876 case SIOCAIFADDR: 1877 switch (ifa->ifa_addr->sa_family) { 1878 #ifdef INET 1879 case AF_INET: 1880 sc->sc_if.if_flags |= IFF_UP; 1881 bcopy(ifa->ifa_addr, ifa->ifa_dstaddr, 1882 sizeof(struct sockaddr)); 1883 error = carp_set_addr(sc, satosin(&ifra->ifra_addr)); 1884 break; 1885 #endif /* INET */ 1886 #ifdef INET6 1887 case AF_INET6: 1888 sc->sc_if.if_flags |= IFF_UP; 1889 error = carp_set_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_state != INIT && !(ifr->ifr_flags & IFF_UP)) { 1900 callout_stop(&sc->sc_ad_tmo); 1901 callout_stop(&sc->sc_md_tmo); 1902 callout_stop(&sc->sc_md6_tmo); 1903 if (sc->sc_state == MASTER) { 1904 /* we need the interface up to bow out */ 1905 sc->sc_if.if_flags |= IFF_UP; 1906 sc->sc_bow_out = 1; 1907 carp_send_ad(sc); 1908 } 1909 sc->sc_if.if_flags &= ~IFF_UP; 1910 carp_set_state(sc, INIT); 1911 carp_setrun(sc, 0); 1912 } else if (sc->sc_state == INIT && (ifr->ifr_flags & IFF_UP)) { 1913 sc->sc_if.if_flags |= IFF_UP; 1914 carp_setrun(sc, 0); 1915 } 1916 break; 1917 1918 case SIOCSVH: 1919 if (l == 0 || (error = kauth_authorize_generic(l->l_cred, 1920 KAUTH_GENERIC_ISSUSER, &l->l_acflag))) 1921 break; 1922 if ((error = copyin(ifr->ifr_data, &carpr, sizeof carpr))) 1923 break; 1924 error = 1; 1925 if (carpr.carpr_carpdev[0] != '\0' && 1926 (cdev = ifunit(carpr.carpr_carpdev)) == NULL) 1927 return (EINVAL); 1928 if ((error = carp_set_ifp(sc, cdev))) 1929 return (error); 1930 if (sc->sc_state != INIT && carpr.carpr_state != sc->sc_state) { 1931 switch (carpr.carpr_state) { 1932 case BACKUP: 1933 callout_stop(&sc->sc_ad_tmo); 1934 carp_set_state(sc, BACKUP); 1935 carp_setrun(sc, 0); 1936 carp_setroute(sc, RTM_DELETE); 1937 break; 1938 case MASTER: 1939 carp_master_down(sc); 1940 break; 1941 default: 1942 break; 1943 } 1944 } 1945 if (carpr.carpr_vhid > 0) { 1946 if (carpr.carpr_vhid > 255) { 1947 error = EINVAL; 1948 break; 1949 } 1950 if (sc->sc_carpdev) { 1951 struct carp_if *cif; 1952 cif = (struct carp_if *)sc->sc_carpdev->if_carp; 1953 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) 1954 if (vr != sc && 1955 vr->sc_vhid == carpr.carpr_vhid) 1956 return (EINVAL); 1957 } 1958 sc->sc_vhid = carpr.carpr_vhid; 1959 carp_set_enaddr(sc); 1960 carp_set_state(sc, INIT); 1961 error--; 1962 } 1963 if (carpr.carpr_advbase > 0 || carpr.carpr_advskew > 0) { 1964 if (carpr.carpr_advskew > 254) { 1965 error = EINVAL; 1966 break; 1967 } 1968 if (carpr.carpr_advbase > 255) { 1969 error = EINVAL; 1970 break; 1971 } 1972 sc->sc_advbase = carpr.carpr_advbase; 1973 sc->sc_advskew = carpr.carpr_advskew; 1974 error--; 1975 } 1976 bcopy(carpr.carpr_key, sc->sc_key, sizeof(sc->sc_key)); 1977 if (error > 0) 1978 error = EINVAL; 1979 else { 1980 error = 0; 1981 carp_setrun(sc, 0); 1982 } 1983 break; 1984 1985 case SIOCGVH: 1986 bzero(&carpr, sizeof(carpr)); 1987 if (sc->sc_carpdev != NULL) 1988 strlcpy(carpr.carpr_carpdev, sc->sc_carpdev->if_xname, 1989 IFNAMSIZ); 1990 carpr.carpr_state = sc->sc_state; 1991 carpr.carpr_vhid = sc->sc_vhid; 1992 carpr.carpr_advbase = sc->sc_advbase; 1993 carpr.carpr_advskew = sc->sc_advskew; 1994 1995 if (l != 0 || !