xref: /freebsd/sys/netinet/ip_carp.c (revision 4fd0d10e0fe684211328bc148edf89a792425b39)
1 /*-
2  * Copyright (c) 2002 Michael Shalayeff.
3  * Copyright (c) 2003 Ryan McBride.
4  * Copyright (c) 2011 Gleb Smirnoff <glebius@FreeBSD.org>
5  * All rights reserved.
6  *
7  * Redistribution and use in source and binary forms, with or without
8  * modification, are permitted provided that the following conditions
9  * are met:
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
17  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
18  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
19  * IN NO EVENT SHALL THE AUTHOR OR HIS RELATIVES BE LIABLE FOR ANY DIRECT,
20  * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
21  * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
22  * SERVICES; LOSS OF MIND, USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
24  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
25  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
26  * THE POSSIBILITY OF SUCH DAMAGE.
27  */
28 
29 #include <sys/cdefs.h>
30 __FBSDID("$FreeBSD$");
31 
32 #include "opt_bpf.h"
33 #include "opt_inet.h"
34 #include "opt_inet6.h"
35 
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/bus.h>
39 #include <sys/jail.h>
40 #include <sys/kernel.h>
41 #include <sys/limits.h>
42 #include <sys/malloc.h>
43 #include <sys/mbuf.h>
44 #include <sys/module.h>
45 #include <sys/priv.h>
46 #include <sys/proc.h>
47 #include <sys/protosw.h>
48 #include <sys/socket.h>
49 #include <sys/sockio.h>
50 #include <sys/sysctl.h>
51 #include <sys/syslog.h>
52 #include <sys/taskqueue.h>
53 #include <sys/counter.h>
54 
55 #include <net/ethernet.h>
56 #include <net/fddi.h>
57 #include <net/if.h>
58 #include <net/if_dl.h>
59 #include <net/if_llatbl.h>
60 #include <net/if_types.h>
61 #include <net/iso88025.h>
62 #include <net/route.h>
63 #include <net/vnet.h>
64 
65 #if defined(INET) || defined(INET6)
66 #include <netinet/in.h>
67 #include <netinet/in_var.h>
68 #include <netinet/ip_carp.h>
69 #include <netinet/ip.h>
70 #include <machine/in_cksum.h>
71 #endif
72 #ifdef INET
73 #include <netinet/ip_var.h>
74 #include <netinet/if_ether.h>
75 #endif
76 
77 #ifdef INET6
78 #include <netinet/icmp6.h>
79 #include <netinet/ip6.h>
80 #include <netinet6/ip6protosw.h>
81 #include <netinet6/in6_var.h>
82 #include <netinet6/ip6_var.h>
83 #include <netinet6/scope6_var.h>
84 #include <netinet6/nd6.h>
85 #endif
86 
87 #include <crypto/sha1.h>
88 
89 static MALLOC_DEFINE(M_CARP, "CARP", "CARP addresses");
90 
91 struct carp_softc {
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. */
96 #ifdef INET
97 	struct callout		sc_md_tmo;	/* Master down timeout. */
98 #endif
99 #ifdef INET6
100 	struct callout 		sc_md6_tmo;	/* XXX: Master down timeout. */
101 #endif
102 	struct mtx		sc_mtx;
103 
104 	int			sc_vhid;
105 	int			sc_advskew;
106 	int			sc_advbase;
107 
108 	int			sc_naddrs;
109 	int			sc_naddrs6;
110 	int			sc_ifasiz;
111 	enum { INIT = 0, BACKUP, MASTER }	sc_state;
112 	int			sc_suppress;
113 	int			sc_sendad_errors;
114 #define	CARP_SENDAD_MAX_ERRORS	3
115 	int			sc_sendad_success;
116 #define	CARP_SENDAD_MIN_SUCCESS 3
117 
118 	int			sc_init_counter;
119 	uint64_t		sc_counter;
120 
121 	/* authentication */
122 #define	CARP_HMAC_PAD	64
123 	unsigned char sc_key[CARP_KEY_LEN];
124 	unsigned char sc_pad[CARP_HMAC_PAD];
125 	SHA1_CTX sc_sha1;
126 
127 	TAILQ_ENTRY(carp_softc)	sc_list;	/* On the carp_if list. */
128 	LIST_ENTRY(carp_softc)	sc_next;	/* On the global list. */
129 };
130 
131 struct carp_if {
132 #ifdef INET
133 	int	cif_naddrs;
134 #endif
135 #ifdef INET6
136 	int	cif_naddrs6;
137 #endif
138 	TAILQ_HEAD(, carp_softc) cif_vrs;
139 #ifdef INET
140 	struct ip_moptions 	 cif_imo;
141 #endif
142 #ifdef INET6
143 	struct ip6_moptions 	 cif_im6o;
144 #endif
145 	struct ifnet	*cif_ifp;
146 	struct mtx	cif_mtx;
147 };
148 
149 #define	CARP_INET	0
150 #define	CARP_INET6	1
151 static int proto_reg[] = {-1, -1};
152 
153 /*
154  * Brief design of carp(4).
155  *
156  * Any carp-capable ifnet may have a list of carp softcs hanging off
157  * its ifp->if_carp pointer. Each softc represents one unique virtual
158  * host id, or vhid. The softc has a back pointer to the ifnet. All
159  * softcs are joined in a global list, which has quite limited use.
160  *
161  * Any interface address that takes part in CARP negotiation has a
162  * pointer to the softc of its vhid, ifa->ifa_carp. That could be either
163  * AF_INET or AF_INET6 address.
164  *
165  * Although, one can get the softc's backpointer to ifnet and traverse
166  * through its ifp->if_addrhead queue to find all interface addresses
167  * involved in CARP, we keep a growable array of ifaddr pointers. This
168  * allows us to avoid grabbing the IF_ADDR_LOCK() in many traversals that
169  * do calls into the network stack, thus avoiding LORs.
170  *
171  * Locking:
172  *
173  * Each softc has a lock sc_mtx. It is used to synchronise carp_input_c(),
174  * callout-driven events and ioctl()s.
175  *
176  * To traverse the list of softcs on an ifnet we use CIF_LOCK(), to
177  * traverse the global list we use the mutex carp_mtx.
178  *
179  * Known issues with locking:
180  *
181  * - There is no protection for races between two ioctl() requests,
182  *   neither SIOCSVH, nor SIOCAIFADDR & SIOCAIFADDR_IN6. I think that all
183  *   interface ioctl()s should be serialized right in net/if.c.
184  * - Sending ad, we put the pointer to the softc in an mtag, and no reference
185  *   counting is done on the softc.
186  * - On module unload we may race (?) with packet processing thread
187  *   dereferencing our function pointers.
188  */
189 
190 /* Accept incoming CARP packets. */
191 static VNET_DEFINE(int, carp_allow) = 1;
192 #define	V_carp_allow	VNET(carp_allow)
193 
194 /* Preempt slower nodes. */
195 static VNET_DEFINE(int, carp_preempt) = 0;
196 #define	V_carp_preempt	VNET(carp_preempt)
197 
198 /* Log level. */
199 static VNET_DEFINE(int, carp_log) = 1;
200 #define	V_carp_log	VNET(carp_log)
201 
202 /* Global advskew demotion. */
203 static VNET_DEFINE(int, carp_demotion) = 0;
204 #define	V_carp_demotion	VNET(carp_demotion)
205 
206 /* Send error demotion factor. */
207 static VNET_DEFINE(int, carp_senderr_adj) = CARP_MAXSKEW;
208 #define	V_carp_senderr_adj	VNET(carp_senderr_adj)
209 
210 /* Iface down demotion factor. */
211 static VNET_DEFINE(int, carp_ifdown_adj) = CARP_MAXSKEW;
212 #define	V_carp_ifdown_adj	VNET(carp_ifdown_adj)
213 
214 static int carp_demote_adj_sysctl(SYSCTL_HANDLER_ARGS);
215 
216 SYSCTL_NODE(_net_inet, IPPROTO_CARP,	carp,	CTLFLAG_RW, 0,	"CARP");
217 SYSCTL_VNET_INT(_net_inet_carp, OID_AUTO, allow, CTLFLAG_RW,
218     &VNET_NAME(carp_allow), 0, "Accept incoming CARP packets");
219 SYSCTL_VNET_INT(_net_inet_carp, OID_AUTO, preempt, CTLFLAG_RW,
220     &VNET_NAME(carp_preempt), 0, "High-priority backup preemption mode");
221 SYSCTL_VNET_INT(_net_inet_carp, OID_AUTO, log, CTLFLAG_RW,
222     &VNET_NAME(carp_log), 0, "CARP log level");
223 SYSCTL_VNET_PROC(_net_inet_carp, OID_AUTO, demotion, CTLTYPE_INT|CTLFLAG_RW,
224     0, 0, carp_demote_adj_sysctl, "I",
225     "Adjust demotion factor (skew of advskew)");
226 SYSCTL_VNET_INT(_net_inet_carp, OID_AUTO, senderr_demotion_factor, CTLFLAG_RW,
227     &VNET_NAME(carp_senderr_adj), 0, "Send error demotion factor adjustment");
228 SYSCTL_VNET_INT(_net_inet_carp, OID_AUTO, ifdown_demotion_factor, CTLFLAG_RW,
229     &VNET_NAME(carp_ifdown_adj), 0,
230     "Interface down demotion factor adjustment");
231 
232 VNET_PCPUSTAT_DEFINE(struct carpstats, carpstats);
233 VNET_PCPUSTAT_SYSINIT(carpstats);
234 VNET_PCPUSTAT_SYSUNINIT(carpstats);
235 
236 #define	CARPSTATS_ADD(name, val)	\
237     counter_u64_add(VNET(carpstats)[offsetof(struct carpstats, name) / \
238 	sizeof(uint64_t)], (val))
239 #define	CARPSTATS_INC(name)		CARPSTATS_ADD(name, 1)
240 
241 SYSCTL_VNET_PCPUSTAT(_net_inet_carp, OID_AUTO, stats, struct carpstats,
242     carpstats, "CARP statistics (struct carpstats, netinet/ip_carp.h)");
243 
244 #define	CARP_LOCK_INIT(sc)	mtx_init(&(sc)->sc_mtx, "carp_softc",   \
245 	NULL, MTX_DEF)
246 #define	CARP_LOCK_DESTROY(sc)	mtx_destroy(&(sc)->sc_mtx)
247 #define	CARP_LOCK_ASSERT(sc)	mtx_assert(&(sc)->sc_mtx, MA_OWNED)
248 #define	CARP_LOCK(sc)		mtx_lock(&(sc)->sc_mtx)
249 #define	CARP_UNLOCK(sc)		mtx_unlock(&(sc)->sc_mtx)
250 #define	CIF_LOCK_INIT(cif)	mtx_init(&(cif)->cif_mtx, "carp_if",   \
251 	NULL, MTX_DEF)
252 #define	CIF_LOCK_DESTROY(cif)	mtx_destroy(&(cif)->cif_mtx)
253 #define	CIF_LOCK_ASSERT(cif)	mtx_assert(&(cif)->cif_mtx, MA_OWNED)
254 #define	CIF_LOCK(cif)		mtx_lock(&(cif)->cif_mtx)
255 #define	CIF_UNLOCK(cif)		mtx_unlock(&(cif)->cif_mtx)
256 #define	CIF_FREE(cif)	do {				\
257 		CIF_LOCK_ASSERT(cif);			\
258 		if (TAILQ_EMPTY(&(cif)->cif_vrs))	\
259 			carp_free_if(cif);		\
260 		else					\
261 			CIF_UNLOCK(cif);		\
262 } while (0)
263 
264 #define	CARP_LOG(...)	do {				\
265 	if (V_carp_log > 0)				\
266 		log(LOG_INFO, "carp: " __VA_ARGS__);	\
267 } while (0)
268 
269 #define	CARP_DEBUG(...)	do {				\
270 	if (V_carp_log > 1)				\
271 		log(LOG_DEBUG, __VA_ARGS__);		\
272 } while (0)
273 
274 #define	IFNET_FOREACH_IFA(ifp, ifa)					\
275 	IF_ADDR_LOCK_ASSERT(ifp);					\
276 	TAILQ_FOREACH((ifa), &(ifp)->if_addrhead, ifa_link)		\
277 		if ((ifa)->ifa_carp != NULL)
278 
279 #define	CARP_FOREACH_IFA(sc, ifa)					\
280 	CARP_LOCK_ASSERT(sc);						\
281 	for (int _i = 0;						\
282 		_i < (sc)->sc_naddrs + (sc)->sc_naddrs6 &&		\
283 		((ifa) = sc->sc_ifas[_i]) != NULL;			\
284 		++_i)
285 
286 #define	IFNET_FOREACH_CARP(ifp, sc)					\
287 	CIF_LOCK_ASSERT(ifp->if_carp);					\
288 	TAILQ_FOREACH((sc), &(ifp)->if_carp->cif_vrs, sc_list)
289 
290 #define	DEMOTE_ADVSKEW(sc)					\
291     (((sc)->sc_advskew + V_carp_demotion > CARP_MAXSKEW) ?	\
292     CARP_MAXSKEW : ((sc)->sc_advskew + V_carp_demotion))
293 
294 static void	carp_input_c(struct mbuf *, struct carp_header *, sa_family_t);
295 static struct carp_softc
296 		*carp_alloc(struct ifnet *);
297 static void	carp_detach_locked(struct ifaddr *);
298 static void	carp_destroy(struct carp_softc *);
299 static struct carp_if
300 		*carp_alloc_if(struct ifnet *);
301 static void	carp_free_if(struct carp_if *);
302 static void	carp_set_state(struct carp_softc *, int);
303 static void	carp_sc_state(struct carp_softc *);
304 static void	carp_setrun(struct carp_softc *, sa_family_t);
305 static void	carp_master_down(void *);
306 static void	carp_master_down_locked(struct carp_softc *);
307 static void	carp_send_ad(void *);
308 static void	carp_send_ad_locked(struct carp_softc *);
309 static void	carp_addroute(struct carp_softc *);
310 static void	carp_ifa_addroute(struct ifaddr *);
311 static void	carp_delroute(struct carp_softc *);
312 static void	carp_ifa_delroute(struct ifaddr *);
313 static void	carp_send_ad_all(void *, int);
314 static void	carp_demote_adj(int, char *);
315 
316 static LIST_HEAD(, carp_softc) carp_list;
317 static struct mtx carp_mtx;
318 static struct task carp_sendall_task =
319     TASK_INITIALIZER(0, carp_send_ad_all, NULL);
320 
321 static void
322 carp_hmac_prepare(struct carp_softc *sc)
323 {
324 	uint8_t version = CARP_VERSION, type = CARP_ADVERTISEMENT;
325 	uint8_t vhid = sc->sc_vhid & 0xff;
326 	struct ifaddr *ifa;
327 	int i, found;
328 #ifdef INET
329 	struct in_addr last, cur, in;
330 #endif
331 #ifdef INET6
332 	struct in6_addr last6, cur6, in6;
333 #endif
334 
335 	CARP_LOCK_ASSERT(sc);
336 
337 	/* Compute ipad from key. */
338 	bzero(sc->sc_pad, sizeof(sc->sc_pad));
339 	bcopy(sc->sc_key, sc->sc_pad, sizeof(sc->sc_key));
340 	for (i = 0; i < sizeof(sc->sc_pad); i++)
341 		sc->sc_pad[i] ^= 0x36;
342 
343 	/* Precompute first part of inner hash. */
344 	SHA1Init(&sc->sc_sha1);
345 	SHA1Update(&sc->sc_sha1, sc->sc_pad, sizeof(sc->sc_pad));
346 	SHA1Update(&sc->sc_sha1, (void *)&version, sizeof(version));
347 	SHA1Update(&sc->sc_sha1, (void *)&type, sizeof(type));
348 	SHA1Update(&sc->sc_sha1, (void *)&vhid, sizeof(vhid));
349 #ifdef INET
350 	cur.s_addr = 0;
351 	do {
352 		found = 0;
353 		last = cur;
354 		cur.s_addr = 0xffffffff;
355 		CARP_FOREACH_IFA(sc, ifa) {
356 			in.s_addr = ifatoia(ifa)->ia_addr.sin_addr.s_addr;
357 			if (ifa->ifa_addr->sa_family == AF_INET &&
358 			    ntohl(in.s_addr) > ntohl(last.s_addr) &&
359 			    ntohl(in.s_addr) < ntohl(cur.s_addr)) {
360 				cur.s_addr = in.s_addr;
361 				found++;
362 			}
363 		}
364 		if (found)
365 			SHA1Update(&sc->sc_sha1, (void *)&cur, sizeof(cur));
366 	} while (found);
367 #endif /* INET */
368 #ifdef INET6
369 	memset(&cur6, 0, sizeof(cur6));
370 	do {
371 		found = 0;
372 		last6 = cur6;
373 		memset(&cur6, 0xff, sizeof(cur6));
374 		CARP_FOREACH_IFA(sc, ifa) {
375 			in6 = ifatoia6(ifa)->ia_addr.sin6_addr;
376 			if (IN6_IS_SCOPE_EMBED(&in6))
377 				in6.s6_addr16[1] = 0;
378 			if (ifa->ifa_addr->sa_family == AF_INET6 &&
379 			    memcmp(&in6, &last6, sizeof(in6)) > 0 &&
380 			    memcmp(&in6, &cur6, sizeof(in6)) < 0) {
381 				cur6 = in6;
382 				found++;
383 			}
384 		}
385 		if (found)
386 			SHA1Update(&sc->sc_sha1, (void *)&cur6, sizeof(cur6));
387 	} while (found);
388 #endif /* INET6 */
389 
390 	/* convert ipad to opad */
391 	for (i = 0; i < sizeof(sc->sc_pad); i++)
392 		sc->sc_pad[i] ^= 0x36 ^ 0x5c;
393 }
394 
395 static void
396 carp_hmac_generate(struct carp_softc *sc, uint32_t counter[2],
397     unsigned char md[20])
398 {
399 	SHA1_CTX sha1ctx;
400 
401 	CARP_LOCK_ASSERT(sc);
402 
403 	/* fetch first half of inner hash */
404 	bcopy(&sc->sc_sha1, &sha1ctx, sizeof(sha1ctx));
405 
406 	SHA1Update(&sha1ctx, (void *)counter, sizeof(sc->sc_counter));
407 	SHA1Final(md, &sha1ctx);
408 
409 	/* outer hash */
410 	SHA1Init(&sha1ctx);
411 	SHA1Update(&sha1ctx, sc->sc_pad, sizeof(sc->sc_pad));
412 	SHA1Update(&sha1ctx, md, 20);
413 	SHA1Final(md, &sha1ctx);
414 }
415 
416 static int
417 carp_hmac_verify(struct carp_softc *sc, uint32_t counter[2],
418     unsigned char md[20])
419 {
420 	unsigned char md2[20];
421 
422 	CARP_LOCK_ASSERT(sc);
423 
424 	carp_hmac_generate(sc, counter, md2);
425 
426 	return (bcmp(md, md2, sizeof(md2)));
427 }
428 
429 /*
430  * process input packet.
431  * we have rearranged checks order compared to the rfc,
432  * but it seems more efficient this way or not possible otherwise.
433  */
434 #ifdef INET
435 void
436 carp_input(struct mbuf *m, int hlen)
437 {
438 	struct ip *ip = mtod(m, struct ip *);
439 	struct carp_header *ch;
440 	int iplen, len;
441 
442 	CARPSTATS_INC(carps_ipackets);
443 
444 	if (!V_carp_allow) {
445 		m_freem(m);
446 		return;
447 	}
448 
449 	/* verify that the IP TTL is 255.  */
450 	if (ip->ip_ttl != CARP_DFLTTL) {
451 		CARPSTATS_INC(carps_badttl);
452 		CARP_DEBUG("%s: received ttl %d != 255 on %s\n", __func__,
453 		    ip->ip_ttl,
454 		    m->m_pkthdr.rcvif->if_xname);
455 		m_freem(m);
456 		return;
457 	}
458 
459 	iplen = ip->ip_hl << 2;
460 
461 	if (m->m_pkthdr.len < iplen + sizeof(*ch)) {
462 		CARPSTATS_INC(carps_badlen);
463 		CARP_DEBUG("%s: received len %zd < sizeof(struct carp_header) "
464 		    "on %s\n", __func__, m->m_len - sizeof(struct ip),
465 		    m->m_pkthdr.rcvif->if_xname);
466 		m_freem(m);
467 		return;
468 	}
469 
470 	if (iplen + sizeof(*ch) < m->m_len) {
471 		if ((m = m_pullup(m, iplen + sizeof(*ch))) == NULL) {
472 			CARPSTATS_INC(carps_hdrops);
473 			CARP_DEBUG("%s: pullup failed\n", __func__);
474 			return;
475 		}
476 		ip = mtod(m, struct ip *);
477 	}
478 	ch = (struct carp_header *)((char *)ip + iplen);
479 
480 	/*
481 	 * verify that the received packet length is
482 	 * equal to the CARP header
483 	 */
484 	len = iplen + sizeof(*ch);
485 	if (len > m->m_pkthdr.len) {
486 		CARPSTATS_INC(carps_badlen);
487 		CARP_DEBUG("%s: packet too short %d on %s\n", __func__,
488 		    m->m_pkthdr.len,
489 		    m->m_pkthdr.rcvif->if_xname);
490 		m_freem(m);
491 		return;
492 	}
493 
494 	if ((m = m_pullup(m, len)) == NULL) {
495 		CARPSTATS_INC(carps_hdrops);
496 		return;
497 	}
498 	ip = mtod(m, struct ip *);
499 	ch = (struct carp_header *)((char *)ip + iplen);
500 
501 	/* verify the CARP checksum */
502 	m->m_data += iplen;
503 	if (in_cksum(m, len - iplen)) {
504 		CARPSTATS_INC(carps_badsum);
505 		CARP_DEBUG("%s: checksum failed on %s\n", __func__,
506 		    m->m_pkthdr.rcvif->if_xname);
507 		m_freem(m);
508 		return;
509 	}
510 	m->m_data -= iplen;
511 
512 	carp_input_c(m, ch, AF_INET);
513 }
514 #endif
515 
516 #ifdef INET6
517 int
518 carp6_input(struct mbuf **mp, int *offp, int proto)
519 {
520 	struct mbuf *m = *mp;
521 	struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
522 	struct carp_header *ch;
523 	u_int len;
524 
525 	CARPSTATS_INC(carps_ipackets6);
526 
527 	if (!V_carp_allow) {
528 		m_freem(m);
529 		return (IPPROTO_DONE);
530 	}
531 
532 	/* check if received on a valid carp interface */
533 	if (m->m_pkthdr.rcvif->if_carp == NULL) {
534 		CARPSTATS_INC(carps_badif);
535 		CARP_DEBUG("%s: packet received on non-carp interface: %s\n",
536 		    __func__, m->m_pkthdr.rcvif->if_xname);
537 		m_freem(m);
538 		return (IPPROTO_DONE);
539 	}
540 
541 	/* verify that the IP TTL is 255 */
542 	if (ip6->ip6_hlim != CARP_DFLTTL) {
543 		CARPSTATS_INC(carps_badttl);
544 		CARP_DEBUG("%s: received ttl %d != 255 on %s\n", __func__,
545 		    ip6->ip6_hlim, m->m_pkthdr.rcvif->if_xname);
546 		m_freem(m);
547 		return (IPPROTO_DONE);
548 	}
549 
550 	/* verify that we have a complete carp packet */
551 	len = m->m_len;
552 	IP6_EXTHDR_GET(ch, struct carp_header *, m, *offp, sizeof(*ch));
553 	if (ch == NULL) {
554 		CARPSTATS_INC(carps_badlen);
555 		CARP_DEBUG("%s: packet size %u too small\n", __func__, len);
556 		return (IPPROTO_DONE);
557 	}
558 
559 
560 	/* verify the CARP checksum */
561 	m->m_data += *offp;
562 	if (in_cksum(m, sizeof(*ch))) {
563 		CARPSTATS_INC(carps_badsum);
564 		CARP_DEBUG("%s: checksum failed, on %s\n", __func__,
565 		    m->m_pkthdr.rcvif->if_xname);
566 		m_freem(m);
567 		return (IPPROTO_DONE);
568 	}
569 	m->m_data -= *offp;
570 
571 	carp_input_c(m, ch, AF_INET6);
572 	return (IPPROTO_DONE);
573 }
574 #endif /* INET6 */
575 
576 static void
577 carp_input_c(struct mbuf *m, struct carp_header *ch, sa_family_t af)
578 {
579 	struct ifnet *ifp = m->m_pkthdr.rcvif;
580 	struct ifaddr *ifa;
581 	struct carp_softc *sc;
582 	uint64_t tmp_counter;
583 	struct timeval sc_tv, ch_tv;
584 
585 	/* verify that the VHID is valid on the receiving interface */
586 	IF_ADDR_RLOCK(ifp);
587 	IFNET_FOREACH_IFA(ifp, ifa)
588 		if (ifa->ifa_addr->sa_family == af &&
589 		    ifa->ifa_carp->sc_vhid == ch->carp_vhid) {
590 			ifa_ref(ifa);
591 			break;
592 		}
593 	IF_ADDR_RUNLOCK(ifp);
594 
595 	if (ifa == NULL) {
596 		CARPSTATS_INC(carps_badvhid);
597 		m_freem(m);
598 		return;
599 	}
600 
601 	/* verify the CARP version. */
602 	if (ch->carp_version != CARP_VERSION) {
603 		CARPSTATS_INC(carps_badver);
604 		CARP_DEBUG("%s: invalid version %d\n", ifp->if_xname,
605 		    ch->carp_version);
606 		ifa_free(ifa);
607 		m_freem(m);
608 		return;
609 	}
610 
611 	sc = ifa->ifa_carp;
612 	CARP_LOCK(sc);
613 	ifa_free(ifa);
614 
615 	if (carp_hmac_verify(sc, ch->carp_counter, ch->carp_md)) {
616 		CARPSTATS_INC(carps_badauth);
617 		CARP_DEBUG("%s: incorrect hash for VHID %u@%s\n", __func__,
618 		    sc->sc_vhid, ifp->if_xname);
619 		goto out;
620 	}
621 
622 	tmp_counter = ntohl(ch->carp_counter[0]);
623 	tmp_counter = tmp_counter<<32;
624 	tmp_counter += ntohl(ch->carp_counter[1]);
625 
626 	/* XXX Replay protection goes here */
627 
628 	sc->sc_init_counter = 0;
629 	sc->sc_counter = tmp_counter;
630 
631 	sc_tv.tv_sec = sc->sc_advbase;
632 	sc_tv.tv_usec = DEMOTE_ADVSKEW(sc) * 1000000 / 256;
633 	ch_tv.tv_sec = ch->carp_advbase;
634 	ch_tv.tv_usec = ch->carp_advskew * 1000000 / 256;
635 
636 	switch (sc->sc_state) {
637 	case INIT:
638 		break;
639 	case MASTER:
640 		/*
641 		 * If we receive an advertisement from a master who's going to
642 		 * be more frequent than us, go into BACKUP state.
643 		 */
644 		if (timevalcmp(&sc_tv, &ch_tv, >) ||
645 		    timevalcmp(&sc_tv, &ch_tv, ==)) {
646 			callout_stop(&sc->sc_ad_tmo);
647 			CARP_LOG("VHID %u@%s: MASTER -> BACKUP "
648 			    "(more frequent advertisement received)\n",
649 			    sc->sc_vhid,
650 			    sc->sc_carpdev->if_xname);
651 			carp_set_state(sc, BACKUP);
652 			carp_setrun(sc, 0);
653 			carp_delroute(sc);
654 		}
655 		break;
656 	case BACKUP:
657 		/*
658 		 * If we're pre-empting masters who advertise slower than us,
659 		 * and this one claims to be slower, treat him as down.
660 		 */
661 		if (V_carp_preempt && timevalcmp(&sc_tv, &ch_tv, <)) {
662 			CARP_LOG("VHID %u@%s: BACKUP -> MASTER "
663 			    "(preempting a slower master)\n",
664 			    sc->sc_vhid,
665 			    sc->sc_carpdev->if_xname);
666 			carp_master_down_locked(sc);
667 			break;
668 		}
669 
670 		/*
671 		 *  If the master is going to advertise at such a low frequency
672 		 *  that he's guaranteed to time out, we'd might as well just
673 		 *  treat him as timed out now.
674 		 */
675 		sc_tv.tv_sec = sc->sc_advbase * 3;
676 		if (timevalcmp(&sc_tv, &ch_tv, <)) {
677 			CARP_LOG("VHID %u@%s: BACKUP -> MASTER "
678 			    "(master timed out)\n",
679 			    sc->sc_vhid,
680 			    sc->sc_carpdev->if_xname);
681 			carp_master_down_locked(sc);
682 			break;
683 		}
684 
685 		/*
686 		 * Otherwise, we reset the counter and wait for the next
687 		 * advertisement.
688 		 */
689 		carp_setrun(sc, af);
690 		break;
691 	}
692 
693 out:
694 	CARP_UNLOCK(sc);
695 	m_freem(m);
696 }
697 
698 static int
699 carp_prepare_ad(struct mbuf *m, struct carp_softc *sc, struct carp_header *ch)
700 {
701 	struct m_tag *mtag;
702 
703 	if (sc->sc_init_counter) {
704 		/* this could also be seconds since unix epoch */
705 		sc->sc_counter = arc4random();
706 		sc->sc_counter = sc->sc_counter << 32;
707 		sc->sc_counter += arc4random();
708 	} else
709 		sc->sc_counter++;
710 
711 	ch->carp_counter[0] = htonl((sc->sc_counter>>32)&0xffffffff);
712 	ch->carp_counter[1] = htonl(sc->sc_counter&0xffffffff);
713 
714 	carp_hmac_generate(sc, ch->carp_counter, ch->carp_md);
715 
716 	/* Tag packet for carp_output */
717 	if ((mtag = m_tag_get(PACKET_TAG_CARP, sizeof(struct carp_softc *),
718 	    M_NOWAIT)) == NULL) {
719 		m_freem(m);
720 		CARPSTATS_INC(carps_onomem);
721 		return (ENOMEM);
722 	}
723 	bcopy(&sc, mtag + 1, sizeof(sc));
724 	m_tag_prepend(m, mtag);
725 
726 	return (0);
727 }
728 
729 /*
730  * To avoid LORs and possible recursions this function shouldn't
731  * be called directly, but scheduled via taskqueue.
