ip_carp.c revision 1.24
1 /* $NetBSD: ip_carp.c,v 1.24 2008/04/15 06:03:28 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 1.24 2008/04/15 06:03:28 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 #include <sys/percpu.h> 57 58 #include <sys/cpu.h> 59 60 #include <net/if.h> 61 #include <net/pfil.h> 62 #include <net/if_types.h> 63 #include <net/if_ether.h> 64 #include <net/route.h> 65 #include <net/netisr.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) \ 163 do { \ 164 uint64_t *_carps_ = percpu_getref(carpstat_percpu); \ 165 _carps_[x]++; \ 166 percpu_putref(carpstat_percpu); \ 167 } while (/*CONSTCOND*/0) 168 169 struct carp_if { 170 TAILQ_HEAD(, carp_softc) vhif_vrs; 171 int vhif_nvrs; 172 173 struct ifnet *vhif_ifp; 174 }; 175 176 #define CARP_LOG(sc, s) \ 177 if (carp_opts[CARPCTL_LOG]) { \ 178 if (sc) \ 179 log(LOG_INFO, "%s: ", \ 180 (sc)->sc_if.if_xname); \ 181 else \ 182 log(LOG_INFO, "carp: "); \ 183 addlog s; \ 184 addlog("\n"); \ 185 } 186 187 void carp_hmac_prepare(struct carp_softc *); 188 void carp_hmac_generate(struct carp_softc *, u_int32_t *, 189 unsigned char *); 190 int carp_hmac_verify(struct carp_softc *, u_int32_t *, 191 unsigned char *); 192 void carp_setroute(struct carp_softc *, int); 193 void carp_proto_input_c(struct mbuf *, struct carp_header *, sa_family_t); 194 void carpattach(int); 195 void carpdetach(struct carp_softc *); 196 int carp_prepare_ad(struct mbuf *, struct carp_softc *, 197 struct carp_header *); 198 void carp_send_ad_all(void); 199 void carp_send_ad(void *); 200 void carp_send_arp(struct carp_softc *); 201 void carp_master_down(void *); 202 int carp_ioctl(struct ifnet *, u_long, void *); 203 void carp_start(struct ifnet *); 204 void carp_setrun(struct carp_softc *, sa_family_t); 205 void carp_set_state(struct carp_softc *, int); 206 int carp_addrcount(struct carp_if *, struct in_ifaddr *, int); 207 enum { CARP_COUNT_MASTER, CARP_COUNT_RUNNING }; 208 209 void carp_multicast_cleanup(struct carp_softc *); 210 int carp_set_ifp(struct carp_softc *, struct ifnet *); 211 void carp_set_enaddr(struct carp_softc *); 212 void carp_addr_updated(void *); 213 u_int32_t carp_hash(struct carp_softc *, u_char *); 214 int carp_set_addr(struct carp_softc *, struct sockaddr_in *); 215 int carp_join_multicast(struct carp_softc *); 216 #ifdef INET6 217 void carp_send_na(struct carp_softc *); 218 int carp_set_addr6(struct carp_softc *, struct sockaddr_in6 *); 219 int carp_join_multicast6(struct carp_softc *); 220 #endif 221 int carp_clone_create(struct if_clone *, int); 222 int carp_clone_destroy(struct ifnet *); 223 int carp_ether_addmulti(struct carp_softc *, struct ifreq *); 224 int carp_ether_delmulti(struct carp_softc *, struct ifreq *); 225 void carp_ether_purgemulti(struct carp_softc *); 226 227 struct if_clone carp_cloner = 228 IF_CLONE_INITIALIZER("carp", carp_clone_create, carp_clone_destroy); 229 230 static __inline u_int16_t 231 carp_cksum(struct mbuf *m, int len) 232 { 233 return (in_cksum(m, len)); 234 } 235 236 void 237 carp_hmac_prepare(struct carp_softc *sc) 238 { 239 u_int8_t carp_version = CARP_VERSION, type = CARP_ADVERTISEMENT; 240 u_int8_t vhid = sc->sc_vhid & 0xff; 241 SHA1_CTX sha1ctx; 242 u_int32_t kmd[5]; 243 struct ifaddr *ifa; 244 int i, found; 245 struct in_addr last, cur, in; 246 #ifdef INET6 247 struct in6_addr last6, cur6, in6; 248 #endif /* INET6 */ 249 250 /* compute ipad from key */ 251 bzero(sc->sc_pad, sizeof(sc->sc_pad)); 252 bcopy(sc->sc_key, sc->sc_pad, sizeof(sc->sc_key)); 253 for (i = 0; i < sizeof(sc->sc_pad); i++) 254 sc->sc_pad[i] ^= 0x36; 255 256 /* precompute first part of inner hash */ 257 SHA1Init(&sc->sc_sha1); 258 SHA1Update(&sc->sc_sha1, sc->sc_pad, sizeof(sc->sc_pad)); 259 SHA1Update(&sc->sc_sha1, (void *)&carp_version, sizeof(carp_version)); 260 SHA1Update(&sc->sc_sha1, (void *)&type, sizeof(type)); 261 262 /* generate a key for the arpbalance hash, before the vhid is hashed */ 263 bcopy(&sc->sc_sha1, &sha1ctx, sizeof(sha1ctx)); 264 SHA1Final((unsigned char *)kmd, &sha1ctx); 265 sc->sc_hashkey[0] = kmd[0] ^ kmd[1]; 266 sc->sc_hashkey[1] = kmd[2] ^ kmd[3]; 267 268 /* the rest of the precomputation */ 269 SHA1Update(&sc->sc_sha1, (void *)&vhid, sizeof(vhid)); 270 271 /* Hash the addresses from smallest to largest, not interface order */ 272 #ifdef INET 273 cur.s_addr = 0; 274 do { 275 found = 0; 276 last = cur; 277 cur.s_addr = 0xffffffff; 278 IFADDR_FOREACH(ifa, &sc->sc_if) { 279 in.s_addr = ifatoia(ifa)->ia_addr.sin_addr.s_addr; 280 if (ifa->ifa_addr->sa_family == AF_INET && 281 ntohl(in.s_addr) > ntohl(last.s_addr) && 282 ntohl(in.s_addr) < ntohl(cur.s_addr)) { 283 cur.s_addr = in.s_addr; 284 found++; 285 } 286 } 287 if (found) 288 SHA1Update(&sc->sc_sha1, (void *)&cur, sizeof(cur)); 289 } while (found); 290 #endif /* INET */ 291 292 #ifdef INET6 293 memset(&cur6, 0x00, sizeof(cur6)); 294 do { 295 found = 0; 296 last6 = cur6; 297 memset(&cur6, 0xff, sizeof(cur6)); 298 IFADDR_FOREACH(ifa, &sc->sc_if) { 299 in6 = ifatoia6(ifa)->ia_addr.sin6_addr; 300 if (IN6_IS_ADDR_LINKLOCAL(&in6)) 301 in6.s6_addr16[1] = 0; 302 if (ifa->ifa_addr->sa_family == AF_INET6 && 303 memcmp(&in6, &last6, sizeof(in6)) > 0 && 304 memcmp(&in6, &cur6, sizeof(in6)) < 0) { 305 cur6 = in6; 306 found++; 307 } 308 } 309 if (found) 310 SHA1Update(&sc->sc_sha1, (void *)&cur6, sizeof(cur6)); 311 } while (found); 312 #endif /* INET6 */ 313 314 /* convert ipad to opad */ 315 for (i = 0; i < sizeof(sc->sc_pad); i++) 316 sc->sc_pad[i] ^= 0x36 ^ 0x5c; 317 } 318 319 void 320 carp_hmac_generate(struct carp_softc *sc, u_int32_t counter[2], 321 unsigned char md[20]) 322 { 323 SHA1_CTX sha1ctx; 324 325 /* fetch first half of inner hash */ 326 bcopy(&sc->sc_sha1, &sha1ctx, sizeof(sha1ctx)); 327 328 SHA1Update(&sha1ctx, (void *)counter, sizeof(sc->sc_counter)); 329 SHA1Final(md, &sha1ctx); 330 331 /* outer hash */ 332 SHA1Init(&sha1ctx); 333 SHA1Update(&sha1ctx, sc->sc_pad, sizeof(sc->sc_pad)); 334 SHA1Update(&sha1ctx, md, 20); 335 SHA1Final(md, &sha1ctx); 336 } 337 338 int 339 carp_hmac_verify(struct carp_softc *sc, u_int32_t counter[2], 340 unsigned char md[20]) 341 { 342 unsigned char md2[20]; 343 344 carp_hmac_generate(sc, counter, md2); 345 346 return (bcmp(md, md2, sizeof(md2))); 347 } 348 349 void 350 carp_setroute(struct carp_softc *sc, int cmd) 351 { 352 struct ifaddr *ifa; 353 int s; 354 355 s = splsoftnet(); 356 IFADDR_FOREACH(ifa, &sc->sc_if) { 357 switch (ifa->ifa_addr->sa_family) { 358 case AF_INET: { 359 int count = 0; 360 struct rtentry *rt; 361 int hr_otherif, nr_ourif; 362 363 /* 364 * Avoid screwing with the routes if there are other 365 * carp interfaces which are master and have the same 366 * address. 367 */ 368 if (sc->sc_carpdev != NULL && 369 sc->sc_carpdev->if_carp != NULL) { 370 count = carp_addrcount( 371 (struct carp_if *)sc->sc_carpdev->if_carp, 372 ifatoia(ifa), CARP_COUNT_MASTER); 373 if ((cmd == RTM_ADD && count != 1) || 374 (cmd == RTM_DELETE && count != 0)) 375 continue; 376 } 377 378 /* Remove the existing host route, if any */ 379 rtrequest(RTM_DELETE, ifa->ifa_addr, 380 ifa->ifa_addr, ifa->ifa_netmask, 381 RTF_HOST, NULL); 382 383 rt = NULL; 384 (void)rtrequest(RTM_GET, ifa->ifa_addr, ifa->ifa_addr, 385 ifa->ifa_netmask, RTF_HOST, &rt); 386 hr_otherif = (rt && rt->rt_ifp != &sc->sc_if && 387 rt->rt_flags & (RTF_CLONING|RTF_CLONED)); 388 if (rt != NULL) { 389 RTFREE(rt); 390 rt = NULL; 391 } 392 393 /* Check for a network route on our interface */ 394 395 rt = NULL; 396 (void)rtrequest(RTM_GET, ifa->ifa_addr, ifa->ifa_addr, 397 ifa->ifa_netmask, 0, &rt); 398 nr_ourif = (rt && rt->rt_ifp == &sc->sc_if); 399 400 switch (cmd) { 401 case RTM_ADD: 402 if (hr_otherif) { 403 ifa->ifa_rtrequest = NULL; 404 ifa->ifa_flags &= ~RTF_CLONING; 405 406 rtrequest(RTM_ADD, ifa->ifa_addr, 407 ifa->ifa_addr, ifa->ifa_netmask, 408 RTF_UP | RTF_HOST, NULL); 409 } 410 if (!hr_otherif || nr_ourif || !rt) { 411 if (nr_ourif && !