2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2002 Michael Shalayeff.
5 * Copyright (c) 2003 Ryan McBride.
6 * Copyright (c) 2011 Gleb Smirnoff <glebius@FreeBSD.org>
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
19 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
20 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
21 * IN NO EVENT SHALL THE AUTHOR OR HIS RELATIVES BE LIABLE FOR ANY DIRECT,
22 * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
23 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
24 * SERVICES; LOSS OF MIND, USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
26 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
27 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
28 * THE POSSIBILITY OF SUCH DAMAGE.
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
36 #include "opt_inet6.h"
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/devctl.h>
42 #include <sys/kernel.h>
43 #include <sys/limits.h>
44 #include <sys/malloc.h>
46 #include <sys/module.h>
49 #include <sys/protosw.h>
50 #include <sys/socket.h>
51 #include <sys/sockio.h>
52 #include <sys/sysctl.h>
53 #include <sys/syslog.h>
54 #include <sys/taskqueue.h>
55 #include <sys/counter.h>
57 #include <net/ethernet.h>
59 #include <net/if_var.h>
60 #include <net/if_dl.h>
61 #include <net/if_llatbl.h>
62 #include <net/if_types.h>
63 #include <net/route.h>
66 #if defined(INET) || defined(INET6)
67 #include <netinet/in.h>
68 #include <netinet/in_var.h>
69 #include <netinet/ip_carp.h>
70 #include <netinet/ip.h>
71 #include <machine/in_cksum.h>
74 #include <netinet/ip_var.h>
75 #include <netinet/if_ether.h>
79 #include <netinet/icmp6.h>
80 #include <netinet/ip6.h>
81 #include <netinet6/in6_var.h>
82 #include <netinet6/ip6_var.h>
83 #include <netinet6/scope6_var.h>
84 #include <netinet6/nd6.h>
87 #include <crypto/sha1.h>
89 static MALLOC_DEFINE(M_CARP, "CARP", "CARP addresses");
92 struct ifnet *sc_carpdev; /* Pointer to parent ifnet. */
93 struct ifaddr **sc_ifas; /* Our ifaddrs. */
94 struct sockaddr_dl sc_addr; /* Our link level address. */
95 struct callout sc_ad_tmo; /* Advertising timeout. */
97 struct callout sc_md_tmo; /* Master down timeout. */
100 struct callout sc_md6_tmo; /* XXX: Master down timeout. */
111 enum { INIT = 0, BACKUP, MASTER } sc_state;
113 int sc_sendad_errors;
114 #define CARP_SENDAD_MAX_ERRORS 3
115 int sc_sendad_success;
116 #define CARP_SENDAD_MIN_SUCCESS 3
122 #define CARP_HMAC_PAD 64
123 unsigned char sc_key[CARP_KEY_LEN];
124 unsigned char sc_pad[CARP_HMAC_PAD];
127 TAILQ_ENTRY(carp_softc) sc_list; /* On the carp_if list. */
128 LIST_ENTRY(carp_softc) sc_next; /* On the global list. */
138 TAILQ_HEAD(, carp_softc) cif_vrs;
140 struct ip_moptions cif_imo;
143 struct ip6_moptions cif_im6o;
145 struct ifnet *cif_ifp;
148 #define CIF_PROMISC 0x00000001
153 static int proto_reg[] = {-1, -1};
156 * Brief design of carp(4).
158 * Any carp-capable ifnet may have a list of carp softcs hanging off
159 * its ifp->if_carp pointer. Each softc represents one unique virtual
160 * host id, or vhid. The softc has a back pointer to the ifnet. All
161 * softcs are joined in a global list, which has quite limited use.
163 * Any interface address that takes part in CARP negotiation has a
164 * pointer to the softc of its vhid, ifa->ifa_carp. That could be either
165 * AF_INET or AF_INET6 address.
167 * Although, one can get the softc's backpointer to ifnet and traverse
168 * through its ifp->if_addrhead queue to find all interface addresses
169 * involved in CARP, we keep a growable array of ifaddr pointers. This
170 * allows us to avoid grabbing the IF_ADDR_LOCK() in many traversals that
171 * do calls into the network stack, thus avoiding LORs.
175 * Each softc has a lock sc_mtx. It is used to synchronise carp_input_c(),
176 * callout-driven events and ioctl()s.
178 * To traverse the list of softcs on an ifnet we use CIF_LOCK() or carp_sx.
179 * To traverse the global list we use the mutex carp_mtx.
181 * Known issues with locking:
183 * - Sending ad, we put the pointer to the softc in an mtag, and no reference
184 * counting is done on the softc.
185 * - On module unload we may race (?) with packet processing thread
186 * dereferencing our function pointers.
189 /* Accept incoming CARP packets. */
190 VNET_DEFINE_STATIC(int, carp_allow) = 1;
191 #define V_carp_allow VNET(carp_allow)
193 /* Set DSCP in outgoing CARP packets. */
194 VNET_DEFINE_STATIC(int, carp_dscp) = 56;
195 #define V_carp_dscp VNET(carp_dscp)
197 /* Preempt slower nodes. */
198 VNET_DEFINE_STATIC(int, carp_preempt) = 0;
199 #define V_carp_preempt VNET(carp_preempt)
202 VNET_DEFINE_STATIC(int, carp_log) = 1;
203 #define V_carp_log VNET(carp_log)
205 /* Global advskew demotion. */
206 VNET_DEFINE_STATIC(int, carp_demotion) = 0;
207 #define V_carp_demotion VNET(carp_demotion)
209 /* Send error demotion factor. */
210 VNET_DEFINE_STATIC(int, carp_senderr_adj) = CARP_MAXSKEW;
211 #define V_carp_senderr_adj VNET(carp_senderr_adj)
213 /* Iface down demotion factor. */
214 VNET_DEFINE_STATIC(int, carp_ifdown_adj) = CARP_MAXSKEW;
215 #define V_carp_ifdown_adj VNET(carp_ifdown_adj)
217 static int carp_allow_sysctl(SYSCTL_HANDLER_ARGS);
218 static int carp_dscp_sysctl(SYSCTL_HANDLER_ARGS);
219 static int carp_demote_adj_sysctl(SYSCTL_HANDLER_ARGS);
221 SYSCTL_NODE(_net_inet, IPPROTO_CARP, carp, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
223 SYSCTL_PROC(_net_inet_carp, OID_AUTO, allow,
224 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
225 0, 0, carp_allow_sysctl, "I",
226 "Accept incoming CARP packets");
227 SYSCTL_PROC(_net_inet_carp, OID_AUTO, dscp,
228 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
229 0, 0, carp_dscp_sysctl, "I",
230 "DSCP value for carp packets");
231 SYSCTL_INT(_net_inet_carp, OID_AUTO, preempt, CTLFLAG_VNET | CTLFLAG_RW,
232 &VNET_NAME(carp_preempt), 0, "High-priority backup preemption mode");
233 SYSCTL_INT(_net_inet_carp, OID_AUTO, log, CTLFLAG_VNET | CTLFLAG_RW,
234 &VNET_NAME(carp_log), 0, "CARP log level");
235 SYSCTL_PROC(_net_inet_carp, OID_AUTO, demotion,
236 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
