ip_carp.c revision 1.23
1 /* $NetBSD: ip_carp.c,v 1.23 2008/03/15 16:44:03 ws 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 #include <sys/cdefs.h> 31 __KERNEL_RCSID(0, "$NetBSD: ip_carp.c,v 1.23 2008/03/15 16:44:03 ws Exp $"); 32 33 /* 34 * TODO: 35 * - iface reconfigure 36 * - support for hardware checksum calculations; 37 * 38 */ 39 40 #include <sys/param.h> 41 #include <sys/proc.h> 42 #include <sys/mbuf.h> 43 #include <sys/socket.h> 44 #include <sys/socketvar.h> 45 #include <sys/callout.h> 46 #include <sys/ioctl.h> 47 #include <sys/errno.h> 48 #include <sys/device.h> 49 #include <sys/time.h> 50 #include <sys/kernel.h> 51 #include <sys/kauth.h> 52 #include <sys/sysctl.h> 53 #include <sys/ucred.h> 54 #include <sys/syslog.h> 55 #include <sys/acct.h> 56 57 #include <sys/cpu.h> 58 59 #include <net/if.h> 60 #include <net/pfil.h> 61 #include <net/if_types.h> 62 #include <net/if_ether.h> 63 #include <net/route.h> 64 #include <net/netisr.h> 65 #include <netinet/if_inarp.h> 66 67 #include <machine/stdarg.h> 68 69 #if NFDDI > 0 70 #include <net/if_fddi.h> 71 #endif 72 #if NTOKEN > 0 73 #include <net/if_token.h> 74 #endif 75 76 #ifdef INET 77 #include <netinet/in.h> 78 #include <netinet/in_systm.h> 79 #include <netinet/in_var.h> 80 #include <netinet/ip.h> 81 #include <netinet/ip_var.h> 82 83 #include <net/if_dl.h> 84 #endif 85 86 #ifdef INET6 87 #include <netinet/icmp6.h> 88 #include <netinet/ip6.h> 89 #include <netinet6/ip6_var.h> 90 #include <netinet6/nd6.h> 91 #include <netinet6/scope6_var.h> 92 #endif 93 94 #include "bpfilter.h" 95 #if NBPFILTER > 0 96 #include <net/bpf.h> 97 #endif 98 99 #include <sys/sha1.h> 100 101 #include <netinet/ip_carp.h> 102 103 struct carp_mc_entry { 104 LIST_ENTRY(carp_mc_entry) mc_entries; 105 union { 106 struct ether_multi *mcu_enm; 107 } mc_u; 108 struct sockaddr_storage mc_addr; 109 }; 110 #define mc_enm mc_u.mcu_enm 111 112 struct carp_softc { 113 struct ethercom sc_ac; 114 #define sc_if sc_ac.ec_if 115 #define sc_carpdev sc_ac.ec_if.if_carpdev 116 int ah_cookie; 117 int lh_cookie; 118 struct ip_moptions sc_imo; 119 #ifdef INET6 120 struct ip6_moptions sc_im6o; 121 #endif /* INET6 */ 122 TAILQ_ENTRY(carp_softc) sc_list; 123 124 enum { INIT = 0, BACKUP, MASTER } sc_state; 125 126 int sc_suppress; 127 int sc_bow_out; 128 129 int sc_sendad_errors; 130 #define CARP_SENDAD_MAX_ERRORS 3 131 int sc_sendad_success; 132 #define CARP_SENDAD_MIN_SUCCESS 3 133 134 int sc_vhid; 135 int sc_advskew; 136 int sc_naddrs; 137 int sc_naddrs6; 138 int sc_advbase; /* seconds */ 139 int sc_init_counter; 140 u_int64_t sc_counter; 141 142 /* authentication */ 143 #define CARP_HMAC_PAD 64 144 unsigned char sc_key[CARP_KEY_LEN]; 145 unsigned char sc_pad[CARP_HMAC_PAD]; 146 SHA1_CTX sc_sha1; 147 u_int32_t sc_hashkey[2]; 148 149 struct callout sc_ad_tmo; /* advertisement timeout */ 150 struct callout sc_md_tmo; /* master down timeout */ 151 struct callout sc_md6_tmo; /* master down timeout */ 152 153 LIST_HEAD(__carp_mchead, carp_mc_entry) carp_mc_listhead; 154 }; 155 156 int carp_suppress_preempt = 0; 157 int carp_opts[CARPCTL_MAXID] = { 0, 1, 0, 0, 0 }; /* XXX for now */ 158 struct carpstats carpstats; 159 160 struct carp_if { 161 TAILQ_HEAD(, carp_softc) vhif_vrs; 162 int vhif_nvrs; 163 164 struct ifnet *vhif_ifp; 165 }; 166 167 #define CARP_LOG(sc, s) \ 168 if (carp_opts[CARPCTL_LOG]) { \ 169 if (sc) \ 170 log(LOG_INFO, "%s: ", \ 171 (sc)->sc_if.if_xname); \ 172 else \ 173 log(LOG_INFO, "carp: "); \ 174 addlog s; \ 175 addlog("\n"); \ 176 } 177 178 void carp_hmac_prepare(struct carp_softc *); 179 void carp_hmac_generate(struct carp_softc *, u_int32_t *, 180 unsigned char *); 181 int carp_hmac_verify(struct carp_softc *, u_int32_t *, 182 unsigned char *); 183 void carp_setroute(struct carp_softc *, int); 184 void carp_proto_input_c(struct mbuf *, struct carp_header *, sa_family_t); 185 void carpattach(int); 186 void carpdetach(struct carp_softc *); 187 int carp_prepare_ad(struct mbuf *, struct carp_softc *, 188 struct carp_header *); 189 void carp_send_ad_all(void); 190 void carp_send_ad(void *); 191 void carp_send_arp(struct carp_softc *); 192 void carp_master_down(void *); 193 int carp_ioctl(struct ifnet *, u_long, void *); 194 void carp_start(struct ifnet *); 195 void carp_setrun(struct carp_softc *, sa_family_t); 196 void carp_set_state(struct carp_softc *, int); 197 int carp_addrcount(struct carp_if *, struct in_ifaddr *, int); 198 enum { CARP_COUNT_MASTER, CARP_COUNT_RUNNING }; 199 200 void carp_multicast_cleanup(struct carp_softc *); 201 int carp_set_ifp(struct carp_softc *, struct ifnet *); 202 void carp_set_enaddr(struct carp_softc *); 203 void carp_addr_updated(void *); 204 u_int32_t carp_hash(struct carp_softc *, u_char *); 205 int carp_set_addr(struct carp_softc *, struct sockaddr_in *); 206 int carp_join_multicast(struct carp_softc *); 207 #ifdef INET6 208 void carp_send_na(struct carp_softc *); 209 int carp_set_addr6(struct carp_softc *, struct sockaddr_in6 *); 210 int carp_join_multicast6(struct carp_softc *); 211 #endif 212 int carp_clone_create(struct if_clone *, int); 213 int carp_clone_destroy(struct ifnet *); 214 int carp_ether_addmulti(struct carp_softc *, struct ifreq *); 215 int carp_ether_delmulti(struct carp_softc *, struct ifreq *); 216 void carp_ether_purgemulti(struct carp_softc *); 217 218 struct if_clone carp_cloner = 219 IF_CLONE_INITIALIZER("carp", carp_clone_create, carp_clone_destroy); 220 221 static __inline u_int16_t 222 carp_cksum(struct mbuf *m, int len) 223 { 224 return (in_cksum(m, len)); 225 } 226 227 void 228 carp_hmac_prepare(struct carp_softc *sc) 229 { 230 u_int8_t carp_version = CARP_VERSION, type = CARP_ADVERTISEMENT; 231 u_int8_t vhid = sc->sc_vhid & 0xff; 232 SHA1_CTX sha1ctx; 233 u_int32_t kmd[5]; 234 struct ifaddr *ifa; 235 int i, found; 236 struct in_addr last, cur, in; 237 #ifdef INET6 238 struct in6_addr last6, cur6, in6; 239 #endif /* INET6 */ 240 241 /* compute ipad from key */ 242 bzero(sc->sc_pad, sizeof(sc->sc_pad)); 243 bcopy(sc->sc_key, sc->sc_pad, sizeof(sc->sc_key)); 244 for (i = 0; i < sizeof(sc->sc_pad); i++) 245 sc->sc_pad[i] ^= 0x36; 246 247 /* precompute first part of inner hash */ 248 SHA1Init(&sc->sc_sha1); 249 SHA1Update(&sc->sc_sha1, sc->sc_pad, sizeof(sc->sc_pad)); 250 SHA1Update(&sc->sc_sha1, (void *)&carp_version, sizeof(carp_version)); 251 SHA1Update(&sc->sc_sha1, (void *)&type, sizeof(type)); 252 253 /* generate a key for the arpbalance hash, before the vhid is hashed */ 254 bcopy(&sc->sc_sha1, &sha1ctx, sizeof(sha1ctx)); 255 SHA1Final((unsigned char *)kmd, &sha1ctx); 256 sc->sc_hashkey[0] = kmd[0] ^ kmd[1]; 257 sc->sc_hashkey[1] = kmd[2] ^ kmd[3]; 258 259 /* the rest of the precomputation */ 260 SHA1Update(&sc->sc_sha1, (void *)&vhid, sizeof(vhid)); 261 262 /* Hash the addresses from smallest to largest, not interface order */ 263 #ifdef INET 264 cur.s_addr = 0; 265 do { 266 found = 0; 267 last = cur; 268 cur.s_addr = 0xffffffff; 269 IFADDR_FOREACH(ifa, &sc->sc_if) { 270 in.s_addr = ifatoia(ifa)->ia_addr.sin_addr.s_addr; 271 if (ifa->ifa_addr->sa_family == AF_INET && 272 ntohl(in.s_addr) > ntohl(last.s_addr) && 273 ntohl(in.s_addr) < ntohl(cur.s_addr)) { 274 cur.s_addr = in.s_addr; 275 found++; 276 } 277 } 278 if (found) 279 SHA1Update(&sc->sc_sha1, (void *)&cur, sizeof(cur)); 280 } while (found); 281 #endif /* INET */ 282 283 #ifdef INET6 284 memset(&cur6, 0x00, sizeof(cur6)); 285 do { 286 found = 0; 287 last6 = cur6; 288 memset(&cur6, 0xff, sizeof(cur6)); 289 IFADDR_FOREACH(ifa, &sc->sc_if) { 290 in6 = ifatoia6(ifa)->ia_addr.sin6_addr; 291 if (IN6_IS_ADDR_LINKLOCAL(&in6)) 292 in6.s6_addr16[1] = 0; 293 if (ifa->ifa_addr->sa_family == AF_INET6 && 294 memcmp(&in6, &last6, sizeof(in6)) > 0 && 295 memcmp(&in6, &cur6, sizeof(in6)) < 0) { 296 cur6 = in6; 297 found++; 298 } 299 } 300 if (found) 301 SHA1Update(&sc->sc_sha1, (void *)&cur6, sizeof(cur6)); 302 } while (found); 303 #endif /* INET6 */ 304 305 /* convert ipad to opad */ 306 for (i = 0; i < sizeof(sc->sc_pad); i++) 307 sc->sc_pad[i] ^= 0x36 ^ 0x5c; 308 } 309 310 void 311 carp_hmac_generate(struct carp_softc *sc, u_int32_t counter[2], 312 unsigned char md[20]) 313 { 314 SHA1_CTX sha1ctx; 315 316 /* fetch first half of inner hash */ 317 bcopy(&sc->sc_sha1, &sha1ctx, sizeof(sha1ctx)); 318 319 SHA1Update(&sha1ctx, (void *)counter, sizeof(sc->sc_counter)); 320 SHA1Final(md, &sha1ctx); 321 322 /* outer hash */ 323 SHA1Init(&sha1ctx); 324 SHA1Update(&sha1ctx, sc->sc_pad, sizeof(sc->sc_pad)); 325 SHA1Update(&sha1ctx, md, 20); 326 SHA1Final(md, &sha1ctx); 327 } 328 329 int 330 carp_hmac_verify(struct carp_softc *sc, u_int32_t counter[2], 331 unsigned char md[20]) 332 { 333 unsigned char md2[20]; 334 335 carp_hmac_generate(sc, counter, md2); 336 337 return (bcmp(md, md2, sizeof(md2))); 338 } 339 340 void 341 carp_setroute(struct carp_softc *sc, int cmd) 342 { 343 struct ifaddr *ifa; 344 int s; 345 346 s = splsoftnet(); 347 IFADDR_FOREACH(ifa, &sc->sc_if) { 348 switch (ifa->ifa_addr->sa_family) { 349 case AF_INET: { 350 int count = 0; 351 struct rtentry *rt; 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 rt = NULL; 375 (void)rtrequest(RTM_GET, ifa->ifa_addr, ifa->ifa_addr, 376 ifa->ifa_netmask, RTF_HOST, &rt); 377 hr_otherif = (rt && rt->rt_ifp != &sc->sc_if && 378 rt->rt_flags & (RTF_CLONING|RTF_CLONED)); 379 if (rt != NULL) { 380 RTFREE(rt); 381 rt = NULL; 382 } 383 384 /* Check for a network route on our interface */ 385 386 rt = NULL; 387 (void)rtrequest(RTM_GET, ifa->ifa_addr, ifa->ifa_addr, 388 ifa->ifa_netmask, 0, &rt); 389 nr_ourif = (rt && rt->rt_ifp == &sc->sc_if); 390 391 switch (cmd) { 392 case RTM_ADD: 393 if (hr_otherif) { 394 ifa->ifa_rtrequest = NULL; 395 ifa->ifa_flags &= ~RTF_CLONING; 396 397 rtrequest(RTM_ADD, ifa->ifa_addr, 398 ifa->ifa_addr, ifa->ifa_netmask, 399 RTF_UP | RTF_HOST, NULL); 400 } 401 if (!hr_otherif || nr_ourif || !rt) { 402 if (nr_ourif && !(rt->rt_flags & 403 RTF_CLONING)) 404 rtrequest(RTM_DELETE, 405 ifa->ifa_addr, 406 ifa->ifa_addr, 407 ifa->ifa_netmask, 0, NULL); 408 409 ifa->ifa_rtrequest = arp_rtrequest; 410 ifa->ifa_flags |= RTF_CLONING; 411 412 if (rtrequest(RTM_ADD, ifa->ifa_addr, 413 ifa->ifa_addr, ifa->ifa_netmask, 0, 414 NULL) == 0) 415 ifa->ifa_flags |= IFA_ROUTE; 416 } 417 break; 418 case RTM_DELETE: 419 break; 420 default: 421 break; 422 } 423 if (rt != NULL) { 424 RTFREE(rt); 425 rt = NULL; 426 } 427 break; 428 } 429 430 #ifdef INET6 431 case AF_INET6: 432 if (cmd == RTM_ADD) 433 in6_ifaddloop(ifa); 434 else 435 in6_ifremloop(ifa); 436 break; 437 #endif /* INET6 */ 438 default: 439 break; 440 } 441 } 442 splx(s); 443 } 444 445 /* 446 * process input packet. 447 * we have rearranged checks order compared to the rfc, 448 * but it seems more efficient this way or not possible otherwise. 449 */ 450 void 451 carp_proto_input(struct mbuf *m, ...) 452 { 453 struct ip *ip = mtod(m, struct ip *); 454 struct carp_softc *sc = NULL; 455 struct carp_header *ch; 456 int iplen, len, hlen; 457 va_list ap; 458 459 va_start(ap, m); 460 hlen = va_arg(ap, int); 461 va_end(ap); 462 463 carpstats.carps_ipackets++; 464 465 if (!carp_opts[CARPCTL_ALLOW]) { 466 m_freem(m); 467 return; 468 } 469 470 /* check if received on a valid carp interface */ 471 if (m->m_pkthdr.rcvif->if_type != IFT_CARP) { 472 carpstats.carps_badif++; 473 CARP_LOG(sc, ("packet received on non-carp interface: %s", 474 m->m_pkthdr.rcvif->if_xname)); 475 m_freem(m); 476 return; 477 } 478 479 /* verify that the IP TTL is 255. */ 480 if (ip->ip_ttl != CARP_DFLTTL) { 481 carpstats.carps_badttl++; 482 CARP_LOG(sc, ("received ttl %d != %d on %s", ip->ip_ttl, 483 CARP_DFLTTL, m->m_pkthdr.rcvif->if_xname)); 484 m_freem(m); 485 return; 486 } 487 488 /* 489 * verify that the received packet length is 490 * equal to the CARP header 491 */ 492 iplen = ip->ip_hl << 2; 493 len = iplen + sizeof(*ch); 494 if (len > m->m_pkthdr.len) { 495 carpstats.carps_badlen++; 496 CARP_LOG(sc, ("packet too short %d on %s", m->m_pkthdr.len, 497 m->m_pkthdr.rcvif->if_xname)); 498 m_freem(m); 499 return; 500 } 501 502 if ((m = m_pullup(m, len)) == NULL) { 503 carpstats.carps_hdrops++; 504 return; 505 } 506 ip = mtod(m, struct ip *); 507 ch = (struct carp_header *)((char *)ip + iplen); 508 /* verify the CARP checksum */ 509 m->m_data += iplen; 510 if (carp_cksum(m, len - iplen)) { 511 carpstats.carps_badsum++; 512 CARP_LOG(sc, ("checksum failed on %s", 513 m->m_pkthdr.rcvif->if_xname)); 514 m_freem(m); 515 return; 516 } 517 m->m_data -= iplen; 518 519 carp_proto_input_c(m, ch, AF_INET); 520 } 521 522 #ifdef INET6 523 int 524 carp6_proto_input(struct mbuf **mp, int *offp, int proto) 525 { 526 struct mbuf *m = *mp; 527 struct carp_softc *sc = NULL; 528 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 529 struct carp_header *ch; 530 u_int len; 531 532 carpstats.carps_ipackets6++; 533 534 if (!carp_opts[CARPCTL_ALLOW]) { 535 m_freem(m); 536 return (IPPROTO_DONE); 537 } 538 539 /* check if received on a valid carp interface */ 540 if (m->m_pkthdr.rcvif->if_type != IFT_CARP) { 541 carpstats.carps_badif++; 542 CARP_LOG(sc, ("packet received on non-carp interface: %s", 543 m->m_pkthdr.rcvif->if_xname)); 544 m_freem(m); 545 return (IPPROTO_DONE); 546 } 547 548 /* verify that the IP TTL is 255 */ 549 if (ip6->ip6_hlim != CARP_DFLTTL) { 550 carpstats.carps_badttl++; 551 CARP_LOG(sc, ("received ttl %d != %d on %s", ip6->ip6_hlim, 552 CARP_DFLTTL, m->m_pkthdr.rcvif->if_xname)); 553 m_freem(m); 554 return (IPPROTO_DONE); 555 } 556 557 /* verify that we have a complete carp packet */ 558 len = m->m_len; 559 IP6_EXTHDR_GET(ch, struct carp_header *, m, *offp, sizeof(*ch)); 560 if (ch == NULL) { 561 carpstats.carps_badlen++; 562 CARP_LOG(sc, ("packet size %u too small", len)); 563 return (IPPROTO_DONE); 564 } 565 566 567 /* verify the CARP checksum */ 568 m->m_data += *offp; 569 if (carp_cksum(m, sizeof(*ch))) { 570 carpstats.carps_badsum++; 571 CARP_LOG(sc, ("checksum failed, on %s", 572 m->m_pkthdr.rcvif->if_xname)); 573 m_freem(m); 574 return (IPPROTO_DONE); 575 } 576 m->m_data -= *offp; 577 578 carp_proto_input_c(m, ch, AF_INET6); 579 return (IPPROTO_DONE); 580 } 581 #endif /* INET6 */ 582 583 void 584 carp_proto_input_c(struct mbuf *m, struct carp_header *ch, sa_family_t af) 585 { 586 struct carp_softc *sc; 587 u_int64_t tmp_counter; 588 struct timeval sc_tv, ch_tv; 589 590 TAILQ_FOREACH(sc, &((struct carp_if *) 591 m->m_pkthdr.rcvif->if_carpdev->if_carp)->vhif_vrs, sc_list) 592 if (sc->sc_vhid == ch->carp_vhid) 593 break; 594 595 if (!sc || (sc->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) != 596 (IFF_UP|IFF_RUNNING)) { 597 carpstats.carps_badvhid++; 598 m_freem(m); 599 return; 600 } 601 602 /* 603 * Check if our own advertisement was duplicated 604 * from a non simplex interface. 605 * XXX If there is no address on our physical interface 606 * there is no way to distinguish our ads from the ones 607 * another carp host might have sent us. 608 */ 609 if ((sc->sc_carpdev->if_flags & IFF_SIMPLEX) == 0) { 610 struct sockaddr sa; 611 struct ifaddr *ifa; 612 613 bzero(&sa, sizeof(sa)); 614 sa.sa_family = af; 615 ifa = ifaof_ifpforaddr(&sa, sc->sc_carpdev); 616 617 if (ifa && af == AF_INET) { 618 struct ip *ip = mtod(m, struct ip *); 619 if (ip->ip_src.s_addr == 620 ifatoia(ifa)->ia_addr.sin_addr.s_addr) { 621 m_freem(m); 622 return; 623 } 624 } 625 #ifdef INET6 626 if (ifa && af == AF_INET6) { 627 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 628 struct in6_addr in6_src, in6_found; 629 630 in6_src = ip6->ip6_src; 631 in6_found = ifatoia6(ifa)->ia_addr.sin6_addr; 632 if (IN6_IS_ADDR_LINKLOCAL(&in6_src)) 633 in6_src.s6_addr16[1] = 0; 634 if (IN6_IS_ADDR_LINKLOCAL(&in6_found)) 635 in6_found.s6_addr16[1] = 0; 636 if (IN6_ARE_ADDR_EQUAL(&in6_src, &in6_found)) { 637 m_freem(m); 638 return; 639 } 640 } 641 #endif /* INET6 */ 642 } 643 644 microtime(&sc->sc_if.if_lastchange); 645 sc->sc_if.if_ipackets++; 646 sc->sc_if.if_ibytes += m->m_pkthdr.len; 647 648 /* verify the CARP version. */ 649 if (ch->carp_version != CARP_VERSION) { 650 carpstats.carps_badver++; 651 sc->sc_if.if_ierrors++; 652 CARP_LOG(sc, ("invalid version %d != %d", 653 ch->carp_version, CARP_VERSION)); 654 m_freem(m); 655 return; 656 } 657 658 /* verify the hash */ 659 if (carp_hmac_verify(sc, ch->carp_counter, ch->carp_md)) { 660 carpstats.carps_badauth++; 661 sc->sc_if.if_ierrors++; 662 CARP_LOG(sc, ("incorrect hash")); 663 m_freem(m); 664 return; 665 } 666 667 tmp_counter = ntohl(ch->carp_counter[0]); 668 tmp_counter = tmp_counter<<32; 669 tmp_counter += ntohl(ch->carp_counter[1]); 670 671 /* XXX Replay protection goes here */ 672 673 sc->sc_init_counter = 0; 674 sc->sc_counter = tmp_counter; 675 676 677 sc_tv.tv_sec = sc->sc_advbase; 678 if (carp_suppress_preempt && sc->sc_advskew < 240) 679 sc_tv.tv_usec = 240 * 1000000 / 256; 680 else 681 sc_tv.tv_usec = sc->sc_advskew * 1000000 / 256; 682 ch_tv.tv_sec = ch->carp_advbase; 683 ch_tv.tv_usec = ch->carp_advskew * 1000000 / 256; 684 685 switch (sc->sc_state) { 686 case INIT: 687 break; 688 case MASTER: 689 /* 690 * If we receive an advertisement from a backup who's going to 691 * be more frequent than us, go into BACKUP state. 692 */ 693 if (timercmp(&sc_tv, &ch_tv, >) || 694 timercmp(&sc_tv, &ch_tv, ==)) { 695 callout_stop(&sc->sc_ad_tmo); 696 carp_set_state(sc, BACKUP); 697 carp_setrun(sc, 0); 698 carp_setroute(sc, RTM_DELETE); 699 } 700 break; 701 case BACKUP: 702 /* 703 * If we're pre-empting masters who advertise slower than us, 704 * and this one claims to be slower, treat him as down. 705 */ 706 if (carp_opts[CARPCTL_PREEMPT] && timercmp(&sc_tv, &ch_tv, <)) { 707 carp_master_down(sc); 708 break; 709 } 710 711 /* 712 * If the master is going to advertise at such a low frequency 713 * that he's guaranteed to time out, we'd might as well just 714 * treat him as timed out now. 715 */ 716 sc_tv.tv_sec = sc->sc_advbase * 3; 717 if (timercmp(&sc_tv, &ch_tv, <)) { 718 carp_master_down(sc); 719 break; 720 } 721 722 /* 723 * Otherwise, we reset the counter and wait for the next 724 * advertisement. 725 */ 726 carp_setrun(sc, af); 727 break; 728 } 729 730 m_freem(m); 731 return; 732 } 733 734 /* 735 * Interface side of the CARP implementation. 736 */ 737 738 /* ARGSUSED */ 739 void 740 carpattach(int n) 741 { 742 if_clone_attach(&carp_cloner); 743 } 744 745 int 746 carp_clone_create(struct if_clone *ifc, int unit) 747 { 748 extern int ifqmaxlen; 749 struct carp_softc *sc; 750 struct ifnet *ifp; 751 752 sc = malloc(sizeof(*sc), M_DEVBUF, M_NOWAIT); 753 if (!sc) 754 return (ENOMEM); 755 bzero(sc, sizeof(*sc)); 756 757 sc->sc_suppress = 0; 758 sc->sc_advbase = CARP_DFLTINTV; 759 sc->sc_vhid = -1; /* required setting */ 760 sc->sc_advskew = 0; 761 sc->sc_init_counter = 1; 762 sc->sc_naddrs = sc->sc_naddrs6 = 0; 763 #ifdef INET6 764 sc->sc_im6o.im6o_multicast_hlim = CARP_DFLTTL; 765 #endif /* INET6 */ 766 767 callout_init(&sc->sc_ad_tmo, 0); 768 callout_init(&sc->sc_md_tmo, 0); 769 callout_init(&sc->sc_md6_tmo, 0); 770 771 callout_setfunc(&sc->sc_ad_tmo, carp_send_ad, sc); 772 callout_setfunc(&sc->sc_md_tmo, carp_master_down, sc); 773 callout_setfunc(&sc->sc_md6_tmo, carp_master_down, sc); 774 775 LIST_INIT(&sc->carp_mc_listhead); 776 ifp = &sc->sc_if; 777 ifp->if_softc = sc; 778 snprintf(ifp->if_xname, sizeof ifp->if_xname, "%s%d", ifc->ifc_name, 779 unit); 780 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST; 781 ifp->if_ioctl = carp_ioctl; 782 ifp->if_start = carp_start; 783 ifp->if_output = carp_output; 784 ifp->if_type = IFT_CARP; 785 ifp->if_addrlen = ETHER_ADDR_LEN; 786 ifp->if_hdrlen = ETHER_HDR_LEN; 787 ifp->if_mtu = ETHERMTU; 788 IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen); 789 IFQ_SET_READY(&ifp->if_snd); 790 if_attach(ifp); 791 792 if_alloc_sadl(ifp); 793 ifp->if_broadcastaddr = etherbroadcastaddr; 794 carp_set_enaddr(sc); 795 LIST_INIT(&sc->sc_ac.ec_multiaddrs); 796 #if NBPFILTER > 0 797 bpfattach(ifp, DLT_EN10MB, ETHER_HDR_LEN); 798 #endif 799 return (0); 800 } 801 802 int 803 carp_clone_destroy(struct ifnet *ifp) 804 { 805 struct carp_softc *sc = ifp->if_softc; 806 807 carpdetach(ifp->if_softc); 808 ether_ifdetach(ifp); 809 if_detach(ifp); 810 callout_destroy(&sc->sc_ad_tmo); 811 callout_destroy(&sc->sc_md_tmo); 812 callout_destroy(&sc->sc_md6_tmo); 813 free(ifp->if_softc, M_DEVBUF); 814 815 return (0); 816 } 817 818 void 819 carpdetach(struct carp_softc *sc) 820 { 821 struct carp_if *cif; 822 int s; 823 824 callout_stop(&sc->sc_ad_tmo); 825 callout_stop(&sc->sc_md_tmo); 826 callout_stop(&sc->sc_md6_tmo); 827 828 if (sc->sc_suppress) 829 carp_suppress_preempt--; 830 sc->sc_suppress = 0; 831 832 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) 833 carp_suppress_preempt--; 834 sc->sc_sendad_errors = 0; 835 836 carp_set_state(sc, INIT); 837 sc->sc_if.if_flags &= ~IFF_UP; 838 carp_setrun(sc, 0); 839 carp_multicast_cleanup(sc); 840 841 s = splnet(); 842 if (sc->sc_carpdev != NULL) { 843 /* XXX linkstatehook removal */ 844 cif = (struct carp_if *)sc->sc_carpdev->if_carp; 845 TAILQ_REMOVE(&cif->vhif_vrs, sc, sc_list); 846 if (!--cif->vhif_nvrs) { 847 ifpromisc(sc->sc_carpdev, 0); 848 sc->sc_carpdev->if_carp = NULL; 849 FREE(cif, M_IFADDR); 850 } 851 } 852 sc->sc_carpdev = NULL; 853 splx(s); 854 } 855 856 /* Detach an interface from the carp. */ 857 void 858 carp_ifdetach(struct ifnet *ifp) 859 { 860 struct carp_softc *sc, *nextsc; 861 struct carp_if *cif = (struct carp_if *)ifp->if_carp; 862 863 for (sc = TAILQ_FIRST(&cif->vhif_vrs); sc; sc = nextsc) { 864 nextsc = TAILQ_NEXT(sc, sc_list); 865 carpdetach(sc); 866 } 867 } 868 869 int 870 carp_prepare_ad(struct mbuf *m, struct carp_softc *sc, 871 struct carp_header *ch) 872 { 873 if (sc->sc_init_counter) { 874 /* this could also be seconds since unix epoch */ 875 sc->sc_counter = arc4random(); 876 sc->sc_counter = sc->sc_counter << 32; 877 sc->sc_counter += arc4random(); 878 } else 879 sc->sc_counter++; 880 881 ch->carp_counter[0] = htonl((sc->sc_counter>>32)&0xffffffff); 882 ch->carp_counter[1] = htonl(sc->sc_counter&0xffffffff); 883 884 carp_hmac_generate(sc, ch->carp_counter, ch->carp_md); 885 886 return (0); 887 } 888 889 void 890 carp_send_ad_all(void) 891 { 892 struct ifnet *ifp; 893 struct carp_if *cif; 894 struct carp_softc *vh; 895 896 TAILQ_FOREACH(ifp, &ifnet, if_list) { 897 if (ifp->if_carp == NULL || ifp->if_type == IFT_CARP) 898 continue; 899 900 cif = (struct carp_if *)ifp->if_carp; 901 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 902 if ((vh->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) == 903 (IFF_UP|IFF_RUNNING) && vh->sc_state == MASTER) 904 carp_send_ad(vh); 905 } 906 } 907 } 908 909 910 void 911 carp_send_ad(void *v) 912 { 913 struct carp_header ch; 914 struct timeval tv; 915 struct carp_softc *sc = v; 916 struct carp_header *ch_ptr; 917 struct mbuf *m; 918 int error, len, advbase, advskew, s; 919 struct ifaddr *ifa; 920 struct sockaddr sa; 921 922 s = splsoftnet(); 923 924 advbase = advskew = 0; /* Sssssh compiler */ 925 if (sc->sc_carpdev == NULL) { 926 sc->sc_if.if_oerrors++; 927 goto retry_later; 928 } 929 930 /* bow out if we've gone to backup (the carp interface is going down) */ 931 if (sc->sc_bow_out) { 932 sc->sc_bow_out = 0; 933 advbase = 255; 934 advskew = 255; 935 } else { 936 advbase = sc->sc_advbase; 937 if (!carp_suppress_preempt || sc->sc_advskew > 240) 938 advskew = sc->sc_advskew; 939 else 940 advskew = 240; 941 tv.tv_sec = advbase; 942 tv.tv_usec = advskew * 1000000 / 256; 943 } 944 945 ch.carp_version = CARP_VERSION; 946 ch.carp_type = CARP_ADVERTISEMENT; 947 ch.carp_vhid = sc->sc_vhid; 948 ch.carp_advbase = advbase; 949 ch.carp_advskew = advskew; 950 ch.carp_authlen = 7; /* XXX DEFINE */ 951 ch.carp_pad1 = 0; /* must be zero */ 952 ch.carp_cksum = 0; 953 954 955 #ifdef INET 956 if (sc->sc_naddrs) { 957 struct ip *ip; 958 959 MGETHDR(m, M_DONTWAIT, MT_HEADER); 960 if (m == NULL) { 961 sc->sc_if.if_oerrors++; 962 carpstats.carps_onomem++; 963 /* XXX maybe less ? */ 964 goto retry_later; 965 } 966 len = sizeof(*ip) + sizeof(ch); 967 m->m_pkthdr.len = len; 968 m->m_pkthdr.rcvif = NULL; 969 m->m_len = len; 970 MH_ALIGN(m, m->m_len); 971 m->m_flags |= M_MCAST; 972 ip = mtod(m, struct ip *); 973 ip->ip_v = IPVERSION; 974 ip->ip_hl = sizeof(*ip) >> 2; 975 ip->ip_tos = IPTOS_LOWDELAY; 976 ip->ip_len = htons(len); 977 ip->ip_id = 0; /* no need for id, we don't support fragments */ 978 ip->ip_off = htons(IP_DF); 979 ip->ip_ttl = CARP_DFLTTL; 980 ip->ip_p = IPPROTO_CARP; 981 ip->ip_sum = 0; 982 983 bzero(&sa, sizeof(sa)); 984 sa.sa_family = AF_INET; 985 ifa = ifaof_ifpforaddr(&sa, sc->sc_carpdev); 986 if (ifa == NULL) 987 ip->ip_src.s_addr = 0; 988 else 989 ip->ip_src.s_addr = 990 ifatoia(ifa)->ia_addr.sin_addr.s_addr; 991 ip->ip_dst.s_addr = INADDR_CARP_GROUP; 992 993 ch_ptr = (struct carp_header *)(&ip[1]); 994 bcopy(&ch, ch_ptr, sizeof(ch)); 995 if (carp_prepare_ad(m, sc, ch_ptr)) 996 goto retry_later; 997 998 m->m_data += sizeof(*ip); 999 ch_ptr->carp_cksum = carp_cksum(m, len - sizeof(*ip)); 1000 m->m_data -= sizeof(*ip); 1001 1002 microtime(&sc->sc_if.if_lastchange); 1003 sc->sc_if.if_opackets++; 1004 sc->sc_if.if_obytes += len; 1005 carpstats.carps_opackets++; 1006 1007 error = ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo, 1008 NULL); 1009 if (error) { 1010 if (error == ENOBUFS) 1011 carpstats.carps_onomem++; 1012 else 1013 CARP_LOG(sc, ("ip_output failed: %d", error)); 1014 sc->sc_if.if_oerrors++; 1015 if (sc->sc_sendad_errors < INT_MAX) 1016 sc->sc_sendad_errors++; 1017 if (sc->sc_sendad_errors == CARP_SENDAD_MAX_ERRORS) { 1018 carp_suppress_preempt++; 1019 if (carp_suppress_preempt == 1) 1020 carp_send_ad_all(); 1021 } 1022 sc->sc_sendad_success = 0; 1023 } else { 1024 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) { 1025 if (++sc->sc_sendad_success >= 1026 CARP_SENDAD_MIN_SUCCESS) { 1027 carp_suppress_preempt--; 1028 sc->sc_sendad_errors = 0; 1029 } 1030 } else 1031 sc->sc_sendad_errors = 0; 1032 } 1033 } 1034 #endif /* INET */ 1035 #ifdef INET6 1036 if (sc->sc_naddrs6) { 1037 struct ip6_hdr *ip6; 1038 1039 MGETHDR(m, M_DONTWAIT, MT_HEADER); 1040 if (m == NULL) { 1041 sc->sc_if.if_oerrors++; 1042 carpstats.carps_onomem++; 1043 /* XXX maybe less ? */ 1044 goto retry_later; 1045 } 1046 len = sizeof(*ip6) + sizeof(ch); 1047 m->m_pkthdr.len = len; 1048 m->m_pkthdr.rcvif = NULL; 1049 m->m_len = len; 1050 MH_ALIGN(m, m->m_len); 1051 m->m_flags |= M_MCAST; 1052 ip6 = mtod(m, struct ip6_hdr *); 1053 bzero(ip6, sizeof(*ip6)); 1054 ip6->ip6_vfc |= IPV6_VERSION; 1055 ip6->ip6_hlim = CARP_DFLTTL; 1056 ip6->ip6_nxt = IPPROTO_CARP; 1057 1058 /* set the source address */ 1059 bzero(&sa, sizeof(sa)); 1060 sa.sa_family = AF_INET6; 1061 ifa = ifaof_ifpforaddr(&sa, sc->sc_carpdev); 1062 if (ifa == NULL) /* This should never happen with IPv6 */ 1063 bzero(&ip6->ip6_src, sizeof(struct in6_addr)); 1064 else 1065 bcopy(ifatoia6(ifa)->ia_addr.sin6_addr.s6_addr, 1066 &ip6->ip6_src, sizeof(struct in6_addr)); 1067 /* set the multicast destination */ 1068 1069 ip6->ip6_dst.s6_addr16[0] = htons(0xff02); 1070 ip6->ip6_dst.s6_addr8[15] = 0x12; 1071 if (in6_setscope(&ip6->ip6_dst, sc->sc_carpdev, NULL) != 0) { 1072 sc->sc_if.if_oerrors++; 1073 m_freem(m); 1074 CARP_LOG(sc, ("in6_setscope failed")); 1075 goto retry_later; 1076 } 1077 1078 ch_ptr = (struct carp_header *)(&ip6[1]); 1079 bcopy(&ch, ch_ptr, sizeof(ch)); 1080 if (carp_prepare_ad(m, sc, ch_ptr)) 1081 goto retry_later; 1082 1083 m->m_data += sizeof(*ip6); 1084 ch_ptr->carp_cksum = carp_cksum(m, len - sizeof(*ip6)); 1085 m->m_data -= sizeof(*ip6); 1086 1087 microtime(&sc->sc_if.if_lastchange); 1088 sc->sc_if.if_opackets++; 1089 sc->sc_if.if_obytes += len; 1090 carpstats.carps_opackets6++; 1091 1092 error = ip6_output(m, NULL, NULL, 0, &sc->sc_im6o, NULL, NULL); 1093 if (error) { 1094 if (error == ENOBUFS) 1095 carpstats.carps_onomem++; 1096 else 1097 CARP_LOG(sc, ("ip6_output failed: %d", error)); 1098 sc->sc_if.if_oerrors++; 1099 if (sc->sc_sendad_errors < INT_MAX) 1100 sc->sc_sendad_errors++; 1101 if (sc->sc_sendad_errors == CARP_SENDAD_MAX_ERRORS) { 1102 carp_suppress_preempt++; 1103 if (carp_suppress_preempt == 1) 1104 carp_send_ad_all(); 1105 } 1106 sc->sc_sendad_success = 0; 1107 } else { 1108 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) { 1109 if (++sc->sc_sendad_success >= 1110 CARP_SENDAD_MIN_SUCCESS) { 1111 carp_suppress_preempt--; 1112 sc->sc_sendad_errors = 0; 1113 } 1114 } else 1115 sc->sc_sendad_errors = 0; 1116 } 1117 } 1118 #endif /* INET6 */ 1119 1120 retry_later: 1121 splx(s); 1122 if (advbase != 255 || advskew != 255) 1123 callout_schedule(&sc->sc_ad_tmo, tvtohz(&tv)); 1124 } 1125 1126 /* 1127 * Broadcast a gratuitous ARP request containing 1128 * the virtual router MAC address for each IP address 1129 * associated with the virtual router. 1130 */ 1131 void 1132 carp_send_arp(struct carp_softc *sc) 1133 { 1134 struct ifaddr *ifa; 1135 struct in_addr *in; 1136 int s = splsoftnet(); 1137 1138 IFADDR_FOREACH(ifa, &sc->sc_if) { 1139 1140 if (ifa->ifa_addr->sa_family != AF_INET) 1141 continue; 1142 1143 in = &ifatoia(ifa)->ia_addr.sin_addr; 1144 arprequest(sc->sc_carpdev, in, in, CLLADDR(sc->sc_if.if_sadl)); 1145 DELAY(1000); /* XXX */ 1146 } 1147 splx(s); 1148 } 1149 1150 #ifdef INET6 1151 void 1152 carp_send_na(struct carp_softc *sc) 1153 { 1154 struct ifaddr *ifa; 1155 struct in6_addr *in6; 1156 static struct in6_addr mcast = IN6ADDR_LINKLOCAL_ALLNODES_INIT; 1157 int s = splsoftnet(); 1158 1159 IFADDR_FOREACH(ifa, &sc->sc_if) { 1160 1161 if (ifa->ifa_addr->sa_family != AF_INET6) 1162 continue; 1163 1164 in6 = &ifatoia6(ifa)->ia_addr.sin6_addr; 1165 nd6_na_output(sc->sc_carpdev, &mcast, in6, 1166 ND_NA_FLAG_OVERRIDE, 1, NULL); 1167 DELAY(1000); /* XXX */ 1168 } 1169 splx(s); 1170 } 1171 #endif /* INET6 */ 1172 1173 /* 1174 * Based on bridge_hash() in if_bridge.c 1175 */ 1176 #define mix(a,b,c) \ 1177 do { \ 1178 a -= b; a -= c; a ^= (c >> 13); \ 1179 b -= c; b -= a; b ^= (a << 8); \ 1180 c -= a; c -= b; c ^= (b >> 13); \ 1181 a -= b; a -= c; a ^= (c >> 12); \ 1182 b -= c; b -= a; b ^= (a << 16); \ 1183 c -= a; c -= b; c ^= (b >> 5); \ 1184 a -= b; a -= c; a ^= (c >> 3); \ 1185 b -= c; b -= a; b ^= (a << 10); \ 1186 c -= a; c -= b; c ^= (b >> 15); \ 1187 } while (0) 1188 1189 u_int32_t 1190 carp_hash(struct carp_softc *sc, u_char *src) 1191 { 1192 u_int32_t a = 0x9e3779b9, b = sc->sc_hashkey[0], c = sc->sc_hashkey[1]; 1193 1194 c += sc->sc_key[3] << 24; 1195 c += sc->sc_key[2] << 16; 1196 c += sc->sc_key[1] << 8; 1197 c += sc->sc_key[0]; 1198 b += src[5] << 8; 1199 b += src[4]; 1200 a += src[3] << 24; 1201 a += src[2] << 16; 1202 a += src[1] << 8; 1203 a += src[0]; 1204 1205 mix(a, b, c); 1206 return (c); 1207 } 1208 1209 int 1210 carp_addrcount(struct carp_if *cif, struct in_ifaddr *ia, int type) 1211 { 1212 struct carp_softc *vh; 1213 struct ifaddr *ifa; 1214 int count = 0; 1215 1216 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 1217 if ((type == CARP_COUNT_RUNNING && 1218 (vh->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) == 1219 (IFF_UP|IFF_RUNNING)) || 1220 (type == CARP_COUNT_MASTER && vh->sc_state == MASTER)) { 1221 IFADDR_FOREACH(ifa, &vh->sc_if) { 1222 if (ifa->ifa_addr->sa_family == AF_INET && 1223 ia->ia_addr.sin_addr.s_addr == 1224 ifatoia(ifa)->ia_addr.sin_addr.s_addr) 1225 count++; 1226 } 1227 } 1228 } 1229 return (count); 1230 } 1231 1232 int 1233 carp_iamatch(struct in_ifaddr *ia, u_char *src, 1234 u_int32_t *count, u_int32_t index) 1235 { 1236 struct carp_softc *sc = ia->ia_ifp->if_softc; 1237 1238 if (carp_opts[CARPCTL_ARPBALANCE]) { 1239 /* 1240 * We use the source ip to decide which virtual host should 1241 * handle the request. If we're master of that virtual host, 1242 * then we respond, otherwise, just drop the arp packet on 1243 * the floor. 1244 */ 1245 1246 /* Count the elegible carp interfaces with this address */ 1247 if (*count == 0) 1248 *count = carp_addrcount( 1249 (struct carp_if *)ia->ia_ifp->if_carpdev->if_carp, 1250 ia, CARP_COUNT_RUNNING); 1251 1252 /* This should never happen, but... */ 1253 if (*count == 0) 1254 return (0); 1255 1256 if (carp_hash(sc, src) % *count == index - 1 && 1257 sc->sc_state == MASTER) { 1258 return (1); 1259 } 1260 } else { 1261 if (sc->sc_state == MASTER) 1262 return (1); 1263 } 1264 1265 return (0); 1266 } 1267 1268 #ifdef INET6 1269 struct ifaddr * 1270 carp_iamatch6(void *v, struct in6_addr *taddr) 1271 { 1272 struct carp_if *cif = v; 1273 struct carp_softc *vh; 1274 struct ifaddr *ifa; 1275 1276 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 1277 IFADDR_FOREACH(ifa, &vh->sc_if) { 1278 if (IN6_ARE_ADDR_EQUAL(taddr, 1279 &ifatoia6(ifa)->ia_addr.sin6_addr) && 1280 ((vh->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) == 1281 (IFF_UP|IFF_RUNNING)) && vh->sc_state == MASTER) 1282 return (ifa); 1283 } 1284 } 1285 1286 return (NULL); 1287 } 1288 #endif /* INET6 */ 1289 1290 struct ifnet * 1291 carp_ourether(void *v, struct ether_header *eh, u_char iftype, int src) 1292 { 1293 struct carp_if *cif = (struct carp_if *)v; 1294 struct carp_softc *vh; 1295 u_int8_t *ena; 1296 1297 if (src) 1298 ena = (u_int8_t *)&eh->ether_shost; 1299 else 1300 ena = (u_int8_t *)&eh->ether_dhost; 1301 1302 switch (iftype) { 1303 case IFT_ETHER: 1304 case IFT_FDDI: 1305 if (ena[0] || ena[1] || ena[2] != 0x5e || ena[3] || ena[4] != 1) 1306 return (NULL); 1307 break; 1308 case IFT_ISO88025: 1309 if (ena[0] != 3 || ena[1] || ena[4] || ena[5]) 1310 return (NULL); 1311 break; 1312 default: 1313 return (NULL); 1314 break; 1315 } 1316 1317 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) 1318 if ((vh->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) == 1319 (IFF_UP|IFF_RUNNING) && vh->sc_state == MASTER && 1320 !bcmp(ena, CLLADDR(vh->sc_if.if_sadl), 1321 ETHER_ADDR_LEN)) { 1322 return (&vh->sc_if); 1323 } 1324 1325 return (NULL); 1326 } 1327 1328 int 1329 carp_input(struct mbuf *m, u_int8_t *shost, u_int8_t *dhost, u_int16_t etype) 1330 { 1331 struct ether_header eh; 1332 struct carp_if *cif = (struct carp_if *)m->m_pkthdr.rcvif->if_carp; 1333 struct ifnet *ifp; 1334 1335 bcopy(shost, &eh.ether_shost, sizeof(eh.ether_shost)); 1336 bcopy(dhost, &eh.ether_dhost, sizeof(eh.ether_dhost)); 1337 eh.ether_type = etype; 1338 1339 if (m->m_flags & (M_BCAST|M_MCAST)) { 1340 struct carp_softc *vh; 1341 struct mbuf *m0; 1342 1343 /* 1344 * XXX Should really check the list of multicast addresses 1345 * for each CARP interface _before_ copying. 1346 */ 1347 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 1348 m0 = m_copym(m, 0, M_COPYALL, M_DONTWAIT); 1349 if (m0 == NULL) 1350 continue; 1351 m0->m_pkthdr.rcvif = &vh->sc_if; 1352 ether_input(&vh->sc_if, m0); 1353 } 1354 return (1); 1355 } 1356 1357 ifp = carp_ourether(cif, &eh, m->m_pkthdr.rcvif->if_type, 0); 1358 if (ifp == NULL) { 1359 return (1); 1360 } 1361 1362 m->m_pkthdr.rcvif = ifp; 1363 1364 #if NBPFILTER > 0 1365 if (ifp->if_bpf) 1366 bpf_mtap(ifp->if_bpf, m); 1367 #endif 1368 ifp->if_ipackets++; 1369 ether_input(ifp, m); 1370 return (0); 1371 } 1372 1373 void 1374 carp_master_down(void *v) 1375 { 1376 struct carp_softc *sc = v; 1377 1378 switch (sc->sc_state) { 1379 case INIT: 1380 printf("%s: master_down event in INIT state\n", 1381 sc->sc_if.if_xname); 1382 break; 1383 case MASTER: 1384 break; 1385 case BACKUP: 1386 carp_set_state(sc, MASTER); 1387 carp_send_ad(sc); 1388 carp_send_arp(sc); 1389 #ifdef INET6 1390 carp_send_na(sc); 1391 #endif /* INET6 */ 1392 carp_setrun(sc, 0); 1393 carp_setroute(sc, RTM_ADD); 1394 break; 1395 } 1396 } 1397 1398 /* 1399 * When in backup state, af indicates whether to reset the master down timer 1400 * for v4 or v6. If it's set to zero, reset the ones which are already pending. 1401 */ 1402 void 1403 carp_setrun(struct carp_softc *sc, sa_family_t af) 1404 { 1405 struct timeval tv; 1406 1407 if (sc->sc_carpdev == NULL) { 1408 sc->sc_if.if_flags &= ~IFF_RUNNING; 1409 carp_set_state(sc, INIT); 1410 return; 1411 } 1412 1413 if (sc->sc_if.if_flags & IFF_UP && sc->sc_vhid > 0 && 1414 (sc->sc_naddrs || sc->sc_naddrs6) && !sc->sc_suppress) { 1415 sc->sc_if.if_flags |= IFF_RUNNING; 1416 } else { 1417 sc->sc_if.if_flags &= ~IFF_RUNNING; 1418 carp_setroute(sc, RTM_DELETE); 1419 return; 1420 } 1421 1422 switch (sc->sc_state) { 1423 case INIT: 1424 carp_set_state(sc, BACKUP); 1425 carp_setroute(sc, RTM_DELETE); 1426 carp_setrun(sc, 0); 1427 break; 1428 case BACKUP: 1429 callout_stop(&sc->sc_ad_tmo); 1430 tv.tv_sec = 3 * sc->sc_advbase; 1431 tv.tv_usec = sc->sc_advskew * 1000000 / 256; 1432 switch (af) { 1433 #ifdef INET 1434 case AF_INET: 1435 callout_schedule(&sc->sc_md_tmo, tvtohz(&tv)); 1436 break; 1437 #endif /* INET */ 1438 #ifdef INET6 1439 case AF_INET6: 1440 callout_schedule(&sc->sc_md6_tmo, tvtohz(&tv)); 1441 break; 1442 #endif /* INET6 */ 1443 default: 1444 if (sc->sc_naddrs) 1445 callout_schedule(&sc->sc_md_tmo, tvtohz(&tv)); 1446 if (sc->sc_naddrs6) 1447 callout_schedule(&sc->sc_md6_tmo, tvtohz(&tv)); 1448 break; 1449 } 1450 break; 1451 case MASTER: 1452 tv.tv_sec = sc->sc_advbase; 1453 tv.tv_usec = sc->sc_advskew * 1000000 / 256; 1454 callout_schedule(&sc->sc_ad_tmo, tvtohz(&tv)); 1455 break; 1456 } 1457 } 1458 1459 void 1460 carp_multicast_cleanup(struct carp_softc *sc) 1461 { 1462 struct ip_moptions *imo = &sc->sc_imo; 1463 #ifdef INET6 1464 struct ip6_moptions *im6o = &sc->sc_im6o; 1465 #endif 1466 u_int16_t n = imo->imo_num_memberships; 1467 1468 /* Clean up our own multicast memberships */ 1469 while (n-- > 0) { 1470 if (imo->imo_membership[n] != NULL) { 1471 in_delmulti(imo->imo_membership[n]); 1472 imo->imo_membership[n] = NULL; 1473 } 1474 } 1475 imo->imo_num_memberships = 0; 1476 imo->imo_multicast_ifp = NULL; 1477 1478 #ifdef INET6 1479 while (!LIST_EMPTY(&im6o->im6o_memberships)) { 1480 struct in6_multi_mship *imm = 1481 LIST_FIRST(&im6o->im6o_memberships); 1482 1483 LIST_REMOVE(imm, i6mm_chain); 1484 in6_leavegroup(imm); 1485 } 1486 im6o->im6o_multicast_ifp = NULL; 1487 #endif 1488 1489 /* And any other multicast memberships */ 1490 carp_ether_purgemulti(sc); 1491 } 1492 1493 int 1494 carp_set_ifp(struct carp_softc *sc, struct ifnet *ifp) 1495 { 1496 struct carp_if *cif, *ncif = NULL; 1497 struct carp_softc *vr, *after = NULL; 1498 int myself = 0, error = 0; 1499 int s; 1500 1501 if (ifp == sc->sc_carpdev) 1502 return (0); 1503 1504 if (ifp != NULL) { 1505 if ((ifp->if_flags & IFF_MULTICAST) == 0) 1506 return (EADDRNOTAVAIL); 1507 1508 if (ifp->if_type == IFT_CARP) 1509 return (EINVAL); 1510 1511 if (ifp->if_carp == NULL) { 1512 MALLOC(ncif, struct carp_if *, sizeof(*cif), 1513 M_IFADDR, M_NOWAIT); 1514 if (ncif == NULL) 1515 return (ENOBUFS); 1516 if ((error = ifpromisc(ifp, 1))) { 1517 FREE(ncif, M_IFADDR); 1518 return (error); 1519 } 1520 1521 ncif->vhif_ifp = ifp; 1522 TAILQ_INIT(&ncif->vhif_vrs); 1523 } else { 1524 cif = (struct carp_if *)ifp->if_carp; 1525 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) 1526 if (vr != sc && vr->sc_vhid == sc->sc_vhid) 1527 return (EINVAL); 1528 } 1529 1530 /* detach from old interface */ 1531 if (sc->sc_carpdev != NULL) 1532 carpdetach(sc); 1533 1534 /* join multicast groups */ 1535 if (sc->sc_naddrs < 0 && 1536 (error = carp_join_multicast(sc)) != 0) { 1537 if (ncif != NULL) 1538 FREE(ncif, M_IFADDR); 1539 return (error); 1540 } 1541 1542 #ifdef INET6 1543 if (sc->sc_naddrs6 < 0 && 1544 (error = carp_join_multicast6(sc)) != 0) { 1545 if (ncif != NULL) 1546 FREE(ncif, M_IFADDR); 1547 carp_multicast_cleanup(sc); 1548 return (error); 1549 } 1550 #endif 1551 1552 /* attach carp interface to physical interface */ 1553 if (ncif != NULL) 1554 ifp->if_carp = (void *)ncif; 1555 sc->sc_carpdev = ifp; 1556 cif = (struct carp_if *)ifp->if_carp; 1557 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) { 1558 if (vr == sc) 1559 myself = 1; 1560 if (vr->sc_vhid < sc->sc_vhid) 1561 after = vr; 1562 } 1563 1564 if (!myself) { 1565 /* We're trying to keep things in order */ 1566 if (after == NULL) { 1567 TAILQ_INSERT_TAIL(&cif->vhif_vrs, sc, sc_list); 1568 } else { 1569 TAILQ_INSERT_AFTER(&cif->vhif_vrs, after, 1570 sc, sc_list); 1571 } 1572 cif->vhif_nvrs++; 1573 } 1574 if (sc->sc_naddrs || sc->sc_naddrs6) 1575 sc->sc_if.if_flags |= IFF_UP; 1576 carp_set_enaddr(sc); 1577 s = splnet(); 1578 /* XXX linkstatehooks establish */ 1579 carp_carpdev_state(ifp); 1580 splx(s); 1581 } else { 1582 carpdetach(sc); 1583 sc->sc_if.if_flags &= ~(IFF_UP|IFF_RUNNING); 1584 } 1585 return (0); 1586 } 1587 1588 void 1589 carp_set_enaddr(struct carp_softc *sc) 1590 { 1591 uint8_t enaddr[ETHER_ADDR_LEN]; 1592 if (sc->sc_carpdev && sc->sc_carpdev->if_type == IFT_ISO88025) { 1593 enaddr[0] = 3; 1594 enaddr[1] = 0; 1595 enaddr[2] = 0x40 >> (sc->sc_vhid - 1); 1596 enaddr[3] = 0x40000 >> (sc->sc_vhid - 1); 1597 enaddr[4] = 0; 1598 enaddr[5] = 0; 1599 } else { 1600 enaddr[0] = 0; 1601 enaddr[1] = 0; 1602 enaddr[2] = 0x5e; 1603 enaddr[3] = 0; 1604 enaddr[4] = 1; 1605 enaddr[5] = sc->sc_vhid; 1606 } 1607 if_set_sadl(&sc->sc_if, enaddr, sizeof(enaddr)); 1608 } 1609 1610 void 1611 carp_addr_updated(void *v) 1612 { 1613 struct carp_softc *sc = (struct carp_softc *) v; 1614 struct ifaddr *ifa; 1615 int new_naddrs = 0, new_naddrs6 = 0; 1616 1617 IFADDR_FOREACH(ifa, &sc->sc_if) { 1618 if (ifa->ifa_addr->sa_family == AF_INET) 1619 new_naddrs++; 1620 else if (ifa->ifa_addr->sa_family == AF_INET6) 1621 new_naddrs6++; 1622 } 1623 1624 /* Handle a callback after SIOCDIFADDR */ 1625 if (new_naddrs < sc->sc_naddrs || new_naddrs6 < sc->sc_naddrs6) { 1626 struct in_addr mc_addr; 1627 struct in_multi *inm; 1628 1629 sc->sc_naddrs = new_naddrs; 1630 sc->sc_naddrs6 = new_naddrs6; 1631 1632 /* Re-establish multicast membership removed by in_control */ 1633 mc_addr.s_addr = INADDR_CARP_GROUP; 1634 IN_LOOKUP_MULTI(mc_addr, &sc->sc_if, inm); 1635 if (inm == NULL) { 1636 bzero(&sc->sc_imo, sizeof(sc->sc_imo)); 1637 1638 if (sc->sc_carpdev != NULL && sc->sc_naddrs > 0) 1639 carp_join_multicast(sc); 1640 } 1641 1642 if (sc->sc_naddrs == 0 && sc->sc_naddrs6 == 0) { 1643 sc->sc_if.if_flags &= ~IFF_UP; 1644 carp_set_state(sc, INIT); 1645 } else 1646 carp_hmac_prepare(sc); 1647 } 1648 1649 carp_setrun(sc, 0); 1650 } 1651 1652 int 1653 carp_set_addr(struct carp_softc *sc, struct sockaddr_in *sin) 1654 { 1655 struct ifnet *ifp = sc->sc_carpdev; 1656 struct in_ifaddr *ia, *ia_if; 1657 int error = 0; 1658 1659 if (sin->sin_addr.s_addr == 0) { 1660 if (!(sc->sc_if.if_flags & IFF_UP)) 1661 carp_set_state(sc, INIT); 1662 if (sc->sc_naddrs) 1663 sc->sc_if.if_flags |= IFF_UP; 1664 carp_setrun(sc, 0); 1665 return (0); 1666 } 1667 1668 /* we have to do this by hand to ensure we don't match on ourselves */ 1669 ia_if = NULL; 1670 for (ia = TAILQ_FIRST(&in_ifaddrhead); ia; 1671 ia = TAILQ_NEXT(ia, ia_list)) { 1672 1673 /* and, yeah, we need a multicast-capable iface too */ 1674 if (ia->ia_ifp != &sc->sc_if && 1675 ia->ia_ifp->if_type != IFT_CARP && 1676 (ia->ia_ifp->if_flags & IFF_MULTICAST) && 1677 (sin->sin_addr.s_addr & ia->ia_subnetmask) == 1678 ia->ia_subnet) { 1679 if (!ia_if) 1680 ia_if = ia; 1681 } 1682 } 1683 1684 if (ia_if) { 1685 ia = ia_if; 1686 if (ifp) { 1687 if (ifp != ia->ia_ifp) 1688 return (EADDRNOTAVAIL); 1689 } else { 1690 ifp = ia->ia_ifp; 1691 } 1692 } 1693 1694 if ((error = carp_set_ifp(sc, ifp))) 1695 return (error); 1696 1697 if (sc->sc_carpdev == NULL) 1698 return (EADDRNOTAVAIL); 1699 1700 if (sc->sc_naddrs == 0 && (error = carp_join_multicast(sc)) != 0) 1701 return (error); 1702 1703 sc->sc_naddrs++; 1704 if (sc->sc_carpdev != NULL) 1705 sc->sc_if.if_flags |= IFF_UP; 1706 1707 carp_set_state(sc, INIT); 1708 carp_setrun(sc, 0); 1709 1710 /* 1711 * Hook if_addrhooks so that we get a callback after in_ifinit has run, 1712 * to correct any inappropriate routes that it inserted. 1713 */ 1714 if (sc->ah_cookie == 0) { 1715 /* XXX link address hook */ 1716 } 1717 1718 return (0); 1719 } 1720 1721 int 1722 carp_join_multicast(struct carp_softc *sc) 1723 { 1724 struct ip_moptions *imo = &sc->sc_imo, tmpimo; 1725 struct in_addr addr; 1726 1727 bzero(&tmpimo, sizeof(tmpimo)); 1728 addr.s_addr = INADDR_CARP_GROUP; 1729 if ((tmpimo.imo_membership[0] = 1730 in_addmulti(&addr, &sc->sc_if)) == NULL) { 1731 return (ENOBUFS); 1732 } 1733 1734 imo->imo_membership[0] = tmpimo.imo_membership[0]; 1735 imo->imo_num_memberships = 1; 1736 imo->imo_multicast_ifp = &sc->sc_if; 1737 imo->imo_multicast_ttl = CARP_DFLTTL; 1738 imo->imo_multicast_loop = 0; 1739 return (0); 1740 } 1741 1742 1743 #ifdef INET6 1744 int 1745 carp_set_addr6(struct carp_softc *sc, struct sockaddr_in6 *sin6) 1746 { 1747 struct ifnet *ifp = sc->sc_carpdev; 1748 struct in6_ifaddr *ia, *ia_if; 1749 int error = 0; 1750 1751 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) { 1752 if (!(sc->sc_if.if_flags & IFF_UP)) 1753 carp_set_state(sc, INIT); 1754 if (sc->sc_naddrs6) 1755 sc->sc_if.if_flags |= IFF_UP; 1756 carp_setrun(sc, 0); 1757 return (0); 1758 } 1759 1760 /* we have to do this by hand to ensure we don't match on ourselves */ 1761 ia_if = NULL; 1762 for (ia = in6_ifaddr; ia; ia = ia->ia_next) { 1763 int i; 1764 1765 for (i = 0; i < 4; i++) { 1766 if ((sin6->sin6_addr.s6_addr32[i] & 1767 ia->ia_prefixmask.sin6_addr.s6_addr32[i]) != 1768 (ia->ia_addr.sin6_addr.s6_addr32[i] & 1769 ia->ia_prefixmask.sin6_addr.s6_addr32[i])) 1770 break; 1771 } 1772 /* and, yeah, we need a multicast-capable iface too */ 1773 if (ia->ia_ifp != &sc->sc_if && 1774 ia->ia_ifp->if_type != IFT_CARP && 1775 (ia->ia_ifp->if_flags & IFF_MULTICAST) && 1776 (i == 4)) { 1777 if (!ia_if) 1778 ia_if = ia; 1779 } 1780 } 1781 1782 if (ia_if) { 1783 ia = ia_if; 1784 if (sc->sc_carpdev) { 1785 if (sc->sc_carpdev != ia->ia_ifp) 1786 return (EADDRNOTAVAIL); 1787 } else { 1788 ifp = ia->ia_ifp; 1789 } 1790 } 1791 1792 if ((error = carp_set_ifp(sc, ifp))) 1793 return (error); 1794 1795 if (sc->sc_carpdev == NULL) 1796 return (EADDRNOTAVAIL); 1797 1798 if (sc->sc_naddrs6 == 0 && (error = carp_join_multicast6(sc)) != 0) 1799 return (error); 1800 1801 sc->sc_naddrs6++; 1802 if (sc->sc_carpdev != NULL) 1803 sc->sc_if.if_flags |= IFF_UP; 1804 carp_set_state(sc, INIT); 1805 carp_setrun(sc, 0); 1806 1807 return (0); 1808 } 1809 1810 int 1811 carp_join_multicast6(struct carp_softc *sc) 1812 { 1813 struct in6_multi_mship *imm, *imm2; 1814 struct ip6_moptions *im6o = &sc->sc_im6o; 1815 struct sockaddr_in6 addr6; 1816 int error; 1817 1818 /* Join IPv6 CARP multicast group */ 1819 bzero(&addr6, sizeof(addr6)); 1820 addr6.sin6_family = AF_INET6; 1821 addr6.sin6_len = sizeof(addr6); 1822 addr6.sin6_addr.s6_addr16[0] = htons(0xff02); 1823 addr6.sin6_addr.s6_addr16[1] = htons(sc->sc_if.if_index); 1824 addr6.sin6_addr.s6_addr8[15] = 0x12; 1825 if ((imm = in6_joingroup(&sc->sc_if, 1826 &addr6.sin6_addr, &error, 0)) == NULL) { 1827 return (error); 1828 } 1829 /* join solicited multicast address */ 1830 bzero(&addr6.sin6_addr, sizeof(addr6.sin6_addr)); 1831 addr6.sin6_addr.s6_addr16[0] = htons(0xff02); 1832 addr6.sin6_addr.s6_addr16[1] = htons(sc->sc_if.if_index); 1833 addr6.sin6_addr.s6_addr32[1] = 0; 1834 addr6.sin6_addr.s6_addr32[2] = htonl(1); 1835 addr6.sin6_addr.s6_addr32[3] = 0; 1836 addr6.sin6_addr.s6_addr8[12] = 0xff; 1837 if ((imm2 = in6_joingroup(&sc->sc_if, 1838 &addr6.sin6_addr, &error, 0)) == NULL) { 1839 in6_leavegroup(imm); 1840 return (error); 1841 } 1842 1843 /* apply v6 multicast membership */ 1844 im6o->im6o_multicast_ifp = &sc->sc_if; 1845 if (imm) 1846 LIST_INSERT_HEAD(&im6o->im6o_memberships, imm, 1847 i6mm_chain); 1848 if (imm2) 1849 LIST_INSERT_HEAD(&im6o->im6o_memberships, imm2, 1850 i6mm_chain); 1851 1852 return (0); 1853 } 1854 1855 #endif /* INET6 */ 1856 1857 int 1858 carp_ioctl(struct ifnet *ifp, u_long cmd, void *addr) 1859 { 1860 struct lwp *l = curlwp; /* XXX */ 1861 struct carp_softc *sc = ifp->if_softc, *vr; 1862 struct carpreq carpr; 1863 struct ifaddr *ifa; 1864 struct ifreq *ifr; 1865 struct ifnet *cdev = NULL; 1866 int error = 0; 1867 1868 ifa = (struct ifaddr *)addr; 1869 ifr = (struct ifreq *)addr; 1870 1871 switch (cmd) { 1872 case SIOCSIFADDR: 1873 switch (ifa->ifa_addr->sa_family) { 1874 #ifdef INET 1875 case AF_INET: 1876 sc->sc_if.if_flags |= IFF_UP; 1877 bcopy(ifa->ifa_addr, ifa->ifa_dstaddr, 1878 sizeof(struct sockaddr)); 1879 error = carp_set_addr(sc, satosin(ifa->ifa_addr)); 1880 break; 1881 #endif /* INET */ 1882 #ifdef INET6 1883 case AF_INET6: 1884 sc->sc_if.if_flags|= IFF_UP; 1885 error = carp_set_addr6(sc, satosin6(ifa->ifa_addr)); 1886 break; 1887 #endif /* INET6 */ 1888 default: 1889 error = EAFNOSUPPORT; 1890 break; 1891 } 1892 break; 1893 1894 case SIOCSIFFLAGS: 1895 if (sc->sc_state != INIT && !(ifr->ifr_flags & IFF_UP)) { 1896 callout_stop(&sc->sc_ad_tmo); 1897 callout_stop(&sc->sc_md_tmo); 1898 callout_stop(&sc->sc_md6_tmo); 1899 if (sc->sc_state == MASTER) { 1900 /* we need the interface up to bow out */ 1901 sc->sc_if.if_flags |= IFF_UP; 1902 sc->sc_bow_out = 1; 1903 carp_send_ad(sc); 1904 } 1905 sc->sc_if.if_flags &= ~IFF_UP; 1906 carp_set_state(sc, INIT); 1907 carp_setrun(sc, 0); 1908 } else if (sc->sc_state == INIT && (ifr->ifr_flags & IFF_UP)) { 1909 sc->sc_if.if_flags |= IFF_UP; 1910 carp_setrun(sc, 0); 1911 } 1912 break; 1913 1914 case SIOCSVH: 1915 if (l == NULL) 1916 break; 1917 if ((error = kauth_authorize_network(l->l_cred, 1918 KAUTH_NETWORK_INTERFACE, 1919 KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd, 1920 NULL)) != 0) 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 == NULL) || (error = kauth_authorize_network(l->l_cred, 1996 KAUTH_NETWORK_INTERFACE, 1997 KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd, 1998 NULL)) != 0) 1999 bcopy(sc->sc_key, carpr.carpr_key, 2000 sizeof(carpr.carpr_key)); 2001 error = copyout(&carpr, ifr->ifr_data, sizeof(carpr)); 2002 break; 2003 2004 case SIOCADDMULTI: 2005 error = carp_ether_addmulti(sc, ifr); 2006 break; 2007 2008 case SIOCDELMULTI: 2009 error = carp_ether_delmulti(sc, ifr); 2010 break; 2011 2012 default: 2013 error = EINVAL; 2014 } 2015 2016 carp_hmac_prepare(sc); 2017 return (error); 2018 } 2019 2020 2021 /* 2022 * Start output on carp interface. This function should never be called. 2023 */ 2024 void 2025 carp_start(struct ifnet *ifp) 2026 { 2027 #ifdef DEBUG 2028 printf("%s: start called\n", ifp->if_xname); 2029 #endif 2030 } 2031 2032 int 2033 carp_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *sa, 2034 struct rtentry *rt) 2035 { 2036 struct carp_softc *sc = ((struct carp_softc *)ifp->if_softc); 2037 2038 if (sc->sc_carpdev != NULL && sc->sc_state == MASTER) { 2039 return (sc->sc_carpdev->if_output(ifp, m, sa, rt)); 2040 } else { 2041 m_freem(m); 2042 return (ENETUNREACH); 2043 } 2044 } 2045 2046 void 2047 carp_set_state(struct carp_softc *sc, int state) 2048 { 2049 if (sc->sc_state == state) 2050 return; 2051 2052 sc->sc_state = state; 2053 switch (state) { 2054 case BACKUP: 2055 sc->sc_if.if_link_state = LINK_STATE_DOWN; 2056 break; 2057 case MASTER: 2058 sc->sc_if.if_link_state = LINK_STATE_UP; 2059 break; 2060 default: 2061 sc->sc_if.if_link_state = LINK_STATE_UNKNOWN; 2062 break; 2063 } 2064 rt_ifmsg(&sc->sc_if); 2065 } 2066 2067 void 2068 carp_carpdev_state(void *v) 2069 { 2070 struct carp_if *cif; 2071 struct carp_softc *sc; 2072 struct ifnet *ifp = v; 2073 2074 if (ifp->if_type == IFT_CARP) 2075 return; 2076 2077 cif = (struct carp_if *)ifp->if_carp; 2078 2079 TAILQ_FOREACH(sc, &cif->vhif_vrs, sc_list) { 2080 int suppressed = sc->sc_suppress; 2081 2082 if (sc->sc_carpdev->if_link_state == LINK_STATE_DOWN || 2083 !(sc->sc_carpdev->if_flags & IFF_UP)) { 2084 sc->sc_if.if_flags &= ~IFF_RUNNING; 2085 callout_stop(&sc->sc_ad_tmo); 2086 callout_stop(&sc->sc_md_tmo); 2087 callout_stop(&sc->sc_md6_tmo); 2088 carp_set_state(sc, INIT); 2089 sc->sc_suppress = 1; 2090 carp_setrun(sc, 0); 2091 if (!suppressed) { 2092 carp_suppress_preempt++; 2093 if (carp_suppress_preempt == 1) 2094 carp_send_ad_all(); 2095 } 2096 } else { 2097 carp_set_state(sc, INIT); 2098 sc->sc_suppress = 0; 2099 carp_setrun(sc, 0); 2100 if (suppressed) 2101 carp_suppress_preempt--; 2102 } 2103 } 2104 } 2105 2106 int 2107 carp_ether_addmulti(struct carp_softc *sc, struct ifreq *ifr) 2108 { 2109 const struct sockaddr *sa = ifreq_getaddr(SIOCADDMULTI, ifr); 2110 struct ifnet *ifp; 2111 struct carp_mc_entry *mc; 2112 u_int8_t addrlo[ETHER_ADDR_LEN], addrhi[ETHER_ADDR_LEN]; 2113 int error; 2114 2115 ifp = sc->sc_carpdev; 2116 if (ifp == NULL) 2117 return (EINVAL); 2118 2119 error = ether_addmulti(sa, &sc->sc_ac); 2120 if (error != ENETRESET) 2121 return (error); 2122 2123 /* 2124 * This is new multicast address. We have to tell parent 2125 * about it. Also, remember this multicast address so that 2126 * we can delete them on unconfigure. 2127 */ 2128 MALLOC(mc, struct carp_mc_entry *, sizeof(struct carp_mc_entry), 2129 M_DEVBUF, M_NOWAIT); 2130 if (mc == NULL) { 2131 error = ENOMEM; 2132 goto alloc_failed; 2133 } 2134 2135 /* 2136 * As ether_addmulti() returns ENETRESET, following two 2137 * statement shouldn't fail. 2138 */ 2139 (void)ether_multiaddr(sa, addrlo, addrhi); 2140 ETHER_LOOKUP_MULTI(addrlo, addrhi, &sc->sc_ac, mc->mc_enm); 2141 memcpy(&mc->mc_addr, sa, sa->sa_len); 2142 LIST_INSERT_HEAD(&sc->carp_mc_listhead, mc, mc_entries); 2143 2144 error = (*ifp->if_ioctl)(ifp, SIOCADDMULTI, (void *)ifr); 2145 if (error != 0) 2146 goto ioctl_failed; 2147 2148 return (error); 2149 2150 ioctl_failed: 2151 LIST_REMOVE(mc, mc_entries); 2152 FREE(mc, M_DEVBUF); 2153 alloc_failed: 2154 (void)ether_delmulti(sa, &sc->sc_ac); 2155 2156 return (error); 2157 } 2158 2159 int 2160 carp_ether_delmulti(struct carp_softc *sc, struct ifreq *ifr) 2161 { 2162 const struct sockaddr *sa = ifreq_getaddr(SIOCDELMULTI, ifr); 2163 struct ifnet *ifp; 2164 struct ether_multi *enm; 2165 struct carp_mc_entry *mc; 2166 u_int8_t addrlo[ETHER_ADDR_LEN], addrhi[ETHER_ADDR_LEN]; 2167 int error; 2168 2169 ifp = sc->sc_carpdev; 2170 if (ifp == NULL) 2171 return (EINVAL); 2172 2173 /* 2174 * Find a key to lookup carp_mc_entry. We have to do this 2175 * before calling ether_delmulti for obvious reason. 2176 */ 2177 if ((error = ether_multiaddr(sa, addrlo, addrhi)) != 0) 2178 return (error); 2179 ETHER_LOOKUP_MULTI(addrlo, addrhi, &sc->sc_ac, enm); 2180 if (enm == NULL) 2181 return (EINVAL); 2182 2183 LIST_FOREACH(mc, &sc->carp_mc_listhead, mc_entries) 2184 if (mc->mc_enm == enm) 2185 break; 2186 2187 /* We won't delete entries we didn't add */ 2188 if (mc == NULL) 2189 return (EINVAL); 2190 2191 error = ether_delmulti(sa, &sc->sc_ac); 2192 if (error != ENETRESET) 2193 return (error); 2194 2195 /* We no longer use this multicast address. Tell parent so. */ 2196 error = (*ifp->if_ioctl)(ifp, SIOCDELMULTI, (void *)ifr); 2197 if (error == 0) { 2198 /* And forget about this address. */ 2199 LIST_REMOVE(mc, mc_entries); 2200 FREE(mc, M_DEVBUF); 2201 } else 2202 (void)ether_addmulti(sa, &sc->sc_ac); 2203 return (error); 2204 } 2205 2206 /* 2207 * Delete any multicast address we have asked to add from parent 2208 * interface. Called when the carp is being unconfigured. 2209 */ 2210 void 2211 carp_ether_purgemulti(struct carp_softc *sc) 2212 { 2213 struct ifnet *ifp = sc->sc_carpdev; /* Parent. */ 2214 struct carp_mc_entry *mc; 2215 union { 2216 struct ifreq ifreq; 2217 struct { 2218 char ifr_name[IFNAMSIZ]; 2219 struct sockaddr_storage ifr_ss; 2220 } ifreq_storage; 2221 } u; 2222 struct ifreq *ifr = &u.ifreq; 2223 2224 if (ifp == NULL) 2225 return; 2226 2227 memcpy(ifr->ifr_name, ifp->if_xname, IFNAMSIZ); 2228 while ((mc = LIST_FIRST(&sc->carp_mc_listhead)) != NULL) { 2229 memcpy(&ifr->ifr_addr, &mc->mc_addr, mc->mc_addr.ss_len); 2230 (void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, (void *)ifr); 2231 LIST_REMOVE(mc, mc_entries); 2232 FREE(mc, M_DEVBUF); 2233 } 2234 } 2235 2236 SYSCTL_SETUP(sysctl_net_inet_carp_setup, "sysctl net.inet.carp subtree setup") 2237 { 2238 2239 sysctl_createv(clog, 0, NULL, NULL, 2240 CTLFLAG_PERMANENT, 2241 CTLTYPE_NODE, "net", NULL, 2242 NULL, 0, NULL, 0, 2243 CTL_NET, CTL_EOL); 2244 sysctl_createv(clog, 0, NULL, NULL, 2245 CTLFLAG_PERMANENT, 2246 CTLTYPE_NODE, "inet", NULL, 2247 NULL, 0, NULL, 0, 2248 CTL_NET, PF_INET, CTL_EOL); 2249 sysctl_createv(clog, 0, NULL, NULL, 2250 CTLFLAG_PERMANENT, 2251 CTLTYPE_NODE, "carp", 2252 SYSCTL_DESCR("CARP related settings"), 2253 NULL, 0, NULL, 0, 2254 CTL_NET, PF_INET, IPPROTO_CARP, CTL_EOL); 2255 2256 sysctl_createv(clog, 0, NULL, NULL, 2257 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2258 CTLTYPE_INT, "preempt", 2259 SYSCTL_DESCR("Enable CARP Preempt"), 2260 NULL, 0, &carp_opts[CARPCTL_PREEMPT], 0, 2261 CTL_NET, PF_INET, IPPROTO_CARP, 2262 CTL_CREATE, CTL_EOL); 2263 sysctl_createv(clog, 0, NULL, NULL, 2264 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2265 CTLTYPE_INT, "arpbalance", 2266 SYSCTL_DESCR("Enable ARP balancing"), 2267 NULL, 0, &carp_opts[CARPCTL_ARPBALANCE], 0, 2268 CTL_NET, PF_INET, IPPROTO_CARP, 2269 CTL_CREATE, CTL_EOL); 2270 sysctl_createv(clog, 0, NULL, NULL, 2271 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2272 CTLTYPE_INT, "allow", 2273 SYSCTL_DESCR("Enable CARP"), 2274 NULL, 0, &carp_opts[CARPCTL_ALLOW], 0, 2275 CTL_NET, PF_INET, IPPROTO_CARP, 2276 CTL_CREATE, CTL_EOL); 2277 sysctl_createv(clog, 0, NULL, NULL, 2278 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2279 CTLTYPE_INT, "log", 2280 SYSCTL_DESCR("CARP logging"), 2281 NULL, 0, &carp_opts[CARPCTL_LOG], 0, 2282 CTL_NET, PF_INET, IPPROTO_CARP, 2283 CTL_CREATE, CTL_EOL); 2284 sysctl_createv(clog, 0, NULL, NULL, 2285 CTLFLAG_PERMANENT, 2286 CTLTYPE_STRUCT, "stats", 2287 SYSCTL_DESCR("CARP statistics"), 2288 NULL, 0, &carpstats, sizeof(carpstats), 2289 CTL_NET, PF_INET, IPPROTO_CARP, CARPCTL_STATS, 2290 CTL_EOL); 2291 } 2292
Indexes created Tue Sep 23 23:09:58 GMT 2025