(error = kauth_authorize_generic(l->l_cred, 1996 KAUTH_GENERIC_ISSUSER, &l->l_acflag))) 1997 bcopy(sc->sc_key, carpr.carpr_key, 1998 sizeof(carpr.carpr_key)); 1999 error = copyout(&carpr, ifr->ifr_data, sizeof(carpr)); 2000 break; 2001 2002 case SIOCADDMULTI: 2003 error = carp_ether_addmulti(sc, ifr); 2004 break; 2005 2006 case SIOCDELMULTI: 2007 error = carp_ether_delmulti(sc, ifr); 2008 break; 2009 2010 default: 2011 error = EINVAL; 2012 } 2013 2014 carp_hmac_prepare(sc); 2015 return (error); 2016 } 2017 2018 2019 /* 2020 * Start output on carp interface. This function should never be called. 2021 */ 2022 void 2023 carp_start(struct ifnet *ifp) 2024 { 2025 #ifdef DEBUG 2026 printf("%s: start called\n", ifp->if_xname); 2027 #endif 2028 } 2029 2030 int 2031 carp_output(struct ifnet *ifp, struct mbuf *m, struct sockaddr *sa, 2032 struct rtentry *rt) 2033 { 2034 struct carp_softc *sc = ((struct carp_softc *)ifp->if_softc); 2035 2036 if (sc->sc_carpdev != NULL && sc->sc_state == MASTER) { 2037 return (sc->sc_carpdev->if_output(ifp, m, sa, rt)); 2038 } else { 2039 m_freem(m); 2040 return (ENETUNREACH); 2041 } 2042 } 2043 2044 void 2045 carp_set_state(struct carp_softc *sc, int state) 2046 { 2047 if (sc->sc_state == state) 2048 return; 2049 2050 sc->sc_state = state; 2051 switch (state) { 2052 case BACKUP: 2053 sc->sc_if.if_link_state = LINK_STATE_DOWN; 2054 break; 2055 case MASTER: 2056 sc->sc_if.if_link_state = LINK_STATE_UP; 2057 break; 2058 default: 2059 sc->sc_if.if_link_state = LINK_STATE_UNKNOWN; 2060 break; 2061 } 2062 rt_ifmsg(&sc->sc_if); 2063 } 2064 2065 void 2066 carp_carpdev_state(void *v) 2067 { 2068 struct carp_if *cif; 2069 struct carp_softc *sc; 2070 struct ifnet *ifp = v; 2071 2072 if (ifp->if_type == IFT_CARP) 2073 return; 2074 2075 cif = (struct carp_if *)ifp->if_carp; 2076 2077 TAILQ_FOREACH(sc, &cif->vhif_vrs, sc_list) { 2078 int suppressed = sc->sc_suppress; 2079 2080 if (sc->sc_carpdev->if_link_state == LINK_STATE_DOWN || 2081 !(sc->sc_carpdev->if_flags & IFF_UP)) { 2082 sc->sc_if.if_flags &= ~IFF_RUNNING; 2083 callout_stop(&sc->sc_ad_tmo); 2084 callout_stop(&sc->sc_md_tmo); 2085 callout_stop(&sc->sc_md6_tmo); 2086 carp_set_state(sc, INIT); 2087 sc->sc_suppress = 1; 2088 carp_setrun(sc, 0); 2089 if (!suppressed) { 2090 carp_suppress_preempt++; 2091 if (carp_suppress_preempt == 1) 2092 carp_send_ad_all(); 2093 } 2094 } else { 2095 carp_set_state(sc, INIT); 2096 sc->sc_suppress = 0; 2097 carp_setrun(sc, 0); 2098 if (suppressed) 2099 