237 0, 0, carp_demote_adj_sysctl, "I",
238 "Adjust demotion factor (skew of advskew)");
239 SYSCTL_INT(_net_inet_carp, OID_AUTO, senderr_demotion_factor,
240 CTLFLAG_VNET | CTLFLAG_RW,
241 &VNET_NAME(carp_senderr_adj), 0, "Send error demotion factor adjustment");
242 SYSCTL_INT(_net_inet_carp, OID_AUTO, ifdown_demotion_factor,
243 CTLFLAG_VNET | CTLFLAG_RW,
244 &VNET_NAME(carp_ifdown_adj), 0,
245 "Interface down demotion factor adjustment");
247 VNET_PCPUSTAT_DEFINE(struct carpstats, carpstats);
248 VNET_PCPUSTAT_SYSINIT(carpstats);
249 VNET_PCPUSTAT_SYSUNINIT(carpstats);
251 #define CARPSTATS_ADD(name, val) \
252 counter_u64_add(VNET(carpstats)[offsetof(struct carpstats, name) / \
253 sizeof(uint64_t)], (val))
254 #define CARPSTATS_INC(name) CARPSTATS_ADD(name, 1)
256 SYSCTL_VNET_PCPUSTAT(_net_inet_carp, OID_AUTO, stats, struct carpstats,
257 carpstats, "CARP statistics (struct carpstats, netinet/ip_carp.h)");
259 #define CARP_LOCK_INIT(sc) mtx_init(&(sc)->sc_mtx, "carp_softc", \
261 #define CARP_LOCK_DESTROY(sc) mtx_destroy(&(sc)->sc_mtx)
262 #define CARP_LOCK_ASSERT(sc) mtx_assert(&(sc)->sc_mtx, MA_OWNED)
263 #define CARP_LOCK(sc) mtx_lock(&(sc)->sc_mtx)
264 #define CARP_UNLOCK(sc) mtx_unlock(&(sc)->sc_mtx)
265 #define CIF_LOCK_INIT(cif) mtx_init(&(cif)->cif_mtx, "carp_if", \
267 #define CIF_LOCK_DESTROY(cif) mtx_destroy(&(cif)->cif_mtx)
268 #define CIF_LOCK_ASSERT(cif) mtx_assert(&(cif)->cif_mtx, MA_OWNED)
269 #define CIF_LOCK(cif) mtx_lock(&(cif)->cif_mtx)
270 #define CIF_UNLOCK(cif) mtx_unlock(&(cif)->cif_mtx)
271 #define CIF_FREE(cif) do { \
273 if (TAILQ_EMPTY(&(cif)->cif_vrs)) \
279 #define CARP_LOG(...) do { \
280 if (V_carp_log > 0) \
281 log(LOG_INFO, "carp: " __VA_ARGS__); \
284 #define CARP_DEBUG(...) do { \
285 if (V_carp_log > 1) \
286 log(LOG_DEBUG, __VA_ARGS__); \
289 #define IFNET_FOREACH_IFA(ifp, ifa) \
290 CK_STAILQ_FOREACH((ifa), &(ifp)->if_addrhead, ifa_link) \
291 if ((ifa)->ifa_carp != NULL)
293 #define CARP_FOREACH_IFA(sc, ifa) \
294 CARP_LOCK_ASSERT(sc); \
296 _i < (sc)->sc_naddrs + (sc)->sc_naddrs6 && \
297 ((ifa) = sc->sc_ifas[_i]) != NULL; \
300 #define IFNET_FOREACH_CARP(ifp, sc) \
301 KASSERT(mtx_owned(&ifp->if_carp->cif_mtx) || \
302 sx_xlocked(&carp_sx), ("cif_vrs not locked")); \
303 TAILQ_FOREACH((sc), &(ifp)->if_carp->cif_vrs, sc_list)
305 #define DEMOTE_ADVSKEW(sc) \
306 (((sc)->sc_advskew + V_carp_demotion > CARP_MAXSKEW) ? \
308 (((sc)->sc_advskew + V_carp_demotion < 0) ? \
309 0 : ((sc)->sc_advskew + V_carp_demotion)))
311 static void carp_input_c(struct mbuf *, struct carp_header *, sa_family_t);
312 static struct carp_softc
313 *carp_alloc(struct ifnet *);
314 static void carp_destroy(struct carp_softc *);
315 static struct carp_if
316 *carp_alloc_if(struct ifnet *);
317 static void carp_free_if(struct carp_if *);
318 static void carp_set_state(struct carp_softc *, int, const char* reason);
319 static void carp_sc_state(struct carp_softc *);
320 static void carp_setrun(struct carp_softc *, sa_family_t);
321 static void carp_master_down(void *);
322 static void carp_master_down_locked(struct carp_softc *,
324 static void carp_send_ad(void *);
325 static void carp_send_ad_locked(struct carp_softc *);
326 static void carp_addroute(struct carp_softc *);
327 static void carp_ifa_addroute(struct ifaddr *);
328 static void carp_delroute(struct carp_softc *);
329 static void carp_ifa_delroute(struct ifaddr *);
330 static void carp_send_ad_all(void *, int);
331 static void carp_demote_adj(int, char *);
333 static LIST_HEAD(, carp_softc) carp_list;
334 static struct mtx carp_mtx;
335 static struct sx carp_sx;
336 static struct task carp_sendall_task =
337 TASK_INITIALIZER(0, carp_send_ad_all, NULL);
340 carp_hmac_prepare(struct carp_softc *sc)
342 uint8_t version = CARP_VERSION, type = CARP_ADVERTISEMENT;
343 uint8_t vhid = sc->sc_vhid & 0xff;
347 struct in_addr last, cur, in;
350 struct in6_addr last6, cur6, in6;
353 CARP_LOCK_ASSERT(sc);
355 /* Compute ipad from key. */
356 bzero(sc->sc_pad, sizeof(sc->sc_pad));
357 bcopy(sc->sc_key, sc->sc_pad, sizeof(sc->sc_key));
358 for (i = 0; i < sizeof(sc->sc_pad); i++)
359 sc->sc_pad[i] ^= 0x36;
361 /* Precompute first part of inner hash. */
362 SHA1Init(&sc->sc_sha1);
363 SHA1Update(&sc->sc_sha1, sc->sc_pad, sizeof(sc->sc_pad));
364 SHA1Update(&sc->sc_sha1, (void *)&version, sizeof(version));
365 SHA1Update(&sc->sc_sha1, (void *)&type, sizeof(type));
366 SHA1Update(&sc->sc_sha1, (void *)&vhid, sizeof(vhid));
372 cur.s_addr = 0xffffffff;
373 CARP_FOREACH_IFA(sc, ifa) {
374 in.s_addr = ifatoia(ifa)->ia_addr.sin_addr.s_addr;
375 if (ifa->ifa_addr->sa_family == AF_INET &&
376 ntohl(in.s_addr) > ntohl(last.s_addr) &&
377 ntohl(in.s_addr) < ntohl(cur.s_addr)) {
378 cur.s_addr = in.s_addr;
383 SHA1Update(&sc->sc_sha1, (void *)&cur, sizeof(cur));
387 memset(&cur6, 0, sizeof(cur6));
391 memset(&cur6, 0xff, sizeof(cur6));
392 CARP_FOREACH_IFA(sc, ifa) {
393 in6 = ifatoia6(ifa)->ia_addr.sin6_addr;
394 if (IN6_IS_SCOPE_EMBED(&in6))
395 in6.s6_addr16[1] = 0;
396 if (ifa->ifa_addr->sa_family == AF_INET6 &&
397 memcmp(&in6, &last6, sizeof(in6)) > 0 &&
398 memcmp(&in6, &cur6, sizeof(in6)) < 0) {
404 SHA1Update(&sc->sc_sha1, (void *)&cur6, sizeof(cur6));
408 /* convert ipad to opad */
409 for (i = 0; i < sizeof(sc->sc_pad); i++)
410 sc->sc_pad[i] ^= 0x36 ^ 0x5c;
414 carp_hmac_generate(struct carp_softc *sc, uint32_t counter[2],
415 unsigned char md[20])
419 CARP_LOCK_ASSERT(sc);
421 /* fetch first half of inner hash */
422 bcopy(&sc->sc_sha1, &sha1ctx, sizeof(sha1ctx));
424 SHA1Update(&sha1ctx, (void *)counter, sizeof(sc->sc_counter));
425 SHA1Final(md, &sha1ctx);
429 SHA1Update(&sha1ctx, sc->sc_pad, sizeof(sc->sc_pad));
430 SHA1Update(&sha1ctx, md, 20);
431 SHA1Final(md, &sha1ctx);
435 carp_hmac_verify(struct carp_softc *sc, uint32_t counter[2],
436 unsigned char md[20])
438 unsigned char md2[20];
440 CARP_LOCK_ASSERT(sc);
442 carp_hmac_generate(sc, counter, md2);
444 return (bcmp(md, md2, sizeof(md2)));
448 * process input packet.
449 * we have rearranged checks order compared to the rfc,
450 * but it seems more efficient this way or not possible otherwise.
