ip_carp.c revision 1.22
1 /* $NetBSD: ip_carp.c,v 1.22 2007/12/21 02:07:55 matt 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 1.22 2007/12/21 02:07:55 matt Exp $"); 32 33 /* 34 * TODO: 35 * - iface reconfigure 36 * - support for hardware checksum calculations; 37 * 38 */ 39 40 #include <sys/param.h> 41 #include <sys/proc.h> 42 #include <sys/mbuf.h> 43 #include <sys/socket.h> 44 #include <sys/socketvar.h> 45 #include <sys/callout.h> 46 #include <sys/ioctl.h> 47 #include <sys/errno.h> 48 #include <sys/device.h> 49 #include <sys/time.h> 50 #include <sys/kernel.h> 51 #include <sys/kauth.h> 52 #include <sys/sysctl.h> 53 #include <sys/ucred.h> 54 #include <sys/syslog.h> 55 #include <sys/acct.h> 56 57 #include <sys/cpu.h> 58 59 #include <net/if.h> 60 #include <net/pfil.h> 61 #include <net/if_types.h> 62 #include <net/if_ether.h> 63 #include <net/route.h> 64 #include <net/netisr.h> 65 #include <netinet/if_inarp.h> 66 67 #include <machine/stdarg.h> 68 69 #if NFDDI > 0 70 #include <net/if_fddi.h> 71 #endif 72 #if NTOKEN > 0 73 #include <net/if_token.h> 74 #endif 75 76 #ifdef INET 77 #include <netinet/in.h> 78 #include <netinet/in_systm.h> 79 #include <netinet/in_var.h> 80 #include <netinet/ip.h> 81 #include <netinet/ip_var.h> 82 83 #include <net/if_dl.h> 84 #endif 85 86 #ifdef INET6 87 #include <netinet/icmp6.h> 88 #include <netinet/ip6.h> 89 #include <netinet6/ip6_var.h> 90 #include <netinet6/nd6.h> 91 #endif 92 93 #include "bpfilter.h" 94 #if NBPFILTER > 0 95 #include <net/bpf.h> 96 #endif 97 98 #include <sys/sha1.h> 99 100 #include <netinet/ip_carp.h> 101 102 struct carp_mc_entry { 103 LIST_ENTRY(carp_mc_entry) mc_entries; 104 union { 105 struct ether_multi *mcu_enm; 106 } mc_u; 107 struct sockaddr_storage mc_addr; 108 }; 109 #define mc_enm mc_u.mcu_enm 110 111 struct carp_softc { 112 struct ethercom sc_ac; 113 #define sc_if sc_ac.ec_if 114 #define sc_carpdev sc_ac.ec_if.if_carpdev 115 int ah_cookie; 116 int lh_cookie; 117 struct ip_moptions sc_imo; 118 #ifdef INET6 119 struct ip6_moptions sc_im6o; 120 #endif /* INET6 */ 121 TAILQ_ENTRY(carp_softc) sc_list; 122 123 enum { INIT = 0, BACKUP, MASTER } sc_state; 124 125 int sc_suppress; 126 int sc_bow_out; 127 128 int sc_sendad_errors; 129 #define CARP_SENDAD_MAX_ERRORS 3 130 int sc_sendad_success; 131 #define CARP_SENDAD_MIN_SUCCESS 3 132 133 int sc_vhid; 134 int sc_advskew; 135 int sc_naddrs; 136 int sc_naddrs6; 137 int sc_advbase; /* seconds */ 138 int sc_init_counter; 139 u_int64_t sc_counter; 140 141 /* authentication */ 142 #define CARP_HMAC_PAD 64 143 unsigned char sc_key[CARP_KEY_LEN]; 144 unsigned char sc_pad[CARP_HMAC_PAD]; 145 SHA1_CTX sc_sha1; 146 u_int32_t sc_hashkey[2]; 147 148 struct callout sc_ad_tmo; /* advertisement timeout */ 149 struct callout sc_md_tmo; /* master down timeout */ 150 struct callout sc_md6_tmo; /* master down timeout */ 151 152 LIST_HEAD(__carp_mchead, carp_mc_entry) carp_mc_listhead; 153 }; 154 155 int carp_suppress_preempt = 0; 156 int carp_opts[CARPCTL_MAXID] = { 0, 1, 0, 0, 0 }; /* XXX for now */ 157 struct carpstats carpstats; 158 159 struct carp_if { 160 TAILQ_HEAD(, carp_softc) vhif_vrs; 161 int vhif_nvrs; 162 163 struct ifnet *vhif_ifp; 164 }; 165 166 #define CARP_LOG(sc, s) \ 167 if (carp_opts[CARPCTL_LOG]) { \ 168 if (sc) \ 169 log(LOG_INFO, "%s: ", \ 170 (sc)->sc_if.if_xname); \ 171 else \ 172 log(LOG_INFO, "carp: "); \ 173 addlog s; \ 174 addlog("\n"); \ 175 } 176 177 void carp_hmac_prepare(struct carp_softc *); 178 void carp_hmac_generate(struct carp_softc *, u_int32_t *, 179 unsigned char *); 180 int carp_hmac_verify(struct carp_softc *, u_int32_t *, 181 unsigned char *); 182 void carp_setroute(struct carp_softc *, int); 183 void carp_proto_input_c(struct mbuf *, struct carp_header *, sa_family_t); 184 void carpattach(int); 185 void carpdetach(struct carp_softc *); 186 int carp_prepare_ad(struct mbuf *, struct carp_softc *, 187 struct carp_header *); 188 void carp_send_ad_all(void); 189 void carp_send_ad(void *); 190 void carp_send_arp(struct carp_softc *); 191 void carp_master_down(void *); 192 int carp_ioctl(struct ifnet *, u_long, void *); 193 void carp_start(struct ifnet *); 194 void carp_setrun(struct carp_softc *, sa_family_t); 195 void carp_set_state(struct carp_softc *, int); 196 int carp_addrcount(struct carp_if *, struct in_ifaddr *, int); 197 enum { CARP_COUNT_MASTER, CARP_COUNT_RUNNING }; 198 199 void carp_multicast_cleanup(struct carp_softc *); 200 int carp_set_ifp(struct carp_softc *, struct ifnet *); 201 void carp_set_enaddr(struct carp_softc *); 202 void carp_addr_updated(void *); 203 u_int32_t carp_hash(struct carp_softc *, u_char *); 204 int carp_set_addr(struct carp_softc *, struct sockaddr_in *); 205 int carp_join_multicast(struct carp_softc *); 206 #ifdef INET6 207 void carp_send_na(struct carp_softc *); 208 int carp_set_addr6(struct carp_softc *, struct sockaddr_in6 *); 209 int carp_join_multicast6(struct carp_softc *); 210 #endif 211 int carp_clone_create(struct if_clone *, int); 212 int carp_clone_destroy(struct ifnet *); 213 int carp_ether_addmulti(struct carp_softc *, struct ifreq *); 214 int carp_ether_delmulti(struct carp_softc *, struct ifreq *); 215 void carp_ether_purgemulti(struct carp_softc *); 216 217 struct if_clone carp_cloner = 218 IF_CLONE_INITIALIZER("carp", carp_clone_create, carp_clone_destroy); 219 220 static __inline u_int16_t 221 carp_cksum(struct mbuf *m, int len) 222 { 223 return (in_cksum(m, len)); 224 } 225 226 void 227 carp_hmac_prepare(struct carp_softc *sc) 228 { 229 u_int8_t carp_version = CARP_VERSION, type = CARP_ADVERTISEMENT; 230 u_int8_t vhid = sc->sc_vhid & 0xff; 231 SHA1_CTX sha1ctx; 232 u_int32_t kmd[5]; 233 struct ifaddr *ifa; 234 int i, found; 235 struct in_addr last, cur, in; 236 #ifdef INET6 237 struct in6_addr last6, cur6, in6; 238 #endif /* INET6 */ 239 240 /* compute ipad from key */ 241 bzero(sc->sc_pad, sizeof(sc->sc_pad)); 242 bcopy(sc->sc_key, sc->sc_pad, sizeof(sc->sc_key)); 243 for (i = 0; i < sizeof(sc->sc_pad); i++) 244 sc->sc_pad[i] ^= 0x36; 245 246 /* precompute first part of inner hash */ 247 SHA1Init(&sc->sc_sha1); 248 SHA1Update(&sc->sc_sha1, sc->sc_pad, sizeof(sc->sc_pad)); 249 SHA1Update(&sc->sc_sha1, (void *)&carp_version, sizeof(carp_version)); 250 SHA1Update(&sc->sc_sha1, (void *)&type, sizeof(type)); 251 252 /* generate a key for the arpbalance hash, before the vhid is hashed */ 253 bcopy(&sc->sc_sha1, &sha1ctx, sizeof(sha1ctx)); 254 SHA1Final((unsigned char *)kmd, &sha1ctx); 255 sc->sc_hashkey[0] = kmd[0] ^ kmd[1]; 256 sc->sc_hashkey[1] = kmd[2] ^ kmd[3]; 257 258 /* the rest of the precomputation */ 259 SHA1Update(&sc->sc_sha1, (void *)&vhid, sizeof(vhid)); 260 261 /* Hash the addresses from smallest to largest, not interface order */ 262 #ifdef INET 263 cur.s_addr = 0; 264 do { 265 found = 0; 266 last = cur; 267 cur.s_addr = 0xffffffff; 268 IFADDR_FOREACH(ifa, &sc->sc_if) { 269 in.s_addr = ifatoia(ifa)->ia_addr.sin_addr.s_addr; 270 if (ifa->ifa_addr->sa_family == AF_INET && 271 ntohl(in.s_addr) > ntohl(last.s_addr) && 272 ntohl(in.s_addr) < ntohl(cur.s_addr)) { 273 cur.s_addr = in.s_addr; 274 found++; 275 } 276 } 277 if (found) 278 SHA1Update(&sc->sc_sha1, (void *)&cur, sizeof(cur)); 279 } while (found); 280 #endif /* INET */ 281 282 #ifdef INET6 283 memset(&cur6, 0x00, sizeof(cur6)); 284 do { 285 found = 0; 286 last6 = cur6; 287 memset(&cur6, 0xff, sizeof(cur6)); 288 IFADDR_FOREACH(ifa, &sc->sc_if) { 289 in6 = ifatoia6(ifa)->ia_addr.sin6_addr; 290 if (IN6_IS_ADDR_LINKLOCAL(&in6)) 