carp_suppress_preempt--; 2100 } 2101 } 2102 } 2103 2104 int 2105 carp_ether_addmulti(struct carp_softc *sc, struct ifreq *ifr) 2106 { 2107 struct ifnet *ifp; 2108 struct carp_mc_entry *mc; 2109 u_int8_t addrlo[ETHER_ADDR_LEN], addrhi[ETHER_ADDR_LEN]; 2110 int error; 2111 2112 ifp = sc->sc_carpdev; 2113 if (ifp == NULL) 2114 return (EINVAL); 2115 2116 error = ether_addmulti(ifr, &sc->sc_ac); 2117 if (error != ENETRESET) 2118 return (error); 2119 2120 /* 2121 * This is new multicast address. We have to tell parent 2122 * about it. Also, remember this multicast address so that 2123 * we can delete them on unconfigure. 2124 */ 2125 MALLOC(mc, struct carp_mc_entry *, sizeof(struct carp_mc_entry), 2126 M_DEVBUF, M_NOWAIT); 2127 if (mc == NULL) { 2128 error = ENOMEM; 2129 goto alloc_failed; 2130 } 2131 2132 /* 2133 * As ether_addmulti() returns ENETRESET, following two 2134 * statement shouldn't fail. 2135 */ 2136 (void)ether_multiaddr(&ifr->ifr_addr, addrlo, addrhi); 2137 ETHER_LOOKUP_MULTI(addrlo, addrhi, &sc->sc_ac, mc->mc_enm); 2138 memcpy(&mc->mc_addr, &ifr->ifr_addr, ifr->ifr_addr.sa_len); 2139 LIST_INSERT_HEAD(&sc->carp_mc_listhead, mc, mc_entries); 2140 2141 error = (*ifp->if_ioctl)(ifp, SIOCADDMULTI, (caddr_t)ifr); 2142 if (error != 0) 2143 goto ioctl_failed; 2144 2145 return (error); 2146 2147 ioctl_failed: 2148 LIST_REMOVE(mc, mc_entries); 2149 FREE(mc, M_DEVBUF); 2150 alloc_failed: 2151 (void)ether_delmulti(ifr, &sc->sc_ac); 2152 2153 return (error); 2154 } 2155 2156 int 2157 carp_ether_delmulti(struct carp_softc *sc, struct ifreq *ifr) 2158 { 2159 struct ifnet *ifp; 2160 struct ether_multi *enm; 2161 struct carp_mc_entry *mc; 2162 u_int8_t addrlo[ETHER_ADDR_LEN], addrhi[ETHER_ADDR_LEN]; 2163 int error; 2164 2165 ifp = sc->sc_carpdev; 2166 if (ifp == NULL) 2167 return (EINVAL); 2168 2169 /* 2170 * Find a key to lookup carp_mc_entry. We have to do this 2171 * before calling ether_delmulti for obvious reason. 2172 */ 2173 if ((error = ether_multiaddr(&ifr->ifr_addr, addrlo, addrhi)) != 0) 2174 return (error); 2175 ETHER_LOOKUP_MULTI(addrlo, addrhi, &sc->sc_ac, enm); 2176 if (enm == NULL) 2177 return (EINVAL); 2178 2179 LIST_FOREACH(mc, &sc->carp_mc_listhead, mc_entries) 2180 if (mc->mc_enm == enm) 2181 break; 2182 2183 /* We won't delete entries we didn't add */ 2184 if (mc == NULL) 2185 return (EINVAL); 2186 2187 error = ether_delmulti(ifr, &sc->sc_ac); 2188 if (error != ENETRESET) 2189 return (error); 2190 2191 /* We no longer use this multicast address. Tell parent so. */ 2192 error = (*ifp->if_ioctl)(ifp, SIOCDELMULTI, (caddr_t)ifr); 2193 if (error == 0) { 2194 /* And forget about this address. */ 2195 LIST_REMOVE(mc, mc_entries); 2196 FREE(mc, M_DEVBUF); 2197 } else 2198 (void)ether_addmulti(ifr, &sc->sc_ac); 2199 return (error); 2200 } 2201 2202 /* 2203 * Delete any multicast address we have asked to add from parent 2204 * interface. Called when the carp is being unconfigured. 