454 carp_input(struct mbuf **mp, int *offp, int proto)
456 struct mbuf *m = *mp;
457 struct ip *ip = mtod(m, struct ip *);
458 struct carp_header *ch;
464 CARPSTATS_INC(carps_ipackets);
468 return (IPPROTO_DONE);
471 /* verify that the IP TTL is 255. */
472 if (ip->ip_ttl != CARP_DFLTTL) {
473 CARPSTATS_INC(carps_badttl);
474 CARP_DEBUG("%s: received ttl %d != 255 on %s\n", __func__,
476 m->m_pkthdr.rcvif->if_xname);
478 return (IPPROTO_DONE);
481 iplen = ip->ip_hl << 2;
483 if (m->m_pkthdr.len < iplen + sizeof(*ch)) {
484 CARPSTATS_INC(carps_badlen);
485 CARP_DEBUG("%s: received len %zd < sizeof(struct carp_header) "
486 "on %s\n", __func__, m->m_len - sizeof(struct ip),
487 m->m_pkthdr.rcvif->if_xname);
489 return (IPPROTO_DONE);
492 if (iplen + sizeof(*ch) < m->m_len) {
493 if ((m = m_pullup(m, iplen + sizeof(*ch))) == NULL) {
494 CARPSTATS_INC(carps_hdrops);
495 CARP_DEBUG("%s: pullup failed\n", __func__);
496 return (IPPROTO_DONE);
498 ip = mtod(m, struct ip *);
500 ch = (struct carp_header *)((char *)ip + iplen);
503 * verify that the received packet length is
504 * equal to the CARP header
506 len = iplen + sizeof(*ch);
507 if (len > m->m_pkthdr.len) {
508 CARPSTATS_INC(carps_badlen);
509 CARP_DEBUG("%s: packet too short %d on %s\n", __func__,
511 m->m_pkthdr.rcvif->if_xname);
513 return (IPPROTO_DONE);
516 if ((m = m_pullup(m, len)) == NULL) {
517 CARPSTATS_INC(carps_hdrops);
518 return (IPPROTO_DONE);
520 ip = mtod(m, struct ip *);
521 ch = (struct carp_header *)((char *)ip + iplen);
523 /* verify the CARP checksum */
525 if (in_cksum(m, len - iplen)) {
526 CARPSTATS_INC(carps_badsum);
527 CARP_DEBUG("%s: checksum failed on %s\n", __func__,
528 m->m_pkthdr.rcvif->if_xname);
530 return (IPPROTO_DONE);
534 carp_input_c(m, ch, AF_INET);
535 return (IPPROTO_DONE);
541 carp6_input(struct mbuf **mp, int *offp, int proto)
543 struct mbuf *m = *mp;
544 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
545 struct carp_header *ch;
548 CARPSTATS_INC(carps_ipackets6);
552 return (IPPROTO_DONE);
555 /* check if received on a valid carp interface */
556 if (m->m_pkthdr.rcvif->if_carp == NULL) {
557 CARPSTATS_INC(carps_badif);
558 CARP_DEBUG("%s: packet received on non-carp interface: %s\n",
559 __func__, m->m_pkthdr.rcvif->if_xname);
561 return (IPPROTO_DONE);
564 /* verify that the IP TTL is 255 */
565 if (ip6->ip6_hlim != CARP_DFLTTL) {
566 CARPSTATS_INC(carps_badttl);
567 CARP_DEBUG("%s: received ttl %d != 255 on %s\n", __func__,
568 ip6->ip6_hlim, m->m_pkthdr.rcvif->if_xname);
570 return (IPPROTO_DONE);
573 /* verify that we have a complete carp packet */
574 if (m->m_len < *offp + sizeof(*ch)) {
576 m = m_pullup(m, *offp + sizeof(*ch));
578 CARPSTATS_INC(carps_badlen);
579 CARP_DEBUG("%s: packet size %u too small\n", __func__, len);
580 return (IPPROTO_DONE);
583 ch = (struct carp_header *)(mtod(m, char *) + *offp);
585 /* verify the CARP checksum */
587 if (in_cksum(m, sizeof(*ch))) {
588 CARPSTATS_INC(carps_badsum);
589 CARP_DEBUG("%s: checksum failed, on %s\n", __func__,
590 m->m_pkthdr.rcvif->if_xname);
592 return (IPPROTO_DONE);
596 carp_input_c(m, ch, AF_INET6);
597 return (IPPROTO_DONE);
602 * This routine should not be necessary at all, but some switches
603 * (VMWare ESX vswitches) can echo our own packets back at us,
604 * and we must ignore them or they will cause us to drop out of
607 * We cannot catch all cases of network loops. Instead, what we
608 * do here is catch any packet that arrives with a carp header
609 * with a VHID of 0, that comes from an address that is our own.
610 * These packets are by definition "from us" (even if they are from
611 * a misconfigured host that is pretending to be us).
613 * The VHID test is outside this mini-function.
616 carp_source_is_self(struct mbuf *m, struct ifaddr *ifa, sa_family_t af)
630 ip4 = mtod(m, struct ip *);
631 in4 = ifatoia(ifa)->ia_addr.sin_addr;
632 return (in4.s_addr == ip4->ip_src.s_addr);
636 ip6 = mtod(m, struct ip6_hdr *);
637 in6 = ifatoia6(ifa)->ia_addr.sin6_addr;
638 return (memcmp(&in6, &ip6->ip6_src, sizeof(in6)) == 0);
647 carp_input_c(struct mbuf *m, struct carp_header *ch, sa_family_t af)
649 struct ifnet *ifp = m->m_pkthdr.rcvif;
650 struct ifaddr *ifa, *match;
651 struct carp_softc *sc;
652 uint64_t tmp_counter;
653 struct timeval sc_tv, ch_tv;
659 * Verify that the VHID is valid on the receiving interface.
661 * There should be just one match. If there are none
662 * the VHID is not valid and we drop the packet. If
663 * there are multiple VHID matches, take just the first
664 * one, for compatibility with previous code. While we're
665 * scanning, check for obvious loops in the network topology
666 * (these should never happen, and as noted above, we may
667 * miss real loops; this is just a double-check).
671 IFNET_FOREACH_IFA(ifp, ifa) {
672 if (match == NULL && ifa->ifa_carp != NULL &&
673 ifa->ifa_addr->sa_family == af &&
674 ifa->ifa_carp->sc_vhid == ch->carp_vhid)
676 if (ch->carp_vhid == 0 && carp_source_is_self(m, ifa, af))
679 ifa = error ? NULL : match;
684 if (error == ELOOP) {
685 CARP_DEBUG("dropping looped packet on interface %s\n",
687 CARPSTATS_INC(carps_badif); /* ??? */
689 CARPSTATS_INC(carps_badvhid);
695 /* verify the CARP version. */
696 if (ch->carp_version != CARP_VERSION) {
697 CARPSTATS_INC(carps_badver);
698 CARP_DEBUG("%s: invalid version %d\n", ifp->if_xname,
709 if (carp_hmac_verify(sc, ch->carp_counter, ch->carp_md)) {
710 CARPSTATS_INC(carps_badauth);
711 CARP_DEBUG("%s: incorrect hash for VHID %u@%s\n", __func__,
712 sc->sc_vhid, ifp->if_xname);
716 tmp_counter = ntohl(ch->carp_counter[0]);
717 tmp_counter = tmp_counter<<32;
718 tmp_counter += ntohl(ch->carp_counter[1]);
720 /* XXX Replay protection goes here */
722 sc->sc_init_counter = 0;
723 sc->sc_counter = tmp_counter;
725 sc_tv.tv_sec = sc->sc_advbase;
726 sc_tv.tv_usec = DEMOTE_ADVSKEW(sc) * 1000000 / 256;
727 ch_tv.tv_sec = ch->carp_advbase;
728 ch_tv.tv_usec = ch->carp_advskew * 1000000 / 256;
730 switch (sc->sc_state) {
735 * If we receive an advertisement from a master who's going to
736 * be more frequent than us, go into BACKUP state.
738 if (timevalcmp(&sc_tv, &ch_tv, >) ||
739 timevalcmp(&sc_tv, &ch_tv, ==)) {
740 callout_stop(&sc->sc_ad_tmo);
741 carp_set_state(sc, BACKUP,
742 "more frequent advertisement received");
749 * If we're pre-empting masters who advertise slower than us,
750 * and this one claims to be slower, treat him as down.
752 if (V_carp_preempt && timevalcmp(&sc_tv, &ch_tv, <)) {
753 carp_master_down_locked(sc,
754 "preempting a slower master");
759 * If the master is going to advertise at such a low frequency
760 * that he's guaranteed to time out, we'd might as well just
761 * treat him as timed out now.
763 sc_tv.tv_sec = sc->sc_advbase * 3;
764 if (timevalcmp(&sc_tv, &ch_tv, <)) {
765 carp_master_down_locked(sc, "master will time out");
770 * Otherwise, we reset the counter and wait for the next
783 carp_prepare_ad(struct mbuf *m, struct carp_softc *sc, struct carp_header *ch)
787 if (sc->sc_init_counter) {
788 /* this could also be seconds since unix epoch */
789 sc->sc_counter = arc4random();
790 sc->sc_counter = sc->sc_counter << 32;
791 sc->sc_counter += arc4random();
795 ch->carp_counter[0] = htonl((sc->sc_counter>>32)&0xffffffff);
796 ch->carp_counter[1] = htonl(sc->sc_counter&0xffffffff);
798 carp_hmac_generate(sc, ch->carp_counter, ch->carp_md);
800 /* Tag packet for carp_output */
801 if ((mtag = m_tag_get(PACKET_TAG_CARP, sizeof(struct carp_softc *),
802 M_NOWAIT)) == NULL) {
804 CARPSTATS_INC(carps_onomem);
807 bcopy(&sc, mtag + 1, sizeof(sc));
808 m_tag_prepend(m, mtag);
814 * To avoid LORs and possible recursions this function shouldn't
815 * be called directly, but scheduled via taskqueue.
818 carp_send_ad_all(void *ctx __unused, int pending __unused)
820 struct carp_softc *sc;
821 struct epoch_tracker et;
825 LIST_FOREACH(sc, &carp_list, sc_next)
826 if (sc->sc_state == MASTER) {
828 CURVNET_SET(sc->sc_carpdev->if_vnet);
829 carp_send_ad_locked(sc);
833 mtx_unlock(&carp_mtx);
837 /* Send a periodic advertisement, executed in callout context. */
839 carp_send_ad(void *v)
841 struct carp_softc *sc = v;
842 struct epoch_tracker et;
845 CARP_LOCK_ASSERT(sc);
846 CURVNET_SET(sc->sc_carpdev->if_vnet);
847 carp_send_ad_locked(sc);
854 carp_send_ad_error(struct carp_softc *sc, int error)
858 * We track errors and successfull sends with this logic:
859 * - Any error resets success counter to 0.
860 * - MAX_ERRORS triggers demotion.
861 * - MIN_SUCCESS successes resets error counter to 0.
862 * - MIN_SUCCESS reverts demotion, if it was triggered before.
865 if (sc->sc_sendad_errors < INT_MAX)
866 sc->sc_sendad_errors++;
867 if (sc->sc_sendad_errors == CARP_SENDAD_MAX_ERRORS) {
868 static const char fmt[] = "send error %d on %s";
869 char msg[sizeof(fmt) + IFNAMSIZ];
871 sprintf(msg, fmt, error, sc->sc_carpdev->if_xname);
872 carp_demote_adj(V_carp_senderr_adj, msg);
874 sc->sc_sendad_success = 0;
875 } else if (sc->sc_sendad_errors > 0) {
876 if (++sc->sc_sendad_success >= CARP_SENDAD_MIN_SUCCESS) {
877 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) {
878 static const char fmt[] = "send ok on %s";
879 char msg[sizeof(fmt) + IFNAMSIZ];
881 sprintf(msg, fmt, sc->sc_carpdev->if_xname);
882 carp_demote_adj(-V_carp_senderr_adj, msg);
884 sc->sc_sendad_errors = 0;
890 * Pick the best ifaddr on the given ifp for sending CARP
893 * "Best" here is defined by ifa_preferred(). This function is much
894 * much like ifaof_ifpforaddr() except that we just use ifa_preferred().
896 * (This could be simplified to return the actual address, except that
897 * it has a different format in AF_INET and AF_INET6.)
899 static struct ifaddr *
900 carp_best_ifa(int af, struct ifnet *ifp)
902 struct ifaddr *ifa, *best;
909 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
910 if (ifa->ifa_addr->sa_family == af &&
911 (best == NULL || ifa_preferred(best, ifa)))
920 carp_send_ad_locked(struct carp_softc *sc)
922 struct carp_header ch;
925 struct carp_header *ch_ptr;
930 CARP_LOCK_ASSERT(sc);
932 advskew = DEMOTE_ADVSKEW(sc);
933 tv.tv_sec = sc->sc_advbase;
934 tv.tv_usec = advskew * 1000000 / 256;
936 ch.carp_version = CARP_VERSION;
937 ch.carp_type = CARP_ADVERTISEMENT;
938 ch.carp_vhid = sc->sc_vhid;
939 ch.carp_advbase = sc->sc_advbase;
940 ch.carp_advskew = advskew;
941 ch.carp_authlen = 7; /* XXX DEFINE */
942 ch.carp_pad1 = 0; /* must be zero */
945 /* XXXGL: OpenBSD picks first ifaddr with needed family. */
951 m = m_gethdr(M_NOWAIT, MT_DATA);
953 CARPSTATS_INC(carps_onomem);
956 len = sizeof(*ip) + sizeof(ch);
957 m->m_pkthdr.len = len;
958 m->m_pkthdr.rcvif = NULL;
960 M_ALIGN(m, m->m_len);
961 m->m_flags |= M_MCAST;
962 ip = mtod(m, struct ip *);
963 ip->ip_v = IPVERSION;
964 ip->ip_hl = sizeof(*ip) >> 2;
965 ip->ip_tos = V_carp_dscp << IPTOS_DSCP_OFFSET;
966 ip->ip_len = htons(len);
967 ip->ip_off = htons(IP_DF);
968 ip->ip_ttl = CARP_DFLTTL;
969 ip->ip_p = IPPROTO_CARP;
973 ifa = carp_best_ifa(AF_INET, sc->sc_carpdev);
976 ifatoia(ifa)->ia_addr.sin_addr.s_addr;
979 ip->ip_src.s_addr = 0;
980 ip->ip_dst.s_addr = htonl(INADDR_CARP_GROUP);
982 ch_ptr = (struct carp_header *)(&ip[1]);
983 bcopy(&ch, ch_ptr, sizeof(ch));
984 if (carp_prepare_ad(m, sc, ch_ptr))
987 m->m_data += sizeof(*ip);
988 ch_ptr->carp_cksum = in_cksum(m, len - sizeof(*ip));
989 m->m_data -= sizeof(*ip);
991 CARPSTATS_INC(carps_opackets);
993 carp_send_ad_error(sc, ip_output(m, NULL, NULL, IP_RAWOUTPUT,
994 &sc->sc_carpdev->if_carp->cif_imo, NULL));
998 if (sc->sc_naddrs6) {
1001 m = m_gethdr(M_NOWAIT, MT_DATA);
1003 CARPSTATS_INC(carps_onomem);
1006 len = sizeof(*ip6) + sizeof(ch);
1007 m->m_pkthdr.len = len;
1008 m->m_pkthdr.rcvif = NULL;
1010 M_ALIGN(m, m->m_len);
1011 m->m_flags |= M_MCAST;
1012 ip6 = mtod(m, struct ip6_hdr *);
1013 bzero(ip6, sizeof(*ip6));
1014 ip6->ip6_vfc |= IPV6_VERSION;
1015 /* Traffic class isn't defined in ip6 struct instead
1016 * it gets offset into flowid field */
1017 ip6->ip6_flow |= htonl(V_carp_dscp << (IPV6_FLOWLABEL_LEN +
1018 IPTOS_DSCP_OFFSET));
1019 ip6->ip6_hlim = CARP_DFLTTL;
1020 ip6->ip6_nxt = IPPROTO_CARP;
1022 /* set the source address */
1023 ifa = carp_best_ifa(AF_INET6, sc->sc_carpdev);
1025 bcopy(IFA_IN6(ifa), &ip6->ip6_src,
1026 sizeof(struct in6_addr));
1029 /* This should never happen with IPv6. */
1030 bzero(&ip6->ip6_src, sizeof(struct in6_addr));
1032 /* Set the multicast destination. */
1033 ip6->ip6_dst.s6_addr16[0] = htons(0xff02);
1034 ip6->ip6_dst.s6_addr8[15] = 0x12;
1035 if (in6_setscope(&ip6->ip6_dst, sc->sc_carpdev, NULL) != 0) {
1037 CARP_DEBUG("%s: in6_setscope failed\n", __func__);
1041 ch_ptr = (struct carp_header *)(&ip6[1]);
1042 bcopy(&ch, ch_ptr, sizeof(ch));
1043 if (carp_prepare_ad(m, sc, ch_ptr))
1046 m->m_data += sizeof(*ip6);
1047 ch_ptr->carp_cksum = in_cksum(m, len - sizeof(*ip6));
1048 m->m_data -= sizeof(*ip6);
1050 CARPSTATS_INC(carps_opackets6);
1052 carp_send_ad_error(sc, ip6_output(m, NULL, NULL, 0,
1053 &sc->sc_carpdev->if_carp->cif_im6o, NULL, NULL));
1058 callout_reset(&sc->sc_ad_tmo, tvtohz(&tv), carp_send_ad, sc);
1062 carp_addroute(struct carp_softc *sc)
1066 CARP_FOREACH_IFA(sc, ifa)
1067 carp_ifa_addroute(ifa);
1071 carp_ifa_addroute(struct ifaddr *ifa)
1074 switch (ifa->ifa_addr->sa_family) {
1077 in_addprefix(ifatoia(ifa));
1078 ifa_add_loopback_route(ifa,
1079 (struct sockaddr *)&ifatoia(ifa)->ia_addr);
1084 ifa_add_loopback_route(ifa,
1085 (struct sockaddr *)&ifatoia6(ifa)->ia_addr);
1086 nd6_add_ifa_lle(ifatoia6(ifa));
1093 carp_delroute(struct carp_softc *sc)
1097 CARP_FOREACH_IFA(sc, ifa)
1098 carp_ifa_delroute(ifa);
1102 carp_ifa_delroute(struct ifaddr *ifa)
1105 switch (ifa->ifa_addr->sa_family) {
1108 ifa_del_loopback_route(ifa,
1109 (struct sockaddr *)&ifatoia(ifa)->ia_addr);
1110 in_scrubprefix(ifatoia(ifa), LLE_STATIC);
1115 ifa_del_loopback_route(ifa,
1116 (struct sockaddr *)&ifatoia6(ifa)->ia_addr);
1117 nd6_rem_ifa_lle(ifatoia6(ifa), 1);
1124 carp_master(struct ifaddr *ifa)
1126 struct carp_softc *sc = ifa->ifa_carp;
1128 return (sc->sc_state == MASTER);
1133 * Broadcast a gratuitous ARP request containing
1134 * the virtual router MAC address for each IP address
1135 * associated with the virtual router.
1138 carp_send_arp(struct carp_softc *sc)
1141 struct in_addr addr;
1145 CARP_FOREACH_IFA(sc, ifa) {
1146 if (ifa->ifa_addr->sa_family != AF_INET)
1148 addr = ((struct sockaddr_in *)ifa->ifa_addr)->sin_addr;
1149 arp_announce_ifaddr(sc->sc_carpdev, addr, LLADDR(&sc->sc_addr));
1154 carp_iamatch(struct ifaddr *ifa, uint8_t **enaddr)
1156 struct carp_softc *sc = ifa->ifa_carp;
1158 if (sc->sc_state == MASTER) {
1159 *enaddr = LLADDR(&sc->sc_addr);
1169 carp_send_na(struct carp_softc *sc)
1171 static struct in6_addr mcast = IN6ADDR_LINKLOCAL_ALLNODES_INIT;
1173 struct in6_addr *in6;
1175 CARP_FOREACH_IFA(sc, ifa) {
1176 if (ifa->ifa_addr->sa_family != AF_INET6)
1180 nd6_na_output(sc->sc_carpdev, &mcast, in6,
1181 ND_NA_FLAG_OVERRIDE, 1, NULL);
1182 DELAY(1000); /* XXX */
1187 * Returns ifa in case it's a carp address and it is MASTER, or if the address
1188 * matches and is not a carp address. Returns NULL otherwise.
1191 carp_iamatch6(struct ifnet *ifp, struct in6_addr *taddr)
1198 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1199 if (ifa->ifa_addr->sa_family != AF_INET6)
1201 if (!IN6_ARE_ADDR_EQUAL(taddr, IFA_IN6(ifa)))
1203 if (ifa->ifa_carp && ifa->ifa_carp->sc_state != MASTER)
1214 carp_macmatch6(struct ifnet *ifp, struct mbuf *m, const struct in6_addr *taddr)
1220 IFNET_FOREACH_IFA(ifp, ifa)
1221 if (ifa->ifa_addr->sa_family == AF_INET6 &&
1222 IN6_ARE_ADDR_EQUAL(taddr, IFA_IN6(ifa))) {
1223 struct carp_softc *sc = ifa->ifa_carp;
1226 mtag = m_tag_get(PACKET_TAG_CARP,
1227 sizeof(struct carp_softc *), M_NOWAIT);
1229 /* Better a bit than nothing. */
1230 return (LLADDR(&sc->sc_addr));
1232 bcopy(&sc, mtag + 1, sizeof(sc));
1233 m_tag_prepend(m, mtag);
1235 return (LLADDR(&sc->sc_addr));
1243 carp_forus(struct ifnet *ifp, u_char *dhost)
1245 struct carp_softc *sc;
1246 uint8_t *ena = dhost;
1248 if (ena[0] || ena[1] || ena[2] != 0x5e || ena[3] || ena[4] != 1)
1251 CIF_LOCK(ifp->if_carp);
1252 IFNET_FOREACH_CARP(ifp, sc) {
1254 * CARP_LOCK() is not here, since would protect nothing, but
1255 * cause deadlock with if_bridge, calling this under its lock.
1257 if (sc->sc_state == MASTER && !bcmp(dhost, LLADDR(&sc->sc_addr),
1259 CIF_UNLOCK(ifp->if_carp);
1263 CIF_UNLOCK(ifp->if_carp);
1268 /* Master down timeout event, executed in callout context. */
1270 carp_master_down(void *v)
1272 struct carp_softc *sc = v;
1273 struct epoch_tracker et;
1275 NET_EPOCH_ENTER(et);
1276 CARP_LOCK_ASSERT(sc);
1278 CURVNET_SET(sc->sc_carpdev->if_vnet);
1279 if (sc->sc_state == BACKUP) {
1280 carp_master_down_locked(sc, "master timed out");
1289 carp_master_down_locked(struct carp_softc *sc, const char *reason)
1293 CARP_LOCK_ASSERT(sc);
1295 switch (sc->sc_state) {
1297 carp_set_state(sc, MASTER, reason);
1298 carp_send_ad_locked(sc);
1311 panic("carp: VHID %u@%s: master_down event in %s state\n",
1313 sc->sc_carpdev->if_xname,
1314 sc->sc_state ? "MASTER" : "INIT");
1321 * When in backup state, af indicates whether to reset the master down timer
1322 * for v4 or v6. If it's set to zero, reset the ones which are already pending.
1325 carp_setrun(struct carp_softc *sc, sa_family_t af)
1329 CARP_LOCK_ASSERT(sc);
1331 if ((sc->sc_carpdev->if_flags & IFF_UP) == 0 ||
1332 sc->sc_carpdev->if_link_state != LINK_STATE_UP ||
1333 (sc->sc_naddrs == 0 && sc->sc_naddrs6 == 0) ||
1337 switch (sc->sc_state) {
1339 carp_set_state(sc, BACKUP, "initialization complete");
1343 callout_stop(&sc->sc_ad_tmo);
1344 tv.tv_sec = 3 * sc->sc_advbase;
1345 tv.tv_usec = sc->sc_advskew * 1000000 / 256;
1349 callout_reset(&sc->sc_md_tmo, tvtohz(&tv),
1350 carp_master_down, sc);
1355 callout_reset(&sc->sc_md6_tmo, tvtohz(&tv),
1356 carp_master_down, sc);
1362 callout_reset(&sc->sc_md_tmo, tvtohz(&tv),
1363 carp_master_down, sc);
1367 callout_reset(&sc->sc_md6_tmo, tvtohz(&tv),
1368 carp_master_down, sc);
1374 tv.tv_sec = sc->sc_advbase;
1375 tv.tv_usec = sc->sc_advskew * 1000000 / 256;
1376 callout_reset(&sc->sc_ad_tmo, tvtohz(&tv),
1383 * Setup multicast structures.
1386 carp_multicast_setup(struct carp_if *cif, sa_family_t sa)
1388 struct ifnet *ifp = cif->cif_ifp;
1395 struct ip_moptions *imo = &cif->cif_imo;
1396 struct in_mfilter *imf;
1397 struct in_addr addr;
1399 if (ip_mfilter_first(&imo->imo_head) != NULL)
1402 imf = ip_mfilter_alloc(M_WAITOK, 0, 0);
1403 ip_mfilter_init(&imo->imo_head);
1404 imo->imo_multicast_vif = -1;
1406 addr.s_addr = htonl(INADDR_CARP_GROUP);
1407 if ((error = in_joingroup(ifp, &addr, NULL,
1408 &imf->imf_inm)) != 0) {
1409 ip_mfilter_free(imf);
1413 ip_mfilter_insert(&imo->imo_head, imf);
1414 imo->imo_multicast_ifp = ifp;
1415 imo->imo_multicast_ttl = CARP_DFLTTL;
1416 imo->imo_multicast_loop = 0;
1423 struct ip6_moptions *im6o = &cif->cif_im6o;
1424 struct in6_mfilter *im6f[2];
1425 struct in6_addr in6;
1427 if (ip6_mfilter_first(&im6o->im6o_head))
1430 im6f[0] = ip6_mfilter_alloc(M_WAITOK, 0, 0);
1431 im6f[1] = ip6_mfilter_alloc(M_WAITOK, 0, 0);
1433 ip6_mfilter_init(&im6o->im6o_head);
1434 im6o->im6o_multicast_hlim = CARP_DFLTTL;
1435 im6o->im6o_multicast_ifp = ifp;
1437 /* Join IPv6 CARP multicast group. */
1438 bzero(&in6, sizeof(in6));
1439 in6.s6_addr16[0] = htons(0xff02);
1440 in6.s6_addr8[15] = 0x12;
1441 if ((error = in6_setscope(&in6, ifp, NULL)) != 0) {
1442 ip6_mfilter_free(im6f[0]);
1443 ip6_mfilter_free(im6f[1]);
1446 if ((error = in6_joingroup(ifp, &in6, NULL, &im6f[0]->im6f_in6m, 0)) != 0) {
1447 ip6_mfilter_free(im6f[0]);
1448 ip6_mfilter_free(im6f[1]);
1452 /* Join solicited multicast address. */
1453 bzero(&in6, sizeof(in6));
1454 in6.s6_addr16[0] = htons(0xff02);
1455 in6.s6_addr32[1] = 0;
1456 in6.s6_addr32[2] = htonl(1);
1457 in6.s6_addr32[3] = 0;
1458 in6.s6_addr8[12] = 0xff;
1460 if ((error = in6_setscope(&in6, ifp, NULL)) != 0) {
1461 ip6_mfilter_free(im6f[0]);
1462 ip6_mfilter_free(im6f[1]);
1466 if ((error = in6_joingroup(ifp, &in6, NULL, &im6f[1]->im6f_in6m, 0)) != 0) {
1467 in6_leavegroup(im6f[0]->im6f_in6m, NULL);
1468 ip6_mfilter_free(im6f[0]);
1469 ip6_mfilter_free(im6f[1]);
1472 ip6_mfilter_insert(&im6o->im6o_head, im6f[0]);
1473 ip6_mfilter_insert(&im6o->im6o_head, im6f[1]);
1483 * Free multicast structures.
1486 carp_multicast_cleanup(struct carp_if *cif, sa_family_t sa)
1489 struct ip_moptions *imo = &cif->cif_imo;
1490 struct in_mfilter *imf;
1493 struct ip6_moptions *im6o = &cif->cif_im6o;
1494 struct in6_mfilter *im6f;
1496 sx_assert(&carp_sx, SA_XLOCKED);
1501 if (cif->cif_naddrs != 0)
1504 while ((imf = ip_mfilter_first(&imo->imo_head)) != NULL) {
1505 ip_mfilter_remove(&imo->imo_head, imf);
1506 in_leavegroup(imf->imf_inm, NULL);
1507 ip_mfilter_free(imf);
1513 if (cif->cif_naddrs6 != 0)
1516 while ((im6f = ip6_mfilter_first(&im6o->im6o_head)) != NULL) {
1517 ip6_mfilter_remove(&im6o->im6o_head, im6f);
1518 in6_leavegroup(im6f->im6f_in6m, NULL);
1519 ip6_mfilter_free(im6f);
1527 carp_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *sa)
1530 struct carp_softc *sc;
1535 switch (sa->sa_family) {
1548 mtag = m_tag_find(m, PACKET_TAG_CARP, NULL);
1552 bcopy(mtag + 1, &sc, sizeof(sc));
1554 /* Set the source MAC address to the Virtual Router MAC Address. */
1555 switch (ifp->if_type) {
1559 struct ether_header *eh;
1561 eh = mtod(m, struct ether_header *);
1562 eh->ether_shost[0] = 0;
1563 eh->ether_shost[1] = 0;
1564 eh->ether_shost[2] = 0x5e;
1565 eh->ether_shost[3] = 0;
1566 eh->ether_shost[4] = 1;
1567 eh->ether_shost[5] = sc->sc_vhid;
1571 printf("%s: carp is not supported for the %d interface type\n",
1572 ifp->if_xname, ifp->if_type);
1573 return (EOPNOTSUPP);
1579 static struct carp_softc*
1580 carp_alloc(struct ifnet *ifp)
1582 struct carp_softc *sc;
1583 struct carp_if *cif;
1585 sx_assert(&carp_sx, SA_XLOCKED);
1587 if ((cif = ifp->if_carp) == NULL)
1588 cif = carp_alloc_if(ifp);
1590 sc = malloc(sizeof(*sc), M_CARP, M_WAITOK|M_ZERO);
1592 sc->sc_advbase = CARP_DFLTINTV;
1593 sc->sc_vhid = -1; /* required setting */
1594 sc->sc_init_counter = 1;
1595 sc->sc_state = INIT;
1597 sc->sc_ifasiz = sizeof(struct ifaddr *);
1598 sc->sc_ifas = malloc(sc->sc_ifasiz, M_CARP, M_WAITOK|M_ZERO);
1599 sc->sc_carpdev = ifp;
1603 callout_init_mtx(&sc->sc_md_tmo, &sc->sc_mtx, CALLOUT_RETURNUNLOCKED);
1606 callout_init_mtx(&sc->sc_md6_tmo, &sc->sc_mtx, CALLOUT_RETURNUNLOCKED);
1608 callout_init_mtx(&sc->sc_ad_tmo, &sc->sc_mtx, CALLOUT_RETURNUNLOCKED);
1611 TAILQ_INSERT_TAIL(&cif->cif_vrs, sc, sc_list);
1614 mtx_lock(&carp_mtx);
1615 LIST_INSERT_HEAD(&carp_list, sc, sc_next);
1616 mtx_unlock(&carp_mtx);
1622 carp_grow_ifas(struct carp_softc *sc)
1624 struct ifaddr **new;
1626 new = malloc(sc->sc_ifasiz * 2, M_CARP, M_WAITOK | M_ZERO);
1628 bcopy(sc->sc_ifas, new, sc->sc_ifasiz);
1629 free(sc->sc_ifas, M_CARP);
1636 carp_destroy(struct carp_softc *sc)
1638 struct ifnet *ifp = sc->sc_carpdev;
1639 struct carp_if *cif = ifp->if_carp;
1641 sx_assert(&carp_sx, SA_XLOCKED);
1643 if (sc->sc_suppress)
1644 carp_demote_adj(-V_carp_ifdown_adj, "vhid removed");
1648 TAILQ_REMOVE(&cif->cif_vrs, sc, sc_list);
1651 mtx_lock(&carp_mtx);
1652 LIST_REMOVE(sc, sc_next);
1653 mtx_unlock(&carp_mtx);
1655 callout_drain(&sc->sc_ad_tmo);
1657 callout_drain(&sc->sc_md_tmo);
1660 callout_drain(&sc->sc_md6_tmo);
1662 CARP_LOCK_DESTROY(sc);
1664 free(sc->sc_ifas, M_CARP);
1668 static struct carp_if*
1669 carp_alloc_if(struct ifnet *ifp)
1671 struct carp_if *cif;
1674 cif = malloc(sizeof(*cif), M_CARP, M_WAITOK|M_ZERO);
1676 if ((error = ifpromisc(ifp, 1)) != 0)
1677 printf("%s: ifpromisc(%s) failed: %d\n",
1678 __func__, ifp->if_xname, error);
1680 cif->cif_flags |= CIF_PROMISC;
1684 TAILQ_INIT(&cif->cif_vrs);
1689 IF_ADDR_WUNLOCK(ifp);
1695 carp_free_if(struct carp_if *cif)
1697 struct ifnet *ifp = cif->cif_ifp;
1699 CIF_LOCK_ASSERT(cif);
1700 KASSERT(TAILQ_EMPTY(&cif->cif_vrs), ("%s: softc list not empty",
1704 ifp->if_carp = NULL;
1705 IF_ADDR_WUNLOCK(ifp);
1707 CIF_LOCK_DESTROY(cif);
1709 if (cif->cif_flags & CIF_PROMISC)
1717 carp_carprcp(struct carpreq *carpr, struct carp_softc *sc, int priv)
1721 carpr->carpr_state = sc->sc_state;
1722 carpr->carpr_vhid = sc->sc_vhid;
1723 carpr->carpr_advbase = sc->sc_advbase;
1724 carpr->carpr_advskew = sc->sc_advskew;
1726 bcopy(sc->sc_key, carpr->carpr_key, sizeof(carpr->carpr_key));
1728 bzero(carpr->carpr_key, sizeof(carpr->carpr_key));
1733 carp_ioctl(struct ifreq *ifr, u_long cmd, struct thread *td)
1735 struct carpreq carpr;
1737 struct carp_softc *sc = NULL;
1738 int error = 0, locked = 0;
1740 if ((error = copyin(ifr_data_get_ptr(ifr), &carpr, sizeof carpr)))
1743 ifp = ifunit_ref(ifr->ifr_name);
1747 switch (ifp->if_type) {
1757 if ((ifp->if_flags & IFF_MULTICAST) == 0) {
1758 error = EADDRNOTAVAIL;
1765 if ((error = priv_check(td, PRIV_NETINET_CARP)))
1767 if (carpr.carpr_vhid <= 0 || carpr.carpr_vhid > CARP_MAXVHID ||
1768 carpr.carpr_advbase < 0 || carpr.carpr_advskew < 0) {
1774 IFNET_FOREACH_CARP(ifp, sc)
1775 if (sc->sc_vhid == carpr.carpr_vhid)
1779 sc = carp_alloc(ifp);
1781 sc->sc_vhid = carpr.carpr_vhid;
1782 LLADDR(&sc->sc_addr)[0] = 0;
1783 LLADDR(&sc->sc_addr)[1] = 0;
1784 LLADDR(&sc->sc_addr)[2] = 0x5e;
1785 LLADDR(&sc->sc_addr)[3] = 0;
1786 LLADDR(&sc->sc_addr)[4] = 1;
1787 LLADDR(&sc->sc_addr)[5] = sc->sc_vhid;
1791 if (carpr.carpr_advbase > 0) {
1792 if (carpr.carpr_advbase > 255 ||
1793 carpr.carpr_advbase < CARP_DFLTINTV) {
1797 sc->sc_advbase = carpr.carpr_advbase;
1799 if (carpr.carpr_advskew >= 255) {
1803 sc->sc_advskew = carpr.carpr_advskew;
1804 if (carpr.carpr_key[0] != '\0') {
1805 bcopy(carpr.carpr_key, sc->sc_key, sizeof(sc->sc_key));
1806 carp_hmac_prepare(sc);
1808 if (sc->sc_state != INIT &&
1809 carpr.carpr_state != sc->sc_state) {
1810 switch (carpr.carpr_state) {
1812 callout_stop(&sc->sc_ad_tmo);
1813 carp_set_state(sc, BACKUP,
1814 "user requested via ifconfig");
1819 carp_master_down_locked(sc,
1820 "user requested via ifconfig");
1832 if (carpr.carpr_vhid < 0 || carpr.carpr_vhid > CARP_MAXVHID) {
1836 if (carpr.carpr_count < 1) {
1840 if (ifp->if_carp == NULL) {
1845 priveleged = (priv_check(td, PRIV_NETINET_CARP) == 0);
1846 if (carpr.carpr_vhid != 0) {
1847 IFNET_FOREACH_CARP(ifp, sc)
1848 if (sc->sc_vhid == carpr.carpr_vhid)
1854 carp_carprcp(&carpr, sc, priveleged);
1855 error = copyout(&carpr, ifr_data_get_ptr(ifr),
1861 IFNET_FOREACH_CARP(ifp, sc)
1864 if (count > carpr.carpr_count) {
1865 CIF_UNLOCK(ifp->if_carp);
1871 IFNET_FOREACH_CARP(ifp, sc) {
1872 carp_carprcp(&carpr, sc, priveleged);
1873 carpr.carpr_count = count;
1874 error = copyout(&carpr,
1875 (char *)ifr_data_get_ptr(ifr) +
1876 (i * sizeof(carpr)), sizeof(carpr));
1878 CIF_UNLOCK(ifp->if_carp);
1889 sx_xunlock(&carp_sx);
1900 carp_get_vhid(struct ifaddr *ifa)
1903 if (ifa == NULL || ifa->ifa_carp == NULL)
1906 return (ifa->ifa_carp->sc_vhid);
1910 carp_attach(struct ifaddr *ifa, int vhid)
1912 struct ifnet *ifp = ifa->ifa_ifp;
1913 struct carp_if *cif = ifp->if_carp;
1914 struct carp_softc *sc;
1917 KASSERT(ifa->ifa_carp == NULL, ("%s: ifa %p attached", __func__, ifa));
1919 switch (ifa->ifa_addr->sa_family) {
1928 return (EPROTOTYPE);
1932 if (ifp->if_carp == NULL) {
1933 sx_xunlock(&carp_sx);
1934 return (ENOPROTOOPT);
1937 IFNET_FOREACH_CARP(ifp, sc)
1938 if (sc->sc_vhid == vhid)
1941 sx_xunlock(&carp_sx);
1945 error = carp_multicast_setup(cif, ifa->ifa_addr->sa_family);
1948 sx_xunlock(&carp_sx);
1952 index = sc->sc_naddrs + sc->sc_naddrs6 + 1;
1953 if (index > sc->sc_ifasiz / sizeof(struct ifaddr *))
1956 switch (ifa->ifa_addr->sa_family) {
1974 sc->sc_ifas[index - 1] = ifa;
1976 carp_hmac_prepare(sc);
1980 sx_xunlock(&carp_sx);
1986 carp_detach(struct ifaddr *ifa, bool keep_cif)
1988 struct ifnet *ifp = ifa->ifa_ifp;
1989 struct carp_if *cif = ifp->if_carp;
1990 struct carp_softc *sc = ifa->ifa_carp;
1993 KASSERT(sc != NULL, ("%s: %p not attached", __func__, ifa));
1999 index = sc->sc_naddrs + sc->sc_naddrs6;
2000 for (i = 0; i < index; i++)
2001 if (sc->sc_ifas[i] == ifa)
2003 KASSERT(i < index, ("%s: %p no backref", __func__, ifa));
2004 for (; i < index - 1; i++)
2005 sc->sc_ifas[i] = sc->sc_ifas[i+1];
2006 sc->sc_ifas[index - 1] = NULL;
2008 switch (ifa->ifa_addr->sa_family) {
2023 carp_ifa_delroute(ifa);
2024 carp_multicast_cleanup(cif, ifa->ifa_addr->sa_family);
2026 ifa->ifa_carp = NULL;
2029 carp_hmac_prepare(sc);
2032 if (!keep_cif && sc->sc_naddrs == 0 && sc->sc_naddrs6 == 0)
2040 sx_xunlock(&carp_sx);
2044 carp_set_state(struct carp_softc *sc, int state, const char *reason)
2047 CARP_LOCK_ASSERT(sc);
2049 if (sc->sc_state != state) {
2050 const char *carp_states[] = { CARP_STATES };
2051 char subsys[IFNAMSIZ+5];
2053 snprintf(subsys, IFNAMSIZ+5, "%u@%s", sc->sc_vhid,
2054 sc->sc_carpdev->if_xname);
2056 CARP_LOG("%s: %s -> %s (%s)\n", subsys,
2057 carp_states[sc->sc_state], carp_states[state], reason);
2059 sc->sc_state = state;
2061 devctl_notify("CARP", subsys, carp_states[state], NULL);
2066 carp_linkstate(struct ifnet *ifp)
2068 struct carp_softc *sc;
2070 CIF_LOCK(ifp->if_carp);
2071 IFNET_FOREACH_CARP(ifp, sc) {
2076 CIF_UNLOCK(ifp->if_carp);
2080 carp_sc_state(struct carp_softc *sc)
2083 CARP_LOCK_ASSERT(sc);
2085 if (sc->sc_carpdev->if_link_state != LINK_STATE_UP ||
2086 !(sc->sc_carpdev->if_flags & IFF_UP) ||
2088 callout_stop(&sc->sc_ad_tmo);
2090 callout_stop(&sc->sc_md_tmo);
2093 callout_stop(&sc->sc_md6_tmo);
2095 carp_set_state(sc, INIT, "hardware interface down");
2097 if (!sc->sc_suppress)
2098 carp_demote_adj(V_carp_ifdown_adj, "interface down");
2099 sc->sc_suppress = 1;
2101 carp_set_state(sc, INIT, "hardware interface up");
2103 if (sc->sc_suppress)
2104 carp_demote_adj(-V_carp_ifdown_adj, "interface up");
2105 sc->sc_suppress = 0;
2110 carp_demote_adj(int adj, char *reason)
2112 atomic_add_int(&V_carp_demotion, adj);
2113 CARP_LOG("demoted by %d to %d (%s)\n", adj, V_carp_demotion, reason);
2114 taskqueue_enqueue(taskqueue_swi, &carp_sendall_task);
2118 carp_allow_sysctl(SYSCTL_HANDLER_ARGS)
2121 struct carp_softc *sc;
2124 error = sysctl_handle_int(oidp, &new, 0, req);
2125 if (error || !req->newptr)
2128 if (V_carp_allow != new) {
2131 mtx_lock(&carp_mtx);
2132 LIST_FOREACH(sc, &carp_list, sc_next) {
2134 if (curvnet == sc->sc_carpdev->if_vnet)
2138 mtx_unlock(&carp_mtx);
2145 carp_dscp_sysctl(SYSCTL_HANDLER_ARGS)
2150 error = sysctl_handle_int(oidp, &new, 0, req);
2151 if (error || !req->newptr)
2154 if (new < 0 || new > 63)
2163 carp_demote_adj_sysctl(SYSCTL_HANDLER_ARGS)
2167 new = V_carp_demotion;
2168 error = sysctl_handle_int(oidp, &new, 0, req);
2169 if (error || !req->newptr)
2172 carp_demote_adj(new, "sysctl");
2178 extern struct domain inetdomain;
2179 static struct protosw in_carp_protosw = {
2180 .pr_type = SOCK_RAW,
2181 .pr_domain = &inetdomain,
2182 .pr_protocol = IPPROTO_CARP,
2183 .pr_flags = PR_ATOMIC|PR_ADDR,
2184 .pr_input = carp_input,
2185 .pr_output = rip_output,
2186 .pr_ctloutput = rip_ctloutput,
2187 .pr_usrreqs = &rip_usrreqs
2192 extern struct domain inet6domain;
2193 static struct protosw in6_carp_protosw = {
2194 .pr_type = SOCK_RAW,
2195 .pr_domain = &inet6domain,
2196 .pr_protocol = IPPROTO_CARP,
2197 .pr_flags = PR_ATOMIC|PR_ADDR,
2198 .pr_input = carp6_input,
2199 .pr_output = rip6_output,
2200 .pr_ctloutput = rip6_ctloutput,
2201 .pr_usrreqs = &rip6_usrreqs
2206 carp_mod_cleanup(void)
2210 if (proto_reg[CARP_INET] == 0) {
2211 (void)ipproto_unregister(IPPROTO_CARP);
2212 pf_proto_unregister(PF_INET, IPPROTO_CARP, SOCK_RAW);
2213 proto_reg[CARP_INET] = -1;
2215 carp_iamatch_p = NULL;
2218 if (proto_reg[CARP_INET6] == 0) {
2219 (void)ip6proto_unregister(IPPROTO_CARP);
2220 pf_proto_unregister(PF_INET6, IPPROTO_CARP, SOCK_RAW);
2221 proto_reg[CARP_INET6] = -1;
2223 carp_iamatch6_p = NULL;
2224 carp_macmatch6_p = NULL;
2226 carp_ioctl_p = NULL;
2227 carp_attach_p = NULL;
2228 carp_detach_p = NULL;
2229 carp_get_vhid_p = NULL;
2230 carp_linkstate_p = NULL;
2231 carp_forus_p = NULL;
2232 carp_output_p = NULL;
2233 carp_demote_adj_p = NULL;
2234 carp_master_p = NULL;
2235 mtx_unlock(&carp_mtx);
2236 taskqueue_drain(taskqueue_swi, &carp_sendall_task);
2237 mtx_destroy(&carp_mtx);
2238 sx_destroy(&carp_sx);
2246 mtx_init(&carp_mtx, "carp_mtx", NULL, MTX_DEF);
2247 sx_init(&carp_sx, "carp_sx");
2248 LIST_INIT(&carp_list);
2249 carp_get_vhid_p = carp_get_vhid;
2250 carp_forus_p = carp_forus;
2251 carp_output_p = carp_output;
2252 carp_linkstate_p = carp_linkstate;
2253 carp_ioctl_p = carp_ioctl;
2254 carp_attach_p = carp_attach;
2255 carp_detach_p = carp_detach;
2256 carp_demote_adj_p = carp_demote_adj;
2257 carp_master_p = carp_master;
2259 carp_iamatch6_p = carp_iamatch6;
2260 carp_macmatch6_p = carp_macmatch6;
2261 proto_reg[CARP_INET6] = pf_proto_register(PF_INET6,
2262 (struct protosw *)&in6_carp_protosw);
2263 if (proto_reg[CARP_INET6]) {
2264 printf("carp: error %d attaching to PF_INET6\n",
2265 proto_reg[CARP_INET6]);
2267 return (proto_reg[CARP_INET6]);
2269 err = ip6proto_register(IPPROTO_CARP);
2271 printf("carp: error %d registering with INET6\n", err);
2277 carp_iamatch_p = carp_iamatch;
2278 proto_reg[CARP_INET] = pf_proto_register(PF_INET, &in_carp_protosw);
2279 if (proto_reg[CARP_INET]) {
2280 printf("carp: error %d attaching to PF_INET\n",
2281 proto_reg[CARP_INET]);
2283 return (proto_reg[CARP_INET]);
2285 err = ipproto_register(IPPROTO_CARP);
2287 printf("carp: error %d registering with INET\n", err);
2296 carp_modevent(module_t mod, int type, void *data)
2300 return carp_mod_load();
2303 mtx_lock(&carp_mtx);
2304 if (LIST_EMPTY(&carp_list))
2307 mtx_unlock(&carp_mtx);
2319 static moduledata_t carp_mod = {
2325 DECLARE_MODULE(carp, carp_mod, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY);