1 /* $NetBSD: ip_carp.c,v 1.30 2009/01/11 02:45:54 christos Exp $ */ 2 /* $OpenBSD: ip_carp.c,v 1.113 2005/11/04 08:11:54 mcbride Exp $ */ 3 4 /* 5 * Copyright (c) 2002 Michael Shalayeff. All rights reserved. 6 * Copyright (c) 2003 Ryan McBride. All rights reserved. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 11 * 1. Redistributions of source code must retain the above copyright 12 * notice, this list of conditions and the following disclaimer. 13 * 2. Redistributions in binary form must reproduce the above copyright 14 * notice, this list of conditions and the following disclaimer in the 15 * documentation and/or other materials provided with the distribution. 16 * 17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR 18 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 19 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 20 * IN NO EVENT SHALL THE AUTHOR OR HIS RELATIVES BE LIABLE FOR ANY DIRECT, 21 * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES 22 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR 23 * SERVICES; LOSS OF MIND, USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, 25 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING 26 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF 27 * THE POSSIBILITY OF SUCH DAMAGE. 28 */ 29 30 #include <sys/cdefs.h> 31 __KERNEL_RCSID(0, "$NetBSD: ip_carp.c,v 1.30 2009/01/11 02:45:54 christos 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 nanotime(&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|M_ZERO); 759 if (!sc) 760 return (ENOMEM); 761 762 sc->sc_suppress = 0; 763 sc->sc_advbase = CARP_DFLTINTV; 764 sc->sc_vhid = -1; /* required setting */ 765 sc->sc_advskew = 0; 766 sc->sc_init_counter = 1; 767 sc->sc_naddrs = sc->sc_naddrs6 = 0; 768 #ifdef INET6 769 sc->sc_im6o.im6o_multicast_hlim = CARP_DFLTTL; 770 #endif /* INET6 */ 771 772 callout_init(&sc->sc_ad_tmo, 0); 773 callout_init(&sc->sc_md_tmo, 0); 774 callout_init(&sc->sc_md6_tmo, 0); 775 776 callout_setfunc(&sc->sc_ad_tmo, carp_send_ad, sc); 777 callout_setfunc(&sc->sc_md_tmo, carp_master_down, sc); 778 callout_setfunc(&sc->sc_md6_tmo, carp_master_down, sc); 779 780 LIST_INIT(&sc->carp_mc_listhead); 781 ifp = &sc->sc_if; 782 ifp->if_softc = sc; 783 snprintf(ifp->if_xname, sizeof ifp->if_xname, "%s%d", ifc->ifc_name, 784 unit); 785 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST; 786 ifp->if_ioctl = carp_ioctl; 787 ifp->if_start = carp_start; 788 ifp->if_output = carp_output; 789 ifp->if_type = IFT_CARP; 790 ifp->if_addrlen = ETHER_ADDR_LEN; 791 ifp->if_hdrlen = ETHER_HDR_LEN; 792 ifp->if_mtu = ETHERMTU; 793 IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen); 794 IFQ_SET_READY(&ifp->if_snd); 795 if_attach(ifp); 796 797 if_alloc_sadl(ifp); 798 ifp->if_broadcastaddr = etherbroadcastaddr; 799 carp_set_enaddr(sc); 800 LIST_INIT(&sc->sc_ac.ec_multiaddrs); 801 #if NBPFILTER > 0 802 bpfattach(ifp, DLT_EN10MB, ETHER_HDR_LEN); 803 #endif 804 return (0); 805 } 806 807 int 808 carp_clone_destroy(struct ifnet *ifp) 809 { 810 struct carp_softc *sc = ifp->if_softc; 811 812 carpdetach(ifp->if_softc); 813 ether_ifdetach(ifp); 814 if_detach(ifp); 815 callout_destroy(&sc->sc_ad_tmo); 816 callout_destroy(&sc->sc_md_tmo); 817 callout_destroy(&sc->sc_md6_tmo); 818 free(ifp->if_softc, M_DEVBUF); 819 820 return (0); 821 } 822 823 void 824 carpdetach(struct carp_softc *sc) 825 { 826 struct carp_if *cif; 827 int s; 828 829 callout_stop(&sc->sc_ad_tmo); 830 callout_stop(&sc->sc_md_tmo); 831 callout_stop(&sc->sc_md6_tmo); 832 833 if (sc->sc_suppress) 834 carp_suppress_preempt--; 835 sc->sc_suppress = 0; 836 837 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) 838 carp_suppress_preempt--; 839 sc->sc_sendad_errors = 0; 840 841 carp_set_state(sc, INIT); 842 sc->sc_if.if_flags &= ~IFF_UP; 843 carp_setrun(sc, 0); 844 carp_multicast_cleanup(sc); 845 846 s = splnet(); 847 if (sc->sc_carpdev != NULL) { 848 /* XXX linkstatehook removal */ 849 cif = (struct carp_if *)sc->sc_carpdev->if_carp; 850 TAILQ_REMOVE(&cif->vhif_vrs, sc, sc_list); 851 if (!