2205 */ 2206 void 2207 carp_ether_purgemulti(struct carp_softc *sc) 2208 { 2209 struct ifnet *ifp = sc->sc_carpdev; /* Parent. */ 2210 struct carp_mc_entry *mc; 2211 union { 2212 struct ifreq ifreq; 2213 struct { 2214 char ifr_name[IFNAMSIZ]; 2215 struct sockaddr_storage ifr_ss; 2216 } ifreq_storage; 2217 } u; 2218 struct ifreq *ifr = &u.ifreq; 2219 2220 if (ifp == NULL) 2221 return; 2222 2223 memcpy(ifr->ifr_name, ifp->if_xname, IFNAMSIZ); 2224 while ((mc = LIST_FIRST(&sc->carp_mc_listhead)) != NULL) { 2225 memcpy(&ifr->ifr_addr, &mc->mc_addr, mc->mc_addr.ss_len); 2226 (void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, (caddr_t)ifr); 2227 LIST_REMOVE(mc, mc_entries); 2228 FREE(mc, M_DEVBUF); 2229 } 2230 } 2231 2232 SYSCTL_SETUP(sysctl_net_inet_carp_setup, "sysctl net.inet.carp subtree setup") 2233 { 2234 2235 sysctl_createv(clog, 0, NULL, NULL, 2236 CTLFLAG_PERMANENT, 2237 CTLTYPE_NODE, "net", NULL, 2238 NULL, 0, NULL, 0, 2239 CTL_NET, CTL_EOL); 2240 sysctl_createv(clog, 0, NULL, NULL, 2241 CTLFLAG_PERMANENT, 2242 CTLTYPE_NODE, "inet", NULL, 2243 NULL, 0, NULL, 0, 2244 CTL_NET, PF_INET, CTL_EOL); 2245 sysctl_createv(clog, 0, NULL, NULL, 2246 CTLFLAG_PERMANENT, 2247 CTLTYPE_NODE, "carp", 2248 SYSCTL_DESCR("CARP related settings"), 2249 NULL, 0, NULL, 0, 2250 CTL_NET, PF_INET, IPPROTO_CARP, CTL_EOL); 2251 2252 sysctl_createv(clog, 0, NULL, NULL, 2253 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2254 CTLTYPE_INT, "preempt", 2255 SYSCTL_DESCR("Enable CARP Preempt"), 2256 NULL, 0, &carp_opts[CARPCTL_PREEMPT], 0, 2257 CTL_NET, PF_INET, IPPROTO_CARP, 2258 CTL_CREATE, CTL_EOL); 2259 sysctl_createv(clog, 0, NULL, NULL, 2260 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2261 CTLTYPE_INT, "arpbalance", 2262 SYSCTL_DESCR("Enable ARP balancing"), 2263 NULL, 0, &carp_opts[CARPCTL_ARPBALANCE], 0, 2264 CTL_NET, PF_INET, IPPROTO_CARP, 2265 CTL_CREATE, CTL_EOL); 2266 sysctl_createv(clog, 0, NULL, NULL, 2267 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2268 CTLTYPE_INT, "allow", 2269 SYSCTL_DESCR("Enable CARP"), 2270 NULL, 0, &carp_opts[CARPCTL_ALLOW], 0, 2271 CTL_NET, PF_INET, IPPROTO_CARP, 2272 CTL_CREATE, CTL_EOL); 2273 sysctl_createv(clog, 0, NULL, NULL, 2274 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2275 CTLTYPE_INT, "log", 2276 SYSCTL_DESCR("CARP logging"), 2277 NULL, 0, &carp_opts[CARPCTL_LOG], 0, 2278 CTL_NET, PF_INET, IPPROTO_CARP, 2279 CTL_CREATE, CTL_EOL); 2280 sysctl_createv(clog, 0, NULL, NULL, 2281 CTLFLAG_PERMANENT, 2282 CTLTYPE_STRUCT, "stats", 2283 SYSCTL_DESCR("CARP statistics"), 2284 NULL, 0, &carpstats, sizeof(carpstats), 2285 CTL_NET, PF_INET, IPPROTO_CARP, CARPCTL_STATS, 2286 CTL_EOL); 2287 } 2288