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>
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, 0, "CARP");
222 SYSCTL_PROC(_net_inet_carp, OID_AUTO, allow,
223 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, 0, 0, carp_allow_sysctl, "I",
224 "Accept incoming CARP packets");
225 SYSCTL_PROC(_net_inet_carp, OID_AUTO, dscp,
226 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW, 0, 0, carp_dscp_sysctl, "I",
227 "DSCP value for carp packets");
228 SYSCTL_INT(_net_inet_carp, OID_AUTO, preempt, CTLFLAG_VNET | CTLFLAG_RW,
229 &VNET_NAME(carp_preempt), 0, "High-priority backup preemption mode");
230 SYSCTL_INT(_net_inet_carp, OID_AUTO, log, CTLFLAG_VNET | CTLFLAG_RW,
231 &VNET_NAME(carp_log), 0, "CARP log level");
232 SYSCTL_PROC(_net_inet_carp, OID_AUTO, demotion,
233 CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW,
234 0, 0, carp_demote_adj_sysctl, "I",
235 "Adjust demotion factor (skew of advskew)");
236 SYSCTL_INT(_net_inet_carp, OID_AUTO, senderr_demotion_factor,
237 CTLFLAG_VNET | CTLFLAG_RW,
238 &VNET_NAME(carp_senderr_adj), 0, "Send error demotion factor adjustment");
239 SYSCTL_INT(_net_inet_carp, OID_AUTO, ifdown_demotion_factor,
240 CTLFLAG_VNET | CTLFLAG_RW,
241 &VNET_NAME(carp_ifdown_adj), 0,
242 "Interface down demotion factor adjustment");
244 VNET_PCPUSTAT_DEFINE(struct carpstats, carpstats);
245 VNET_PCPUSTAT_SYSINIT(carpstats);
246 VNET_PCPUSTAT_SYSUNINIT(carpstats);
248 #define CARPSTATS_ADD(name, val) \
249 counter_u64_add(VNET(carpstats)[offsetof(struct carpstats, name) / \
250 sizeof(uint64_t)], (val))
251 #define CARPSTATS_INC(name) CARPSTATS_ADD(name, 1)
253 SYSCTL_VNET_PCPUSTAT(_net_inet_carp, OID_AUTO, stats, struct carpstats,
254 carpstats, "CARP statistics (struct carpstats, netinet/ip_carp.h)");
256 #define CARP_LOCK_INIT(sc) mtx_init(&(sc)->sc_mtx, "carp_softc", \
258 #define CARP_LOCK_DESTROY(sc) mtx_destroy(&(sc)->sc_mtx)
259 #define CARP_LOCK_ASSERT(sc) mtx_assert(&(sc)->sc_mtx, MA_OWNED)
260 #define CARP_LOCK(sc) mtx_lock(&(sc)->sc_mtx)
261 #define CARP_UNLOCK(sc) mtx_unlock(&(sc)->sc_mtx)
262 #define CIF_LOCK_INIT(cif) mtx_init(&(cif)->cif_mtx, "carp_if", \
264 #define CIF_LOCK_DESTROY(cif) mtx_destroy(&(cif)->cif_mtx)
265 #define CIF_LOCK_ASSERT(cif) mtx_assert(&(cif)->cif_mtx, MA_OWNED)
266 #define CIF_LOCK(cif) mtx_lock(&(cif)->cif_mtx)
267 #define CIF_UNLOCK(cif) mtx_unlock(&(cif)->cif_mtx)
268 #define CIF_FREE(cif) do { \
270 if (TAILQ_EMPTY(&(cif)->cif_vrs)) \
276 #define CARP_LOG(...) do { \
277 if (V_carp_log > 0) \
278 log(LOG_INFO, "carp: " __VA_ARGS__); \
281 #define CARP_DEBUG(...) do { \
282 if (V_carp_log > 1) \
283 log(LOG_DEBUG, __VA_ARGS__); \
286 #define IFNET_FOREACH_IFA(ifp, ifa) \
287 CK_STAILQ_FOREACH((ifa), &(ifp)->if_addrhead, ifa_link) \
288 if ((ifa)->ifa_carp != NULL)
290 #define CARP_FOREACH_IFA(sc, ifa) \
291 CARP_LOCK_ASSERT(sc); \
293 _i < (sc)->sc_naddrs + (sc)->sc_naddrs6 && \
294 ((ifa) = sc->sc_ifas[_i]) != NULL; \
297 #define IFNET_FOREACH_CARP(ifp, sc) \
298 KASSERT(mtx_owned(&ifp->if_carp->cif_mtx) || \
299 sx_xlocked(&carp_sx), ("cif_vrs not locked")); \
300 TAILQ_FOREACH((sc), &(ifp)->if_carp->cif_vrs, sc_list)
302 #define DEMOTE_ADVSKEW(sc) \
303 (((sc)->sc_advskew + V_carp_demotion > CARP_MAXSKEW) ? \
304 CARP_MAXSKEW : ((sc)->sc_advskew + V_carp_demotion))
306 static void carp_input_c(struct mbuf *, struct carp_header *, sa_family_t);
307 static struct carp_softc
308 *carp_alloc(struct ifnet *);
309 static void carp_destroy(struct carp_softc *);
310 static struct carp_if
311 *carp_alloc_if(struct ifnet *);
312 static void carp_free_if(struct carp_if *);
313 static void carp_set_state(struct carp_softc *, int, const char* reason);
314 static void carp_sc_state(struct carp_softc *);
315 static void carp_setrun(struct carp_softc *, sa_family_t);
316 static void carp_master_down(void *);
317 static void carp_master_down_locked(struct carp_softc *,
319 static void carp_send_ad(void *);
320 static void carp_send_ad_locked(struct carp_softc *);
321 static void carp_addroute(struct carp_softc *);
322 static void carp_ifa_addroute(struct ifaddr *);
323 static void carp_delroute(struct carp_softc *);
324 static void carp_ifa_delroute(struct ifaddr *);
325 static void carp_send_ad_all(void *, int);
326 static void carp_demote_adj(int, char *);
328 static LIST_HEAD(, carp_softc) carp_list;
329 static struct mtx carp_mtx;
330 static struct sx carp_sx;
331 static struct task carp_sendall_task =
332 TASK_INITIALIZER(0, carp_send_ad_all, NULL);
335 carp_hmac_prepare(struct carp_softc *sc)
337 uint8_t version = CARP_VERSION, type = CARP_ADVERTISEMENT;
338 uint8_t vhid = sc->sc_vhid & 0xff;
342 struct in_addr last, cur, in;
345 struct in6_addr last6, cur6, in6;
348 CARP_LOCK_ASSERT(sc);
350 /* Compute ipad from key. */
351 bzero(sc->sc_pad, sizeof(sc->sc_pad));
352 bcopy(sc->sc_key, sc->sc_pad, sizeof(sc->sc_key));
353 for (i = 0; i < sizeof(sc->sc_pad); i++)
354 sc->sc_pad[i] ^= 0x36;
356 /* Precompute first part of inner hash. */
357 SHA1Init(&sc->sc_sha1);
358 SHA1Update(&sc->sc_sha1, sc->sc_pad, sizeof(sc->sc_pad));
359 SHA1Update(&sc->sc_sha1, (void *)&version, sizeof(version));
360 SHA1Update(&sc->sc_sha1, (void *)&type, sizeof(type));
361 SHA1Update(&sc->sc_sha1, (void *)&vhid, sizeof(vhid));
367 cur.s_addr = 0xffffffff;
368 CARP_FOREACH_IFA(sc, ifa) {