291 in6.s6_addr16[1] = 0; 292 if (ifa->ifa_addr->sa_family == AF_INET6 && 293 memcmp(&in6, &last6, sizeof(in6)) > 0 && 294 memcmp(&in6, &cur6, sizeof(in6)) < 0) { 295 cur6 = in6; 296 found++; 297 } 298 } 299 if (found) 300 SHA1Update(&sc->sc_sha1, (void *)&cur6, sizeof(cur6)); 301 } while (found); 302 #endif /* INET6 */ 303 304 /* convert ipad to opad */ 305 for (i = 0; i < sizeof(sc->sc_pad); i++) 306 sc->sc_pad[i] ^= 0x36 ^ 0x5c; 307 } 308 309 void 310 carp_hmac_generate(struct carp_softc *sc, u_int32_t counter[2], 311 unsigned char md[20]) 312 { 313 SHA1_CTX sha1ctx; 314 315 /* fetch first half of inner hash */ 316 bcopy(&sc->sc_sha1, &sha1ctx, sizeof(sha1ctx)); 317 318 SHA1Update(&sha1ctx, (void *)counter, sizeof(sc->sc_counter)); 319 SHA1Final(md, &sha1ctx); 320 321 /* outer hash */ 322 SHA1Init(&sha1ctx); 323 SHA1Update(&sha1ctx, sc->sc_pad, sizeof(sc->sc_pad)); 324 SHA1Update(&sha1ctx, md, 20); 325 SHA1Final(md, &sha1ctx); 326 } 327 328 int 329 carp_hmac_verify(struct carp_softc *sc, u_int32_t counter[2], 330 unsigned char md[20]) 331 { 332 unsigned char md2[20]; 333 334 carp_hmac_generate(sc, counter, md2); 335 336 return (bcmp(md, md2, sizeof(md2))); 337 } 338 339 void 340 carp_setroute(struct carp_softc *sc, int cmd) 341 { 342 struct ifaddr *ifa; 343 int s; 344 345 s = splsoftnet(); 346 IFADDR_FOREACH(ifa, &sc->sc_if) { 347 switch (ifa->ifa_addr->sa_family) { 348 case AF_INET: { 349 int count = 0; 350 struct rtentry *rt; 351 int hr_otherif, nr_ourif; 352 353 /* 354 * Avoid screwing with the routes if there are other 355 * carp interfaces which are master and have the same 356 * address. 357 */ 358 if (sc->sc_carpdev != NULL && 359 sc->sc_carpdev->if_carp != NULL) { 360 count = carp_addrcount( 361 (struct carp_if *)sc->sc_carpdev->if_carp, 362 ifatoia(ifa), CARP_COUNT_MASTER); 363 if ((cmd == RTM_ADD && count != 1) || 364 (cmd == RTM_DELETE && count != 0)) 365 continue; 366 } 367 368 /* Remove the existing host route, if any */ 369 rtrequest(RTM_DELETE, ifa->ifa_addr, 370 ifa->ifa_addr, ifa->ifa_netmask, 371 RTF_HOST, NULL); 372 373 rt = NULL; 374 (void)rtrequest(RTM_GET, ifa->ifa_addr, ifa->ifa_addr, 375 ifa->ifa_netmask, RTF_HOST, &rt); 376 hr_otherif = (rt && rt->rt_ifp != &sc->sc_if && 377 rt->rt_flags & (RTF_CLONING|RTF_CLONED)); 378 if (rt != NULL) { 379 RTFREE(rt); 380 rt = NULL; 381 } 382 383 /* Check for a network route on our interface */ 384 385 rt = NULL; 386 (void)rtrequest(RTM_GET, ifa->ifa_addr, ifa->ifa_addr, 387 ifa->ifa_netmask, 0, &rt); 388 nr_ourif = (rt && rt->rt_ifp == &sc->sc_if); 389 390 switch (cmd) { 391 case RTM_ADD: 392 if (hr_otherif) { 393 ifa->ifa_rtrequest = NULL; 394 ifa->ifa_flags &= ~RTF_CLONING; 395 396 rtrequest(RTM_ADD, ifa->ifa_addr, 397 ifa->ifa_addr, ifa->ifa_netmask, 398 RTF_UP | RTF_HOST, NULL); 399 } 400 if (!hr_otherif || nr_ourif || !rt) { 401 if (nr_ourif && !(rt->rt_flags & 402 RTF_CLONING)) 403 rtrequest(RTM_DELETE, 404 ifa->ifa_addr, 405 ifa->ifa_addr, 406 ifa->ifa_netmask, 0, NULL); 407 408 ifa->ifa_rtrequest = arp_rtrequest; 409 ifa->ifa_flags |= RTF_CLONING; 410 411 if (rtrequest(RTM_ADD, ifa->ifa_addr, 412 ifa->ifa_addr, ifa->ifa_netmask, 0, 413 NULL) == 0) 414 ifa->ifa_flags |= IFA_ROUTE; 415 } 416 break; 417 case RTM_DELETE: 418 break; 419 default: 420 break; 421 } 422 if (rt != NULL) { 423 RTFREE(rt); 424 rt = NULL; 425 } 426 break; 427 } 428 429 #ifdef INET6 430 case AF_INET6: 431 if (cmd == RTM_ADD) 432 in6_ifaddloop(ifa); 433 else 434 in6_ifremloop(ifa); 435 break; 436 #endif /* INET6 */ 437 default: 438 break; 439 } 440 } 441 splx(s); 442 } 443 444 /* 445 * process input packet. 446 * we have rearranged checks order compared to the rfc, 447 * but it seems more efficient this way or not possible otherwise. 448 */ 449 void 450 carp_proto_input(struct mbuf *m, ...) 451 { 452 struct ip *ip = mtod(m, struct ip *); 453 struct carp_softc *sc = NULL; 454 struct carp_header *ch; 455 int iplen, len, hlen; 456 va_list ap; 457 458 va_start(ap, m); 459 hlen = va_arg(ap, int); 460 va_end(ap); 461 462 carpstats.carps_ipackets++; 463 464 if (!carp_opts[CARPCTL_ALLOW]) { 465 m_freem(m); 466 return; 467 } 468 469 /* check if received on a valid carp interface */ 470 if (m->m_pkthdr.rcvif->if_type != IFT_CARP) { 471 carpstats.carps_badif++; 472 CARP_LOG(sc, ("packet received on non-carp interface: %s", 473 m->m_pkthdr.rcvif->if_xname)); 474 m_freem(m); 475 return; 476 } 477 478 /* verify that the IP TTL is 255. */ 479 if (ip->ip_ttl != CARP_DFLTTL) { 480 carpstats.carps_badttl++; 481 CARP_LOG(sc, ("received ttl %d != %d on %s", ip->ip_ttl, 482 CARP_DFLTTL, m->m_pkthdr.rcvif->if_xname)); 483 m_freem(m); 484 return; 485 } 486 487 /* 488 * verify that the received packet length is 489 * equal to the CARP header 490 */ 491 iplen = ip->ip_hl << 2; 492 len = iplen + sizeof(*ch); 493 if (len > m->m_pkthdr.len) { 494 carpstats.carps_badlen++; 495 CARP_LOG(sc, ("packet too short %d on %s", m->m_pkthdr.len, 496 m->m_pkthdr.rcvif->if_xname)); 497 m_freem(m); 498 return; 499 } 500 501 if ((m = m_pullup(m, len)) == NULL) { 502 carpstats.carps_hdrops++; 503 return; 504 } 505 ip = mtod(m, struct ip *); 506 ch = (struct carp_header *)((char *)ip + iplen); 507 /* verify the CARP checksum */ 508 m->m_data += iplen; 509 if (carp_cksum(m, len - iplen)) { 510 carpstats.carps_badsum++; 511 CARP_LOG(sc, ("checksum failed on %s", 512 m->m_pkthdr.rcvif->if_xname)); 513 m_freem(m); 514 return; 515 } 516 m->m_data -= iplen; 517 518 carp_proto_input_c(m, ch, AF_INET); 519 } 520 521 #ifdef INET6 522 int 523 carp6_proto_input(struct mbuf **mp, int *offp, int proto) 524 { 525 struct mbuf *m = *mp; 526 struct carp_softc *sc = NULL; 527 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 528 struct carp_header *ch; 529 u_int len; 530 531 carpstats.carps_ipackets6++; 532 533 if (!carp_opts[CARPCTL_ALLOW]) { 534 m_freem(m); 535 return (IPPROTO_DONE); 536 } 537 538 /* check if received on a valid carp interface */ 539 if (m->m_pkthdr.rcvif->if_type != IFT_CARP) { 540 carpstats.carps_badif++; 541 CARP_LOG(sc, ("packet received on non-carp interface: %s", 542 m->m_pkthdr.rcvif->if_xname)); 543 m_freem(m); 544 return (IPPROTO_DONE); 545 } 546 547 /* verify that the IP TTL is 255 */ 548 if (ip6->ip6_hlim != CARP_DFLTTL) { 549 carpstats.carps_badttl++; 550 CARP_LOG(sc, ("received ttl %d != %d on %s", ip6->ip6_hlim, 551 CARP_DFLTTL, m->m_pkthdr.rcvif->if_xname)); 552 m_freem(m); 553 return (IPPROTO_DONE); 554 } 555 556 /* verify that we have a complete carp packet */ 557 len = m->m_len; 558 IP6_EXTHDR_GET(ch, struct carp_header *, m, *offp, sizeof(*ch)); 559 if (ch == NULL) { 560 carpstats.carps_badlen++; 561 CARP_LOG(sc, ("packet size %u too small", len)); 562 return (IPPROTO_DONE); 563 } 564 565 566 /* verify the CARP checksum */ 567 m->m_data += *offp; 568 if (carp_cksum(m, sizeof(*ch))) { 569 carpstats.carps_badsum++; 570 CARP_LOG(sc, ("checksum failed, on %s", 571 m->m_pkthdr.rcvif->if_xname)); 572 m_freem(m); 573 return (IPPROTO_DONE); 574 } 575 m->m_data -= *offp; 576 577 carp_proto_input_c(m, ch, AF_INET6); 578 return (IPPROTO_DONE); 579 } 580 #endif /* INET6 */ 581 582 void 583 carp_proto_input_c(struct mbuf *m, struct carp_header *ch, sa_family_t af) 584 { 585 struct carp_softc *sc; 586 u_int64_t tmp_counter; 587 struct timeval sc_tv, ch_tv; 588 589 TAILQ_FOREACH(sc, &((struct carp_if *) 590 m->m_pkthdr.rcvif->if_carpdev->if_carp)->vhif_vrs, sc_list) 591 if (sc->sc_vhid == ch->carp_vhid) 592 break; 593 594 if (!sc || (sc->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) != 595 (IFF_UP|IFF_RUNNING)) { 596 carpstats.carps_badvhid++; 597 m_freem(m); 598 return; 599 } 600 601 /* 602 * Check if our own advertisement was duplicated 603 * from a non simplex interface. 