--cif->vhif_nvrs) { 852 ifpromisc(sc->sc_carpdev, 0); 853 sc->sc_carpdev->if_carp = NULL; 854 free(cif, M_IFADDR); 855 } 856 } 857 sc->sc_carpdev = NULL; 858 splx(s); 859 } 860 861 /* Detach an interface from the carp. */ 862 void 863 carp_ifdetach(struct ifnet *ifp) 864 { 865 struct carp_softc *sc, *nextsc; 866 struct carp_if *cif = (struct carp_if *)ifp->if_carp; 867 868 for (sc = TAILQ_FIRST(&cif->vhif_vrs); sc; sc = nextsc) { 869 nextsc = TAILQ_NEXT(sc, sc_list); 870 carpdetach(sc); 871 } 872 } 873 874 int 875 carp_prepare_ad(struct mbuf *m, struct carp_softc *sc, 876 struct carp_header *ch) 877 { 878 if (sc->sc_init_counter) { 879 /* this could also be seconds since unix epoch */ 880 sc->sc_counter = arc4random(); 881 sc->sc_counter = sc->sc_counter << 32; 882 sc->sc_counter += arc4random(); 883 } else 884 sc->sc_counter++; 885 886 ch->carp_counter[0] = htonl((sc->sc_counter>>32)&0xffffffff); 887 ch->carp_counter[1] = htonl(sc->sc_counter&0xffffffff); 888 889 carp_hmac_generate(sc, ch->carp_counter, ch->carp_md); 890 891 return (0); 892 } 893 894 void 895 carp_send_ad_all(void) 896 { 897 struct ifnet *ifp; 898 struct carp_if *cif; 899 struct carp_softc *vh; 900 901 TAILQ_FOREACH(ifp, &ifnet, if_list) { 902 if (ifp->if_carp == NULL || ifp->if_type == IFT_CARP) 903 continue; 904 905 cif = (struct carp_if *)ifp->if_carp; 906 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 907 if ((vh->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) == 908 (IFF_UP|IFF_RUNNING) && vh->sc_state == MASTER) 909 carp_send_ad(vh); 910 } 911 } 912 } 913 914 915 void 916 carp_send_ad(void *v) 917 { 918 struct carp_header ch; 919 struct timeval tv; 920 struct carp_softc *sc = v; 921 struct carp_header *ch_ptr; 922 struct mbuf *m; 923 int error, len, advbase, advskew, s; 924 struct ifaddr *ifa; 925 struct sockaddr sa; 926 927 s = splsoftnet(); 928 929 advbase = advskew = 0; /* Sssssh compiler */ 930 if (sc->sc_carpdev == NULL) { 931 sc->sc_if.if_oerrors++; 932 goto retry_later; 933 } 934 935 /* bow out if we've gone to backup (the carp interface is going down) */ 936 if (sc->sc_bow_out) { 937 sc->sc_bow_out = 0; 938 advbase = 255; 939 advskew = 255; 940 } else { 941 advbase = sc->sc_advbase; 942 if (!carp_suppress_preempt || sc->sc_advskew > 240) 943 advskew = sc->sc_advskew; 944 else 945 advskew = 240; 946 tv.tv_sec = advbase; 947 tv.tv_usec = advskew * 1000000 / 256; 948 } 949 950 ch.carp_version = CARP_VERSION; 951 ch.carp_type = CARP_ADVERTISEMENT; 952 ch.carp_vhid = sc->sc_vhid; 953 ch.carp_advbase = advbase; 954 ch.carp_advskew = advskew; 955 ch.carp_authlen = 7; /* XXX DEFINE */ 956 ch.carp_pad1 = 0; /* must be zero */ 957 ch.carp_cksum = 0; 958 959 960 #ifdef INET 961 if (sc->sc_naddrs) { 962 struct ip *ip; 963 964 MGETHDR(m, M_DONTWAIT, MT_HEADER); 965 if (m == NULL) { 966 sc->sc_if.if_oerrors++; 967 CARP_STATINC(CARP_STAT_ONOMEM); 968 /* XXX maybe less ? */ 969 goto retry_later; 970 } 971 len = sizeof(*ip) + sizeof(ch); 972 m->m_pkthdr.len = len; 973 m->m_pkthdr.rcvif = NULL; 974 m->m_len = len; 975 MH_ALIGN(m, m->m_len); 976 m->m_flags |= M_MCAST; 977 ip = mtod(m, struct ip *); 978 ip->ip_v = IPVERSION; 979 ip->ip_hl = sizeof(*ip) >> 2; 980 ip->ip_tos = IPTOS_LOWDELAY; 981 ip->ip_len = htons(len); 982 ip->ip_id = 0; /* no need for id, we don't support fragments */ 983 ip->ip_off = htons(IP_DF); 984 ip->ip_ttl = CARP_DFLTTL; 985 ip->ip_p = IPPROTO_CARP; 986 ip->ip_sum = 0; 987 988 bzero(&sa, sizeof(sa)); 989 sa.sa_family = AF_INET; 990 ifa = ifaof_ifpforaddr(&sa, sc->sc_carpdev); 991 if (ifa == NULL) 992 ip->ip_src.s_addr = 0; 993 else 994 ip->ip_src.s_addr = 995 ifatoia(ifa)->ia_addr.sin_addr.s_addr; 996 ip->ip_dst.s_addr = INADDR_CARP_GROUP; 997 998 ch_ptr = (struct carp_header *)(&ip[1]); 999 bcopy(&ch, ch_ptr, sizeof(ch)); 1000 if (carp_prepare_ad(m, sc, ch_ptr)) 1001 goto retry_later; 1002 1003 m->m_data += sizeof(*ip); 1004 ch_ptr->carp_cksum = carp_cksum(m, len - sizeof(*ip)); 1005 m->m_data -= sizeof(*ip); 1006 1007 nanotime(&sc->sc_if.if_lastchange); 1008 sc->sc_if.if_opackets++; 1009 sc->sc_if.if_obytes += len; 1010 CARP_STATINC(CARP_STAT_OPACKETS); 1011 1012 error = ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo, 1013 NULL); 1014 if (error) { 1015 if (error == ENOBUFS) 1016 CARP_STATINC(CARP_STAT_ONOMEM); 1017 else 1018 CARP_LOG(sc, ("ip_output failed: %d", error)); 1019 sc->sc_if.if_oerrors++; 1020 if (sc->sc_sendad_errors < INT_MAX) 1021 sc->sc_sendad_errors++; 1022 if (sc->sc_sendad_errors == CARP_SENDAD_MAX_ERRORS) { 1023 carp_suppress_preempt++; 1024 if (carp_suppress_preempt == 1) 1025 carp_send_ad_all(); 1026 } 1027 sc->sc_sendad_success = 0; 1028 } else { 1029 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) { 1030 if (++sc->sc_sendad_success >= 1031 CARP_SENDAD_MIN_SUCCESS) { 1032 carp_suppress_preempt--; 1033 sc->sc_sendad_errors = 0; 1034 } 1035 } else 1036 sc->sc_sendad_errors = 0; 1037 } 1038 } 1039 #endif /* INET */ 1040 #ifdef INET6 1041 if (sc->sc_naddrs6) { 1042 struct ip6_hdr *ip6; 1043 1044 MGETHDR(m, M_DONTWAIT, MT_HEADER); 1045 if (m == NULL) { 1046 sc->sc_if.if_oerrors++; 1047 CARP_STATINC(CARP_STAT_ONOMEM); 1048 /* XXX maybe less ? */ 