369 in.s_addr = ifatoia(ifa)->ia_addr.sin_addr.s_addr;
370 if (ifa->ifa_addr->sa_family == AF_INET &&
371 ntohl(in.s_addr) > ntohl(last.s_addr) &&
372 ntohl(in.s_addr) < ntohl(cur.s_addr)) {
373 cur.s_addr = in.s_addr;
378 SHA1Update(&sc->sc_sha1, (void *)&cur, sizeof(cur));
382 memset(&cur6, 0, sizeof(cur6));
386 memset(&cur6, 0xff, sizeof(cur6));
387 CARP_FOREACH_IFA(sc, ifa) {
388 in6 = ifatoia6(ifa)->ia_addr.sin6_addr;
389 if (IN6_IS_SCOPE_EMBED(&in6))
390 in6.s6_addr16[1] = 0;
391 if (ifa->ifa_addr->sa_family == AF_INET6 &&
392 memcmp(&in6, &last6, sizeof(in6)) > 0 &&
393 memcmp(&in6, &cur6, sizeof(in6)) < 0) {
399 SHA1Update(&sc->sc_sha1, (void *)&cur6, sizeof(cur6));
403 /* convert ipad to opad */
404 for (i = 0; i < sizeof(sc->sc_pad); i++)
405 sc->sc_pad[i] ^= 0x36 ^ 0x5c;
409 carp_hmac_generate(struct carp_softc *sc, uint32_t counter[2],
410 unsigned char md[20])
414 CARP_LOCK_ASSERT(sc);
416 /* fetch first half of inner hash */
417 bcopy(&sc->sc_sha1, &sha1ctx, sizeof(sha1ctx));
419 SHA1Update(&sha1ctx, (void *)counter, sizeof(sc->sc_counter));
420 SHA1Final(md, &sha1ctx);
424 SHA1Update(&sha1ctx, sc->sc_pad, sizeof(sc->sc_pad));
425 SHA1Update(&sha1ctx, md, 20);
426 SHA1Final(md, &sha1ctx);
430 carp_hmac_verify(struct carp_softc *sc, uint32_t counter[2],
431 unsigned char md[20])
433 unsigned char md2[20];
435 CARP_LOCK_ASSERT(sc);
437 carp_hmac_generate(sc, counter, md2);
439 return (bcmp(md, md2, sizeof(md2)));
443 * process input packet.
444 * we have rearranged checks order compared to the rfc,
445 * but it seems more efficient this way or not possible otherwise.
449 carp_input(struct mbuf **mp, int *offp, int proto)
451 struct mbuf *m = *mp;
452 struct ip *ip = mtod(m, struct ip *);
453 struct carp_header *ch;
459 CARPSTATS_INC(carps_ipackets);
463 return (IPPROTO_DONE);
466 /* verify that the IP TTL is 255. */
467 if (ip->ip_ttl != CARP_DFLTTL) {
468 CARPSTATS_INC(carps_badttl);
469 CARP_DEBUG("%s: received ttl %d != 255 on %s\n", __func__,
471 m->m_pkthdr.rcvif->if_xname);
473 return (IPPROTO_DONE);
476 iplen = ip->ip_hl << 2;
478 if (m->m_pkthdr.len < iplen + sizeof(*ch)) {
479 CARPSTATS_INC(carps_badlen);
480 CARP_DEBUG("%s: received len %zd < sizeof(struct carp_header) "
481 "on %s\n", __func__, m->m_len - sizeof(struct ip),
482 m->m_pkthdr.rcvif->if_xname);
484 return (IPPROTO_DONE);
487 if (iplen + sizeof(*ch) < m->m_len) {
488 if ((m = m_pullup(m, iplen + sizeof(*ch))) == NULL) {
489 CARPSTATS_INC(carps_hdrops);
490 CARP_DEBUG("%s: pullup failed\n", __func__);
491 return (IPPROTO_DONE);
493 ip = mtod(m, struct ip *);
495 ch = (struct carp_header *)((char *)ip + iplen);
498 * verify that the received packet length is
499 * equal to the CARP header
501 len = iplen + sizeof(*ch);
502 if (len > m->m_pkthdr.len) {
503 CARPSTATS_INC(carps_badlen);
504 CARP_DEBUG("%s: packet too short %d on %s\n", __func__,
506 m->m_pkthdr.rcvif->if_xname);
508 return (IPPROTO_DONE);
511 if ((m = m_pullup(m, len)) == NULL) {
512 CARPSTATS_INC(carps_hdrops);
513 return (IPPROTO_DONE);
515 ip = mtod(m, struct ip *);
516 ch = (struct carp_header *)((char *)ip + iplen);
518 /* verify the CARP checksum */
520 if (in_cksum(m, len - iplen)) {
521 CARPSTATS_INC(carps_badsum);
522 CARP_DEBUG("%s: checksum failed on %s\n", __func__,
523 m->m_pkthdr.rcvif->if_xname);
525 return (IPPROTO_DONE);
529 carp_input_c(m, ch, AF_INET);
530 return (IPPROTO_DONE);
536 carp6_input(struct mbuf **mp, int *offp, int proto)
538 struct mbuf *m = *mp;
539 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
540 struct carp_header *ch;
543 CARPSTATS_INC(carps_ipackets6);
547 return (IPPROTO_DONE);
550 /* check if received on a valid carp interface */
551 if (m->m_pkthdr.rcvif->if_carp == NULL) {
552 CARPSTATS_INC(carps_badif);
553 CARP_DEBUG("%s: packet received on non-carp interface: %s\n",
554 __func__, m->m_pkthdr.rcvif->if_xname);
556 return (IPPROTO_DONE);
559 /* verify that the IP TTL is 255 */
560 if (ip6->ip6_hlim != CARP_DFLTTL) {
561 CARPSTATS_INC(carps_badttl);
562 CARP_DEBUG("%s: received ttl %d != 255 on %s\n", __func__,
563 ip6->ip6_hlim, m->m_pkthdr.rcvif->if_xname);
565 return (IPPROTO_DONE);
568 /* verify that we have a complete carp packet */
569 if (m->m_len < *offp + sizeof(*ch)) {
571 m = m_pullup(m, *offp + sizeof(*ch));
573 CARPSTATS_INC(carps_badlen);
574 CARP_DEBUG("%s: packet size %u too small\n", __func__, len);
575 return (IPPROTO_DONE);
578 ch = (struct carp_header *)(mtod(m, char *) + *offp);
581 /* verify the CARP checksum */
583 if (in_cksum(m, sizeof(*ch))) {
584 CARPSTATS_INC(carps_badsum);
585 CARP_DEBUG("%s: checksum failed, on %s\n", __func__,
586 m->m_pkthdr.rcvif->if_xname);
588 return (IPPROTO_DONE);
592 carp_input_c(m, ch, AF_INET6);
593 return (IPPROTO_DONE);
598 * This routine should not be necessary at all, but some switches
599 * (VMWare ESX vswitches) can echo our own packets back at us,
600 * and we must ignore them or they will cause us to drop out of
603 * We cannot catch all cases of network loops. Instead, what we
604 * do here is catch any packet that arrives with a carp header
605 * with a VHID of 0, that comes from an address that is our own.
606 * These packets are by definition "from us" (even if they are from
607 * a misconfigured host that is pretending to be us).
609 * The VHID test is outside this mini-function.
