2 * Copyright (c) 2002 Michael Shalayeff. All rights reserved.
3 * Copyright (c) 2003 Ryan McBride. All rights reserved.
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
15 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
16 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
17 * IN NO EVENT SHALL THE AUTHOR OR HIS RELATIVES BE LIABLE FOR ANY DIRECT,
18 * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
19 * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
20 * SERVICES; LOSS OF MIND, USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
22 * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
23 * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
24 * THE POSSIBILITY OF SUCH DAMAGE.
27 * $FreeBSD: src/sys/netinet/ip_carp.c,v 1.48 2007/02/02 09:39:09 glebius Exp $
32 #include "opt_inet6.h"
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/kernel.h>
37 #include <sys/in_cksum.h>
38 #include <sys/limits.h>
39 #include <sys/malloc.h>
41 #include <sys/msgport2.h>
45 #include <sys/sockio.h>
46 #include <sys/socket.h>
47 #include <sys/sysctl.h>
48 #include <sys/syslog.h>
49 #include <sys/thread.h>
51 #include <machine/stdarg.h>
52 #include <crypto/sha1.h>
55 #include <net/ethernet.h>
57 #include <net/if_dl.h>
58 #include <net/if_types.h>
59 #include <net/route.h>
60 #include <net/if_clone.h>
61 #include <net/if_var.h>
62 #include <net/ifq_var.h>
63 #include <net/netmsg2.h>
66 #include <netinet/in.h>
67 #include <netinet/in_var.h>
68 #include <netinet/in_systm.h>
69 #include <netinet/ip.h>
70 #include <netinet/ip_var.h>
71 #include <netinet/if_ether.h>
75 #include <netinet/icmp6.h>
76 #include <netinet/ip6.h>
77 #include <netinet6/ip6_var.h>
78 #include <netinet6/scope6_var.h>
79 #include <netinet6/nd6.h>
82 #include <netinet/ip_carp.h>
84 #define CARP_IFNAME "carp"
85 #define CARP_IS_RUNNING(ifp) \
86 (((ifp)->if_flags & (IFF_UP | IFF_RUNNING)) == (IFF_UP | IFF_RUNNING))
89 uint32_t vha_flags; /* CARP_VHAF_ */
90 struct in_ifaddr *vha_ia; /* carp address */
91 struct in_ifaddr *vha_iaback; /* backing address */
92 TAILQ_ENTRY(carp_vhaddr) vha_link;
94 TAILQ_HEAD(carp_vhaddr_list, carp_vhaddr);
98 struct ifnet *sc_carpdev; /* parent interface */
99 struct carp_vhaddr_list sc_vha_list; /* virtual addr list */
101 const struct in_ifaddr *sc_ia; /* primary iface address v4 */
102 struct ip_moptions sc_imo;
105 struct in6_ifaddr *sc_ia6; /* primary iface address v6 */
106 struct ip6_moptions sc_im6o;
108 TAILQ_ENTRY(carp_softc) sc_list;
110 enum { INIT = 0, BACKUP, MASTER }
116 int sc_sendad_errors;
117 #define CARP_SENDAD_MAX_ERRORS 3
118 int sc_sendad_success;
119 #define CARP_SENDAD_MIN_SUCCESS 3
123 int sc_naddrs; /* actually used IPv4 vha */
125 int sc_advbase; /* seconds */
130 #define CARP_HMAC_PAD 64
131 unsigned char sc_key[CARP_KEY_LEN];
132 unsigned char sc_pad[CARP_HMAC_PAD];
135 struct callout sc_ad_tmo; /* advertisement timeout */
136 struct callout sc_md_tmo; /* master down timeout */
137 struct callout sc_md6_tmo; /* master down timeout */
139 LIST_ENTRY(carp_softc) sc_next; /* Interface clue */
142 #define sc_if arpcom.ac_if
145 TAILQ_HEAD(, carp_softc) vhif_vrs;
149 struct netmsg_base base;
150 struct ifnet *nc_carpdev;
151 struct carp_softc *nc_softc;
154 SYSCTL_DECL(_net_inet_carp);
156 static int carp_opts[CARPCTL_MAXID] = { 0, 1, 0, 1, 0, 0 }; /* XXX for now */
157 SYSCTL_INT(_net_inet_carp, CARPCTL_ALLOW, allow, CTLFLAG_RW,
158 &carp_opts[CARPCTL_ALLOW], 0, "Accept incoming CARP packets");
159 SYSCTL_INT(_net_inet_carp, CARPCTL_PREEMPT, preempt, CTLFLAG_RW,
160 &carp_opts[CARPCTL_PREEMPT], 0, "high-priority backup preemption mode");
161 SYSCTL_INT(_net_inet_carp, CARPCTL_LOG, log, CTLFLAG_RW,
162 &carp_opts[CARPCTL_LOG], 0, "log bad carp packets");
163 SYSCTL_INT(_net_inet_carp, CARPCTL_ARPBALANCE, arpbalance, CTLFLAG_RW,
164 &carp_opts[CARPCTL_ARPBALANCE], 0, "balance arp responses");
166 static int carp_suppress_preempt = 0;
167 SYSCTL_INT(_net_inet_carp, OID_AUTO, suppress_preempt, CTLFLAG_RD,
168 &carp_suppress_preempt, 0, "Preemption is suppressed");
170 static struct carpstats carpstats;
171 SYSCTL_STRUCT(_net_inet_carp, CARPCTL_STATS, stats, CTLFLAG_RW,
172 &carpstats, carpstats,
173 "CARP statistics (struct carpstats, netinet/ip_carp.h)");
175 #define CARP_LOG(...) do { \
176 if (carp_opts[CARPCTL_LOG] > 0) \
177 log(LOG_INFO, __VA_ARGS__); \
180 #define CARP_DEBUG(...) do { \
181 if (carp_opts[CARPCTL_LOG] > 1) \
182 log(LOG_DEBUG, __VA_ARGS__); \
185 static struct lwkt_token carp_tok = LWKT_TOKEN_INITIALIZER(carp_token);
187 static void carp_hmac_prepare(struct carp_softc *);
188 static void carp_hmac_generate(struct carp_softc *, uint32_t *,
190 static int carp_hmac_verify(struct carp_softc *, uint32_t *,
192 static void carp_setroute(struct carp_softc *, int);
193 static void carp_proto_input_c(struct carp_softc *, struct mbuf *,
194 struct carp_header *, sa_family_t);
195 static int carp_clone_create(struct if_clone *, int, caddr_t);
196 static int carp_clone_destroy(struct ifnet *);
197 static void carp_detach(struct carp_softc *, int, boolean_t);
198 static void carp_prepare_ad(struct carp_softc *, struct carp_header *);
199 static void carp_send_ad_all(void);
200 static void carp_send_ad_timeout(void *);
201 static void carp_send_ad(struct carp_softc *);
202 static void carp_send_arp(struct carp_softc *);
203 static void carp_master_down_timeout(void *);
204 static void carp_master_down(struct carp_softc *);
205 static void carp_setrun(struct carp_softc *, sa_family_t);
206 static void carp_set_state(struct carp_softc *, int);
207 static struct ifnet *carp_forus(struct carp_if *, const uint8_t *);
209 static void carp_init(void *);
210 static int carp_ioctl(struct ifnet *, u_long, caddr_t, struct ucred *);
211 static int carp_output(struct ifnet *, struct mbuf *, struct sockaddr *,
213 static void carp_start(struct ifnet *);
214 static void carp_serialize(struct ifnet *, enum ifnet_serialize);
215 static void carp_deserialize(struct ifnet *, enum ifnet_serialize);
216 static int carp_tryserialize(struct ifnet *, enum ifnet_serialize);
218 static void carp_serialize_assert(struct ifnet *, enum ifnet_serialize,
222 static void carp_multicast_cleanup(struct carp_softc *);
223 static void carp_add_addr(struct carp_softc *, struct ifaddr *);
224 static void carp_del_addr(struct carp_softc *, struct ifaddr *);
225 static void carp_config_addr(struct carp_softc *, struct ifaddr *);
226 static void carp_link_addrs(struct carp_softc *, struct ifnet *,
228 static void carp_unlink_addrs(struct carp_softc *, struct ifnet *,
230 static void carp_update_addrs(struct carp_softc *, struct ifaddr *);
232 static int carp_get_vhaddr(struct carp_softc *, struct ifdrv *);
233 static int carp_config_vhaddr(struct carp_softc *, struct carp_vhaddr *,
235 static int carp_activate_vhaddr(struct carp_softc *, struct carp_vhaddr *,
236 struct ifnet *, struct in_ifaddr *, int);
237 static void carp_deactivate_vhaddr(struct carp_softc *,
238 struct carp_vhaddr *, boolean_t);
239 static int carp_addroute_vhaddr(struct carp_softc *, struct carp_vhaddr *);
240 static void carp_delroute_vhaddr(struct carp_softc *, struct carp_vhaddr *,
243 static void carp_sc_state(struct carp_softc *);
245 static void carp_send_na(struct carp_softc *);
247 static int carp_set_addr6(struct carp_softc *, struct sockaddr_in6 *);
248 static int carp_del_addr6(struct carp_softc *, struct sockaddr_in6 *);