1049 goto retry_later; 1050 } 1051 len = sizeof(*ip6) + sizeof(ch); 1052 m->m_pkthdr.len = len; 1053 m->m_pkthdr.rcvif = NULL; 1054 m->m_len = len; 1055 MH_ALIGN(m, m->m_len); 1056 m->m_flags |= M_MCAST; 1057 ip6 = mtod(m, struct ip6_hdr *); 1058 bzero(ip6, sizeof(*ip6)); 1059 ip6->ip6_vfc |= IPV6_VERSION; 1060 ip6->ip6_hlim = CARP_DFLTTL; 1061 ip6->ip6_nxt = IPPROTO_CARP; 1062 1063 /* set the source address */ 1064 bzero(&sa, sizeof(sa)); 1065 sa.sa_family = AF_INET6; 1066 ifa = ifaof_ifpforaddr(&sa, sc->sc_carpdev); 1067 if (ifa == NULL) /* This should never happen with IPv6 */ 1068 bzero(&ip6->ip6_src, sizeof(struct in6_addr)); 1069 else 1070 bcopy(ifatoia6(ifa)->ia_addr.sin6_addr.s6_addr, 1071 &ip6->ip6_src, sizeof(struct in6_addr)); 1072 /* set the multicast destination */ 1073 1074 ip6->ip6_dst.s6_addr16[0] = htons(0xff02); 1075 ip6->ip6_dst.s6_addr8[15] = 0x12; 1076 if (in6_setscope(&ip6->ip6_dst, sc->sc_carpdev, NULL) != 0) { 1077 sc->sc_if.if_oerrors++; 1078 m_freem(m); 1079 CARP_LOG(sc, ("in6_setscope failed")); 1080 goto retry_later; 1081 } 1082 1083 ch_ptr = (struct carp_header *)(&ip6[1]); 1084 bcopy(&ch, ch_ptr, sizeof(ch)); 1085 if (carp_prepare_ad(m, sc, ch_ptr)) 1086 goto retry_later; 1087 1088 m->m_data += sizeof(*ip6); 1089 ch_ptr->carp_cksum = carp_cksum(m, len - sizeof(*ip6)); 1090 m->m_data -= sizeof(*ip6); 1091 1092 nanotime(&sc->sc_if.if_lastchange); 1093 sc->sc_if.if_opackets++; 1094 sc->sc_if.if_obytes += len; 1095 CARP_STATINC(CARP_STAT_OPACKETS6); 1096 1097 error = ip6_output(m, NULL, NULL, 0, &sc->sc_im6o, NULL, NULL); 1098 if (error) { 1099 if (error == ENOBUFS) 1100 CARP_STATINC(CARP_STAT_ONOMEM); 1101 else 1102 CARP_LOG(sc, ("ip6_output failed: %d", error)); 1103 sc->sc_if.if_oerrors++; 1104 if (sc->sc_sendad_errors < INT_MAX) 1105 sc->sc_sendad_errors++; 1106 if (sc->sc_sendad_errors == CARP_SENDAD_MAX_ERRORS) { 1107 carp_suppress_preempt++; 1108 if (carp_suppress_preempt == 1) 1109 carp_send_ad_all(); 1110 } 1111 sc->sc_sendad_success = 0; 1112 } else { 1113 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) { 1114 if (++sc->sc_sendad_success >= 1115 CARP_SENDAD_MIN_SUCCESS) { 1116 carp_suppress_preempt--; 1117 sc->sc_sendad_errors = 0; 1118 } 1119 } else 1120 sc->sc_sendad_errors = 0; 1121 } 1122 } 1123 #endif /* INET6 */ 1124 1125 retry_later: 1126 splx(s); 1127 if (advbase != 255 || advskew != 255) 1128 callout_schedule(&sc->sc_ad_tmo, tvtohz(&tv)); 1129 } 1130 1131 /* 1132 * Broadcast a gratuitous ARP request containing 1133 * the virtual router MAC address for each IP address 1134 * associated with the virtual router. 1135 */ 1136 void 1137 carp_send_arp(struct carp_softc *sc) 1138 { 1139 struct ifaddr *ifa; 1140 struct in_addr *in; 1141 int s = splsoftnet(); 1142 1143 IFADDR_FOREACH(ifa, &sc->sc_if) { 1144 1145 if (ifa->ifa_addr->sa_family != AF_INET) 1146 continue; 1147 1148 in = &ifatoia(ifa)->ia_addr.sin_addr; 1149 arprequest(sc->sc_carpdev, in, in, CLLADDR(sc->sc_if.if_sadl)); 1150 DELAY(1000); /* XXX */ 1151 } 1152 splx(s); 1153 } 1154 1155 #ifdef INET6 1156 void 1157 carp_send_na(struct carp_softc *sc) 1158 { 1159 struct ifaddr *ifa; 1160 struct in6_addr *in6; 1161 static struct in6_addr mcast = IN6ADDR_LINKLOCAL_ALLNODES_INIT; 1162 int s = splsoftnet(); 1163 1164 IFADDR_FOREACH(ifa, &sc->sc_if) { 1165 1166 if (ifa->ifa_addr->sa_family != AF_INET6) 1167 continue; 1168 1169 in6 = &ifatoia6(ifa)->ia_addr.sin6_addr; 1170 nd6_na_output(sc->sc_carpdev, &mcast, in6, 1171 ND_NA_FLAG_OVERRIDE, 1, NULL); 1172 DELAY(1000); /* XXX */ 1173 } 1174 splx(s); 1175 } 1176 #endif /* INET6 */ 1177 1178 /* 1179 * Based on bridge_hash() in if_bridge.c 1180 */ 1181 #define mix(a,b,c) \ 1182 do { \ 1183 a -= b; a -= c; a ^= (c >> 13); \ 1184 b -= c; b -= a; b ^= (a << 8); \ 1185 c -= a; c -= b; c ^= (b >> 13); \ 1186 a -= b; a -= c; a ^= (c >> 12); \ 1187 b -= c; b -= a; b ^= (a << 16); \ 1188 c -= a; c -= b; c ^= (b >> 5); \ 1189 a -= b; a -= c; a ^= (c >> 3); \ 1190 b -= c; b -= a; b ^= (a << 10); \ 1191 c -= a; c -= b; c ^= (b >> 15); \ 1192 } while (0) 1193 1194 u_int32_t 1195 carp_hash(struct carp_softc *sc, u_char *src) 1196 { 1197 u_int32_t a = 0x9e3779b9, b = sc->sc_hashkey[0], c = sc->sc_hashkey[1]; 1198 1199 c += sc->sc_key[3] << 24; 1200 c += sc->sc_key[2] << 16; 1201 c += sc->sc_key[1] << 8; 1202 c += sc->sc_key[0]; 1203 b += src[5] << 8; 1204 b += src[4]; 1205 a += src[3] << 24; 1206 a += src[2] << 16; 1207 a += src[1] << 8; 1208 a += src[0]; 1209 1210 mix(a, b, c); 1211 return (c); 1212 } 1213 1214 int 1215 carp_addrcount(struct carp_if *cif, struct in_ifaddr *ia, int type) 1216 { 1217 struct