612 carp_source_is_self(struct mbuf *m, struct ifaddr *ifa, sa_family_t af)
626 ip4 = mtod(m, struct ip *);
627 in4 = ifatoia(ifa)->ia_addr.sin_addr;
628 return (in4.s_addr == ip4->ip_src.s_addr);
632 ip6 = mtod(m, struct ip6_hdr *);
633 in6 = ifatoia6(ifa)->ia_addr.sin6_addr;
634 return (memcmp(&in6, &ip6->ip6_src, sizeof(in6)) == 0);
643 carp_input_c(struct mbuf *m, struct carp_header *ch, sa_family_t af)
645 struct ifnet *ifp = m->m_pkthdr.rcvif;
646 struct ifaddr *ifa, *match;
647 struct carp_softc *sc;
648 uint64_t tmp_counter;
649 struct timeval sc_tv, ch_tv;
653 * Verify that the VHID is valid on the receiving interface.
655 * There should be just one match. If there are none
656 * the VHID is not valid and we drop the packet. If
657 * there are multiple VHID matches, take just the first
658 * one, for compatibility with previous code. While we're
659 * scanning, check for obvious loops in the network topology
660 * (these should never happen, and as noted above, we may
661 * miss real loops; this is just a double-check).
666 IFNET_FOREACH_IFA(ifp, ifa) {
667 if (match == NULL && ifa->ifa_carp != NULL &&
668 ifa->ifa_addr->sa_family == af &&
669 ifa->ifa_carp->sc_vhid == ch->carp_vhid)
671 if (ch->carp_vhid == 0 && carp_source_is_self(m, ifa, af))
674 ifa = error ? NULL : match;
677 IF_ADDR_RUNLOCK(ifp);
680 if (error == ELOOP) {
681 CARP_DEBUG("dropping looped packet on interface %s\n",
683 CARPSTATS_INC(carps_badif); /* ??? */
685 CARPSTATS_INC(carps_badvhid);
691 /* verify the CARP version. */
692 if (ch->carp_version != CARP_VERSION) {
693 CARPSTATS_INC(carps_badver);
694 CARP_DEBUG("%s: invalid version %d\n", ifp->if_xname,
705 if (carp_hmac_verify(sc, ch->carp_counter, ch->carp_md)) {
706 CARPSTATS_INC(carps_badauth);
707 CARP_DEBUG("%s: incorrect hash for VHID %u@%s\n", __func__,
708 sc->sc_vhid, ifp->if_xname);
712 tmp_counter = ntohl(ch->carp_counter[0]);
713 tmp_counter = tmp_counter<<32;
714 tmp_counter += ntohl(ch->carp_counter[1]);
716 /* XXX Replay protection goes here */
718 sc->sc_init_counter = 0;
719 sc->sc_counter = tmp_counter;
721 sc_tv.tv_sec = sc->sc_advbase;
722 sc_tv.tv_usec = DEMOTE_ADVSKEW(sc) * 1000000 / 256;
723 ch_tv.tv_sec = ch->carp_advbase;
724 ch_tv.tv_usec = ch->carp_advskew * 1000000 / 256;
726 switch (sc->sc_state) {
731 * If we receive an advertisement from a master who's going to
732 * be more frequent than us, go into BACKUP state.
734 if (timevalcmp(&sc_tv, &ch_tv, >) ||
735 timevalcmp(&sc_tv, &ch_tv, ==)) {
736 callout_stop(&sc->sc_ad_tmo);
737 carp_set_state(sc, BACKUP,
738 "more frequent advertisement received");
745 * If we're pre-empting masters who advertise slower than us,
746 * and this one claims to be slower, treat him as down.
748 if (V_carp_preempt && timevalcmp(&sc_tv, &ch_tv, <)) {
749 carp_master_down_locked(sc,
750 "preempting a slower master");
755 * If the master is going to advertise at such a low frequency
756 * that he's guaranteed to time out, we'd might as well just
757 * treat him as timed out now.
759 sc_tv.tv_sec = sc->sc_advbase * 3;
760 if (timevalcmp(&sc_tv, &ch_tv, <)) {
761 carp_master_down_locked(sc, "master will time out");
766 * Otherwise, we reset the counter and wait for the next
779 carp_prepare_ad(struct mbuf *m, struct carp_softc *sc, struct carp_header *ch)
783 if (sc->sc_init_counter) {
784 /* this could also be seconds since unix epoch */
785 sc->sc_counter = arc4random();
786 sc->sc_counter = sc->sc_counter << 32;
787 sc->sc_counter += arc4random();
791 ch->carp_counter[0] = htonl((sc->sc_counter>>32)&0xffffffff);
792 ch->carp_counter[1] = htonl(sc->sc_counter&0xffffffff);
794 carp_hmac_generate(sc, ch->carp_counter, ch->carp_md);
796 /* Tag packet for carp_output */
797 if ((mtag = m_tag_get(PACKET_TAG_CARP, sizeof(struct carp_softc *),
798 M_NOWAIT)) == NULL) {
800 CARPSTATS_INC(carps_onomem);
803 bcopy(&sc, mtag + 1, sizeof(sc));
804 m_tag_prepend(m, mtag);
810 * To avoid LORs and possible recursions this function shouldn't
811 * be called directly, but scheduled via taskqueue.
814 carp_send_ad_all(void *ctx __unused, int pending __unused)
816 struct carp_softc *sc;
819 LIST_FOREACH(sc, &carp_list, sc_next)
820 if (sc->sc_state == MASTER) {
822 CURVNET_SET(sc->sc_carpdev->if_vnet);
823 carp_send_ad_locked(sc);
827 mtx_unlock(&carp_mtx);
830 /* Send a periodic advertisement, executed in callout context. */
832 carp_send_ad(void *v)
834 struct carp_softc *sc = v;
836 CARP_LOCK_ASSERT(sc);
837 CURVNET_SET(sc->sc_carpdev->if_vnet);
838 carp_send_ad_locked(sc);
844 carp_send_ad_error(struct carp_softc *sc, int error)
848 if (sc->sc_sendad_errors < INT_MAX)
849 sc->sc_sendad_errors++;
850 if (sc->sc_sendad_errors == CARP_SENDAD_MAX_ERRORS) {
851 static const char fmt[] = "send error %d on %s";
852 char msg[sizeof(fmt) + IFNAMSIZ];
854 sprintf(msg, fmt, error, sc->sc_carpdev->if_xname);
855 carp_demote_adj(V_carp_senderr_adj, msg);
857 sc->sc_sendad_success = 0;
859 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS &&
860 ++sc->sc_sendad_success >= CARP_SENDAD_MIN_SUCCESS) {
861 static const char fmt[] = "send ok on %s";
862 char msg[sizeof(fmt) + IFNAMSIZ];
864 sprintf(msg, fmt, sc->sc_carpdev->if_xname);
865 carp_demote_adj(-V_carp_senderr_adj, msg);
866 sc->sc_sendad_errors = 0;
868 sc->sc_sendad_errors = 0;
873 * Pick the best ifaddr on the given ifp for sending CARP
876 * "Best" here is defined by ifa_preferred(). This function is much
877 * much like ifaof_ifpforaddr() except that we just use ifa_preferred().
879 * (This could be simplified to return the actual address, except that
880 * it has a different format in AF_INET and AF_INET6.)