250 static void carp_multicast6_cleanup(struct carp_softc *);
252 static void carp_stop(struct carp_softc *, int);
253 static void carp_suspend(struct carp_softc *, int);
254 static void carp_ioctl_stop(struct carp_softc *);
256 static void carp_ifaddr(void *, struct ifnet *, enum ifaddr_event,
258 static void carp_ifdetach(void *, struct ifnet *);
260 static void carp_ifdetach_dispatch(netmsg_t);
261 static void carp_clone_destroy_dispatch(netmsg_t);
262 static void carp_init_dispatch(netmsg_t);
263 static void carp_ioctl_stop_dispatch(netmsg_t);
265 static MALLOC_DEFINE(M_CARP, "CARP", "CARP interfaces");
267 static LIST_HEAD(, carp_softc) carpif_list;
269 static struct if_clone carp_cloner =
270 IF_CLONE_INITIALIZER(CARP_IFNAME, carp_clone_create, carp_clone_destroy,
273 static uint8_t carp_etheraddr[ETHER_ADDR_LEN] = { 0, 0, 0x5e, 0, 1, 0 };
275 static eventhandler_tag carp_ifdetach_event;
276 static eventhandler_tag carp_ifaddr_event;
279 carp_insert_vhaddr(struct carp_softc *sc, struct carp_vhaddr *vha_new)
281 struct carp_vhaddr *vha;
282 u_long new_addr, addr;
284 KKASSERT((vha_new->vha_flags & CARP_VHAF_ONLIST) == 0);
287 * Virtual address list is sorted; smaller one first
289 new_addr = ntohl(vha_new->vha_ia->ia_addr.sin_addr.s_addr);
291 TAILQ_FOREACH(vha, &sc->sc_vha_list, vha_link) {
292 addr = ntohl(vha->vha_ia->ia_addr.sin_addr.s_addr);
298 TAILQ_INSERT_TAIL(&sc->sc_vha_list, vha_new, vha_link);
300 TAILQ_INSERT_BEFORE(vha, vha_new, vha_link);
301 vha_new->vha_flags |= CARP_VHAF_ONLIST;
305 carp_remove_vhaddr(struct carp_softc *sc, struct carp_vhaddr *vha)
307 KKASSERT(vha->vha_flags & CARP_VHAF_ONLIST);
308 vha->vha_flags &= ~CARP_VHAF_ONLIST;
309 TAILQ_REMOVE(&sc->sc_vha_list, vha, vha_link);
313 carp_hmac_prepare(struct carp_softc *sc)
315 uint8_t version = CARP_VERSION, type = CARP_ADVERTISEMENT;
316 uint8_t vhid = sc->sc_vhid & 0xff;
319 struct ifaddr_container *ifac;
323 struct carp_vhaddr *vha;
326 /* XXX: possible race here */
328 /* compute ipad from key */
329 bzero(sc->sc_pad, sizeof(sc->sc_pad));
330 bcopy(sc->sc_key, sc->sc_pad, sizeof(sc->sc_key));
331 for (i = 0; i < sizeof(sc->sc_pad); i++)
332 sc->sc_pad[i] ^= 0x36;
334 /* precompute first part of inner hash */
335 SHA1Init(&sc->sc_sha1);
336 SHA1Update(&sc->sc_sha1, sc->sc_pad, sizeof(sc->sc_pad));
337 SHA1Update(&sc->sc_sha1, (void *)&version, sizeof(version));
338 SHA1Update(&sc->sc_sha1, (void *)&type, sizeof(type));
339 SHA1Update(&sc->sc_sha1, (void *)&vhid, sizeof(vhid));
341 TAILQ_FOREACH(vha, &sc->sc_vha_list, vha_link) {
342 SHA1Update(&sc->sc_sha1,
343 (const uint8_t *)&vha->vha_ia->ia_addr.sin_addr,
344 sizeof(struct in_addr));
348 TAILQ_FOREACH(ifac, &sc->sc_if.if_addrheads[mycpuid], ifa_link) {
349 struct ifaddr *ifa = ifac->ifa;
351 if (ifa->ifa_addr->sa_family == AF_INET6) {
352 in6 = ifatoia6(ifa)->ia_addr.sin6_addr;
353 in6_clearscope(&in6);
354 SHA1Update(&sc->sc_sha1, (void *)&in6, sizeof(in6));
359 /* convert ipad to opad */
360 for (i = 0; i < sizeof(sc->sc_pad); i++)
361 sc->sc_pad[i] ^= 0x36 ^ 0x5c;
365 carp_hmac_generate(struct carp_softc *sc, uint32_t counter[2],
366 unsigned char md[20])
370 /* fetch first half of inner hash */
371 bcopy(&sc->sc_sha1, &sha1ctx, sizeof(sha1ctx));
373 SHA1Update(&sha1ctx, (void *)counter, sizeof(sc->sc_counter));
374 SHA1Final(md, &sha1ctx);
378 SHA1Update(&sha1ctx, sc->sc_pad, sizeof(sc->sc_pad));
379 SHA1Update(&sha1ctx, md, 20);
380 SHA1Final(md, &sha1ctx);
384 carp_hmac_verify(struct carp_softc *sc, uint32_t counter[2],
385 unsigned char md[20])
387 unsigned char md2[20];
389 carp_hmac_generate(sc, counter, md2);
390 return (bcmp(md, md2, sizeof(md2)));
394 carp_setroute(struct carp_softc *sc, int cmd)
397 struct ifaddr_container *ifac;
399 struct carp_vhaddr *vha;
401 KKASSERT(cmd == RTM_DELETE || cmd == RTM_ADD);
403 TAILQ_FOREACH(vha, &sc->sc_vha_list, vha_link) {
404 if (vha->vha_iaback == NULL)
406 if (cmd == RTM_DELETE)
407 carp_delroute_vhaddr(sc, vha, FALSE);
409 carp_addroute_vhaddr(sc, vha);
413 TAILQ_FOREACH(ifac, &sc->sc_if.if_addrheads[mycpuid], ifa_link) {
414 struct ifaddr *ifa = ifac->ifa;
416 if (ifa->ifa_addr->sa_family == AF_INET6) {
427 carp_clone_create(struct if_clone *ifc, int unit, caddr_t param __unused)
429 struct carp_softc *sc;
432 sc = kmalloc(sizeof(*sc), M_CARP, M_WAITOK | M_ZERO);
436 sc->sc_advbase = CARP_DFLTINTV;
437 sc->sc_vhid = -1; /* required setting */
439 sc->sc_init_counter = 1;
443 TAILQ_INIT(&sc->sc_vha_list);
446 sc->sc_im6o.im6o_multicast_hlim = CARP_DFLTTL;
449 callout_init_mp(&sc->sc_ad_tmo);
450 callout_init_mp(&sc->sc_md_tmo);
451 callout_init_mp(&sc->sc_md6_tmo);
453 if_initname(ifp, CARP_IFNAME, unit);
455 ifp->if_flags = IFF_BROADCAST | IFF_SIMPLEX | IFF_MULTICAST;
456 ifp->if_init = carp_init;
457 ifp->if_ioctl = carp_ioctl;
458 ifp->if_start = carp_start;
459 ifp->if_serialize = carp_serialize;
460 ifp->if_deserialize = carp_deserialize;
461 ifp->if_tryserialize = carp_tryserialize;
463 ifp->if_serialize_assert = carp_serialize_assert;
465 ifq_set_maxlen(&ifp->if_snd, ifqmaxlen);
466 ifq_set_ready(&ifp->if_snd);
468 ether_ifattach(ifp, carp_etheraddr, NULL);
470 ifp->if_type = IFT_CARP;
471 ifp->if_output = carp_output;
474 LIST_INSERT_HEAD(&carpif_list, sc, sc_next);
481 carp_clone_destroy_dispatch(netmsg_t msg)
483 struct netmsg_carp *cmsg = (struct netmsg_carp *)msg;
484 struct carp_softc *sc = cmsg->nc_softc;
489 carp_detach(sc, 1, FALSE);
493 lwkt_replymsg(&cmsg->base.lmsg, 0);
497 carp_clone_destroy(struct ifnet *ifp)
499 struct carp_softc *sc = ifp->if_softc;
500 struct netmsg_carp cmsg;
502 bzero(&cmsg, sizeof(cmsg));
503 netmsg_init(&cmsg.base, NULL, &curthread->td_msgport, 0,
504 carp_clone_destroy_dispatch);
507 lwkt_domsg(cpu_portfn(0), &cmsg.base.lmsg, 0);
510 LIST_REMOVE(sc, sc_next);
516 KASSERT(sc->sc_naddrs == 0, ("certain inet address is still active\n"));
523 carp_detach(struct carp_softc *sc, int detach, boolean_t del_iaback)
527 carp_suspend(sc, detach);
529 carp_multicast_cleanup(sc);
531 carp_multicast6_cleanup(sc);
534 if (!sc->sc_dead && detach) {
535 struct carp_vhaddr *vha;
537 TAILQ_FOREACH(vha, &sc->sc_vha_list, vha_link)
538 carp_deactivate_vhaddr(sc, vha, del_iaback);
539 KKASSERT(sc->sc_naddrs == 0);
542 if (sc->sc_carpdev != NULL) {
543 cif = sc->sc_carpdev->if_carp;
544 TAILQ_REMOVE(&cif->vhif_vrs, sc, sc_list);
545 if (TAILQ_EMPTY(&cif->vhif_vrs)) {
546 ifpromisc(sc->sc_carpdev, 0);
547 sc->sc_carpdev->if_carp = NULL;
550 sc->sc_carpdev = NULL;
556 carp_ifdetach_dispatch(netmsg_t msg)
558 struct netmsg_carp *cmsg = (struct netmsg_carp *)msg;
559 struct ifnet *ifp = cmsg->nc_carpdev;
560 struct carp_if *cif = ifp->if_carp;
561 struct carp_softc *sc;
565 while (ifp->if_carp &&
566 (sc = TAILQ_FIRST(&cif->vhif_vrs)) != NULL)
567 carp_detach(sc, 1, TRUE);
571 lwkt_replymsg(&cmsg->base.lmsg, 0);
574 /* Detach an interface from the carp. */
576 carp_ifdetach(void *arg __unused, struct ifnet *ifp)
578 struct netmsg_carp cmsg;
580 ASSERT_IFNET_NOT_SERIALIZED_ALL(ifp);
582 bzero(&cmsg, sizeof(cmsg));
583 netmsg_init(&cmsg.base, NULL, &curthread->td_msgport, 0,
584 carp_ifdetach_dispatch);
585 cmsg.nc_carpdev = ifp;
587 lwkt_domsg(cpu_portfn(0), &cmsg.base.lmsg, 0);
591 * process input packet.
592 * we have rearranged checks order compared to the rfc,
593 * but it seems more efficient this way or not possible otherwise.