carp_softc *vh; 1218 struct ifaddr *ifa; 1219 int count = 0; 1220 1221 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 1222 if ((type == CARP_COUNT_RUNNING && 1223 (vh->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) == 1224 (IFF_UP|IFF_RUNNING)) || 1225 (type == CARP_COUNT_MASTER && vh->sc_state == MASTER)) { 1226 IFADDR_FOREACH(ifa, &vh->sc_if) { 1227 if (ifa->ifa_addr->sa_family == AF_INET && 1228 ia->ia_addr.sin_addr.s_addr == 1229 ifatoia(ifa)->ia_addr.sin_addr.s_addr) 1230 count++; 1231 } 1232 } 1233 } 1234 return (count); 1235 } 1236 1237 int 1238 carp_iamatch(struct in_ifaddr *ia, u_char *src, 1239 u_int32_t *count, u_int32_t index) 1240 { 1241 struct carp_softc *sc = ia->ia_ifp->if_softc; 1242 1243 if (carp_opts[CARPCTL_ARPBALANCE]) { 1244 /* 1245 * We use the source ip to decide which virtual host should 1246 * handle the request. If we're master of that virtual host, 1247 * then we respond, otherwise, just drop the arp packet on 1248 * the floor. 1249 */ 1250 1251 /* Count the elegible carp interfaces with this address */ 1252 if (*count == 0) 1253 *count = carp_addrcount( 1254 (struct carp_if *)ia->ia_ifp->if_carpdev->if_carp, 1255 ia, CARP_COUNT_RUNNING); 1256 1257 /* This should never happen, but... */ 1258 if (*count == 0) 1259 return (0); 1260 1261 if (carp_hash(sc, src) % *count == index - 1 && 1262 sc->sc_state == MASTER) { 1263 return (1); 1264 } 1265 } else { 1266 if (sc->sc_state == MASTER) 1267 return (1); 1268 } 1269 1270 return (0); 1271 } 1272 1273 #ifdef INET6 1274 struct ifaddr * 1275 carp_iamatch6(void *v, struct in6_addr *taddr) 1276 { 1277 struct carp_if *cif = v; 1278 struct carp_softc *vh; 1279 struct ifaddr *ifa; 1280 1281 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 1282 IFADDR_FOREACH(ifa, &vh->sc_if) { 1283 if (IN6_ARE_ADDR_EQUAL(taddr, 1284 &ifatoia6(ifa)->ia_addr.sin6_addr) && 1285 ((vh->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) == 1286 (IFF_UP|IFF_RUNNING)) && vh->sc_state == MASTER) 1287 return (ifa); 1288 } 1289 } 1290 1291 return (NULL); 1292 } 1293 #endif /* INET6 */ 1294 1295 struct ifnet * 1296 carp_ourether(void *v, struct ether_header *eh, u_char iftype, int src) 1297 { 1298 struct carp_if *cif = (struct carp_if *)v; 1299 struct carp_softc *vh; 1300 u_int8_t *ena; 1301 1302 if (src) 1303 ena = (u_int8_t *)&eh->ether_shost; 1304 else 1305 ena = (u_int8_t *)&eh->ether_dhost; 1306 1307 switch (iftype) { 1308 case IFT_ETHER: 1309 case IFT_FDDI: 1310 if (ena[0] || ena[1] || ena[2] != 0x5e || ena[3] || ena[4] != 1) 1311 return (NULL); 1312 break; 1313 case IFT_ISO88025: 1314 if (ena[0] != 3 || ena[1] || ena[4] || ena[5]) 1315 return (NULL); 1316 break; 1317 default: 1318 return (NULL); 1319 break; 1320 } 1321 1322 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) 1323 if ((vh->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) == 1324 (IFF_UP|IFF_RUNNING) && vh->sc_state == MASTER && 1325 !bcmp(ena, CLLADDR(vh->sc_if.if_sadl), 1326 ETHER_ADDR_LEN)) { 1327 return (&vh->sc_if); 1328 } 1329 1330 return (NULL); 1331 } 1332 1333 int 1334 carp_input(struct mbuf *m, u_int8_t *shost, u_int8_t *dhost, u_int16_t etype) 1335 { 1336 struct ether_header eh; 1337 struct carp_if *cif = (struct carp_if *)m->m_pkthdr.rcvif->if_carp; 1338 struct ifnet *ifp; 1339 1340 bcopy(shost, &eh.ether_shost, sizeof(eh.ether_shost)); 1341 bcopy(dhost, &eh.ether_dhost, sizeof(eh.ether_dhost)); 1342 eh.ether_type = etype; 1343 1344 if (m->m_flags & (M_BCAST|M_MCAST)) { 1345 struct carp_softc *vh; 1346 struct mbuf *m0; 1347 1348 /* 1349 * XXX Should really check the list of multicast addresses 1350 * for each CARP interface _before_ copying. 1351 */ 1352 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 1353 m0 = m_copym(m, 0, M_COPYALL, M_DONTWAIT); 1354 if (m0 == NULL) 1355 continue; 1356 m0->m_pkthdr.rcvif = &vh->sc_if; 1357 ether_input(&vh->sc_if, m0); 1358 } 1359 return (1); 1360 } 1361 1362 ifp = carp_ourether(cif, &eh, m->m_pkthdr.rcvif->if_type, 0); 1363 if (ifp == NULL) { 1364 return (1); 1365 } 1366 1367 m->m_pkthdr.rcvif = ifp; 1368 1369 #if NBPFILTER > 0 1370 if (ifp->if_bpf) 1371 bpf_mtap(ifp->if_bpf, m); 1372 #endif 1373 ifp->if_ipackets++; 1374 ether_input(ifp, m); 1375 return (0); 1376 } 1377 1378 void 1379 carp_master_down(void *v) 1380 { 1381 struct carp_softc *sc = v; 1382 1383 switch (sc->sc_state) { 1384 case INIT: 1385 printf("%s: master_down event in INIT state\n", 1386 sc->sc_if.if_xname); 1387 break; 1388 case MASTER: 1389 break; 1390 case BACKUP: 1391 carp_set_state(sc, MASTER); 1392 carp_send_ad(sc); 1393 carp_send_arp(sc); 1394 #ifdef INET6 1395 carp_send_na(sc); 1396 #endif /* INET6 */ 1397 carp_setrun(sc, 0); 1398 carp_setroute(sc, RTM_ADD); 1399 break; 1400 } 1401 } 1402 1403 /* 1404 * When in backup state, af indicates whether to reset the master down timer 1405 * for v4 or v6. If it's set to zero, reset the ones which are already pending. 1406 */ 1407 void 1408 carp_setrun(struct carp_softc *sc, sa_family_t af) 1409 { 1410 struct timeval tv; 1411 1412 if (sc->sc_carpdev == NULL) { 1413 sc->sc_if.if_flags &= ~IFF_RUNNING; 1414 carp_set_state(sc, INIT); 1415 return; 1416 } 1417 1418 if (sc->sc_if.if_flags & IFF_UP && sc->sc_vhid > 0 && 1419 (sc->sc_naddrs || sc->sc_naddrs6) && !sc->sc_suppress) { 1420 sc->sc_if.if_flags |= IFF_RUNNING; 1421 } else { 1422 sc->sc_if.if_flags &= ~IFF_RUNNING; 1423 carp_setroute(sc, RTM_DELETE); 1424 return; 1425 } 1426 1427 switch (sc->sc_state) { 1428 case INIT: 1429 carp_set_state(sc, BACKUP); 1430 carp_setroute(sc, RTM_DELETE); 1431 carp_setrun(sc, 0); 1432 break; 1433 case BACKUP: 1434 callout_stop(&sc->sc_ad_tmo); 1435 tv.tv_sec = 3 * sc->sc_advbase; 1436 tv.tv_usec = sc->sc_advskew * 1000000 / 256; 1437 switch (af) { 1438 #ifdef INET 1439 case AF_INET: 1440 callout_schedule(&sc->sc_md_tmo, tvtohz(&tv)); 1441 break; 1442 #endif /* INET */ 1443 #ifdef INET6 1444 case AF_INET6: 1445 callout_schedule(&sc->sc_md6_tmo, tvtohz(&tv)); 1446 break; 1447 #endif /* INET6 */ 1448 default: 1449 if (sc->sc_naddrs) 1450 callout_schedule(&sc->sc_md_tmo, tvtohz(&tv)); 1451 if (sc->sc_naddrs6) 1452 callout_schedule(&sc->sc_md6_tmo, tvtohz(&tv)); 1453 break; 1454 } 1455 break; 1456 case MASTER: 1457 tv.tv_sec = sc->sc_advbase; 1458 tv.tv_usec = sc->sc_advskew * 1000000 / 256; 1459 callout_schedule(&sc->sc_ad_tmo, tvtohz(&tv)); 1460 break; 1461 } 1462 } 1463 1464 void 1465 carp_multicast_cleanup(struct carp_softc *sc) 1466 { 1467 struct ip_moptions *imo = &sc->sc_imo; 1468 #ifdef INET6 1469 struct ip6_moptions *im6o = &sc->sc_im6o; 1470 #endif 1471 u_int16_t n = imo->imo_num_memberships; 1472 1473 /* Clean up our own multicast memberships */ 1474 while (n-- > 0) { 1475 if (imo->imo_membership[n] != NULL) { 1476 in_delmulti(imo->imo_membership[n]); 1477 imo->imo_membership[n] = NULL; 1478 } 1479 } 1480 imo->imo_num_memberships = 0; 1481 imo->imo_multicast_ifp = NULL; 1482 1483 #ifdef INET6 1484 while (!LIST_EMPTY(&im6o->im6o_memberships)) { 1485 struct in6_multi_mship *imm = 1486 LIST_FIRST(&im6o->im6o_memberships); 1487 1488 LIST_REMOVE(imm, i6mm_chain); 1489 in6_leavegroup(imm); 1490 } 1491 im6o->im6o_multicast_ifp = NULL; 1492 #endif 1493 1494 /* And any other multicast memberships */ 1495 carp_ether_purgemulti(sc); 1496 } 1497 1498 int 1499 carp_set_ifp(struct carp_softc *sc, struct ifnet *ifp) 1500 { 1501 struct carp_if *cif, *ncif = NULL; 1502 struct carp_softc *vr, *after = NULL; 1503 int myself = 0, error = 0; 1504 int s; 1505 1506 if (ifp == sc->sc_carpdev) 1507 return (0); 1508 1509 if (ifp != NULL) { 1510 if ((ifp->if_flags & IFF_MULTICAST) == 0) 1511 return (EADDRNOTAVAIL); 1512 1513 if (ifp->if_type == IFT_CARP) 1514 return (EINVAL); 1515 1516 if (ifp->if_carp == NULL) { 1517 ncif = malloc(sizeof(*cif), M_IFADDR, M_NOWAIT); 1518 if (ncif == NULL) 1519 return (ENOBUFS); 1520 if ((error = ifpromisc(ifp, 1))) { 1521 free(ncif, M_IFADDR); 1522 return (error); 1523 } 1524 1525 ncif->vhif_ifp = ifp; 1526 TAILQ_INIT(&ncif->vhif_vrs); 1527 } else { 1528 cif = (struct carp_if *)ifp->if_carp; 1529 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) 1530 if (vr != sc && vr->sc_vhid == sc->sc_vhid) 1531 return (EINVAL); 1532 } 1533 1534 /* detach from old interface */ 1535 if (sc->sc_carpdev != NULL) 1536 carpdetach(sc); 1537 1538 /* join multicast groups */ 1539 if (sc->sc_naddrs < 0 && 1540 (error = carp_join_multicast(sc)) != 0) { 1541 if (ncif != NULL) 1542 free(ncif, M_IFADDR); 1543 return (error); 1544 } 1545 1546 #ifdef INET6 1547 if (sc->sc_naddrs6 < 0 && 1548 (error = carp_join_multicast6(sc)) != 0) { 1549 if (ncif != NULL) 1550 free(ncif, M_IFADDR); 1551 carp_multicast_cleanup(sc); 1552 return (error); 1553 } 1554 #endif 1555 1556 /* attach carp interface to physical interface */ 1557 if (ncif != NULL) 1558 ifp->if_carp = (void *)ncif; 1559 sc->sc_carpdev = ifp; 1560 cif = (struct carp_if *)ifp->if_carp; 1561 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) { 1562 if (vr == sc) 1563 myself = 1; 1564 if (vr->sc_vhid < sc->sc_vhid) 1565 after = vr; 1566 } 1567 1568 if (!myself) { 1569 /* We're trying to keep things in order */ 1570 if (after == NULL) { 1571 TAILQ_INSERT_TAIL(&cif->vhif_vrs, sc, sc_list); 1572 } else { 1573 TAILQ_INSERT_AFTER(&cif->vhif_vrs, after, 1574 sc, sc_list); 1575 } 1576 cif->vhif_nvrs++; 1577 } 1578 if (sc->sc_naddrs || sc->sc_naddrs6) 1579 sc->sc_if.if_flags |= IFF_UP; 1580 carp_set_enaddr(sc); 1581 s = splnet(); 1582 /* XXX linkstatehooks establish */ 1583 carp_carpdev_state(ifp); 1584 splx(s); 1585 } else { 1586 carpdetach(sc); 1587 sc->sc_if.if_flags &= ~(IFF_UP|IFF_RUNNING); 1588 } 1589 return (0); 1590 } 1591 1592 void 1593 carp_set_enaddr(struct carp_softc *sc) 1594 { 1595 uint8_t enaddr[ETHER_ADDR_LEN]; 1596 if (sc->sc_carpdev && sc->sc_carpdev->if_type == IFT_ISO88025) { 1597 enaddr[0] = 3; 1598 enaddr[1] = 0; 1599 enaddr[2] = 0x40 >> (sc->sc_vhid - 1); 1600 enaddr[3] = 0x40000 >> (sc->sc_vhid - 1); 1601 enaddr[4] = 0; 1602 enaddr[5] = 0; 1603 } else { 1604 enaddr[0] = 0; 1605 enaddr[1] = 0; 1606 enaddr[2] = 0x5e; 1607 enaddr[3] = 0; 1608 enaddr[4] = 1; 1609 enaddr[5] = sc->sc_vhid; 1610 } 1611 if_set_sadl(&sc->sc_if, enaddr, sizeof(enaddr), false); 1612 } 1613 1614 void 1615 carp_addr_updated(void *v) 1616 { 1617 struct carp_softc *sc = (struct carp_softc *) v; 1618 struct ifaddr *ifa; 1619 int new_naddrs = 0, new_naddrs6 = 0; 1620 1621 IFADDR_FOREACH(ifa, &sc->sc_if) { 1622 if (ifa->ifa_addr->sa_family == AF_INET) 1623 new_naddrs++; 1624 else if (ifa->ifa_addr->sa_family == AF_INET6) 1625 new_naddrs6++; 1626 } 1627 1628 /* Handle a callback after SIOCDIFADDR */ 1629 if (new_naddrs < sc->sc_naddrs || new_naddrs6 < sc->sc_naddrs6) { 1630 struct in_addr mc_addr; 1631 struct in_multi *inm; 1632 1633 sc->sc_naddrs = new_naddrs; 1634 sc->sc_naddrs6 = new_naddrs6; 1635 1636 /* Re-establish multicast membership removed by in_control */ 1637 mc_addr.s_addr = INADDR_CARP_GROUP; 1638 IN_LOOKUP_MULTI(mc_addr, &sc->sc_if, inm); 1639 if (inm == NULL) { 1640 bzero(&sc->sc_imo, sizeof(sc->sc_imo)); 1641 1642 if (sc->sc_carpdev != NULL && sc->sc_naddrs > 0) 1643 carp_join_multicast(sc); 1644 } 1645 1646 if (sc->sc_naddrs == 0 && sc->sc_naddrs6 == 0) { 1647 sc->sc_if.if_flags &= ~IFF_UP; 1648 carp_set_state(sc, INIT); 1649 } else 1650 carp_hmac_prepare(sc); 1651 } 1652 1653 carp_setrun(sc, 0); 1654 } 1655 1656 int 1657 carp_set_addr(struct carp_softc *sc, struct sockaddr_in *sin) 1658 { 1659 struct ifnet *ifp = sc->sc_carpdev; 1660 struct in_ifaddr *ia, *ia_if; 1661 int error = 0; 1662 1663 if (sin->sin_addr.s_addr == 0) { 1664 if (!(sc->sc_if.if_flags & IFF_UP)) 1665 carp_set_state(sc, INIT); 1666 if (sc->sc_naddrs) 1667 sc->sc_if.if_flags |= IFF_UP; 1668 carp_setrun(sc, 0); 1669 return (0); 1670 } 1671 1672 /* we have to do this by hand to ensure we don't match on ourselves */ 1673 ia_if = NULL; 1674 for (ia = TAILQ_FIRST(&in_ifaddrhead); ia; 1675 ia = TAILQ_NEXT(ia, ia_list)) { 1676 1677 /* and, yeah, we need a multicast-capable iface too */ 1678 if (ia->ia_ifp != &sc->sc_if && 1679 ia->ia_ifp->if_type != IFT_CARP && 1680 (ia->ia_ifp->if_flags & IFF_MULTICAST) && 1681 (sin->sin_addr.s_addr & ia->ia_subnetmask) == 1682 ia->ia_subnet) { 1683 if (!ia_if) 1684 ia_if = ia; 1685 } 1686 } 1687 1688 if (ia_if) { 1689 ia = ia_if; 1690 if (ifp) { 1691 if (ifp != ia->ia_ifp) 1692 return (EADDRNOTAVAIL); 1693 } else { 1694 ifp = ia->ia_ifp; 1695 } 1696 } 1697 1698 if ((error = carp_set_ifp(sc, ifp))) 1699 return (error); 1700 1701 if (sc->sc_carpdev == NULL) 1702 return (EADDRNOTAVAIL); 1703 1704 if (sc->sc_naddrs == 0 && (error = carp_join_multicast(sc)) != 0) 1705 return (error); 1706 1707 sc->sc_naddrs++; 1708 if (sc->sc_carpdev != NULL) 1709 sc->sc_if.if_flags |= IFF_UP; 1710 1711 carp_set_state(sc, INIT); 1712 carp_setrun(sc, 0); 1713 1714 /* 1715 * Hook if_addrhooks so that we get a callback after in_ifinit has run, 1716 * to correct any inappropriate routes that it inserted. 1717 */ 1718 if (sc->ah_cookie == 0) { 1719 /* XXX link address hook */ 1720 } 1721 1722 return (0); 1723 } 1724 1725 int 1726 carp_join_multicast(struct carp_softc *sc) 1727 { 1728 struct ip_moptions *imo = &sc->sc_imo, tmpimo; 1729 struct in_addr addr; 1730 1731 bzero(&tmpimo, sizeof(tmpimo)); 1732 addr.s_addr = INADDR_CARP_GROUP; 1733 if ((tmpimo.imo_membership[0] = 1734 in_addmulti(&addr, &sc->sc_if)) == NULL) { 1735 return (ENOBUFS); 1736 } 1737 1738 imo->imo_membership[0] = tmpimo.imo_membership[0]; 1739 imo->imo_num_memberships = 1; 1740 imo->imo_multicast_ifp = &sc->sc_if; 1741 imo->imo_multicast_ttl = CARP_DFLTTL; 1742 imo->imo_multicast_loop = 0; 1743 return (0); 1744 } 1745 1746 1747 #ifdef INET6 1748 int 1749 carp_set_addr6(struct carp_softc *sc, struct sockaddr_in6 *sin6) 1750 { 1751 struct ifnet *ifp = sc->sc_carpdev; 1752 struct in6_ifaddr *ia, *ia_if; 1753 int error = 0; 1754 1755 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) { 1756 if (!(sc->sc_if.if_flags & IFF_UP)) 1757 carp_set_state(sc, INIT); 1758 if (sc->sc_naddrs6) 1759 sc->sc_if.if_flags |= IFF_UP; 1760 carp_setrun(sc, 0); 1761 return (0); 1762 } 1763 1764 /* we have to do this by hand to ensure we don't match on ourselves */ 1765 ia_if = NULL; 1766 for (ia = in6_ifaddr; ia; ia = ia->ia_next) { 1767 int i; 1768 1769 for (i = 0; i < 4; i++) { 1770 if ((sin6->sin6_addr.s6_addr32[i] & 1771 ia->ia_prefixmask.sin6_addr.s6_addr32[i]) != 1772 (ia->ia_addr.sin6_addr.s6_addr32[i] & 1773 ia->ia_prefixmask.sin6_addr.s6_addr32[i])) 1774 break; 1775 } 1776 /* and, yeah, we need a multicast-capable iface too */ 1777 if (ia->ia_ifp != &sc->sc_if && 1778 ia->ia_ifp->if_type != IFT_CARP && 1779 (ia->ia_ifp->if_flags & IFF_MULTICAST) && 1780 (i == 4)) { 1781 if (!ia_if) 1782 ia_if = ia; 1783 } 1784 } 1785 1786 if (ia_if) { 1787 ia = ia_if; 1788 if (sc->sc_carpdev) { 1789 if (sc->sc_carpdev != ia->ia_ifp) 1790 return (EADDRNOTAVAIL); 1791 } else { 1792 ifp = ia->ia_ifp; 1793 } 1794 } 1795 1796 if ((error = carp_set_ifp(sc, ifp))) 1797 return (error); 1798 1799 if (sc->sc_carpdev == NULL) 1800 return (EADDRNOTAVAIL); 1801 1802 if (sc->sc_naddrs6 == 0 && (error = carp_join_multicast6(sc)) != 0) 1803 return (error); 1804 1805 sc->sc_naddrs6++; 1806 if (sc->sc_carpdev != NULL) 1807 sc->sc_if.if_flags |= IFF_UP; 1808 carp_set_state(sc, INIT); 1809 carp_setrun(sc, 0); 1810 1811 return (0); 1812 } 1813 1814 int 1815 carp_join_multicast6(struct carp_softc *sc) 1816 { 1817 struct in6_multi_mship *imm, *imm2; 1818 struct ip6_moptions *im6o = &sc->sc_im6o; 1819 struct sockaddr_in6 addr6; 1820 int error; 1821 1822 /* Join IPv6 CARP multicast group */ 1823 bzero(&addr6, sizeof(addr6)); 1824 addr6.sin6_family = AF_INET6; 1825 addr6.sin6_len = sizeof(addr6); 1826 addr6.sin6_addr.s6_addr16[0] = htons(0xff02); 1827 addr6.sin6_addr.s6_addr16[1] = htons(sc->sc_if.if_index); 1828 addr6.sin6_addr.s6_addr8[15] = 0x12; 1829 if ((imm = in6_joingroup(&sc->sc_if, 1830 &addr6.sin6_addr, &error, 0)) == NULL) { 1831 return (error); 1832 } 1833 /* join solicited multicast address */ 1834 bzero(&addr6.sin6_addr, sizeof(addr6.sin6_addr)); 1835 addr6.sin6_addr.s6_addr16[0] = htons(0xff02); 1836 addr6.sin6_addr.s6_addr16[1] = htons(sc->sc_if.if_index); 1837 addr6.sin6_addr.s6_addr32[1] = 0; 1838 addr6.sin6_addr.s6_addr32[2] = htonl(1); 1839 addr6.sin6_addr.s6_addr32[3] = 0; 1840 addr6.sin6_addr.s6_addr8[12] = 0xff; 1841 if ((imm2 = in6_joingroup(&sc->sc_if, 1842 &addr6.sin6_addr, &error, 0)) == NULL) { 1843 in6_leavegroup(imm); 1844 return (error); 1845 } 1846 1847 /* apply v6 multicast membership */ 1848 im6o->im6o_multicast_ifp = &sc->sc_if; 1849 if (imm) 1850 LIST_INSERT_HEAD(&im6o->im6o_memberships, imm, 1851 i6mm_chain); 1852 if (imm2) 1853 LIST_INSERT_HEAD(&im6o->im6o_memberships, imm2, 1854 i6mm_chain); 1855 1856 return (0); 1857 } 1858 1859 #endif /* INET6 */ 1860 1861 int 1862 carp_ioctl(struct ifnet *ifp, u_long cmd, void *data) 1863 { 1864 struct lwp *l = curlwp; /* XXX */ 1865 struct carp_softc *sc = ifp->if_softc, *vr; 1866 struct carpreq carpr; 1867 struct ifaddr *ifa; 1868 struct ifreq *ifr; 1869 struct ifnet *cdev = NULL; 1870 int error = 0; 1871 1872 ifa = (struct ifaddr *)data; 1873 ifr = (struct ifreq *)data; 1874 1875 switch (cmd) { 1876 case SIOCINITIFADDR: 1877 switch (ifa->ifa_addr->sa_family) { 1878 #ifdef INET 1879 case AF_INET: 1880 sc->sc_if.if_flags |= IFF_UP; 1881 bcopy(ifa->ifa_addr, ifa->ifa_dstaddr, 1882 sizeof(struct sockaddr)); 1883 error = carp_set_addr(sc, satosin(ifa->ifa_addr)); 1884 break; 1885 #endif /* INET */ 1886 #ifdef INET6 1887 case AF_INET6: 1888 sc->sc_if.if_flags|= IFF_UP; 1889 error = carp_set_addr6(sc, satosin6(ifa->ifa_addr)); 1890 break; 1891 #endif /* INET6 */ 1892 default: 1893 error = EAFNOSUPPORT; 1894 break; 1895 } 1896 break; 1897 1898 case SIOCSIFFLAGS: 1899 if ((error = ifioctl_common(ifp, cmd, data)) != 0) 1900 break; 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 = ether_ioctl(ifp, cmd, data); 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 mc = malloc(sizeof(struct carp_mc_entry), M_DEVBUF, M_NOWAIT); 2135 if (mc == NULL) { 2136 error = ENOMEM; 2137 goto alloc_failed; 2138 } 2139 2140 /* 2141 * As ether_addmulti() returns ENETRESET, following two 2142 * statement shouldn't fail. 2143 */ 2144 (void)ether_multiaddr(sa, addrlo, addrhi); 2145 ETHER_LOOKUP_MULTI(addrlo, addrhi, &sc->sc_ac, mc->mc_enm); 2146 memcpy(&mc->mc_addr, sa, sa->sa_len); 2147 LIST_INSERT_HEAD(&sc->carp_mc_listhead, mc, mc_entries); 2148 2149 error = (*ifp->if_ioctl)(ifp, SIOCADDMULTI, ifr); 2150 if (error != 0) 2151 goto ioctl_failed; 2152 2153 return (error); 2154 2155 ioctl_failed: 2156 LIST_REMOVE(mc, mc_entries); 2157 free(mc, M_DEVBUF); 2158 alloc_failed: 2159 (void)ether_delmulti(sa, &sc->sc_ac); 2160 2161 return (error); 2162 } 2163 2164 int 2165 carp_ether_delmulti(struct carp_softc *sc, struct ifreq *ifr) 2166 { 2167 const struct sockaddr *sa = ifreq_getaddr(SIOCDELMULTI, ifr); 2168 struct ifnet *ifp; 2169 struct ether_multi *enm; 2170 struct carp_mc_entry *mc; 2171 u_int8_t addrlo[ETHER_ADDR_LEN], addrhi[ETHER_ADDR_LEN]; 2172 int error; 2173 2174 ifp = sc->sc_carpdev; 2175 if (ifp == NULL) 2176 return (EINVAL); 2177 2178 /* 2179 * Find a key to lookup carp_mc_entry. We have to do this 2180 * before calling ether_delmulti for obvious reason. 2181 */ 2182 if ((error = ether_multiaddr(sa, addrlo, addrhi)) != 0) 2183 return (error); 2184 ETHER_LOOKUP_MULTI(addrlo, addrhi, &sc->sc_ac, enm); 2185 if (enm == NULL) 2186 return (EINVAL); 2187 2188 LIST_FOREACH(mc, &sc->carp_mc_listhead, mc_entries) 2189 if (mc->mc_enm == enm) 2190 break; 2191 2192 /* We won't delete entries we didn't add */ 2193 if (mc == NULL) 2194 return (EINVAL); 2195 2196 error = ether_delmulti(sa, &sc->sc_ac); 2197 if (error != ENETRESET) 2198 return (error); 2199 2200 /* We no longer use this multicast address. Tell parent so. */ 2201 error = (*ifp->if_ioctl)(ifp, SIOCDELMULTI, ifr); 2202 if (error == 0) { 2203 /* And forget about this address. */ 2204 LIST_REMOVE(mc, mc_entries); 2205 free(mc, M_DEVBUF); 2206 } else 2207 (void)ether_addmulti(sa, &sc->sc_ac); 2208 return (error); 2209 } 2210 2211 /* 2212 * Delete any multicast address we have asked to add from parent 2213 * interface. Called when the carp is being unconfigured. 2214 */ 2215 void 2216 carp_ether_purgemulti(struct carp_softc *sc) 2217 { 2218 struct ifnet *ifp = sc->sc_carpdev; /* Parent. */ 2219 struct carp_mc_entry *mc; 2220 union { 2221 struct ifreq ifreq; 2222 struct { 2223 char ifr_name[IFNAMSIZ]; 2224 struct sockaddr_storage ifr_ss; 2225 } ifreq_storage; 2226 } u; 2227 struct ifreq *ifr = &u.ifreq; 2228 2229 if (ifp == NULL) 2230 return; 2231 2232 memcpy(ifr->ifr_name, ifp->if_xname, IFNAMSIZ); 2233 while ((mc = LIST_FIRST(&sc->carp_mc_listhead)) != NULL) { 2234 memcpy(&ifr->ifr_addr, &mc->mc_addr, mc->mc_addr.ss_len); 2235 (void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, ifr); 2236 LIST_REMOVE(mc, mc_entries); 2237 free(mc, M_DEVBUF); 2238 } 2239 } 2240 2241 static int 2242 sysctl_net_inet_carp_stats(SYSCTLFN_ARGS) 2243 { 2244 2245 return (NETSTAT_SYSCTL(carpstat_percpu, CARP_NSTATS)); 2246 } 2247 2248 SYSCTL_SETUP(sysctl_net_inet_carp_setup, "sysctl net.inet.carp subtree setup") 2249 { 2250 2251 sysctl_createv(clog, 0, NULL, NULL, 2252 CTLFLAG_PERMANENT, 2253 CTLTYPE_NODE, "net", NULL, 2254 NULL, 0, NULL, 0, 2255 CTL_NET, CTL_EOL); 2256 sysctl_createv(clog, 0, NULL, NULL, 2257 CTLFLAG_PERMANENT, 2258 CTLTYPE_NODE, "inet", NULL, 2259 NULL, 0, NULL, 0, 2260 CTL_NET, PF_INET, CTL_EOL); 2261 sysctl_createv(clog, 0, NULL, NULL, 2262 CTLFLAG_PERMANENT, 2263 CTLTYPE_NODE, "carp", 2264 SYSCTL_DESCR("CARP related settings"), 2265 NULL, 0, NULL, 0, 2266 CTL_NET, PF_INET, IPPROTO_CARP, CTL_EOL); 2267 2268 sysctl_createv(clog, 0, NULL, NULL, 2269 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2270 CTLTYPE_INT, "preempt", 2271 SYSCTL_DESCR("Enable CARP Preempt"), 2272 NULL, 0, &carp_opts[CARPCTL_PREEMPT], 0, 2273 CTL_NET, PF_INET, IPPROTO_CARP, 2274 CTL_CREATE, CTL_EOL); 2275 sysctl_createv(clog, 0, NULL, NULL, 2276 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2277 CTLTYPE_INT, "arpbalance", 2278 SYSCTL_DESCR("Enable ARP balancing"), 2279 NULL, 0, &carp_opts[CARPCTL_ARPBALANCE], 0, 2280 CTL_NET, PF_INET, IPPROTO_CARP, 2281 CTL_CREATE, CTL_EOL); 2282 sysctl_createv(clog, 0, NULL, NULL, 2283 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2284 CTLTYPE_INT, "allow", 2285 SYSCTL_DESCR("Enable CARP"), 2286 NULL, 0, &carp_opts[CARPCTL_ALLOW], 0, 2287 CTL_NET, PF_INET, IPPROTO_CARP, 2288 CTL_CREATE, CTL_EOL); 2289 sysctl_createv(clog, 0, NULL, NULL, 2290 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2291 CTLTYPE_INT, "log", 2292 SYSCTL_DESCR("CARP logging"), 2293 NULL, 0, &carp_opts[CARPCTL_LOG], 0, 2294 CTL_NET, PF_INET, IPPROTO_CARP, 2295 CTL_CREATE, CTL_EOL); 2296 sysctl_createv(clog, 0, NULL, NULL, 2297 CTLFLAG_PERMANENT, 2298 CTLTYPE_STRUCT, "stats", 2299 SYSCTL_DESCR("CARP statistics"), 2300 sysctl_net_inet_carp_stats, 0, NULL, 0, 2301 CTL_NET, PF_INET, IPPROTO_CARP, CARPCTL_STATS, 2302 CTL_EOL); 2303 } 2304