604 * XXX If there is no address on our physical interface 605 * there is no way to distinguish our ads from the ones 606 * another carp host might have sent us. 607 */ 608 if ((sc->sc_carpdev->if_flags & IFF_SIMPLEX) == 0) { 609 struct sockaddr sa; 610 struct ifaddr *ifa; 611 612 bzero(&sa, sizeof(sa)); 613 sa.sa_family = af; 614 ifa = ifaof_ifpforaddr(&sa, sc->sc_carpdev); 615 616 if (ifa && af == AF_INET) { 617 struct ip *ip = mtod(m, struct ip *); 618 if (ip->ip_src.s_addr == 619 ifatoia(ifa)->ia_addr.sin_addr.s_addr) { 620 m_freem(m); 621 return; 622 } 623 } 624 #ifdef INET6 625 if (ifa && af == AF_INET6) { 626 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *); 627 struct in6_addr in6_src, in6_found; 628 629 in6_src = ip6->ip6_src; 630 in6_found = ifatoia6(ifa)->ia_addr.sin6_addr; 631 if (IN6_IS_ADDR_LINKLOCAL(&in6_src)) 632 in6_src.s6_addr16[1] = 0; 633 if (IN6_IS_ADDR_LINKLOCAL(&in6_found)) 634 in6_found.s6_addr16[1] = 0; 635 if (IN6_ARE_ADDR_EQUAL(&in6_src, &in6_found)) { 636 m_freem(m); 637 return; 638 } 639 } 640 #endif /* INET6 */ 641 } 642 643 microtime(&sc->sc_if.if_lastchange); 644 sc->sc_if.if_ipackets++; 645 sc->sc_if.if_ibytes += m->m_pkthdr.len; 646 647 /* verify the CARP version. */ 648 if (ch->carp_version != CARP_VERSION) { 649 carpstats.carps_badver++; 650 sc->sc_if.if_ierrors++; 651 CARP_LOG(sc, ("invalid version %d != %d", 652 ch->carp_version, CARP_VERSION)); 653 m_freem(m); 654 return; 655 } 656 657 /* verify the hash */ 658 if (carp_hmac_verify(sc, ch->carp_counter, ch->carp_md)) { 659 carpstats.carps_badauth++; 660 sc->sc_if.if_ierrors++; 661 CARP_LOG(sc, ("incorrect hash")); 662 m_freem(m); 663 return; 664 } 665 666 tmp_counter = ntohl(ch->carp_counter[0]); 667 tmp_counter = tmp_counter<<32; 668 tmp_counter += ntohl(ch->carp_counter[1]); 669 670 /* XXX Replay protection goes here */ 671 672 sc->sc_init_counter = 0; 673 sc->sc_counter = tmp_counter; 674 675 676 sc_tv.tv_sec = sc->sc_advbase; 677 if (carp_suppress_preempt && sc->sc_advskew < 240) 678 sc_tv.tv_usec = 240 * 1000000 / 256; 679 else 680 sc_tv.tv_usec = sc->sc_advskew * 1000000 / 256; 681 ch_tv.tv_sec = ch->carp_advbase; 682 ch_tv.tv_usec = ch->carp_advskew * 1000000 / 256; 683 684 switch (sc->sc_state) { 685 case INIT: 686 break; 687 case MASTER: 688 /* 689 * If we receive an advertisement from a backup who's going to 690 * be more frequent than us, go into BACKUP state. 691 */ 692 if (timercmp(&sc_tv, &ch_tv, >) || 693 timercmp(&sc_tv, &ch_tv, ==)) { 694 callout_stop(&sc->sc_ad_tmo); 695 carp_set_state(sc, BACKUP); 696 carp_setrun(sc, 0); 697 carp_setroute(sc, RTM_DELETE); 698 } 699 break; 700 case BACKUP: 701 /* 702 * If we're pre-empting masters who advertise slower than us, 703 * and this one claims to be slower, treat him as down. 704 */ 705 if (carp_opts[CARPCTL_PREEMPT] && timercmp(&sc_tv, &ch_tv, <)) { 706 carp_master_down(sc); 707 break; 708 } 709 710 /* 711 * If the master is going to advertise at such a low frequency 712 * that he's guaranteed to time out, we'd might as well just 713 * treat him as timed out now. 714 */ 715 sc_tv.tv_sec = sc->sc_advbase * 3; 716 if (timercmp(&sc_tv, &ch_tv, <)) { 717 carp_master_down(sc); 718 break; 719 } 720 721 /* 722 * Otherwise, we reset the counter and wait for the next 723 * advertisement. 724 */ 725 carp_setrun(sc, af); 726 break; 727 } 728 729 m_freem(m); 730 return; 731 } 732 733 /* 734 * Interface side of the CARP implementation. 735 */ 736 737 /* ARGSUSED */ 738 void 739 carpattach(int n) 740 { 741 if_clone_attach(&carp_cloner); 742 } 743 744 int 745 carp_clone_create(struct if_clone *ifc, int unit) 746 { 747 extern int ifqmaxlen; 748 struct carp_softc *sc; 749 struct ifnet *ifp; 750 751 sc = malloc(sizeof(*sc), M_DEVBUF, M_NOWAIT); 752 if (!sc) 753 return (ENOMEM); 754 bzero(sc, sizeof(*sc)); 755 756 sc->sc_suppress = 0; 757 sc->sc_advbase = CARP_DFLTINTV; 758 sc->sc_vhid = -1; /* required setting */ 759 sc->sc_advskew = 0; 760 sc->sc_init_counter = 1; 761 sc->sc_naddrs = sc->sc_naddrs6 = 0; 762 #ifdef INET6 763 sc->sc_im6o.im6o_multicast_hlim = CARP_DFLTTL; 764 #endif /* INET6 */ 765 766 callout_init(&sc->sc_ad_tmo, 0); 767 callout_init(&sc->sc_md_tmo, 0); 768 callout_init(&sc->sc_md6_tmo, 0); 769 770 callout_setfunc(&sc->sc_ad_tmo, carp_send_ad, sc); 771 callout_setfunc(&sc->sc_md_tmo, carp_master_down, sc); 772 callout_setfunc(&sc->sc_md6_tmo, carp_master_down, sc); 773 774 LIST_INIT(&sc->carp_mc_listhead); 775 ifp = &sc->sc_if; 776 ifp->if_softc = sc; 777 snprintf(ifp->if_xname, sizeof ifp->if_xname, "%s%d", ifc->ifc_name, 778 unit); 779 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST; 780 ifp->if_ioctl = carp_ioctl; 781 ifp->if_start = carp_start; 782 ifp->if_output = carp_output; 783 ifp->if_type = IFT_CARP; 784 ifp->if_addrlen = ETHER_ADDR_LEN; 785 ifp->if_hdrlen = ETHER_HDR_LEN; 786 ifp->if_mtu = ETHERMTU; 787 IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen); 788 IFQ_SET_READY(&ifp->if_snd); 789 if_attach(ifp); 790 791 if_alloc_sadl(ifp); 792 ifp->if_broadcastaddr = etherbroadcastaddr; 793 carp_set_enaddr(sc); 794 LIST_INIT(&sc->sc_ac.ec_multiaddrs); 795 #if NBPFILTER > 0 796 bpfattach(ifp, DLT_EN10MB, ETHER_HDR_LEN); 797 #endif 798 return (0); 799 } 800 801 int 802 carp_clone_destroy(struct ifnet *ifp) 803 { 804 struct carp_softc *sc = ifp->if_softc; 805 806 carpdetach(ifp->if_softc); 807 ether_ifdetach(ifp); 808 if_detach(ifp); 809 callout_destroy(&sc->sc_ad_tmo); 810 callout_destroy(&sc->sc_md_tmo); 811 callout_destroy(&sc->sc_md6_tmo); 812 free(ifp->if_softc, M_DEVBUF); 813 814 return (0); 815 } 816 817 void 818 carpdetach(struct carp_softc *sc) 819 { 820 struct carp_if *cif; 821 int s; 822 823 callout_stop(&sc->sc_ad_tmo); 824 callout_stop(&sc->sc_md_tmo); 825 callout_stop(&sc->sc_md6_tmo); 826 827 if (sc->sc_suppress) 828 carp_suppress_preempt--; 829 sc->sc_suppress = 0; 830 831 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) 832 carp_suppress_preempt--; 833 sc->sc_sendad_errors = 0; 834 835 carp_set_state(sc, INIT); 836 sc->sc_if.if_flags &= ~IFF_UP; 837 carp_setrun(sc, 0); 838 carp_multicast_cleanup(sc); 839 840 s = splnet(); 841 if (sc->sc_carpdev != NULL) { 842 /* XXX linkstatehook removal */ 843 cif = (struct carp_if *)sc->sc_carpdev->if_carp; 844 TAILQ_REMOVE(&cif->vhif_vrs, sc, sc_list); 845 if (!--cif->vhif_nvrs) { 846 ifpromisc(sc->sc_carpdev, 0); 847 sc->sc_carpdev->if_carp = NULL; 848 FREE(cif, M_IFADDR); 849 } 850 } 851 sc->sc_carpdev = NULL; 852 splx(s); 853 } 854 855 /* Detach an interface from the carp. */ 856 void 857 carp_ifdetach(struct ifnet *ifp) 858 { 859 struct carp_softc *sc, *nextsc; 860 struct carp_if *cif = (struct carp_if *)ifp->if_carp; 861 862 for (sc = TAILQ_FIRST(&cif->vhif_vrs); sc; sc = nextsc) { 863 nextsc = TAILQ_NEXT(sc, sc_list); 864 carpdetach(sc); 865 } 866 } 867 868 int 869 carp_prepare_ad(struct mbuf *m, struct carp_softc *sc, 870 struct carp_header *ch) 871 { 872 if (sc->sc_init_counter) { 873 /* this could also be seconds since unix epoch */ 874 sc->sc_counter = arc4random(); 875 sc->sc_counter = sc->sc_counter << 32; 876 sc->sc_counter += arc4random(); 877 } else 878 sc->sc_counter++; 879 880 ch->carp_counter[0] = htonl((sc->sc_counter>>32)&0xffffffff); 881 ch->carp_counter[1] = htonl(sc->sc_counter&0xffffffff); 882 883 carp_hmac_generate(sc, ch->carp_counter, ch->carp_md); 884 885 return (0); 886 } 887 888 void 889 carp_send_ad_all(void) 890 { 891 struct ifnet *ifp; 892 struct carp_if *cif; 893 struct carp_softc *vh; 894 895 TAILQ_FOREACH(ifp, &ifnet, if_list) { 896 if (ifp->if_carp == NULL || ifp->if_type == IFT_CARP) 897 continue; 898 899 cif = (struct carp_if *)ifp->if_carp; 900 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 901 if ((vh->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) == 902 (IFF_UP|IFF_RUNNING) && vh->sc_state == MASTER) 903 carp_send_ad(vh); 904 } 905 } 906 } 907 908 909 void 910 carp_send_ad(void *v) 911 { 912 struct carp_header ch; 913 struct timeval tv; 914 struct carp_softc *sc = v; 915 struct carp_header *ch_ptr; 916 struct mbuf *m; 917 int error, len, advbase, advskew, s; 918 struct ifaddr *ifa; 919 struct sockaddr sa; 920 921 s = splsoftnet(); 922 923 advbase = advskew = 0; /* Sssssh compiler */ 924 if (sc->sc_carpdev == NULL) { 925 sc->sc_if.if_oerrors++; 926 goto retry_later; 927 } 928 929 /* bow out if we've gone to backup (the carp interface is going down) */ 930 if (sc->sc_bow_out) { 931 sc->sc_bow_out = 0; 932 advbase = 255; 933 advskew = 255; 934 } else { 935 advbase = sc->sc_advbase; 936 if (!carp_suppress_preempt || sc->sc_advskew > 240) 937 advskew = sc->sc_advskew; 938 else 939 advskew = 240; 940 tv.tv_sec = advbase; 941 tv.tv_usec = advskew * 1000000 / 256; 942 } 943 944 ch.carp_version = CARP_VERSION; 945 ch.carp_type = CARP_ADVERTISEMENT; 946 ch.carp_vhid = sc->sc_vhid; 947 ch.carp_advbase = advbase; 948 ch.carp_advskew = advskew; 949 ch.carp_authlen = 7; /* XXX DEFINE */ 950 ch.carp_pad1 = 0; /* must be zero */ 951 ch.carp_cksum = 0; 952 953 954 #ifdef INET 955 if (sc->sc_naddrs) { 956 struct ip *ip; 957 958 MGETHDR(m, M_DONTWAIT, MT_HEADER); 959 if (m == NULL) { 960 sc->sc_if.if_oerrors++; 961 carpstats.carps_onomem++; 962 /* XXX maybe less ? */ 963 goto retry_later; 964 } 965 len = sizeof(*ip) + sizeof(ch); 966 m->m_pkthdr.len = len; 967 m->m_pkthdr.rcvif = NULL; 968 m->m_len = len; 969 MH_ALIGN(m, m->m_len); 970 m->m_flags |= M_MCAST; 971 ip = mtod(m, struct ip *); 972 ip->ip_v = IPVERSION; 973 ip->ip_hl = sizeof(*ip) >> 2; 974 ip->ip_tos = IPTOS_LOWDELAY; 975 ip->ip_len = htons(len); 976 ip->ip_id = 0; /* no need for id, we don't support fragments */ 977 ip->ip_off = htons(IP_DF); 978 ip->ip_ttl = CARP_DFLTTL; 979 ip->ip_p = IPPROTO_CARP; 980 ip->ip_sum = 0; 981 982 bzero(&sa, sizeof(sa)); 983 sa.sa_family = AF_INET; 984 ifa = ifaof_ifpforaddr(&sa, sc->sc_carpdev); 985 if (ifa == NULL) 986 ip->ip_src.s_addr = 0; 987 else 988 ip->ip_src.s_addr = 989 ifatoia(ifa)->ia_addr.sin_addr.s_addr; 990 ip->ip_dst.s_addr = INADDR_CARP_GROUP; 991 992 ch_ptr = (struct carp_header *)(&ip[1]); 993 bcopy(&ch, ch_ptr, sizeof(ch)); 994 if (carp_prepare_ad(m, sc, ch_ptr)) 995 goto retry_later; 996 997 m->m_data += sizeof(*ip); 998 ch_ptr->carp_cksum = carp_cksum(m, len - sizeof(*ip)); 999 m->m_data -= sizeof(*ip); 1000 1001 microtime(&sc->sc_if.if_lastchange); 1002 sc->sc_if.if_opackets++; 1003 sc->sc_if.if_obytes += len; 1004 carpstats.carps_opackets++; 1005 1006 error = ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo, 1007 NULL); 1008 if (error) { 1009 if (error == ENOBUFS) 1010 carpstats.carps_onomem++; 1011 else 1012 CARP_LOG(sc, ("ip_output failed: %d", error)); 1013 sc->sc_if.if_oerrors++; 1014 if (sc->sc_sendad_errors < INT_MAX) 1015 sc->sc_sendad_errors++; 1016 if (sc->sc_sendad_errors == CARP_SENDAD_MAX_ERRORS) { 1017 carp_suppress_preempt++; 1018 if (carp_suppress_preempt == 1) 1019 carp_send_ad_all(); 1020 } 1021 sc->sc_sendad_success = 0; 1022 } else { 1023 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) { 1024 if (++sc->sc_sendad_success >= 1025 CARP_SENDAD_MIN_SUCCESS) { 1026 carp_suppress_preempt--; 1027 sc->sc_sendad_errors = 0; 1028 } 1029 } else 1030 sc->sc_sendad_errors = 0; 1031 } 1032 } 1033 #endif /* INET */ 1034 #ifdef INET6 1035 if (sc->sc_naddrs6) { 1036 struct ip6_hdr *ip6; 1037 1038 MGETHDR(m, M_DONTWAIT, MT_HEADER); 1039 if (m == NULL) { 1040 sc->sc_if.if_oerrors++; 1041 carpstats.carps_onomem++; 1042 /* XXX maybe less ? */ 1043 goto retry_later; 1044 } 1045 len = sizeof(*ip6) + sizeof(ch); 1046 m->m_pkthdr.len = len; 1047 m->m_pkthdr.rcvif = NULL; 1048 m->m_len = len; 1049 MH_ALIGN(m, m->m_len); 1050 m->m_flags |= M_MCAST; 1051 ip6 = mtod(m, struct ip6_hdr *); 1052 bzero(ip6, sizeof(*ip6)); 1053 ip6->ip6_vfc |= IPV6_VERSION; 1054 ip6->ip6_hlim = CARP_DFLTTL; 1055 ip6->ip6_nxt = IPPROTO_CARP; 1056 1057 /* set the source address */ 1058 bzero(&sa, sizeof(sa)); 1059 sa.sa_family = AF_INET6; 1060 ifa = ifaof_ifpforaddr(&sa, sc->sc_carpdev); 1061 if (ifa == NULL) /* This should never happen with IPv6 */ 1062 bzero(&ip6->ip6_src, sizeof(struct in6_addr)); 1063 else 1064 bcopy(ifatoia6(ifa)->ia_addr.sin6_addr.s6_addr, 1065 &ip6->ip6_src, sizeof(struct in6_addr)); 1066 /* set the multicast destination */ 1067 1068 ip6->ip6_dst.s6_addr8[0] = 0xff; 1069 ip6->ip6_dst.s6_addr8[1] = 0x02; 1070 ip6->ip6_dst.s6_addr8[15] = 0x12; 1071 1072 ch_ptr = (struct carp_header *)(&ip6[1]); 1073 bcopy(&ch, ch_ptr, sizeof(ch)); 1074 if (carp_prepare_ad(m, sc, ch_ptr)) 1075 goto retry_later; 1076 1077 m->m_data += sizeof(*ip6); 1078 ch_ptr->carp_cksum = carp_cksum(m, len - sizeof(*ip6)); 1079 m->m_data -= sizeof(*ip6); 1080 1081 microtime(&sc->sc_if.if_lastchange); 1082 sc->sc_if.if_opackets++; 1083 sc->sc_if.if_obytes += len; 1084 carpstats.carps_opackets6++; 1085 1086 error = ip6_output(m, NULL, NULL, 0, &sc->sc_im6o, NULL, NULL); 1087 if (error) { 1088 if (error == ENOBUFS) 1089 carpstats.carps_onomem++; 1090 else 1091 CARP_LOG(sc, ("ip6_output failed: %d", error)); 1092 sc->sc_if.if_oerrors++; 1093 if (sc->sc_sendad_errors < INT_MAX) 1094 sc->sc_sendad_errors++; 1095 if (sc->sc_sendad_errors == CARP_SENDAD_MAX_ERRORS) { 1096 carp_suppress_preempt++; 1097 if (carp_suppress_preempt == 1) 1098 carp_send_ad_all(); 1099 } 1100 sc->sc_sendad_success = 0; 1101 } else { 1102 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) { 1103 if (++sc->sc_sendad_success >= 1104 CARP_SENDAD_MIN_SUCCESS) { 1105 carp_suppress_preempt--; 1106 sc->sc_sendad_errors = 0; 1107 } 1108 } else 1109 sc->sc_sendad_errors = 0; 1110 } 1111 } 1112 #endif /* INET6 */ 1113 1114 retry_later: 1115 splx(s); 1116 if (advbase != 255 || advskew != 255) 1117 callout_schedule(&sc->sc_ad_tmo, tvtohz(&tv)); 1118 } 1119 1120 /* 1121 * Broadcast a gratuitous ARP request containing 1122 * the virtual router MAC address for each IP address 1123 * associated with the virtual router. 