882 static struct ifaddr *
883 carp_best_ifa(int af, struct ifnet *ifp)
885 struct ifaddr *ifa, *best;
891 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
892 if (ifa->ifa_addr->sa_family == af &&
893 (best == NULL || ifa_preferred(best, ifa)))
896 IF_ADDR_RUNLOCK(ifp);
903 carp_send_ad_locked(struct carp_softc *sc)
905 struct carp_header ch;
908 struct carp_header *ch_ptr;
912 CARP_LOCK_ASSERT(sc);
914 advskew = DEMOTE_ADVSKEW(sc);
915 tv.tv_sec = sc->sc_advbase;
916 tv.tv_usec = advskew * 1000000 / 256;
918 ch.carp_version = CARP_VERSION;
919 ch.carp_type = CARP_ADVERTISEMENT;
920 ch.carp_vhid = sc->sc_vhid;
921 ch.carp_advbase = sc->sc_advbase;
922 ch.carp_advskew = advskew;
923 ch.carp_authlen = 7; /* XXX DEFINE */
924 ch.carp_pad1 = 0; /* must be zero */
927 /* XXXGL: OpenBSD picks first ifaddr with needed family. */
933 m = m_gethdr(M_NOWAIT, MT_DATA);
935 CARPSTATS_INC(carps_onomem);
938 len = sizeof(*ip) + sizeof(ch);
939 m->m_pkthdr.len = len;
940 m->m_pkthdr.rcvif = NULL;
942 M_ALIGN(m, m->m_len);
943 m->m_flags |= M_MCAST;
944 ip = mtod(m, struct ip *);
945 ip->ip_v = IPVERSION;
946 ip->ip_hl = sizeof(*ip) >> 2;
947 ip->ip_tos = V_carp_dscp << IPTOS_DSCP_OFFSET;
948 ip->ip_len = htons(len);
949 ip->ip_off = htons(IP_DF);
950 ip->ip_ttl = CARP_DFLTTL;
951 ip->ip_p = IPPROTO_CARP;
955 ifa = carp_best_ifa(AF_INET, sc->sc_carpdev);
958 ifatoia(ifa)->ia_addr.sin_addr.s_addr;
961 ip->ip_src.s_addr = 0;
962 ip->ip_dst.s_addr = htonl(INADDR_CARP_GROUP);
964 ch_ptr = (struct carp_header *)(&ip[1]);
965 bcopy(&ch, ch_ptr, sizeof(ch));
966 if (carp_prepare_ad(m, sc, ch_ptr))
969 m->m_data += sizeof(*ip);
970 ch_ptr->carp_cksum = in_cksum(m, len - sizeof(*ip));
971 m->m_data -= sizeof(*ip);
973 CARPSTATS_INC(carps_opackets);
975 carp_send_ad_error(sc, ip_output(m, NULL, NULL, IP_RAWOUTPUT,
976 &sc->sc_carpdev->if_carp->cif_imo, NULL));
980 if (sc->sc_naddrs6) {
983 m = m_gethdr(M_NOWAIT, MT_DATA);
985 CARPSTATS_INC(carps_onomem);
988 len = sizeof(*ip6) + sizeof(ch);
989 m->m_pkthdr.len = len;
990 m->m_pkthdr.rcvif = NULL;
992 M_ALIGN(m, m->m_len);
993 m->m_flags |= M_MCAST;
994 ip6 = mtod(m, struct ip6_hdr *);
995 bzero(ip6, sizeof(*ip6));
996 ip6->ip6_vfc |= IPV6_VERSION;
997 /* Traffic class isn't defined in ip6 struct instead
998 * it gets offset into flowid field */
999 ip6->ip6_flow |= htonl(V_carp_dscp << (IPV6_FLOWLABEL_LEN +
1000 IPTOS_DSCP_OFFSET));
1001 ip6->ip6_hlim = CARP_DFLTTL;
1002 ip6->ip6_nxt = IPPROTO_CARP;
1004 /* set the source address */
1005 ifa = carp_best_ifa(AF_INET6, sc->sc_carpdev);
1007 bcopy(IFA_IN6(ifa), &ip6->ip6_src,
1008 sizeof(struct in6_addr));
1011 /* This should never happen with IPv6. */
1012 bzero(&ip6->ip6_src, sizeof(struct in6_addr));
1014 /* Set the multicast destination. */
1015 ip6->ip6_dst.s6_addr16[0] = htons(0xff02);
1016 ip6->ip6_dst.s6_addr8[15] = 0x12;
1017 if (in6_setscope(&ip6->ip6_dst, sc->sc_carpdev, NULL) != 0) {
1019 CARP_DEBUG("%s: in6_setscope failed\n", __func__);
1023 ch_ptr = (struct carp_header *)(&ip6[1]);
1024 bcopy(&ch, ch_ptr, sizeof(ch));
1025 if (carp_prepare_ad(m, sc, ch_ptr))
1028 m->m_data += sizeof(*ip6);
1029 ch_ptr->carp_cksum = in_cksum(m, len - sizeof(*ip6));
1030 m->m_data -= sizeof(*ip6);
1032 CARPSTATS_INC(carps_opackets6);
1034 carp_send_ad_error(sc, ip6_output(m, NULL, NULL, 0,
1035 &sc->sc_carpdev->if_carp->cif_im6o, NULL, NULL));
1040 callout_reset(&sc->sc_ad_tmo, tvtohz(&tv), carp_send_ad, sc);
1044 carp_addroute(struct carp_softc *sc)
1048 CARP_FOREACH_IFA(sc, ifa)
1049 carp_ifa_addroute(ifa);
1053 carp_ifa_addroute(struct ifaddr *ifa)
1056 switch (ifa->ifa_addr->sa_family) {
1059 in_addprefix(ifatoia(ifa), RTF_UP);
1060 ifa_add_loopback_route(ifa,
1061 (struct sockaddr *)&ifatoia(ifa)->ia_addr);
1066 ifa_add_loopback_route(ifa,
1067 (struct sockaddr *)&ifatoia6(ifa)->ia_addr);
1068 nd6_add_ifa_lle(ifatoia6(ifa));
1075 carp_delroute(struct carp_softc *sc)
1079 CARP_FOREACH_IFA(sc, ifa)
1080 carp_ifa_delroute(ifa);
1084 carp_ifa_delroute(struct ifaddr *ifa)
1087 switch (ifa->ifa_addr->sa_family) {
1090 ifa_del_loopback_route(ifa,
1091 (struct sockaddr *)&ifatoia(ifa)->ia_addr);
1092 in_scrubprefix(ifatoia(ifa), LLE_STATIC);
1097 ifa_del_loopback_route(ifa,
1098 (struct sockaddr *)&ifatoia6(ifa)->ia_addr);
1099 nd6_rem_ifa_lle(ifatoia6(ifa), 1);
1106 carp_master(struct ifaddr *ifa)
1108 struct carp_softc *sc = ifa->ifa_carp;
1110 return (sc->sc_state == MASTER);
1115 * Broadcast a gratuitous ARP request containing
1116 * the virtual router MAC address for each IP address
1117 * associated with the virtual router.
