1 /* $NetBSD: ip_carp.c,v 1.26 2008/05/04 07:22:14 thorpej 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 1.26 2008/05/04 07:22:14 thorpej 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 <net/net_stats.h> 66 #include <netinet/if_inarp.h> 67 68 #include <machine/stdarg.h> 69 70 #if NFDDI > 0 71 #include <net/if_fddi.h> 72 #endif 73 #if NTOKEN > 0 74 #include <net/if_token.h> 75 #endif 76 77 #ifdef INET 78 #include <netinet/in.h> 79 #include <netinet/in_systm.h> 80 #include <netinet/in_var.h> 81 #include <netinet/ip.h> 82 #include <netinet/ip_var.h> 83 84 #include <net/if_dl.h> 85 #endif 86 87 #ifdef INET6 88 #include <netinet/icmp6.h> 89 #include <netinet/ip6.h> 90 #include <netinet6/ip6_var.h> 91 #include <netinet6/nd6.h> 92 #include <netinet6/scope6_var.h> 93 #endif 94 95 #include "bpfilter.h" 96 #if NBPFILTER > 0 97 #include <net/bpf.h> 98 #endif 99 100 #include <sys/sha1.h> 101 102 #include <netinet/ip_carp.h> 103 104 struct carp_mc_entry { 105 LIST_ENTRY(carp_mc_entry) mc_entries; 106 union { 107 struct ether_multi *mcu_enm; 108 } mc_u; 109 struct sockaddr_storage mc_addr; 110 }; 111 #define mc_enm mc_u.mcu_enm 112 113 struct carp_softc { 114 struct ethercom sc_ac; 115 #define sc_if sc_ac.ec_if 116 #define sc_carpdev sc_ac.ec_if.if_carpdev 117 int ah_cookie; 118 int lh_cookie; 119 struct ip_moptions sc_imo; 120 #ifdef INET6 121 struct ip6_moptions sc_im6o; 122 #endif /* INET6 */ 123 TAILQ_ENTRY(carp_softc) sc_list; 124 125 enum { INIT = 0, BACKUP, MASTER } sc_state; 126 127 int sc_suppress; 128 int sc_bow_out; 129 130 int sc_sendad_errors; 131 #define CARP_SENDAD_MAX_ERRORS 3 132 int sc_sendad_success; 133 #define CARP_SENDAD_MIN_SUCCESS 3 134 135 int sc_vhid; 136 int sc_advskew; 137 int sc_naddrs; 138 int sc_naddrs6; 139 int sc_advbase; /* seconds */ 140 int sc_init_counter; 141 u_int64_t sc_counter; 142 143 /* authentication */ 144 #define CARP_HMAC_PAD 64 145 unsigned char sc_key[CARP_KEY_LEN]; 146 unsigned char sc_pad[CARP_HMAC_PAD]; 147 SHA1_CTX sc_sha1; 148 u_int32_t sc_hashkey[2]; 149 150 struct callout sc_ad_tmo; /* advertisement timeout */ 151 struct callout sc_md_tmo; /* master down timeout */ 152 struct callout sc_md6_tmo; /* master down timeout */ 153 154 LIST_HEAD(__carp_mchead, carp_mc_entry) carp_mc_listhead; 155 }; 156 157 int carp_suppress_preempt = 0; 158 int carp_opts[CARPCTL_MAXID] = { 0, 1, 0, 0, 0 }; /* XXX for now */ 159 160 static percpu_t *carpstat_percpu; 161 162 #define CARP_STATINC(x) _NET_STATINC(carpstat_percpu, x) 163 164 struct carp_if { 165 TAILQ_HEAD(, carp_softc) vhif_vrs; 166 int vhif_nvrs; 167 168 struct ifnet *vhif_ifp; 169 }; 170 171 #define CARP_LOG(sc, s) \ 172 if (carp_opts[CARPCTL_LOG]) { \ 173 if (sc) \ 174 log(LOG_INFO, "%s: ", \ 175 (sc)->sc_if.if_xname); \ 176 else \ 177 log(LOG_INFO, "carp: "); \ 178 addlog s; \ 179 addlog("\n"); \ 180 } 181 182 void carp_hmac_prepare(struct carp_softc *); 183 void carp_hmac_generate(struct carp_softc *, u_int32_t *, 184 unsigned char *); 185 int carp_hmac_verify(struct carp_softc *, u_int32_t *, 186 unsigned char *); 187 void carp_setroute(struct carp_softc *, int); 188 void carp_proto_input_c(struct mbuf *, struct carp_header *, sa_family_t); 189 void carpattach(int); 190 void carpdetach(struct carp_softc *); 191 int carp_prepare_ad(struct mbuf *, struct carp_softc *, 192 struct carp_header *); 193 void carp_send_ad_all(void); 194 void carp_send_ad(void *); 195 void carp_send_arp(struct carp_softc *); 196 void carp_master_down(void *); 197 int carp_ioctl(struct ifnet *, u_long, void *); 198 void carp_start(struct ifnet *); 199 void carp_setrun(struct carp_softc *, sa_family_t); 200 void carp_set_state(struct carp_softc *, int); 201 int carp_addrcount(struct carp_if *, struct in_ifaddr *, int); 202 enum { CARP_COUNT_MASTER, CARP_COUNT_RUNNING }; 203 204 void carp_multicast_cleanup(struct carp_softc *); 205 int carp_set_ifp(struct carp_softc *, struct ifnet *); 206 void carp_set_enaddr(struct carp_softc *); 207 void carp_addr_updated(void *); 208 u_int32_t carp_hash(struct carp_softc *, u_char *); 209 int carp_set_addr(struct carp_softc *, struct sockaddr_in *); 210 int carp_join_multicast(struct carp_softc *); 211 #ifdef INET6 212 void carp_send_na(struct carp_softc *); 213 int carp_set_addr6(struct carp_softc *, struct sockaddr_in6 *); 214 int carp_join_multicast6(struct carp_softc *); 215 #endif 216 int carp_clone_create(struct if_clone *, int); 217 int carp_clone_destroy(struct ifnet *); 218 int carp_ether_addmulti(struct carp_softc *, struct ifreq *); 219 int carp_ether_delmulti(struct carp_softc *, struct ifreq *); 220 void carp_ether_purgemulti(struct carp_softc *); 221 222 struct if_clone carp_cloner = 223 IF_CLONE_INITIALIZER("carp", carp_clone_create, carp_clone_destroy); 224 225 static __inline u_int16_t 226 carp_cksum(struct mbuf *m, int len) 227 { 228 return (in_cksum(m, len)); 229 } 230 231 void 232 carp_hmac_prepare(struct carp_softc *sc) 233 { 234 u_int8_t carp_version = CARP_VERSION, type = CARP_ADVERTISEMENT; 235 u_int8_t vhid = sc->sc_vhid & 0xff; 236 SHA1_CTX sha1ctx; 237 u_int32_t kmd[5]; 238 struct ifaddr *ifa; 239 int i, found; 240 struct in_addr last, cur, in; 241 #ifdef INET6 242 struct in6_addr last6, cur6, in6; 243 #endif /* INET6 */ 244 245 /* compute ipad from key */ 246 bzero(sc->sc_pad, sizeof(sc->sc_pad)); 247 bcopy(sc->sc_key, sc->sc_pad, sizeof(sc->sc_key)); 248 for (i = 0; i < sizeof(sc->sc_pad); i++) 249 sc->sc_pad[i] ^= 0x36; 250 251 /* precompute first part of inner hash */ 252 SHA1Init(&sc->sc_sha1); 253 SHA1Update(&sc->sc_sha1, sc->sc_pad, sizeof(sc->sc_pad)); 254 SHA1Update(&sc->sc_sha1, (void *)&carp_version, sizeof(carp_version)); 255 SHA1Update(&sc->sc_sha1, (void *)&type, sizeof(type)); 256 257 /* generate a key for the arpbalance hash, before the vhid is hashed */ 258 bcopy(&sc->sc_sha1, &sha1ctx, sizeof(sha1ctx)); 259 SHA1Final((unsigned char *)kmd, &sha1ctx); 260 sc->sc_hashkey[0] = kmd[0] ^ kmd[1]; 261 sc->sc_hashkey[1] = kmd[2] ^ kmd[3]; 262 263 /* the rest of the precomputation */ 264 SHA1Update(&sc->sc_sha1, (void *)&vhid, sizeof(vhid)); 265 266 /* Hash the addresses from smallest to largest, not interface order */ 267 #ifdef INET 268 cur.s_addr = 0; 269 do { 270 found = 0; 271 last = cur; 272 cur.s_addr = 0xffffffff; 273 IFADDR_FOREACH(ifa, &sc->sc_if) { 274 in.s_addr = ifatoia(ifa)->ia_addr.sin_addr.s_addr; 275 if (ifa->ifa_addr->sa_family == AF_INET && 276 ntohl(in.s_addr) > ntohl(last.s_addr) && 277 ntohl(in.s_addr) < ntohl(cur.s_addr)) { 278 cur.s_addr = in.s_addr; 279 found++; 280 } 281 } 282 if (found) 283 SHA1Update(&sc->sc_sha1, (void *)&cur, sizeof(cur)); 284 } while (found); 285 #endif /* INET */ 286 287 #ifdef INET6 288 memset(&cur6, 0x00, sizeof(cur6)); 289 do { 290 found = 0; 291 last6 = cur6; 292 memset(&cur6, 0xff, sizeof(cur6)); 293 IFADDR_FOREACH(ifa, &sc->sc_if) { 294 in6 = ifatoia6(ifa)->ia_addr.sin6_addr; 295 if (IN6_IS_ADDR_LINKLOCAL(&in6)) 296 in6.s6_addr16[1] = 0; 297 if (ifa->ifa_addr->sa_family == AF_INET6 && 298 memcmp(&in6, &last6, sizeof(in6)) > 0 && 299 memcmp(&in6, &cur6, sizeof(in6)) < 0) { 300 cur6 = in6; 301 found++; 302 } 303 } 304 if (found) 305 SHA1Update(&sc->sc_sha1, (void *)&cur6, sizeof(cur6)); 306 } while (found); 307 #endif /* INET6 */ 308 309 /* convert ipad to opad */ 310 for (i = 0; i < sizeof(sc->sc_pad); i++) 311 sc->sc_pad[i] ^= 0x36 ^ 0x5c; 312 } 313 314 void 315 carp_hmac_generate(struct carp_softc *sc, u_int32_t counter[2], 316 unsigned char md[20]) 317 { 318 SHA1_CTX sha1ctx; 319 320 /* fetch first half of inner hash */ 321 bcopy(&sc->sc_sha1, &sha1ctx, sizeof(sha1ctx)); 322 323 SHA1Update(&sha1ctx, (void *)counter, sizeof(sc->sc_counter)); 324 SHA1Final(md, &sha1ctx); 325 326 /* outer hash */ 327 SHA1Init(&sha1ctx); 328 SHA1Update(&sha1ctx, sc->sc_pad, sizeof(sc->sc_pad)); 329 SHA1Update(&sha1ctx, md, 20); 330 SHA1Final(md, &sha1ctx); 331 } 332 333 int 334 carp_hmac_verify(struct carp_softc *sc, u_int32_t counter[2], 335 unsigned char md[20]) 336 { 337 unsigned char md2[20]; 338 339 carp_hmac_generate(sc, counter, md2); 340 341 return (bcmp(md, md2, sizeof(md2))); 342 } 343 344 void 345 carp_setroute(struct carp_softc *sc, int cmd) 346 { 347 struct ifaddr *ifa; 348 int s; 349 350 s = splsoftnet(); 351 IFADDR_FOREACH(ifa, &sc->sc_if) { 352 switch (ifa->ifa_addr->sa_family) { 353 case AF_INET: { 354 int count = 0; 355 struct rtentry *rt; 356 int hr_otherif, nr_ourif; 357 358 /* 359 * Avoid screwing with the routes if there are other 360 * carp interfaces which are master and have the same 361 * address. 362 */ 363 if (sc->sc_carpdev != NULL && 364 sc->sc_carpdev->if_carp != NULL) { 365 count = carp_addrcount( 366 (struct carp_if *)sc->sc_carpdev->if_carp, 367 ifatoia(ifa), CARP_COUNT_MASTER); 368 if ((cmd == RTM_ADD && count != 1) || 369 (cmd == RTM_DELETE && count != 0)) 370 continue; 371 } 372 373 /* Remove the existing host route, if any */ 374 rtrequest(RTM_DELETE, ifa->ifa_addr, 375 ifa->ifa_addr, ifa->ifa_netmask, 376 RTF_HOST, NULL); 377 378 rt = NULL; 379 (void)rtrequest(RTM_GET, ifa->ifa_addr, ifa->ifa_addr, 380 ifa->ifa_netmask, RTF_HOST, &rt); 381 hr_otherif = (rt && rt->rt_ifp != &sc->sc_if && 382 rt->rt_flags & (RTF_CLONING|RTF_CLONED)); 383 if (rt != NULL) { 384 RTFREE(rt); 385 rt = NULL; 386 } 387 388 /* Check for a network route on our interface */ 389 390 rt = NULL; 391 (void)rtrequest(RTM_GET, ifa->ifa_addr, ifa->ifa_addr, 392 ifa->ifa_netmask, 0, &rt); 393 nr_ourif = (rt && rt->rt_ifp == &sc->sc_if); 394 395 switch (cmd) { 396 case RTM_ADD: 397 if (hr_otherif) { 398 ifa->ifa_rtrequest = NULL; 399 ifa->ifa_flags &= ~RTF_CLONING; 400 401 rtrequest(RTM_ADD, ifa->ifa_addr, 402 ifa->ifa_addr, ifa->ifa_netmask, 403 RTF_UP | RTF_HOST, NULL); 404 } 405 if (!hr_otherif || nr_ourif || !rt) { 406 if (nr_ourif && !(rt->rt_flags & 407 RTF_CLONING)) 408 rtrequest(RTM_DELETE, 409 ifa->ifa_addr, 410 ifa->ifa_addr, 411 ifa->ifa_netmask, 0, NULL); 412 413 ifa->ifa_rtrequest = arp_rtrequest; 414 ifa->ifa_flags |= RTF_CLONING; 415 416 if (rtrequest(RTM_ADD, ifa->ifa_addr, 417 ifa->ifa_addr, ifa->ifa_netmask, 0, 418 NULL) == 0) 419 ifa->ifa_flags |= IFA_ROUTE; 420 } 421 break; 422 case RTM_DELETE: 423 break; 424 default: 425 break; 426 } 427 if (rt != NULL) { 428 RTFREE(rt); 429 rt = NULL; 430 } 431 break; 432 } 433 434 #ifdef INET6 435 case AF_INET6: 436 if (cmd == RTM_ADD) 437 in6_ifaddloop(ifa); 438 else 439 in6_ifremloop(ifa); 440 break; 441 #endif /* INET6 */ 442 default: 443 break; 444 } 445 } 446 splx(s); 447 } 448 449 /* 450 * process input packet. 451 * we have rearranged checks order compared to the rfc, 452 * but it seems more efficient this way or not possible otherwise. 453 */ 454 void 455 carp_proto_input(struct mbuf *m, ...) 456 { 457 struct ip *ip = mtod(m, struct ip *); 458 struct carp_softc *sc = NULL; 459 struct carp_header *ch; 460 int iplen, len, hlen; 461 va_list ap; 462 463 va_start(ap, m); 464 hlen = va_arg(ap, int); 465 va_end(ap); 466 467 CARP_STATINC(CARP_STAT_IPACKETS); 468 469 if (!carp_opts[CARPCTL_ALLOW]) { 470 m_freem(m); 471 return; 472 } 473 474 /* check if received on a valid carp interface */ 475 if (m->m_pkthdr.rcvif->if_type != IFT_CARP) { 476 CARP_STATINC(CARP_STAT_BADIF); 477 CARP_LOG(sc, ("packet received on non-carp interface: %s", 478 m->m_pkthdr.rcvif->if_xname)); 479 m_freem(m); 480 return; 481 } 482 483 /* verify that the IP TTL is 255. */ 484 if (ip->ip_ttl != CARP_DFLTTL) { 485 CARP_STATINC(CARP_STAT_BADTTL); 486 CARP_LOG(sc, ("received ttl %d != %d on %s", ip->ip_ttl, 487 CARP_DFLTTL, m->m_pkthdr.rcvif->if_xname)); 488 m_freem(m); 489 return; 490 } 491 492 /* 493 * verify that the received packet length is 494 * equal to the CARP header 495 */ 496 iplen = ip->ip_hl << 2; 497 len = iplen + sizeof(*ch); 498 if (len > m->m_pkthdr.len) { 499 CARP_STATINC(CARP_STAT_BADLEN); 500 CARP_LOG(sc, ("packet too short %d on %s", m->m_pkthdr.len, 501 m->m_pkthdr.rcvif->if_xname)); 502 m_freem(m); 503 return; 504 } 505 506 if ((m = m_pullup(m, len)) == NULL) { 507 CARP_STATINC(CARP_STAT_HDROPS); 508 return; 509 } 510 ip = mtod(m, struct ip *); 511 ch = (struct carp_header *)((char *)ip + iplen); 512 /* verify the CARP checksum */ 513 m->m_data += iplen; 514 if (carp_cksum(m, len - iplen)) { 515 CARP_STATINC(CARP_STAT_BADSUM); 516 CARP_LOG(sc, ("checksum failed on %s", 517 m->m_pkthdr.rcvif->if_xname)); 518 m_freem(m); 519 return; 520 } 521 m->m_data -= iplen; 522 523 carp_proto_input_c(m, ch, AF_INET); 524 } 525 526 #ifdef INET6 527 int 528 carp6_proto_input(struct mbuf **mp, int *offp, int proto) 529 { 530 struct mbuf *m = *mp; 531 struct carp_softc *sc = NULL; 532 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 533 struct carp_header *ch; 534 u_int len; 535 536 CARP_STATINC(CARP_STAT_IPACKETS6); 537 538 if (!carp_opts[CARPCTL_ALLOW]) { 539 m_freem(m); 540 return (IPPROTO_DONE); 541 } 542 543 /* check if received on a valid carp interface */ 544 if (m->m_pkthdr.rcvif->if_type != IFT_CARP) { 545 CARP_STATINC(CARP_STAT_BADIF); 546 CARP_LOG(sc, ("packet received on non-carp interface: %s", 547 m->m_pkthdr.rcvif->if_xname)); 548 m_freem(m); 549 return (IPPROTO_DONE); 550 } 551 552 /* verify that the IP TTL is 255 */ 553 if (ip6->ip6_hlim != CARP_DFLTTL) { 554 CARP_STATINC(CARP_STAT_BADTTL); 555 CARP_LOG(sc, ("received ttl %d != %d on %s", ip6->ip6_hlim, 556 CARP_DFLTTL, m->m_pkthdr.rcvif->if_xname)); 557 m_freem(m); 558 return (IPPROTO_DONE); 559 } 560 561 /* verify that we have a complete carp packet */ 562 len = m->m_len; 563 IP6_EXTHDR_GET(ch, struct carp_header *, m, *offp, sizeof(*ch)); 564 if (ch == NULL) { 565 CARP_STATINC(CARP_STAT_BADLEN); 566 CARP_LOG(sc, ("packet size %u too small", len)); 567 return (IPPROTO_DONE); 568 } 569 570 571 /* verify the CARP checksum */ 572 m->m_data += *offp; 573 if (carp_cksum(m, sizeof(*ch))) { 574 CARP_STATINC(CARP_STAT_BADSUM); 575 CARP_LOG(sc, ("checksum failed, on %s", 576 m->m_pkthdr.rcvif->if_xname)); 577 m_freem(m); 578 return (IPPROTO_DONE); 579 } 580 m->m_data -= *offp; 581 582 carp_proto_input_c(m, ch, AF_INET6); 583 return (IPPROTO_DONE); 584 } 585 #endif /* INET6 */ 586 587 void 588 carp_proto_input_c(struct mbuf *m, struct carp_header *ch, sa_family_t af) 589 { 590 struct carp_softc *sc; 591 u_int64_t tmp_counter; 592 struct timeval sc_tv, ch_tv; 593 594 TAILQ_FOREACH(sc, &((struct carp_if *) 595 m->m_pkthdr.rcvif->if_carpdev->if_carp)->vhif_vrs, sc_list) 596 if (sc->sc_vhid == ch->carp_vhid) 597 break; 598 599 if (!sc || (sc->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) != 600 (IFF_UP|IFF_RUNNING)) { 601 CARP_STATINC(CARP_STAT_BADVHID); 602 m_freem(m); 603 return; 604 } 605 606 /* 607 * Check if our own advertisement was duplicated 608 * from a non simplex interface. 609 * XXX If there is no address on our physical interface 610 * there is no way to distinguish our ads from the ones 611 * another carp host might have sent us. 612 */ 613 if ((sc->sc_carpdev->if_flags & IFF_SIMPLEX) == 0) { 614 struct sockaddr sa; 615 struct ifaddr *ifa; 616 617 bzero(&sa, sizeof(sa)); 618 sa.sa_family = af; 619 ifa = ifaof_ifpforaddr(&sa, sc->sc_carpdev); 620 621 if (ifa && af == AF_INET) { 622 struct ip *ip = mtod(m, struct ip *); 623 if (ip->ip_src.s_addr == 624 ifatoia(ifa)->ia_addr.sin_addr.s_addr) { 625 m_freem(m); 626 return; 627 } 628 } 629 #ifdef INET6 630 if (ifa && af == AF_INET6) { 631 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 632 struct in6_addr in6_src, in6_found; 633 634 in6_src = ip6->ip6_src; 635 in6_found = ifatoia6(ifa)->ia_addr.sin6_addr; 636 if (IN6_IS_ADDR_LINKLOCAL(&in6_src)) 637 in6_src.s6_addr16[1] = 0; 638 if (IN6_IS_ADDR_LINKLOCAL(&in6_found)) 639 in6_found.s6_addr16[1] = 0; 640 if (IN6_ARE_ADDR_EQUAL(&in6_src, &in6_found)) { 641 m_freem(m); 642 return; 643 } 644 } 645 #endif /* INET6 */ 646 } 647 648 microtime(&sc->sc_if.if_lastchange); 649 sc->sc_if.if_ipackets++; 650 sc->sc_if.if_ibytes += m->m_pkthdr.len; 651 652 /* verify the CARP version. */ 653 if (ch->carp_version != CARP_VERSION) { 654 CARP_STATINC(CARP_STAT_BADVER); 655 sc->sc_if.if_ierrors++; 656 CARP_LOG(sc, ("invalid version %d != %d", 657 ch->carp_version, CARP_VERSION)); 658 m_freem(m); 659 return; 660 } 661 662 /* verify the hash */ 663 if (carp_hmac_verify(sc, ch->carp_counter, ch->carp_md)) { 664 CARP_STATINC(CARP_STAT_BADAUTH); 665 sc->sc_if.if_ierrors++; 666 CARP_LOG(sc, ("incorrect hash")); 667 m_freem(m); 668 return; 669 } 670 671 tmp_counter = ntohl(ch->carp_counter[0]); 672 tmp_counter = tmp_counter<<32; 673 tmp_counter += ntohl(ch->carp_counter[1]); 674 675 /* XXX Replay protection goes here */ 676 677 sc->sc_init_counter = 0; 678 sc->sc_counter = tmp_counter; 679 680 681 sc_tv.tv_sec = sc->sc_advbase; 682 if (carp_suppress_preempt && sc->sc_advskew < 240) 683 sc_tv.tv_usec = 240 * 1000000 / 256; 684 else 685 sc_tv.tv_usec = sc->sc_advskew * 1000000 / 256; 686 ch_tv.tv_sec = ch->carp_advbase; 687 ch_tv.tv_usec = ch->carp_advskew * 1000000 / 256; 688 689 switch (sc->sc_state) { 690 case INIT: 691 break; 692 case MASTER: 693 /* 694 * If we receive an advertisement from a backup who's going to 695 * be more frequent than us, go into BACKUP state. 696 */ 697 if (timercmp(&sc_tv, &ch_tv, >) || 698 timercmp(&sc_tv, &ch_tv, ==)) { 699 callout_stop(&sc->sc_ad_tmo); 700 carp_set_state(sc, BACKUP); 701 carp_setrun(sc, 0); 702 carp_setroute(sc, RTM_DELETE); 703 } 704 break; 705 case BACKUP: 706 /* 707 * If we're pre-empting masters who advertise slower than us, 708 * and this one claims to be slower, treat him as down. 709 */ 710 if (carp_opts[CARPCTL_PREEMPT] && timercmp(&sc_tv, &ch_tv, <)) { 711 carp_master_down(sc); 712 break; 713 } 714 715 /* 716 * If the master is going to advertise at such a low frequency 717 * that he's guaranteed to time out, we'd might as well just 718 * treat him as timed out now. 719 */ 720 sc_tv.tv_sec = sc->sc_advbase * 3; 721 if (timercmp(&sc_tv, &ch_tv, <)) { 722 carp_master_down(sc); 723 break; 724 } 725 726 /* 727 * Otherwise, we reset the counter and wait for the next 728 * advertisement. 729 */ 730 carp_setrun(sc, af); 731 break; 732 } 733 734 m_freem(m); 735 return; 736 } 737 738 /* 739 * Interface side of the CARP implementation. 740 */ 741 742 /* ARGSUSED */ 743 void 744 carpattach(int n) 745 { 746 if_clone_attach(&carp_cloner); 747 748 carpstat_percpu = percpu_alloc(sizeof(uint64_t) * CARP_NSTATS); 749 } 750 751 int 752 carp_clone_create(struct if_clone *ifc, int unit) 753 { 754 extern int ifqmaxlen; 755 struct carp_softc *sc; 756 struct ifnet *ifp; 757 758 sc = malloc(sizeof(*sc), M_DEVBUF, M_NOWAIT); 759 if (!sc) 760 return (ENOMEM); 761 bzero(sc, sizeof(*sc)); 762 763 sc->sc_suppress = 0; 764 sc->sc_advbase = CARP_DFLTINTV; 765 sc->sc_vhid = -1; /* required setting */ 766 sc->sc_advskew = 0; 767 sc->sc_init_counter = 1; 768 sc->sc_naddrs = sc->sc_naddrs6 = 0; 769 #ifdef INET6 770 sc->sc_im6o.im6o_multicast_hlim = CARP_DFLTTL; 771 #endif /* INET6 */ 772 773 callout_init(&sc->sc_ad_tmo, 0); 774 callout_init(&sc->sc_md_tmo, 0); 775 callout_init(&sc->sc_md6_tmo, 0); 776 777 callout_setfunc(&sc->sc_ad_tmo, carp_send_ad, sc); 778 callout_setfunc(&sc->sc_md_tmo, carp_master_down, sc); 779 callout_setfunc(&sc->sc_md6_tmo, carp_master_down, sc); 780 781 LIST_INIT(&sc->carp_mc_listhead); 782 ifp = &sc->sc_if; 783 ifp->if_softc = sc; 784 snprintf(ifp->if_xname, sizeof ifp->if_xname, "%s%d", ifc->ifc_name, 785 unit); 786 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST; 787 ifp->if_ioctl = carp_ioctl; 788 ifp->if_start = carp_start; 789 ifp->if_output = carp_output; 790 ifp->if_type = IFT_CARP; 791 ifp->if_addrlen = ETHER_ADDR_LEN; 792 ifp->if_hdrlen = ETHER_HDR_LEN; 793 ifp->if_mtu = ETHERMTU; 794 IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen); 795 IFQ_SET_READY(&ifp->if_snd); 796 if_attach(ifp); 797 798 if_alloc_sadl(ifp); 799 ifp->if_broadcastaddr = etherbroadcastaddr; 800 carp_set_enaddr(sc); 801 LIST_INIT(&sc->sc_ac.ec_multiaddrs); 802 #if NBPFILTER > 0 803 bpfattach(ifp, DLT_EN10MB, ETHER_HDR_LEN); 804 #endif 805 return (0); 806 } 807 808 int 809 carp_clone_destroy(struct ifnet *ifp) 810 { 811 struct carp_softc *sc = ifp->if_softc; 812 813 carpdetach(ifp->if_softc); 814 ether_ifdetach(ifp); 815 if_detach(ifp); 816 callout_destroy(&sc->sc_ad_tmo); 817 callout_destroy(&sc->sc_md_tmo); 818 callout_destroy(&sc->sc_md6_tmo); 819 free(ifp->if_softc, M_DEVBUF); 820 821 return (0); 822 } 823 824 void 825 carpdetach(struct carp_softc *sc) 826 { 827 struct carp_if *cif; 828 int s; 829 830 callout_stop(&sc->sc_ad_tmo); 831 callout_stop(&sc->sc_md_tmo); 832 callout_stop(&sc->sc_md6_tmo); 833 834 if (sc->sc_suppress) 835 carp_suppress_preempt--; 836 sc->sc_suppress = 0; 837 838 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) 839 carp_suppress_preempt--; 840 sc->sc_sendad_errors = 0; 841 842 carp_set_state(sc, INIT); 843 sc->sc_if.if_flags &= ~IFF_UP; 844 carp_setrun(sc, 0); 845 carp_multicast_cleanup(sc); 846 847 s = splnet(); 848 if (sc->sc_carpdev != NULL) { 849 /* XXX linkstatehook removal */ 850 cif = (struct carp_if *)sc->sc_carpdev->if_carp; 851 TAILQ_REMOVE(&cif->vhif_vrs, sc, sc_list); 852 if (!--cif->vhif_nvrs) { 853 ifpromisc(sc->sc_carpdev, 0); 854 sc->sc_carpdev->if_carp = NULL; 855 FREE(cif, M_IFADDR); 856 } 857 } 858 sc->sc_carpdev = NULL; 859 splx(s); 860 } 861 862 /* Detach an interface from the carp. */ 863 void 864 carp_ifdetach(struct ifnet *ifp) 865 { 866 struct carp_softc *sc, *nextsc; 867 struct carp_if *cif = (struct carp_if *)ifp->if_carp; 868 869 for (sc = TAILQ_FIRST(&cif->vhif_vrs); sc; sc = nextsc) { 870 nextsc = TAILQ_NEXT(sc, sc_list); 871 carpdetach(sc); 872 } 873 } 874 875 int 876 carp_prepare_ad(struct mbuf *m, struct carp_softc *sc, 877 struct carp_header *ch) 878 { 879 if (sc->sc_init_counter) { 880 /* this could also be seconds since unix epoch */ 881 sc->sc_counter = arc4random(); 882 sc->sc_counter = sc->sc_counter << 32; 883 sc->sc_counter += arc4random(); 884 } else 885 sc->sc_counter++; 886 887 ch->carp_counter[0] = htonl((sc->sc_counter>>32)&0xffffffff); 888 ch->carp_counter[1] = htonl(sc->sc_counter&0xffffffff); 889 890 carp_hmac_generate(sc, ch->carp_counter, ch->carp_md); 891 892 return (0); 893 } 894 895 void 896 carp_send_ad_all(void) 897 { 898 struct ifnet *ifp; 899 struct carp_if *cif; 900 struct carp_softc *vh; 901 902 TAILQ_FOREACH(ifp, &ifnet, if_list) { 903 if (ifp->if_carp == NULL || ifp->if_type == IFT_CARP) 904 continue; 905 906 cif = (struct carp_if *)ifp->if_carp; 907 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 908 if ((vh->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) == 909 (IFF_UP|IFF_RUNNING) && vh->sc_state == MASTER) 910 carp_send_ad(vh); 911 } 912 } 913 } 914 915 916 void 917 carp_send_ad(void *v) 918 { 919 struct carp_header ch; 920 struct timeval tv; 921 struct carp_softc *sc = v; 922 struct carp_header *ch_ptr; 923 struct mbuf *m; 924 int error, len, advbase, advskew, s; 925 struct ifaddr *ifa; 926 struct sockaddr sa; 927 928 s = splsoftnet(); 929 930 advbase = advskew = 0; /* Sssssh compiler */ 931 if (sc->sc_carpdev == NULL) { 932 sc->sc_if.if_oerrors++; 933 goto retry_later; 934 } 935 936 /* bow out if we've gone to backup (the carp interface is going down) */ 937 if (sc->sc_bow_out) { 938 sc->sc_bow_out = 0; 939 advbase = 255; 940 advskew = 255; 941 } else { 942 advbase = sc->sc_advbase; 943 if (!carp_suppress_preempt || sc->sc_advskew > 240) 944 advskew = sc->sc_advskew; 945 else 946 advskew = 240; 947 tv.tv_sec = advbase; 948 tv.tv_usec = advskew * 1000000 / 256; 949 } 950 951 ch.carp_version = CARP_VERSION; 952 ch.carp_type = CARP_ADVERTISEMENT; 953 ch.carp_vhid = sc->sc_vhid; 954 ch.carp_advbase = advbase; 955 ch.carp_advskew = advskew; 956 ch.carp_authlen = 7; /* XXX DEFINE */ 957 ch.carp_pad1 = 0; /* must be zero */ 958 ch.carp_cksum = 0; 959 960 961 #ifdef INET 962 if (sc->sc_naddrs) { 963 struct ip *ip; 964 965 MGETHDR(m, M_DONTWAIT, MT_HEADER); 966 if (m == NULL) { 967 sc->sc_if.if_oerrors++; 968 CARP_STATINC(CARP_STAT_ONOMEM); 969 /* XXX maybe less ? */ 970 goto retry_later; 971 } 972 len = sizeof(*ip) + sizeof(ch); 973 m->m_pkthdr.len = len; 974 m->m_pkthdr.rcvif = NULL; 975 m->m_len = len; 976 MH_ALIGN(m, m->m_len); 977 m->m_flags |= M_MCAST; 978 ip = mtod(m, struct ip *); 979 ip->ip_v = IPVERSION; 980 ip->ip_hl = sizeof(*ip) >> 2; 981 ip->ip_tos = IPTOS_LOWDELAY; 982 ip->ip_len = htons(len); 983 ip->ip_id = 0; /* no need for id, we don't support fragments */ 984 ip->ip_off = htons(IP_DF); 985 ip->ip_ttl = CARP_DFLTTL; 986 ip->ip_p = IPPROTO_CARP; 987 ip->ip_sum = 0; 988 989 bzero(&sa, sizeof(sa)); 990 sa.sa_family = AF_INET; 991 ifa = ifaof_ifpforaddr(&sa, sc->sc_carpdev); 992 if (ifa == NULL) 993 ip->ip_src.s_addr = 0; 994 else 995 ip->ip_src.s_addr = 996 ifatoia(ifa)->ia_addr.sin_addr.s_addr; 997 ip->ip_dst.s_addr = INADDR_CARP_GROUP; 998 999 ch_ptr = (struct carp_header *)(&ip[1]); 1000 bcopy(&ch, ch_ptr, sizeof(ch)); 1001 if (carp_prepare_ad(m, sc, ch_ptr)) 1002 goto retry_later; 1003 1004 m->m_data += sizeof(*ip); 1005 ch_ptr->carp_cksum = carp_cksum(m, len - sizeof(*ip)); 1006 m->m_data -= sizeof(*ip); 1007 1008 microtime(&sc->sc_if.if_lastchange); 1009 sc->sc_if.if_opackets++; 1010 sc->sc_if.if_obytes += len; 1011 CARP_STATINC(CARP_STAT_OPACKETS); 1012 1013 error = ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo, 1014 NULL); 1015 if (error) { 1016 if (error == ENOBUFS) 1017 CARP_STATINC(CARP_STAT_ONOMEM); 1018 else 1019 CARP_LOG(sc, ("ip_output failed: %d", error)); 1020 sc->sc_if.if_oerrors++; 1021 if (sc->sc_sendad_errors < INT_MAX) 1022 sc->sc_sendad_errors++; 1023 if (sc->sc_sendad_errors == CARP_SENDAD_MAX_ERRORS) { 1024 carp_suppress_preempt++; 1025 if (carp_suppress_preempt == 1) 1026 carp_send_ad_all(); 1027 } 1028 sc->sc_sendad_success = 0; 1029 } else { 1030 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) { 1031 if (++sc->sc_sendad_success >= 1032 CARP_SENDAD_MIN_SUCCESS) { 1033 carp_suppress_preempt--; 1034 sc->sc_sendad_errors = 0; 1035 } 1036 } else 1037 sc->sc_sendad_errors = 0; 1038 } 1039 } 1040 #endif /* INET */ 1041 #ifdef INET6 1042 if (sc->sc_naddrs6) { 1043 struct ip6_hdr *ip6; 1044 1045 MGETHDR(m, M_DONTWAIT, MT_HEADER); 1046 if (m == NULL) { 1047 sc->sc_if.if_oerrors++; 1048 CARP_STATINC(CARP_STAT_ONOMEM); 1049 /* XXX maybe less ? */ 1050 goto retry_later; 1051 } 1052 len = sizeof(*ip6) + sizeof(ch); 1053 m->m_pkthdr.len = len; 1054 m->m_pkthdr.rcvif = NULL; 1055 m->m_len = len; 1056 MH_ALIGN(m, m->m_len); 1057 m->m_flags |= M_MCAST; 1058 ip6 = mtod(m, struct ip6_hdr *); 1059 bzero(ip6, sizeof(*ip6)); 1060 ip6->ip6_vfc |= IPV6_VERSION; 1061 ip6->ip6_hlim = CARP_DFLTTL; 1062 ip6->ip6_nxt = IPPROTO_CARP; 1063 1064 /* set the source address */ 1065 bzero(&sa, sizeof(sa)); 1066 sa.sa_family = AF_INET6; 1067 ifa = ifaof_ifpforaddr(&sa, sc->sc_carpdev); 1068 if (ifa == NULL) /* This should never happen with IPv6 */ 1069 bzero(&ip6->ip6_src, sizeof(struct in6_addr)); 1070 else 1071 bcopy(ifatoia6(ifa)->ia_addr.sin6_addr.s6_addr, 1072 &ip6->ip6_src, sizeof(struct in6_addr)); 1073 /* set the multicast destination */ 1074 1075 ip6->ip6_dst.s6_addr16[0] = htons(0xff02); 1076 ip6->ip6_dst.s6_addr8[15] = 0x12; 1077 if (in6_setscope(&ip6->ip6_dst, sc->sc_carpdev, NULL) != 0) { 1078 sc->sc_if.if_oerrors++; 1079 m_freem(m); 1080 CARP_LOG(sc, ("in6_setscope failed")); 1081 goto retry_later; 1082 } 1083 1084 ch_ptr = (struct carp_header *)(&ip6[1]); 1085 bcopy(&ch, ch_ptr, sizeof(ch)); 1086 if (carp_prepare_ad(m, sc, ch_ptr)) 1087 goto retry_later; 1088 1089 m->m_data += sizeof(*ip6); 1090 ch_ptr->carp_cksum = carp_cksum(m, len - sizeof(*ip6)); 1091 m->m_data -= sizeof(*ip6); 1092 1093 microtime(&sc->sc_if.if_lastchange); 1094 sc->sc_if.if_opackets++; 1095 sc->sc_if.if_obytes += len; 1096 CARP_STATINC(CARP_STAT_OPACKETS6); 1097 1098 error = ip6_output(m, NULL, NULL, 0, &sc->sc_im6o, NULL, NULL); 1099 if (error) { 1100 if (error == ENOBUFS) 1101 CARP_STATINC(CARP_STAT_ONOMEM); 1102 else 1103 CARP_LOG(sc, ("ip6_output failed: %d", error)); 1104 sc->sc_if.if_oerrors++; 1105 if (sc->sc_sendad_errors < INT_MAX) 1106 sc->sc_sendad_errors++; 1107 if (sc->sc_sendad_errors == CARP_SENDAD_MAX_ERRORS) { 1108 carp_suppress_preempt++; 1109 if (carp_suppress_preempt == 1) 1110 carp_send_ad_all(); 1111 } 1112 sc->sc_sendad_success = 0; 1113 } else { 1114 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) { 1115 if (++sc->sc_sendad_success >= 1116 CARP_SENDAD_MIN_SUCCESS) { 1117 carp_suppress_preempt--; 1118 sc->sc_sendad_errors = 0; 1119 } 1120 } else 1121 sc->sc_sendad_errors = 0; 1122 } 1123 } 1124 #endif /* INET6 */ 1125 1126 retry_later: 1127 splx(s); 1128 if (advbase != 255 || advskew != 255) 1129 callout_schedule(&sc->sc_ad_tmo, tvtohz(&tv)); 1130 } 1131 1132 /* 1133 * Broadcast a gratuitous ARP request containing 1134 * the virtual router MAC address for each IP address 1135 * associated with the virtual router. 1136 */ 1137 void 1138 carp_send_arp(struct carp_softc *sc) 1139 { 1140 struct ifaddr *ifa; 1141 struct in_addr *in; 1142 int s = splsoftnet(); 1143 1144 IFADDR_FOREACH(ifa, &sc->sc_if) { 1145 1146 if (ifa->ifa_addr->sa_family != AF_INET) 1147 continue; 1148 1149 in = &ifatoia(ifa)->ia_addr.sin_addr; 1150 arprequest(sc->sc_carpdev, in, in, CLLADDR(sc->sc_if.if_sadl)); 1151 DELAY(1000); /* XXX */ 1152 } 1153 splx(s); 1154 } 1155 1156 #ifdef INET6 1157 void 1158 carp_send_na(struct carp_softc *sc) 1159 { 1160 struct ifaddr *ifa; 1161 struct in6_addr *in6; 1162 static struct in6_addr mcast = IN6ADDR_LINKLOCAL_ALLNODES_INIT; 1163 int s = splsoftnet(); 1164 1165 IFADDR_FOREACH(ifa, &sc->sc_if) { 1166 1167 if (ifa->ifa_addr->sa_family != AF_INET6) 1168 continue; 1169 1170 in6 = &ifatoia6(ifa)->ia_addr.sin6_addr; 1171 nd6_na_output(sc->sc_carpdev, &mcast, in6, 1172 ND_NA_FLAG_OVERRIDE, 1, NULL); 1173 DELAY(1000); /* XXX */ 1174 } 1175 splx(s); 1176 } 1177 #endif /* INET6 */ 1178 1179 /* 1180 * Based on bridge_hash() in if_bridge.c 1181 */ 1182 #define mix(a,b,c) \ 1183 do { \ 1184 a -= b; a -= c; a ^= (c >> 13); \ 1185 b -= c; b -= a; b ^= (a << 8); \ 1186 c -= a; c -= b; c ^= (b >> 13); \ 1187 a -= b; a -= c; a ^= (c >> 12); \ 1188 b -= c; b -= a; b ^= (a << 16); \ 1189 c -= a; c -= b; c ^= (b >> 5); \ 1190 a -= b; a -= c; a ^= (c >> 3); \ 1191 b -= c; b -= a; b ^= (a << 10); \ 1192 c -= a; c -= b; c ^= (b >> 15); \ 1193 } while (0) 1194 1195 u_int32_t 1196 carp_hash(struct carp_softc *sc, u_char *src) 1197 { 1198 u_int32_t a = 0x9e3779b9, b = sc->sc_hashkey[0], c = sc->sc_hashkey[1]; 1199 1200 c += sc->sc_key[3] << 24; 1201 c += sc->sc_key[2] << 16; 1202 c += sc->sc_key[1] << 8; 1203 c += sc->sc_key[0]; 1204 b += src[5] << 8; 1205 b += src[4]; 1206 a += src[3] << 24; 1207 a += src[2] << 16; 1208 a += src[1] << 8; 1209 a += src[0]; 1210 1211 mix(a, b, c); 1212 return (c); 1213 } 1214 1215 int 1216 carp_addrcount(struct carp_if *cif, struct in_ifaddr *ia, int type) 1217 { 1218 struct carp_softc *vh; 1219 struct ifaddr *ifa; 1220 int count = 0; 1221 1222 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 1223 if ((type == CARP_COUNT_RUNNING && 1224 (vh->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) == 1225 (IFF_UP|IFF_RUNNING)) || 1226 (type == CARP_COUNT_MASTER && vh->sc_state == MASTER)) { 1227 IFADDR_FOREACH(ifa, &vh->sc_if) { 1228 if (ifa->ifa_addr->sa_family == AF_INET && 1229 ia->ia_addr.sin_addr.s_addr == 1230 ifatoia(ifa)->ia_addr.sin_addr.s_addr) 1231 count++; 1232 } 1233 } 1234 } 1235 return (count); 1236 } 1237 1238 int 1239 carp_iamatch(struct in_ifaddr *ia, u_char *src, 1240 u_int32_t *count, u_int32_t index) 1241 { 1242 struct carp_softc *sc = ia->ia_ifp->if_softc; 1243 1244 if (carp_opts[CARPCTL_ARPBALANCE]) { 1245 /* 1246 * We use the source ip to decide which virtual host should 1247 * handle the request. If we're master of that virtual host, 1248 * then we respond, otherwise, just drop the arp packet on 1249 * the floor. 1250 */ 1251 1252 /* Count the elegible carp interfaces with this address */ 1253 if (*count == 0) 1254 *count = carp_addrcount( 1255 (struct carp_if *)ia->ia_ifp->if_carpdev->if_carp, 1256 ia, CARP_COUNT_RUNNING); 1257 1258 /* This should never happen, but... */ 1259 if (*count == 0) 1260 return (0); 1261 1262 if (carp_hash(sc, src) % *count == index - 1 && 1263 sc->sc_state == MASTER) { 1264 return (1); 1265 } 1266 } else { 1267 if (sc->sc_state == MASTER) 1268 return (1); 1269 } 1270 1271 return (0); 1272 } 1273 1274 #ifdef INET6 1275 struct ifaddr * 1276 carp_iamatch6(void *v, struct in6_addr *taddr) 1277 { 1278 struct carp_if *cif = v; 1279 struct carp_softc *vh; 1280 struct ifaddr *ifa; 1281 1282 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 1283 IFADDR_FOREACH(ifa, &vh->sc_if) { 1284 if (IN6_ARE_ADDR_EQUAL(taddr, 1285 &ifatoia6(ifa)->ia_addr.sin6_addr) && 1286 ((vh->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) == 1287 (IFF_UP|IFF_RUNNING)) && vh->sc_state == MASTER) 1288 return (ifa); 1289 } 1290 } 1291 1292 return (NULL); 1293 } 1294 #endif /* INET6 */ 1295 1296 struct ifnet * 1297 carp_ourether(void *v, struct ether_header *eh, u_char iftype, int src) 1298 { 1299 struct carp_if *cif = (struct carp_if *)v; 1300 struct carp_softc *vh; 1301 u_int8_t *ena; 1302 1303 if (src) 1304 ena = (u_int8_t *)&eh->ether_shost; 1305 else 1306 ena = (u_int8_t *)&eh->ether_dhost; 1307 1308 switch (iftype) { 1309 case IFT_ETHER: 1310 case IFT_FDDI: 1311 if (ena[0] || ena[1] || ena[2] != 0x5e || ena[3] || ena[4] != 1) 1312 return (NULL); 1313 break; 1314 case IFT_ISO88025: 1315 if (ena[0] != 3 || ena[1] || ena[4] || ena[5]) 1316 return (NULL); 1317 break; 1318 default: 1319 return (NULL); 1320 break; 1321 } 1322 1323 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) 1324 if ((vh->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) == 1325 (IFF_UP|IFF_RUNNING) && vh->sc_state == MASTER && 1326 !bcmp(ena, CLLADDR(vh->sc_if.if_sadl), 1327 ETHER_ADDR_LEN)) { 1328 return (&vh->sc_if); 1329 } 1330 1331 return (NULL); 1332 } 1333 1334 int 1335 carp_input(struct mbuf *m, u_int8_t *shost, u_int8_t *dhost, u_int16_t etype) 1336 { 1337 struct ether_header eh; 1338 struct carp_if *cif = (struct carp_if *)m->m_pkthdr.rcvif->if_carp; 1339 struct ifnet *ifp; 1340 1341 bcopy(shost, &eh.ether_shost, sizeof(eh.ether_shost)); 1342 bcopy(dhost, &eh.ether_dhost, sizeof(eh.ether_dhost)); 1343 eh.ether_type = etype; 1344 1345 if (m->m_flags & (M_BCAST|M_MCAST)) { 1346 struct carp_softc *vh; 1347 struct mbuf *m0; 1348 1349 /* 1350 * XXX Should really check the list of multicast addresses 1351 * for each CARP interface _before_ copying. 1352 */ 1353 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 1354 m0 = m_copym(m, 0, M_COPYALL, M_DONTWAIT); 1355 if (m0 == NULL) 1356 continue; 1357 m0->m_pkthdr.rcvif = &vh->sc_if; 1358 ether_input(&vh->sc_if, m0); 1359 } 1360 return (1); 1361 } 1362 1363 ifp = carp_ourether(cif, &eh, m->m_pkthdr.rcvif->if_type, 0); 1364 if (ifp == NULL) { 1365 return (1); 1366 } 1367 1368 m->m_pkthdr.rcvif = ifp; 1369 1370 #if NBPFILTER > 0 1371 if (ifp->if_bpf) 1372 bpf_mtap(ifp->if_bpf, m); 1373 #endif 1374 ifp->if_ipackets++; 1375 ether_input(ifp, m); 1376 return (0); 1377 } 1378 1379 void 1380 carp_master_down(void *v) 1381 { 1382 struct carp_softc *sc = v; 1383 1384 switch (sc->sc_state) { 1385 case INIT: 1386 printf("%s: master_down event in INIT state\n", 1387 sc->sc_if.if_xname); 1388 break; 1389 case MASTER: 1390 break; 1391 case BACKUP: 1392 carp_set_state(sc, MASTER); 1393 carp_send_ad(sc); 1394 carp_send_arp(sc); 1395 #ifdef INET6 1396 carp_send_na(sc); 1397 #endif /* INET6 */ 1398 carp_setrun(sc, 0); 1399 carp_setroute(sc, RTM_ADD); 1400 break; 1401 } 1402 } 1403 1404 /* 1405 * When in backup state, af indicates whether to reset the master down timer 1406 * for v4 or v6. If it's set to zero, reset the ones which are already pending. 1407 */ 1408 void 1409 carp_setrun(struct carp_softc *sc, sa_family_t af) 1410 { 1411 struct timeval tv; 1412 1413 if (sc->sc_carpdev == NULL) { 1414 sc->sc_if.if_flags &= ~IFF_RUNNING; 1415 carp_set_state(sc, INIT); 1416 return; 1417 } 1418 1419 if (sc->sc_if.if_flags & IFF_UP && sc->sc_vhid > 0 && 1420 (sc->sc_naddrs || sc->sc_naddrs6) && !sc->sc_suppress) { 1421 sc->sc_if.if_flags |= IFF_RUNNING; 1422 } else { 1423 sc->sc_if.if_flags &= ~IFF_RUNNING; 1424 carp_setroute(sc, RTM_DELETE); 1425 return; 1426 } 1427 1428 switch (sc->sc_state) { 1429 case INIT: 1430 carp_set_state(sc, BACKUP); 1431 carp_setroute(sc, RTM_DELETE); 1432 carp_setrun(sc, 0); 1433 break; 1434 case BACKUP: 1435 callout_stop(&sc->sc_ad_tmo); 1436 tv.tv_sec = 3 * sc->sc_advbase; 1437 tv.tv_usec = sc->sc_advskew * 1000000 / 256; 1438 switch (af) { 1439 #ifdef INET 1440 case AF_INET: 1441 callout_schedule(&sc->sc_md_tmo, tvtohz(&tv)); 1442 break; 1443 #endif /* INET */ 1444 #ifdef INET6 1445 case AF_INET6: 1446 callout_schedule(&sc->sc_md6_tmo, tvtohz(&tv)); 1447 break; 1448 #endif /* INET6 */ 1449 default: 1450 if (sc->sc_naddrs) 1451 callout_schedule(&sc->sc_md_tmo, tvtohz(&tv)); 1452 if (sc->sc_naddrs6) 1453 callout_schedule(&sc->sc_md6_tmo, tvtohz(&tv)); 1454 break; 1455 } 1456 break; 1457 case MASTER: 1458 tv.tv_sec = sc->sc_advbase; 1459 tv.tv_usec = sc->sc_advskew * 1000000 / 256; 1460 callout_schedule(&sc->sc_ad_tmo, tvtohz(&tv)); 1461 break; 1462 } 1463 } 1464 1465 void 1466 carp_multicast_cleanup(struct carp_softc *sc) 1467 { 1468 struct ip_moptions *imo = &sc->sc_imo; 1469 #ifdef INET6 1470 struct ip6_moptions *im6o = &sc->sc_im6o; 1471 #endif 1472 u_int16_t n = imo->imo_num_memberships; 1473 1474 /* Clean up our own multicast memberships */ 1475 while (n-- > 0) { 1476 if (imo->imo_membership[n] != NULL) { 1477 in_delmulti(imo->imo_membership[n]); 1478 imo->imo_membership[n] = NULL; 1479 } 1480 } 1481 imo->imo_num_memberships = 0; 1482 imo->imo_multicast_ifp = NULL; 1483 1484 #ifdef INET6 1485 while (!LIST_EMPTY(&im6o->im6o_memberships)) { 1486 struct in6_multi_mship *imm = 1487 LIST_FIRST(&im6o->im6o_memberships); 1488 1489 LIST_REMOVE(imm, i6mm_chain); 1490 in6_leavegroup(imm); 1491 } 1492 im6o->im6o_multicast_ifp = NULL; 1493 #endif 1494 1495 /* And any other multicast memberships */ 1496 carp_ether_purgemulti(sc); 1497 } 1498 1499 int 1500 carp_set_ifp(struct carp_softc *sc, struct ifnet *ifp) 1501 { 1502 struct carp_if *cif, *ncif = NULL; 1503 struct carp_softc *vr, *after = NULL; 1504 int myself = 0, error = 0; 1505 int s; 1506 1507 if (ifp == sc->sc_carpdev) 1508 return (0); 1509 1510 if (ifp != NULL) { 1511 if ((ifp->if_flags & IFF_MULTICAST) == 0) 1512 return (EADDRNOTAVAIL); 1513 1514 if (ifp->if_type == IFT_CARP) 1515 return (EINVAL); 1516 1517 if (ifp->if_carp == NULL) { 1518 MALLOC(ncif, struct carp_if *, sizeof(*cif), 1519 M_IFADDR, M_NOWAIT); 1520 if (ncif == NULL) 1521 return (ENOBUFS); 1522 if ((error = ifpromisc(ifp, 1))) { 1523 FREE(ncif, M_IFADDR); 1524 return (error); 1525 } 1526 1527 ncif->vhif_ifp = ifp; 1528 TAILQ_INIT(&ncif->vhif_vrs); 1529 } else { 1530 cif = (struct carp_if *)ifp->if_carp; 1531 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) 1532 if (vr != sc && vr->sc_vhid == sc->sc_vhid) 1533 return (EINVAL); 1534 } 1535 1536 /* detach from old interface */ 1537 if (sc->sc_carpdev != NULL) 1538 carpdetach(sc); 1539 1540 /* join multicast groups */ 1541 if (sc->sc_naddrs < 0 && 1542 (error = carp_join_multicast(sc)) != 0) { 1543 if (ncif != NULL) 1544 FREE(ncif, M_IFADDR); 1545 return (error); 1546 } 1547 1548 #ifdef INET6 1549 if (sc->sc_naddrs6 < 0 && 1550 (error = carp_join_multicast6(sc)) != 0) { 1551 if (ncif != NULL) 1552 FREE(ncif, M_IFADDR); 1553 carp_multicast_cleanup(sc); 1554 return (error); 1555 } 1556 #endif 1557 1558 /* attach carp interface to physical interface */ 1559 if (ncif != NULL) 1560 ifp->if_carp = (void *)ncif; 1561 sc->sc_carpdev = ifp; 1562 cif = (struct carp_if *)ifp->if_carp; 1563 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) { 1564 if (vr == sc) 1565 myself = 1; 1566 if (vr->sc_vhid < sc->sc_vhid) 1567 after = vr; 1568 } 1569 1570 if (!myself) { 1571 /* We're trying to keep things in order */ 1572 if (after == NULL) { 1573 TAILQ_INSERT_TAIL(&cif->vhif_vrs, sc, sc_list); 1574 } else { 1575 TAILQ_INSERT_AFTER(&cif->vhif_vrs, after, 1576 sc, sc_list); 1577 } 1578 cif->vhif_nvrs++; 1579 } 1580 if (sc->sc_naddrs || sc->sc_naddrs6) 1581 sc->sc_if.if_flags |= IFF_UP; 1582 carp_set_enaddr(sc); 1583 s = splnet(); 1584 /* XXX linkstatehooks establish */ 1585 carp_carpdev_state(ifp); 1586 splx(s); 1587 } else { 1588 carpdetach(sc); 1589 sc->sc_if.if_flags &= ~(IFF_UP|IFF_RUNNING); 1590 } 1591 return (0); 1592 } 1593 1594 void 1595 carp_set_enaddr(struct carp_softc *sc) 1596 { 1597 uint8_t enaddr[ETHER_ADDR_LEN]; 1598 if (sc->sc_carpdev && sc->sc_carpdev->if_type == IFT_ISO88025) { 1599 enaddr[0] = 3; 1600 enaddr[1] = 0; 1601 enaddr[2] = 0x40 >> (sc->sc_vhid - 1); 1602 enaddr[3] = 0x40000 >> (sc->sc_vhid - 1); 1603 enaddr[4] = 0; 1604 enaddr[5] = 0; 1605 } else { 1606 enaddr[0] = 0; 1607 enaddr[1] = 0; 1608 enaddr[2] = 0x5e; 1609 enaddr[3] = 0; 1610 enaddr[4] = 1; 1611 enaddr[5] = sc->sc_vhid; 1612 } 1613 if_set_sadl(&sc->sc_if, enaddr, sizeof(enaddr)); 1614 } 1615 1616 void 1617 carp_addr_updated(void *v) 1618 { 1619 struct carp_softc *sc = (struct carp_softc *) v; 1620 struct ifaddr *ifa; 1621 int new_naddrs = 0, new_naddrs6 = 0; 1622 1623 IFADDR_FOREACH(ifa, &sc->sc_if) { 1624 if (ifa->ifa_addr->sa_family == AF_INET) 1625 new_naddrs++; 1626 else if (ifa->ifa_addr->sa_family == AF_INET6) 1627 new_naddrs6++; 1628 } 1629 1630 /* Handle a callback after SIOCDIFADDR */ 1631 if (new_naddrs < sc->sc_naddrs || new_naddrs6 < sc->sc_naddrs6) { 1632 struct in_addr mc_addr; 1633 struct in_multi *inm; 1634 1635 sc->sc_naddrs = new_naddrs; 1636 sc->sc_naddrs6 = new_naddrs6; 1637 1638 /* Re-establish multicast membership removed by in_control */ 1639 mc_addr.s_addr = INADDR_CARP_GROUP; 1640 IN_LOOKUP_MULTI(mc_addr, &sc->sc_if, inm); 1641 if (inm == NULL) { 1642 bzero(&sc->sc_imo, sizeof(sc->sc_imo)); 1643 1644 if (sc->sc_carpdev != NULL && sc->sc_naddrs > 0) 1645 carp_join_multicast(sc); 1646 } 1647 1648 if (sc->sc_naddrs == 0 && sc->sc_naddrs6 == 0) { 1649 sc->sc_if.if_flags &= ~IFF_UP; 1650 carp_set_state(sc, INIT); 1651 } else 1652 carp_hmac_prepare(sc); 1653 } 1654 1655 carp_setrun(sc, 0); 1656 } 1657 1658 int 1659 carp_set_addr(struct carp_softc *sc, struct sockaddr_in *sin) 1660 { 1661 struct ifnet *ifp = sc->sc_carpdev; 1662 struct in_ifaddr *ia, *ia_if; 1663 int error = 0; 1664 1665 if (sin->sin_addr.s_addr == 0) { 1666 if (!(sc->sc_if.if_flags & IFF_UP)) 1667 carp_set_state(sc, INIT); 1668 if (sc->sc_naddrs) 1669 sc->sc_if.if_flags |= IFF_UP; 1670 carp_setrun(sc, 0); 1671 return (0); 1672 } 1673 1674 /* we have to do this by hand to ensure we don't match on ourselves */ 1675 ia_if = NULL; 1676 for (ia = TAILQ_FIRST(&in_ifaddrhead); ia; 1677 ia = TAILQ_NEXT(ia, ia_list)) { 1678 1679 /* and, yeah, we need a multicast-capable iface too */ 1680 if (ia->ia_ifp != &sc->sc_if && 1681 ia->ia_ifp->if_type != IFT_CARP && 1682 (ia->ia_ifp->if_flags & IFF_MULTICAST) && 1683 (sin->sin_addr.s_addr & ia->ia_subnetmask) == 1684 ia->ia_subnet) { 1685 if (!ia_if) 1686 ia_if = ia; 1687 } 1688 } 1689 1690 if (ia_if) { 1691 ia = ia_if; 1692 if (ifp) { 1693 if (ifp != ia->ia_ifp) 1694 return (EADDRNOTAVAIL); 1695 } else { 1696 ifp = ia->ia_ifp; 1697 } 1698 } 1699 1700 if ((error = carp_set_ifp(sc, ifp))) 1701 return (error); 1702 1703 if (sc->sc_carpdev == NULL) 1704 return (EADDRNOTAVAIL); 1705 1706 if (sc->sc_naddrs == 0 && (error = carp_join_multicast(sc)) != 0) 1707 return (error); 1708 1709 sc->sc_naddrs++; 1710 if (sc->sc_carpdev != NULL) 1711 sc->sc_if.if_flags |= IFF_UP; 1712 1713 carp_set_state(sc, INIT); 1714 carp_setrun(sc, 0); 1715 1716 /* 1717 * Hook if_addrhooks so that we get a callback after in_ifinit has run, 1718 * to correct any inappropriate routes that it inserted. 1719 */ 1720 if (sc->ah_cookie == 0) { 1721 /* XXX link address hook */ 1722 } 1723 1724 return (0); 1725 } 1726 1727 int 1728 carp_join_multicast(struct carp_softc *sc) 1729 { 1730 struct ip_moptions *imo = &sc->sc_imo, tmpimo; 1731 struct in_addr addr; 1732 1733 bzero(&tmpimo, sizeof(tmpimo)); 1734 addr.s_addr = INADDR_CARP_GROUP; 1735 if ((tmpimo.imo_membership[0] = 1736 in_addmulti(&addr, &sc->sc_if)) == NULL) { 1737 return (ENOBUFS); 1738 } 1739 1740 imo->imo_membership[0] = tmpimo.imo_membership[0]; 1741 imo->imo_num_memberships = 1; 1742 imo->imo_multicast_ifp = &sc->sc_if; 1743 imo->imo_multicast_ttl = CARP_DFLTTL; 1744 imo->imo_multicast_loop = 0; 1745 return (0); 1746 } 1747 1748 1749 #ifdef INET6 1750 int 1751 carp_set_addr6(struct carp_softc *sc, struct sockaddr_in6 *sin6) 1752 { 1753 struct ifnet *ifp = sc->sc_carpdev; 1754 struct in6_ifaddr *ia, *ia_if; 1755 int error = 0; 1756 1757 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) { 1758 if (!(sc->sc_if.if_flags & IFF_UP)) 1759 carp_set_state(sc, INIT); 1760 if (sc->sc_naddrs6) 1761 sc->sc_if.if_flags |= IFF_UP; 1762 carp_setrun(sc, 0); 1763 return (0); 1764 } 1765 1766 /* we have to do this by hand to ensure we don't match on ourselves */ 1767 ia_if = NULL; 1768 for (ia = in6_ifaddr; ia; ia = ia->ia_next) { 1769 int i; 1770 1771 for (i = 0; i < 4; i++) { 1772 if ((sin6->sin6_addr.s6_addr32[i] & 1773 ia->ia_prefixmask.sin6_addr.s6_addr32[i]) != 1774 (ia->ia_addr.sin6_addr.s6_addr32[i] & 1775 ia->ia_prefixmask.sin6_addr.s6_addr32[i])) 1776 break; 1777 } 1778 /* and, yeah, we need a multicast-capable iface too */ 1779 if (ia->ia_ifp != &sc->sc_if && 1780 ia->ia_ifp->if_type != IFT_CARP && 1781 (ia->ia_ifp->if_flags & IFF_MULTICAST) && 1782 (i == 4)) { 1783 if (!ia_if) 1784 ia_if = ia; 1785 } 1786 } 1787 1788 if (ia_if) { 1789 ia = ia_if; 1790 if (sc->sc_carpdev) { 1791 if (sc->sc_carpdev != ia->ia_ifp) 1792 return (EADDRNOTAVAIL); 1793 } else { 1794 ifp = ia->ia_ifp; 1795 } 1796 } 1797 1798 if ((error = carp_set_ifp(sc, ifp))) 1799 return (error); 1800 1801 if (sc->sc_carpdev == NULL) 1802 return (EADDRNOTAVAIL); 1803 1804 if (sc->sc_naddrs6 == 0 && (error = carp_join_multicast6(sc)) != 0) 1805 return (error); 1806 1807 sc->sc_naddrs6++; 1808 if (sc->sc_carpdev != NULL) 1809 sc->sc_if.if_flags |= IFF_UP; 1810 carp_set_state(sc, INIT); 1811 carp_setrun(sc, 0); 1812 1813 return (0); 1814 } 1815 1816 int 1817 carp_join_multicast6(struct carp_softc *sc) 1818 { 1819 struct in6_multi_mship *imm, *imm2; 1820 struct ip6_moptions *im6o = &sc->sc_im6o; 1821 struct sockaddr_in6 addr6; 1822 int error; 1823 1824 /* Join IPv6 CARP multicast group */ 1825 bzero(&addr6, sizeof(addr6)); 1826 addr6.sin6_family = AF_INET6; 1827 addr6.sin6_len = sizeof(addr6); 1828 addr6.sin6_addr.s6_addr16[0] = htons(0xff02); 1829 addr6.sin6_addr.s6_addr16[1] = htons(sc->sc_if.if_index); 1830 addr6.sin6_addr.s6_addr8[15] = 0x12; 1831 if ((imm = in6_joingroup(&sc->sc_if, 1832 &addr6.sin6_addr, &error, 0)) == NULL) { 1833 return (error); 1834 } 1835 /* join solicited multicast address */ 1836 bzero(&addr6.sin6_addr, sizeof(addr6.sin6_addr)); 1837 addr6.sin6_addr.s6_addr16[0] = htons(0xff02); 1838 addr6.sin6_addr.s6_addr16[1] = htons(sc->sc_if.if_index); 1839 addr6.sin6_addr.s6_addr32[1] = 0; 1840 addr6.sin6_addr.s6_addr32[2] = htonl(1); 1841 addr6.sin6_addr.s6_addr32[3] = 0; 1842 addr6.sin6_addr.s6_addr8[12] = 0xff; 1843 if ((imm2 = in6_joingroup(&sc->sc_if, 1844 &addr6.sin6_addr, &error, 0)) == NULL) { 1845 in6_leavegroup(imm); 1846 return (error); 1847 } 1848 1849 /* apply v6 multicast membership */ 1850 im6o->im6o_multicast_ifp = &sc->sc_if; 1851 if (imm) 1852 LIST_INSERT_HEAD(&im6o->im6o_memberships, imm, 1853 i6mm_chain); 1854 if (imm2) 1855 LIST_INSERT_HEAD(&im6o->im6o_memberships, imm2, 1856 i6mm_chain); 1857 1858 return (0); 1859 } 1860 1861 #endif /* INET6 */ 1862 1863 int 1864 carp_ioctl(struct ifnet *ifp, u_long cmd, void *addr) 1865 { 1866 struct lwp *l = curlwp; /* XXX */ 1867 struct carp_softc *sc = ifp->if_softc, *vr; 1868 struct carpreq carpr; 1869 struct ifaddr *ifa; 1870 struct ifreq *ifr; 1871 struct ifnet *cdev = NULL; 1872 int error = 0; 1873 1874 ifa = (struct ifaddr *)addr; 1875 ifr = (struct ifreq *)addr; 1876 1877 switch (cmd) { 1878 case SIOCSIFADDR: 1879 switch (ifa->ifa_addr->sa_family) { 1880 #ifdef INET 1881 case AF_INET: 1882 sc->sc_if.if_flags |= IFF_UP; 1883 bcopy(ifa->ifa_addr, ifa->ifa_dstaddr, 1884 sizeof(struct sockaddr)); 1885 error = carp_set_addr(sc, satosin(ifa->ifa_addr)); 1886 break; 1887 #endif /* INET */ 1888 #ifdef INET6 1889 case AF_INET6: 1890 sc->sc_if.if_flags|= IFF_UP; 1891 error = carp_set_addr6(sc, satosin6(ifa->ifa_addr)); 1892 break; 1893 #endif /* INET6 */ 1894 default: 1895 error = EAFNOSUPPORT; 1896 break; 1897 } 1898 break; 1899 1900 case SIOCSIFFLAGS: 1901 if (sc->sc_state != INIT && !(ifr->ifr_flags & IFF_UP)) { 1902 callout_stop(&sc->sc_ad_tmo); 1903 callout_stop(&sc->sc_md_tmo); 1904 callout_stop(&sc->sc_md6_tmo); 1905 if (sc->sc_state == MASTER) { 1906 /* we need the interface up to bow out */ 1907 sc->sc_if.if_flags |= IFF_UP; 1908 sc->sc_bow_out = 1; 1909 carp_send_ad(sc); 1910 } 1911 sc->sc_if.if_flags &= ~IFF_UP; 1912 carp_set_state(sc, INIT); 1913 carp_setrun(sc, 0); 1914 } else if (sc->sc_state == INIT && (ifr->ifr_flags & IFF_UP)) { 1915 sc->sc_if.if_flags |= IFF_UP; 1916 carp_setrun(sc, 0); 1917 } 1918 break; 1919 1920 case SIOCSVH: 1921 if (l == NULL) 1922 break; 1923 if ((error = kauth_authorize_network(l->l_cred, 1924 KAUTH_NETWORK_INTERFACE, 1925 KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd, 1926 NULL)) != 0) 1927 break; 1928 if ((error = copyin(ifr->ifr_data, &carpr, sizeof carpr))) 1929 break; 1930 error = 1; 1931 if (carpr.carpr_carpdev[0] != '\0' && 1932 (cdev = ifunit(carpr.carpr_carpdev)) == NULL) 1933 return (EINVAL); 1934 if ((error = carp_set_ifp(sc, cdev))) 1935 return (error); 1936 if (sc->sc_state != INIT && carpr.carpr_state != sc->sc_state) { 1937 switch (carpr.carpr_state) { 1938 case BACKUP: 1939 callout_stop(&sc->sc_ad_tmo); 1940 carp_set_state(sc, BACKUP); 1941 carp_setrun(sc, 0); 1942 carp_setroute(sc, RTM_DELETE); 1943 break; 1944 case MASTER: 1945 carp_master_down(sc); 1946 break; 1947 default: 1948 break; 1949 } 1950 } 1951 if (carpr.carpr_vhid > 0) { 1952 if (carpr.carpr_vhid > 255) { 1953 error = EINVAL; 1954 break; 1955 } 1956 if (sc->sc_carpdev) { 1957 struct carp_if *cif; 1958 cif = (struct carp_if *)sc->sc_carpdev->if_carp; 1959 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) 1960 if (vr != sc && 1961 vr->sc_vhid == carpr.carpr_vhid) 1962 return (EINVAL); 1963 } 1964 sc->sc_vhid = carpr.carpr_vhid; 1965 carp_set_enaddr(sc); 1966 carp_set_state(sc, INIT); 1967 error--; 1968 } 1969 if (carpr.carpr_advbase > 0 || carpr.carpr_advskew > 0) { 1970 if (carpr.carpr_advskew > 254) { 1971 error = EINVAL; 1972 break; 1973 } 1974 if (carpr.carpr_advbase > 255) { 1975 error = EINVAL; 1976 break; 1977 } 1978 sc->sc_advbase = carpr.carpr_advbase; 1979 sc->sc_advskew = carpr.carpr_advskew; 1980 error--; 1981 } 1982 bcopy(carpr.carpr_key, sc->sc_key, sizeof(sc->sc_key)); 1983 if (error > 0) 1984 error = EINVAL; 1985 else { 1986 error = 0; 1987 carp_setrun(sc, 0); 1988 } 1989 break; 1990 1991 case SIOCGVH: 1992 bzero(&carpr, sizeof(carpr)); 1993 if (sc->sc_carpdev != NULL) 1994 strlcpy(carpr.carpr_carpdev, sc->sc_carpdev->if_xname, 1995 IFNAMSIZ); 1996 carpr.carpr_state = sc->sc_state; 1997 carpr.carpr_vhid = sc->sc_vhid; 1998 carpr.carpr_advbase = sc->sc_advbase; 1999 carpr.carpr_advskew = sc->sc_advskew; 2000 2001 if ((l == NULL) || (error = kauth_authorize_network(l->l_cred, 2002 KAUTH_NETWORK_INTERFACE, 2003 KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd, 2004 NULL)) != 0) 2005 bcopy(sc->sc_key, carpr.carpr_key, 2006 sizeof(carpr.carpr_key)); 2007 error = copyout(&carpr, ifr->ifr_data, sizeof(carpr)); 2008 break; 2009 2010 case SIOCADDMULTI: 2011 error = carp_ether_addmulti(sc, ifr); 2012 break; 2013 2014 case SIOCDELMULTI: 2015 error = carp_ether_delmulti(sc, ifr); 2016 break; 2017 2018 default: 2019 error = EINVAL; 2020 } 2021 2022 carp_hmac_prepare(sc); 2023 return (error); 2024 } 2025 2026 2027 /* 2028 * Start output on carp interface. This function should never be called. 2029 */ 2030 void 2031 carp_start(struct ifnet *ifp) 2032 { 2033 #ifdef DEBUG 2034 printf("%s: start called\n", ifp->if_xname); 2035 #endif 2036 } 2037 2038 int 2039 carp_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *sa, 2040 struct rtentry *rt) 2041 { 2042 struct carp_softc *sc = ((struct carp_softc *)ifp->if_softc); 2043 2044 if (sc->sc_carpdev != NULL && sc->sc_state == MASTER) { 2045 return (sc->sc_carpdev->if_output(ifp, m, sa, rt)); 2046 } else { 2047 m_freem(m); 2048 return (ENETUNREACH); 2049 } 2050 } 2051 2052 void 2053 carp_set_state(struct carp_softc *sc, int state) 2054 { 2055 if (sc->sc_state == state) 2056 return; 2057 2058 sc->sc_state = state; 2059 switch (state) { 2060 case BACKUP: 2061 sc->sc_if.if_link_state = LINK_STATE_DOWN; 2062 break; 2063 case MASTER: 2064 sc->sc_if.if_link_state = LINK_STATE_UP; 2065 break; 2066 default: 2067 sc->sc_if.if_link_state = LINK_STATE_UNKNOWN; 2068 break; 2069 } 2070 rt_ifmsg(&sc->sc_if); 2071 } 2072 2073 void 2074 carp_carpdev_state(void *v) 2075 { 2076 struct carp_if *cif; 2077 struct carp_softc *sc; 2078 struct ifnet *ifp = v; 2079 2080 if (ifp->if_type == IFT_CARP) 2081 return; 2082 2083 cif = (struct carp_if *)ifp->if_carp; 2084 2085 TAILQ_FOREACH(sc, &cif->vhif_vrs, sc_list) { 2086 int suppressed = sc->sc_suppress; 2087 2088 if (sc->sc_carpdev->if_link_state == LINK_STATE_DOWN || 2089 !(sc->sc_carpdev->if_flags & IFF_UP)) { 2090 sc->sc_if.if_flags &= ~IFF_RUNNING; 2091 callout_stop(&sc->sc_ad_tmo); 2092 callout_stop(&sc->sc_md_tmo); 2093 callout_stop(&sc->sc_md6_tmo); 2094 carp_set_state(sc, INIT); 2095 sc->sc_suppress = 1; 2096 carp_setrun(sc, 0); 2097 if (!suppressed) { 2098 carp_suppress_preempt++; 2099 if (carp_suppress_preempt == 1) 2100 carp_send_ad_all(); 2101 } 2102 } else { 2103 carp_set_state(sc, INIT); 2104 sc->sc_suppress = 0; 2105 carp_setrun(sc, 0); 2106 if (suppressed) 2107 carp_suppress_preempt--; 2108 } 2109 } 2110 } 2111 2112 int 2113 carp_ether_addmulti(struct carp_softc *sc, struct ifreq *ifr) 2114 { 2115 const struct sockaddr *sa = ifreq_getaddr(SIOCADDMULTI, ifr); 2116 struct ifnet *ifp; 2117 struct carp_mc_entry *mc; 2118 u_int8_t addrlo[ETHER_ADDR_LEN], addrhi[ETHER_ADDR_LEN]; 2119 int error; 2120 2121 ifp = sc->sc_carpdev; 2122 if (ifp == NULL) 2123 return (EINVAL); 2124 2125 error = ether_addmulti(sa, &sc->sc_ac); 2126 if (error != ENETRESET) 2127 return (error); 2128 2129 /* 2130 * This is new multicast address. We have to tell parent 2131 * about it. Also, remember this multicast address so that 2132 * we can delete them on unconfigure. 2133 */ 2134 MALLOC(mc, struct carp_mc_entry *, sizeof(struct carp_mc_entry), 2135 M_DEVBUF, M_NOWAIT); 2136 if (mc == NULL) { 2137 error = ENOMEM; 2138 goto alloc_failed; 2139 } 2140 2141 /* 2142 * As ether_addmulti() returns ENETRESET, following two 2143 * statement shouldn't fail. 2144 */ 2145 (void)ether_multiaddr(sa, addrlo, addrhi); 2146 ETHER_LOOKUP_MULTI(addrlo, addrhi, &sc->sc_ac, mc->mc_enm); 2147 memcpy(&mc->mc_addr, sa, sa->sa_len); 2148 LIST_INSERT_HEAD(&sc->carp_mc_listhead, mc, mc_entries); 2149 2150 error = (*ifp->if_ioctl)(ifp, SIOCADDMULTI, (void *)ifr); 2151 if (error != 0) 2152 goto ioctl_failed; 2153 2154 return (error); 2155 2156 ioctl_failed: 2157 LIST_REMOVE(mc, mc_entries); 2158 FREE(mc, M_DEVBUF); 2159 alloc_failed: 2160 (void)ether_delmulti(sa, &sc->sc_ac); 2161 2162 return (error); 2163 } 2164 2165 int 2166 carp_ether_delmulti(struct carp_softc *sc, struct ifreq *ifr) 2167 { 2168 const struct sockaddr *sa = ifreq_getaddr(SIOCDELMULTI, ifr); 2169 struct ifnet *ifp; 2170 struct ether_multi *enm; 2171 struct carp_mc_entry *mc; 2172 u_int8_t addrlo[ETHER_ADDR_LEN], addrhi[ETHER_ADDR_LEN]; 2173 int error; 2174 2175 ifp = sc->sc_carpdev; 2176 if (ifp == NULL) 2177 return (EINVAL); 2178 2179 /* 2180 * Find a key to lookup carp_mc_entry. We have to do this 2181 * before calling ether_delmulti for obvious reason. 2182 */ 2183 if ((error = ether_multiaddr(sa, addrlo, addrhi)) != 0) 2184 return (error); 2185 ETHER_LOOKUP_MULTI(addrlo, addrhi, &sc->sc_ac, enm); 2186 if (enm == NULL) 2187 return (EINVAL); 2188 2189 LIST_FOREACH(mc, &sc->carp_mc_listhead, mc_entries) 2190 if (mc->mc_enm == enm) 2191 break; 2192 2193 /* We won't delete entries we didn't add */ 2194 if (mc == NULL) 2195 return (EINVAL); 2196 2197 error = ether_delmulti(sa, &sc->sc_ac); 2198 if (error != ENETRESET) 2199 return (error); 2200 2201 /* We no longer use this multicast address. Tell parent so. */ 2202 error = (*ifp->if_ioctl)(ifp, SIOCDELMULTI, (void *)ifr); 2203 if (error == 0) { 2204 /* And forget about this address. */ 2205 LIST_REMOVE(mc, mc_entries); 2206 FREE(mc, M_DEVBUF); 2207 } else 2208 (void)ether_addmulti(sa, &sc->sc_ac); 2209 return (error); 2210 } 2211 2212 /* 2213 * Delete any multicast address we have asked to add from parent 2214 * interface. Called when the carp is being unconfigured. 2215 */ 2216 void 2217 carp_ether_purgemulti(struct carp_softc *sc) 2218 { 2219 struct ifnet *ifp = sc->sc_carpdev; /* Parent. */ 2220 struct carp_mc_entry *mc; 2221 union { 2222 struct ifreq ifreq; 2223 struct { 2224 char ifr_name[IFNAMSIZ]; 2225 struct sockaddr_storage ifr_ss; 2226 } ifreq_storage; 2227 } u; 2228 struct ifreq *ifr = &u.ifreq; 2229 2230 if (ifp == NULL) 2231 return; 2232 2233 memcpy(ifr->ifr_name, ifp->if_xname, IFNAMSIZ); 2234 while ((mc = LIST_FIRST(&sc->carp_mc_listhead)) != NULL) { 2235 memcpy(&ifr->ifr_addr, &mc->mc_addr, mc->mc_addr.ss_len); 2236 (void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, (void *)ifr); 2237 LIST_REMOVE(mc, mc_entries); 2238 FREE(mc, M_DEVBUF); 2239 } 2240 } 2241 2242 static int 2243 sysctl_net_inet_carp_stats(SYSCTLFN_ARGS) 2244 { 2245 2246 return (NETSTAT_SYSCTL(carpstat_percpu, CARP_NSTATS)); 2247 } 2248 2249 SYSCTL_SETUP(sysctl_net_inet_carp_setup, "sysctl net.inet.carp subtree setup") 2250 { 2251 2252 sysctl_createv(clog, 0, NULL, NULL, 2253 CTLFLAG_PERMANENT, 2254 CTLTYPE_NODE, "net", NULL, 2255 NULL, 0, NULL, 0, 2256 CTL_NET, CTL_EOL); 2257 sysctl_createv(clog, 0, NULL, NULL, 2258 CTLFLAG_PERMANENT, 2259 CTLTYPE_NODE, "inet", NULL, 2260 NULL, 0, NULL, 0, 2261 CTL_NET, PF_INET, CTL_EOL); 2262 sysctl_createv(clog, 0, NULL, NULL, 2263 CTLFLAG_PERMANENT, 2264 CTLTYPE_NODE, "carp", 2265 SYSCTL_DESCR("CARP related settings"), 2266 NULL, 0, NULL, 0, 2267 CTL_NET, PF_INET, IPPROTO_CARP, CTL_EOL); 2268 2269 sysctl_createv(clog, 0, NULL, NULL, 2270 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2271 CTLTYPE_INT, "preempt", 2272 SYSCTL_DESCR("Enable CARP Preempt"), 2273 NULL, 0, &carp_opts[CARPCTL_PREEMPT], 0, 2274 CTL_NET, PF_INET, IPPROTO_CARP, 2275 CTL_CREATE, CTL_EOL); 2276 sysctl_createv(clog, 0, NULL, NULL, 2277 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2278 CTLTYPE_INT, "arpbalance", 2279 SYSCTL_DESCR("Enable ARP balancing"), 2280 NULL, 0, &carp_opts[CARPCTL_ARPBALANCE], 0, 2281 CTL_NET, PF_INET, IPPROTO_CARP, 2282 CTL_CREATE, CTL_EOL); 2283 sysctl_createv(clog, 0, NULL, NULL, 2284 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2285 CTLTYPE_INT, "allow", 2286 SYSCTL_DESCR("Enable CARP"), 2287 NULL, 0, &carp_opts[CARPCTL_ALLOW], 0, 2288 CTL_NET, PF_INET, IPPROTO_CARP, 2289 CTL_CREATE, CTL_EOL); 2290 sysctl_createv(clog, 0, NULL, NULL, 2291 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2292 CTLTYPE_INT, "log", 2293 SYSCTL_DESCR("CARP logging"), 2294 NULL, 0, &carp_opts[CARPCTL_LOG], 0, 2295 CTL_NET, PF_INET, IPPROTO_CARP, 2296 CTL_CREATE, CTL_EOL); 2297 sysctl_createv(clog, 0, NULL, NULL, 2298 CTLFLAG_PERMANENT, 2299 CTLTYPE_STRUCT, "stats", 2300 SYSCTL_DESCR("CARP statistics"), 2301 sysctl_net_inet_carp_stats, 0, NULL, 0, 2302 CTL_NET, PF_INET, IPPROTO_CARP, CARPCTL_STATS, 2303 CTL_EOL); 2304 } 2305