732  */
733 static void
734 carp_send_ad_all(void *ctx __unused, int pending __unused)
735 {
736 	struct carp_softc *sc;
737 
738 	mtx_lock(&carp_mtx);
739 	LIST_FOREACH(sc, &carp_list, sc_next)
740 		if (sc->sc_state == MASTER) {
741 			CARP_LOCK(sc);
742 			CURVNET_SET(sc->sc_carpdev->if_vnet);
743 			carp_send_ad_locked(sc);
744 			CURVNET_RESTORE();
745 			CARP_UNLOCK(sc);
746 		}
747 	mtx_unlock(&carp_mtx);
748 }
749 
750 /* Send a periodic advertisement, executed in callout context. */
751 static void
752 carp_send_ad(void *v)
753 {
754 	struct carp_softc *sc = v;
755 
756 	CARP_LOCK_ASSERT(sc);
757 	CURVNET_SET(sc->sc_carpdev->if_vnet);
758 	carp_send_ad_locked(sc);
759 	CURVNET_RESTORE();
760 	CARP_UNLOCK(sc);
761 }
762 
763 static void
764 carp_send_ad_locked(struct carp_softc *sc)
765 {
766 	struct carp_header ch;
767 	struct timeval tv;
768 	struct sockaddr sa;
769 	struct ifaddr *ifa;
770 	struct carp_header *ch_ptr;
771 	struct mbuf *m;
772 	int len, advskew;
773 
774 	CARP_LOCK_ASSERT(sc);
775 
776 	advskew = DEMOTE_ADVSKEW(sc);
777 	tv.tv_sec = sc->sc_advbase;
778 	tv.tv_usec = advskew * 1000000 / 256;
779 
780 	ch.carp_version = CARP_VERSION;
781 	ch.carp_type = CARP_ADVERTISEMENT;
782 	ch.carp_vhid = sc->sc_vhid;
783 	ch.carp_advbase = sc->sc_advbase;
784 	ch.carp_advskew = advskew;
785 	ch.carp_authlen = 7;	/* XXX DEFINE */
786 	ch.carp_pad1 = 0;	/* must be zero */
787 	ch.carp_cksum = 0;
788 
789 	/* XXXGL: OpenBSD picks first ifaddr with needed family. */
790 
791 #ifdef INET
792 	if (sc->sc_naddrs) {
793 		struct ip *ip;
794 
795 		m = m_gethdr(M_NOWAIT, MT_DATA);
796 		if (m == NULL) {
797 			CARPSTATS_INC(carps_onomem);
798 			goto resched;
799 		}
800 		len = sizeof(*ip) + sizeof(ch);
801 		m->m_pkthdr.len = len;
802 		m->m_pkthdr.rcvif = NULL;
803 		m->m_len = len;
804 		MH_ALIGN(m, m->m_len);
805 		m->m_flags |= M_MCAST;
806 		ip = mtod(m, struct ip *);
807 		ip->ip_v = IPVERSION;
808 		ip->ip_hl = sizeof(*ip) >> 2;
809 		ip->ip_tos = IPTOS_LOWDELAY;
810 		ip->ip_len = htons(len);
811 		ip->ip_id = ip_newid();
812 		ip->ip_off = htons(IP_DF);
813 		ip->ip_ttl = CARP_DFLTTL;
814 		ip->ip_p = IPPROTO_CARP;
815 		ip->ip_sum = 0;
816 
817 		bzero(&sa, sizeof(sa));
818 		sa.sa_family = AF_INET;
819 		ifa = ifaof_ifpforaddr(&sa, sc->sc_carpdev);
820 		if (ifa != NULL) {
821 			ip->ip_src.s_addr =
822 			    ifatoia(ifa)->ia_addr.sin_addr.s_addr;
823 			ifa_free(ifa);
824 		} else
825 			ip->ip_src.s_addr = 0;
826 		ip->ip_dst.s_addr = htonl(INADDR_CARP_GROUP);
827 
828 		ch_ptr = (struct carp_header *)(&ip[1]);
829 		bcopy(&ch, ch_ptr, sizeof(ch));
830 		if (carp_prepare_ad(m, sc, ch_ptr))
831 			goto resched;
832 
833 		m->m_data += sizeof(*ip);
834 		ch_ptr->carp_cksum = in_cksum(m, len - sizeof(*ip));
835 		m->m_data -= sizeof(*ip);
836 
837 		CARPSTATS_INC(carps_opackets);
838 
839 		if (ip_output(m, NULL, NULL, IP_RAWOUTPUT,
840 		    &sc->sc_carpdev->if_carp->cif_imo, NULL)) {
841 			if (sc->sc_sendad_errors < INT_MAX)
842 				sc->sc_sendad_errors++;
843 			if (sc->sc_sendad_errors == CARP_SENDAD_MAX_ERRORS)
844 				carp_demote_adj(V_carp_senderr_adj,
845 				    "send error");
846 			sc->sc_sendad_success = 0;
847 		} else {
848 			if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) {
849 				if (++sc->sc_sendad_success >=
850 				    CARP_SENDAD_MIN_SUCCESS) {
851 					carp_demote_adj(-V_carp_senderr_adj,
852 					    "send ok");
853 					sc->sc_sendad_errors = 0;
854 				}
855 			} else
856 				sc->sc_sendad_errors = 0;
857 		}
858 	}
859 #endif /* INET */
860 #ifdef INET6
861 	if (sc->sc_naddrs6) {
862 		struct ip6_hdr *ip6;
863 
864 		m = m_gethdr(M_NOWAIT, MT_DATA);
865 		if (m == NULL) {
866 			CARPSTATS_INC(carps_onomem);
867 			goto resched;
868 		}
869 		len = sizeof(*ip6) + sizeof(ch);
870 		m->m_pkthdr.len = len;
871 		m->m_pkthdr.rcvif = NULL;
872 		m->m_len = len;
873 		MH_ALIGN(m, m->m_len);
874 		m->m_flags |= M_MCAST;
875 		ip6 = mtod(m, struct ip6_hdr *);
876 		bzero(ip6, sizeof(*ip6));
877 		ip6->ip6_vfc |= IPV6_VERSION;
878 		ip6->ip6_hlim = CARP_DFLTTL;
879 		ip6->ip6_nxt = IPPROTO_CARP;
880 		bzero(&sa, sizeof(sa));
881 
882 		/* set the source address */
883 		sa.sa_family = AF_INET6;
884 		ifa = ifaof_ifpforaddr(&sa, sc->sc_carpdev);
885 		if (ifa != NULL) {
886 			bcopy(IFA_IN6(ifa), &ip6->ip6_src,
887 			    sizeof(struct in6_addr));
888 			ifa_free(ifa);
889 		} else
890 			/* This should never happen with IPv6. */
891 			bzero(&ip6->ip6_src, sizeof(struct in6_addr));
892 
893 		/* Set the multicast destination. */
894 		ip6->ip6_dst.s6_addr16[0] = htons(0xff02);
895 		ip6->ip6_dst.s6_addr8[15] = 0x12;
896 		if (in6_setscope(&ip6->ip6_dst, sc->sc_carpdev, NULL) != 0) {
897 			m_freem(m);
898 			CARP_DEBUG("%s: in6_setscope failed\n", __func__);
899 			goto resched;
900 		}
901 
902 		ch_ptr = (struct carp_header *)(&ip6[1]);
903 		bcopy(&ch, ch_ptr, sizeof(ch));
904 		if (carp_prepare_ad(m, sc, ch_ptr))
905 			goto resched;
906 
907 		m->m_data += sizeof(*ip6);
908 		ch_ptr->carp_cksum = in_cksum(m, len - sizeof(*ip6));
909 		m->m_data -= sizeof(*ip6);
910 
911 		CARPSTATS_INC(carps_opackets6);
912 
913 		if (ip6_output(m, NULL, NULL, 0,
914 		    &sc->sc_carpdev->if_carp->cif_im6o, NULL, NULL)) {
915 			if (sc->sc_sendad_errors < INT_MAX)
916 				sc->sc_sendad_errors++;
917 			if (sc->sc_sendad_errors == CARP_SENDAD_MAX_ERRORS)
918 				carp_demote_adj(V_carp_senderr_adj,
919 				    "send6 error");
920 			sc->sc_sendad_success = 0;
921 		} else {
922 			if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) {
923 				if (++sc->sc_sendad_success >=
924 				    CARP_SENDAD_MIN_SUCCESS) {
925 					carp_demote_adj(-V_carp_senderr_adj,
926 					    "send6 ok");
927 					sc->sc_sendad_errors = 0;
928 				}
929 			} else
930 				sc->sc_sendad_errors = 0;
931 		}
932 	}
933 #endif /* INET6 */
934 
935 resched:
936 	callout_reset(&sc->sc_ad_tmo, tvtohz(&tv), carp_send_ad, sc);
937 }
938 
939 static void
940 carp_addroute(struct carp_softc *sc)
941 {
942 	struct ifaddr *ifa;
943 
944 	CARP_FOREACH_IFA(sc, ifa)
945 		carp_ifa_addroute(ifa);
946 }
947 
948 static void
949 carp_ifa_addroute(struct ifaddr *ifa)
950 {
951 
952 	switch (ifa->ifa_addr->sa_family) {
953 #ifdef INET
954 	case AF_INET:
955 		in_addprefix(ifatoia(ifa), RTF_UP);
956 		ifa_add_loopback_route(ifa,
957 		    (struct sockaddr *)&ifatoia(ifa)->ia_addr);
958 		break;
959 #endif
960 #ifdef INET6
961 	case AF_INET6:
962 		ifa_add_loopback_route(ifa,
963 		    (struct sockaddr *)&ifatoia6(ifa)->ia_addr);
964 		in6_ifaddloop(ifa);
965 		break;
966 #endif
967 	}
968 }
969 
970 static void
971 carp_delroute(struct carp_softc *sc)
972 {
973 	struct ifaddr *ifa;
974 
975 	CARP_FOREACH_IFA(sc, ifa)
976 		carp_ifa_delroute(ifa);
977 }
978 
979 static void
980 carp_ifa_delroute(struct ifaddr *ifa)
981 {
982 
983 	switch (ifa->ifa_addr->sa_family) {
984 #ifdef INET
985 	case AF_INET:
986 		ifa_del_loopback_route(ifa,
987 		    (struct sockaddr *)&ifatoia(ifa)->ia_addr);
988 		in_scrubprefix(ifatoia(ifa), LLE_STATIC);
989 		break;
990 #endif
991 #ifdef INET6
992 	case AF_INET6:
993 		ifa_del_loopback_route(ifa,
994 		    (struct sockaddr *)&ifatoia6(ifa)->ia_addr);
995 		in6_ifremloop(ifa);
996 		break;
997 #endif
998 	}
999 }
1000 
1001 int
1002 carp_master(struct ifaddr *ifa)
1003 {
1004 	struct carp_softc *sc = ifa->ifa_carp;
1005 
1006 	return (sc->sc_state == MASTER);
1007 }
1008 
1009 #ifdef INET
1010 /*
1011  * Broadcast a gratuitous ARP request containing
1012  * the virtual router MAC address for each IP address
1013  * associated with the virtual router.
1014  */
1015 static void
1016 carp_send_arp(struct carp_softc *sc)
1017 {
1018 	struct ifaddr *ifa;
1019 
1020 	CARP_FOREACH_IFA(sc, ifa)
1021 		if (ifa->ifa_addr->sa_family == AF_INET)
1022 			arp_ifinit2(sc->sc_carpdev, ifa, LLADDR(&sc->sc_addr));
1023 }
1024 
1025 int
1026 carp_iamatch(struct ifaddr *ifa, uint8_t **enaddr)
1027 {
1028 	struct carp_softc *sc = ifa->ifa_carp;
1029 
1030 	if (sc->sc_state == MASTER) {
1031 		*enaddr = LLADDR(&sc->sc_addr);
1032 		return (1);
1033 	}
1034 
1035 	return (0);
1036 }
1037 #endif
1038 
1039 #ifdef INET6
1040 static void
1041 carp_send_na(struct carp_softc *sc)
1042 {
1043 	static struct in6_addr mcast = IN6ADDR_LINKLOCAL_ALLNODES_INIT;
1044 	struct ifaddr *ifa;
1045 	struct in6_addr *in6;
1046 
1047 	CARP_FOREACH_IFA(sc, ifa) {
1048 		if (ifa->ifa_addr->sa_family != AF_INET6)
1049 			continue;
1050 
1051 		in6 = IFA_IN6(ifa);
1052 		nd6_na_output(sc->sc_carpdev, &mcast, in6,
1053 		    ND_NA_FLAG_OVERRIDE, 1, NULL);
1054 		DELAY(1000);	/* XXX */
1055 	}
1056 }
1057 
1058 /*
1059  * Returns ifa in case it's a carp address and it is MASTER, or if the address
1060  * matches and is not a carp address.  Returns NULL otherwise.
1061  */
1062 struct ifaddr *
1063 carp_iamatch6(struct ifnet *ifp, struct in6_addr *taddr)
1064 {
1065 	struct ifaddr *ifa;
1066 
1067 	ifa = NULL;
1068 	IF_ADDR_RLOCK(ifp);
1069 	TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1070 		if (ifa->ifa_addr->sa_family != AF_INET6)
1071 			continue;
1072 		if (!IN6_ARE_ADDR_EQUAL(taddr, IFA_IN6(ifa)))
1073 			continue;
1074 		if (ifa->ifa_carp && ifa->ifa_carp->sc_state != MASTER)
1075 			ifa = NULL;
1076 		else
1077 			ifa_ref(ifa);
1078 		break;
1079 	}
1080 	IF_ADDR_RUNLOCK(ifp);
1081 
1082 	return (ifa);
1083 }
1084 
1085 caddr_t
1086 carp_macmatch6(struct ifnet *ifp, struct mbuf *m, const struct in6_addr *taddr)
1087 {
1088 	struct ifaddr *ifa;
1089 
1090 	IF_ADDR_RLOCK(ifp);
1091 	IFNET_FOREACH_IFA(ifp, ifa)
1092 		if (ifa->ifa_addr->sa_family == AF_INET6 &&
1093 		    IN6_ARE_ADDR_EQUAL(taddr, IFA_IN6(ifa))) {
1094 			struct carp_softc *sc = ifa->ifa_carp;
1095 			struct m_tag *mtag;
1096 
1097 			IF_ADDR_RUNLOCK(ifp);
1098 
1099 			mtag = m_tag_get(PACKET_TAG_CARP,
1100 			    sizeof(struct carp_softc *), M_NOWAIT);
1101 			if (mtag == NULL)
1102 				/* Better a bit than nothing. */
1103 				return (LLADDR(&sc->sc_addr));
1104 
1105 			bcopy(&sc, mtag + 1, sizeof(sc));
1106 			m_tag_prepend(m, mtag);
1107 
1108 			return (LLADDR(&sc->sc_addr));
1109 		}
1110 	IF_ADDR_RUNLOCK(ifp);
1111 
1112 	return (NULL);
1113 }
1114 #endif /* INET6 */
1115 
1116 int
1117 carp_forus(struct ifnet *ifp, u_char *dhost)
1118 {
1119 	struct carp_softc *sc;
1120 	uint8_t *ena = dhost;
1121 
1122 	if (ena[0] || ena[1] || ena[2] != 0x5e || ena[3] || ena[4] != 1)
1123 		return (0);
1124 
1125 	CIF_LOCK(ifp->if_carp);
1126 	IFNET_FOREACH_CARP(ifp, sc) {
1127 		CARP_LOCK(sc);
1128 		if (sc->sc_state == MASTER && !bcmp(dhost, LLADDR(&sc->sc_addr),
1129 		    ETHER_ADDR_LEN)) {
1130 			CARP_UNLOCK(sc);
1131 			CIF_UNLOCK(ifp->if_carp);
1132 			return (1);
1133 		}
1134 		CARP_UNLOCK(sc);
1135 	}
1136 	CIF_UNLOCK(ifp->if_carp);
1137 
1138 	return (0);
1139 }
1140 
1141 /* Master down timeout event, executed in callout context. */
1142 static void
1143 carp_master_down(void *v)
1144 {
1145 	struct carp_softc *sc = v;
1146 
1147 	CARP_LOCK_ASSERT(sc);
1148 
1149 	CURVNET_SET(sc->sc_carpdev->if_vnet);
1150 	if (sc->sc_state == BACKUP) {
1151 		CARP_LOG("VHID %u@%s: BACKUP -> MASTER (master down)\n",
1152 		    sc->sc_vhid,
1153 		    sc->sc_carpdev->if_xname);
1154 		carp_master_down_locked(sc);
1155 	}
1156 	CURVNET_RESTORE();
1157 
1158 	CARP_UNLOCK(sc);
1159 }
1160 
1161 static void
1162 carp_master_down_locked(struct carp_softc *sc)
1163 {
1164 
1165 	CARP_LOCK_ASSERT(sc);
1166 
1167 	switch (sc->sc_state) {
1168 	case BACKUP:
1169 		carp_set_state(sc, MASTER);
1170 		carp_send_ad_locked(sc);
1171 #ifdef INET
1172 		carp_send_arp(sc);
1173 #endif
1174 #ifdef INET6
1175 		carp_send_na(sc);
1176 #endif
1177 		carp_setrun(sc, 0);
1178 		carp_addroute(sc);
1179 		break;
1180 	case INIT:
1181 	case MASTER:
1182 #ifdef INVARIANTS
1183 		panic("carp: VHID %u@%s: master_down event in %s state\n",
1184 		    sc->sc_vhid,
1185 		    sc->sc_carpdev->if_xname,
1186 		    sc->sc_state ? "MASTER" : "INIT");
1187 #endif
1188 		break;
1189 	}
1190 }
1191 
1192 /*
1193  * When in backup state, af indicates whether to reset the master down timer
1194  * for v4 or v6. If it's set to zero, reset the ones which are already pending.
1195  */
1196 static void
1197 carp_setrun(struct carp_softc *sc, sa_family_t af)
1198 {
1199 	struct timeval tv;
1200 
1201 	CARP_LOCK_ASSERT(sc);
1202 
1203 	if ((sc->sc_carpdev->if_flags & IFF_UP) == 0 ||
1204 	    sc->sc_carpdev->if_link_state != LINK_STATE_UP ||
1205 	    (sc->sc_naddrs == 0 && sc->sc_naddrs6 == 0))
1206 		return;
1207 
1208 	switch (sc->sc_state) {
1209 	case INIT:
1210 		CARP_LOG("VHID %u@%s: INIT -> BACKUP\n",
1211 		    sc->sc_vhid,
1212 		    sc->sc_carpdev->if_xname);
1213 		carp_set_state(sc, BACKUP);
1214 		carp_setrun(sc, 0);
1215 		break;
1216 	case BACKUP:
1217 		callout_stop(&sc->sc_ad_tmo);
1218 		tv.tv_sec = 3 * sc->sc_advbase;
1219 		tv.tv_usec = sc->sc_advskew * 1000000 / 256;
1220 		switch (af) {
1221 #ifdef INET
1222 		case AF_INET:
1223 			callout_reset(&sc->sc_md_tmo, tvtohz(&tv),
1224 			    carp_master_down, sc);
1225 			break;
1226 #endif
1227 #ifdef INET6
1228 		case AF_INET6:
1229 			callout_reset(&sc->sc_md6_tmo, tvtohz(&tv),
1230 			    carp_master_down, sc);
1231 			break;
1232 #endif
1233 		default:
1234 #ifdef INET
1235 			if (sc->sc_naddrs)
1236 				callout_reset(&sc->sc_md_tmo, tvtohz(&tv),
1237 				    carp_master_down, sc);
1238 #endif
1239 #ifdef INET6
1240 			if (sc->sc_naddrs6)
1241 				callout_reset(&sc->sc_md6_tmo, tvtohz(&tv),
1242 				    carp_master_down, sc);
1243 #endif
1244 			break;
1245 		}
1246 		break;
1247 	case MASTER:
1248 		tv.tv_sec = sc->sc_advbase;
1249 		tv.tv_usec = sc->sc_advskew * 1000000 / 256;
1250 		callout_reset(&sc->sc_ad_tmo, tvtohz(&tv),
1251 		    carp_send_ad, sc);
1252 		break;
1253 	}
1254 }
1255 
1256 /*
1257  * Setup multicast structures.
1258  */
1259 static int
1260 carp_multicast_setup(struct carp_if *cif, sa_family_t sa)
1261 {
1262 	struct ifnet *ifp = cif->cif_ifp;
1263 	int error = 0;
1264 
1265 	CIF_LOCK_ASSERT(cif);
1266 
1267 	switch (sa) {
1268 #ifdef INET
1269 	case AF_INET:
1270 	    {
1271 		struct ip_moptions *imo = &cif->cif_imo;
1272 		struct in_addr addr;
1273 
1274 		if (imo->imo_membership)
1275 			return (0);
1276 
1277 		imo->imo_membership = (struct in_multi **)malloc(
1278 		    (sizeof(struct in_multi *) * IP_MIN_MEMBERSHIPS), M_CARP,
1279 		    M_NOWAIT);
1280 		if (imo->imo_membership == NULL)
1281 			return (ENOMEM);
1282 		imo->imo_mfilters = NULL;
1283 		imo->imo_max_memberships = IP_MIN_MEMBERSHIPS;
1284 		imo->imo_multicast_vif = -1;
1285 
1286 		addr.s_addr = htonl(INADDR_CARP_GROUP);
1287 		if ((error = in_joingroup(ifp, &addr, NULL,
1288 		    &imo->imo_membership[0])) != 0) {
1289 			free(imo->imo_membership, M_CARP);
1290 			break;
1291 		}
1292 		imo->imo_num_memberships++;
1293 		imo->imo_multicast_ifp = ifp;
1294 		imo->imo_multicast_ttl = CARP_DFLTTL;
1295 		imo->imo_multicast_loop = 0;
1296 		break;
1297 	   }
1298 #endif
1299 #ifdef INET6
1300 	case AF_INET6:
1301 	    {
1302 		struct ip6_moptions *im6o = &cif->cif_im6o;
1303 		struct in6_addr in6;
1304 		struct in6_multi *in6m;
1305 
1306 		if (im6o->im6o_membership)
1307 			return (0);
1308 
1309 		im6o->im6o_membership = (struct in6_multi **)malloc(
1310 		    (sizeof(struct in6_multi *) * IPV6_MIN_MEMBERSHIPS), M_CARP,
1311 		    M_ZERO | M_NOWAIT);
1312 		if (im6o->im6o_membership == NULL)
1313 			return (ENOMEM);
1314 		im6o->im6o_mfilters = NULL;
1315 		im6o->im6o_max_memberships = IPV6_MIN_MEMBERSHIPS;
1316 		im6o->im6o_multicast_hlim = CARP_DFLTTL;
1317 		im6o->im6o_multicast_ifp = ifp;
1318 
1319 		/* Join IPv6 CARP multicast group. */
1320 		bzero(&in6, sizeof(in6));
1321 		in6.s6_addr16[0] = htons(0xff02);
1322 		in6.s6_addr8[15] = 0x12;
1323 		if ((error = in6_setscope(&in6, ifp, NULL)) != 0) {
1324 			free(im6o->im6o_membership, M_CARP);
1325 			break;
1326 		}
1327 		in6m = NULL;
1328 		if ((error = in6_mc_join(ifp, &in6, NULL, &in6m, 0)) != 0) {
1329 			free(im6o->im6o_membership, M_CARP);
1330 			break;
1331 		}
1332 		im6o->im6o_membership[0] = in6m;
1333 		im6o->im6o_num_memberships++;
1334 
1335 		/* Join solicited multicast address. */
1336 		bzero(&in6, sizeof(in6));
1337 		in6.s6_addr16[0] = htons(0xff02);
1338 		in6.s6_addr32[1] = 0;
1339 		in6.s6_addr32[2] = htonl(1);
1340 		in6.s6_addr32[3] = 0;
1341 		in6.s6_addr8[12] = 0xff;
1342 		if ((error = in6_setscope(&in6, ifp, NULL)) != 0) {
1343 			in6_mc_leave(im6o->im6o_membership[0], NULL);
1344 			free(im6o->im6o_membership, M_CARP);
1345 			break;
1346 		}
1347 		in6m = NULL;
1348 		if ((error = in6_mc_join(ifp, &in6, NULL, &in6m, 0)) != 0) {
1349 			in6_mc_leave(im6o->im6o_membership[0], NULL);
1350 			free(im6o->im6o_membership, M_CARP);
1351 			break;
1352 		}
1353 		im6o->im6o_membership[1] = in6m;
1354 		im6o->im6o_num_memberships++;
1355 		break;
1356 	    }
1357 #endif
1358 	}
1359 
1360 	return (error);
1361 }
1362 
1363 /*
1364  * Free multicast structures.
1365  */
1366 static void
1367 carp_multicast_cleanup(struct carp_if *cif, sa_family_t sa)
1368 {
1369 
1370 	CIF_LOCK_ASSERT(cif);
1371 	switch (sa) {
1372 #ifdef INET
1373 	case AF_INET:
1374 		if (cif->cif_naddrs == 0) {
1375 			struct ip_moptions *imo = &cif->cif_imo;
1376 
1377 			in_leavegroup(imo->imo_membership[0], NULL);
1378 			KASSERT(imo->imo_mfilters == NULL,
1379 			    ("%s: imo_mfilters != NULL", __func__));
1380 			free(imo->imo_membership, M_CARP);
1381 			imo->imo_membership = NULL;
1382 
1383 		}
1384 		break;
1385 #endif
1386 #ifdef INET6
1387 	case AF_INET6:
1388 		if (cif->cif_naddrs6 == 0) {
1389 			struct ip6_moptions *im6o = &cif->cif_im6o;
1390 
1391 			in6_mc_leave(im6o->im6o_membership[0], NULL);
1392 			in6_mc_leave(im6o->im6o_membership[1], NULL);
1393 			KASSERT(im6o->im6o_mfilters == NULL,
1394 			    ("%s: im6o_mfilters != NULL", __func__));
1395 			free(im6o->im6o_membership, M_CARP);
1396 			im6o->im6o_membership = NULL;
1397 		}
1398 		break;
1399 #endif
1400 	}
1401 }
1402 
1403 int
1404 carp_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *sa)
1405 {
1406 	struct m_tag *mtag;
1407 	struct carp_softc *sc;
1408 
1409 	if (!sa)
1410 		return (0);
1411 
1412 	switch (sa->sa_family) {
1413 #ifdef INET
1414 	case AF_INET:
1415 		break;
1416 #endif
1417 #ifdef INET6
1418 	case AF_INET6:
1419 		break;
1420 #endif
1421 	default:
1422 		return (0);
1423 	}
1424 
1425 	mtag = m_tag_find(m, PACKET_TAG_CARP, NULL);
1426 	if (mtag == NULL)
1427 		return (0);
1428 
1429 	bcopy(mtag + 1, &sc, sizeof(sc));
1430 
1431 	/* Set the source MAC address to the Virtual Router MAC Address. */
1432 	switch (ifp->if_type) {
1433 	case IFT_ETHER:
1434 	case IFT_BRIDGE:
1435 	case IFT_L2VLAN: {
1436 			struct ether_header *eh;
1437 
1438 			eh = mtod(m, struct ether_header *);
1439 			eh->ether_shost[0] = 0;
1440 			eh->ether_shost[1] = 0;
1441 			eh->ether_shost[2] = 0x5e;
1442 			eh->ether_shost[3] = 0;
1443 			eh->ether_shost[4] = 1;
1444 			eh->ether_shost[5] = sc->sc_vhid;
1445 		}
1446 		break;
1447 	case IFT_FDDI: {
1448 			struct fddi_header *fh;
1449 
1450 			fh = mtod(m, struct fddi_header *);
1451 			fh->fddi_shost[0] = 0;
1452 			fh->fddi_shost[1] = 0;
1453 			fh->fddi_shost[2] = 0x5e;
1454 			fh->fddi_shost[3] = 0;
1455 			fh->fddi_shost[4] = 1;
1456 			fh->fddi_shost[5] = sc->sc_vhid;
1457 		}
1458 		break;
1459 	case IFT_ISO88025: {
1460  			struct iso88025_header *th;
1461  			th = mtod(m, struct iso88025_header *);
1462 			th->iso88025_shost[0] = 3;
1463 			th->iso88025_shost[1] = 0;
1464 			th->iso88025_shost[2] = 0x40 >> (sc->sc_vhid - 1);
1465 			th->iso88025_shost[3] = 0x40000 >> (sc->sc_vhid - 1);
1466 			th->iso88025_shost[4] = 0;
1467 			th->iso88025_shost[5] = 0;
1468 		}
1469 		break;
1470 	default:
1471 		printf("%s: carp is not supported for the %d interface type\n",
1472 		    ifp->if_xname, ifp->if_type);
1473 		return (EOPNOTSUPP);
1474 	}
1475 
1476 	return (0);
1477 }
1478 
1479 static struct carp_softc*
1480 carp_alloc(struct ifnet *ifp)
1481 {
1482 	struct carp_softc *sc;
1483 	struct carp_if *cif;
1484 
1485 	if ((cif = ifp->if_carp) == NULL) {
1486 		cif = carp_alloc_if(ifp);
1487 		if (cif == NULL)
1488 			return (NULL);
1489 	}
1490 
1491 	sc = malloc(sizeof(*sc), M_CARP, M_WAITOK|M_ZERO);
1492 
1493 	sc->sc_advbase = CARP_DFLTINTV;
1494 	sc->sc_vhid = -1;	/* required setting */
1495 	sc->sc_init_counter = 1;
1496 	sc->sc_state = INIT;
1497 
1498 	sc->sc_ifasiz = sizeof(struct ifaddr *);
1499 	sc->sc_ifas = malloc(sc->sc_ifasiz, M_CARP, M_WAITOK|M_ZERO);
1500 	sc->sc_carpdev = ifp;
1501 
1502 	CARP_LOCK_INIT(sc);
1503 #ifdef INET
1504 	callout_init_mtx(&sc->sc_md_tmo, &sc->sc_mtx, CALLOUT_RETURNUNLOCKED);
1505 #endif
1506 #ifdef INET6
1507 	callout_init_mtx(&sc->sc_md6_tmo, &sc->sc_mtx, CALLOUT_RETURNUNLOCKED);
1508 #endif
1509 	callout_init_mtx(&sc->sc_ad_tmo, &sc->sc_mtx, CALLOUT_RETURNUNLOCKED);
1510 
1511 	CIF_LOCK(cif);
1512 	TAILQ_INSERT_TAIL(&cif->cif_vrs, sc, sc_list);
1513 	CIF_UNLOCK(cif);
1514 
1515 	mtx_lock(&carp_mtx);
1516 	LIST_INSERT_HEAD(&carp_list, sc, sc_next);
1517 	mtx_unlock(&carp_mtx);
1518 
1519 	return (sc);
1520 }
1521 
1522 static int
1523 carp_grow_ifas(struct carp_softc *sc)
1524 {
1525 	struct ifaddr **new;
1526 
1527 	CARP_LOCK_ASSERT(sc);
1528 
1529 	new = malloc(sc->sc_ifasiz * 2, M_CARP, M_NOWAIT|M_ZERO);
1530 	if (new == NULL)
1531 		return (ENOMEM);
1532 	bcopy(sc->sc_ifas, new, sc->sc_ifasiz);
1533 	free(sc->sc_ifas, M_CARP);
1534 	sc->sc_ifas = new;
1535 	sc->sc_ifasiz *= 2;
1536 
1537 	return (0);
1538 }
1539 
1540 static void
1541 carp_destroy(struct carp_softc *sc)
1542 {
1543 	struct ifnet *ifp = sc->sc_carpdev;
1544 	struct carp_if *cif = ifp->if_carp;
1545 
1546 	CIF_LOCK_ASSERT(cif);
1547 
1548 	TAILQ_REMOVE(&cif->cif_vrs, sc, sc_list);
1549 
1550 	mtx_lock(&carp_mtx);
1551 	LIST_REMOVE(sc, sc_next);
1552 	mtx_unlock(&carp_mtx);
1553 
1554 	CARP_LOCK(sc);
1555 	if (sc->sc_suppress)
1556 		carp_demote_adj(-V_carp_ifdown_adj, "vhid removed");
1557 	callout_drain(&sc->sc_ad_tmo);
1558 #ifdef INET
1559 	callout_drain(&sc->sc_md_tmo);
1560 #endif
1561 #ifdef INET6
1562 	callout_drain(&sc->sc_md6_tmo);
1563 #endif
1564 	CARP_LOCK_DESTROY(sc);
1565 
1566 	free(sc->sc_ifas, M_CARP);
1567 	free(sc, M_CARP);
1568 }
1569 
1570 static struct carp_if*
1571 carp_alloc_if(struct ifnet *ifp)
1572 {
1573 	struct carp_if *cif;
1574 
1575 	cif = malloc(sizeof(*cif), M_CARP, M_WAITOK|M_ZERO);
1576 
1577 	if (ifpromisc(ifp, 1) != 0)
1578 		goto cleanup;
1579 
1580 	CIF_LOCK_INIT(cif);
1581 	cif->cif_ifp = ifp;
1582 	TAILQ_INIT(&cif->cif_vrs);
1583 
1584 	IF_ADDR_WLOCK(ifp);
1585 	ifp->if_carp = cif;
1586 	if_ref(ifp);
1587 	IF_ADDR_WUNLOCK(ifp);
1588 
1589 	return (cif);
1590 
1591 cleanup:
1592 	free(cif, M_CARP);
1593 
1594 	return (NULL);
1595 }
1596 
1597 static void
1598 carp_free_if(struct carp_if *cif)
1599 {
1600 	struct ifnet *ifp = cif->cif_ifp;
1601 
1602 	CIF_LOCK_ASSERT(cif);
1603 	KASSERT(TAILQ_EMPTY(&cif->cif_vrs), ("%s: softc list not empty",
1604 	    __func__));
1605 
1606 	IF_ADDR_WLOCK(ifp);
1607 	ifp->if_carp = NULL;
1608 	IF_ADDR_WUNLOCK(ifp);
1609 
1610 	CIF_LOCK_DESTROY(cif);
1611 
1612 	ifpromisc(ifp, 0);
1613 	if_rele(ifp);
1614 
1615 	free(cif, M_CARP);
1616 }
1617 
1618 static void
1619 carp_carprcp(struct carpreq *carpr, struct carp_softc *sc, int priv)
1620 {
1621 
1622 	CARP_LOCK(sc);
1623 	carpr->carpr_state = sc->sc_state;
1624 	carpr->carpr_vhid = sc->sc_vhid;
1625 	carpr->carpr_advbase = sc->sc_advbase;
1626 	carpr->carpr_advskew = sc->sc_advskew;
1627 	if (priv)
1628 		bcopy(sc->sc_key, carpr->carpr_key, sizeof(carpr->carpr_key));
1629 	else
1630 		bzero(carpr->carpr_key, sizeof(carpr->carpr_key));
1631 	CARP_UNLOCK(sc);
1632 }
1633 
1634 int
1635 carp_ioctl(struct ifreq *ifr, u_long cmd, struct thread *td)
1636 {
1637 	struct carpreq carpr;
1638 	struct ifnet *ifp;
1639 	struct carp_softc *sc = NULL;
1640 	int error = 0, locked = 0;
1641 
1642 	if ((error = copyin(ifr->ifr_data, &carpr, sizeof carpr)))
1643 		return (error);
1644 
1645 	ifp = ifunit_ref(ifr->ifr_name);
1646 	if (ifp == NULL)
1647 		return (ENXIO);
1648 
1649 	switch (ifp->if_type) {
1650 	case IFT_ETHER:
1651 	case IFT_L2VLAN:
1652 	case IFT_BRIDGE:
1653 	case IFT_FDDI:
1654 	case IFT_ISO88025:
1655 		break;
1656 	default:
1657 		error = EOPNOTSUPP;
1658 		goto out;
1659 	}
1660 
1661 	if ((ifp->if_flags & IFF_MULTICAST) == 0) {
1662 		error = EADDRNOTAVAIL;
1663 		goto out;
1664 	}
1665 
1666 	switch (cmd) {
1667 	case SIOCSVH:
1668 		if ((error = priv_check(td, PRIV_NETINET_CARP)))
1669 			break;
1670 		if (carpr.carpr_vhid <= 0 || carpr.carpr_vhid > CARP_MAXVHID ||
1671 		    carpr.carpr_advbase < 0 || carpr.carpr_advskew < 0) {
1672 			error = EINVAL;
1673 			break;
1674 		}
1675 
1676 		if (ifp->if_carp) {
1677 			CIF_LOCK(ifp->if_carp);
1678 			IFNET_FOREACH_CARP(ifp, sc)
1679 				if (sc->sc_vhid == carpr.carpr_vhid)
1680 					break;
1681 			CIF_UNLOCK(ifp->if_carp);
1682 		}
1683 		if (sc == NULL) {
1684 			sc = carp_alloc(ifp);
1685 			if (sc == NULL) {
1686 				error = EINVAL; /* XXX: ifpromisc failed */
1687 				break;
1688 			}
1689 
1690 			CARP_LOCK(sc);
1691 			sc->sc_vhid = carpr.carpr_vhid;
1692 			LLADDR(&sc->sc_addr)[0] = 0;
1693 			LLADDR(&sc->sc_addr)[1] = 0;
1694 			LLADDR(&sc->sc_addr)[2] = 0x5e;
1695 			LLADDR(&sc->sc_addr)[3] = 0;
1696 			LLADDR(&sc->sc_addr)[4] = 1;
1697 			LLADDR(&sc->sc_addr)[5] = sc->sc_vhid;
1698 		} else
1699 			CARP_LOCK(sc);
1700 		locked = 1;
1701 		if (carpr.carpr_advbase > 0) {
1702 			if (carpr.carpr_advbase > 255 ||
1703 			    carpr.carpr_advbase < CARP_DFLTINTV) {
1704 				error = EINVAL;
1705 				break;
1706 			}
1707 			sc->sc_advbase = carpr.carpr_advbase;
1708 		}
1709 		if (carpr.carpr_advskew > 0) {
1710 			if (carpr.carpr_advskew >= 255) {
1711 				error = EINVAL;
1712 				break;
1713 			}
1714 			sc->sc_advskew = carpr.carpr_advskew;
1715 		}
1716 		if (carpr.carpr_key[0] != '\0') {
1717 			bcopy(carpr.carpr_key, sc->sc_key, sizeof(sc->sc_key));
1718 			carp_hmac_prepare(sc);
1719 		}
1720 		if (sc->sc_state != INIT &&
1721 		    carpr.carpr_state != sc->sc_state) {
1722 			switch (carpr.carpr_state) {
1723 			case BACKUP:
1724 				callout_stop(&sc->sc_ad_tmo);
1725 				carp_set_state(sc, BACKUP);
1726 				carp_setrun(sc, 0);
1727 				carp_delroute(sc);
1728 				break;
1729 			case MASTER:
1730 				carp_master_down_locked(sc);
1731 				break;
1732 			default:
1733 				break;
1734 			}
1735 		}
1736 		break;
1737 
1738 	case SIOCGVH:
1739 	    {
1740 		int priveleged;
1741 
1742 		if (carpr.carpr_vhid < 0 || carpr.carpr_vhid > CARP_MAXVHID) {
1743 			error = EINVAL;
1744 			break;
1745 		}
1746 		if (carpr.carpr_count < 1) {
1747 			error = EMSGSIZE;
1748 			break;
1749 		}
1750 		if (ifp->if_carp == NULL) {
1751 			error = ENOENT;
1752 			break;
1753 		}
1754 
1755 		priveleged = (priv_check(td, PRIV_NETINET_CARP) == 0);
1756 		if (carpr.carpr_vhid != 0) {
1757 			CIF_LOCK(ifp->if_carp);
1758 			IFNET_FOREACH_CARP(ifp, sc)
1759 				if (sc->sc_vhid == carpr.carpr_vhid)
1760 					break;
1761 			CIF_UNLOCK(ifp->if_carp);
1762 			if (sc == NULL) {
1763 				error = ENOENT;
1764 				break;
1765 			}
1766 			carp_carprcp(&carpr, sc, priveleged);
1767 			error = copyout(&carpr, ifr->ifr_data, sizeof(carpr));
1768 		} else  {
1769 			int i, count;
1770 
1771 			count = 0;
1772 			CIF_LOCK(ifp->if_carp);
1773 			IFNET_FOREACH_CARP(ifp, sc)
1774 				count++;
1775 
1776 			if (count > carpr.carpr_count) {
1777 				CIF_UNLOCK(ifp->if_carp);
1778 				error = EMSGSIZE;
1779 				break;
1780 			}
1781 
1782 			i = 0;
1783 			IFNET_FOREACH_CARP(ifp, sc) {
1784 				carp_carprcp(&carpr, sc, priveleged);
1785 				carpr.carpr_count = count;
1786 				error = copyout(&carpr, ifr->ifr_data +
1787 				    (i * sizeof(carpr)), sizeof(carpr));
1788 				if (error) {
1789 					CIF_UNLOCK(ifp->if_carp);
1790 					break;
1791 				}
1792 				i++;
1793 			}
1794 			CIF_UNLOCK(ifp->if_carp);
1795 		}
1796 		break;
1797 	    }
1798 	default:
1799 		error = EINVAL;
1800 	}
1801 
1802 out:
1803 	if (locked)
1804 		CARP_UNLOCK(sc);
1805 	if_rele(ifp);
1806 
1807 	return (error);
1808 }
1809 
1810 static int
1811 carp_get_vhid(struct ifaddr *ifa)
1812 {
1813 
1814 	if (ifa == NULL || ifa->ifa_carp == NULL)
1815 		return (0);
1816 
1817 	return (ifa->ifa_carp->sc_vhid);
1818 }
1819 
1820 int
1821 carp_attach(struct ifaddr *ifa, int vhid)
1822 {
1823 	struct ifnet *ifp = ifa->ifa_ifp;
1824 	struct carp_if *cif = ifp->if_carp;
1825 	struct carp_softc *sc;
1826 	int index, error;
1827 
1828 	if (ifp->if_carp == NULL)
1829 		return (ENOPROTOOPT);
1830 
1831 	switch (ifa->ifa_addr->sa_family) {
1832 #ifdef INET
1833 	case AF_INET:
1834 #endif
1835 #ifdef INET6
1836 	case AF_INET6:
1837 #endif
1838 		break;
1839 	default:
1840 		return (EPROTOTYPE);
1841 	}
1842 
1843 	CIF_LOCK(cif);
1844 	IFNET_FOREACH_CARP(ifp, sc)
1845 		if (sc->sc_vhid == vhid)
1846 			break;
1847 	if (sc == NULL) {
1848 		CIF_UNLOCK(cif);
1849 		return (ENOENT);
1850 	}
1851 
1852 	if (ifa->ifa_carp) {
1853 		if (ifa->ifa_carp->sc_vhid != vhid)
1854 			carp_detach_locked(ifa);
1855 		else {
1856 			CIF_UNLOCK(cif);
1857 			return (0);
1858 		}
1859 	}
1860 
1861 	error = carp_multicast_setup(cif, ifa->ifa_addr->sa_family);
1862 	if (error) {
1863 		CIF_FREE(cif);
1864 		return (error);
1865 	}
1866 
1867 	CARP_LOCK(sc);
1868 	index = sc->sc_naddrs + sc->sc_naddrs6 + 1;
1869 	if (index > sc->sc_ifasiz / sizeof(struct ifaddr *))
1870 		if ((error = carp_grow_ifas(sc)) != 0) {
1871 			carp_multicast_cleanup(cif,
1872 			    ifa->ifa_addr->sa_family);
1873 			CARP_UNLOCK(sc);
1874 			CIF_FREE(cif);
1875 			return (error);
1876 		}
1877 
1878 	switch (ifa->ifa_addr->sa_family) {
1879 #ifdef INET
1880 	case AF_INET:
1881 		cif->cif_naddrs++;
1882 		sc->sc_naddrs++;
1883 		break;
1884 #endif
1885 #ifdef INET6
1886 	case AF_INET6:
1887 		cif->cif_naddrs6++;
1888 		sc->sc_naddrs6++;
1889 		break;
1890 #endif
1891 	}
1892 
1893 	ifa_ref(ifa);
1894 	sc->sc_ifas[index - 1] = ifa;
1895 	ifa->ifa_carp = sc;
1896 
1897 	carp_hmac_prepare(sc);
1898 	carp_sc_state(sc);
1899 
1900 	CARP_UNLOCK(sc);
1901 	CIF_UNLOCK(cif);
1902 
1903 	return (0);
1904 }
1905 
1906 void
1907 carp_detach(struct ifaddr *ifa)
1908 {
1909 	struct ifnet *ifp = ifa->ifa_ifp;
1910 	struct carp_if *cif = ifp->if_carp;
1911 
1912 	CIF_LOCK(cif);
1913 	carp_detach_locked(ifa);
1914 	CIF_FREE(cif);
1915 }
1916 
1917 static void
1918 carp_detach_locked(struct ifaddr *ifa)
1919 {
1920 	struct ifnet *ifp = ifa->ifa_ifp;
1921 	struct carp_if *cif = ifp->if_carp;
1922 	struct carp_softc *sc = ifa->ifa_carp;
1923 	int i, index;
1924 
1925 	KASSERT(sc != NULL, ("%s: %p not attached", __func__, ifa));
1926 
1927 	CIF_LOCK_ASSERT(cif);
1928 	CARP_LOCK(sc);
1929 
1930 	/* Shift array. */
1931 	index = sc->sc_naddrs + sc->sc_naddrs6;
1932 	for (i = 0; i < index; i++)
1933 		if (sc->sc_ifas[i] == ifa)
1934 			break;
1935 	KASSERT(i < index, ("%s: %p no backref", __func__, ifa));
1936 	for (; i < index - 1; i++)
1937 		sc->sc_ifas[i] = sc->sc_ifas[i+1];
1938 	sc->sc_ifas[index - 1] = NULL;
1939 
1940 	switch (ifa->ifa_addr->sa_family) {
1941 #ifdef INET
1942 	case AF_INET:
1943 		cif->cif_naddrs--;
1944 		sc->sc_naddrs--;
1945 		break;
1946 #endif
1947 #ifdef INET6
1948 	case AF_INET6:
1949 		cif->cif_naddrs6--;
1950 		sc->sc_naddrs6--;
1951 		break;
1952 #endif
1953 	}
1954 
1955 	carp_ifa_delroute(ifa);
1956 	carp_multicast_cleanup(cif, ifa->ifa_addr->sa_family);
1957 
1958 	ifa->ifa_carp = NULL;
1959 	ifa_free(ifa);
1960 
1961 	carp_hmac_prepare(sc);
1962 	carp_sc_state(sc);
1963 
1964 	if (sc->sc_naddrs == 0 && sc->sc_naddrs6 == 0) {
1965 		CARP_UNLOCK(sc);
1966 		carp_destroy(sc);
1967 	} else
1968 		CARP_UNLOCK(sc);
1969 }
1970 
1971 static void
1972 carp_set_state(struct carp_softc *sc, int state)
1973 {
1974 
1975 	CARP_LOCK_ASSERT(sc);
1976 
1977 	if (sc->sc_state != state) {
1978 		const char *carp_states[] = { CARP_STATES };
1979 		char subsys[IFNAMSIZ+5];
1980 
1981 		sc->sc_state = state;
1982 
1983 		snprintf(subsys, IFNAMSIZ+5, "%u@%s", sc->sc_vhid,
1984 		    sc->sc_carpdev->if_xname);
1985 		devctl_notify("CARP", subsys, carp_states[state], NULL);
1986 	}
1987 }
1988 
1989 static void
1990 carp_linkstate(struct ifnet *ifp)
1991 {
1992 	struct carp_softc *sc;
1993 
1994 	CIF_LOCK(ifp->if_carp);
1995 	IFNET_FOREACH_CARP(ifp, sc) {
1996 		CARP_LOCK(sc);
1997 		carp_sc_state(sc);
1998 		CARP_UNLOCK(sc);
1999 	}
2000 	CIF_UNLOCK(ifp->if_carp);
2001 }
2002 
2003 static void
2004 carp_sc_state(struct carp_softc *sc)
2005 {
2006 
2007 	CARP_LOCK_ASSERT(sc);
2008 
2009 	if (sc->sc_carpdev->if_link_state != LINK_STATE_UP ||
2010 	    !(sc->sc_carpdev->if_flags & IFF_UP)) {
2011 		callout_stop(&sc->sc_ad_tmo);
2012 #ifdef INET
2013 		callout_stop(&sc->sc_md_tmo);
2014 #endif
2015 #ifdef INET6
2016 		callout_stop(&sc->sc_md6_tmo);
2017 #endif
2018 		carp_set_state(sc, INIT);
2019 		carp_setrun(sc, 0);
2020 		if (!sc->sc_suppress)
2021 			carp_demote_adj(V_carp_ifdown_adj, "interface down");
2022 		sc->sc_suppress = 1;
2023 	} else {
2024 		carp_set_state(sc, INIT);
2025 		carp_setrun(sc, 0);
2026 		if (sc->sc_suppress)
2027 			carp_demote_adj(-V_carp_ifdown_adj, "interface up");
2028 		sc->sc_suppress = 0;
2029 	}
2030 }
2031 
2032 static void
2033 carp_demote_adj(int adj, char *reason)
2034 {
2035 	atomic_add_int(&V_carp_demotion, adj);
2036 	CARP_LOG("demoted by %d to %d (%s)\n", adj, V_carp_demotion, reason);
2037 	taskqueue_enqueue(taskqueue_swi, &carp_sendall_task);
2038 }
2039 
2040 static int
2041 carp_demote_adj_sysctl(SYSCTL_HANDLER_ARGS)
2042 {
2043 	int new, error;
2044 
2045 	new = V_carp_demotion;
2046 	error = sysctl_handle_int(oidp, &new, 0, req);
2047 	if (error || !req->newptr)
2048 		return (error);
2049 
2050 	carp_demote_adj(new, "sysctl");
2051 
2052 	return (0);
2053 }
2054 
2055 #ifdef INET
2056 extern  struct domain inetdomain;
2057 static struct protosw in_carp_protosw = {
2058 	.pr_type =		SOCK_RAW,
2059 	.pr_domain =		&inetdomain,
2060 	.pr_protocol =		IPPROTO_CARP,
2061 	.pr_flags =		PR_ATOMIC|PR_ADDR,
2062 	.pr_input =		carp_input,
2063 	.pr_output =		(pr_output_t *)rip_output,
2064 	.pr_ctloutput =		rip_ctloutput,
2065 	.pr_usrreqs =		&rip_usrreqs
2066 };
2067 #endif
2068 
2069 #ifdef INET6
2070 extern	struct domain inet6domain;
2071 static struct ip6protosw in6_carp_protosw = {
2072 	.pr_type =		SOCK_RAW,
2073 	.pr_domain =		&inet6domain,
2074 	.pr_protocol =		IPPROTO_CARP,
2075 	.pr_flags =		PR_ATOMIC|PR_ADDR,
2076 	.pr_input =		carp6_input,
2077 	.pr_output =		rip6_output,
2078 	.pr_ctloutput =		rip6_ctloutput,
2079 	.pr_usrreqs =		&rip6_usrreqs
2080 };
2081 #endif
2082 
2083 static void
2084 carp_mod_cleanup(void)
2085 {
2086 
2087 #ifdef INET
2088 	if (proto_reg[CARP_INET] == 0) {
2089 		(void)ipproto_unregister(IPPROTO_CARP);
2090 		pf_proto_unregister(PF_INET, IPPROTO_CARP, SOCK_RAW);
2091 		proto_reg[CARP_INET] = -1;
2092 	}
2093 	carp_iamatch_p = NULL;
2094 #endif
2095 #ifdef INET6
2096 	if (proto_reg[CARP_INET6] == 0) {
2097 		(void)ip6proto_unregister(IPPROTO_CARP);
2098 		pf_proto_unregister(PF_INET6, IPPROTO_CARP, SOCK_RAW);
2099 		proto_reg[CARP_INET6] = -1;
2100 	}
2101 	carp_iamatch6_p = NULL;
2102 	carp_macmatch6_p = NULL;
2103 #endif
2104 	carp_ioctl_p = NULL;
2105 	carp_attach_p = NULL;
2106 	carp_detach_p = NULL;
2107 	carp_get_vhid_p = NULL;
2108 	carp_linkstate_p = NULL;
2109 	carp_forus_p = NULL;
2110 	carp_output_p = NULL;
2111 	carp_demote_adj_p = NULL;
2112 	carp_master_p = NULL;
2113 	mtx_unlock(&carp_mtx);
2114 	taskqueue_drain(taskqueue_swi, &carp_sendall_task);
2115 	mtx_destroy(&carp_mtx);
2116 }
2117 
2118 static int
2119 carp_mod_load(void)
2120 {
2121 	int err;
2122 
2123 	mtx_init(&carp_mtx, "carp_mtx", NULL, MTX_DEF);
2124 	LIST_INIT(&carp_list);
2125 	carp_get_vhid_p = carp_get_vhid;
2126 	carp_forus_p = carp_forus;
2127 	carp_output_p = carp_output;
2128 	carp_linkstate_p = carp_linkstate;
2129 	carp_ioctl_p = carp_ioctl;
2130 	carp_attach_p = carp_attach;
2131 	carp_detach_p = carp_detach;
2132 	carp_demote_adj_p = carp_demote_adj;
2133 	carp_master_p = carp_master;
2134 #ifdef INET6
2135 	carp_iamatch6_p = carp_iamatch6;
2136 	carp_macmatch6_p = carp_macmatch6;
2137 	proto_reg[CARP_INET6] = pf_proto_register(PF_INET6,
2138 	    (struct protosw *)&in6_carp_protosw);
2139 	if (proto_reg[CARP_INET6]) {
2140 		printf("carp: error %d attaching to PF_INET6\n",
2141 		    proto_reg[CARP_INET6]);
2142 		carp_mod_cleanup();
2143 		return (proto_reg[CARP_INET6]);
2144 	}
2145 	err = ip6proto_register(IPPROTO_CARP);
2146 	if (err) {
2147 		printf("carp: error %d registering with INET6\n", err);
2148 		carp_mod_cleanup();
2149 		return (err);
2150 	}
2151 #endif
2152 #ifdef INET
2153 	carp_iamatch_p = carp_iamatch;
2154 	proto_reg[CARP_INET] = pf_proto_register(PF_INET, &in_carp_protosw);
2155 	if (proto_reg[CARP_INET]) {
2156 		printf("carp: error %d attaching to PF_INET\n",
2157 		    proto_reg[CARP_INET]);
2158 		carp_mod_cleanup();
2159 		return (proto_reg[CARP_INET]);
2160 	}
2161 	err = ipproto_register(IPPROTO_CARP);
2162 	if (err) {
2163 		printf("carp: error %d registering with INET\n", err);
2164 		carp_mod_cleanup();
2165 		return (err);
2166 	}
2167 #endif
2168 	return (0);
2169 }
2170 
2171 static int
2172 carp_modevent(module_t mod, int type, void *data)
2173 {
2174 	switch (type) {
2175 	case MOD_LOAD:
2176 		return carp_mod_load();
2177 		/* NOTREACHED */
2178 	case MOD_UNLOAD:
2179 		mtx_lock(&carp_mtx);
2180 		if (LIST_EMPTY(&carp_list))
2181 			carp_mod_cleanup();
2182 		else {
2183 			mtx_unlock(&carp_mtx);
2184 			return (EBUSY);
2185 		}
2186 		break;
2187 
2188 	default:
2189 		return (EINVAL);
2190 	}
2191 
2192 	return (0);
2193 }
2194 
2195 static moduledata_t carp_mod = {
2196 	"carp",
2197 	carp_modevent,
2198 	0
2199 };
2200 
2201 DECLARE_MODULE(carp, carp_mod, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY);
2202