1120 carp_send_arp(struct carp_softc *sc)
1123 struct in_addr addr;
1125 CARP_FOREACH_IFA(sc, ifa) {
1126 if (ifa->ifa_addr->sa_family != AF_INET)
1128 addr = ((struct sockaddr_in *)ifa->ifa_addr)->sin_addr;
1129 arp_announce_ifaddr(sc->sc_carpdev, addr, LLADDR(&sc->sc_addr));
1134 carp_iamatch(struct ifaddr *ifa, uint8_t **enaddr)
1136 struct carp_softc *sc = ifa->ifa_carp;
1138 if (sc->sc_state == MASTER) {
1139 *enaddr = LLADDR(&sc->sc_addr);
1149 carp_send_na(struct carp_softc *sc)
1151 static struct in6_addr mcast = IN6ADDR_LINKLOCAL_ALLNODES_INIT;
1153 struct in6_addr *in6;
1155 CARP_FOREACH_IFA(sc, ifa) {
1156 if (ifa->ifa_addr->sa_family != AF_INET6)
1160 nd6_na_output(sc->sc_carpdev, &mcast, in6,
1161 ND_NA_FLAG_OVERRIDE, 1, NULL);
1162 DELAY(1000); /* XXX */
1167 * Returns ifa in case it's a carp address and it is MASTER, or if the address
1168 * matches and is not a carp address. Returns NULL otherwise.
1171 carp_iamatch6(struct ifnet *ifp, struct in6_addr *taddr)
1177 CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1178 if (ifa->ifa_addr->sa_family != AF_INET6)
1180 if (!IN6_ARE_ADDR_EQUAL(taddr, IFA_IN6(ifa)))
1182 if (ifa->ifa_carp && ifa->ifa_carp->sc_state != MASTER)
1188 IF_ADDR_RUNLOCK(ifp);
1194 carp_macmatch6(struct ifnet *ifp, struct mbuf *m, const struct in6_addr *taddr)
1199 IFNET_FOREACH_IFA(ifp, ifa)
1200 if (ifa->ifa_addr->sa_family == AF_INET6 &&
1201 IN6_ARE_ADDR_EQUAL(taddr, IFA_IN6(ifa))) {
1202 struct carp_softc *sc = ifa->ifa_carp;
1205 IF_ADDR_RUNLOCK(ifp);
1207 mtag = m_tag_get(PACKET_TAG_CARP,
1208 sizeof(struct carp_softc *), M_NOWAIT);
1210 /* Better a bit than nothing. */
1211 return (LLADDR(&sc->sc_addr));
1213 bcopy(&sc, mtag + 1, sizeof(sc));
1214 m_tag_prepend(m, mtag);
1216 return (LLADDR(&sc->sc_addr));
1218 IF_ADDR_RUNLOCK(ifp);
1225 carp_forus(struct ifnet *ifp, u_char *dhost)
1227 struct carp_softc *sc;
1228 uint8_t *ena = dhost;
1230 if (ena[0] || ena[1] || ena[2] != 0x5e || ena[3] || ena[4] != 1)
1233 CIF_LOCK(ifp->if_carp);
1234 IFNET_FOREACH_CARP(ifp, sc) {
1236 * CARP_LOCK() is not here, since would protect nothing, but
1237 * cause deadlock with if_bridge, calling this under its lock.
1239 if (sc->sc_state == MASTER && !bcmp(dhost, LLADDR(&sc->sc_addr),
1241 CIF_UNLOCK(ifp->if_carp);
1245 CIF_UNLOCK(ifp->if_carp);
1250 /* Master down timeout event, executed in callout context. */
1252 carp_master_down(void *v)
1254 struct carp_softc *sc = v;
1256 CARP_LOCK_ASSERT(sc);
1258 CURVNET_SET(sc->sc_carpdev->if_vnet);
1259 if (sc->sc_state == BACKUP) {
1260 carp_master_down_locked(sc, "master timed out");
1268 carp_master_down_locked(struct carp_softc *sc, const char *reason)
1271 CARP_LOCK_ASSERT(sc);
1273 switch (sc->sc_state) {
1275 carp_set_state(sc, MASTER, reason);
1276 carp_send_ad_locked(sc);
1289 panic("carp: VHID %u@%s: master_down event in %s state\n",
1291 sc->sc_carpdev->if_xname,
1292 sc->sc_state ? "MASTER" : "INIT");
1299 * When in backup state, af indicates whether to reset the master down timer
1300 * for v4 or v6. If it's set to zero, reset the ones which are already pending.
1303 carp_setrun(struct carp_softc *sc, sa_family_t af)
1307 CARP_LOCK_ASSERT(sc);
1309 if ((sc->sc_carpdev->if_flags & IFF_UP) == 0 ||
1310 sc->sc_carpdev->if_link_state != LINK_STATE_UP ||
1311 (sc->sc_naddrs == 0 && sc->sc_naddrs6 == 0) ||
1315 switch (sc->sc_state) {
1317 carp_set_state(sc, BACKUP, "initialization complete");
1321 callout_stop(&sc->sc_ad_tmo);
1322 tv.tv_sec = 3 * sc->sc_advbase;
1323 tv.tv_usec = sc->sc_advskew * 1000000 / 256;
1327 callout_reset(&sc->sc_md_tmo, tvtohz(&tv),
1328 carp_master_down, sc);
1333 callout_reset(&sc->sc_md6_tmo, tvtohz(&tv),
1334 carp_master_down, sc);
1340 callout_reset(&sc->sc_md_tmo, tvtohz(&tv),
1341 carp_master_down, sc);
1345 callout_reset(&sc->sc_md6_tmo, tvtohz(&tv),
1346 carp_master_down, sc);
1352 tv.tv_sec = sc->sc_advbase;
1353 tv.tv_usec = sc->sc_advskew * 1000000 / 256;
1354 callout_reset(&sc->sc_ad_tmo, tvtohz(&tv),
1361 * Setup multicast structures.
1364 carp_multicast_setup(struct carp_if *cif, sa_family_t sa)
1366 struct ifnet *ifp = cif->cif_ifp;
1373 struct ip_moptions *imo = &cif->cif_imo;
1374 struct in_mfilter *imf;
1375 struct in_addr addr;
1377 if (ip_mfilter_first(&imo->imo_head) != NULL)
1380 imf = ip_mfilter_alloc(M_WAITOK, 0, 0);
1381 ip_mfilter_init(&imo->imo_head);
1382 imo->imo_multicast_vif = -1;
1384 addr.s_addr = htonl(INADDR_CARP_GROUP);
1385 if ((error = in_joingroup(ifp, &addr, NULL,
1386 &imf->imf_inm)) != 0) {
1387 ip_mfilter_free(imf);
1391 ip_mfilter_insert(&imo->imo_head, imf);
1392 imo->imo_multicast_ifp = ifp;
1393 imo->imo_multicast_ttl = CARP_DFLTTL;
1394 imo->imo_multicast_loop = 0;
1401 struct ip6_moptions *im6o = &cif->cif_im6o;
1402 struct in6_mfilter *im6f[2];
1403 struct in6_addr in6;
1405 if (ip6_mfilter_first(&im6o->im6o_head))
1408 im6f[0] = ip6_mfilter_alloc(M_WAITOK, 0, 0);
1409 im6f[1] = ip6_mfilter_alloc(M_WAITOK, 0, 0);
1411 ip6_mfilter_init(&im6o->im6o_head);
1412 im6o->im6o_multicast_hlim = CARP_DFLTTL;
1413 im6o->im6o_multicast_ifp = ifp;
1415 /* Join IPv6 CARP multicast group. */
1416 bzero(&in6, sizeof(in6));
1417 in6.s6_addr16[0] = htons(0xff02);
1418 in6.s6_addr8[15] = 0x12;
1419 if ((error = in6_setscope(&in6, ifp, NULL)) != 0) {
1420 ip6_mfilter_free(im6f[0]);
1421 ip6_mfilter_free(im6f[1]);
1424 if ((error = in6_joingroup(ifp, &in6, NULL, &im6f[0]->im6f_in6m, 0)) != 0) {
1425 ip6_mfilter_free(im6f[0]);
1426 ip6_mfilter_free(im6f[1]);
1430 /* Join solicited multicast address. */
1431 bzero(&in6, sizeof(in6));
1432 in6.s6_addr16[0] = htons(0xff02);
1433 in6.s6_addr32[1] = 0;
1434 in6.s6_addr32[2] = htonl(1);
1435 in6.s6_addr32[3] = 0;
1436 in6.s6_addr8[12] = 0xff;
1438 if ((error = in6_setscope(&in6, ifp, NULL)) != 0) {
1439 ip6_mfilter_free(im6f[0]);
1440 ip6_mfilter_free(im6f[1]);
1444 if ((error = in6_joingroup(ifp, &in6, NULL, &im6f[1]->im6f_in6m, 0)) != 0) {
1445 in6_leavegroup(im6f[0]->im6f_in6m, NULL);
1446 ip6_mfilter_free(im6f[0]);
1447 ip6_mfilter_free(im6f[1]);
1450 ip6_mfilter_insert(&im6o->im6o_head, im6f[0]);
1451 ip6_mfilter_insert(&im6o->im6o_head, im6f[1]);
1461 * Free multicast structures.
