ip_carp.c revision 1.13
1 /* $NetBSD: ip_carp.c,v 1.13 2007/07/09 21:11:11 ad 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, void *); 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 /* 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) 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, 0); 760 callout_init(&sc->sc_md_tmo, 0); 761 callout_init(&sc->sc_md6_tmo, 0); 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 struct carp_softc *sc = ifp->if_softc; 798 799 carpdetach(ifp->if_softc); 800 ether_ifdetach(ifp); 801 if_detach(ifp); 802 callout_destroy(&sc->sc_ad_tmo); 803 callout_destroy(&sc->sc_md_tmo); 804 callout_destroy(&sc->sc_md6_tmo); 805 free(ifp->if_softc, M_DEVBUF); 806 807 return (0); 808 } 809 810 void 811 carpdetach(struct carp_softc *sc) 812 { 813 struct carp_if *cif; 814 int s; 815 816 callout_stop(&sc->sc_ad_tmo); 817 callout_stop(&sc->sc_md_tmo); 818 callout_stop(&sc->sc_md6_tmo); 819 820 if (sc->sc_suppress) 821 carp_suppress_preempt--; 822 sc->sc_suppress = 0; 823 824 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) 825 carp_suppress_preempt--; 826 sc->sc_sendad_errors = 0; 827 828 carp_set_state(sc, INIT); 829 sc->sc_if.if_flags &= ~IFF_UP; 830 carp_setrun(sc, 0); 831 carp_multicast_cleanup(sc); 832 833 s = splnet(); 834 if (sc->sc_carpdev != NULL) { 835 /* XXX linkstatehook removal */ 836 cif = (struct carp_if *)sc->sc_carpdev->if_carp; 837 TAILQ_REMOVE(&cif->vhif_vrs, sc, sc_list); 838 if (!--cif->vhif_nvrs) { 839 ifpromisc(sc->sc_carpdev, 0); 840 sc->sc_carpdev->if_carp = NULL; 841 FREE(cif, M_IFADDR); 842 } 843 } 844 sc->sc_carpdev = NULL; 845 splx(s); 846 } 847 848 /* Detach an interface from the carp. */ 849 void 850 carp_ifdetach(struct ifnet *ifp) 851 { 852 struct carp_softc *sc, *nextsc; 853 struct carp_if *cif = (struct carp_if *)ifp->if_carp; 854 855 for (sc = TAILQ_FIRST(&cif->vhif_vrs); sc; sc = nextsc) { 856 nextsc = TAILQ_NEXT(sc, sc_list); 857 carpdetach(sc); 858 } 859 } 860 861 int 862 carp_prepare_ad(struct mbuf *m, struct carp_softc *sc, 863 struct carp_header *ch) 864 { 865 if (sc->sc_init_counter) { 866 /* this could also be seconds since unix epoch */ 867 sc->sc_counter = arc4random(); 868 sc->sc_counter = sc->sc_counter << 32; 869 sc->sc_counter += arc4random(); 870 } else 871 sc->sc_counter++; 872 873 ch->carp_counter[0] = htonl((sc->sc_counter>>32)&0xffffffff); 874 ch->carp_counter[1] = htonl(sc->sc_counter&0xffffffff); 875 876 carp_hmac_generate(sc, ch->carp_counter, ch->carp_md); 877 878 return (0); 879 } 880 881 void 882 carp_send_ad_all(void) 883 { 884 struct ifnet *ifp; 885 struct carp_if *cif; 886 struct carp_softc *vh; 887 888 TAILQ_FOREACH(ifp, &ifnet, if_list) { 889 if (ifp->if_carp == NULL || ifp->if_type == IFT_CARP) 890 continue; 891 892 cif = (struct carp_if *)ifp->if_carp; 893 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 894 if ((vh->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) == 895 (IFF_UP|IFF_RUNNING) && vh->sc_state == MASTER) 896 carp_send_ad(vh); 897 } 898 } 899 } 900 901 902 void 903 carp_send_ad(void *v) 904 { 905 struct carp_header ch; 906 struct timeval tv; 907 struct carp_softc *sc = v; 908 struct carp_header *ch_ptr; 909 struct mbuf *m; 910 int error, len, advbase, advskew, s; 911 struct ifaddr *ifa; 912 struct sockaddr sa; 913 914 s = splsoftnet(); 915 916 advbase = advskew = 0; /* Sssssh compiler */ 917 if (sc->sc_carpdev == NULL) { 918 sc->sc_if.if_oerrors++; 919 goto retry_later; 920 } 921 922 /* bow out if we've gone to backup (the carp interface is going down) */ 923 if (sc->sc_bow_out) { 924 sc->sc_bow_out = 0; 925 advbase = 255; 926 advskew = 255; 927 } else { 928 advbase = sc->sc_advbase; 929 if (!carp_suppress_preempt || sc->sc_advskew > 240) 930 advskew = sc->sc_advskew; 931 else 932 advskew = 240; 933 tv.tv_sec = advbase; 934 tv.tv_usec = advskew * 1000000 / 256; 935 } 936 937 ch.carp_version = CARP_VERSION; 938 ch.carp_type = CARP_ADVERTISEMENT; 939 ch.carp_vhid = sc->sc_vhid; 940 ch.carp_advbase = advbase; 941 ch.carp_advskew = advskew; 942 ch.carp_authlen = 7; /* XXX DEFINE */ 943 ch.carp_pad1 = 0; /* must be zero */ 944 ch.carp_cksum = 0; 945 946 947 #ifdef INET 948 if (sc->sc_naddrs) { 949 struct ip *ip; 950 951 MGETHDR(m, M_DONTWAIT, MT_HEADER); 952 if (m == NULL) { 953 sc->sc_if.if_oerrors++; 954 carpstats.carps_onomem++; 955 /* XXX maybe less ? */ 956 goto retry_later; 957 } 958 len = sizeof(*ip) + sizeof(ch); 959 m->m_pkthdr.len = len; 960 m->m_pkthdr.rcvif = NULL; 961 m->m_len = len; 962 MH_ALIGN(m, m->m_len); 963 m->m_flags |= M_MCAST; 964 ip = mtod(m, struct ip *); 965 ip->ip_v = IPVERSION; 966 ip->ip_hl = sizeof(*ip) >> 2; 967 ip->ip_tos = IPTOS_LOWDELAY; 968 ip->ip_len = htons(len); 969 ip->ip_id = htons(ip_randomid()); 970 ip->ip_off = htons(IP_DF); 971 ip->ip_ttl = CARP_DFLTTL; 972 ip->ip_p = IPPROTO_CARP; 973 ip->ip_sum = 0; 974 975 bzero(&sa, sizeof(sa)); 976 sa.sa_family = AF_INET; 977 ifa = ifaof_ifpforaddr(&sa, sc->sc_carpdev); 978 if (ifa == NULL) 979 ip->ip_src.s_addr = 0; 980 else 981 ip->ip_src.s_addr = 982 ifatoia(ifa)->ia_addr.sin_addr.s_addr; 983 ip->ip_dst.s_addr = INADDR_CARP_GROUP; 984 985 ch_ptr = (struct carp_header *)(&ip[1]); 986 bcopy(&ch, ch_ptr, sizeof(ch)); 987 if (carp_prepare_ad(m, sc, ch_ptr)) 988 goto retry_later; 989 990 m->m_data += sizeof(*ip); 991 ch_ptr->carp_cksum = carp_cksum(m, len - sizeof(*ip)); 992 m->m_data -= sizeof(*ip); 993 994 microtime(&sc->sc_if.if_lastchange); 995 sc->sc_if.if_opackets++; 996 sc->sc_if.if_obytes += len; 997 carpstats.carps_opackets++; 998 999 error = ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo, 1000 NULL); 1001 if (error) { 1002 if (error == ENOBUFS) 1003 carpstats.carps_onomem++; 1004 else 1005 CARP_LOG(sc, ("ip_output failed: %d", error)); 1006 sc->sc_if.if_oerrors++; 1007 if (sc->sc_sendad_errors < INT_MAX) 1008 sc->sc_sendad_errors++; 1009 if (sc->sc_sendad_errors == CARP_SENDAD_MAX_ERRORS) { 1010 carp_suppress_preempt++; 1011 if (carp_suppress_preempt == 1) 1012 carp_send_ad_all(); 1013 } 1014 sc->sc_sendad_success = 0; 1015 } else { 1016 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) { 1017 if (++sc->sc_sendad_success >= 1018 CARP_SENDAD_MIN_SUCCESS) { 1019 carp_suppress_preempt--; 1020 sc->sc_sendad_errors = 0; 1021 } 1022 } else 1023 sc->sc_sendad_errors = 0; 1024 } 1025 } 1026 #endif /* INET */ 1027 #ifdef INET6 1028 if (sc->sc_naddrs6) { 1029 struct ip6_hdr *ip6; 1030 1031 MGETHDR(m, M_DONTWAIT, MT_HEADER); 1032 if (m == NULL) { 1033 sc->sc_if.if_oerrors++; 1034 carpstats.carps_onomem++; 1035 /* XXX maybe less ? */ 1036 goto retry_later; 1037 } 1038 len = sizeof(*ip6) + sizeof(ch); 1039 m->m_pkthdr.len = len; 1040 m->m_pkthdr.rcvif = NULL; 1041 m->m_len = len; 1042 MH_ALIGN(m, m->m_len); 1043 m->m_flags |= M_MCAST; 1044 ip6 = mtod(m, struct ip6_hdr *); 1045 bzero(ip6, sizeof(*ip6)); 1046 ip6->ip6_vfc |= IPV6_VERSION; 1047 ip6->ip6_hlim = CARP_DFLTTL; 1048 ip6->ip6_nxt = IPPROTO_CARP; 1049 1050 /* set the source address */ 1051 bzero(&sa, sizeof(sa)); 1052 sa.sa_family = AF_INET6; 1053 ifa = ifaof_ifpforaddr(&sa, sc->sc_carpdev); 1054 if (ifa == NULL) /* This should never happen with IPv6 */ 1055 bzero(&ip6->ip6_src, sizeof(struct in6_addr)); 1056 else 1057 bcopy(ifatoia6(ifa)->ia_addr.sin6_addr.s6_addr, 1058 &ip6->ip6_src, sizeof(struct in6_addr)); 1059 /* set the multicast destination */ 1060 1061 ip6->ip6_dst.s6_addr8[0] = 0xff; 1062 ip6->ip6_dst.s6_addr8[1] = 0x02; 1063 ip6->ip6_dst.s6_addr8[15] = 0x12; 1064 1065 ch_ptr = (struct carp_header *)(&ip6[1]); 1066 bcopy(&ch, ch_ptr, sizeof(ch)); 1067 if (carp_prepare_ad(m, sc, ch_ptr)) 1068 goto retry_later; 1069 1070 m->m_data += sizeof(*ip6); 1071 ch_ptr->carp_cksum = carp_cksum(m, len - sizeof(*ip6)); 1072 m->m_data -= sizeof(*ip6); 1073 1074 microtime(&sc->sc_if.if_lastchange); 1075 sc->sc_if.if_opackets++; 1076 sc->sc_if.if_obytes += len; 1077 carpstats.carps_opackets6++; 1078 1079 error = ip6_output(m, NULL, NULL, 0, &sc->sc_im6o, NULL, NULL); 1080 if (error) { 1081 if (error == ENOBUFS) 1082 carpstats.carps_onomem++; 1083 else 1084 CARP_LOG(sc, ("ip6_output failed: %d", error)); 1085 sc->sc_if.if_oerrors++; 1086 if (sc->sc_sendad_errors < INT_MAX) 1087 sc->sc_sendad_errors++; 1088 if (sc->sc_sendad_errors == CARP_SENDAD_MAX_ERRORS) { 1089 carp_suppress_preempt++; 1090 if (carp_suppress_preempt == 1) 1091 carp_send_ad_all(); 1092 } 1093 sc->sc_sendad_success = 0; 1094 } else { 1095 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) { 1096 if (++sc->sc_sendad_success >= 1097 CARP_SENDAD_MIN_SUCCESS) { 1098 carp_suppress_preempt--; 1099 sc->sc_sendad_errors = 0; 1100 } 1101 } else 1102 sc->sc_sendad_errors = 0; 1103 } 1104 } 1105 #endif /* INET6 */ 1106 1107 retry_later: 1108 splx(s); 1109 if (advbase != 255 || advskew != 255) 1110 callout_schedule(&sc->sc_ad_tmo, tvtohz(&tv)); 1111 } 1112 1113 /* 1114 * Broadcast a gratuitous ARP request containing 1115 * the virtual router MAC address for each IP address 1116 * associated with the virtual router. 1117 */ 1118 void 1119 carp_send_arp(struct carp_softc *sc) 1120 { 1121 struct ifaddr *ifa; 1122 struct in_addr *in; 1123 int s = splsoftnet(); 1124 1125 TAILQ_FOREACH(ifa, &sc->sc_if.if_addrlist, ifa_list) { 1126 1127 if (ifa->ifa_addr->sa_family != AF_INET) 1128 continue; 1129 1130 in = &ifatoia(ifa)->ia_addr.sin_addr; 1131 arprequest(sc->sc_carpdev, in, in, LLADDR(sc->sc_if.if_sadl)); 1132 DELAY(1000); /* XXX */ 1133 } 1134 splx(s); 1135 } 1136 1137 #ifdef INET6 1138 void 1139 carp_send_na(struct carp_softc *sc) 1140 { 1141 struct ifaddr *ifa; 1142 struct in6_addr *in6; 1143 static struct in6_addr mcast = IN6ADDR_LINKLOCAL_ALLNODES_INIT; 1144 int s = splsoftnet(); 1145 1146 TAILQ_FOREACH(ifa, &sc->sc_if.if_addrlist, ifa_list) { 1147 1148 if (ifa->ifa_addr->sa_family != AF_INET6) 1149 continue; 1150 1151 in6 = &ifatoia6(ifa)->ia_addr.sin6_addr; 1152 nd6_na_output(sc->sc_carpdev, &mcast, in6, 1153 ND_NA_FLAG_OVERRIDE, 1, NULL); 1154 DELAY(1000); /* XXX */ 1155 } 1156 splx(s); 1157 } 1158 #endif /* INET6 */ 1159 1160 /* 1161 * Based on bridge_hash() in if_bridge.c 1162 */ 1163 #define mix(a,b,c) \ 1164 do { \ 1165 a -= b; a -= c; a ^= (c >> 13); \ 1166 b -= c; b -= a; b ^= (a << 8); \ 1167 c -= a; c -= b; c ^= (b >> 13); \ 1168 a -= b; a -= c; a ^= (c >> 12); \ 1169 b -= c; b -= a; b ^= (a << 16); \ 1170 c -= a; c -= b; c ^= (b >> 5); \ 1171 a -= b; a -= c; a ^= (c >> 3); \ 1172 b -= c; b -= a; b ^= (a << 10); \ 1173 c -= a; c -= b; c ^= (b >> 15); \ 1174 } while (0) 1175 1176 u_int32_t 1177 carp_hash(struct carp_softc *sc, u_char *src) 1178 { 1179 u_int32_t a = 0x9e3779b9, b = sc->sc_hashkey[0], c = sc->sc_hashkey[1]; 1180 1181 c += sc->sc_key[3] << 24; 1182 c += sc->sc_key[2] << 16; 1183 c += sc->sc_key[1] << 8; 1184 c += sc->sc_key[0]; 1185 b += src[5] << 8; 1186 b += src[4]; 1187 a += src[3] << 24; 1188 a += src[2] << 16; 1189 a += src[1] << 8; 1190 a += src[0]; 1191 1192 mix(a, b, c); 1193 return (c); 1194 } 1195 1196 int 1197 carp_addrcount(struct carp_if *cif, struct in_ifaddr *ia, int type) 1198 { 1199 struct carp_softc *vh; 1200 struct ifaddr *ifa; 1201 int count = 0; 1202 1203 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 1204 if ((type == CARP_COUNT_RUNNING && 1205 (vh->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) == 1206 (IFF_UP|IFF_RUNNING)) || 1207 (type == CARP_COUNT_MASTER && vh->sc_state == MASTER)) { 1208 TAILQ_FOREACH(ifa, &vh->sc_if.if_addrlist, ifa_list) { 1209 if (ifa->ifa_addr->sa_family == AF_INET && 1210 ia->ia_addr.sin_addr.s_addr == 1211 ifatoia(ifa)->ia_addr.sin_addr.s_addr) 1212 count++; 1213 } 1214 } 1215 } 1216 return (count); 1217 } 1218 1219 int 1220 carp_iamatch(struct in_ifaddr *ia, u_char *src, 1221 u_int32_t *count, u_int32_t index) 1222 { 1223 struct carp_softc *sc = ia->ia_ifp->if_softc; 1224 1225 if (carp_opts[CARPCTL_ARPBALANCE]) { 1226 /* 1227 * We use the source ip to decide which virtual host should 1228 * handle the request. If we're master of that virtual host, 1229 * then we respond, otherwise, just drop the arp packet on 1230 * the floor. 1231 */ 1232 1233 /* Count the elegible carp interfaces with this address */ 1234 if (*count == 0) 1235 *count = carp_addrcount( 1236 (struct carp_if *)ia->ia_ifp->if_carpdev->if_carp, 1237 ia, CARP_COUNT_RUNNING); 1238 1239 /* This should never happen, but... */ 1240 if (*count == 0) 1241 return (0); 1242 1243 if (carp_hash(sc, src) % *count == index - 1 && 1244 sc->sc_state == MASTER) { 1245 return (1); 1246 } 1247 } else { 1248 if (sc->sc_state == MASTER) 1249 return (1); 1250 } 1251 1252 return (0); 1253 } 1254 1255 #ifdef INET6 1256 struct ifaddr * 1257 carp_iamatch6(void *v, struct in6_addr *taddr) 1258 { 1259 struct carp_if *cif = v; 1260 struct carp_softc *vh; 1261 struct ifaddr *ifa; 1262 1263 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 1264 TAILQ_FOREACH(ifa, &vh->sc_if.if_addrlist, ifa_list) { 1265 if (IN6_ARE_ADDR_EQUAL(taddr, 1266 &ifatoia6(ifa)->ia_addr.sin6_addr) && 1267 ((vh->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) == 1268 (IFF_UP|IFF_RUNNING)) && vh->sc_state == MASTER) 1269 return (ifa); 1270 } 1271 } 1272 1273 return (NULL); 1274 } 1275 #endif /* INET6 */ 1276 1277 struct ifnet * 1278 carp_ourether(void *v, struct ether_header *eh, u_char iftype, int src) 1279 { 1280 struct carp_if *cif = (struct carp_if *)v; 1281 struct carp_softc *vh; 1282 u_int8_t *ena; 1283 1284 if (src) 1285 ena = (u_int8_t *)&eh->ether_shost; 1286 else 1287 ena = (u_int8_t *)&eh->ether_dhost; 1288 1289 switch (iftype) { 1290 case IFT_ETHER: 1291 case IFT_FDDI: 1292 if (ena[0] || ena[1] || ena[2] != 0x5e || ena[3] || ena[4] != 1) 1293 return (NULL); 1294 break; 1295 case IFT_ISO88025: 1296 if (ena[0] != 3 || ena[1] || ena[4] || ena[5]) 1297 return (NULL); 1298 break; 1299 default: 1300 return (NULL); 1301 break; 1302 } 1303 1304 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) 1305 if ((vh->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) == 1306 (IFF_UP|IFF_RUNNING) && vh->sc_state == MASTER && 1307 !bcmp(ena, LLADDR(vh->sc_if.if_sadl), 1308 ETHER_ADDR_LEN)) { 1309 return (&vh->sc_if); 1310 } 1311 1312 return (NULL); 1313 } 1314 1315 int 1316 carp_input(struct mbuf *m, u_int8_t *shost, u_int8_t *dhost, u_int16_t etype) 1317 { 1318 struct ether_header eh; 1319 struct carp_if *cif = (struct carp_if *)m->m_pkthdr.rcvif->if_carp; 1320 struct ifnet *ifp; 1321 1322 bcopy(shost, &eh.ether_shost, sizeof(eh.ether_shost)); 1323 bcopy(dhost, &eh.ether_dhost, sizeof(eh.ether_dhost)); 1324 eh.ether_type = etype; 1325 1326 if (m->m_flags & (M_BCAST|M_MCAST)) { 1327 struct carp_softc *vh; 1328 struct mbuf *m0; 1329 1330 /* 1331 * XXX Should really check the list of multicast addresses 1332 * for each CARP interface _before_ copying. 1333 */ 1334 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 1335 m0 = m_copym(m, 0, M_COPYALL, M_DONTWAIT); 1336 if (m0 == NULL) 1337 continue; 1338 m0->m_pkthdr.rcvif = &vh->sc_if; 1339 ether_input(&vh->sc_if, m0); 1340 } 1341 return (1); 1342 } 1343 1344 ifp = carp_ourether(cif, &eh, m->m_pkthdr.rcvif->if_type, 0); 1345 if (ifp == NULL) { 1346 return (1); 1347 } 1348 1349 m->m_pkthdr.rcvif = ifp; 1350 1351 #if NBPFILTER > 0 1352 if (ifp->if_bpf) 1353 bpf_mtap(ifp->if_bpf, m); 1354 #endif 1355 ifp->if_ipackets++; 1356 ether_input(ifp, m); 1357 return (0); 1358 } 1359 1360 void 1361 carp_master_down(void *v) 1362 { 1363 struct carp_softc *sc = v; 1364 1365 switch (sc->sc_state) { 1366 case INIT: 1367 printf("%s: master_down event in INIT state\n", 1368 sc->sc_if.if_xname); 1369 break; 1370 case MASTER: 1371 break; 1372 case BACKUP: 1373 carp_set_state(sc, MASTER); 1374 carp_send_ad(sc); 1375 carp_send_arp(sc); 1376 #ifdef INET6 1377 carp_send_na(sc); 1378 #endif /* INET6 */ 1379 carp_setrun(sc, 0); 1380 carp_setroute(sc, RTM_ADD); 1381 break; 1382 } 1383 } 1384 1385 /* 1386 * When in backup state, af indicates whether to reset the master down timer 1387 * for v4 or v6. If it's set to zero, reset the ones which are already pending. 1388 */ 1389 void 1390 carp_setrun(struct carp_softc *sc, sa_family_t af) 1391 { 1392 struct timeval tv; 1393 1394 if (sc->sc_carpdev == NULL) { 1395 sc->sc_if.if_flags &= ~IFF_RUNNING; 1396 carp_set_state(sc, INIT); 1397 return; 1398 } 1399 1400 if (sc->sc_if.if_flags & IFF_UP && sc->sc_vhid > 0 && 1401 (sc->sc_naddrs || sc->sc_naddrs6) && !sc->sc_suppress) { 1402 sc->sc_if.if_flags |= IFF_RUNNING; 1403 } else { 1404 sc->sc_if.if_flags &= ~IFF_RUNNING; 1405 carp_setroute(sc, RTM_DELETE); 1406 return; 1407 } 1408 1409 switch (sc->sc_state) { 1410 case INIT: 1411 carp_set_state(sc, BACKUP); 1412 carp_setroute(sc, RTM_DELETE); 1413 carp_setrun(sc, 0); 1414 break; 1415 case BACKUP: 1416 callout_stop(&sc->sc_ad_tmo); 1417 tv.tv_sec = 3 * sc->sc_advbase; 1418 tv.tv_usec = sc->sc_advskew * 1000000 / 256; 1419 switch (af) { 1420 #ifdef INET 1421 case AF_INET: 1422 callout_schedule(&sc->sc_md_tmo, tvtohz(&tv)); 1423 break; 1424 #endif /* INET */ 1425 #ifdef INET6 1426 case AF_INET6: 1427 callout_schedule(&sc->sc_md6_tmo, tvtohz(&tv)); 1428 break; 1429 #endif /* INET6 */ 1430 default: 1431 if (sc->sc_naddrs) 1432 callout_schedule(&sc->sc_md_tmo, tvtohz(&tv)); 1433 if (sc->sc_naddrs6) 1434 callout_schedule(&sc->sc_md6_tmo, tvtohz(&tv)); 1435 break; 1436 } 1437 break; 1438 case MASTER: 1439 tv.tv_sec = sc->sc_advbase; 1440 tv.tv_usec = sc->sc_advskew * 1000000 / 256; 1441 callout_schedule(&sc->sc_ad_tmo, tvtohz(&tv)); 1442 break; 1443 } 1444 } 1445 1446 void 1447 carp_multicast_cleanup(struct carp_softc *sc) 1448 { 1449 struct ip_moptions *imo = &sc->sc_imo; 1450 #ifdef INET6 1451 struct ip6_moptions *im6o = &sc->sc_im6o; 1452 #endif 1453 u_int16_t n = imo->imo_num_memberships; 1454 1455 /* Clean up our own multicast memberships */ 1456 while (n-- > 0) { 1457 if (imo->imo_membership[n] != NULL) { 1458 in_delmulti(imo->imo_membership[n]); 1459 imo->imo_membership[n] = NULL; 1460 } 1461 } 1462 imo->imo_num_memberships = 0; 1463 imo->imo_multicast_ifp = NULL; 1464 1465 #ifdef INET6 1466 while (!LIST_EMPTY(&im6o->im6o_memberships)) { 1467 struct in6_multi_mship *imm = 1468 LIST_FIRST(&im6o->im6o_memberships); 1469 1470 LIST_REMOVE(imm, i6mm_chain); 1471 in6_leavegroup(imm); 1472 } 1473 im6o->im6o_multicast_ifp = NULL; 1474 #endif 1475 1476 /* And any other multicast memberships */ 1477 carp_ether_purgemulti(sc); 1478 } 1479 1480 int 1481 carp_set_ifp(struct carp_softc *sc, struct ifnet *ifp) 1482 { 1483 struct carp_if *cif, *ncif = NULL; 1484 struct carp_softc *vr, *after = NULL; 1485 int myself = 0, error = 0; 1486 int s; 1487 1488 if (ifp == sc->sc_carpdev) 1489 return (0); 1490 1491 if (ifp != NULL) { 1492 if ((ifp->if_flags & IFF_MULTICAST) == 0) 1493 return (EADDRNOTAVAIL); 1494 1495 if (ifp->if_type == IFT_CARP) 1496 return (EINVAL); 1497 1498 if (ifp->if_carp == NULL) { 1499 MALLOC(ncif, struct carp_if *, sizeof(*cif), 1500 M_IFADDR, M_NOWAIT); 1501 if (ncif == NULL) 1502 return (ENOBUFS); 1503 if ((error = ifpromisc(ifp, 1))) { 1504 FREE(ncif, M_IFADDR); 1505 return (error); 1506 } 1507 1508 ncif->vhif_ifp = ifp; 1509 TAILQ_INIT(&ncif->vhif_vrs); 1510 } else { 1511 cif = (struct carp_if *)ifp->if_carp; 1512 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) 1513 if (vr != sc && vr->sc_vhid == sc->sc_vhid) 1514 return (EINVAL); 1515 } 1516 1517 /* detach from old interface */ 1518 if (sc->sc_carpdev != NULL) 1519 carpdetach(sc); 1520 1521 /* join multicast groups */ 1522 if (sc->sc_naddrs < 0 && 1523 (error = carp_join_multicast(sc)) != 0) { 1524 if (ncif != NULL) 1525 FREE(ncif, M_IFADDR); 1526 return (error); 1527 } 1528 1529 #ifdef INET6 1530 if (sc->sc_naddrs6 < 0 && 1531 (error = carp_join_multicast6(sc)) != 0) { 1532 if (ncif != NULL) 1533 FREE(ncif, M_IFADDR); 1534 carp_multicast_cleanup(sc); 1535 return (error); 1536 } 1537 #endif 1538 1539 /* attach carp interface to physical interface */ 1540 if (ncif != NULL) 1541 ifp->if_carp = (void *)ncif; 1542 sc->sc_carpdev = ifp; 1543 cif = (struct carp_if *)ifp->if_carp; 1544 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) { 1545 if (vr == sc) 1546 myself = 1; 1547 if (vr->sc_vhid < sc->sc_vhid) 1548 after = vr; 1549 } 1550 1551 if (!myself) { 1552 /* We're trying to keep things in order */ 1553 if (after == NULL) { 1554 TAILQ_INSERT_TAIL(&cif->vhif_vrs, sc, sc_list); 1555 } else { 1556 TAILQ_INSERT_AFTER(&cif->vhif_vrs, after, 1557 sc, sc_list); 1558 } 1559 cif->vhif_nvrs++; 1560 } 1561 if (sc->sc_naddrs || sc->sc_naddrs6) 1562 sc->sc_if.if_flags |= IFF_UP; 1563 carp_set_enaddr(sc); 1564 s = splnet(); 1565 /* XXX linkstatehooks establish */ 1566 carp_carpdev_state(ifp); 1567 splx(s); 1568 } else { 1569 carpdetach(sc); 1570 sc->sc_if.if_flags &= ~(IFF_UP|IFF_RUNNING); 1571 } 1572 return (0); 1573 } 1574 1575 void 1576 carp_set_enaddr(struct carp_softc *sc) 1577 { 1578 if (sc->sc_carpdev && sc->sc_carpdev->if_type == IFT_ISO88025) { 1579 LLADDR(sc->sc_if.if_sadl)[0] = 3; 1580 LLADDR(sc->sc_if.if_sadl)[1] = 0; 1581 LLADDR(sc->sc_if.if_sadl)[2] = 0x40 >> (sc->sc_vhid - 1); 1582 LLADDR(sc->sc_if.if_sadl)[3] = 0x40000 >> (sc->sc_vhid - 1); 1583 LLADDR(sc->sc_if.if_sadl)[4] = 0; 1584 LLADDR(sc->sc_if.if_sadl)[5] = 0; 1585 } else { 1586 LLADDR(sc->sc_if.if_sadl)[0] = 0; 1587 LLADDR(sc->sc_if.if_sadl)[1] = 0; 1588 LLADDR(sc->sc_if.if_sadl)[2] = 0x5e; 1589 LLADDR(sc->sc_if.if_sadl)[3] = 0; 1590 LLADDR(sc->sc_if.if_sadl)[4] = 1; 1591 LLADDR(sc->sc_if.if_sadl)[5] = sc->sc_vhid; 1592 } 1593 } 1594 1595 void 1596 carp_addr_updated(void *v) 1597 { 1598 struct carp_softc *sc = (struct carp_softc *) v; 1599 struct ifaddr *ifa; 1600 int new_naddrs = 0, new_naddrs6 = 0; 1601 1602 TAILQ_FOREACH(ifa, &sc->sc_if.if_addrlist, ifa_list) { 1603 if (ifa->ifa_addr->sa_family == AF_INET) 1604 new_naddrs++; 1605 else if (ifa->ifa_addr->sa_family == AF_INET6) 1606 new_naddrs6++; 1607 } 1608 1609 /* Handle a callback after SIOCDIFADDR */ 1610 if (new_naddrs < sc->sc_naddrs || new_naddrs6 < sc->sc_naddrs6) { 1611 struct in_addr mc_addr; 1612 struct in_multi *inm; 1613 1614 sc->sc_naddrs = new_naddrs; 1615 sc->sc_naddrs6 = new_naddrs6; 1616 1617 /* Re-establish multicast membership removed by in_control */ 1618 mc_addr.s_addr = INADDR_CARP_GROUP; 1619 IN_LOOKUP_MULTI(mc_addr, &sc->sc_if, inm); 1620 if (inm == NULL) { 1621 bzero(&sc->sc_imo, sizeof(sc->sc_imo)); 1622 1623 if (sc->sc_carpdev != NULL && sc->sc_naddrs > 0) 1624 carp_join_multicast(sc); 1625 } 1626 1627 if (sc->sc_naddrs == 0 && sc->sc_naddrs6 == 0) { 1628 sc->sc_if.if_flags &= ~IFF_UP; 1629 carp_set_state(sc, INIT); 1630 } else 1631 carp_hmac_prepare(sc); 1632 } 1633 1634 carp_setrun(sc, 0); 1635 } 1636 1637 int 1638 carp_set_addr(struct carp_softc *sc, struct sockaddr_in *sin) 1639 { 1640 struct ifnet *ifp = sc->sc_carpdev; 1641 struct in_ifaddr *ia, *ia_if; 1642 int error = 0; 1643 1644 if (sin->sin_addr.s_addr == 0) { 1645 if (!(sc->sc_if.if_flags & IFF_UP)) 1646 carp_set_state(sc, INIT); 1647 if (sc->sc_naddrs) 1648 sc->sc_if.if_flags |= IFF_UP; 1649 carp_setrun(sc, 0); 1650 return (0); 1651 } 1652 1653 /* we have to do this by hand to ensure we don't match on ourselves */ 1654 ia_if = NULL; 1655 for (ia = TAILQ_FIRST(&in_ifaddrhead); ia; 1656 ia = TAILQ_NEXT(ia, ia_list)) { 1657 1658 /* and, yeah, we need a multicast-capable iface too */ 1659 if (ia->ia_ifp != &sc->sc_if && 1660 ia->ia_ifp->if_type != IFT_CARP && 1661 (ia->ia_ifp->if_flags & IFF_MULTICAST) && 1662 (sin->sin_addr.s_addr & ia->ia_subnetmask) == 1663 ia->ia_subnet) { 1664 if (!ia_if) 1665 ia_if = ia; 1666 } 1667 } 1668 1669 if (ia_if) { 1670 ia = ia_if; 1671 if (ifp) { 1672 if (ifp != ia->ia_ifp) 1673 return (EADDRNOTAVAIL); 1674 } else { 1675 ifp = ia->ia_ifp; 1676 } 1677 } 1678 1679 if ((error = carp_set_ifp(sc, ifp))) 1680 return (error); 1681 1682 if (sc->sc_carpdev == NULL) 1683 return (EADDRNOTAVAIL); 1684 1685 if (sc->sc_naddrs == 0 && (error = carp_join_multicast(sc)) != 0) 1686 return (error); 1687 1688 sc->sc_naddrs++; 1689 if (sc->sc_carpdev != NULL) 1690 sc->sc_if.if_flags |= IFF_UP; 1691 1692 carp_set_state(sc, INIT); 1693 carp_setrun(sc, 0); 1694 1695 /* 1696 * Hook if_addrhooks so that we get a callback after in_ifinit has run, 1697 * to correct any inappropriate routes that it inserted. 1698 */ 1699 if (sc->ah_cookie == 0) { 1700 /* XXX link address hook */ 1701 } 1702 1703 return (0); 1704 } 1705 1706 int 1707 carp_join_multicast(struct carp_softc *sc) 1708 { 1709 struct ip_moptions *imo = &sc->sc_imo, tmpimo; 1710 struct in_addr addr; 1711 1712 bzero(&tmpimo, sizeof(tmpimo)); 1713 addr.s_addr = INADDR_CARP_GROUP; 1714 if ((tmpimo.imo_membership[0] = 1715 in_addmulti(&addr, &sc->sc_if)) == NULL) { 1716 return (ENOBUFS); 1717 } 1718 1719 imo->imo_membership[0] = tmpimo.imo_membership[0]; 1720 imo->imo_num_memberships = 1; 1721 imo->imo_multicast_ifp = &sc->sc_if; 1722 imo->imo_multicast_ttl = CARP_DFLTTL; 1723 imo->imo_multicast_loop = 0; 1724 return (0); 1725 } 1726 1727 1728 #ifdef INET6 1729 int 1730 carp_set_addr6(struct carp_softc *sc, struct sockaddr_in6 *sin6) 1731 { 1732 struct ifnet *ifp = sc->sc_carpdev; 1733 struct in6_ifaddr *ia, *ia_if; 1734 int error = 0; 1735 1736 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) { 1737 if (!(sc->sc_if.if_flags & IFF_UP)) 1738 carp_set_state(sc, INIT); 1739 if (sc->sc_naddrs6) 1740 sc->sc_if.if_flags |= IFF_UP; 1741 carp_setrun(sc, 0); 1742 return (0); 1743 } 1744 1745 /* we have to do this by hand to ensure we don't match on ourselves */ 1746 ia_if = NULL; 1747 for (ia = in6_ifaddr; ia; ia = ia->ia_next) { 1748 int i; 1749 1750 for (i = 0; i < 4; i++) { 1751 if ((sin6->sin6_addr.s6_addr32[i] & 1752 ia->ia_prefixmask.sin6_addr.s6_addr32[i]) != 1753 (ia->ia_addr.sin6_addr.s6_addr32[i] & 1754 ia->ia_prefixmask.sin6_addr.s6_addr32[i])) 1755 break; 1756 } 1757 /* and, yeah, we need a multicast-capable iface too */ 1758 if (ia->ia_ifp != &sc->sc_if && 1759 ia->ia_ifp->if_type != IFT_CARP && 1760 (ia->ia_ifp->if_flags & IFF_MULTICAST) && 1761 (i == 4)) { 1762 if (!ia_if) 1763 ia_if = ia; 1764 } 1765 } 1766 1767 if (ia_if) { 1768 ia = ia_if; 1769 if (sc->sc_carpdev) { 1770 if (sc->sc_carpdev != ia->ia_ifp) 1771 return (EADDRNOTAVAIL); 1772 } else { 1773 ifp = ia->ia_ifp; 1774 } 1775 } 1776 1777 if ((error = carp_set_ifp(sc, ifp))) 1778 return (error); 1779 1780 if (sc->sc_carpdev == NULL) 1781 return (EADDRNOTAVAIL); 1782 1783 if (sc->sc_naddrs6 == 0 && (error = carp_join_multicast6(sc)) != 0) 1784 return (error); 1785 1786 sc->sc_naddrs6++; 1787 if (sc->sc_carpdev != NULL) 1788 sc->sc_if.if_flags |= IFF_UP; 1789 carp_set_state(sc, INIT); 1790 carp_setrun(sc, 0); 1791 1792 return (0); 1793 } 1794 1795 int 1796 carp_join_multicast6(struct carp_softc *sc) 1797 { 1798 struct in6_multi_mship *imm, *imm2; 1799 struct ip6_moptions *im6o = &sc->sc_im6o; 1800 struct sockaddr_in6 addr6; 1801 int error; 1802 1803 /* Join IPv6 CARP multicast group */ 1804 bzero(&addr6, sizeof(addr6)); 1805 addr6.sin6_family = AF_INET6; 1806 addr6.sin6_len = sizeof(addr6); 1807 addr6.sin6_addr.s6_addr16[0] = htons(0xff02); 1808 addr6.sin6_addr.s6_addr16[1] = htons(sc->sc_if.if_index); 1809 addr6.sin6_addr.s6_addr8[15] = 0x12; 1810 if ((imm = in6_joingroup(&sc->sc_if, 1811 &addr6.sin6_addr, &error, 0)) == NULL) { 1812 return (error); 1813 } 1814 /* join solicited multicast address */ 1815 bzero(&addr6.sin6_addr, sizeof(addr6.sin6_addr)); 1816 addr6.sin6_addr.s6_addr16[0] = htons(0xff02); 1817 addr6.sin6_addr.s6_addr16[1] = htons(sc->sc_if.if_index); 1818 addr6.sin6_addr.s6_addr32[1] = 0; 1819 addr6.sin6_addr.s6_addr32[2] = htonl(1); 1820 addr6.sin6_addr.s6_addr32[3] = 0; 1821 addr6.sin6_addr.s6_addr8[12] = 0xff; 1822 if ((imm2 = in6_joingroup(&sc->sc_if, 1823 &addr6.sin6_addr, &error, 0)) == NULL) { 1824 in6_leavegroup(imm); 1825 return (error); 1826 } 1827 1828 /* apply v6 multicast membership */ 1829 im6o->im6o_multicast_ifp = &sc->sc_if; 1830 if (imm) 1831 LIST_INSERT_HEAD(&im6o->im6o_memberships, imm, 1832 i6mm_chain); 1833 if (imm2) 1834 LIST_INSERT_HEAD(&im6o->im6o_memberships, imm2, 1835 i6mm_chain); 1836 1837 return (0); 1838 } 1839 1840 #endif /* INET6 */ 1841 1842 int 1843 carp_ioctl(struct ifnet *ifp, u_long cmd, void *addr) 1844 { 1845 struct lwp *l = curlwp; /* XXX */ 1846 struct carp_softc *sc = ifp->if_softc, *vr; 1847 struct carpreq carpr; 1848 struct ifaddr *ifa; 1849 struct ifreq *ifr; 1850 struct ifnet *cdev = NULL; 1851 int error = 0; 1852 1853 ifa = (struct ifaddr *)addr; 1854 ifr = (struct ifreq *)addr; 1855 1856 switch (cmd) { 1857 case SIOCSIFADDR: 1858 switch (ifa->ifa_addr->sa_family) { 1859 #ifdef INET 1860 case AF_INET: 1861 sc->sc_if.if_flags |= IFF_UP; 1862 bcopy(ifa->ifa_addr, ifa->ifa_dstaddr, 1863 sizeof(struct sockaddr)); 1864 error = carp_set_addr(sc, satosin(ifa->ifa_addr)); 1865 break; 1866 #endif /* INET */ 1867 #ifdef INET6 1868 case AF_INET6: 1869 sc->sc_if.if_flags|= IFF_UP; 1870 error = carp_set_addr6(sc, satosin6(ifa->ifa_addr)); 1871 break; 1872 #endif /* INET6 */ 1873 default: 1874 error = EAFNOSUPPORT; 1875 break; 1876 } 1877 break; 1878 1879 case SIOCSIFFLAGS: 1880 if (sc->sc_state != INIT && !(ifr->ifr_flags & IFF_UP)) { 1881 callout_stop(&sc->sc_ad_tmo); 1882 callout_stop(&sc->sc_md_tmo); 1883 callout_stop(&sc->sc_md6_tmo); 1884 if (sc->sc_state == MASTER) { 1885 /* we need the interface up to bow out */ 1886 sc->sc_if.if_flags |= IFF_UP; 1887 sc->sc_bow_out = 1; 1888 carp_send_ad(sc); 1889 } 1890 sc->sc_if.if_flags &= ~IFF_UP; 1891 carp_set_state(sc, INIT); 1892 carp_setrun(sc, 0); 1893 } else if (sc->sc_state == INIT && (ifr->ifr_flags & IFF_UP)) { 1894 sc->sc_if.if_flags |= IFF_UP; 1895 carp_setrun(sc, 0); 1896 } 1897 break; 1898 1899 case SIOCSVH: 1900 if (l == NULL) 1901 break; 1902 if ((error = kauth_authorize_network(l->l_cred, 1903 KAUTH_NETWORK_INTERFACE, 1904 KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd, 1905 NULL)) != 0) 1906 break; 1907 if ((error = copyin(ifr->ifr_data, &carpr, sizeof carpr))) 1908 break; 1909 error = 1; 1910 if (carpr.carpr_carpdev[0] != '\0' && 1911 (cdev = ifunit(carpr.carpr_carpdev)) == NULL) 1912 return (EINVAL); 1913 if ((error = carp_set_ifp(sc, cdev))) 1914 return (error); 1915 if (sc->sc_state != INIT && carpr.carpr_state != sc->sc_state) { 1916 switch (carpr.carpr_state) { 1917 case BACKUP: 1918 callout_stop(&sc->sc_ad_tmo); 1919 carp_set_state(sc, BACKUP); 1920 carp_setrun(sc, 0); 1921 carp_setroute(sc, RTM_DELETE); 1922 break; 1923 case MASTER: 1924 carp_master_down(sc); 1925 break; 1926 default: 1927 break; 1928 } 1929 } 1930 if (carpr.carpr_vhid > 0) { 1931 if (carpr.carpr_vhid > 255) { 1932 error = EINVAL; 1933 break; 1934 } 1935 if (sc->sc_carpdev) { 1936 struct carp_if *cif; 1937 cif = (struct carp_if *)sc->sc_carpdev->if_carp; 1938 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) 1939 if (vr != sc && 1940 vr->sc_vhid == carpr.carpr_vhid) 1941 return (EINVAL); 1942 } 1943 sc->sc_vhid = carpr.carpr_vhid; 1944 carp_set_enaddr(sc); 1945 carp_set_state(sc, INIT); 1946 error--; 1947 } 1948 if (carpr.carpr_advbase > 0 || carpr.carpr_advskew > 0) { 1949 if (carpr.carpr_advskew > 254) { 1950 error = EINVAL; 1951 break; 1952 } 1953 if (carpr.carpr_advbase > 255) { 1954 error = EINVAL; 1955 break; 1956 } 1957 sc->sc_advbase = carpr.carpr_advbase; 1958 sc->sc_advskew = carpr.carpr_advskew; 1959 error--; 1960 } 1961 bcopy(carpr.carpr_key, sc->sc_key, sizeof(sc->sc_key)); 1962 if (error > 0) 1963 error = EINVAL; 1964 else { 1965 error = 0; 1966 carp_setrun(sc, 0); 1967 } 1968 break; 1969 1970 case SIOCGVH: 1971 bzero(&carpr, sizeof(carpr)); 1972 if (sc->sc_carpdev != NULL) 1973 strlcpy(carpr.carpr_carpdev, sc->sc_carpdev->if_xname, 1974 IFNAMSIZ); 1975 carpr.carpr_state = sc->sc_state; 1976 carpr.carpr_vhid = sc->sc_vhid; 1977 carpr.carpr_advbase = sc->sc_advbase; 1978 carpr.carpr_advskew = sc->sc_advskew; 1979 1980 if ((l == NULL) || (error = kauth_authorize_network(l->l_cred, 1981 KAUTH_NETWORK_INTERFACE, 1982 KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd, 1983 NULL)) != 0) 1984 bcopy(sc->sc_key, carpr.carpr_key, 1985 sizeof(carpr.carpr_key)); 1986 error = copyout(&carpr, ifr->ifr_data, sizeof(carpr)); 1987 break; 1988 1989 case SIOCADDMULTI: 1990 error = carp_ether_addmulti(sc, ifr); 1991 break; 1992 1993 case SIOCDELMULTI: 1994 error = carp_ether_delmulti(sc, ifr); 1995 break; 1996 1997 default: 1998 error = EINVAL; 1999 } 2000 2001 carp_hmac_prepare(sc); 2002 return (error); 2003 } 2004 2005 2006 /* 2007 * Start output on carp interface. This function should never be called. 2008 */ 2009 void 2010 carp_start(struct ifnet *ifp) 2011 { 2012 #ifdef DEBUG 2013 printf("%s: start called\n", ifp->if_xname); 2014 #endif 2015 } 2016 2017 int 2018 carp_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *sa, 2019 struct rtentry *rt) 2020 { 2021 struct carp_softc *sc = ((struct carp_softc *)ifp->if_softc); 2022 2023 if (sc->sc_carpdev != NULL && sc->sc_state == MASTER) { 2024 return (sc->sc_carpdev->if_output(ifp, m, sa, rt)); 2025 } else { 2026 m_freem(m); 2027 return (ENETUNREACH); 2028 } 2029 } 2030 2031 void 2032 carp_set_state(struct carp_softc *sc, int state) 2033 { 2034 if (sc->sc_state == state) 2035 return; 2036 2037 sc->sc_state = state; 2038 switch (state) { 2039 case BACKUP: 2040 sc->sc_if.if_link_state = LINK_STATE_DOWN; 2041 break; 2042 case MASTER: 2043 sc->sc_if.if_link_state = LINK_STATE_UP; 2044 break; 2045 default: 2046 sc->sc_if.if_link_state = LINK_STATE_UNKNOWN; 2047 break; 2048 } 2049 rt_ifmsg(&sc->sc_if); 2050 } 2051 2052 void 2053 carp_carpdev_state(void *v) 2054 { 2055 struct carp_if *cif; 2056 struct carp_softc *sc; 2057 struct ifnet *ifp = v; 2058 2059 if (ifp->if_type == IFT_CARP) 2060 return; 2061 2062 cif = (struct carp_if *)ifp->if_carp; 2063 2064 TAILQ_FOREACH(sc, &cif->vhif_vrs, sc_list) { 2065 int suppressed = sc->sc_suppress; 2066 2067 if (sc->sc_carpdev->if_link_state == LINK_STATE_DOWN || 2068 !(sc->sc_carpdev->if_flags & IFF_UP)) { 2069 sc->sc_if.if_flags &= ~IFF_RUNNING; 2070 callout_stop(&sc->sc_ad_tmo); 2071 callout_stop(&sc->sc_md_tmo); 2072 callout_stop(&sc->sc_md6_tmo); 2073 carp_set_state(sc, INIT); 2074 sc->sc_suppress = 1; 2075 carp_setrun(sc, 0); 2076 if (!suppressed) { 2077 carp_suppress_preempt++; 2078 if (carp_suppress_preempt == 1) 2079 carp_send_ad_all(); 2080 } 2081 } else { 2082 carp_set_state(sc, INIT); 2083 sc->sc_suppress = 0; 2084 carp_setrun(sc, 0); 2085 if (suppressed) 2086 carp_suppress_preempt--; 2087 } 2088 } 2089 } 2090 2091 int 2092 carp_ether_addmulti(struct carp_softc *sc, struct ifreq *ifr) 2093 { 2094 struct ifnet *ifp; 2095 struct carp_mc_entry *mc; 2096 u_int8_t addrlo[ETHER_ADDR_LEN], addrhi[ETHER_ADDR_LEN]; 2097 int error; 2098 2099 ifp = sc->sc_carpdev; 2100 if (ifp == NULL) 2101 return (EINVAL); 2102 2103 error = ether_addmulti(ifr, &sc->sc_ac); 2104 if (error != ENETRESET) 2105 return (error); 2106 2107 /* 2108 * This is new multicast address. We have to tell parent 2109 * about it. Also, remember this multicast address so that 2110 * we can delete them on unconfigure. 2111 */ 2112 MALLOC(mc, struct carp_mc_entry *, sizeof(struct carp_mc_entry), 2113 M_DEVBUF, M_NOWAIT); 2114 if (mc == NULL) { 2115 error = ENOMEM; 2116 goto alloc_failed; 2117 } 2118 2119 /* 2120 * As ether_addmulti() returns ENETRESET, following two 2121 * statement shouldn't fail. 2122 */ 2123 (void)ether_multiaddr(&ifr->ifr_addr, addrlo, addrhi); 2124 ETHER_LOOKUP_MULTI(addrlo, addrhi, &sc->sc_ac, mc->mc_enm); 2125 memcpy(&mc->mc_addr, &ifr->ifr_addr, ifr->ifr_addr.sa_len); 2126 LIST_INSERT_HEAD(&sc->carp_mc_listhead, mc, mc_entries); 2127 2128 error = (*ifp->if_ioctl)(ifp, SIOCADDMULTI, (void *)ifr); 2129 if (error != 0) 2130 goto ioctl_failed; 2131 2132 return (error); 2133 2134 ioctl_failed: 2135 LIST_REMOVE(mc, mc_entries); 2136 FREE(mc, M_DEVBUF); 2137 alloc_failed: 2138 (void)ether_delmulti(ifr, &sc->sc_ac); 2139 2140 return (error); 2141 } 2142 2143 int 2144 carp_ether_delmulti(struct carp_softc *sc, struct ifreq *ifr) 2145 { 2146 struct ifnet *ifp; 2147 struct ether_multi *enm; 2148 struct carp_mc_entry *mc; 2149 u_int8_t addrlo[ETHER_ADDR_LEN], addrhi[ETHER_ADDR_LEN]; 2150 int error; 2151 2152 ifp = sc->sc_carpdev; 2153 if (ifp == NULL) 2154 return (EINVAL); 2155 2156 /* 2157 * Find a key to lookup carp_mc_entry. We have to do this 2158 * before calling ether_delmulti for obvious reason. 2159 */ 2160 if ((error = ether_multiaddr(&ifr->ifr_addr, addrlo, addrhi)) != 0) 2161 return (error); 2162 ETHER_LOOKUP_MULTI(addrlo, addrhi, &sc->sc_ac, enm); 2163 if (enm == NULL) 2164 return (EINVAL); 2165 2166 LIST_FOREACH(mc, &sc->carp_mc_listhead, mc_entries) 2167 if (mc->mc_enm == enm) 2168 break; 2169 2170 /* We won't delete entries we didn't add */ 2171 if (mc == NULL) 2172 return (EINVAL); 2173 2174 error = ether_delmulti(ifr, &sc->sc_ac); 2175 if (error != ENETRESET) 2176 return (error); 2177 2178 /* We no longer use this multicast address. Tell parent so. */ 2179 error = (*ifp->if_ioctl)(ifp, SIOCDELMULTI, (void *)ifr); 2180 if (error == 0) { 2181 /* And forget about this address. */ 2182 LIST_REMOVE(mc, mc_entries); 2183 FREE(mc, M_DEVBUF); 2184 } else 2185 (void)ether_addmulti(ifr, &sc->sc_ac); 2186 return (error); 2187 } 2188 2189 /* 2190 * Delete any multicast address we have asked to add from parent 2191 * interface. Called when the carp is being unconfigured. 2192 */ 2193 void 2194 carp_ether_purgemulti(struct carp_softc *sc) 2195 { 2196 struct ifnet *ifp = sc->sc_carpdev; /* Parent. */ 2197 struct carp_mc_entry *mc; 2198 union { 2199 struct ifreq ifreq; 2200 struct { 2201 char ifr_name[IFNAMSIZ]; 2202 struct sockaddr_storage ifr_ss; 2203 } ifreq_storage; 2204 } u; 2205 struct ifreq *ifr = &u.ifreq; 2206 2207 if (ifp == NULL) 2208 return; 2209 2210 memcpy(ifr->ifr_name, ifp->if_xname, IFNAMSIZ); 2211 while ((mc = LIST_FIRST(&sc->carp_mc_listhead)) != NULL) { 2212 memcpy(&ifr->ifr_addr, &mc->mc_addr, mc->mc_addr.ss_len); 2213 (void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, (void *)ifr); 2214 LIST_REMOVE(mc, mc_entries); 2215 FREE(mc, M_DEVBUF); 2216 } 2217 } 2218 2219 SYSCTL_SETUP(sysctl_net_inet_carp_setup, "sysctl net.inet.carp subtree setup") 2220 { 2221 2222 sysctl_createv(clog, 0, NULL, NULL, 2223 CTLFLAG_PERMANENT, 2224 CTLTYPE_NODE, "net", NULL, 2225 NULL, 0, NULL, 0, 2226 CTL_NET, CTL_EOL); 2227 sysctl_createv(clog, 0, NULL, NULL, 2228 CTLFLAG_PERMANENT, 2229 CTLTYPE_NODE, "inet", NULL, 2230 NULL, 0, NULL, 0, 2231 CTL_NET, PF_INET, CTL_EOL); 2232 sysctl_createv(clog, 0, NULL, NULL, 2233 CTLFLAG_PERMANENT, 2234 CTLTYPE_NODE, "carp", 2235 SYSCTL_DESCR("CARP related settings"), 2236 NULL, 0, NULL, 0, 2237 CTL_NET, PF_INET, IPPROTO_CARP, CTL_EOL); 2238 2239 sysctl_createv(clog, 0, NULL, NULL, 2240 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2241 CTLTYPE_INT, "preempt", 2242 SYSCTL_DESCR("Enable CARP Preempt"), 2243 NULL, 0, &carp_opts[CARPCTL_PREEMPT], 0, 2244 CTL_NET, PF_INET, IPPROTO_CARP, 2245 CTL_CREATE, CTL_EOL); 2246 sysctl_createv(clog, 0, NULL, NULL, 2247 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2248 CTLTYPE_INT, "arpbalance", 2249 SYSCTL_DESCR("Enable ARP balancing"), 2250 NULL, 0, &carp_opts[CARPCTL_ARPBALANCE], 0, 2251 CTL_NET, PF_INET, IPPROTO_CARP, 2252 CTL_CREATE, CTL_EOL); 2253 sysctl_createv(clog, 0, NULL, NULL, 2254 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2255 CTLTYPE_INT, "allow", 2256 SYSCTL_DESCR("Enable CARP"), 2257 NULL, 0, &carp_opts[CARPCTL_ALLOW], 0, 2258 CTL_NET, PF_INET, IPPROTO_CARP, 2259 CTL_CREATE, CTL_EOL); 2260 sysctl_createv(clog, 0, NULL, NULL, 2261 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2262 CTLTYPE_INT, "log", 2263 SYSCTL_DESCR("CARP logging"), 2264 NULL, 0, &carp_opts[CARPCTL_LOG], 0, 2265 CTL_NET, PF_INET, IPPROTO_CARP, 2266 CTL_CREATE, CTL_EOL); 2267 sysctl_createv(clog, 0, NULL, NULL, 2268 CTLFLAG_PERMANENT, 2269 CTLTYPE_STRUCT, "stats", 2270 SYSCTL_DESCR("CARP statistics"), 2271 NULL, 0, &carpstats, sizeof(carpstats), 2272 CTL_NET, PF_INET, IPPROTO_CARP, CARPCTL_STATS, 2273 CTL_EOL); 2274 } 2275
Indexes created Mon Sep 22 13:09:51 GMT 2025