596 carp_proto_input(struct mbuf **mp, int *offp, int proto)
598 struct mbuf *m = *mp;
599 struct ip *ip = mtod(m, struct ip *);
600 struct ifnet *ifp = m->m_pkthdr.rcvif;
601 struct carp_header *ch;
602 struct carp_softc *sc;
610 carpstats.carps_ipackets++;
612 if (!carp_opts[CARPCTL_ALLOW]) {
617 /* Check if received on a valid carp interface */
618 if (ifp->if_type != IFT_CARP) {
619 carpstats.carps_badif++;
620 CARP_LOG("carp_proto_input: packet received on non-carp "
621 "interface: %s\n", ifp->if_xname);
626 if (!CARP_IS_RUNNING(ifp)) {
627 carpstats.carps_badif++;
628 CARP_LOG("carp_proto_input: packet received on stopped carp "
629 "interface: %s\n", ifp->if_xname);
635 if (sc->sc_carpdev == NULL) {
636 carpstats.carps_badif++;
637 CARP_LOG("carp_proto_input: packet received on defunc carp "
638 "interface: %s\n", ifp->if_xname);
643 if (!IN_MULTICAST(ntohl(ip->ip_dst.s_addr))) {
644 carpstats.carps_badif++;
645 CARP_LOG("carp_proto_input: non-mcast packet on "
646 "interface: %s\n", ifp->if_xname);
651 /* Verify that the IP TTL is CARP_DFLTTL. */
652 if (ip->ip_ttl != CARP_DFLTTL) {
653 carpstats.carps_badttl++;
654 CARP_LOG("carp_proto_input: received ttl %d != %d on %s\n",
655 ip->ip_ttl, CARP_DFLTTL, ifp->if_xname);
660 /* Minimal CARP packet size */
661 len = iphlen + sizeof(*ch);
664 * Verify that the received packet length is
665 * not less than the CARP header
667 if (m->m_pkthdr.len < len) {
668 carpstats.carps_badlen++;
669 CARP_LOG("packet too short %d on %s\n", m->m_pkthdr.len,
675 /* Make sure that CARP header is contiguous */
676 if (len > m->m_len) {
677 m = m_pullup(m, len);
679 carpstats.carps_hdrops++;
680 CARP_LOG("carp_proto_input: m_pullup failed\n");
683 ip = mtod(m, struct ip *);
685 ch = (struct carp_header *)((uint8_t *)ip + iphlen);
687 /* Verify the CARP checksum */
688 if (in_cksum_skip(m, len, iphlen)) {
689 carpstats.carps_badsum++;
690 CARP_LOG("carp_proto_input: checksum failed on %s\n",
695 carp_proto_input_c(sc, m, ch, AF_INET);
698 return(IPPROTO_DONE);
703 carp6_proto_input(struct mbuf **mp, int *offp, int proto)
705 struct mbuf *m = *mp;
706 struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
707 struct ifnet *ifp = m->m_pkthdr.rcvif;
708 struct carp_header *ch;
709 struct carp_softc *sc;
714 carpstats.carps_ipackets6++;
716 if (!carp_opts[CARPCTL_ALLOW]) {
721 /* check if received on a valid carp interface */
722 if (ifp->if_type != IFT_CARP) {
723 carpstats.carps_badif++;
724 CARP_LOG("carp6_proto_input: packet received on non-carp "
725 "interface: %s\n", ifp->if_xname);
730 if (!CARP_IS_RUNNING(ifp)) {
731 carpstats.carps_badif++;
732 CARP_LOG("carp_proto_input: packet received on stopped carp "
733 "interface: %s\n", ifp->if_xname);
739 if (sc->sc_carpdev == NULL) {
740 carpstats.carps_badif++;
741 CARP_LOG("carp6_proto_input: packet received on defunc-carp "
742 "interface: %s\n", ifp->if_xname);
747 /* verify that the IP TTL is 255 */
748 if (ip6->ip6_hlim != CARP_DFLTTL) {
749 carpstats.carps_badttl++;
750 CARP_LOG("carp6_proto_input: received ttl %d != 255 on %s\n",
751 ip6->ip6_hlim, ifp->if_xname);
756 /* verify that we have a complete carp packet */
758 IP6_EXTHDR_GET(ch, struct carp_header *, m, *offp, sizeof(*ch));
760 carpstats.carps_badlen++;
761 CARP_LOG("carp6_proto_input: packet size %u too small\n", len);
765 /* verify the CARP checksum */
766 if (in_cksum_range(m, 0, *offp, sizeof(*ch))) {
767 carpstats.carps_badsum++;
768 CARP_LOG("carp6_proto_input: checksum failed, on %s\n",
774 carp_proto_input_c(sc, m, ch, AF_INET6);
777 return (IPPROTO_DONE);
782 carp_proto_input_c(struct carp_softc *sc, struct mbuf *m,
783 struct carp_header *ch, sa_family_t af)
786 uint64_t tmp_counter;
787 struct timeval sc_tv, ch_tv;
789 if (sc->sc_vhid != ch->carp_vhid) {
791 * CARP uses multicast, however, multicast packets
792 * are tapped to all CARP interfaces on the physical
793 * interface receiving the CARP packets, so we don't
794 * update any stats here.
801 /* verify the CARP version. */
802 if (ch->carp_version != CARP_VERSION) {
803 carpstats.carps_badver++;
804 CARP_LOG("%s; invalid version %d\n", cifp->if_xname,
810 /* verify the hash */
811 if (carp_hmac_verify(sc, ch->carp_counter, ch->carp_md)) {
812 carpstats.carps_badauth++;
813 CARP_LOG("%s: incorrect hash\n", cifp->if_xname);
818 tmp_counter = ntohl(ch->carp_counter[0]);
819 tmp_counter = tmp_counter<<32;
820 tmp_counter += ntohl(ch->carp_counter[1]);
822 /* XXX Replay protection goes here */
824 sc->sc_init_counter = 0;
825 sc->sc_counter = tmp_counter;
827 sc_tv.tv_sec = sc->sc_advbase;
828 if (carp_suppress_preempt && sc->sc_advskew < 240)
829 sc_tv.tv_usec = 240 * 1000000 / 256;
831 sc_tv.tv_usec = sc->sc_advskew * 1000000 / 256;
832 ch_tv.tv_sec = ch->carp_advbase;
833 ch_tv.tv_usec = ch->carp_advskew * 1000000 / 256;
835 switch (sc->sc_state) {
841 * If we receive an advertisement from a master who's going to
842 * be more frequent than us, go into BACKUP state.
844 if (timevalcmp(&sc_tv, &ch_tv, >) ||
845 timevalcmp(&sc_tv, &ch_tv, ==)) {
846 callout_stop(&sc->sc_ad_tmo);
847 CARP_DEBUG("%s: MASTER -> BACKUP "
848 "(more frequent advertisement received)\n",
850 carp_set_state(sc, BACKUP);
852 carp_setroute(sc, RTM_DELETE);
858 * If we're pre-empting masters who advertise slower than us,
859 * and this one claims to be slower, treat him as down.
861 if (carp_opts[CARPCTL_PREEMPT] &&
862 timevalcmp(&sc_tv, &ch_tv, <)) {
863 CARP_DEBUG("%s: BACKUP -> MASTER "
864 "(preempting a slower master)\n", cifp->if_xname);
865 carp_master_down(sc);
870 * If the master is going to advertise at such a low frequency
871 * that he's guaranteed to time out, we'd might as well just
872 * treat him as timed out now.
874 sc_tv.tv_sec = sc->sc_advbase * 3;
875 if (timevalcmp(&sc_tv, &ch_tv, <)) {
876 CARP_DEBUG("%s: BACKUP -> MASTER (master timed out)\n",
878 carp_master_down(sc);
883 * Otherwise, we reset the counter and wait for the next
893 carp_input(void *v, struct mbuf *m)
895 struct carp_if *cif = v;
896 struct ether_header *eh;
897 struct carp_softc *sc;
900 ASSERT_LWKT_TOKEN_HELD(&carp_tok);
902 eh = mtod(m, struct ether_header *);
904 ifp = carp_forus(cif, eh->ether_dhost);
906 ether_reinput_oncpu(ifp, m, REINPUT_RUNBPF);
910 if ((m->m_flags & (M_BCAST | M_MCAST)) == 0)
914 * XXX Should really check the list of multicast addresses
915 * for each CARP interface _before_ copying.
917 TAILQ_FOREACH(sc, &cif->vhif_vrs, sc_list) {
920 if ((sc->sc_if.if_flags & IFF_UP) == 0)
923 m0 = m_dup(m, MB_DONTWAIT);
927 ether_reinput_oncpu(&sc->sc_if, m0, REINPUT_RUNBPF);
933 carp_prepare_ad(struct carp_softc *sc, struct carp_header *ch)
935 if (sc->sc_init_counter) {
936 /* this could also be seconds since unix epoch */
937 sc->sc_counter = karc4random();
938 sc->sc_counter = sc->sc_counter << 32;
939 sc->sc_counter += karc4random();
944 ch->carp_counter[0] = htonl((sc->sc_counter >> 32) & 0xffffffff);
945 ch->carp_counter[1] = htonl(sc->sc_counter & 0xffffffff);
947 carp_hmac_generate(sc, ch->carp_counter, ch->carp_md);
951 carp_send_ad_all(void)
953 struct carp_softc *sc;
955 LIST_FOREACH(sc, &carpif_list, sc_next) {
956 if (sc->sc_carpdev == NULL)
959 if (CARP_IS_RUNNING(&sc->sc_if) && sc->sc_state == MASTER)
965 carp_send_ad_timeout(void *xsc)
973 carp_send_ad(struct carp_softc *sc)
975 struct ifnet *cifp = &sc->sc_if;
976 struct carp_header ch;
978 struct carp_header *ch_ptr;
980 int len, advbase, advskew;
982 if (!CARP_IS_RUNNING(cifp)) {
987 advbase = sc->sc_advbase;
988 if (!carp_suppress_preempt || sc->sc_advskew > 240)
989 advskew = sc->sc_advskew;
993 tv.tv_usec = advskew * 1000000 / 256;
996 ch.carp_version = CARP_VERSION;
997 ch.carp_type = CARP_ADVERTISEMENT;
998 ch.carp_vhid = sc->sc_vhid;
999 ch.carp_advbase = advbase;
1000 ch.carp_advskew = advskew;
1001 ch.carp_authlen = 7; /* XXX DEFINE */
1002 ch.carp_pad1 = 0; /* must be zero */
1006 if (sc->sc_ia != NULL) {
1009 MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1012 carpstats.carps_onomem++;
1013 /* XXX maybe less ? */
1014 if (advbase != 255 || advskew != 255)
1015 callout_reset(&sc->sc_ad_tmo, tvtohz_high(&tv),
1016 carp_send_ad_timeout, sc);
1019 len = sizeof(*ip) + sizeof(ch);
1020 m->m_pkthdr.len = len;
1021 m->m_pkthdr.rcvif = NULL;
1023 MH_ALIGN(m, m->m_len);
1024 m->m_flags |= M_MCAST;
1025 ip = mtod(m, struct ip *);
1026 ip->ip_v = IPVERSION;
1027 ip->ip_hl = sizeof(*ip) >> 2;
1028 ip->ip_tos = IPTOS_LOWDELAY;
1030 ip->ip_id = ip_newid();
1032 ip->ip_ttl = CARP_DFLTTL;
1033 ip->ip_p = IPPROTO_CARP;
1035 ip->ip_src = sc->sc_ia->ia_addr.sin_addr;
1036 ip->ip_dst.s_addr = htonl(INADDR_CARP_GROUP);
1038 ch_ptr = (struct carp_header *)(&ip[1]);
1039 bcopy(&ch, ch_ptr, sizeof(ch));
1040 carp_prepare_ad(sc, ch_ptr);
1041 ch_ptr->carp_cksum = in_cksum_skip(m, len, sizeof(*ip));
1043 getmicrotime(&cifp->if_lastchange);
1044 cifp->if_opackets++;
1045 cifp->if_obytes += len;
1046 carpstats.carps_opackets++;
1048 if (ip_output(m, NULL, NULL, IP_RAWOUTPUT, &sc->sc_imo, NULL)) {
1050 if (sc->sc_sendad_errors < INT_MAX)
1051 sc->sc_sendad_errors++;
1052 if (sc->sc_sendad_errors == CARP_SENDAD_MAX_ERRORS) {
1053 carp_suppress_preempt++;
1054 if (carp_suppress_preempt == 1) {
1058 sc->sc_sendad_success = 0;
1060 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) {
1061 if (++sc->sc_sendad_success >=
1062 CARP_SENDAD_MIN_SUCCESS) {
1063 carp_suppress_preempt--;
1064 sc->sc_sendad_errors = 0;
1067 sc->sc_sendad_errors = 0;
1074 struct ip6_hdr *ip6;
1076 MGETHDR(m, MB_DONTWAIT, MT_HEADER);
1079 carpstats.carps_onomem++;
1080 /* XXX maybe less ? */
1081 if (advbase != 255 || advskew != 255)
1082 callout_reset(&sc->sc_ad_tmo, tvtohz_high(&tv),
1083 carp_send_ad_timeout, sc);
1086 len = sizeof(*ip6) + sizeof(ch);
1087 m->m_pkthdr.len = len;
1088 m->m_pkthdr.rcvif = NULL;
1090 MH_ALIGN(m, m->m_len);
1091 m->m_flags |= M_MCAST;
1092 ip6 = mtod(m, struct ip6_hdr *);
1093 bzero(ip6, sizeof(*ip6));
1094 ip6->ip6_vfc |= IPV6_VERSION;
1095 ip6->ip6_hlim = CARP_DFLTTL;
1096 ip6->ip6_nxt = IPPROTO_CARP;
1097 bcopy(&sc->sc_ia6->ia_addr.sin6_addr, &ip6->ip6_src,
1098 sizeof(struct in6_addr));
1099 /* set the multicast destination */
1101 ip6->ip6_dst.s6_addr16[0] = htons(0xff02);
1102 ip6->ip6_dst.s6_addr8[15] = 0x12;
1103 if (in6_setscope(&ip6->ip6_dst, sc->sc_carpdev, NULL) != 0) {
1106 CARP_LOG("%s: in6_setscope failed\n", __func__);
1110 ch_ptr = (struct carp_header *)(&ip6[1]);
1111 bcopy(&ch, ch_ptr, sizeof(ch));
1112 carp_prepare_ad(sc, ch_ptr);
1113 ch_ptr->carp_cksum = in_cksum_skip(m, len, sizeof(*ip6));
1115 getmicrotime(&cifp->if_lastchange);
1116 cifp->if_opackets++;
1117 cifp->if_obytes += len;
1118 carpstats.carps_opackets6++;
1120 if (ip6_output(m, NULL, NULL, 0, &sc->sc_im6o, NULL, NULL)) {
1122 if (sc->sc_sendad_errors < INT_MAX)
1123 sc->sc_sendad_errors++;
1124 if (sc->sc_sendad_errors == CARP_SENDAD_MAX_ERRORS) {
1125 carp_suppress_preempt++;
1126 if (carp_suppress_preempt == 1) {
1130 sc->sc_sendad_success = 0;
1132 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) {
1133 if (++sc->sc_sendad_success >=
1134 CARP_SENDAD_MIN_SUCCESS) {
1135 carp_suppress_preempt--;
1136 sc->sc_sendad_errors = 0;
1139 sc->sc_sendad_errors = 0;
1145 if (advbase != 255 || advskew != 255)
1146 callout_reset(&sc->sc_ad_tmo, tvtohz_high(&tv),
1147 carp_send_ad_timeout, sc);
1151 * Broadcast a gratuitous ARP request containing
1152 * the virtual router MAC address for each IP address
1153 * associated with the virtual router.
1156 carp_send_arp(struct carp_softc *sc)
1158 const struct carp_vhaddr *vha;
1160 TAILQ_FOREACH(vha, &sc->sc_vha_list, vha_link) {
1161 if (vha->vha_iaback == NULL)
1163 arp_gratuitous(&sc->sc_if, &vha->vha_ia->ia_ifa);
1169 carp_send_na(struct carp_softc *sc)
1171 struct ifaddr_container *ifac;
1172 struct in6_addr *in6;
1173 static struct in6_addr mcast = IN6ADDR_LINKLOCAL_ALLNODES_INIT;
1175 TAILQ_FOREACH(ifac, &sc->sc_if.if_addrheads[mycpuid], ifa_link) {
1176 struct ifaddr *ifa = ifac->ifa;
1178 if (ifa->ifa_addr->sa_family != AF_INET6)
1181 in6 = &ifatoia6(ifa)->ia_addr.sin6_addr;
1182 nd6_na_output(sc->sc_carpdev, &mcast, in6,
1183 ND_NA_FLAG_OVERRIDE, 1, NULL);
1184 DELAY(1000); /* XXX */
1189 static __inline const struct carp_vhaddr *
1190 carp_find_addr(const struct carp_softc *sc, const struct in_addr *addr)
1192 struct carp_vhaddr *vha;
1194 TAILQ_FOREACH(vha, &sc->sc_vha_list, vha_link) {
1195 if (vha->vha_iaback == NULL)
1198 if (vha->vha_ia->ia_addr.sin_addr.s_addr == addr->s_addr)
1206 carp_iamatch_balance(const struct carp_if *cif, const struct in_addr *itaddr,
1207 const struct in_addr *isaddr, uint8_t **enaddr)
1209 const struct carp_softc *vh;
1210 int index, count = 0;
1213 * XXX proof of concept implementation.
1214 * We use the source ip to decide which virtual host should
1215 * handle the request. If we're master of that virtual host,
1216 * then we respond, otherwise, just drop the arp packet on
1220 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) {
1221 if (!CARP_IS_RUNNING(&vh->sc_if))
1224 if (carp_find_addr(vh, itaddr) != NULL)
1230 /* this should be a hash, like pf_hash() */
1231 index = ntohl(isaddr->s_addr) % count;
1234 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) {
1235 if (!CARP_IS_RUNNING(&vh->sc_if))
1238 if (carp_find_addr(vh, itaddr) == NULL)
1241 if (count == index) {
1242 if (vh->sc_state == MASTER) {
1243 *enaddr = IF_LLADDR(&vh->sc_if);
1256 carp_iamatch(const struct in_ifaddr *ia)
1258 const struct carp_softc *sc = ia->ia_ifp->if_softc;
1260 ASSERT_LWKT_TOKEN_HELD(&carp_tok);
1263 if (carp_opts[CARPCTL_ARPBALANCE])
1264 return carp_iamatch_balance(cif, itaddr, isaddr, enaddr);
1267 if (!CARP_IS_RUNNING(&sc->sc_if) || sc->sc_state != MASTER)
1275 carp_iamatch6(void *v, struct in6_addr *taddr)
1277 struct carp_if *cif = v;
1278 struct carp_softc *vh;
1280 ASSERT_LWKT_TOKEN_HELD(&carp_tok);
1282 TAILQ_FOREACH(vh, &cif->vhif_vrs, sc_list) {
1283 struct ifaddr_container *ifac;
1285 TAILQ_FOREACH(ifac, &vh->sc_if.if_addrheads[mycpuid],
1287 struct ifaddr *ifa = ifac->ifa;
1289 if (IN6_ARE_ADDR_EQUAL(taddr,
1290 &ifatoia6(ifa)->ia_addr.sin6_addr) &&
1291 CARP_IS_RUNNING(&vh->sc_if) &&
1292 vh->sc_state == MASTER) {
1301 carp_macmatch6(void *v, struct mbuf *m, const struct in6_addr *taddr)
1304 struct carp_if *cif = v;
1305 struct carp_softc *sc;
1307 ASSERT_LWKT_TOKEN_HELD(&carp_tok);
1309 TAILQ_FOREACH(sc, &cif->vhif_vrs, sc_list) {
1310 struct ifaddr_container *ifac;
1312 TAILQ_FOREACH(ifac, &sc->sc_if.if_addrheads[mycpuid],
1314 struct ifaddr *ifa = ifac->ifa;
1316 if (IN6_ARE_ADDR_EQUAL(taddr,
1317 &ifatoia6(ifa)->ia_addr.sin6_addr) &&
1318 CARP_IS_RUNNING(&sc->sc_if)) {
1319 struct ifnet *ifp = &sc->sc_if;
1321 mtag = m_tag_get(PACKET_TAG_CARP,
1322 sizeof(struct ifnet *), MB_DONTWAIT);
1324 /* better a bit than nothing */
1325 return (IF_LLADDR(ifp));
1327 bcopy(&ifp, (caddr_t)(mtag + 1),
1328 sizeof(struct ifnet *));
1329 m_tag_prepend(m, mtag);
1331 return (IF_LLADDR(ifp));
1339 static struct ifnet *
1340 carp_forus(struct carp_if *cif, const uint8_t *dhost)
1342 struct carp_softc *sc;
1344 ASSERT_LWKT_TOKEN_HELD(&carp_tok);
1346 if (memcmp(dhost, carp_etheraddr, ETHER_ADDR_LEN - 1) != 0)
1349 TAILQ_FOREACH(sc, &cif->vhif_vrs, sc_list) {
1350 struct ifnet *ifp = &sc->sc_if;
1352 if (CARP_IS_RUNNING(ifp) && sc->sc_state == MASTER &&
1353 !bcmp(dhost, IF_LLADDR(ifp), ETHER_ADDR_LEN))
1360 carp_master_down_timeout(void *xsc)
1362 struct carp_softc *sc = xsc;
1364 CARP_DEBUG("%s: BACKUP -> MASTER (master timed out)\n",
1365 sc->sc_if.if_xname);
1367 carp_master_down(sc);
1372 carp_master_down(struct carp_softc *sc)
1374 switch (sc->sc_state) {
1376 kprintf("%s: master_down event in INIT state\n",
1377 sc->sc_if.if_xname);
1384 carp_set_state(sc, MASTER);
1391 carp_setroute(sc, RTM_ADD);
1397 * When in backup state, af indicates whether to reset the master down timer
1398 * for v4 or v6. If it's set to zero, reset the ones which are already pending.
1401 carp_setrun(struct carp_softc *sc, sa_family_t af)
1403 struct ifnet *cifp = &sc->sc_if;
1406 if (sc->sc_carpdev == NULL) {
1407 carp_set_state(sc, INIT);
1411 if ((cifp->if_flags & IFF_RUNNING) && sc->sc_vhid > 0 &&
1412 (sc->sc_naddrs || sc->sc_naddrs6)) {
1415 carp_setroute(sc, RTM_DELETE);
1419 switch (sc->sc_state) {
1421 if (carp_opts[CARPCTL_PREEMPT] && !carp_suppress_preempt) {
1427 CARP_DEBUG("%s: INIT -> MASTER (preempting)\n",
1429 carp_set_state(sc, MASTER);
1430 carp_setroute(sc, RTM_ADD);
1432 CARP_DEBUG("%s: INIT -> BACKUP\n", cifp->if_xname);
1433 carp_set_state(sc, BACKUP);
1434 carp_setroute(sc, RTM_DELETE);
1440 callout_stop(&sc->sc_ad_tmo);
1441 tv.tv_sec = 3 * sc->sc_advbase;
1442 tv.tv_usec = sc->sc_advskew * 1000000 / 256;
1446 callout_reset(&sc->sc_md_tmo, tvtohz_high(&tv),
1447 carp_master_down_timeout, sc);
1452 callout_reset(&sc->sc_md6_tmo, tvtohz_high(&tv),
1453 carp_master_down_timeout, sc);
1458 callout_reset(&sc->sc_md_tmo, tvtohz_high(&tv),
1459 carp_master_down_timeout, sc);
1461 callout_reset(&sc->sc_md6_tmo, tvtohz_high(&tv),
1462 carp_master_down_timeout, sc);
1468 tv.tv_sec = sc->sc_advbase;
1469 tv.tv_usec = sc->sc_advskew * 1000000 / 256;
1470 callout_reset(&sc->sc_ad_tmo, tvtohz_high(&tv),
1471 carp_send_ad_timeout, sc);
1477 carp_multicast_cleanup(struct carp_softc *sc)
1479 struct ip_moptions *imo = &sc->sc_imo;
1481 if (imo->imo_num_memberships == 0)
1483 KKASSERT(imo->imo_num_memberships == 1);
1485 in_delmulti(imo->imo_membership[0]);
1486 imo->imo_membership[0] = NULL;
1487 imo->imo_num_memberships = 0;
1488 imo->imo_multicast_ifp = NULL;
1493 carp_multicast6_cleanup(struct carp_softc *sc)
1495 struct ip6_moptions *im6o = &sc->sc_im6o;
1497 while (!LIST_EMPTY(&im6o->im6o_memberships)) {
1498 struct in6_multi_mship *imm =
1499 LIST_FIRST(&im6o->im6o_memberships);
1501 LIST_REMOVE(imm, i6mm_chain);
1502 in6_leavegroup(imm);
1504 im6o->im6o_multicast_ifp = NULL;
1509 carp_get_vhaddr(struct carp_softc *sc, struct ifdrv *ifd)
1511 const struct carp_vhaddr *vha;
1512 struct ifcarpvhaddr *carpa, *carpa0;
1513 int count, len, error;
1516 TAILQ_FOREACH(vha, &sc->sc_vha_list, vha_link)
1519 if (ifd->ifd_len == 0) {
1520 ifd->ifd_len = count * sizeof(*carpa);
1522 } else if (count == 0 || ifd->ifd_len < sizeof(*carpa)) {
1526 len = min(ifd->ifd_len, sizeof(*carpa) * count);
1527 KKASSERT(len >= sizeof(*carpa));
1529 carpa0 = carpa = kmalloc(len, M_TEMP, M_WAITOK | M_NULLOK | M_ZERO);
1534 TAILQ_FOREACH(vha, &sc->sc_vha_list, vha_link) {
1535 if (len < sizeof(*carpa))
1538 carpa->carpa_flags = vha->vha_flags;
1539 carpa->carpa_addr.sin_family = AF_INET;
1540 carpa->carpa_addr.sin_addr = vha->vha_ia->ia_addr.sin_addr;
1542 carpa->carpa_baddr.sin_family = AF_INET;
1543 if (vha->vha_iaback == NULL) {
1544 carpa->carpa_baddr.sin_addr.s_addr = INADDR_ANY;
1546 carpa->carpa_baddr.sin_addr =
1547 vha->vha_iaback->ia_addr.sin_addr;
1552 len -= sizeof(*carpa);
1554 ifd->ifd_len = sizeof(*carpa) * count;
1555 KKASSERT(ifd->ifd_len > 0);
1557 error = copyout(carpa0, ifd->ifd_data, ifd->ifd_len);
1558 kfree(carpa0, M_TEMP);
1563 carp_config_vhaddr(struct carp_softc *sc, struct carp_vhaddr *vha,
1564 struct in_ifaddr *ia_del)
1567 struct in_ifaddr *ia_if;
1568 struct in_ifaddr_container *iac;
1569 const struct sockaddr_in *sin;
1573 KKASSERT(vha->vha_ia != NULL);
1575 sin = &vha->vha_ia->ia_addr;
1576 iaddr = ntohl(sin->sin_addr.s_addr);
1580 TAILQ_FOREACH(iac, &in_ifaddrheads[mycpuid], ia_link) {
1581 struct in_ifaddr *ia = iac->ia;
1586 if (ia->ia_ifp->if_type == IFT_CARP)
1589 if ((ia->ia_ifp->if_flags & IFF_UP) == 0)
1592 /* and, yeah, we need a multicast-capable iface too */
1593 if ((ia->ia_ifp->if_flags & IFF_MULTICAST) == 0)
1596 if ((iaddr & ia->ia_subnetmask) == ia->ia_subnet) {
1597 if (sin->sin_addr.s_addr ==
1598 ia->ia_addr.sin_addr.s_addr)
1602 else if (sc->sc_carpdev != NULL &&
1603 sc->sc_carpdev == ia->ia_ifp)
1608 carp_deactivate_vhaddr(sc, vha, FALSE);
1612 ifp = ia_if->ia_ifp;
1614 /* XXX Don't allow parent iface to be changed */
1615 if (sc->sc_carpdev != NULL && sc->sc_carpdev != ifp)
1618 return carp_activate_vhaddr(sc, vha, ifp, ia_if, own);
1622 carp_add_addr(struct carp_softc *sc, struct ifaddr *carp_ifa)
1624 struct carp_vhaddr *vha_new;
1625 struct in_ifaddr *carp_ia;
1627 struct carp_vhaddr *vha;
1630 KKASSERT(carp_ifa->ifa_addr->sa_family == AF_INET);
1631 carp_ia = ifatoia(carp_ifa);
1634 TAILQ_FOREACH(vha, &sc->sc_vha_list, vha_link)
1635 KKASSERT(vha->vha_ia != NULL && vha->vha_ia != carp_ia);
1638 vha_new = kmalloc(sizeof(*vha_new), M_CARP, M_WAITOK | M_ZERO);
1639 vha_new->vha_ia = carp_ia;
1640 carp_insert_vhaddr(sc, vha_new);
1642 if (carp_config_vhaddr(sc, vha_new, NULL) != 0) {
1644 * If the above configuration fails, it may only mean
1645 * that the new address is problematic. However, the
1646 * carp(4) interface may already have several working
1647 * addresses. Since the expected behaviour of
1648 * SIOC[AS]IFADDR is to put the NIC into working state,
1649 * we try starting the state machine manually here with
1650 * the hope that the carp(4)'s previously working
1651 * addresses still could be brought up.
1653 carp_hmac_prepare(sc);
1654 carp_set_state(sc, INIT);
1660 carp_del_addr(struct carp_softc *sc, struct ifaddr *carp_ifa)
1662 struct carp_vhaddr *vha;
1663 struct in_ifaddr *carp_ia;
1665 KKASSERT(carp_ifa->ifa_addr->sa_family == AF_INET);
1666 carp_ia = ifatoia(carp_ifa);
1668 TAILQ_FOREACH(vha, &sc->sc_vha_list, vha_link) {
1669 KKASSERT(vha->vha_ia != NULL);
1670 if (vha->vha_ia == carp_ia)
1673 KASSERT(vha != NULL, ("no corresponding vhaddr %p\n", carp_ifa));
1676 * Remove the vhaddr from the list before deactivating
1677 * the vhaddr, so that the HMAC could be correctly
1678 * updated in carp_deactivate_vhaddr()
1680 carp_remove_vhaddr(sc, vha);
1682 carp_deactivate_vhaddr(sc, vha, FALSE);
1687 carp_config_addr(struct carp_softc *sc, struct ifaddr *carp_ifa)
1689 struct carp_vhaddr *vha;
1690 struct in_ifaddr *carp_ia;
1692 KKASSERT(carp_ifa->ifa_addr->sa_family == AF_INET);
1693 carp_ia = ifatoia(carp_ifa);
1695 TAILQ_FOREACH(vha, &sc->sc_vha_list, vha_link) {
1696 KKASSERT(vha->vha_ia != NULL);
1697 if (vha->vha_ia == carp_ia)
1700 KASSERT(vha != NULL, ("no corresponding vhaddr %p\n", carp_ifa));
1702 /* Remove then reinsert, to keep the vhaddr list sorted */
1703 carp_remove_vhaddr(sc, vha);
1704 carp_insert_vhaddr(sc, vha);
1706 if (carp_config_vhaddr(sc, vha, NULL) != 0) {
1707 /* See the comment in carp_add_addr() */
1708 carp_hmac_prepare(sc);
1709 carp_set_state(sc, INIT);
1718 carp_set_addr6(struct carp_softc *sc, struct sockaddr_in6 *sin6)
1721 struct carp_if *cif;
1722 struct in6_ifaddr *ia, *ia_if;
1723 struct ip6_moptions *im6o = &sc->sc_im6o;
1724 struct in6_multi_mship *imm;
1725 struct in6_addr in6;
1728 if (IN6_IS_ADDR_UNSPECIFIED(&sin6->sin6_addr)) {
1733 /* we have to do it by hands to check we won't match on us */
1734 ia_if = NULL; own = 0;
1735 for (ia = in6_ifaddr; ia; ia = ia->ia_next) {
1738 for (i = 0; i < 4; i++) {
1739 if ((sin6->sin6_addr.s6_addr32[i] &
1740 ia->ia_prefixmask.sin6_addr.s6_addr32[i]) !=
1741 (ia->ia_addr.sin6_addr.s6_addr32[i] &
1742 ia->ia_prefixmask.sin6_addr.s6_addr32[i]))
1745 /* and, yeah, we need a multicast-capable iface too */
1746 if (ia->ia_ifp != &sc->sc_if &&
1747 (ia->ia_ifp->if_flags & IFF_MULTICAST) &&
1751 if (IN6_ARE_ADDR_EQUAL(&sin6->sin6_addr,
1752 &ia->ia_addr.sin6_addr))
1758 return (EADDRNOTAVAIL);
1762 if (ifp == NULL || (ifp->if_flags & IFF_MULTICAST) == 0 ||
1763 (im6o->im6o_multicast_ifp && im6o->im6o_multicast_ifp != ifp))
1764 return (EADDRNOTAVAIL);
1766 if (!sc->sc_naddrs6) {
1767 im6o->im6o_multicast_ifp = ifp;
1769 /* join CARP multicast address */
1770 bzero(&in6, sizeof(in6));
1771 in6.s6_addr16[0] = htons(0xff02);
1772 in6.s6_addr8[15] = 0x12;
1773 if (in6_setscope(&in6, ifp, NULL) != 0)
1775 if ((imm = in6_joingroup(ifp, &in6, &error)) == NULL)
1777 LIST_INSERT_HEAD(&im6o->im6o_memberships, imm, i6mm_chain);
1779 /* join solicited multicast address */
1780 bzero(&in6, sizeof(in6));
1781 in6.s6_addr16[0] = htons(0xff02);
1782 in6.s6_addr32[1] = 0;
1783 in6.s6_addr32[2] = htonl(1);
1784 in6.s6_addr32[3] = sin6->sin6_addr.s6_addr32[3];
1785 in6.s6_addr8[12] = 0xff;
1786 if (in6_setscope(&in6, ifp, NULL) != 0)
1788 if ((imm = in6_joingroup(ifp, &in6, &error)) == NULL)
1790 LIST_INSERT_HEAD(&im6o->im6o_memberships, imm, i6mm_chain);
1793 if (!ifp->if_carp) {
1794 cif = kmalloc(sizeof(*cif), M_CARP, M_WAITOK | M_ZERO);
1796 if ((error = ifpromisc(ifp, 1))) {
1801 TAILQ_INIT(&cif->vhif_vrs);
1804 struct carp_softc *vr;
1807 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) {
1808 if (vr != sc && vr->sc_vhid == sc->sc_vhid) {
1815 sc->sc_carpdev = ifp;
1817 { /* XXX prevent endless loop if already in queue */
1818 struct carp_softc *vr, *after = NULL;
1822 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) {
1825 if (vr->sc_vhid < sc->sc_vhid)
1830 /* We're trying to keep things in order */
1832 TAILQ_INSERT_TAIL(&cif->vhif_vrs, sc, sc_list);
1834 TAILQ_INSERT_AFTER(&cif->vhif_vrs, after, sc, sc_list);
1847 /* clean up multicast memberships */
1848 if (!sc->sc_naddrs6) {
1849 while (!LIST_EMPTY(&im6o->im6o_memberships)) {
1850 imm = LIST_FIRST(&im6o->im6o_memberships);
1851 LIST_REMOVE(imm, i6mm_chain);
1852 in6_leavegroup(imm);
1859 carp_del_addr6(struct carp_softc *sc, struct sockaddr_in6 *sin6)
1863 if (!--sc->sc_naddrs6) {
1864 struct carp_if *cif = sc->sc_carpdev->if_carp;
1865 struct ip6_moptions *im6o = &sc->sc_im6o;
1867 callout_stop(&sc->sc_ad_tmo);
1869 while (!LIST_EMPTY(&im6o->im6o_memberships)) {
1870 struct in6_multi_mship *imm =
1871 LIST_FIRST(&im6o->im6o_memberships);
1873 LIST_REMOVE(imm, i6mm_chain);
1874 in6_leavegroup(imm);
1876 im6o->im6o_multicast_ifp = NULL;
1877 TAILQ_REMOVE(&cif->vhif_vrs, sc, sc_list);
1878 if (TAILQ_EMPTY(&cif->vhif_vrs)) {
1879 sc->sc_carpdev->if_carp = NULL;
1880 kfree(cif, M_IFADDR);
1890 carp_ioctl(struct ifnet *ifp, u_long cmd, caddr_t addr, struct ucred *cr)
1892 struct carp_softc *sc = ifp->if_softc, *vr;
1893 struct carpreq carpr;
1896 struct ifaliasreq *ifra;
1898 char devname[IFNAMSIZ];
1901 ASSERT_IFNET_SERIALIZED_ALL(ifp);
1905 ifa = (struct ifaddr *)addr;
1906 ifra = (struct ifaliasreq *)addr;
1907 ifr = (struct ifreq *)addr;
1908 ifd = (struct ifdrv *)addr;
1912 if (ifp->if_flags & IFF_UP) {
1913 if ((ifp->if_flags & IFF_RUNNING) == 0)
1915 } else if (ifp->if_flags & IFF_RUNNING) {
1916 carp_ioctl_stop(sc);
1921 error = priv_check_cred(cr, PRIV_ROOT, NULL_CRED_OKAY);
1924 error = copyin(ifr->ifr_data, &carpr, sizeof(carpr));
1929 if ((ifp->if_flags & IFF_RUNNING) &&
1930 sc->sc_state != INIT && carpr.carpr_state != sc->sc_state) {
1931 switch (carpr.carpr_state) {
1933 callout_stop(&sc->sc_ad_tmo);
1934 carp_set_state(sc, BACKUP);
1936 carp_setroute(sc, RTM_DELETE);
1940 carp_master_down(sc);
1947 if (carpr.carpr_vhid > 0) {
1948 if (carpr.carpr_vhid > 255) {
1952 if (sc->sc_carpdev) {
1953 struct carp_if *cif = sc->sc_carpdev->if_carp;
1955 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) {
1957 vr->sc_vhid == carpr.carpr_vhid) {
1963 sc->sc_vhid = carpr.carpr_vhid;
1965 IF_LLADDR(ifp)[5] = sc->sc_vhid;
1966 bcopy(IF_LLADDR(ifp), sc->arpcom.ac_enaddr,
1971 if (carpr.carpr_advbase > 0 || carpr.carpr_advskew > 0) {
1972 if (carpr.carpr_advskew >= 255) {
1976 if (carpr.carpr_advbase > 255) {
1980 sc->sc_advbase = carpr.carpr_advbase;
1981 sc->sc_advskew = carpr.carpr_advskew;
1984 bcopy(carpr.carpr_key, sc->sc_key, sizeof(sc->sc_key));
1994 bzero(&carpr, sizeof(carpr));
1995 carpr.carpr_state = sc->sc_state;
1996 carpr.carpr_vhid = sc->sc_vhid;
1997 carpr.carpr_advbase = sc->sc_advbase;
1998 carpr.carpr_advskew = sc->sc_advskew;
1999 error = priv_check_cred(cr, PRIV_ROOT, NULL_CRED_OKAY);
2001 bcopy(sc->sc_key, carpr.carpr_key,
2002 sizeof(carpr.carpr_key));
2005 error = copyout(&carpr, ifr->ifr_data, sizeof(carpr));
2009 switch (ifd->ifd_cmd) {
2011 if (ifd->ifd_len != sizeof(devname))
2025 switch (ifd->ifd_cmd) {
2027 error = carp_get_vhaddr(sc, ifd);
2031 bzero(devname, sizeof(devname));
2032 if (sc->sc_carpdev != NULL) {
2033 strlcpy(devname, sc->sc_carpdev->if_xname,
2036 error = copyout(devname, ifd->ifd_data,
2043 error = ether_ioctl(ifp, cmd, addr);
2046 carp_hmac_prepare(sc);
2053 carp_ioctl_stop_dispatch(netmsg_t msg)
2055 struct netmsg_carp *cmsg = (struct netmsg_carp *)msg;
2056 struct carp_softc *sc = cmsg->nc_softc;
2062 lwkt_replymsg(&cmsg->base.lmsg, 0);
2066 carp_ioctl_stop(struct carp_softc *sc)
2068 struct ifnet *ifp = &sc->arpcom.ac_if;
2069 struct netmsg_carp cmsg;
2071 ASSERT_IFNET_SERIALIZED_ALL(ifp);
2073 ifnet_deserialize_all(ifp);
2075 bzero(&cmsg, sizeof(cmsg));
2076 netmsg_init(&cmsg.base, NULL, &curthread->td_msgport, 0,
2077 carp_ioctl_stop_dispatch);
2080 lwkt_domsg(cpu_portfn(0), &cmsg.base.lmsg, 0);
2082 ifnet_serialize_all(ifp);
2086 carp_init_dispatch(netmsg_t msg)
2088 struct netmsg_carp *cmsg = (struct netmsg_carp *)msg;
2089 struct carp_softc *sc = cmsg->nc_softc;
2093 sc->sc_if.if_flags |= IFF_RUNNING;
2094 carp_set_state(sc, INIT);
2099 lwkt_replymsg(&cmsg->base.lmsg, 0);
2103 carp_init(void *xsc)
2105 struct carp_softc *sc = xsc;
2106 struct ifnet *ifp = &sc->arpcom.ac_if;
2107 struct netmsg_carp cmsg;
2109 ASSERT_IFNET_SERIALIZED_ALL(ifp);
2111 ifnet_deserialize_all(ifp);
2113 bzero(&cmsg, sizeof(cmsg));
2114 netmsg_init(&cmsg.base, NULL, &curthread->td_msgport, 0,
2115 carp_init_dispatch);
2118 lwkt_domsg(cpu_portfn(0), &cmsg.base.lmsg, 0);
2120 ifnet_serialize_all(ifp);
2124 carp_output(struct ifnet *ifp, struct mbuf *m, struct sockaddr *dst,
2127 struct carp_softc *sc = ifp->if_softc;
2131 if (sc->sc_carpdev) {
2134 * CARP's ifp is passed to backing device's
2137 sc->sc_carpdev->if_output(ifp, m, dst, rt);
2140 error = ENETUNREACH;
2148 * Start output on carp interface. This function should never be called.
2151 carp_start(struct ifnet *ifp)
2153 panic("%s: start called\n", ifp->if_xname);
2157 carp_serialize(struct ifnet *ifp __unused,
2158 enum ifnet_serialize slz __unused)
2163 carp_deserialize(struct ifnet *ifp __unused,
2164 enum ifnet_serialize slz __unused)
2169 carp_tryserialize(struct ifnet *ifp __unused,
2170 enum ifnet_serialize slz __unused)
2178 carp_serialize_assert(struct ifnet *ifp __unused,
2179 enum ifnet_serialize slz __unused, boolean_t serialized __unused)
2183 #endif /* INVARIANTS */
2186 carp_set_state(struct carp_softc *sc, int state)
2188 struct ifnet *cifp = &sc->sc_if;
2190 if (sc->sc_state == state)
2192 sc->sc_state = state;
2194 switch (sc->sc_state) {
2196 cifp->if_link_state = LINK_STATE_DOWN;
2200 cifp->if_link_state = LINK_STATE_UP;
2204 cifp->if_link_state = LINK_STATE_UNKNOWN;
2211 carp_group_demote_adj(struct ifnet *ifp, int adj)
2213 struct ifg_list *ifgl;
2218 TAILQ_FOREACH(ifgl, &ifp->if_groups, ifgl_next) {
2219 if (!strcmp(ifgl->ifgl_group->ifg_group, IFG_ALL))
2221 dm = &ifgl->ifgl_group->ifg_carp_demoted;
2228 if (adj > 0 && *dm == 1)
2230 CARP_LOG("%s demoted group %s to %d", ifp->if_xname,
2231 ifgl->ifgl_group->ifg_group, *dm);
2238 carp_carpdev_state(void *v)
2240 struct carp_if *cif = v;
2241 struct carp_softc *sc;
2245 TAILQ_FOREACH(sc, &cif->vhif_vrs, sc_list)
2252 carp_sc_state(struct carp_softc *sc)
2254 if (!(sc->sc_carpdev->if_flags & IFF_UP)) {
2255 callout_stop(&sc->sc_ad_tmo);
2256 callout_stop(&sc->sc_md_tmo);
2257 callout_stop(&sc->sc_md6_tmo);
2258 carp_set_state(sc, INIT);
2260 if (!sc->sc_suppress) {
2261 carp_suppress_preempt++;
2262 if (carp_suppress_preempt == 1)
2265 sc->sc_suppress = 1;
2267 carp_set_state(sc, INIT);
2269 if (sc->sc_suppress)
2270 carp_suppress_preempt--;
2271 sc->sc_suppress = 0;
2276 carp_stop(struct carp_softc *sc, int detach)
2278 sc->sc_if.if_flags &= ~IFF_RUNNING;
2280 callout_stop(&sc->sc_ad_tmo);
2281 callout_stop(&sc->sc_md_tmo);
2282 callout_stop(&sc->sc_md6_tmo);
2284 if (!detach && sc->sc_state == MASTER)
2287 if (sc->sc_suppress)
2288 carp_suppress_preempt--;
2289 sc->sc_suppress = 0;
2291 if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS)
2292 carp_suppress_preempt--;
2293 sc->sc_sendad_errors = 0;
2294 sc->sc_sendad_success = 0;
2296 carp_set_state(sc, INIT);
2301 carp_suspend(struct carp_softc *sc, int detach)
2303 struct ifnet *cifp = &sc->sc_if;
2305 carp_stop(sc, detach);
2307 /* Retain the running state, if we are not dead yet */
2308 if (!sc->sc_dead && (cifp->if_flags & IFF_UP))
2309 cifp->if_flags |= IFF_RUNNING;
2313 carp_activate_vhaddr(struct carp_softc *sc, struct carp_vhaddr *vha,
2314 struct ifnet *ifp, struct in_ifaddr *ia_if, int own)
2316 struct ip_moptions *imo = &sc->sc_imo;
2317 struct carp_if *cif;
2318 struct carp_softc *vr, *after = NULL;
2324 KKASSERT(vha->vha_ia != NULL);
2326 KASSERT(ia_if != NULL, ("NULL backing address\n"));
2327 KASSERT(vha->vha_iaback == NULL, ("%p is already activated\n", vha));
2328 KASSERT((vha->vha_flags & CARP_VHAF_OWNER) == 0,
2329 ("inactive vhaddr %p is the address owner\n", vha));
2331 KASSERT(sc->sc_carpdev == NULL || sc->sc_carpdev == ifp,
2332 ("%s is already on %s\n", sc->sc_if.if_xname,
2333 sc->sc_carpdev->if_xname));
2335 if (!ifp->if_carp) {
2336 KASSERT(sc->sc_carpdev == NULL,
2337 ("%s is already on %s\n", sc->sc_if.if_xname,
2338 sc->sc_carpdev->if_xname));
2340 cif = kmalloc(sizeof(*cif), M_CARP, M_WAITOK | M_ZERO);
2342 error = ifpromisc(ifp, 1);
2348 TAILQ_INIT(&cif->vhif_vrs);
2352 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) {
2353 if (vr != sc && vr->sc_vhid == sc->sc_vhid)
2359 if (sc->sc_carpdev != NULL)
2365 sc->sc_carpdev = ifp;
2369 TAILQ_FOREACH(vr, &cif->vhif_vrs, sc_list) {
2372 if (vr->sc_vhid < sc->sc_vhid)
2377 if (assert_onlist) {
2378 KASSERT(onlist, ("%s is not on %s carp list\n",
2379 sc->sc_if.if_xname, ifp->if_xname));
2381 KASSERT(!onlist, ("%s is already on %s carp list\n",
2382 sc->sc_if.if_xname, ifp->if_xname));
2387 /* We're trying to keep things in order */
2389 TAILQ_INSERT_TAIL(&cif->vhif_vrs, sc, sc_list);
2391 TAILQ_INSERT_AFTER(&cif->vhif_vrs, after, sc, sc_list);
2394 vha->vha_iaback = ia_if;
2398 vha->vha_flags |= CARP_VHAF_OWNER;
2400 /* XXX save user configured advskew? */
2404 carp_addroute_vhaddr(sc, vha);
2407 * Join the multicast group only after the backing interface
2408 * has been hooked with the CARP interface.
2410 KASSERT(imo->imo_multicast_ifp == NULL ||
2411 imo->imo_multicast_ifp == &sc->sc_if,
2412 ("%s didn't leave mcast group on %s\n",
2413 sc->sc_if.if_xname, imo->imo_multicast_ifp->if_xname));
2415 if (imo->imo_num_memberships == 0) {
2416 struct in_addr addr;
2418 addr.s_addr = htonl(INADDR_CARP_GROUP);
2419 imo->imo_membership[0] = in_addmulti(&addr, &sc->sc_if);
2420 if (imo->imo_membership[0] == NULL) {
2421 carp_deactivate_vhaddr(sc, vha, FALSE);
2425 imo->imo_num_memberships++;
2426 imo->imo_multicast_ifp = &sc->sc_if;
2427 imo->imo_multicast_ttl = CARP_DFLTTL;
2428 imo->imo_multicast_loop = 0;
2431 carp_hmac_prepare(sc);
2432 carp_set_state(sc, INIT);
2438 carp_deactivate_vhaddr(struct carp_softc *sc, struct carp_vhaddr *vha,
2439 boolean_t del_iaback)
2441 KKASSERT(vha->vha_ia != NULL);
2443 carp_hmac_prepare(sc);
2445 if (vha->vha_iaback == NULL) {
2446 KASSERT((vha->vha_flags & CARP_VHAF_OWNER) == 0,
2447 ("inactive vhaddr %p is the address owner\n", vha));
2451 vha->vha_flags &= ~CARP_VHAF_OWNER;
2452 carp_delroute_vhaddr(sc, vha, del_iaback);
2454 KKASSERT(sc->sc_naddrs > 0);
2455 vha->vha_iaback = NULL;
2457 if (!sc->sc_naddrs) {
2458 if (sc->sc_naddrs6) {
2459 carp_multicast_cleanup(sc);
2462 carp_detach(sc, 0, del_iaback);
2468 carp_link_addrs(struct carp_softc *sc, struct ifnet *ifp, struct ifaddr *ifa_if)
2470 struct carp_vhaddr *vha;
2471 struct in_ifaddr *ia_if;
2473 KKASSERT(ifa_if->ifa_addr->sa_family == AF_INET);
2474 ia_if = ifatoia(ifa_if);
2477 * Test each inactive vhaddr against the newly added address.
2478 * If the newly added address could be the backing address,
2479 * then activate the matching vhaddr.
2481 TAILQ_FOREACH(vha, &sc->sc_vha_list, vha_link) {
2482 const struct in_ifaddr *ia;
2486 if (vha->vha_iaback != NULL)
2490 iaddr = ntohl(ia->ia_addr.sin_addr.s_addr);
2492 if ((iaddr & ia_if->ia_subnetmask) != ia_if->ia_subnet)
2496 if (ia->ia_addr.sin_addr.s_addr ==
2497 ia_if->ia_addr.sin_addr.s_addr)
2500 carp_activate_vhaddr(sc, vha, ifp, ia_if, own);
2505 carp_unlink_addrs(struct carp_softc *sc, struct ifnet *ifp,
2506 struct ifaddr *ifa_if)
2508 struct carp_vhaddr *vha;
2509 struct in_ifaddr *ia_if;
2511 KKASSERT(ifa_if->ifa_addr->sa_family == AF_INET);
2512 ia_if = ifatoia(ifa_if);
2515 * Ad src address is deleted; set it to NULL.
2516 * Following loop will try pick up a new ad src address
2517 * if one of the vhaddr could retain its backing address.
2519 if (sc->sc_ia == ia_if)
2523 * Test each active vhaddr against the deleted address.
2524 * If the deleted address is vhaddr address's backing
2525 * address, then deactivate the vhaddr.
2527 TAILQ_FOREACH(vha, &sc->sc_vha_list, vha_link) {
2528 if (vha->vha_iaback == NULL)
2531 if (vha->vha_iaback == ia_if)
2532 carp_deactivate_vhaddr(sc, vha, TRUE);
2533 else if (sc->sc_ia == NULL)
2534 sc->sc_ia = vha->vha_iaback;
2539 carp_update_addrs(struct carp_softc *sc, struct ifaddr *ifa_del)
2541 struct carp_vhaddr *vha;
2543 KKASSERT(sc->sc_carpdev == NULL);
2545 TAILQ_FOREACH(vha, &sc->sc_vha_list, vha_link)
2546 carp_config_vhaddr(sc, vha, ifatoia(ifa_del));
2550 carp_ifaddr(void *arg __unused, struct ifnet *ifp,
2551 enum ifaddr_event event, struct ifaddr *ifa)
2553 struct carp_softc *sc;
2557 if (ifa->ifa_addr->sa_family != AF_INET)
2560 KASSERT(&curthread->td_msgport == cpu_portfn(0),
2561 ("not in netisr0"));
2563 if (ifp->if_type == IFT_CARP) {
2565 * Address is changed on carp(4) interface
2568 case IFADDR_EVENT_ADD:
2569 carp_add_addr(ifp->if_softc, ifa);
2572 case IFADDR_EVENT_CHANGE:
2573 carp_config_addr(ifp->if_softc, ifa);
2576 case IFADDR_EVENT_DELETE:
2577 carp_del_addr(ifp->if_softc, ifa);
2584 * Address is changed on non-carp(4) interface
2586 if ((ifp->if_flags & IFF_MULTICAST) == 0)
2589 LIST_FOREACH(sc, &carpif_list, sc_next) {
2590 if (sc->sc_carpdev != NULL && sc->sc_carpdev != ifp) {
2591 /* Not the parent iface; skip */
2596 case IFADDR_EVENT_ADD:
2597 carp_link_addrs(sc, ifp, ifa);
2600 case IFADDR_EVENT_DELETE:
2601 if (sc->sc_carpdev != NULL) {
2602 carp_unlink_addrs(sc, ifp, ifa);
2603 if (sc->sc_carpdev == NULL) {
2605 * We no longer have the parent
2606 * interface, however, certain
2607 * virtual addresses, which are
2608 * not used because they can't
2609 * match the previous parent
2610 * interface's addresses, may now
2611 * match different interface's
2614 carp_update_addrs(sc, ifa);
2618 * The carp(4) interface didn't have a
2619 * parent iface, so it is not possible
2620 * that it will contain any address to
2626 case IFADDR_EVENT_CHANGE:
2627 if (sc->sc_carpdev == NULL) {
2629 * The carp(4) interface didn't have a
2630 * parent iface, so it is not possible
2631 * that it will contain any address to
2634 carp_link_addrs(sc, ifp, ifa);
2637 * First try breaking tie with the old
2638 * address. Then see whether we could
2639 * link certain vhaddr to the new address.
2640 * If that fails, i.e. carpdev is NULL,
2641 * we try a global update.
2643 * NOTE: The above order is critical.
2645 carp_unlink_addrs(sc, ifp, ifa);
2646 carp_link_addrs(sc, ifp, ifa);
2647 if (sc->sc_carpdev == NULL) {
2649 * See the comment in the above
2650 * IFADDR_EVENT_DELETE block.
2652 carp_update_addrs(sc, NULL);
2664 carp_proto_ctlinput(netmsg_t msg)
2666 int cmd = msg->ctlinput.nm_cmd;
2667 struct sockaddr *sa = msg->ctlinput.nm_arg;
2668 struct in_ifaddr_container *iac;
2672 TAILQ_FOREACH(iac, &in_ifaddrheads[mycpuid], ia_link) {
2673 struct in_ifaddr *ia = iac->ia;
2674 struct ifnet *ifp = ia->ia_ifp;
2676 if (ifp->if_type == IFT_CARP)
2679 if (ia->ia_ifa.ifa_addr == sa) {
2680 if (cmd == PRC_IFDOWN) {
2681 carp_ifaddr(NULL, ifp, IFADDR_EVENT_DELETE,
2683 } else if (cmd == PRC_IFUP) {
2684 carp_ifaddr(NULL, ifp, IFADDR_EVENT_ADD,
2692 lwkt_replymsg(&msg->lmsg, 0);
2698 lwkt_gettoken(&carp_tok);
2704 lwkt_reltoken(&carp_tok);
2708 carp_parent(struct ifnet *cifp)
2710 struct carp_softc *sc;
2712 ASSERT_LWKT_TOKEN_HELD(&carp_tok);
2714 KKASSERT(cifp->if_type == IFT_CARP);
2715 sc = cifp->if_softc;
2717 return sc->sc_carpdev;
2720 #define rtinitflags(x) \
2721 (((x)->ia_ifp->if_flags & (IFF_LOOPBACK | IFF_POINTOPOINT)) \
2725 carp_addroute_vhaddr(struct carp_softc *sc, struct carp_vhaddr *vha)
2727 struct in_ifaddr *ia, *iaback;
2730 if (sc->sc_state != MASTER)
2734 KKASSERT(ia != NULL);
2736 iaback = vha->vha_iaback;
2737 KKASSERT(iaback != NULL);
2739 rtinit(&iaback->ia_ifa, RTM_DELETE, rtinitflags(iaback));
2740 in_ifadown(&iaback->ia_ifa, 1);
2741 iaback->ia_flags &= ~IFA_ROUTE;
2743 error = rtinit(&ia->ia_ifa, RTM_ADD, rtinitflags(ia) | RTF_UP);
2745 ia->ia_flags |= IFA_ROUTE;
2750 carp_delroute_vhaddr(struct carp_softc *sc, struct carp_vhaddr *vha,
2751 boolean_t del_iaback)
2753 struct in_ifaddr *ia, *iaback;
2756 KKASSERT(ia != NULL);
2758 iaback = vha->vha_iaback;
2759 KKASSERT(iaback != NULL);
2761 rtinit(&ia->ia_ifa, RTM_DELETE, rtinitflags(ia));
2762 in_ifadown(&ia->ia_ifa, 1);
2763 ia->ia_flags &= ~IFA_ROUTE;
2765 if (!del_iaback && (iaback->ia_ifp->if_flags & IFF_UP)) {
2768 error = rtinit(&iaback->ia_ifa, RTM_ADD,
2769 rtinitflags(iaback) | RTF_UP);
2771 iaback->ia_flags |= IFA_ROUTE;
2776 carp_modevent(module_t mod, int type, void *data)
2780 LIST_INIT(&carpif_list);
2781 carp_ifdetach_event =
2782 EVENTHANDLER_REGISTER(ifnet_detach_event, carp_ifdetach, NULL,
2783 EVENTHANDLER_PRI_ANY);
2785 EVENTHANDLER_REGISTER(ifaddr_event, carp_ifaddr, NULL,
2786 EVENTHANDLER_PRI_FIRST);
2787 if_clone_attach(&carp_cloner);
2791 EVENTHANDLER_DEREGISTER(ifnet_detach_event,
2792 carp_ifdetach_event);
2793 EVENTHANDLER_DEREGISTER(ifaddr_event,
2795 if_clone_detach(&carp_cloner);
2804 static moduledata_t carp_mod = {
2809 DECLARE_MODULE(carp, carp_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);