1124 */ 1125 void 1126 carp_send_arp(struct carp_softc *sc) 1127 { 1128 struct ifaddr *ifa; 1129 struct in_addr *in; 1130 int s = splsoftnet(); 1131 1132 IFADDR_FOREACH(ifa, &sc->sc_if) { 1133 1134 if (ifa->ifa_addr->sa_family != AF_INET) 1135 continue; 1136 1137 in = &ifatoia(ifa)->ia_addr.sin_addr; 1138 arprequest(sc->sc_carpdev, in, in, CLLADDR(sc->sc_if.if_sadl)); 1139 DELAY(1000); /* XXX */ 1140 } 1141 splx(s); 1142 } 1143 1144 #ifdef INET6 1145 void 1146 carp_send_na(struct carp_softc *sc) 1147 { 1148 struct ifaddr *ifa; 1149 struct in6_addr *in6; 1150 static struct in6_addr mcast = IN6ADDR_LINKLOCAL_ALLNODES_INIT; 1151 int s = splsoftnet(); 1152 1153 IFADDR_FOREACH(ifa, &sc->sc_if) { 1154 1155 if (ifa->ifa_addr->sa_family != AF_INET6) 1156 continue; 1157 1158 in6 = &ifatoia6(ifa)->ia_addr.sin6_addr; 1159 nd6_na_output(sc->sc_carpdev, &mcast, in6, 1160 ND_NA_FLAG_OVERRIDE, 1, NULL); 1161 DELAY(1000); /* XXX */ 1162 } 1163 splx(s); 1164 } 1165 #endif /* INET6 */ 1166 1167 /* 1168 * Based on bridge_hash() in if_bridge.c 1169 */ 1170 #define mix(a,b,c) \ 1171 do { \ 1172 a -= b; a -= c; a ^= (c >> 13); \ 1173 b -= c; b -= a; b ^= (a << 8); \ 1174 c -= a; c -= b; c ^= (b >> 13); \ 1175 a -= b; a -= c; a ^= (c >> 12); \ 1176 b -= c; b -= a; b ^= (a << 16); \ 1177 c -= a; c -= b; c ^= (b >> 5); \ 1178 a -= b; a -= c; a ^= (c >> 3); \ 1179 b -= c; b -= a; b ^= (a << 10); \ 1180 c -= a; c -= b; c ^= (b >> 15); \ 1181 } while (0) 1182 1183 u_int32_t 1184 carp_hash(struct carp_softc *sc, u_char *src) 1185 { 1186 u_int32_t a = 0x9e3779b9, b = sc->sc_hashkey[0], c = sc->sc_hashkey[1]; 1187 1188 c += sc->sc_key[3] << 24; 1189 c += sc->sc_key[2] << 16; 1190 c += sc->sc_key[1] << 8; 1191 c += sc->sc_key[0]; 1192 b += src[5] << 8; 1193 b += src[4]; 1194 a += src[3] << 24; 1195 a += src[2] << 16; 1196 a += src[1] << 8; 1197 a += src[0]; 1198 1199 mix(a, b, c); 1200 return (c); 1201 } 1202 1203 int 1204 carp_addrcount(struct carp_if *cif, struct in_ifaddr *ia, int type) 1205 { 1206 struct carp_softc *vh; 1207 struct ifaddr *ifa; 1208 int count = 0; 1209 1210 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 1211 if ((type == CARP_COUNT_RUNNING && 1212 (vh->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) == 1213 (IFF_UP|IFF_RUNNING)) || 1214 (type == CARP_COUNT_MASTER && vh->sc_state == MASTER)) { 1215 IFADDR_FOREACH(ifa, &vh->sc_if) { 1216 if (ifa->ifa_addr->sa_family == AF_INET && 1217 ia->ia_addr.sin_addr.s_addr == 1218 ifatoia(ifa)->ia_addr.sin_addr.s_addr) 1219 count++; 1220 } 1221 } 1222 } 1223 return (count); 1224 } 1225 1226 int 1227 carp_iamatch(struct in_ifaddr *ia, u_char *src, 1228 u_int32_t *count, u_int32_t index) 1229 { 1230 struct carp_softc *sc = ia->ia_ifp->if_softc; 1231 1232 if (carp_opts[CARPCTL_ARPBALANCE]) { 1233 /* 1234 * We use the source ip to decide which virtual host should 1235 * handle the request. If we're master of that virtual host, 1236 * then we respond, otherwise, just drop the arp packet on 1237 * the floor. 1238 */ 1239 1240 /* Count the elegible carp interfaces with this address */ 1241 if (*count == 0) 1242 *count = carp_addrcount( 1243 (struct carp_if *)ia->ia_ifp->if_carpdev->if_carp, 1244 ia, CARP_COUNT_RUNNING); 1245 1246 /* This should never happen, but... */ 1247 if (*count == 0) 1248 return (0); 1249 1250 if (carp_hash(sc, src) % *count == index - 1 && 1251 sc->sc_state == MASTER) { 1252 return (1); 1253 } 1254 } else { 1255 if (sc->sc_state == MASTER) 1256 return (1); 1257 } 1258 1259 return (0); 1260 } 1261 1262 #ifdef INET6 1263 struct ifaddr * 1264 carp_iamatch6(void *v, struct in6_addr *taddr) 1265 { 1266 struct carp_if *cif = v; 1267 struct carp_softc *vh; 1268 struct ifaddr *ifa; 1269 1270 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 1271 IFADDR_FOREACH(ifa, &vh->sc_if) { 1272 if (IN6_ARE_ADDR_EQUAL(taddr, 1273 &ifatoia6(ifa)->ia_addr.sin6_addr) && 1274 ((vh->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) == 1275 (IFF_UP|IFF_RUNNING)) && vh->sc_state == MASTER) 1276 return (ifa); 1277 } 1278 } 1279 1280 return (NULL); 1281 } 1282 #endif /* INET6 */ 1283 1284 struct ifnet * 1285 carp_ourether(void *v, struct ether_header *eh, u_char iftype, int src) 1286 { 1287 struct carp_if *cif = (struct carp_if *)v; 1288 struct carp_softc *vh; 1289 u_int8_t *ena; 1290 1291 if (src) 1292 ena = (u_int8_t *)&eh->ether_shost; 1293 else 1294 ena = (u_int8_t *)&eh->ether_dhost; 1295 1296 switch (iftype) { 1297 case IFT_ETHER: 1298 case IFT_FDDI: 1299 if (ena[0] || ena[1] || ena[2] != 0x5e || ena[3] || ena[4] != 1) 1300 return (NULL); 1301 break; 1302 case IFT_ISO88025: 1303 if (ena[0] != 3 || ena[1] || ena[4] || ena[5]) 1304 return (NULL); 1305 break; 1306 default: 1307 return (NULL); 1308 break; 1309 } 1310 1311 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) 1312 if ((vh->sc_if.if_flags & (IFF_UP|IFF_RUNNING)) == 1313 (IFF_UP|IFF_RUNNING) && vh->sc_state == MASTER && 1314 !bcmp(ena, CLLADDR(vh->sc_if.if_sadl), 1315 ETHER_ADDR_LEN)) { 1316 return (&vh->sc_if); 1317 } 1318 1319 return (NULL); 1320 } 1321 1322 int 1323 carp_input(struct mbuf *m, u_int8_t *shost, u_int8_t *dhost, u_int16_t etype) 1324 { 1325 struct ether_header eh; 1326 struct carp_if *cif = (struct carp_if *)m->m_pkthdr.rcvif->if_carp; 1327 struct ifnet *ifp; 1328 1329 bcopy(shost, &eh.ether_shost, sizeof(eh.ether_shost)); 1330 bcopy(dhost, &eh.ether_dhost, sizeof(eh.ether_dhost)); 1331 eh.ether_type = etype; 1332 1333 if (m->m_flags & (M_BCAST|M_MCAST)) { 1334 struct carp_softc *vh; 1335 struct mbuf *m0; 1336 1337 /* 1338 * XXX Should really check the list of multicast addresses 1339 * for each CARP interface _before_ copying. 1340 */ 1341 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) { 1342 m0 = m_copym(m, 0, M_COPYALL, M_DONTWAIT); 1343 if (m0 == NULL) 1344 continue; 1345 m0->m_pkthdr.rcvif = &vh->sc_if; 1346 ether_input(&vh->sc_if, m0); 1347 } 1348 return (1); 1349 } 1350 1351 ifp = carp_ourether(cif, &eh, m->m_pkthdr.rcvif->if_type, 0); 1352 if (ifp == NULL) { 1353 return (1); 1354 } 1355 1356 m->m_pkthdr.rcvif = ifp; 1357 1358 #if NBPFILTER > 0 1359 if (ifp->if_bpf) 1360 bpf_mtap(ifp->if_bpf, m); 1361 #endif 1362 ifp->if_ipackets++; 1363 ether_input(ifp, m); 1364 return (0); 1365 } 1366 1367 void 1368 carp_master_down(void *v) 1369 { 1370 struct carp_softc *sc = v; 1371 1372 switch (sc->sc_state) { 1373 case INIT: 1374 printf("%s: master_down event in INIT state\n", 1375 sc->sc_if.if_xname); 1376 break; 1377 case MASTER: 1378 break; 1379 case BACKUP: 1380 carp_set_state(sc, MASTER); 1381 carp_send_ad(sc); 1382 carp_send_arp(sc); 1383 #ifdef INET6 1384 carp_send_na(sc); 1385 #endif /* INET6 */ 1386 carp_setrun(sc, 0); 1387 carp_setroute(sc, RTM_ADD); 1388 break; 1389 } 1390 } 1391 1392 /* 1393 * When in backup state, af indicates whether to reset the master down timer 1394 * for v4 or v6. If it's set to zero, reset the ones which are already pending. 1395 */ 1396 void 1397 carp_setrun(struct carp_softc *sc, sa_family_t af) 1398 { 1399 struct timeval tv; 1400 1401 if (sc->sc_carpdev == NULL) { 1402 sc->sc_if.if_flags &= ~IFF_RUNNING; 1403 carp_set_state(sc, INIT); 1404 return; 1405 } 1406 1407 if (sc->sc_if.if_flags & IFF_UP && sc->sc_vhid > 0 && 1408 (sc->sc_naddrs || sc->sc_naddrs6) && !sc->sc_suppress) { 1409 sc->sc_if.if_flags |= IFF_RUNNING; 1410 } else { 1411 sc->sc_if.if_flags &= ~IFF_RUNNING; 1412 carp_setroute(sc, RTM_DELETE); 1413 return; 1414 } 1415 1416 switch (sc->sc_state) { 1417 case INIT: 1418 carp_set_state(sc, BACKUP); 1419 carp_setroute(sc, RTM_DELETE); 1420 carp_setrun(sc, 0); 1421 break; 1422 case BACKUP: 1423 callout_stop(&sc->sc_ad_tmo); 1424 tv.tv_sec = 3 * sc->sc_advbase; 1425 tv.tv_usec = sc->sc_advskew * 1000000 / 256; 1426 switch (af) { 1427 #ifdef INET 1428 case AF_INET: 1429 callout_schedule(&sc->sc_md_tmo, tvtohz(&tv)); 1430 break; 1431 #endif /* INET */ 1432 #ifdef INET6 1433 case AF_INET6: 1434 callout_schedule(&sc->sc_md6_tmo, tvtohz(&tv)); 1435 break; 1436 #endif /* INET6 */ 1437 default: 1438 if (sc->sc_naddrs) 1439 callout_schedule(&sc->sc_md_tmo, tvtohz(&tv)); 1440 if (sc->sc_naddrs6) 1441 callout_schedule(&sc->sc_md6_tmo, tvtohz(&tv)); 1442 break; 1443 } 1444 break; 1445 case MASTER: 1446 tv.tv_sec = sc->sc_advbase; 1447 tv.tv_usec = sc->sc_advskew * 1000000 / 256; 1448 callout_schedule(&sc->sc_ad_tmo, tvtohz(&tv)); 1449 break; 