1464 carp_multicast_cleanup(struct carp_if *cif, sa_family_t sa)
1467 struct ip_moptions *imo = &cif->cif_imo;
1468 struct in_mfilter *imf;
1471 struct ip6_moptions *im6o = &cif->cif_im6o;
1472 struct in6_mfilter *im6f;
1474 sx_assert(&carp_sx, SA_XLOCKED);
1479 if (cif->cif_naddrs != 0)
1482 while ((imf = ip_mfilter_first(&imo->imo_head)) != NULL) {
1483 ip_mfilter_remove(&imo->imo_head, imf);
1484 in_leavegroup(imf->imf_inm, NULL);
1485 ip_mfilter_free(imf);
1491 if (cif->cif_naddrs6 != 0)
1494 while ((im6f = ip6_mfilter_first(&im6o->im6o_head)) != NULL) {
1495 ip6_mfilter_remove(&im6o->im6o_head, im6f);
1496 in6_leavegroup(im6f->im6f_in6m, NULL);
1497 ip6_mfilter_free(im6f);
1505 carp_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *sa)
1508 struct carp_softc *sc;
1513 switch (sa->sa_family) {
1526 mtag = m_tag_find(m, PACKET_TAG_CARP, NULL);
1530 bcopy(mtag + 1, &sc, sizeof(sc));
1532 /* Set the source MAC address to the Virtual Router MAC Address. */
1533 switch (ifp->if_type) {
1537 struct ether_header *eh;
1539 eh = mtod(m, struct ether_header *);
1540 eh->ether_shost[0] = 0;
1541 eh->ether_shost[1] = 0;
1542 eh->ether_shost[2] = 0x5e;
1543 eh->ether_shost[3] = 0;
1544 eh->ether_shost[4] = 1;
1545 eh->ether_shost[5] = sc->sc_vhid;
1549 printf("%s: carp is not supported for the %d interface type\n",
1550 ifp->if_xname, ifp->if_type);
1551 return (EOPNOTSUPP);
1557 static struct carp_softc*
1558 carp_alloc(struct ifnet *ifp)
1560 struct carp_softc *sc;
1561 struct carp_if *cif;
1563 sx_assert(&carp_sx, SA_XLOCKED);
1565 if ((cif = ifp->if_carp) == NULL)
1566 cif = carp_alloc_if(ifp);
1568 sc = malloc(sizeof(*sc), M_CARP, M_WAITOK|M_ZERO);
1570 sc->sc_advbase = CARP_DFLTINTV;
1571 sc->sc_vhid = -1; /* required setting */
1572 sc->sc_init_counter = 1;
1573 sc->sc_state = INIT;
1575 sc->sc_ifasiz = sizeof(struct ifaddr *);
1576 sc->sc_ifas = malloc(sc->sc_ifasiz, M_CARP, M_WAITOK|M_ZERO);
1577 sc->sc_carpdev = ifp;
1581 callout_init_mtx(&sc->sc_md_tmo, &sc->sc_mtx, CALLOUT_RETURNUNLOCKED);
1584 callout_init_mtx(&sc->sc_md6_tmo, &sc->sc_mtx, CALLOUT_RETURNUNLOCKED);
1586 callout_init_mtx(&sc->sc_ad_tmo, &sc->sc_mtx, CALLOUT_RETURNUNLOCKED);
1589 TAILQ_INSERT_TAIL(&cif->cif_vrs, sc, sc_list);
1592 mtx_lock(&carp_mtx);
1593 LIST_INSERT_HEAD(&carp_list, sc, sc_next);
1594 mtx_unlock(&carp_mtx);
1600 carp_grow_ifas(struct carp_softc *sc)
1602 struct ifaddr **new;
1604 new = malloc(sc->sc_ifasiz * 2, M_CARP, M_WAITOK | M_ZERO);
1606 bcopy(sc->sc_ifas, new, sc->sc_ifasiz);
1607 free(sc->sc_ifas, M_CARP);
1614 carp_destroy(struct carp_softc *sc)
1616 struct ifnet *ifp = sc->sc_carpdev;
1617 struct carp_if *cif = ifp->if_carp;
1619 sx_assert(&carp_sx, SA_XLOCKED);
1621 if (sc->sc_suppress)
1622 carp_demote_adj(-V_carp_ifdown_adj, "vhid removed");
1626 TAILQ_REMOVE(&cif->cif_vrs, sc, sc_list);
1629 mtx_lock(&carp_mtx);
1630 LIST_REMOVE(sc, sc_next);
1631 mtx_unlock(&carp_mtx);
1633 callout_drain(&sc->sc_ad_tmo);
1635 callout_drain(&sc->sc_md_tmo);
1638 callout_drain(&sc->sc_md6_tmo);
1640 CARP_LOCK_DESTROY(sc);
1642 free(sc->sc_ifas, M_CARP);
1646 static struct carp_if*
1647 carp_alloc_if(struct ifnet *ifp)
1649 struct carp_if *cif;
1652 cif = malloc(sizeof(*cif), M_CARP, M_WAITOK|M_ZERO);
1654 if ((error = ifpromisc(ifp, 1)) != 0)
1655 printf("%s: ifpromisc(%s) failed: %d\n",
1656 __func__, ifp->if_xname, error);
1658 cif->cif_flags |= CIF_PROMISC;
1662 TAILQ_INIT(&cif->cif_vrs);
1667 IF_ADDR_WUNLOCK(ifp);
1673 carp_free_if(struct carp_if *cif)
1675 struct ifnet *ifp = cif->cif_ifp;
1677 CIF_LOCK_ASSERT(cif);
1678 KASSERT(TAILQ_EMPTY(&cif->cif_vrs), ("%s: softc list not empty",
1682 ifp->if_carp = NULL;
1683 IF_ADDR_WUNLOCK(ifp);
1685 CIF_LOCK_DESTROY(cif);
1687 if (cif->cif_flags & CIF_PROMISC)
1695 carp_carprcp(struct carpreq *carpr, struct carp_softc *sc, int priv)
1699 carpr->carpr_state = sc->sc_state;
1700 carpr->carpr_vhid = sc->sc_vhid;
1701 carpr->carpr_advbase = sc->sc_advbase;
1702 carpr->carpr_advskew = sc->sc_advskew;
1704 bcopy(sc->sc_key, carpr->carpr_key, sizeof(carpr->carpr_key));
1706 bzero(carpr->carpr_key, sizeof(carpr->carpr_key));
1711 carp_ioctl(struct ifreq *ifr, u_long cmd, struct thread *td)
1713 struct carpreq carpr;
1715 struct carp_softc *sc = NULL;
1716 int error = 0, locked = 0;
1718 if ((error = copyin(ifr_data_get_ptr(ifr), &carpr, sizeof carpr)))
1721 ifp = ifunit_ref(ifr->ifr_name);
1725 switch (ifp->if_type) {
1735 if ((ifp->if_flags & IFF_MULTICAST) == 0) {
1736 error = EADDRNOTAVAIL;
1743 if ((error = priv_check(td, PRIV_NETINET_CARP)))
1745 if (carpr.carpr_vhid <= 0 || carpr.carpr_vhid > CARP_MAXVHID ||
1746 carpr.carpr_advbase < 0 || carpr.carpr_advskew < 0) {