(rt->rt_flags & 412 RTF_CLONING)) 413 rtrequest(RTM_DELETE, 414 ifa->ifa_addr, 415 ifa->ifa_addr, 416 ifa->ifa_netmask, 0, NULL); 417 418 ifa->ifa_rtrequest = arp_rtrequest; 419 ifa->ifa_flags |= RTF_CLONING; 420 421 if (rtrequest(RTM_ADD, ifa->ifa_addr, 422 ifa->ifa_addr, ifa->ifa_netmask, 0, 423 NULL) == 0) 424 ifa->ifa_flags |= IFA_ROUTE; 425 } 426 break; 427 case RTM_DELETE: 428 break; 429 default: 430 break; 431 } 432 if (rt != NULL) { 433 RTFREE(rt); 434 rt = NULL; 435 } 436 break; 437 } 438 439 #ifdef INET6 440 case AF_INET6: 441 if (cmd == RTM_ADD) 442 in6_ifaddloop(ifa); 443 else 444 in6_ifremloop(ifa); 445 break; 446 #endif /* INET6 */ 447 default: 448 break; 449 } 450 } 451 splx(s); 452 } 453 454 /* 455 * process input packet. 456 * we have rearranged checks order compared to the rfc, 457 * but it seems more efficient this way or not possible otherwise. 458 */ 459 void 460 carp_proto_input(struct mbuf *m, ...) 461 { 462 struct ip *ip = mtod(m, struct ip *); 463 struct carp_softc *sc = NULL; 464 struct carp_header *ch; 465 int iplen, len, hlen; 466 va_list ap; 467 468 va_start(ap, m); 469 hlen = va_arg(ap, int); 470 va_end(ap); 471 472 CARP_STATINC(CARP_STAT_IPACKETS); 473 474 if (!carp_opts[CARPCTL_ALLOW]) { 475 m_freem(m); 476 return; 477 } 478 479 /* check if received on a valid carp interface */ 480 if (m->m_pkthdr.rcvif->if_type != IFT_CARP) { 481 CARP_STATINC(CARP_STAT_BADIF); 482 CARP_LOG(sc, ("packet received on non-carp interface: %s", 483 m->m_pkthdr.rcvif->if_xname)); 484 m_freem(m); 485 return; 486 } 487 488 /* verify that the IP TTL is 255. */ 489 if (ip->ip_ttl != CARP_DFLTTL) { 490 CARP_STATINC(CARP_STAT_BADTTL); 491 CARP_LOG(sc, ("received ttl %d != %d on %s", ip->ip_ttl, 492 CARP_DFLTTL, m->m_pkthdr.rcvif->if_xname)); 493 m_freem(m); 494 return; 495 } 496 497 /* 498 * verify that the received packet length is 499 * equal to the CARP header 500 */ 501 iplen = ip->ip_hl << 2; 502 len = iplen + sizeof(*ch); 503 if (len > m->m_pkthdr.len) { 504 CARP_STATINC(CARP_STAT_BADLEN); 505 CARP_LOG(sc, ("packet too short %d on %s", m->m_pkthdr.len, 506 m->m_pkthdr.rcvif->if_xname)); 507 m_freem(m); 508 return; 509 } 510 511 if ((m = m_pullup(m, len)) == NULL) { 512 CARP_STATINC(CARP_STAT_HDROPS); 513 return; 514 } 515 ip = mtod(m, struct ip *); 516 ch = (struct carp_header *)((char *)ip + iplen); 517 /* verify the CARP checksum */ 518 m->m_data += iplen; 519 if (carp_cksum(m, len - iplen)) { 520 CARP_STATINC(CARP_STAT_BADSUM); 521 CARP_LOG(sc, ("checksum failed on %s", 522 m->m_pkthdr.rcvif->if_xname)); 523 m_freem(m); 524 return; 525 } 526 m->m_data -= iplen; 527 528 carp_proto_input_c(m, ch, AF_INET); 529 } 530 531 #ifdef INET6 532 int 533 carp6_proto_input(struct mbuf **mp, int *offp, int proto) 534 { 535 struct mbuf *m = *mp; 536 struct carp_softc *sc = NULL; 537 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 538 struct carp_header *ch; 539 u_int len; 540 541 CARP_STATINC(CARP_STAT_IPACKETS6); 542 543 if (!carp_opts[CARPCTL_ALLOW]) { 544 m_freem(m); 545 return (IPPROTO_DONE); 546 } 547 548 /* check if received on a valid carp interface */ 549 if (m->m_pkthdr.rcvif->if_type != IFT_CARP) { 550 CARP_STATINC(CARP_STAT_BADIF); 551 CARP_LOG(sc, ("packet received on non-carp interface: %s", 552 m->m_pkthdr.rcvif->if_xname)); 553 m_freem(m); 554 return (IPPROTO_DONE); 555 } 556 557 /* verify that the IP TTL is 255 */ 558 if (ip6->ip6_hlim != CARP_DFLTTL) { 559 CARP_STATINC(CARP_STAT_BADTTL); 560 CARP_LOG(sc, ("received ttl %d != %d on %s", ip6->ip6_hlim, 561 CARP_DFLTTL, m->m_pkthdr.rcvif->if_xname)); 562 m_freem(m); 563 return (IPPROTO_DONE); 564 } 565 566 /* verify that we have a complete carp packet */ 567 len = m->m_len; 568 IP6_EXTHDR_GET(ch, struct carp_header *, m, *offp, sizeof(*ch)); 569 if (ch == NULL) { 570 CARP_STATINC(CARP_STAT_BADLEN); 571 CARP_LOG(sc, ("packet size %u too small", len)); 572 return (IPPROTO_DONE); 573 } 574 575 576 /* verify the CARP checksum */ 577 m->m_data += *offp; 578 if (carp_cksum(m, sizeof(*ch))) { 579 CARP_STATINC(CARP_STAT_BADSUM); 580 CARP_LOG(sc, ("checksum failed, on %s", 581 m->m_pkthdr.rcvif->if_xname)); 582 m_freem(m); 583 return (IPPROTO_DONE); 584 } 585 m->m_data -= *offp; 586 587 carp_proto_input_c(m, ch, AF_INET6); 588 return (IPPROTO_DONE); 589 } 590 #endif /* INET6 */ 591 592 void 593 carp_proto_input_c(struct mbuf *m, struct carp_header *ch, sa_family_t af) 594 { 595 struct carp_softc *sc; 596 u_int64_t tmp_counter; 597 struct timeval sc_tv, ch_tv; 598 599 TAILQ_FOREACH(sc, &((struct carp_if *) 600 m->m_pkthdr.rcvif->if_carpdev->if_carp)->vhif_vrs, sc_list) 601 if (sc->sc_vhid == ch->carp_vhid) 602 break; 603 604 if (!sc || (sc->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) != 605 (IFF_UP|IFF_RUNNING)) { 606 CARP_STATINC(CARP_STAT_BADVHID); 607 m_freem(m); 608 return; 609 } 610 611 /* 612 * Check if our own advertisement was duplicated 613 * from a non simplex interface. 614 * XXX If there is no address on our physical interface 615 * there is no way to distinguish our ads from the ones 616 * another carp host might have sent us. 617 */ 618 if ((sc->sc_carpdev->if_flags & IFF_SIMPLEX) == 0) { 619 struct sockaddr sa; 620 struct ifaddr *ifa; 621 622 bzero(&sa, sizeof(sa)); 623 sa.sa_family = af; 624 ifa = ifaof_ifpforaddr(&sa, sc->sc_carpdev); 625 626 if (ifa && af == AF_INET) { 627 struct ip *ip = mtod(m, struct ip *); 628 if (ip->ip_src.s_addr == 629 ifatoia(ifa)->ia_addr.sin_addr.s_addr) { 630 m_freem(m); 631 return; 632 } 633 } 634 #ifdef INET6 635 if (ifa && af == AF_INET6) { 636 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 637 struct in6_addr in6_src, in6_found; 638 639 in6_src = ip6->ip6_src; 640 in6_found = ifatoia6(ifa)->ia_addr.sin6_addr; 641 if (IN6_IS_ADDR_LINKLOCAL(&in6_src)) 642 in6_src.s6_addr16[1] = 0; 643 if (IN6_IS_ADDR_LINKLOCAL(&in6_found)) 644 in6_found.s6_addr16[1] = 0; 645 if (IN6_ARE_ADDR_EQUAL(&in6_src, &in6_found)) { 646 m_freem(m); 647 return; 648 } 649 } 650 #endif /* INET6 */ 651 } 652 653 microtime(&sc->sc_if.if_lastchange); 654 sc->sc_if.if_ipackets++; 655 sc->sc_if.if_ibytes += m->m_pkthdr.len; 656 657 /* verify the CARP version. */ 658 if (ch->carp_version != CARP_VERSION) { 659 CARP_STATINC(CARP_STAT_BADVER); 660 sc->sc_if.if_ierrors++; 661 CARP_LOG(sc, ("invalid version %d != %d", 662 ch->carp_version, CARP_VERSION)); 663 m_freem(m); 664 return; 665 } 666 667 /* verify the hash */ 668 if (carp_hmac_verify(sc, ch->carp_counter, ch->carp_md)) { 669 CARP_STATINC(CARP_STAT_BADAUTH); 670 sc->sc_if.if_ierrors++; 671 CARP_LOG(sc, ("incorrect hash")); 672 m_freem(m); 673 return; 674 } 675 676 tmp_counter = ntohl(ch->carp_counter[0]); 677 tmp_counter = tmp_counter<<32; 678 tmp_counter += ntohl(ch->carp_counter[1]); 679 680 /* XXX Replay protection goes here */ 681 682 sc->sc_init_counter = 0; 683 sc->sc_counter = tmp_counter; 684 685 686 sc_tv.tv_sec = sc->sc_advbase; 687 if (carp_suppress_preempt && sc->sc_advskew < 240) 688 sc_tv.tv_usec = 240 * 1000000 / 256; 689 else 690 sc_tv.tv_usec = sc->sc_advskew * 1000000 / 256; 691 ch_tv.tv_sec = ch->carp_advbase; 692 ch_tv.tv_usec = ch->carp_advskew * 1000000 / 256; 693 694 switch (sc->sc_state) { 695 case INIT: 696 break; 697 case MASTER: 698 /* 699 * If we receive an advertisement from a backup who's going to 700 * be more frequent than us, go into BACKUP state. 701 */ 702 if (timercmp(&sc_tv, &ch_tv, >) || 703 timercmp(&sc_tv, &ch_tv, ==)) { 704 callout_stop(&sc->sc_ad_tmo); 705 carp_set_state(sc, BACKUP); 706 carp_setrun(sc, 0); 707 carp_setroute(sc, RTM_DELETE); 708 } 709 break; 710 case BACKUP: 711 /* 712 * If we're pre-empting masters who advertise slower than us, 713 * and this one claims to be slower, treat him as down. 714 */ 715 if (carp_opts[CARPCTL_PREEMPT] && timercmp(&sc_tv, &ch_tv, <)) { 716 carp_master_down(sc); 717 break; 718 } 719 720 /* 721 * If the master is going to advertise at such a low frequency 722 * that he's guaranteed to time out, we'd might as well just 723 * treat him as timed out now. 724 */ 725 sc_tv.tv_sec = sc->sc_advbase * 3; 726 if (timercmp(&sc_tv, &ch_tv, <)) { 727 carp_master_down(sc); 728 break; 729 } 730 731 /* 732 * Otherwise, we reset the counter and wait for the next 733 * advertisement. 734 */ 735 carp_setrun(sc, af); 736 break; 737 } 738 739 m_freem(m); 740 return; 741 } 742 743 /* 744 * Interface side of the CARP implementation. 745 */ 746 747 /* ARGSUSED */ 748 void 749 carpattach(int n) 750 { 751 if_clone_attach(&carp_cloner); 752 753 carpstat_percpu = percpu_alloc(sizeof(uint64_t) * CARP_NSTATS); 754 } 755 756 int 757 carp_clone_create(struct if_clone *ifc, int unit) 758 { 759 extern int ifqmaxlen; 760 struct carp_softc *sc; 761 struct ifnet *ifp; 762 763 sc = malloc(sizeof(*sc), M_DEVBUF, M_NOWAIT); 764 if (!sc) 765 return (ENOMEM); 766 bzero(sc, sizeof(*sc)); 767 768 sc->sc_suppress = 0; 769 sc->sc_advbase = CARP_DFLTINTV; 770 sc->sc_vhid = -1; /* required setting */ 771 sc->sc_advskew = 0; 772 sc->sc_init_counter = 1; 773 sc->sc_naddrs = sc->sc_naddrs6 = 0; 774 #ifdef INET6 775 sc->sc_im6o.im6o_multicast_hlim = CARP_DFLTTL; 776 #endif /* INET6 */ 777 778 callout_init(&sc->sc_ad_tmo, 0); 779 callout_init(&sc->sc_md_tmo, 0); 780 callout_init(&sc->sc_md6_tmo, 0); 781 782 callout_setfunc(&sc->sc_ad_tmo, carp_send_ad, sc); 783 callout_setfunc(&sc->sc_md_tmo, carp_master_down, sc); 784 callout_setfunc(&sc->sc_md6_tmo, carp_master_down, sc); 785 786 LIST_INIT(&sc->carp_mc_listhead); 787 ifp = &sc->sc_if; 788 ifp->if_softc = sc; 789 snprintf(ifp->if_xname, sizeof ifp->if_xname, "%s%d", ifc->ifc_name, 790 unit); 791 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST; 792 ifp->if_ioctl = carp_ioctl; 793 ifp->if_start = carp_start; 794 ifp->if_output = carp_output; 795 ifp->if_type = IFT_CARP; 796 ifp->if_addrlen = ETHER_ADDR_LEN; 797 ifp->if_hdrlen = ETHER_HDR_LEN; 798 ifp->if_mtu = ETHERMTU; 799 IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen); 800 IFQ_SET_READY(&ifp->if_snd); 801 if_attach(ifp); 802 803 if_alloc_sadl(ifp); 804 ifp->if_broadcastaddr = etherbroadcastaddr; 805 carp_set_enaddr(sc); 806 LIST_INIT(&sc->sc_ac.ec_multiaddrs); 807 #if NBPFILTER > 0 808 bpfattach(ifp, DLT_EN10MB, ETHER_HDR_LEN); 809 #endif 810 return (0); 811 } 812 813 int 814 carp_clone_destroy(struct ifnet *ifp) 815 { 816 struct carp_softc *sc = ifp->if_softc; 817 818 carpdetach(ifp->if_softc); 819 ether_ifdetach(ifp); 820 if_detach(ifp); 821 callout_destroy(&sc->sc_ad_tmo); 822 callout_destroy(&sc->sc_md_tmo); 823 callout_destroy(&sc->sc_md6_tmo); 824 free(ifp->if_softc, M_DEVBUF); 825 826 return (0); 827 } 828 829 void 830 carpdetach(struct carp_softc *sc) 831 { 832 struct carp_if *cif; 833 int s; 834 835 callout_stop(&sc->sc_ad_tmo); 836 callout_stop(&sc->sc_md_tmo); 837 callout_stop(&sc->sc_md6_tmo); 838 839 if (sc->sc_suppress) 840 carp_suppress_preempt--; 841 sc->sc_suppress = 0; 842 843 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) 844 carp_suppress_preempt--; 845 sc->sc_sendad_errors = 0; 846 847 carp_set_state(sc, INIT); 848 sc->sc_if.if_flags &= ~IFF_UP; 849 carp_setrun(sc, 0); 850 carp_multicast_cleanup(sc); 851 852 s = splnet(); 853 if (sc->sc_carpdev != NULL) { 854 /* XXX linkstatehook removal */ 855 cif = (struct carp_if *)sc->sc_carpdev->if_carp; 856 TAILQ_REMOVE(&cif->vhif_vrs, sc, sc_list); 857 if (!--cif->vhif_nvrs) { 858 ifpromisc(sc->sc_carpdev, 0); 859 sc->sc_carpdev->if_carp = NULL; 860 FREE(cif, M_IFADDR); 861 } 862 } 863 sc->sc_carpdev = NULL; 864 splx(s); 865 } 866 867 /* Detach an interface from the carp. */ 868 void 869 carp_ifdetach(struct ifnet *ifp) 870 { 871 struct carp_softc *sc, *nextsc; 872 struct carp_if *cif = (struct carp_if *)ifp->if_carp; 873 874 for (sc = TAILQ_FIRST(&cif->vhif_vrs); sc; sc = nextsc) { 875 nextsc = TAILQ_NEXT(sc, sc_list); 876 carpdetach(sc); 877 } 878 } 879 880 int 881 carp_prepare_ad(struct mbuf *m, struct carp_softc *sc, 882 struct carp_header *ch) 883 { 884 if (sc->sc_init_counter) { 885 /* this could also be seconds since unix epoch */ 886 sc->sc_counter = arc4random(); 887 sc->sc_counter = sc->sc_counter << 32; 888 sc->sc_counter += arc4random(); 889 } else 890 sc->sc_counter++; 891 892 ch->carp_counter[0] = htonl((sc->sc_counter>>32)&0xffffffff); 893 ch->carp_counter[1] = htonl(sc->sc_counter&0xffffffff); 894 895 carp_hmac_generate(sc, ch->carp_counter, ch->carp_md); 896 897 return (0); 898 } 899 900 void 901 carp_send_ad_all(void) 902 { 903 struct ifnet *ifp; 904 struct carp_if *cif; 905 struct carp_softc *vh; 906 907 TAILQ_FOREACH(ifp, &ifnet, if_list) { 908 if (ifp->if_carp == NULL || ifp->if_type == IFT_CARP) 909 continue; 910 911 cif = (struct carp_if *)ifp->if_carp; 912 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 913 if ((vh->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) == 914 (IFF_UP|IFF_RUNNING) && vh->sc_state == MASTER) 915 carp_send_ad(vh); 916 } 917 } 918 } 919 920 921 void 922 carp_send_ad(void *v) 923 { 924 struct carp_header ch; 925 struct timeval tv; 926 struct carp_softc *sc = v; 927 struct carp_header *ch_ptr; 928 struct mbuf *m; 929 int error, len, advbase, advskew, s; 930 struct ifaddr *ifa; 931 struct sockaddr sa; 932 933 s = splsoftnet(); 934 935 advbase = advskew = 0; /* Sssssh compiler */ 936 if (sc->sc_carpdev == NULL) { 937 sc->sc_if.if_oerrors++; 938 goto retry_later; 939 } 940 941 /* bow out if we've gone to backup (the carp interface is going down) */ 942 if (sc->sc_bow_out) { 943 sc->sc_bow_out = 0; 944 advbase = 255; 945 advskew = 255; 946 } else { 947 advbase = sc->sc_advbase; 948 if (!carp_suppress_preempt || sc->sc_advskew > 240) 949 advskew = sc->sc_advskew; 950 else 951 advskew = 240; 952 tv.tv_sec = advbase; 953 tv.tv_usec = advskew * 1000000 / 256; 954 } 955 956 ch.carp_version = CARP_VERSION; 957 ch.carp_type = CARP_ADVERTISEMENT; 958 ch.carp_vhid = sc->sc_vhid; 959 ch.carp_advbase = advbase; 960 ch.carp_advskew = advskew; 961 ch.carp_authlen = 7; /* XXX DEFINE */ 962 ch.carp_pad1 = 0; /* must be zero */ 963 ch.carp_cksum = 0; 964 965 966 #ifdef INET 967 if (sc->sc_naddrs) { 968 struct ip *ip; 969 970 MGETHDR(m, M_DONTWAIT, MT_HEADER); 971 if (m == NULL) { 972 sc->sc_if.if_oerrors++; 973 CARP_STATINC(CARP_STAT_ONOMEM); 974 /* XXX maybe less ? */ 975 goto retry_later; 976 } 977 len = sizeof(*ip) + sizeof(ch); 978 m->m_pkthdr.len = len; 979 m->m_pkthdr.rcvif = NULL; 980 m->m_len = len; 981 MH_ALIGN(m, m->m_len); 982 m->m_flags |= M_MCAST; 983 ip = mtod(m, struct ip *); 984 ip->ip_v = IPVERSION; 985 ip->ip_hl = sizeof(*ip) >> 2; 986 ip->ip_tos = IPTOS_LOWDELAY; 987 ip->ip_len = htons(len); 988 ip->ip_id = 0; /* no need for id, we don't support fragments */ 989 ip->ip_off = htons(IP_DF); 990 ip->ip_ttl = CARP_DFLTTL; 991 ip->ip_p = IPPROTO_CARP; 992 ip->ip_sum = 0; 993 994 bzero(&sa, sizeof(sa)); 995 sa.sa_family = AF_INET; 996 ifa = ifaof_ifpforaddr(&sa, sc->sc_carpdev); 997 if (ifa == NULL) 998 ip->ip_src.s_addr = 0; 999 else 1000 ip->ip_src.s_addr = 1001 ifatoia(ifa)->ia_addr.sin_addr.s_addr; 1002 ip->ip_dst.s_addr = INADDR_CARP_GROUP; 1003 1004 ch_ptr = (struct carp_header *)(&ip[1]); 1005 bcopy(&ch, ch_ptr, sizeof(ch)); 1006 if (carp_prepare_ad(m, sc, ch_ptr)) 1007 goto retry_later; 1008 1009 m->m_data += sizeof(*ip); 1010 ch_ptr->carp_cksum = carp_cksum(m, len - sizeof(*ip)); 1011 m->m_data -= sizeof(*ip); 1012 1013 microtime(&sc->sc_if.if_lastchange); 1014 sc->sc_if.if_opackets++; 1015 sc->sc_if.if_obytes += len; 1016 CARP_STATINC(CARP_STAT_OPACKETS); 1017 1018 error = ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo, 1019 NULL); 1020 if (error) { 1021 if (error == ENOBUFS) 1022 CARP_STATINC(CARP_STAT_ONOMEM); 1023 else 1024 CARP_LOG(sc, ("ip_output failed: %d", error)); 1025 sc->sc_if.if_oerrors++; 1026 if (sc->sc_sendad_errors < INT_MAX) 1027 sc->sc_sendad_errors++; 1028 if (sc->sc_sendad_errors == CARP_SENDAD_MAX_ERRORS) { 1029 carp_suppress_preempt++; 1030 if (carp_suppress_preempt == 1) 1031 carp_send_ad_all(); 1032 } 1033 sc->sc_sendad_success = 0; 1034 } else { 1035 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) { 1036 if (++sc->sc_sendad_success >= 1037 CARP_SENDAD_MIN_SUCCESS) { 1038 carp_suppress_preempt--; 1039 sc->sc_sendad_errors = 0; 1040 } 1041 } else 1042 sc->sc_sendad_errors = 0; 1043 } 1044 } 1045 #endif /* INET */ 1046 #ifdef INET6 1047 if (sc->sc_naddrs6) { 1048 struct ip6_hdr *ip6; 1049 1050 MGETHDR(m, M_DONTWAIT, MT_HEADER); 1051 if (m == NULL) { 1052 sc->sc_if.if_oerrors++; 1053 CARP_STATINC(CARP_STAT_ONOMEM); 1054 /* XXX maybe less ? */ 1055 goto retry_later; 1056 } 1057 len = sizeof(*ip6) + sizeof(ch); 1058 m->m_pkthdr.len = len; 1059 m->m_pkthdr.rcvif = NULL; 1060 m->m_len = len; 1061 MH_ALIGN(m, m->m_len); 1062 m->m_flags |= M_MCAST; 1063 ip6 = mtod(m, struct ip6_hdr *); 1064 bzero(ip6, sizeof(*ip6)); 1065 ip6->ip6_vfc |= IPV6_VERSION; 1066 ip6->ip6_hlim = CARP_DFLTTL; 1067 ip6->ip6_nxt = IPPROTO_CARP; 1068 1069 /* set the source address */ 1070 bzero(&sa, sizeof(sa)); 1071 sa.sa_family = AF_INET6; 1072 ifa = ifaof_ifpforaddr(&sa, sc->sc_carpdev); 1073 if (ifa == NULL) /* This should never happen with IPv6 */ 1074 bzero(&ip6->ip6_src, sizeof(struct in6_addr)); 1075 else 1076 bcopy(ifatoia6(ifa)->ia_addr.sin6_addr.s6_addr, 1077 &ip6->ip6_src, sizeof(struct in6_addr)); 1078 /* set the multicast destination */ 1079 1080 ip6->ip6_dst.s6_addr16[0] = htons(0xff02); 1081 ip6->ip6_dst.s6_addr8[15] = 0x12; 1082 if (in6_setscope(&ip6->ip6_dst, sc->sc_carpdev, NULL) != 0) { 1083 sc->sc_if.if_oerrors++; 1084 m_freem(m); 1085 CARP_LOG(sc, ("in6_setscope failed")); 1086 goto retry_later; 1087 } 1088 1089 ch_ptr = (struct carp_header *)(&ip6[1]); 1090 bcopy(&ch, ch_ptr, sizeof(ch)); 1091 if (carp_prepare_ad(m, sc, ch_ptr)) 1092 goto retry_later; 1093 1094 m->m_data += sizeof(*ip6); 1095 ch_ptr->carp_cksum = carp_cksum(m, len - sizeof(*ip6)); 1096 m->m_data -= sizeof(*ip6); 1097 1098 microtime(&sc->sc_if.if_lastchange); 1099 sc->sc_if.if_opackets++; 1100 sc->sc_if.if_obytes += len; 1101 CARP_STATINC(CARP_STAT_OPACKETS6); 1102 1103 error = ip6_output(m, NULL, NULL, 0, &sc->sc_im6o, NULL, NULL); 1104 if (error) { 1105 if (error == ENOBUFS) 1106 CARP_STATINC(CARP_STAT_ONOMEM); 1107 else 1108 CARP_LOG(sc, ("ip6_output failed: %d", error)); 1109 sc->sc_if.if_oerrors++; 1110 if (sc->sc_sendad_errors < INT_MAX) 1111 sc->sc_sendad_errors++; 1112 if (sc->sc_sendad_errors == CARP_SENDAD_MAX_ERRORS) { 1113 carp_suppress_preempt++; 1114 if (carp_suppress_preempt == 1) 1115 carp_send_ad_all(); 1116 } 1117 sc->sc_sendad_success = 0; 1118 } else { 1119 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) { 1120 if (++sc->sc_sendad_success >= 1121 CARP_SENDAD_MIN_SUCCESS) { 1122 carp_suppress_preempt--; 1123 sc->sc_sendad_errors = 0; 1124 } 1125 } else 1126 sc->sc_sendad_errors = 0; 1127 } 1128 } 1129 #endif /* INET6 */ 1130 1131 retry_later: 1132 splx(s); 1133 if (advbase != 255 || advskew != 255) 1134 callout_schedule(&sc->sc_ad_tmo, tvtohz(&tv)); 1135 } 1136 1137 /* 1138 * Broadcast a gratuitous ARP request containing 1139 * the virtual router MAC address for each IP address 1140 * associated with the virtual router. 1141 */ 1142 void 1143 carp_send_arp(struct carp_softc *sc) 1144 { 1145 struct ifaddr *ifa; 1146 struct in_addr *in; 1147 int s = splsoftnet(); 1148 1149 IFADDR_FOREACH(ifa, &sc->sc_if) { 1150 1151 if (ifa->ifa_addr->sa_family != AF_INET) 1152 continue; 1153 1154 in = &ifatoia(ifa)->ia_addr.sin_addr; 1155 arprequest(sc->sc_carpdev, in, in, CLLADDR(sc->sc_if.if_sadl)); 1156 DELAY(1000); /* XXX */ 1157 } 1158 splx(s); 1159 } 1160 1161 #ifdef INET6 1162 void 1163 carp_send_na(struct carp_softc *sc) 1164 { 1165 struct ifaddr *ifa; 1166 struct in6_addr *in6; 1167 static struct in6_addr mcast = IN6ADDR_LINKLOCAL_ALLNODES_INIT; 1168 int s = splsoftnet(); 1169 1170 IFADDR_FOREACH(ifa, &sc->sc_if) { 1171 1172 if (ifa->ifa_addr->sa_family != AF_INET6) 1173 continue; 1174 1175 in6 = &ifatoia6(ifa)->ia_addr.sin6_addr; 1176 nd6_na_output(sc->sc_carpdev, &mcast, in6, 1177 ND_NA_FLAG_OVERRIDE, 1, NULL); 1178 DELAY(1000); /* XXX */ 1179 } 1180 splx(s); 1181 } 1182 #endif /* INET6 */ 1183 1184 /* 1185 * Based on bridge_hash() in if_bridge.c 1186 */ 1187 #define mix(a,b,c) \ 1188 do { \ 1189 a -= b; a -= c; a ^= (c >> 13); \ 1190 b -= c; b -= a; b ^= (a << 8); \ 1191 c -= a; c -= b; c ^= (b >> 13); \ 1192 a -= b; a -= c; a ^= (c >> 12); \ 1193 b -= c; b -= a; b ^= (a << 16); \ 1194 c -= a; c -= b; c ^= (b >> 5); \ 1195 a -= b; a -= c; a ^= (c >> 3); \ 1196 b -= c; b -= a; b ^= (a << 10); \ 1197 c -= a; c -= b; c ^= (b >> 15); \ 1198 } while (0) 1199 1200 u_int32_t 1201 carp_hash(struct carp_softc *sc, u_char *src) 1202 { 1203 u_int32_t a = 0x9e3779b9, b = sc->sc_hashkey[0], c = sc->sc_hashkey[1]; 1204 1205 c += sc->sc_key[3] << 24; 1206 c += sc->sc_key[2] << 16; 1207 c += sc->sc_key[1] << 8; 1208 c += sc->sc_key[0]; 1209 b += src[5] << 8; 1210 b += src[4]; 1211 a += src[3] << 24; 1212 a += src[2] << 16; 1213 a += src[1] << 8; 1214 a += src[0]; 1215 1216 mix(a, b, c); 1217 return (c); 1218 } 1219 1220 int 1221 carp_addrcount(struct carp_if *cif, struct in_ifaddr *ia, int type) 1222 { 1223 struct carp_softc *vh; 1224 struct ifaddr *ifa; 1225 int count = 0; 1226 1227 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 1228 if ((type == CARP_COUNT_RUNNING && 1229 (vh->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) == 1230 (IFF_UP|IFF_RUNNING)) || 1231 (type == CARP_COUNT_MASTER && vh->sc_state == MASTER)) { 1232 IFADDR_FOREACH(ifa, &vh->sc_if) { 1233 if (ifa->ifa_addr->sa_family == AF_INET && 1234 ia->ia_addr.sin_addr.s_addr == 1235 ifatoia(ifa)->ia_addr.sin_addr.s_addr) 1236 count++; 1237 } 1238 } 1239 } 1240 return (count); 1241 } 1242 1243 int 1244 carp_iamatch(struct in_ifaddr *ia, u_char *src, 1245 u_int32_t *count, u_int32_t index) 1246 { 1247 struct carp_softc *sc = ia->ia_ifp->if_softc; 1248 1249 if (carp_opts[CARPCTL_ARPBALANCE]) { 1250 /* 1251 * We use the source ip to decide which virtual host should 1252 * handle the request. If we're master of that virtual host, 1253 * then we respond, otherwise, just drop the arp packet on 1254 * the floor. 1255 */ 1256 1257 /* Count the elegible carp interfaces with this address */ 1258 if (*count == 0) 1259 *count = carp_addrcount( 1260 (struct carp_if *)ia->ia_ifp->if_carpdev->if_carp, 1261 ia, CARP_COUNT_RUNNING); 1262 1263 /* This should never happen, but... */ 1264 if (*count == 0) 1265 return (0); 1266 1267 if (carp_hash(sc, src) % *count == index - 1 && 1268 sc->sc_state == MASTER) { 1269 return (1); 1270 } 1271 } else { 1272 if (sc->sc_state == MASTER) 1273 return (1); 1274 } 1275 1276 return (0); 1277 } 1278 1279 #ifdef INET6 1280 struct ifaddr * 1281 carp_iamatch6(void *v, struct in6_addr *taddr) 1282 { 1283 struct carp_if *cif = v; 1284 struct carp_softc *vh; 1285 struct ifaddr *ifa; 1286 1287 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 1288 IFADDR_FOREACH(ifa, &vh->sc_if) { 1289 if (IN6_ARE_ADDR_EQUAL(taddr, 1290 &ifatoia6(ifa)->ia_addr.sin6_addr) && 1291 ((vh->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) == 1292 (IFF_UP|IFF_RUNNING)) && vh->sc_state == MASTER) 1293 return (ifa); 1294 } 1295 } 1296 1297 return (NULL); 1298 } 1299 #endif /* INET6 */ 1300 1301 struct ifnet * 1302 carp_ourether(void *v, struct ether_header *eh, u_char iftype, int src) 1303 { 1304 struct carp_if *cif = (struct carp_if *)v; 1305 struct carp_softc *vh; 1306 u_int8_t *ena; 1307 1308 if (src) 1309 ena = (u_int8_t *)&eh->ether_shost; 1310 else 1311 ena = (u_int8_t *)&eh->ether_dhost; 1312 1313 switch (iftype) { 1314 case IFT_ETHER: 1315 case IFT_FDDI: 1316 if (ena[0] || ena[1] || ena[2] != 0x5e || ena[3] || ena[4] != 1) 1317 return (NULL); 1318 break; 1319 case IFT_ISO88025: 1320 if (ena[0] != 3 || ena[1] || ena[4] || ena[5]) 1321 return (NULL); 1322 break; 1323 default: 1324 return (NULL); 1325 break; 1326 } 1327 1328 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) 1329 if ((vh->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) == 1330 (IFF_UP|IFF_RUNNING) && vh->sc_state == MASTER && 1331 !bcmp(ena, CLLADDR(vh->sc_if.if_sadl), 1332 ETHER_ADDR_LEN)) { 1333 return (&vh->sc_if); 1334 } 1335 1336 return (NULL); 1337 } 1338 1339 int 1340 carp_input(struct mbuf *m, u_int8_t *shost, u_int8_t *dhost, u_int16_t etype) 1341 { 1342 struct ether_header eh; 1343 struct carp_if *cif = (struct carp_if *)m->m_pkthdr.rcvif->if_carp; 1344 struct ifnet *ifp; 1345 1346 bcopy(shost, &eh.ether_shost, sizeof(eh.ether_shost)); 1347 bcopy(dhost, &eh.ether_dhost, sizeof(eh.ether_dhost)); 1348 eh.ether_type = etype; 1349 1350 if (m->m_flags & (M_BCAST|M_MCAST)) { 1351 struct carp_softc *vh; 1352 struct mbuf *m0; 1353 1354 /* 1355 * XXX Should really check the list of multicast addresses 1356 * for each CARP interface _before_ copying. 1357 */ 1358 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 1359 m0 = m_copym(m, 0, M_COPYALL, M_DONTWAIT); 1360 if (m0 == NULL) 1361 continue; 1362 m0->m_pkthdr.rcvif = &vh->sc_if; 1363 ether_input(&vh->sc_if, m0); 1364 } 1365 return (1); 1366 } 1367 1368 ifp = carp_ourether(cif, &eh, m->m_pkthdr.rcvif->if_type, 0); 1369 if (ifp == NULL) { 1370 return (1); 1371 } 1372 1373 m->m_pkthdr.rcvif = ifp; 1374 1375 #if NBPFILTER > 0 1376 if (ifp->if_bpf) 1377 bpf_mtap(ifp->if_bpf, m); 1378 #endif 1379 ifp->if_ipackets++; 1380 ether_input(ifp, m); 1381 return (0); 1382 } 1383 1384 void 1385 carp_master_down(void *v) 1386 { 1387 struct carp_softc *sc = v; 1388 1389 switch (sc->sc_state) { 1390 case INIT: 1391 printf("%s: master_down event in INIT state\n", 1392 sc->sc_if.if_xname); 1393 break; 1394 case MASTER: 1395 break; 1396 case BACKUP: 1397 carp_set_state(sc, MASTER); 1398 carp_send_ad(sc); 1399 carp_send_arp(sc); 1400 #ifdef INET6 1401 carp_send_na(sc); 1402 #endif /* INET6 */ 1403 carp_setrun(sc, 0); 1404 carp_setroute(sc, RTM_ADD); 1405 break; 1406 } 1407 } 1408 1409 /* 1410 * When in backup state, af indicates whether to reset the master down timer 1411 * for v4 or v6. If it's set to zero, reset the ones which are already pending. 1412 */ 1413 void 1414 carp_setrun(struct carp_softc *sc, sa_family_t af) 1415 { 1416 struct timeval tv; 1417 1418 if (sc->sc_carpdev == NULL) { 1419 sc->sc_if.if_flags &= ~IFF_RUNNING; 1420 carp_set_state(sc, INIT); 1421 return; 1422 } 1423 1424 if (sc->sc_if.if_flags & IFF_UP && sc->sc_vhid > 0 && 1425 (sc->sc_naddrs || sc->sc_naddrs6) && !sc->sc_suppress) { 1426 sc->sc_if.if_flags |= IFF_RUNNING; 1427 } else { 1428 sc->sc_if.if_flags &= ~IFF_RUNNING; 1429 carp_setroute(sc, RTM_DELETE); 1430 return; 1431 } 1432 1433 switch (sc->sc_state) { 1434 case INIT: 1435 carp_set_state(sc, BACKUP); 1436 carp_setroute(sc, RTM_DELETE); 1437 carp_setrun(sc, 0); 1438 break; 1439 case BACKUP: 1440 callout_stop(&sc->sc_ad_tmo); 1441 tv.tv_sec = 3 * sc->sc_advbase; 1442 tv.tv_usec = sc->sc_advskew * 1000000 / 256; 1443 switch (af) { 1444 #ifdef INET 1445 case AF_INET: 1446 callout_schedule(&sc->sc_md_tmo, tvtohz(&tv)); 1447 break; 1448 #endif /* INET */ 1449 #ifdef INET6 1450 case AF_INET6: 1451 callout_schedule(&sc->sc_md6_tmo, tvtohz(&tv)); 1452 break; 1453 #endif /* INET6 */ 1454 default: 1455 if (sc->sc_naddrs) 1456 callout_schedule(&sc->sc_md_tmo, tvtohz(&tv)); 1457 if (sc->sc_naddrs6) 1458 callout_schedule(&sc->sc_md6_tmo, tvtohz(&tv)); 1459 break; 1460 } 1461 break; 1462 case MASTER: 1463 tv.tv_sec = sc->sc_advbase; 1464 tv.tv_usec = sc->sc_advskew * 1000000 / 256; 1465 callout_schedule(&sc->sc_ad_tmo, tvtohz(&tv)); 1466 break; 1467 } 1468 } 1469 1470 void 1471 carp_multicast_cleanup(struct carp_softc *sc) 1472 { 1473 struct ip_moptions *imo = &sc->sc_imo; 1474 #ifdef INET6 1475 struct ip6_moptions *im6o = &sc->sc_im6o; 1476 #endif 1477 u_int16_t n = imo->imo_num_memberships; 1478 1479 /* Clean up our own multicast memberships */ 1480 while (n-- > 0) { 1481 if (imo->imo_membership[n] != NULL) { 1482 in_delmulti(imo->imo_membership[n]); 1483 imo->imo_membership[n] = NULL; 1484 } 1485 } 1486 imo->imo_num_memberships = 0; 1487 imo->imo_multicast_ifp = NULL; 1488 1489 #ifdef INET6 1490 while (!LIST_EMPTY(&im6o->im6o_memberships)) { 1491 struct in6_multi_mship *imm = 1492 LIST_FIRST(&im6o->im6o_memberships); 1493 1494 LIST_REMOVE(imm, i6mm_chain); 1495 in6_leavegroup(imm); 1496 } 1497 im6o->im6o_multicast_ifp = NULL; 1498 #endif 1499 1500 /* And any other multicast memberships */ 1501 carp_ether_purgemulti(sc); 1502 } 1503 1504 int 1505 carp_set_ifp(struct carp_softc *sc, struct ifnet *ifp) 1506 { 1507 struct carp_if *cif, *ncif = NULL; 1508 struct carp_softc *vr, *after = NULL; 1509 int myself = 0, error = 0; 1510 int s; 1511 1512 if (ifp == sc->sc_carpdev) 1513 return (0); 1514 1515 if (ifp != NULL) { 1516 if ((ifp->if_flags & IFF_MULTICAST) == 0) 1517 return (EADDRNOTAVAIL); 1518 1519 if (ifp->if_type == IFT_CARP) 1520 return (EINVAL); 1521 1522 if (ifp->if_carp == NULL) { 1523 MALLOC(ncif, struct carp_if *, sizeof(*cif), 1524 M_IFADDR, M_NOWAIT); 1525 if (ncif == NULL) 1526 return (ENOBUFS); 1527 if ((error = ifpromisc(ifp, 1))) { 1528 FREE(ncif, M_IFADDR); 1529 return (error); 1530 } 1531 1532 ncif->vhif_ifp = ifp; 1533 TAILQ_INIT(&ncif->vhif_vrs); 1534 } else { 1535 cif = (struct carp_if *)ifp->if_carp; 1536 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) 1537 if (vr != sc && vr->sc_vhid == sc->sc_vhid) 1538 return (EINVAL); 1539 } 1540 1541 /* detach from old interface */ 1542 if (sc->sc_carpdev != NULL) 1543 carpdetach(sc); 1544 1545 /* join multicast groups */ 1546 if (sc->sc_naddrs < 0 && 1547 (error = carp_join_multicast(sc)) != 0) { 1548 if (ncif != NULL) 1549 FREE(ncif, M_IFADDR); 1550 return (error); 1551 } 1552 1553 #ifdef INET6 1554 if (sc->sc_naddrs6 < 0 && 1555 (error = carp_join_multicast6(sc)) != 0) { 1556 if (ncif != NULL) 1557 FREE(ncif, M_IFADDR); 1558 carp_multicast_cleanup(sc); 1559 return (error); 1560 } 1561 #endif 1562 1563 /* attach carp interface to physical interface */ 1564 if (ncif != NULL) 1565 ifp->if_carp = (void *)ncif; 1566 sc->sc_carpdev = ifp; 1567 cif = (struct carp_if *)ifp->if_carp; 1568 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) { 1569 if (vr == sc) 1570 myself = 1; 1571 if (vr->sc_vhid < sc->sc_vhid) 1572 after = vr; 1573 } 1574 1575 if (!myself) { 1576 /* We're trying to keep things in order */ 1577 if (after == NULL) { 1578 TAILQ_INSERT_TAIL(&cif->vhif_vrs, sc, sc_list); 1579 } else { 1580 TAILQ_INSERT_AFTER(&cif->vhif_vrs, after, 1581 sc, sc_list); 1582 } 1583 cif->vhif_nvrs++; 1584 } 1585 if (sc->sc_naddrs || sc->sc_naddrs6) 1586 sc->sc_if.if_flags |= IFF_UP; 1587 carp_set_enaddr(sc); 1588 s = splnet(); 1589 /* XXX linkstatehooks establish */ 1590 carp_carpdev_state(ifp); 1591 splx(s); 1592 } else { 1593 carpdetach(sc); 1594 sc->sc_if.if_flags &= ~(IFF_UP|IFF_RUNNING); 1595 } 1596 return (0); 1597 } 1598 1599 void 1600 carp_set_enaddr(struct carp_softc *sc) 1601 { 1602 uint8_t enaddr[ETHER_ADDR_LEN]; 1603 if (sc->sc_carpdev && sc->sc_carpdev->if_type == IFT_ISO88025) { 1604 enaddr[0] = 3; 1605 enaddr[1] = 0; 1606 enaddr[2] = 0x40 >> (sc->sc_vhid - 1); 1607 enaddr[3] = 0x40000 >> (sc->sc_vhid - 1); 1608 enaddr[4] = 0; 1609 enaddr[5] = 0; 1610 } else { 1611 enaddr[0] = 0; 1612 enaddr[1] = 0; 1613 enaddr[2] = 0x5e; 1614 enaddr[3] = 0; 1615 enaddr[4] = 1; 1616 enaddr[5] = sc->sc_vhid; 1617 } 1618 if_set_sadl(&sc->sc_if, enaddr, sizeof(enaddr)); 1619 } 1620 1621 void 1622 carp_addr_updated(void *v) 1623 { 1624 struct carp_softc *sc = (struct carp_softc *) v; 1625 struct ifaddr *ifa; 1626 int new_naddrs = 0, new_naddrs6 = 0; 1627 1628 IFADDR_FOREACH(ifa, &sc->sc_if) { 1629 if (ifa->ifa_addr->sa_family == AF_INET) 1630 new_naddrs++; 1631 else if (ifa->ifa_addr->sa_family == AF_INET6) 1632 new_naddrs6++; 1633 } 1634 1635 /* Handle a callback after SIOCDIFADDR */ 1636 if (new_naddrs < sc->sc_naddrs || new_naddrs6 < sc->sc_naddrs6) { 1637 struct in_addr mc_addr; 1638 struct in_multi *inm; 1639 1640 sc->sc_naddrs = new_naddrs; 1641 sc->sc_naddrs6 = new_naddrs6; 1642 1643 /* Re-establish multicast membership removed by in_control */ 1644 mc_addr.s_addr = INADDR_CARP_GROUP; 1645 IN_LOOKUP_MULTI(mc_addr, &sc->sc_if, inm); 1646 if (inm == NULL) { 1647 bzero(&sc->sc_imo, sizeof(sc->sc_imo)); 1648 1649 if (sc->sc_carpdev != NULL && sc->sc_naddrs > 0) 1650 carp_join_multicast(sc); 1651 } 1652 1653 if (sc->sc_naddrs == 0 && sc->sc_naddrs6 == 0) { 1654 sc->sc_if.if_flags &= ~IFF_UP; 1655 carp_set_state(sc, INIT); 1656 } else 1657 carp_hmac_prepare(sc); 1658 } 1659 1660 carp_setrun(sc, 0); 1661 } 1662 1663 int 1664 carp_set_addr(struct carp_softc *sc, struct sockaddr_in *sin) 1665 { 1666 struct ifnet *ifp = sc->sc_carpdev; 1667 struct in_ifaddr *ia, *ia_if; 1668 int error = 0; 1669 1670 if (sin->sin_addr.s_addr == 0) { 1671 if (!(sc->sc_if.if_flags & IFF_UP)) 1672 carp_set_state(sc, INIT); 1673 if (sc->sc_naddrs) 1674 sc->sc_if.if_flags |= IFF_UP; 1675 carp_setrun(sc, 0); 1676 return (0); 1677 } 1678 1679 /* we have to do this by hand to ensure we don't match on ourselves */ 1680 ia_if = NULL; 1681 for (ia = TAILQ_FIRST(&in_ifaddrhead); ia; 1682 ia = TAILQ_NEXT(ia, ia_list)) { 1683 1684 /* and, yeah, we need a multicast-capable iface too */ 1685 if (ia->ia_ifp != &sc->sc_if && 1686 ia->ia_ifp->if_type != IFT_CARP && 1687 (ia->ia_ifp->if_flags & IFF_MULTICAST) && 1688 (sin->sin_addr.s_addr & ia->ia_subnetmask) == 1689 ia->ia_subnet) { 1690 if (!ia_if) 1691 ia_if = ia; 1692 } 1693 } 1694 1695 if (ia_if) { 1696 ia = ia_if; 1697 if (ifp) { 1698 if (ifp != ia->ia_ifp) 1699 return (EADDRNOTAVAIL); 1700 } else { 1701 ifp = ia->ia_ifp; 1702 } 1703 } 1704 1705 if ((error = carp_set_ifp(sc, ifp))) 1706 return (error); 1707 1708 if (sc->sc_carpdev == NULL) 1709 return (EADDRNOTAVAIL); 1710 1711 if (sc->sc_naddrs == 0 && (error = carp_join_multicast(sc)) != 0) 1712 return (error); 1713 1714 sc->sc_naddrs++; 1715 if (sc->sc_carpdev != NULL) 1716 sc->sc_if.if_flags |= IFF_UP; 1717 1718 carp_set_state(sc, INIT); 1719 carp_setrun(sc, 0); 1720 1721 /* 1722 * Hook if_addrhooks so that we get a callback after in_ifinit has run, 1723 * to correct any inappropriate routes that it inserted. 1724 */ 1725 if (sc->ah_cookie == 0) { 1726 /* XXX link address hook */ 1727 } 1728 1729 return (0); 1730 } 1731 1732 int 1733 carp_join_multicast(struct carp_softc *sc) 1734 { 1735 struct ip_moptions *imo = &sc->sc_imo, tmpimo; 1736 struct in_addr addr; 1737 1738 bzero(&tmpimo, sizeof(tmpimo)); 1739 addr.s_addr = INADDR_CARP_GROUP; 1740 if ((tmpimo.imo_membership[0] = 1741 in_addmulti(&addr, &sc->sc_if)) == NULL) { 1742 return (ENOBUFS); 1743 } 1744 1745 imo->imo_membership[0] = tmpimo.imo_membership[0]; 1746 imo->imo_num_memberships = 1; 1747 imo->imo_multicast_ifp = &sc->sc_if; 1748 imo->imo_multicast_ttl = CARP_DFLTTL; 1749 imo->imo_multicast_loop = 0; 1750 return (0); 1751 } 1752 1753 1754 #ifdef INET6 1755 int 1756 carp_set_addr6(struct carp_softc *sc, struct sockaddr_in6 *sin6) 1757 { 1758 struct ifnet *ifp = sc->sc_carpdev; 1759 struct in6_ifaddr *ia, *ia_if; 1760 int error = 0; 1761 1762 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) { 1763 if (!(sc->sc_if.if_flags & IFF_UP)) 1764 carp_set_state(sc, INIT); 1765 if (sc->sc_naddrs6) 1766 sc->sc_if.if_flags |= IFF_UP; 1767 carp_setrun(sc, 0); 1768 return (0); 1769 } 1770 1771 /* we have to do this by hand to ensure we don't match on ourselves */ 1772 ia_if = NULL; 1773 for (ia = in6_ifaddr; ia; ia = ia->ia_next) { 1774 int i; 1775 1776 for (i = 0; i < 4; i++) { 1777 if ((sin6->sin6_addr.s6_addr32[i] & 1778 ia->ia_prefixmask.sin6_addr.s6_addr32[i]) != 1779 (ia->ia_addr.sin6_addr.s6_addr32[i] & 1780 ia->ia_prefixmask.sin6_addr.s6_addr32[i])) 1781 break; 1782 } 1783 /* and, yeah, we need a multicast-capable iface too */ 1784 if (ia->ia_ifp != &sc->sc_if && 1785 ia->ia_ifp->if_type != IFT_CARP && 1786 (ia->ia_ifp->if_flags & IFF_MULTICAST) && 1787 (i == 4)) { 1788 if (!ia_if) 1789 ia_if = ia; 1790 } 1791 } 1792 1793 if (ia_if) { 1794 ia = ia_if; 1795 if (sc->sc_carpdev) { 1796 if (sc->sc_carpdev != ia->ia_ifp) 1797 return (EADDRNOTAVAIL); 1798 } else { 1799 ifp = ia->ia_ifp; 1800 } 1801 } 1802 1803 if ((error = carp_set_ifp(sc, ifp))) 1804 return (error); 1805 1806 if (sc->sc_carpdev == NULL) 1807 return (EADDRNOTAVAIL); 1808 1809 if (sc->sc_naddrs6 == 0 && (error = carp_join_multicast6(sc)) != 0) 1810 return (error); 1811 1812 sc->sc_naddrs6++; 1813 if (sc->sc_carpdev != NULL) 1814 sc->sc_if.if_flags |= IFF_UP; 1815 carp_set_state(sc, INIT); 1816 carp_setrun(sc, 0); 1817 1818 return (0); 1819 } 1820 1821 int 1822 carp_join_multicast6(struct carp_softc *sc) 1823 { 1824 struct in6_multi_mship *imm, *imm2; 1825 struct ip6_moptions *im6o = &sc->sc_im6o; 1826 struct sockaddr_in6 addr6; 1827 int error; 1828 1829 /* Join IPv6 CARP multicast group */ 1830 bzero(&addr6, sizeof(addr6)); 1831 addr6.sin6_family = AF_INET6; 1832 addr6.sin6_len = sizeof(addr6); 1833 addr6.sin6_addr.s6_addr16[0] = htons(0xff02); 1834 addr6.sin6_addr.s6_addr16[1] = htons(sc->sc_if.if_index); 1835 addr6.sin6_addr.s6_addr8[15] = 0x12; 1836 if ((imm = in6_joingroup(&sc->sc_if, 1837 &addr6.sin6_addr, &error, 0)) == NULL) { 1838 return (error); 1839 } 1840 /* join solicited multicast address */ 1841 bzero(&addr6.sin6_addr, sizeof(addr6.sin6_addr)); 1842 addr6.sin6_addr.s6_addr16[0] = htons(0xff02); 1843 addr6.sin6_addr.s6_addr16[1] = htons(sc->sc_if.if_index); 1844 addr6.sin6_addr.s6_addr32[1] = 0; 1845 addr6.sin6_addr.s6_addr32[2] = htonl(1); 1846 addr6.sin6_addr.s6_addr32[3] = 0; 1847 addr6.sin6_addr.s6_addr8[12] = 0xff; 1848 if ((imm2 = in6_joingroup(&sc->sc_if, 1849 &addr6.sin6_addr, &error, 0)) == NULL) { 1850 in6_leavegroup(imm); 1851 return (error); 1852 } 1853 1854 /* apply v6 multicast membership */ 1855 im6o->im6o_multicast_ifp = &sc->sc_if; 1856 if (imm) 1857 LIST_INSERT_HEAD(&im6o->im6o_memberships, imm, 1858 i6mm_chain); 1859 if (imm2) 1860 LIST_INSERT_HEAD(&im6o->im6o_memberships, imm2, 1861 i6mm_chain); 1862 1863 return (0); 1864 } 1865 1866 #endif /* INET6 */ 1867 1868 int 1869 carp_ioctl(struct ifnet *ifp, u_long cmd, void *addr) 1870 { 1871 struct lwp *l = curlwp; /* XXX */ 1872 struct carp_softc *sc = ifp->if_softc, *vr; 1873 struct carpreq carpr; 1874 struct ifaddr *ifa; 1875 struct ifreq *ifr; 1876 struct ifnet *cdev = NULL; 1877 int error = 0; 1878 1879 ifa = (struct ifaddr *)addr; 1880 ifr = (struct ifreq *)addr; 1881 1882 switch (cmd) { 1883 case SIOCSIFADDR: 1884 switch (ifa->ifa_addr->sa_family) { 1885 #ifdef INET 1886 case AF_INET: 1887 sc->sc_if.if_flags |= IFF_UP; 1888 bcopy(ifa->ifa_addr, ifa->ifa_dstaddr, 1889 sizeof(struct sockaddr)); 1890 error = carp_set_addr(sc, satosin(ifa->ifa_addr)); 1891 break; 1892 #endif /* INET */ 1893 #ifdef INET6 1894 case AF_INET6: 1895 sc->sc_if.if_flags|= IFF_UP; 1896 error = carp_set_addr6(sc, satosin6(ifa->ifa_addr)); 1897 break; 1898 #endif /* INET6 */ 1899 default: 1900 error = EAFNOSUPPORT; 1901 break; 1902 } 1903 break; 1904 1905 case SIOCSIFFLAGS: 1906 if (sc->sc_state != INIT && !(ifr->ifr_flags & IFF_UP)) { 1907 callout_stop(&sc->sc_ad_tmo); 1908 callout_stop(&sc->sc_md_tmo); 1909 callout_stop(&sc->sc_md6_tmo); 1910 if (sc->sc_state == MASTER) { 1911 /* we need the interface up to bow out */ 1912 sc->sc_if.if_flags |= IFF_UP; 1913 sc->sc_bow_out = 1; 1914 carp_send_ad(sc); 1915 } 1916 sc->sc_if.if_flags &= ~IFF_UP; 1917 carp_set_state(sc, INIT); 1918 carp_setrun(sc, 0); 1919 } else if (sc->sc_state == INIT && (ifr->ifr_flags & IFF_UP)) { 1920 sc->sc_if.if_flags |= IFF_UP; 1921 carp_setrun(sc, 0); 1922 } 1923 break; 1924 1925 case SIOCSVH: 1926 if (l == NULL) 1927 break; 1928 if ((error = kauth_authorize_network(l->l_cred, 1929 KAUTH_NETWORK_INTERFACE, 1930 KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd, 1931 NULL)) != 0) 1932 break; 1933 if ((error = copyin(ifr->ifr_data, &carpr, sizeof carpr))) 1934 break; 1935 error = 1; 1936 if (carpr.carpr_carpdev[0] != '\0' && 1937 (cdev = ifunit(carpr.carpr_carpdev)) == NULL) 1938 return (EINVAL); 1939 if ((error = carp_set_ifp(sc, cdev))) 1940 return (error); 1941 if (sc->sc_state != INIT && carpr.carpr_state != sc->sc_state) { 1942 switch (carpr.carpr_state) { 1943 case BACKUP: 1944 callout_stop(&sc->sc_ad_tmo); 1945 carp_set_state(sc, BACKUP); 1946 carp_setrun(sc, 0); 1947 carp_setroute(sc, RTM_DELETE); 1948 break; 1949 case MASTER: 1950 carp_master_down(sc); 1951 break; 1952 default: 1953 break; 1954 } 1955 } 1956 if (carpr.carpr_vhid > 0) { 1957 if (carpr.carpr_vhid > 255) { 1958 error = EINVAL; 1959 break; 1960 } 1961 if (sc->sc_carpdev) { 1962 struct carp_if *cif; 1963 cif = (struct carp_if *)sc->sc_carpdev->if_carp; 1964 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) 1965 if (vr != sc && 1966 vr->sc_vhid == carpr.carpr_vhid) 1967 return (EINVAL); 1968 } 1969 sc->sc_vhid = carpr.carpr_vhid; 1970 carp_set_enaddr(sc); 1971 carp_set_state(sc, INIT); 1972 error--; 1973 } 1974 if (carpr.carpr_advbase > 0 || carpr.carpr_advskew > 0) { 1975 if (carpr.carpr_advskew > 254) { 1976 error = EINVAL; 1977 break; 1978 } 1979 if (carpr.carpr_advbase > 255) { 1980 error = EINVAL; 1981 break; 1982 } 1983 sc->sc_advbase = carpr.carpr_advbase; 1984 sc->sc_advskew = carpr.carpr_advskew; 1985 error--; 1986 } 1987 bcopy(carpr.carpr_key, sc->sc_key, sizeof(sc->sc_key)); 1988 if (error > 0) 1989 error = EINVAL; 1990 else { 1991 error = 0; 1992 carp_setrun(sc, 0); 1993 } 1994 break; 1995 1996 case SIOCGVH: 1997 bzero(&carpr, sizeof(carpr)); 1998 if (sc->sc_carpdev != NULL) 1999 strlcpy(carpr.carpr_carpdev, sc->sc_carpdev->if_xname, 2000 IFNAMSIZ); 2001 carpr.carpr_state = sc->sc_state; 2002 carpr.carpr_vhid = sc->sc_vhid; 2003 carpr.carpr_advbase = sc->sc_advbase; 2004 carpr.carpr_advskew = sc->sc_advskew; 2005 2006 if ((l == NULL) || (error = kauth_authorize_network(l->l_cred, 2007 KAUTH_NETWORK_INTERFACE, 2008 KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd, 2009 NULL)) != 0) 2010 bcopy(sc->sc_key, carpr.carpr_key, 2011 sizeof(carpr.carpr_key)); 2012 error = copyout(&carpr, ifr->ifr_data, sizeof(carpr)); 2013 break; 2014 2015 case SIOCADDMULTI: 2016 error = carp_ether_addmulti(sc, ifr); 2017 break; 2018 2019 case SIOCDELMULTI: 2020 error = carp_ether_delmulti(sc, ifr); 2021 break; 2022 2023 default: 2024 error = EINVAL; 2025 } 2026 2027 carp_hmac_prepare(sc); 2028 return (error); 2029 } 2030 2031 2032 /* 2033 * Start output on carp interface. This function should never be called. 2034 */ 2035 void 2036 carp_start(struct ifnet *ifp) 2037 { 2038 #ifdef DEBUG 2039 printf("%s: start called\n", ifp->if_xname); 2040 #endif 2041 } 2042 2043 int 2044 carp_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *sa, 2045 struct rtentry *rt) 2046 { 2047 struct carp_softc *sc = ((struct carp_softc *)ifp->if_softc); 2048 2049 if (sc->sc_carpdev != NULL && sc->sc_state == MASTER) { 2050 return (sc->sc_carpdev->if_output(ifp, m, sa, rt)); 2051 } else { 2052 m_freem(m); 2053 return (ENETUNREACH); 2054 } 2055 } 2056 2057 void 2058 carp_set_state(struct carp_softc *sc, int state) 2059 { 2060 if (sc->sc_state == state) 2061 return; 2062 2063 sc->sc_state = state; 2064 switch (state) { 2065 case BACKUP: 2066 sc->sc_if.if_link_state = LINK_STATE_DOWN; 2067 break; 2068 case MASTER: 2069 sc->sc_if.if_link_state = LINK_STATE_UP; 2070 break; 2071 default: 2072 sc->sc_if.if_link_state = LINK_STATE_UNKNOWN; 2073 break; 2074 } 2075 rt_ifmsg(&sc->sc_if); 2076 } 2077 2078 void 2079 carp_carpdev_state(void *v) 2080 { 2081 struct carp_if *cif; 2082 struct carp_softc *sc; 2083 struct ifnet *ifp = v; 2084 2085 if (ifp->if_type == IFT_CARP) 2086 return; 2087 2088 cif = (struct carp_if *)ifp->if_carp; 2089 2090 TAILQ_FOREACH(sc, &cif->vhif_vrs, sc_list) { 2091 int suppressed = sc->sc_suppress; 2092 2093 if (sc->sc_carpdev->if_link_state == LINK_STATE_DOWN || 2094 !(sc->sc_carpdev->if_flags & IFF_UP)) { 2095 sc->sc_if.if_flags &= ~IFF_RUNNING; 2096 callout_stop(&sc->sc_ad_tmo); 2097 callout_stop(&sc->sc_md_tmo); 2098 callout_stop(&sc->sc_md6_tmo); 2099 carp_set_state(sc, INIT); 2100 sc->sc_suppress = 1; 2101 carp_setrun(sc, 0); 2102 if (!suppressed) { 2103 carp_suppress_preempt++; 2104 if (carp_suppress_preempt == 1) 2105 carp_send_ad_all(); 2106 } 2107 } else { 2108 carp_set_state(sc, INIT); 2109 sc->sc_suppress = 0; 2110 carp_setrun(sc, 0); 2111 if (suppressed) 2112 carp_suppress_preempt--; 2113 } 2114 } 2115 } 2116 2117 int 2118 carp_ether_addmulti(struct carp_softc *sc, struct ifreq *ifr) 2119 { 2120 const struct sockaddr *sa = ifreq_getaddr(SIOCADDMULTI, ifr); 2121 struct ifnet *ifp; 2122 struct carp_mc_entry *mc; 2123 u_int8_t addrlo[ETHER_ADDR_LEN], addrhi[ETHER_ADDR_LEN]; 2124 int error; 2125 2126 ifp = sc->sc_carpdev; 2127 if (ifp == NULL) 2128 return (EINVAL); 2129 2130 error = ether_addmulti(sa, &sc->sc_ac); 2131 if (error != ENETRESET) 2132 return (error); 2133 2134 /* 2135 * This is new multicast address. We have to tell parent 2136 * about it. Also, remember this multicast address so that 2137 * we can delete them on unconfigure. 2138 */ 2139 MALLOC(mc, struct carp_mc_entry *, sizeof(struct carp_mc_entry), 2140 M_DEVBUF, M_NOWAIT); 2141 if (mc == NULL) { 2142 error = ENOMEM; 2143 goto alloc_failed; 2144 } 2145 2146 /* 2147 * As ether_addmulti() returns ENETRESET, following two 2148 * statement shouldn't fail. 2149 */ 2150 (void)ether_multiaddr(sa, addrlo, addrhi); 2151 ETHER_LOOKUP_MULTI(addrlo, addrhi, &sc->sc_ac, mc->mc_enm); 2152 memcpy(&mc->mc_addr, sa, sa->sa_len); 2153 LIST_INSERT_HEAD(&sc->carp_mc_listhead, mc, mc_entries); 2154 2155 error = (*ifp->if_ioctl)(ifp, SIOCADDMULTI, (void *)ifr); 2156 if (error != 0) 2157 goto ioctl_failed; 2158 2159 return (error); 2160 2161 ioctl_failed: 2162 LIST_REMOVE(mc, mc_entries); 2163 FREE(mc, M_DEVBUF); 2164 alloc_failed: 2165 (void)ether_delmulti(sa, &sc->sc_ac); 2166 2167 return (error); 2168 } 2169 2170 int 2171 carp_ether_delmulti(struct carp_softc *sc, struct ifreq *ifr) 2172 { 2173 const struct sockaddr *sa = ifreq_getaddr(SIOCDELMULTI, ifr); 2174 struct ifnet *ifp; 2175 struct ether_multi *enm; 2176 struct carp_mc_entry *mc; 2177 u_int8_t addrlo[ETHER_ADDR_LEN], addrhi[ETHER_ADDR_LEN]; 2178 int error; 2179 2180 ifp = sc->sc_carpdev; 2181 if (ifp == NULL) 2182 return (EINVAL); 2183 2184 /* 2185 * Find a key to lookup carp_mc_entry. We have to do this 2186 * before calling ether_delmulti for obvious reason. 2187 */ 2188 if ((error = ether_multiaddr(sa, addrlo, addrhi)) != 0) 2189 return (error); 2190 ETHER_LOOKUP_MULTI(addrlo, addrhi, &sc->sc_ac, enm); 2191 if (enm == NULL) 2192 return (EINVAL); 2193 2194 LIST_FOREACH(mc, &sc->carp_mc_listhead, mc_entries) 2195 if (mc->mc_enm == enm) 2196 break; 2197 2198 /* We won't delete entries we didn't add */ 2199 if (mc == NULL) 2200 return (EINVAL); 2201 2202 error = ether_delmulti(sa, &sc->sc_ac); 2203 if (error != ENETRESET) 2204 return (error); 2205 2206 /* We no longer use this multicast address. Tell parent so. */ 2207 error = (*ifp->if_ioctl)(ifp, SIOCDELMULTI, (void *)ifr); 2208 if (error == 0) { 2209 /* And forget about this address. */ 2210 LIST_REMOVE(mc, mc_entries); 2211 FREE(mc, M_DEVBUF); 2212 } else 2213 (void)ether_addmulti(sa, &sc->sc_ac); 2214 return (error); 2215 } 2216 2217 /* 2218 * Delete any multicast address we have asked to add from parent 2219 * interface. Called when the carp is being unconfigured. 2220 */ 2221 void 2222 carp_ether_purgemulti(struct carp_softc *sc) 2223 { 2224 struct ifnet *ifp = sc->sc_carpdev; /* Parent. */ 2225 struct carp_mc_entry *mc; 2226 union { 2227 struct ifreq ifreq; 2228 struct { 2229 char ifr_name[IFNAMSIZ]; 2230 struct sockaddr_storage ifr_ss; 2231 } ifreq_storage; 2232 } u; 2233 struct ifreq *ifr = &u.ifreq; 2234 2235 if (ifp == NULL) 2236 return; 2237 2238 memcpy(ifr->ifr_name, ifp->if_xname, IFNAMSIZ); 2239 while ((mc = LIST_FIRST(&sc->carp_mc_listhead)) != NULL) { 2240 memcpy(&ifr->ifr_addr, &mc->mc_addr, mc->mc_addr.ss_len); 2241 (void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, (void *)ifr); 2242 LIST_REMOVE(mc, mc_entries); 2243 FREE(mc, M_DEVBUF); 2244 } 2245 } 2246 2247 static void 2248 carpstat_convert_to_user_cb(void *v1, void *v2, struct cpu_info *ci) 2249 { 2250 uint64_t *carpsc = v1; 2251 uint64_t *carps = v2; 2252 u_int i; 2253 2254 for (i = 0; i < CARP_NSTATS; i++) 2255 carps[i] += carpsc[i]; 2256 } 2257 2258 static void 2259 carpstat_convert_to_user(uint64_t *carps) 2260 { 2261 2262 memset(carps, 0, sizeof(uint64_t) * CARP_NSTATS); 2263 percpu_foreach(carpstat_percpu, carpstat_convert_to_user_cb, carps); 2264 } 2265 2266 static int 2267 sysctl_net_inet_carp_stats(SYSCTLFN_ARGS) 2268 { 2269 struct sysctlnode node; 2270 uint64_t carps[CARP_NSTATS]; 2271 2272 carpstat_convert_to_user(carps); 2273 node = *rnode; 2274 node.sysctl_data = carps; 2275 node.sysctl_size = sizeof(carps); 2276 return (sysctl_lookup(SYSCTLFN_CALL(&node))); 2277 } 2278 2279 SYSCTL_SETUP(sysctl_net_inet_carp_setup, "sysctl net.inet.carp subtree setup") 2280 { 2281 2282 sysctl_createv(clog, 0, NULL, NULL, 2283 CTLFLAG_PERMANENT, 2284 CTLTYPE_NODE, "net", NULL, 2285 NULL, 0, NULL, 0, 2286 CTL_NET, CTL_EOL); 2287 sysctl_createv(clog, 0, NULL, NULL, 2288 CTLFLAG_PERMANENT, 2289 CTLTYPE_NODE, "inet", NULL, 2290 NULL, 0, NULL, 0, 2291 CTL_NET, PF_INET, CTL_EOL); 2292 sysctl_createv(clog, 0, NULL, NULL, 2293 CTLFLAG_PERMANENT, 2294 CTLTYPE_NODE, "carp", 2295 SYSCTL_DESCR("CARP related settings"), 2296 NULL, 0, NULL, 0, 2297 CTL_NET, PF_INET, IPPROTO_CARP, CTL_EOL); 2298 2299 sysctl_createv(clog, 0, NULL, NULL, 2300 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2301 CTLTYPE_INT, "preempt", 2302 SYSCTL_DESCR("Enable CARP Preempt"), 2303 NULL, 0, &carp_opts[CARPCTL_PREEMPT], 0, 2304 CTL_NET, PF_INET, IPPROTO_CARP, 2305 CTL_CREATE, CTL_EOL); 2306 sysctl_createv(clog, 0, NULL, NULL, 2307 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2308 CTLTYPE_INT, "arpbalance", 2309 SYSCTL_DESCR("Enable ARP balancing"), 2310 NULL, 0, &carp_opts[CARPCTL_ARPBALANCE], 0, 2311 CTL_NET, PF_INET, IPPROTO_CARP, 2312 CTL_CREATE, CTL_EOL); 2313 sysctl_createv(clog, 0, NULL, NULL, 2314 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2315 CTLTYPE_INT, "allow", 2316 SYSCTL_DESCR("Enable CARP"), 2317 NULL, 0, &carp_opts[CARPCTL_ALLOW], 0, 2318 CTL_NET, PF_INET, IPPROTO_CARP, 2319 CTL_CREATE, CTL_EOL); 2320 sysctl_createv(clog, 0, NULL, NULL, 2321 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2322 CTLTYPE_INT, "log", 2323 SYSCTL_DESCR("CARP logging"), 2324 NULL, 0, &carp_opts[CARPCTL_LOG], 0, 2325 CTL_NET, PF_INET, IPPROTO_CARP, 2326 CTL_CREATE, CTL_EOL); 2327 sysctl_createv(clog, 0, NULL, NULL, 2328 CTLFLAG_PERMANENT, 2329 CTLTYPE_STRUCT, "stats", 2330 SYSCTL_DESCR("CARP statistics"), 2331 sysctl_net_inet_carp_stats, 0, NULL, 0, 2332 CTL_NET, PF_INET, IPPROTO_CARP, CARPCTL_STATS, 2333 CTL_EOL); 2334 } 2335
Indexes created Fri Sep 26 08:10:20 GMT 2025