1450 } 1451 } 1452 1453 void 1454 carp_multicast_cleanup(struct carp_softc *sc) 1455 { 1456 struct ip_moptions *imo = &sc->sc_imo; 1457 #ifdef INET6 1458 struct ip6_moptions *im6o = &sc->sc_im6o; 1459 #endif 1460 u_int16_t n = imo->imo_num_memberships; 1461 1462 /* Clean up our own multicast memberships */ 1463 while (n-- > 0) { 1464 if (imo->imo_membership[n] != NULL) { 1465 in_delmulti(imo->imo_membership[n]); 1466 imo->imo_membership[n] = NULL; 1467 } 1468 } 1469 imo->imo_num_memberships = 0; 1470 imo->imo_multicast_ifp = NULL; 1471 1472 #ifdef INET6 1473 while (!LIST_EMPTY(&im6o->im6o_memberships)) { 1474 struct in6_multi_mship *imm = 1475 LIST_FIRST(&im6o->im6o_memberships); 1476 1477 LIST_REMOVE(imm, i6mm_chain); 1478 in6_leavegroup(imm); 1479 } 1480 im6o->im6o_multicast_ifp = NULL; 1481 #endif 1482 1483 /* And any other multicast memberships */ 1484 carp_ether_purgemulti(sc); 1485 } 1486 1487 int 1488 carp_set_ifp(struct carp_softc *sc, struct ifnet *ifp) 1489 { 1490 struct carp_if *cif, *ncif = NULL; 1491 struct carp_softc *vr, *after = NULL; 1492 int myself = 0, error = 0; 1493 int s; 1494 1495 if (ifp == sc->sc_carpdev) 1496 return (0); 1497 1498 if (ifp != NULL) { 1499 if ((ifp->if_flags & IFF_MULTICAST) == 0) 1500 return (EADDRNOTAVAIL); 1501 1502 if (ifp->if_type == IFT_CARP) 1503 return (EINVAL); 1504 1505 if (ifp->if_carp == NULL) { 1506 MALLOC(ncif, struct carp_if *, sizeof(*cif), 1507 M_IFADDR, M_NOWAIT); 1508 if (ncif == NULL) 1509 return (ENOBUFS); 1510 if ((error = ifpromisc(ifp, 1))) { 1511 FREE(ncif, M_IFADDR); 1512 return (error); 1513 } 1514 1515 ncif->vhif_ifp = ifp; 1516 TAILQ_INIT(&ncif->vhif_vrs); 1517 } else { 1518 cif = (struct carp_if *)ifp->if_carp; 1519 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) 1520 if (vr != sc && vr->sc_vhid == sc->sc_vhid) 1521 return (EINVAL); 1522 } 1523 1524 /* detach from old interface */ 1525 if (sc->sc_carpdev != NULL) 1526 carpdetach(sc); 1527 1528 /* join multicast groups */ 1529 if (sc->sc_naddrs < 0 && 1530 (error = carp_join_multicast(sc)) != 0) { 1531 if (ncif != NULL) 1532 FREE(ncif, M_IFADDR); 1533 return (error); 1534 } 1535 1536 #ifdef INET6 1537 if (sc->sc_naddrs6 < 0 && 1538 (error = carp_join_multicast6(sc)) != 0) { 1539 if (ncif != NULL) 1540 FREE(ncif, M_IFADDR); 1541 carp_multicast_cleanup(sc); 1542 return (error); 1543 } 1544 #endif 1545 1546 /* attach carp interface to physical interface */ 1547 if (ncif != NULL) 1548 ifp->if_carp = (void *)ncif; 1549 sc->sc_carpdev = ifp; 1550 cif = (struct carp_if *)ifp->if_carp; 1551 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) { 1552 if (vr == sc) 1553 myself = 1; 1554 if (vr->sc_vhid < sc->sc_vhid) 1555 after = vr; 1556 } 1557 1558 if (!myself) { 1559 /* We're trying to keep things in order */ 1560 if (after == NULL) { 1561 TAILQ_INSERT_TAIL(&cif->vhif_vrs, sc, sc_list); 1562 } else { 1563 TAILQ_INSERT_AFTER(&cif->vhif_vrs, after, 1564 sc, sc_list); 1565 } 1566 cif->vhif_nvrs++; 1567 } 1568 if (sc->sc_naddrs || sc->sc_naddrs6) 1569 sc->sc_if.if_flags |= IFF_UP; 1570 carp_set_enaddr(sc); 1571 s = splnet(); 1572 /* XXX linkstatehooks establish */ 1573 carp_carpdev_state(ifp); 1574 splx(s); 1575 } else { 1576 carpdetach(sc); 1577 sc->sc_if.if_flags &= ~(IFF_UP|IFF_RUNNING); 1578 } 1579 return (0); 1580 } 1581 1582 void 1583 carp_set_enaddr(struct carp_softc *sc) 1584 { 1585 uint8_t enaddr[ETHER_ADDR_LEN]; 1586 if (sc->sc_carpdev && sc->sc_carpdev->if_type == IFT_ISO88025) { 1587 enaddr[0] = 3; 1588 enaddr[1] = 0; 1589 enaddr[2] = 0x40 >> (sc->sc_vhid - 1); 1590 enaddr[3] = 0x40000 >> (sc->sc_vhid - 1); 1591 enaddr[4] = 0; 1592 enaddr[5] = 0; 1593 } else { 1594 enaddr[0] = 0; 1595 enaddr[1] = 0; 1596 enaddr[2] = 0x5e; 1597 enaddr[3] = 0; 1598 enaddr[4] = 1; 1599 enaddr[5] = sc->sc_vhid; 1600 } 1601 if_set_sadl(&sc->sc_if, enaddr, sizeof(enaddr)); 1602 } 1603 1604 void 1605 carp_addr_updated(void *v) 1606 { 1607 struct carp_softc *sc = (struct carp_softc *) v; 1608 struct ifaddr *ifa; 1609 int new_naddrs = 0, new_naddrs6 = 0; 1610 1611 IFADDR_FOREACH(ifa, &sc->sc_if) { 1612 if (ifa->ifa_addr->sa_family == AF_INET) 1613 new_naddrs++; 1614 else if (ifa->ifa_addr->sa_family == AF_INET6) 1615 new_naddrs6++; 1616 } 1617 1618 /* Handle a callback after SIOCDIFADDR */ 1619 if (new_naddrs < sc->sc_naddrs || new_naddrs6 < sc->sc_naddrs6) { 1620 struct in_addr mc_addr; 1621 struct in_multi *inm; 1622 1623 sc->sc_naddrs = new_naddrs; 1624 sc->sc_naddrs6 = new_naddrs6; 1625 1626 /* Re-establish multicast membership removed by in_control */ 1627 mc_addr.s_addr = INADDR_CARP_GROUP; 1628 IN_LOOKUP_MULTI(mc_addr, &sc->sc_if, inm); 1629 if (inm == NULL) { 1630 bzero(&sc->sc_imo, sizeof(sc->sc_imo)); 1631 1632 if (sc->sc_carpdev != NULL && sc->sc_naddrs > 0) 1633 carp_join_multicast(sc); 1634 } 1635 1636 if (sc->sc_naddrs == 0 && sc->sc_naddrs6 == 0) { 1637 sc->sc_if.if_flags &= ~IFF_UP; 1638 carp_set_state(sc, INIT); 1639 } else 1640 carp_hmac_prepare(sc); 1641 } 1642 1643 carp_setrun(sc, 0); 1644 } 1645 1646 int 1647 carp_set_addr(struct carp_softc *sc, struct sockaddr_in *sin) 1648 { 1649 struct ifnet *ifp = sc->sc_carpdev; 1650 struct in_ifaddr *ia, *ia_if; 1651 int error = 0; 1652 1653 if (sin->sin_addr.s_addr == 0) { 1654 if (!(sc->sc_if.if_flags & IFF_UP)) 1655 carp_set_state(sc, INIT); 1656 if (sc->sc_naddrs) 1657 sc->sc_if.if_flags |= IFF_UP; 1658 carp_setrun(sc, 0); 1659 return (0); 1660 } 1661 1662 /* we have to do this by hand to ensure we don't match on ourselves */ 1663 ia_if = NULL; 1664 for (ia = TAILQ_FIRST(&in_ifaddrhead); ia; 1665 ia = TAILQ_NEXT(ia, ia_list)) { 1666 1667 /* and, yeah, we need a multicast-capable iface too */ 1668 if (ia->ia_ifp != &sc->sc_if && 1669 ia->ia_ifp->if_type != IFT_CARP && 1670 (ia->ia_ifp->if_flags & IFF_MULTICAST) && 1671 (sin->sin_addr.s_addr & ia->ia_subnetmask) == 1672 ia->ia_subnet) { 1673 if (!ia_if) 1674 ia_if = ia; 1675 } 1676 } 1677 1678 if (ia_if) { 1679 ia = ia_if; 1680 if (ifp) { 1681 if (ifp != ia->ia_ifp) 1682 return (EADDRNOTAVAIL); 1683 } else { 1684 ifp = ia->ia_ifp; 1685 } 1686 } 1687 1688 if ((error = carp_set_ifp(sc, ifp))) 1689 return (error); 1690 1691 if (sc->sc_carpdev == NULL) 1692 return (EADDRNOTAVAIL); 1693 1694 if (sc->sc_naddrs == 0 && (error = carp_join_multicast(sc)) != 0) 1695 return (error); 1696 1697 sc->sc_naddrs++; 1698 if (sc->sc_carpdev != NULL) 1699 sc->sc_if.if_flags |= IFF_UP; 1700 1701 carp_set_state(sc, INIT); 1702 carp_setrun(sc, 0); 1703 1704 /* 1705 * Hook if_addrhooks so that we get a callback after in_ifinit has run, 1706 * to correct any inappropriate routes that it inserted. 1707 */ 1708 if (sc->ah_cookie == 0) { 1709 /* XXX link address hook */ 1710 } 1711 1712 return (0); 1713 } 1714 1715 int 1716 carp_join_multicast(struct carp_softc *sc) 1717 { 1718 struct ip_moptions *imo = &sc->sc_imo, tmpimo; 1719 struct in_addr addr; 1720 1721 bzero(&tmpimo, sizeof(tmpimo)); 1722 addr.s_addr = INADDR_CARP_GROUP; 1723 if ((tmpimo.imo_membership[0] = 1724 in_addmulti(&addr, &sc->sc_if)) == NULL) { 1725 return (ENOBUFS); 1726 } 1727 1728 imo->imo_membership[0] = tmpimo.imo_membership[0]; 1729 imo->imo_num_memberships = 1; 1730 imo->imo_multicast_ifp = &sc->sc_if; 1731 imo->imo_multicast_ttl = CARP_DFLTTL; 1732 imo->imo_multicast_loop = 0; 1733 return (0); 1734 } 1735 1736 1737 #ifdef INET6 1738 int 1739 carp_set_addr6(struct carp_softc *sc, struct sockaddr_in6 *sin6) 1740 { 1741 struct ifnet *ifp = sc->sc_carpdev; 1742 struct in6_ifaddr *ia, *ia_if; 1743 int error = 0; 1744 1745 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) { 1746 if (!(sc->sc_if.if_flags & IFF_UP)) 1747 carp_set_state(sc, INIT); 1748 if (sc->sc_naddrs6) 1749 sc->sc_if.if_flags |= IFF_UP; 1750 carp_setrun(sc, 0); 1751 return (0); 1752 } 1753 1754 /* we have to do this by hand to ensure we don't match on ourselves */ 1755 ia_if = NULL; 1756 for (ia = in6_ifaddr; ia; ia = ia->ia_next) { 1757 int i; 1758 1759 for (i = 0; i < 4; i++) { 1760 if ((sin6->sin6_addr.s6_addr32[i] & 1761 ia->ia_prefixmask.sin6_addr.s6_addr32[i]) != 1762 (ia->ia_addr.sin6_addr.s6_addr32[i] & 1763 ia->ia_prefixmask.sin6_addr.s6_addr32[i])) 1764 break; 1765 } 1766 /* and, yeah, we need a multicast-capable iface too */ 1767 if (ia->ia_ifp != &sc->sc_if && 1768 ia->ia_ifp->if_type != IFT_CARP && 1769 (ia->ia_ifp->if_flags & IFF_MULTICAST) && 1770 (i == 4)) { 1771 if (!ia_if) 1772 ia_if = ia; 1773 } 1774 } 1775 1776 if (ia_if) { 1777 ia = ia_if; 1778 if (sc->sc_carpdev) { 1779 if (sc->sc_carpdev != ia->ia_ifp) 1780 return (EADDRNOTAVAIL); 1781 } else { 1782 ifp = ia->ia_ifp; 1783 } 1784 } 1785 1786 if ((error = carp_set_ifp(sc, ifp))) 1787 return (error); 1788 1789 if (sc->sc_carpdev == NULL) 1790 return (EADDRNOTAVAIL); 1791 1792 if (sc->sc_naddrs6 == 0 && (error = carp_join_multicast6(sc)) != 0) 1793 return (error); 1794 1795 sc->sc_naddrs6++; 1796 if (sc->sc_carpdev != NULL) 1797 sc->sc_if.if_flags |= IFF_UP; 1798 carp_set_state(sc, INIT); 1799 carp_setrun(sc, 0); 1800 1801 return (0); 1802 } 1803 1804 int 1805 carp_join_multicast6(struct carp_softc *sc) 1806 { 1807 struct in6_multi_mship *imm, *imm2; 1808 struct ip6_moptions *im6o = &sc->sc_im6o; 1809 struct sockaddr_in6 addr6; 1810 int error; 1811 1812 /* Join IPv6 CARP multicast group */ 1813 bzero(&addr6, sizeof(addr6)); 1814 addr6.sin6_family = AF_INET6; 1815 addr6.sin6_len = sizeof(addr6); 1816 addr6.sin6_addr.s6_addr16[0] = htons(0xff02); 1817 addr6.sin6_addr.s6_addr16[1] = htons(sc->sc_if.if_index); 1818 addr6.sin6_addr.s6_addr8[15] = 0x12; 1819 if ((imm = in6_joingroup(&sc->sc_if, 1820 &addr6.sin6_addr, &error, 0)) == NULL) { 1821 return (error); 1822 } 1823 /* join solicited multicast address */ 1824 bzero(&addr6.sin6_addr, sizeof(addr6.sin6_addr)); 1825 addr6.sin6_addr.s6_addr16[0] = htons(0xff02); 1826 addr6.sin6_addr.s6_addr16[1] = htons(sc->sc_if.if_index); 1827 addr6.sin6_addr.s6_addr32[1] = 0; 1828 addr6.sin6_addr.s6_addr32[2] = htonl(1); 1829 addr6.sin6_addr.s6_addr32[3] = 0; 1830 addr6.sin6_addr.s6_addr8[12] = 0xff; 1831 if ((imm2 = in6_joingroup(&sc->sc_if, 1832 &addr6.sin6_addr, &error, 0)) == NULL) { 1833 in6_leavegroup(imm); 1834 return (error); 1835 } 1836 1837 /* apply v6 multicast membership */ 1838 im6o->im6o_multicast_ifp = &sc->sc_if; 1839 if (imm) 1840 LIST_INSERT_HEAD(&im6o->im6o_memberships, imm, 1841 i6mm_chain); 1842 if (imm2) 1843 LIST_INSERT_HEAD(&im6o->im6o_memberships, imm2, 1844 i6mm_chain); 1845 1846 return (0); 1847 } 1848 1849 #endif /* INET6 */ 1850 1851 int 1852 carp_ioctl(struct ifnet *ifp, u_long cmd, void *addr) 1853 { 1854 struct lwp *l = curlwp; /* XXX */ 1855 struct carp_softc *sc = ifp->if_softc, *vr; 1856 struct carpreq carpr; 1857 struct ifaddr *ifa; 1858 struct ifreq *ifr; 1859 struct ifnet *cdev = NULL; 1860 int error = 0; 1861 1862 ifa = (struct ifaddr *)addr; 1863 ifr = (struct ifreq *)addr; 1864 1865 switch (cmd) { 1866 case SIOCSIFADDR: 1867 switch (ifa->ifa_addr->sa_family) { 1868 #ifdef INET 1869 case AF_INET: 1870 sc->sc_if.if_flags |= IFF_UP; 1871 bcopy(ifa->ifa_addr, ifa->ifa_dstaddr, 1872 sizeof(struct sockaddr)); 1873 error = carp_set_addr(sc, satosin(ifa->ifa_addr)); 1874 break; 1875 #endif /* INET */ 1876 #ifdef INET6 1877 case AF_INET6: 1878 sc->sc_if.if_flags|= IFF_UP; 1879 error = carp_set_addr6(sc, satosin6(ifa->ifa_addr)); 1880 break; 1881 #endif /* INET6 */ 1882 default: 1883 error = EAFNOSUPPORT; 1884 break; 1885 } 1886 break; 1887 1888 case SIOCSIFFLAGS: 1889 if (sc->sc_state != INIT && !(ifr->ifr_flags & IFF_UP)) { 1890 callout_stop(&sc->sc_ad_tmo); 1891 callout_stop(&sc->sc_md_tmo); 1892 callout_stop(&sc->sc_md6_tmo); 1893 if (sc->sc_state == MASTER) { 1894 /* we need the interface up to bow out */ 1895 sc->sc_if.if_flags |= IFF_UP; 1896 sc->sc_bow_out = 1; 1897 carp_send_ad(sc); 1898 } 1899 sc->sc_if.if_flags &= ~IFF_UP; 1900 carp_set_state(sc, INIT); 1901 carp_setrun(sc, 0); 1902 } else if (sc->sc_state == INIT && (ifr->ifr_flags & IFF_UP)) { 1903 sc->sc_if.if_flags |= IFF_UP; 1904 carp_setrun(sc, 0); 1905 } 1906 break; 1907 1908 case SIOCSVH: 1909 if (l == NULL) 1910 break; 1911 if ((error = kauth_authorize_network(l->l_cred, 1912 KAUTH_NETWORK_INTERFACE, 1913 KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd, 1914 NULL)) != 0) 1915 break; 1916 if ((error = copyin(ifr->ifr_data, &carpr, sizeof carpr))) 1917 break; 1918 error = 1; 1919 if (carpr.carpr_carpdev[0] != '\0' && 1920 (cdev = ifunit(carpr.carpr_carpdev)) == NULL) 1921 return (EINVAL); 1922 if ((error = carp_set_ifp(sc, cdev))) 1923 return (error); 1924 if (sc->sc_state != INIT && carpr.carpr_state != sc->sc_state) { 1925 switch (carpr.carpr_state) { 1926 case BACKUP: 1927 callout_stop(&sc->sc_ad_tmo); 1928 carp_set_state(sc, BACKUP); 1929 carp_setrun(sc, 0); 1930 carp_setroute(sc, RTM_DELETE); 1931 break; 1932 case MASTER: 1933 carp_master_down(sc); 1934 break; 1935 default: 1936 break; 1937 } 1938 } 1939 if (carpr.carpr_vhid > 0) { 1940 if (carpr.carpr_vhid > 255) { 1941 error = EINVAL; 1942 break; 1943 } 1944 if (sc->sc_carpdev) { 1945 struct carp_if *cif; 1946 cif = (struct carp_if *)sc->sc_carpdev->if_carp; 1947 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) 1948 if (vr != sc && 1949 vr->sc_vhid == carpr.carpr_vhid) 1950 return (EINVAL); 1951 } 1952 sc->sc_vhid = carpr.carpr_vhid; 1953 carp_set_enaddr(sc); 1954 carp_set_state(sc, INIT); 1955 error--; 1956 } 1957 if (carpr.carpr_advbase > 0 || carpr.carpr_advskew > 0) { 1958 if (carpr.carpr_advskew > 254) { 1959 error = EINVAL; 1960 break; 1961 } 1962 if (carpr.carpr_advbase > 255) { 1963 error = EINVAL; 1964 break; 1965 } 1966 sc->sc_advbase = carpr.carpr_advbase; 1967 sc->sc_advskew = carpr.carpr_advskew; 1968 error--; 1969 } 1970 bcopy(carpr.carpr_key, sc->sc_key, sizeof(sc->sc_key)); 1971 if (error > 0) 1972 error = EINVAL; 1973 else { 1974 error = 0; 1975 carp_setrun(sc, 0); 1976 } 1977 break; 1978 1979 case SIOCGVH: 1980 bzero(&carpr, sizeof(carpr)); 1981 if (sc->sc_carpdev != NULL) 1982 strlcpy(carpr.carpr_carpdev, sc->sc_carpdev->if_xname, 1983 IFNAMSIZ); 1984 carpr.carpr_state = sc->sc_state; 1985 carpr.carpr_vhid = sc->sc_vhid; 1986 carpr.carpr_advbase = sc->sc_advbase; 1987 carpr.carpr_advskew = sc->sc_advskew; 1988 1989 if ((l == NULL) || (error = kauth_authorize_network(l->l_cred, 1990 KAUTH_NETWORK_INTERFACE, 1991 KAUTH_REQ_NETWORK_INTERFACE_SETPRIV, ifp, (void *)cmd, 1992 NULL)) != 0) 1993 bcopy(sc->sc_key, carpr.carpr_key, 1994 sizeof(carpr.carpr_key)); 1995 error = copyout(&carpr, ifr->ifr_data, sizeof(carpr)); 1996 break; 1997 1998 case SIOCADDMULTI: 1999 error = carp_ether_addmulti(sc, ifr); 2000 break; 2001 2002 case SIOCDELMULTI: 2003 error = carp_ether_delmulti(sc, ifr); 2004 break; 2005 2006 default: 2007 error = EINVAL; 2008 } 2009 2010 carp_hmac_prepare(sc); 2011 return (error); 2012 } 2013 2014 2015 /* 2016 * Start output on carp interface. This function should never be called. 2017 */ 2018 void 2019 carp_start(struct ifnet *ifp) 2020 { 2021 #ifdef DEBUG 2022 printf("%s: start called\n", ifp->if_xname); 2023 #endif 2024 } 2025 2026 int 2027 carp_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *sa, 2028 struct rtentry *rt) 2029 { 2030 struct carp_softc *sc = ((struct carp_softc *)ifp->if_softc); 2031 2032 if (sc->sc_carpdev != NULL && sc->sc_state == MASTER) { 2033 return (sc->sc_carpdev->if_output(ifp, m, sa, rt)); 2034 } else { 2035 m_freem(m); 2036 return (ENETUNREACH); 2037 } 2038 } 2039 2040 void 2041 carp_set_state(struct carp_softc *sc, int state) 2042 { 2043 if (sc->sc_state == state) 2044 return; 2045 2046 sc->sc_state = state; 2047 switch (state) { 2048 case BACKUP: 2049 sc->sc_if.if_link_state = LINK_STATE_DOWN; 2050 break; 2051 case MASTER: 2052 sc->sc_if.if_link_state = LINK_STATE_UP; 2053 break; 2054 default: 2055 sc->sc_if.if_link_state = LINK_STATE_UNKNOWN; 2056 break; 2057 } 2058 rt_ifmsg(&sc->sc_if); 2059 } 2060 2061 void 2062 carp_carpdev_state(void *v) 2063 { 2064 struct carp_if *cif; 2065 struct carp_softc *sc; 2066 struct ifnet *ifp = v; 2067 2068 if (ifp->if_type == IFT_CARP) 2069 return; 2070 2071 cif = (struct carp_if *)ifp->if_carp; 2072 2073 TAILQ_FOREACH(sc, &cif->vhif_vrs, sc_list) { 2074 int suppressed = sc->sc_suppress; 2075 2076 if (sc->sc_carpdev->if_link_state == LINK_STATE_DOWN || 2077 !(sc->sc_carpdev->if_flags & IFF_UP)) { 2078 sc->sc_if.if_flags &= ~IFF_RUNNING; 2079 callout_stop(&sc->sc_ad_tmo); 2080 callout_stop(&sc->sc_md_tmo); 2081 callout_stop(&sc->sc_md6_tmo); 2082 carp_set_state(sc, INIT); 2083 sc->sc_suppress = 1; 2084 carp_setrun(sc, 0); 2085 if (!suppressed) { 2086 carp_suppress_preempt++; 2087 if (carp_suppress_preempt == 1) 2088 carp_send_ad_all(); 2089 } 2090 } else { 2091 carp_set_state(sc, INIT); 2092 sc->sc_suppress = 0; 2093 carp_setrun(sc, 0); 2094 if (suppressed) 2095 carp_suppress_preempt--; 2096 } 2097 } 2098 } 2099 2100 int 2101 carp_ether_addmulti(struct carp_softc *sc, struct ifreq *ifr) 2102 { 2103 const struct sockaddr *sa = ifreq_getaddr(SIOCADDMULTI, ifr); 2104 struct ifnet *ifp; 2105 struct carp_mc_entry *mc; 2106 u_int8_t addrlo[ETHER_ADDR_LEN], addrhi[ETHER_ADDR_LEN]; 2107 int error; 2108 2109 ifp = sc->sc_carpdev; 2110 if (ifp == NULL) 2111 return (EINVAL); 2112 2113 error = ether_addmulti(sa, &sc->sc_ac); 2114 if (error != ENETRESET) 2115 return (error); 2116 2117 /* 2118 * This is new multicast address. We have to tell parent 2119 * about it. Also, remember this multicast address so that 2120 * we can delete them on unconfigure. 2121 */ 2122 MALLOC(mc, struct carp_mc_entry *, sizeof(struct carp_mc_entry), 2123 M_DEVBUF, M_NOWAIT); 2124 if (mc == NULL) { 2125 error = ENOMEM; 2126 goto alloc_failed; 2127 } 2128 2129 /* 2130 * As ether_addmulti() returns ENETRESET, following two 2131 * statement shouldn't fail. 2132 */ 2133 (void)ether_multiaddr(sa, addrlo, addrhi); 2134 ETHER_LOOKUP_MULTI(addrlo, addrhi, &sc->sc_ac, mc->mc_enm); 2135 memcpy(&mc->mc_addr, sa, sa->sa_len); 2136 LIST_INSERT_HEAD(&sc->carp_mc_listhead, mc, mc_entries); 2137 2138 error = (*ifp->if_ioctl)(ifp, SIOCADDMULTI, (void *)ifr); 2139 if (error != 0) 2140 goto ioctl_failed; 2141 2142 return (error); 2143 2144 ioctl_failed: 2145 LIST_REMOVE(mc, mc_entries); 2146 FREE(mc, M_DEVBUF); 2147 alloc_failed: 2148 (void)ether_delmulti(sa, &sc->sc_ac); 2149 2150 return (error); 2151 } 2152 2153 int 2154 carp_ether_delmulti(struct carp_softc *sc, struct ifreq *ifr) 2155 { 2156 const struct sockaddr *sa = ifreq_getaddr(SIOCDELMULTI, ifr); 2157 struct ifnet *ifp; 2158 struct ether_multi *enm; 2159 struct carp_mc_entry *mc; 2160 u_int8_t addrlo[ETHER_ADDR_LEN], addrhi[ETHER_ADDR_LEN]; 2161 int error; 2162 2163 ifp = sc->sc_carpdev; 2164 if (ifp == NULL) 2165 return (EINVAL); 2166 2167 /* 2168 * Find a key to lookup carp_mc_entry. We have to do this 2169 * before calling ether_delmulti for obvious reason. 2170 */ 2171 if ((error = ether_multiaddr(sa, addrlo, addrhi)) != 0) 2172 return (error); 2173 ETHER_LOOKUP_MULTI(addrlo, addrhi, &sc->sc_ac, enm); 2174 if (enm == NULL) 2175 return (EINVAL); 2176 2177 LIST_FOREACH(mc, &sc->carp_mc_listhead, mc_entries) 2178 if (mc->mc_enm == enm) 2179 break; 2180 2181 /* We won't delete entries we didn't add */ 2182 if (mc == NULL) 2183 return (EINVAL); 2184 2185 error = ether_delmulti(sa, &sc->sc_ac); 2186 if (error != ENETRESET) 2187 return (error); 2188 2189 /* We no longer use this multicast address. Tell parent so. */ 2190 error = (*ifp->if_ioctl)(ifp, SIOCDELMULTI, (void *)ifr); 2191 if (error == 0) { 2192 /* And forget about this address. */ 2193 LIST_REMOVE(mc, mc_entries); 2194 FREE(mc, M_DEVBUF); 2195 } else 2196 (void)ether_addmulti(sa, &sc->sc_ac); 2197 return (error); 2198 } 2199 2200 /* 2201 * Delete any multicast address we have asked to add from parent 2202 * interface. Called when the carp is being unconfigured. 2203 */ 2204 void 2205 carp_ether_purgemulti(struct carp_softc *sc) 2206 { 2207 struct ifnet *ifp = sc->sc_carpdev; /* Parent. */ 2208 struct carp_mc_entry *mc; 2209 union { 2210 struct ifreq ifreq; 2211 struct { 2212 char ifr_name[IFNAMSIZ]; 2213 struct sockaddr_storage ifr_ss; 2214 } ifreq_storage; 2215 } u; 2216 struct ifreq *ifr = &u.ifreq; 2217 2218 if (ifp == NULL) 2219 return; 2220 2221 memcpy(ifr->ifr_name, ifp->if_xname, IFNAMSIZ); 2222 while ((mc = LIST_FIRST(&sc->carp_mc_listhead)) != NULL) { 2223 memcpy(&ifr->ifr_addr, &mc->mc_addr, mc->mc_addr.ss_len); 2224 (void)(*ifp->if_ioctl)(ifp, SIOCDELMULTI, (void *)ifr); 2225 LIST_REMOVE(mc, mc_entries); 2226 FREE(mc, M_DEVBUF); 2227 } 2228 } 2229 2230 SYSCTL_SETUP(sysctl_net_inet_carp_setup, "sysctl net.inet.carp subtree setup") 2231 { 2232 2233 sysctl_createv(clog, 0, NULL, NULL, 2234 CTLFLAG_PERMANENT, 2235 CTLTYPE_NODE, "net", NULL, 2236 NULL, 0, NULL, 0, 2237 CTL_NET, CTL_EOL); 2238 sysctl_createv(clog, 0, NULL, NULL, 2239 CTLFLAG_PERMANENT, 2240 CTLTYPE_NODE, "inet", NULL, 2241 NULL, 0, NULL, 0, 2242 CTL_NET, PF_INET, CTL_EOL); 2243 sysctl_createv(clog, 0, NULL, NULL, 2244 CTLFLAG_PERMANENT, 2245 CTLTYPE_NODE, "carp", 2246 SYSCTL_DESCR("CARP related settings"), 2247 NULL, 0, NULL, 0, 2248 CTL_NET, PF_INET, IPPROTO_CARP, CTL_EOL); 2249 2250 sysctl_createv(clog, 0, NULL, NULL, 2251 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2252 CTLTYPE_INT, "preempt", 2253 SYSCTL_DESCR("Enable CARP Preempt"), 2254 NULL, 0, &carp_opts[CARPCTL_PREEMPT], 0, 2255 CTL_NET, PF_INET, IPPROTO_CARP, 2256 CTL_CREATE, CTL_EOL); 2257 sysctl_createv(clog, 0, NULL, NULL, 2258 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2259 CTLTYPE_INT, "arpbalance", 2260 SYSCTL_DESCR("Enable ARP balancing"), 2261 NULL, 0, &carp_opts[CARPCTL_ARPBALANCE], 0, 2262 CTL_NET, PF_INET, IPPROTO_CARP, 2263 CTL_CREATE, CTL_EOL); 2264 sysctl_createv(clog, 0, NULL, NULL, 2265 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2266 CTLTYPE_INT, "allow", 2267 SYSCTL_DESCR("Enable CARP"), 2268 NULL, 0, &carp_opts[CARPCTL_ALLOW], 0, 2269 CTL_NET, PF_INET, IPPROTO_CARP, 2270 CTL_CREATE, CTL_EOL); 2271 sysctl_createv(clog, 0, NULL, NULL, 2272 CTLFLAG_PERMANENT|CTLFLAG_READWRITE, 2273 CTLTYPE_INT, "log", 2274 SYSCTL_DESCR("CARP logging"), 2275 NULL, 0, &carp_opts[CARPCTL_LOG], 0, 2276 CTL_NET, PF_INET, IPPROTO_CARP, 2277 CTL_CREATE, CTL_EOL); 2278 sysctl_createv(clog, 0, NULL, NULL, 2279 CTLFLAG_PERMANENT, 2280 CTLTYPE_STRUCT, "stats", 2281 SYSCTL_DESCR("CARP statistics"), 2282 NULL, 0, &carpstats, sizeof(carpstats), 2283 CTL_NET, PF_INET, IPPROTO_CARP, CARPCTL_STATS, 2284 CTL_EOL); 2285 } 2286
Indexes created Tue Sep 23 22:09:46 GMT 2025