1752 IFNET_FOREACH_CARP(ifp, sc)
1753 if (sc->sc_vhid == carpr.carpr_vhid)
1757 sc = carp_alloc(ifp);
1759 sc->sc_vhid = carpr.carpr_vhid;
1760 LLADDR(&sc->sc_addr)[0] = 0;
1761 LLADDR(&sc->sc_addr)[1] = 0;
1762 LLADDR(&sc->sc_addr)[2] = 0x5e;
1763 LLADDR(&sc->sc_addr)[3] = 0;
1764 LLADDR(&sc->sc_addr)[4] = 1;
1765 LLADDR(&sc->sc_addr)[5] = sc->sc_vhid;
1769 if (carpr.carpr_advbase > 0) {
1770 if (carpr.carpr_advbase > 255 ||
1771 carpr.carpr_advbase < CARP_DFLTINTV) {
1775 sc->sc_advbase = carpr.carpr_advbase;
1777 if (carpr.carpr_advskew >= 255) {
1781 sc->sc_advskew = carpr.carpr_advskew;
1782 if (carpr.carpr_key[0] != '\0') {
1783 bcopy(carpr.carpr_key, sc->sc_key, sizeof(sc->sc_key));
1784 carp_hmac_prepare(sc);
1786 if (sc->sc_state != INIT &&
1787 carpr.carpr_state != sc->sc_state) {
1788 switch (carpr.carpr_state) {
1790 callout_stop(&sc->sc_ad_tmo);
1791 carp_set_state(sc, BACKUP,
1792 "user requested via ifconfig");
1797 carp_master_down_locked(sc,
1798 "user requested via ifconfig");
1810 if (carpr.carpr_vhid < 0 || carpr.carpr_vhid > CARP_MAXVHID) {
1814 if (carpr.carpr_count < 1) {
1818 if (ifp->if_carp == NULL) {
1823 priveleged = (priv_check(td, PRIV_NETINET_CARP) == 0);
1824 if (carpr.carpr_vhid != 0) {
1825 IFNET_FOREACH_CARP(ifp, sc)
1826 if (sc->sc_vhid == carpr.carpr_vhid)
1832 carp_carprcp(&carpr, sc, priveleged);
1833 error = copyout(&carpr, ifr_data_get_ptr(ifr),
1839 IFNET_FOREACH_CARP(ifp, sc)
1842 if (count > carpr.carpr_count) {
1843 CIF_UNLOCK(ifp->if_carp);
1849 IFNET_FOREACH_CARP(ifp, sc) {
1850 carp_carprcp(&carpr, sc, priveleged);
1851 carpr.carpr_count = count;
1852 error = copyout(&carpr,
1853 (char *)ifr_data_get_ptr(ifr) +
1854 (i * sizeof(carpr)), sizeof(carpr));
1856 CIF_UNLOCK(ifp->if_carp);
1867 sx_xunlock(&carp_sx);
1878 carp_get_vhid(struct ifaddr *ifa)
1881 if (ifa == NULL || ifa->ifa_carp == NULL)
1884 return (ifa->ifa_carp->sc_vhid);
1888 carp_attach(struct ifaddr *ifa, int vhid)
1890 struct ifnet *ifp = ifa->ifa_ifp;
1891 struct carp_if *cif = ifp->if_carp;
1892 struct carp_softc *sc;
1895 KASSERT(ifa->ifa_carp == NULL, ("%s: ifa %p attached", __func__, ifa));
1897 switch (ifa->ifa_addr->sa_family) {
1906 return (EPROTOTYPE);
1910 if (ifp->if_carp == NULL) {
1911 sx_xunlock(&carp_sx);
1912 return (ENOPROTOOPT);
1915 IFNET_FOREACH_CARP(ifp, sc)
1916 if (sc->sc_vhid == vhid)
1919 sx_xunlock(&carp_sx);
1923 error = carp_multicast_setup(cif, ifa->ifa_addr->sa_family);
1926 sx_xunlock(&carp_sx);
1930 index = sc->sc_naddrs + sc->sc_naddrs6 + 1;
1931 if (index > sc->sc_ifasiz / sizeof(struct ifaddr *))
1934 switch (ifa->ifa_addr->sa_family) {
1952 sc->sc_ifas[index - 1] = ifa;
1954 carp_hmac_prepare(sc);
1958 sx_xunlock(&carp_sx);
1964 carp_detach(struct ifaddr *ifa, bool keep_cif)
1966 struct ifnet *ifp = ifa->ifa_ifp;
1967 struct carp_if *cif = ifp->if_carp;
1968 struct carp_softc *sc = ifa->ifa_carp;
1971 KASSERT(sc != NULL, ("%s: %p not attached", __func__, ifa));
1977 index = sc->sc_naddrs + sc->sc_naddrs6;
1978 for (i = 0; i < index; i++)
1979 if (sc->sc_ifas[i] == ifa)
1981 KASSERT(i < index, ("%s: %p no backref", __func__, ifa));
1982 for (; i < index - 1; i++)
1983 sc->sc_ifas[i] = sc->sc_ifas[i+1];
1984 sc->sc_ifas[index - 1] = NULL;
1986 switch (ifa->ifa_addr->sa_family) {
2001 carp_ifa_delroute(ifa);
2002 carp_multicast_cleanup(cif, ifa->ifa_addr->sa_family);
2004 ifa->ifa_carp = NULL;
2007 carp_hmac_prepare(sc);
2010 if (!keep_cif && sc->sc_naddrs == 0 && sc->sc_naddrs6 == 0)
2018 sx_xunlock(&carp_sx);
2022 carp_set_state(struct carp_softc *sc, int state, const char *reason)
2025 CARP_LOCK_ASSERT(sc);
2027 if (sc->sc_state != state) {
2028 const char *carp_states[] = { CARP_STATES };
2029 char subsys[IFNAMSIZ+5];
2031 snprintf(subsys, IFNAMSIZ+5, "%u@%s", sc->sc_vhid,
2032 sc->sc_carpdev->if_xname);
2034 CARP_LOG("%s: %s -> %s (%s)\n", subsys,
2035 carp_states[sc->sc_state], carp_states[state], reason);
2037 sc->sc_state = state;
2039 devctl_notify("CARP", subsys, carp_states[state], NULL);
2044 carp_linkstate(struct ifnet *ifp)
2046 struct carp_softc *sc;
2048 CIF_LOCK(ifp->if_carp);
2049 IFNET_FOREACH_CARP(ifp, sc) {
2054 CIF_UNLOCK(ifp->if_carp);
2058 carp_sc_state(struct carp_softc *sc)
2061 CARP_LOCK_ASSERT(sc);
2063 if (sc->sc_carpdev->if_link_state != LINK_STATE_UP ||
2064 !(sc->sc_carpdev->if_flags & IFF_UP) ||
2066 callout_stop(&sc->sc_ad_tmo);
2068 callout_stop(&sc->sc_md_tmo);
2071 callout_stop(&sc->sc_md6_tmo);
2073 carp_set_state(sc, INIT, "hardware interface down");
2075 if (!sc->sc_suppress)
2076 carp_demote_adj(V_carp_ifdown_adj, "interface down");
2077 sc->sc_suppress = 1;
2079 carp_set_state(sc, INIT, "hardware interface up");
2081 if (sc->sc_suppress)
2082 carp_demote_adj(-V_carp_ifdown_adj, "interface up");
2083 sc->sc_suppress = 0;
2088 carp_demote_adj(int adj, char *reason)
2090 atomic_add_int(&V_carp_demotion, adj);
2091 CARP_LOG("demoted by %d to %d (%s)\n", adj, V_carp_demotion, reason);
2092 taskqueue_enqueue(taskqueue_swi, &carp_sendall_task);
2096 carp_allow_sysctl(SYSCTL_HANDLER_ARGS)
2099 struct carp_softc *sc;
2102 error = sysctl_handle_int(oidp, &new, 0, req);
2103 if (error || !req->newptr)
2106 if (V_carp_allow != new) {
2109 mtx_lock(&carp_mtx);
2110 LIST_FOREACH(sc, &carp_list, sc_next) {
2112 if (curvnet == sc->sc_carpdev->if_vnet)
2116 mtx_unlock(&carp_mtx);
2123 carp_dscp_sysctl(SYSCTL_HANDLER_ARGS)
2128 error = sysctl_handle_int(oidp, &new, 0, req);
2129 if (error || !req->newptr)
2132 if (new < 0 || new > 63)
2141 carp_demote_adj_sysctl(SYSCTL_HANDLER_ARGS)
2145 new = V_carp_demotion;
2146 error = sysctl_handle_int(oidp, &new, 0, req);
2147 if (error || !req->newptr)
2150 carp_demote_adj(new, "sysctl");
2156 extern struct domain inetdomain;
2157 static struct protosw in_carp_protosw = {
2158 .pr_type = SOCK_RAW,
2159 .pr_domain = &inetdomain,
2160 .pr_protocol = IPPROTO_CARP,
2161 .pr_flags = PR_ATOMIC|PR_ADDR,
2162 .pr_input = carp_input,
2163 .pr_output = rip_output,
2164 .pr_ctloutput = rip_ctloutput,
2165 .pr_usrreqs = &rip_usrreqs
2170 extern struct domain inet6domain;
2171 static struct protosw in6_carp_protosw = {
2172 .pr_type = SOCK_RAW,
2173 .pr_domain = &inet6domain,
2174 .pr_protocol = IPPROTO_CARP,
2175 .pr_flags = PR_ATOMIC|PR_ADDR,
2176 .pr_input = carp6_input,
2177 .pr_output = rip6_output,
2178 .pr_ctloutput = rip6_ctloutput,
2179 .pr_usrreqs = &rip6_usrreqs
2184 #if defined(__i386__)
2186 * XXX This is a hack to work around an absolute relocation outside
2187 * set_vnet by one (on the stop symbol) for carpstats. Add a dummy variable
2188 * to the end of the file in the hope that the linker will just keep the
2189 * order (as it seems to do at the moment). It is understood to be fragile.
2190 * See PR 230857 for a longer discussion of the problem and the referenced
2191 * review for possible alternate solutions. Each is a hack; we just need
2192 * the least intrusive one for the next release.
2194 VNET_DEFINE(char, carp_zzz) = 0xde;
2199 carp_mod_cleanup(void)
2203 if (proto_reg[CARP_INET] == 0) {
2204 (void)ipproto_unregister(IPPROTO_CARP);
2205 pf_proto_unregister(PF_INET, IPPROTO_CARP, SOCK_RAW);
2206 proto_reg[CARP_INET] = -1;
2208 carp_iamatch_p = NULL;
2211 if (proto_reg[CARP_INET6] == 0) {
2212 (void)ip6proto_unregister(IPPROTO_CARP);
2213 pf_proto_unregister(PF_INET6, IPPROTO_CARP, SOCK_RAW);
2214 proto_reg[CARP_INET6] = -1;
2216 carp_iamatch6_p = NULL;
2217 carp_macmatch6_p = NULL;
2219 carp_ioctl_p = NULL;
2220 carp_attach_p = NULL;
2221 carp_detach_p = NULL;
2222 carp_get_vhid_p = NULL;
2223 carp_linkstate_p = NULL;
2224 carp_forus_p = NULL;
2225 carp_output_p = NULL;
2226 carp_demote_adj_p = NULL;
2227 carp_master_p = NULL;
2228 mtx_unlock(&carp_mtx);
2229 taskqueue_drain(taskqueue_swi, &carp_sendall_task);
2230 mtx_destroy(&carp_mtx);
2231 sx_destroy(&carp_sx);
2239 mtx_init(&carp_mtx, "carp_mtx", NULL, MTX_DEF);
2240 sx_init(&carp_sx, "carp_sx");
2241 LIST_INIT(&carp_list);
2242 carp_get_vhid_p = carp_get_vhid;
2243 carp_forus_p = carp_forus;
2244 carp_output_p = carp_output;
2245 carp_linkstate_p = carp_linkstate;
2246 carp_ioctl_p = carp_ioctl;
2247 carp_attach_p = carp_attach;
2248 carp_detach_p = carp_detach;
2249 carp_demote_adj_p = carp_demote_adj;
2250 carp_master_p = carp_master;
2252 carp_iamatch6_p = carp_iamatch6;
2253 carp_macmatch6_p = carp_macmatch6;
2254 proto_reg[CARP_INET6] = pf_proto_register(PF_INET6,
2255 (struct protosw *)&in6_carp_protosw);
2256 if (proto_reg[CARP_INET6]) {
2257 printf("carp: error %d attaching to PF_INET6\n",
2258 proto_reg[CARP_INET6]);
2260 return (proto_reg[CARP_INET6]);
2262 err = ip6proto_register(IPPROTO_CARP);
2264 printf("carp: error %d registering with INET6\n", err);
2270 carp_iamatch_p = carp_iamatch;
2271 proto_reg[CARP_INET] = pf_proto_register(PF_INET, &in_carp_protosw);
2272 if (proto_reg[CARP_INET]) {
2273 printf("carp: error %d attaching to PF_INET\n",
2274 proto_reg[CARP_INET]);
2276 return (proto_reg[CARP_INET]);
2278 err = ipproto_register(IPPROTO_CARP);
2280 printf("carp: error %d registering with INET\n", err);
2289 carp_modevent(module_t mod, int type, void *data)
2293 return carp_mod_load();
2296 mtx_lock(&carp_mtx);
2297 if (LIST_EMPTY(&carp_list))
2300 mtx_unlock(&carp_mtx);
2312 static moduledata_t carp_mod = {
2318 DECLARE_MODULE(carp, carp_mod, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY);