xref: /freebsd/sys/netinet/ip_carp.c (revision 38a52bd3b5cac3da6f7f6eef3dd050e6aa08ebb3)
1 /*-
2  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3  *
4  * Copyright (c) 2002 Michael Shalayeff.
5  * Copyright (c) 2003 Ryan McBride.
6  * Copyright (c) 2011 Gleb Smirnoff <glebius@FreeBSD.org>
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  *
18  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
19  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
20  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
21  * IN NO EVENT SHALL THE AUTHOR OR HIS RELATIVES BE LIABLE FOR ANY DIRECT,
22  * INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES
23  * (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR
24  * SERVICES; LOSS OF MIND, USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT,
26  * STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING
27  * IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF
28  * THE POSSIBILITY OF SUCH DAMAGE.
29  */
30 
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
33 
34 #include "opt_bpf.h"
35 #include "opt_inet.h"
36 #include "opt_inet6.h"
37 
38 #include <sys/param.h>
39 #include <sys/systm.h>
40 #include <sys/devctl.h>
41 #include <sys/jail.h>
42 #include <sys/kernel.h>
43 #include <sys/limits.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/module.h>
47 #include <sys/priv.h>
48 #include <sys/proc.h>
49 #include <sys/socket.h>
50 #include <sys/sockio.h>
51 #include <sys/sysctl.h>
52 #include <sys/syslog.h>
53 #include <sys/taskqueue.h>
54 #include <sys/counter.h>
55 
56 #include <net/ethernet.h>
57 #include <net/if.h>
58 #include <net/if_var.h>
59 #include <net/if_dl.h>
60 #include <net/if_llatbl.h>
61 #include <net/if_types.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/in6_var.h>
81 #include <netinet6/ip6_var.h>
82 #include <netinet6/scope6_var.h>
83 #include <netinet6/nd6.h>
84 #endif
85 
86 #include <crypto/sha1.h>
87 
88 static MALLOC_DEFINE(M_CARP, "CARP", "CARP addresses");
89 
90 struct carp_softc {
91 	struct ifnet		*sc_carpdev;	/* Pointer to parent ifnet. */
92 	struct ifaddr		**sc_ifas;	/* Our ifaddrs. */
93 	struct sockaddr_dl	sc_addr;	/* Our link level address. */
94 	struct callout		sc_ad_tmo;	/* Advertising timeout. */
95 #ifdef INET
96 	struct callout		sc_md_tmo;	/* Master down timeout. */
97 #endif
98 #ifdef INET6
99 	struct callout 		sc_md6_tmo;	/* XXX: Master down timeout. */
100 #endif
101 	struct mtx		sc_mtx;
102 
103 	int			sc_vhid;
104 	int			sc_advskew;
105 	int			sc_advbase;
106 
107 	int			sc_naddrs;
108 	int			sc_naddrs6;
109 	int			sc_ifasiz;
110 	enum { INIT = 0, BACKUP, MASTER }	sc_state;
111 	int			sc_suppress;
112 	int			sc_sendad_errors;
113 #define	CARP_SENDAD_MAX_ERRORS	3
114 	int			sc_sendad_success;
115 #define	CARP_SENDAD_MIN_SUCCESS 3
116 
117 	int			sc_init_counter;
118 	uint64_t		sc_counter;
119 
120 	/* authentication */
121 #define	CARP_HMAC_PAD	64
122 	unsigned char sc_key[CARP_KEY_LEN];
123 	unsigned char sc_pad[CARP_HMAC_PAD];
124 	SHA1_CTX sc_sha1;
125 
126 	TAILQ_ENTRY(carp_softc)	sc_list;	/* On the carp_if list. */
127 	LIST_ENTRY(carp_softc)	sc_next;	/* On the global list. */
128 };
129 
130 struct carp_if {
131 #ifdef INET
132 	int	cif_naddrs;
133 #endif
134 #ifdef INET6
135 	int	cif_naddrs6;
136 #endif
137 	TAILQ_HEAD(, carp_softc) cif_vrs;
138 #ifdef INET
139 	struct ip_moptions 	 cif_imo;
140 #endif
141 #ifdef INET6
142 	struct ip6_moptions 	 cif_im6o;
143 #endif
144 	struct ifnet	*cif_ifp;
145 	struct mtx	cif_mtx;
146 	uint32_t	cif_flags;
147 #define	CIF_PROMISC	0x00000001
148 };
149 
150 /*
151  * Brief design of carp(4).
152  *
153  * Any carp-capable ifnet may have a list of carp softcs hanging off
154  * its ifp->if_carp pointer. Each softc represents one unique virtual
155  * host id, or vhid. The softc has a back pointer to the ifnet. All
156  * softcs are joined in a global list, which has quite limited use.
157  *
158  * Any interface address that takes part in CARP negotiation has a
159  * pointer to the softc of its vhid, ifa->ifa_carp. That could be either
160  * AF_INET or AF_INET6 address.
161  *
162  * Although, one can get the softc's backpointer to ifnet and traverse
163  * through its ifp->if_addrhead queue to find all interface addresses
164  * involved in CARP, we keep a growable array of ifaddr pointers. This
165  * allows us to avoid grabbing the IF_ADDR_LOCK() in many traversals that
166  * do calls into the network stack, thus avoiding LORs.
167  *
168  * Locking:
169  *
170  * Each softc has a lock sc_mtx. It is used to synchronise carp_input_c(),
171  * callout-driven events and ioctl()s.
172  *
173  * To traverse the list of softcs on an ifnet we use CIF_LOCK() or carp_sx.
174  * To traverse the global list we use the mutex carp_mtx.
175  *
176  * Known issues with locking:
177  *
178  * - Sending ad, we put the pointer to the softc in an mtag, and no reference
179  *   counting is done on the softc.
180  * - On module unload we may race (?) with packet processing thread
181  *   dereferencing our function pointers.
182  */
183 
184 /* Accept incoming CARP packets. */
185 VNET_DEFINE_STATIC(int, carp_allow) = 1;
186 #define	V_carp_allow	VNET(carp_allow)
187 
188 /* Set DSCP in outgoing CARP packets. */
189 VNET_DEFINE_STATIC(int, carp_dscp) = 56;
190 #define	V_carp_dscp	VNET(carp_dscp)
191 
192 /* Preempt slower nodes. */
193 VNET_DEFINE_STATIC(int, carp_preempt) = 0;
194 #define	V_carp_preempt	VNET(carp_preempt)
195 
196 /* Log level. */
197 VNET_DEFINE_STATIC(int, carp_log) = 1;
198 #define	V_carp_log	VNET(carp_log)
199 
200 /* Global advskew demotion. */
201 VNET_DEFINE_STATIC(int, carp_demotion) = 0;
202 #define	V_carp_demotion	VNET(carp_demotion)
203 
204 /* Send error demotion factor. */
205 VNET_DEFINE_STATIC(int, carp_senderr_adj) = CARP_MAXSKEW;
206 #define	V_carp_senderr_adj	VNET(carp_senderr_adj)
207 
208 /* Iface down demotion factor. */
209 VNET_DEFINE_STATIC(int, carp_ifdown_adj) = CARP_MAXSKEW;
210 #define	V_carp_ifdown_adj	VNET(carp_ifdown_adj)
211 
212 static int carp_allow_sysctl(SYSCTL_HANDLER_ARGS);
213 static int carp_dscp_sysctl(SYSCTL_HANDLER_ARGS);
214 static int carp_demote_adj_sysctl(SYSCTL_HANDLER_ARGS);
215 
216 SYSCTL_NODE(_net_inet, IPPROTO_CARP, carp, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
217     "CARP");
218 SYSCTL_PROC(_net_inet_carp, OID_AUTO, allow,
219     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
220     0, 0, carp_allow_sysctl, "I",
221     "Accept incoming CARP packets");
222 SYSCTL_PROC(_net_inet_carp, OID_AUTO, dscp,
223     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
224     0, 0, carp_dscp_sysctl, "I",
225     "DSCP value for carp packets");
226 SYSCTL_INT(_net_inet_carp, OID_AUTO, preempt, CTLFLAG_VNET | CTLFLAG_RW,
227     &VNET_NAME(carp_preempt), 0, "High-priority backup preemption mode");
228 SYSCTL_INT(_net_inet_carp, OID_AUTO, log, CTLFLAG_VNET | CTLFLAG_RW,
229     &VNET_NAME(carp_log), 0, "CARP log level");
230 SYSCTL_PROC(_net_inet_carp, OID_AUTO, demotion,
231     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
232     0, 0, carp_demote_adj_sysctl, "I",
233     "Adjust demotion factor (skew of advskew)");
234 SYSCTL_INT(_net_inet_carp, OID_AUTO, senderr_demotion_factor,
235     CTLFLAG_VNET | CTLFLAG_RW,
236     &VNET_NAME(carp_senderr_adj), 0, "Send error demotion factor adjustment");
237 SYSCTL_INT(_net_inet_carp, OID_AUTO, ifdown_demotion_factor,
238     CTLFLAG_VNET | CTLFLAG_RW,
239     &VNET_NAME(carp_ifdown_adj), 0,
240     "Interface down demotion factor adjustment");
241 
242 VNET_PCPUSTAT_DEFINE(struct carpstats, carpstats);
243 VNET_PCPUSTAT_SYSINIT(carpstats);
244 VNET_PCPUSTAT_SYSUNINIT(carpstats);
245 
246 #define	CARPSTATS_ADD(name, val)	\
247     counter_u64_add(VNET(carpstats)[offsetof(struct carpstats, name) / \
248 	sizeof(uint64_t)], (val))
249 #define	CARPSTATS_INC(name)		CARPSTATS_ADD(name, 1)
250 
251 SYSCTL_VNET_PCPUSTAT(_net_inet_carp, OID_AUTO, stats, struct carpstats,
252     carpstats, "CARP statistics (struct carpstats, netinet/ip_carp.h)");
253 
254 #define	CARP_LOCK_INIT(sc)	mtx_init(&(sc)->sc_mtx, "carp_softc",   \
255 	NULL, MTX_DEF)
256 #define	CARP_LOCK_DESTROY(sc)	mtx_destroy(&(sc)->sc_mtx)
257 #define	CARP_LOCK_ASSERT(sc)	mtx_assert(&(sc)->sc_mtx, MA_OWNED)
258 #define	CARP_LOCK(sc)		mtx_lock(&(sc)->sc_mtx)
259 #define	CARP_UNLOCK(sc)		mtx_unlock(&(sc)->sc_mtx)
260 #define	CIF_LOCK_INIT(cif)	mtx_init(&(cif)->cif_mtx, "carp_if",   \
261 	NULL, MTX_DEF)
262 #define	CIF_LOCK_DESTROY(cif)	mtx_destroy(&(cif)->cif_mtx)
263 #define	CIF_LOCK_ASSERT(cif)	mtx_assert(&(cif)->cif_mtx, MA_OWNED)
264 #define	CIF_LOCK(cif)		mtx_lock(&(cif)->cif_mtx)
265 #define	CIF_UNLOCK(cif)		mtx_unlock(&(cif)->cif_mtx)
266 #define	CIF_FREE(cif)	do {				\
267 		CIF_LOCK(cif);				\
268 		if (TAILQ_EMPTY(&(cif)->cif_vrs))	\
269 			carp_free_if(cif);		\
270 		else					\
271 			CIF_UNLOCK(cif);		\
272 } while (0)
273 
274 #define	CARP_LOG(...)	do {				\
275 	if (V_carp_log > 0)				\
276 		log(LOG_INFO, "carp: " __VA_ARGS__);	\
277 } while (0)
278 
279 #define	CARP_DEBUG(...)	do {				\
280 	if (V_carp_log > 1)				\
281 		log(LOG_DEBUG, __VA_ARGS__);		\
282 } while (0)
283 
284 #define	IFNET_FOREACH_IFA(ifp, ifa)					\
285 	CK_STAILQ_FOREACH((ifa), &(ifp)->if_addrhead, ifa_link)	\
286 		if ((ifa)->ifa_carp != NULL)
287 
288 #define	CARP_FOREACH_IFA(sc, ifa)					\
289 	CARP_LOCK_ASSERT(sc);						\
290 	for (int _i = 0;						\
291 		_i < (sc)->sc_naddrs + (sc)->sc_naddrs6 &&		\
292 		((ifa) = sc->sc_ifas[_i]) != NULL;			\
293 		++_i)
294 
295 #define	IFNET_FOREACH_CARP(ifp, sc)					\
296 	KASSERT(mtx_owned(&ifp->if_carp->cif_mtx) ||			\
297 	    sx_xlocked(&carp_sx), ("cif_vrs not locked"));		\
298 	TAILQ_FOREACH((sc), &(ifp)->if_carp->cif_vrs, sc_list)
299 
300 #define	DEMOTE_ADVSKEW(sc)					\
301     (((sc)->sc_advskew + V_carp_demotion > CARP_MAXSKEW) ?	\
302     CARP_MAXSKEW :						\
303         (((sc)->sc_advskew + V_carp_demotion < 0) ?		\
304         0 : ((sc)->sc_advskew + V_carp_demotion)))
305 
306 static void	carp_input_c(struct mbuf *, struct carp_header *, sa_family_t);
307 static struct carp_softc
308 		*carp_alloc(struct ifnet *);
309 static void	carp_destroy(struct carp_softc *);
310 static struct carp_if
311 		*carp_alloc_if(struct ifnet *);
312 static void	carp_free_if(struct carp_if *);
313 static void	carp_set_state(struct carp_softc *, int, const char* reason);
314 static void	carp_sc_state(struct carp_softc *);
315 static void	carp_setrun(struct carp_softc *, sa_family_t);
316 static void	carp_master_down(void *);
317 static void	carp_master_down_locked(struct carp_softc *,
318     		    const char* reason);
319 static void	carp_send_ad(void *);
320 static void	carp_send_ad_locked(struct carp_softc *);
321 static void	carp_addroute(struct carp_softc *);
322 static void	carp_ifa_addroute(struct ifaddr *);
323 static void	carp_delroute(struct carp_softc *);
324 static void	carp_ifa_delroute(struct ifaddr *);
325 static void	carp_send_ad_all(void *, int);
326 static void	carp_demote_adj(int, char *);
327 
328 static LIST_HEAD(, carp_softc) carp_list;
329 static struct mtx carp_mtx;
330 static struct sx carp_sx;
331 static struct task carp_sendall_task =
332     TASK_INITIALIZER(0, carp_send_ad_all, NULL);
333 
334 static void
335 carp_hmac_prepare(struct carp_softc *sc)
336 {
337 	uint8_t version = CARP_VERSION, type = CARP_ADVERTISEMENT;
338 	uint8_t vhid = sc->sc_vhid & 0xff;
339 	struct ifaddr *ifa;
340 	int i, found;
341 #ifdef INET
342 	struct in_addr last, cur, in;
343 #endif
344 #ifdef INET6
345 	struct in6_addr last6, cur6, in6;
346 #endif
347 
348 	CARP_LOCK_ASSERT(sc);
349 
350 	/* Compute ipad from key. */
351 	bzero(sc->sc_pad, sizeof(sc->sc_pad));
352 	bcopy(sc->sc_key, sc->sc_pad, sizeof(sc->sc_key));
353 	for (i = 0; i < sizeof(sc->sc_pad); i++)
354 		sc->sc_pad[i] ^= 0x36;
355 
356 	/* Precompute first part of inner hash. */
357 	SHA1Init(&sc->sc_sha1);
358 	SHA1Update(&sc->sc_sha1, sc->sc_pad, sizeof(sc->sc_pad));
359 	SHA1Update(&sc->sc_sha1, (void *)&version, sizeof(version));
360 	SHA1Update(&sc->sc_sha1, (void *)&type, sizeof(type));
361 	SHA1Update(&sc->sc_sha1, (void *)&vhid, sizeof(vhid));
362 #ifdef INET
363 	cur.s_addr = 0;
364 	do {
365 		found = 0;
366 		last = cur;
367 		cur.s_addr = 0xffffffff;
368 		CARP_FOREACH_IFA(sc, ifa) {
369 			in.s_addr = ifatoia(ifa)->ia_addr.sin_addr.s_addr;
370 			if (ifa->ifa_addr->sa_family == AF_INET &&
371 			    ntohl(in.s_addr) > ntohl(last.s_addr) &&
372 			    ntohl(in.s_addr) < ntohl(cur.s_addr)) {
373 				cur.s_addr = in.s_addr;
374 				found++;
375 			}
376 		}
377 		if (found)
378 			SHA1Update(&sc->sc_sha1, (void *)&cur, sizeof(cur));
379 	} while (found);
380 #endif /* INET */
381 #ifdef INET6
382 	memset(&cur6, 0, sizeof(cur6));
383 	do {
384 		found = 0;
385 		last6 = cur6;
386 		memset(&cur6, 0xff, sizeof(cur6));
387 		CARP_FOREACH_IFA(sc, ifa) {
388 			in6 = ifatoia6(ifa)->ia_addr.sin6_addr;
389 			if (IN6_IS_SCOPE_EMBED(&in6))
390 				in6.s6_addr16[1] = 0;
391 			if (ifa->ifa_addr->sa_family == AF_INET6 &&
392 			    memcmp(&in6, &last6, sizeof(in6)) > 0 &&
393 			    memcmp(&in6, &cur6, sizeof(in6)) < 0) {
394 				cur6 = in6;
395 				found++;
396 			}
397 		}
398 		if (found)
399 			SHA1Update(&sc->sc_sha1, (void *)&cur6, sizeof(cur6));
400 	} while (found);
401 #endif /* INET6 */
402 
403 	/* convert ipad to opad */
404 	for (i = 0; i < sizeof(sc->sc_pad); i++)
405 		sc->sc_pad[i] ^= 0x36 ^ 0x5c;
406 }
407 
408 static void
409 carp_hmac_generate(struct carp_softc *sc, uint32_t counter[2],
410     unsigned char md[20])
411 {
412 	SHA1_CTX sha1ctx;
413 
414 	CARP_LOCK_ASSERT(sc);
415 
416 	/* fetch first half of inner hash */
417 	bcopy(&sc->sc_sha1, &sha1ctx, sizeof(sha1ctx));
418 
419 	SHA1Update(&sha1ctx, (void *)counter, sizeof(sc->sc_counter));
420 	SHA1Final(md, &sha1ctx);
421 
422 	/* outer hash */
423 	SHA1Init(&sha1ctx);
424 	SHA1Update(&sha1ctx, sc->sc_pad, sizeof(sc->sc_pad));
425 	SHA1Update(&sha1ctx, md, 20);
426 	SHA1Final(md, &sha1ctx);
427 }
428 
429 static int
430 carp_hmac_verify(struct carp_softc *sc, uint32_t counter[2],
431     unsigned char md[20])
432 {
433 	unsigned char md2[20];
434 
435 	CARP_LOCK_ASSERT(sc);
436 
437 	carp_hmac_generate(sc, counter, md2);
438 
439 	return (bcmp(md, md2, sizeof(md2)));
440 }
441 
442 /*
443  * process input packet.
444  * we have rearranged checks order compared to the rfc,
445  * but it seems more efficient this way or not possible otherwise.
446  */
447 #ifdef INET
448 static int
449 carp_input(struct mbuf **mp, int *offp, int proto)
450 {
451 	struct mbuf *m = *mp;
452 	struct ip *ip = mtod(m, struct ip *);
453 	struct carp_header *ch;
454 	int iplen, len;
455 
456 	iplen = *offp;
457 	*mp = NULL;
458 
459 	CARPSTATS_INC(carps_ipackets);
460 
461 	if (!V_carp_allow) {
462 		m_freem(m);
463 		return (IPPROTO_DONE);
464 	}
465 
466 	/* verify that the IP TTL is 255.  */
467 	if (ip->ip_ttl != CARP_DFLTTL) {
468 		CARPSTATS_INC(carps_badttl);
469 		CARP_DEBUG("%s: received ttl %d != 255 on %s\n", __func__,
470 		    ip->ip_ttl,
471 		    m->m_pkthdr.rcvif->if_xname);
472 		m_freem(m);
473 		return (IPPROTO_DONE);
474 	}
475 
476 	iplen = ip->ip_hl << 2;
477 
478 	if (m->m_pkthdr.len < iplen + sizeof(*ch)) {
479 		CARPSTATS_INC(carps_badlen);
480 		CARP_DEBUG("%s: received len %zd < sizeof(struct carp_header) "
481 		    "on %s\n", __func__, m->m_len - sizeof(struct ip),
482 		    m->m_pkthdr.rcvif->if_xname);
483 		m_freem(m);
484 		return (IPPROTO_DONE);
485 	}
486 
487 	if (iplen + sizeof(*ch) < m->m_len) {
488 		if ((m = m_pullup(m, iplen + sizeof(*ch))) == NULL) {
489 			CARPSTATS_INC(carps_hdrops);
490 			CARP_DEBUG("%s: pullup failed\n", __func__);
491 			return (IPPROTO_DONE);
492 		}
493 		ip = mtod(m, struct ip *);
494 	}
495 	ch = (struct carp_header *)((char *)ip + iplen);
496 
497 	/*
498 	 * verify that the received packet length is
499 	 * equal to the CARP header
500 	 */
501 	len = iplen + sizeof(*ch);
502 	if (len > m->m_pkthdr.len) {
503 		CARPSTATS_INC(carps_badlen);
504 		CARP_DEBUG("%s: packet too short %d on %s\n", __func__,
505 		    m->m_pkthdr.len,
506 		    m->m_pkthdr.rcvif->if_xname);
507 		m_freem(m);
508 		return (IPPROTO_DONE);
509 	}
510 
511 	if ((m = m_pullup(m, len)) == NULL) {
512 		CARPSTATS_INC(carps_hdrops);
513 		return (IPPROTO_DONE);
514 	}
515 	ip = mtod(m, struct ip *);
516 	ch = (struct carp_header *)((char *)ip + iplen);
517 
518 	/* verify the CARP checksum */
519 	m->m_data += iplen;
520 	if (in_cksum(m, len - iplen)) {
521 		CARPSTATS_INC(carps_badsum);
522 		CARP_DEBUG("%s: checksum failed on %s\n", __func__,
523 		    m->m_pkthdr.rcvif->if_xname);
524 		m_freem(m);
525 		return (IPPROTO_DONE);
526 	}
527 	m->m_data -= iplen;
528 
529 	carp_input_c(m, ch, AF_INET);
530 	return (IPPROTO_DONE);
531 }
532 #endif
533 
534 #ifdef INET6
535 static int
536 carp6_input(struct mbuf **mp, int *offp, int proto)
537 {
538 	struct mbuf *m = *mp;
539 	struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
540 	struct carp_header *ch;
541 	u_int len;
542 
543 	CARPSTATS_INC(carps_ipackets6);
544 
545 	if (!V_carp_allow) {
546 		m_freem(m);
547 		return (IPPROTO_DONE);
548 	}
549 
550 	/* check if received on a valid carp interface */
551 	if (m->m_pkthdr.rcvif->if_carp == NULL) {
552 		CARPSTATS_INC(carps_badif);
553 		CARP_DEBUG("%s: packet received on non-carp interface: %s\n",
554 		    __func__, m->m_pkthdr.rcvif->if_xname);
555 		m_freem(m);
556 		return (IPPROTO_DONE);
557 	}
558 
559 	/* verify that the IP TTL is 255 */
560 	if (ip6->ip6_hlim != CARP_DFLTTL) {
561 		CARPSTATS_INC(carps_badttl);
562 		CARP_DEBUG("%s: received ttl %d != 255 on %s\n", __func__,
563 		    ip6->ip6_hlim, m->m_pkthdr.rcvif->if_xname);
564 		m_freem(m);
565 		return (IPPROTO_DONE);
566 	}
567 
568 	/* verify that we have a complete carp packet */
569 	if (m->m_len < *offp + sizeof(*ch)) {
570 		len = m->m_len;
571 		m = m_pullup(m, *offp + sizeof(*ch));
572 		if (m == NULL) {
573 			CARPSTATS_INC(carps_badlen);
574 			CARP_DEBUG("%s: packet size %u too small\n", __func__, len);
575 			return (IPPROTO_DONE);
576 		}
577 	}
578 	ch = (struct carp_header *)(mtod(m, char *) + *offp);
579 
580 	/* verify the CARP checksum */
581 	m->m_data += *offp;
582 	if (in_cksum(m, sizeof(*ch))) {
583 		CARPSTATS_INC(carps_badsum);
584 		CARP_DEBUG("%s: checksum failed, on %s\n", __func__,
585 		    m->m_pkthdr.rcvif->if_xname);
586 		m_freem(m);
587 		return (IPPROTO_DONE);
588 	}
589 	m->m_data -= *offp;
590 
591 	carp_input_c(m, ch, AF_INET6);
592 	return (IPPROTO_DONE);
593 }
594 #endif /* INET6 */
595 
596 /*
597  * This routine should not be necessary at all, but some switches
598  * (VMWare ESX vswitches) can echo our own packets back at us,
599  * and we must ignore them or they will cause us to drop out of
600  * MASTER mode.
601  *
602  * We cannot catch all cases of network loops.  Instead, what we
603  * do here is catch any packet that arrives with a carp header
604  * with a VHID of 0, that comes from an address that is our own.
605  * These packets are by definition "from us" (even if they are from
606  * a misconfigured host that is pretending to be us).
607  *
608  * The VHID test is outside this mini-function.
609  */
610 static int
611 carp_source_is_self(struct mbuf *m, struct ifaddr *ifa, sa_family_t af)
612 {
613 #ifdef INET
614 	struct ip *ip4;
615 	struct in_addr in4;
616 #endif
617 #ifdef INET6
618 	struct ip6_hdr *ip6;
619 	struct in6_addr in6;
620 #endif
621 
622 	switch (af) {
623 #ifdef INET
624 	case AF_INET:
625 		ip4 = mtod(m, struct ip *);
626 		in4 = ifatoia(ifa)->ia_addr.sin_addr;
627 		return (in4.s_addr == ip4->ip_src.s_addr);
628 #endif
629 #ifdef INET6
630 	case AF_INET6:
631 		ip6 = mtod(m, struct ip6_hdr *);
632 		in6 = ifatoia6(ifa)->ia_addr.sin6_addr;
633 		return (memcmp(&in6, &ip6->ip6_src, sizeof(in6)) == 0);
634 #endif
635 	default:
636 		break;
637 	}
638 	return (0);
639 }
640 
641 static void
642 carp_input_c(struct mbuf *m, struct carp_header *ch, sa_family_t af)
643 {
644 	struct ifnet *ifp = m->m_pkthdr.rcvif;
645 	struct ifaddr *ifa, *match;
646 	struct carp_softc *sc;
647 	uint64_t tmp_counter;
648 	struct timeval sc_tv, ch_tv;
649 	int error;
650 
651 	NET_EPOCH_ASSERT();
652 
653 	/*
654 	 * Verify that the VHID is valid on the receiving interface.
655 	 *
656 	 * There should be just one match.  If there are none
657 	 * the VHID is not valid and we drop the packet.  If
658 	 * there are multiple VHID matches, take just the first
659 	 * one, for compatibility with previous code.  While we're
660 	 * scanning, check for obvious loops in the network topology
661 	 * (these should never happen, and as noted above, we may
662 	 * miss real loops; this is just a double-check).
663 	 */
664 	error = 0;
665 	match = NULL;
666 	IFNET_FOREACH_IFA(ifp, ifa) {
667 		if (match == NULL && ifa->ifa_carp != NULL &&
668 		    ifa->ifa_addr->sa_family == af &&
669 		    ifa->ifa_carp->sc_vhid == ch->carp_vhid)
670 			match = ifa;
671 		if (ch->carp_vhid == 0 && carp_source_is_self(m, ifa, af))
672 			error = ELOOP;
673 	}
674 	ifa = error ? NULL : match;
675 	if (ifa != NULL)
676 		ifa_ref(ifa);
677 
678 	if (ifa == NULL) {
679 		if (error == ELOOP) {
680 			CARP_DEBUG("dropping looped packet on interface %s\n",
681 			    ifp->if_xname);
682 			CARPSTATS_INC(carps_badif);	/* ??? */
683 		} else {
684 			CARPSTATS_INC(carps_badvhid);
685 		}
686 		m_freem(m);
687 		return;
688 	}
689 
690 	/* verify the CARP version. */
691 	if (ch->carp_version != CARP_VERSION) {
692 		CARPSTATS_INC(carps_badver);
693 		CARP_DEBUG("%s: invalid version %d\n", ifp->if_xname,
694 		    ch->carp_version);
695 		ifa_free(ifa);
696 		m_freem(m);
697 		return;
698 	}
699 
700 	sc = ifa->ifa_carp;
701 	CARP_LOCK(sc);
702 	ifa_free(ifa);
703 
704 	if (carp_hmac_verify(sc, ch->carp_counter, ch->carp_md)) {
705 		CARPSTATS_INC(carps_badauth);
706 		CARP_DEBUG("%s: incorrect hash for VHID %u@%s\n", __func__,
707 		    sc->sc_vhid, ifp->if_xname);
708 		goto out;
709 	}
710 
711 	tmp_counter = ntohl(ch->carp_counter[0]);
712 	tmp_counter = tmp_counter<<32;
713 	tmp_counter += ntohl(ch->carp_counter[1]);
714 
715 	/* XXX Replay protection goes here */
716 
717 	sc->sc_init_counter = 0;
718 	sc->sc_counter = tmp_counter;
719 
720 	sc_tv.tv_sec = sc->sc_advbase;
721 	sc_tv.tv_usec = DEMOTE_ADVSKEW(sc) * 1000000 / 256;
722 	ch_tv.tv_sec = ch->carp_advbase;
723 	ch_tv.tv_usec = ch->carp_advskew * 1000000 / 256;
724 
725 	switch (sc->sc_state) {
726 	case INIT:
727 		break;
728 	case MASTER:
729 		/*
730 		 * If we receive an advertisement from a master who's going to
731 		 * be more frequent than us, go into BACKUP state.
732 		 */
733 		if (timevalcmp(&sc_tv, &ch_tv, >) ||
734 		    timevalcmp(&sc_tv, &ch_tv, ==)) {
735 			callout_stop(&sc->sc_ad_tmo);
736 			carp_set_state(sc, BACKUP,
737 			    "more frequent advertisement received");
738 			carp_setrun(sc, 0);
739 			carp_delroute(sc);
740 		}
741 		break;
742 	case BACKUP:
743 		/*
744 		 * If we're pre-empting masters who advertise slower than us,
745 		 * and this one claims to be slower, treat him as down.
746 		 */
747 		if (V_carp_preempt && timevalcmp(&sc_tv, &ch_tv, <)) {
748 			carp_master_down_locked(sc,
749 			    "preempting a slower master");
750 			break;
751 		}
752 
753 		/*
754 		 *  If the master is going to advertise at such a low frequency
755 		 *  that he's guaranteed to time out, we'd might as well just
756 		 *  treat him as timed out now.
757 		 */
758 		sc_tv.tv_sec = sc->sc_advbase * 3;
759 		if (timevalcmp(&sc_tv, &ch_tv, <)) {
760 			carp_master_down_locked(sc, "master will time out");
761 			break;
762 		}
763 
764 		/*
765 		 * Otherwise, we reset the counter and wait for the next
766 		 * advertisement.
767 		 */
768 		carp_setrun(sc, af);
769 		break;
770 	}
771 
772 out:
773 	CARP_UNLOCK(sc);
774 	m_freem(m);
775 }
776 
777 static int
778 carp_prepare_ad(struct mbuf *m, struct carp_softc *sc, struct carp_header *ch)
779 {
780 	struct m_tag *mtag;
781 
782 	if (sc->sc_init_counter) {
783 		/* this could also be seconds since unix epoch */
784 		sc->sc_counter = arc4random();
785 		sc->sc_counter = sc->sc_counter << 32;
786 		sc->sc_counter += arc4random();
787 	} else
788 		sc->sc_counter++;
789 
790 	ch->carp_counter[0] = htonl((sc->sc_counter>>32)&0xffffffff);
791 	ch->carp_counter[1] = htonl(sc->sc_counter&0xffffffff);
792 
793 	carp_hmac_generate(sc, ch->carp_counter, ch->carp_md);
794 
795 	/* Tag packet for carp_output */
796 	if ((mtag = m_tag_get(PACKET_TAG_CARP, sizeof(struct carp_softc *),
797 	    M_NOWAIT)) == NULL) {
798 		m_freem(m);
799 		CARPSTATS_INC(carps_onomem);
800 		return (ENOMEM);
801 	}
802 	bcopy(&sc, mtag + 1, sizeof(sc));
803 	m_tag_prepend(m, mtag);
804 
805 	return (0);
806 }
807 
808 /*
809  * To avoid LORs and possible recursions this function shouldn't
810  * be called directly, but scheduled via taskqueue.
811  */
812 static void
813 carp_send_ad_all(void *ctx __unused, int pending __unused)
814 {
815 	struct carp_softc *sc;
816 	struct epoch_tracker et;
817 
818 	NET_EPOCH_ENTER(et);
819 	mtx_lock(&carp_mtx);
820 	LIST_FOREACH(sc, &carp_list, sc_next)
821 		if (sc->sc_state == MASTER) {
822 			CARP_LOCK(sc);
823 			CURVNET_SET(sc->sc_carpdev->if_vnet);
824 			carp_send_ad_locked(sc);
825 			CURVNET_RESTORE();
826 			CARP_UNLOCK(sc);
827 		}
828 	mtx_unlock(&carp_mtx);
829 	NET_EPOCH_EXIT(et);
830 }
831 
832 /* Send a periodic advertisement, executed in callout context. */
833 static void
834 carp_send_ad(void *v)
835 {
836 	struct carp_softc *sc = v;
837 	struct epoch_tracker et;
838 
839 	NET_EPOCH_ENTER(et);
840 	CARP_LOCK_ASSERT(sc);
841 	CURVNET_SET(sc->sc_carpdev->if_vnet);
842 	carp_send_ad_locked(sc);
843 	CURVNET_RESTORE();
844 	CARP_UNLOCK(sc);
845 	NET_EPOCH_EXIT(et);
846 }
847 
848 static void
849 carp_send_ad_error(struct carp_softc *sc, int error)
850 {
851 
852 	/*
853 	 * We track errors and successfull sends with this logic:
854 	 * - Any error resets success counter to 0.
855 	 * - MAX_ERRORS triggers demotion.
856 	 * - MIN_SUCCESS successes resets error counter to 0.
857 	 * - MIN_SUCCESS reverts demotion, if it was triggered before.
858 	 */
859 	if (error) {
860 		if (sc->sc_sendad_errors < INT_MAX)
861 			sc->sc_sendad_errors++;
862 		if (sc->sc_sendad_errors == CARP_SENDAD_MAX_ERRORS) {
863 			static const char fmt[] = "send error %d on %s";
864 			char msg[sizeof(fmt) + IFNAMSIZ];
865 
866 			sprintf(msg, fmt, error, sc->sc_carpdev->if_xname);
867 			carp_demote_adj(V_carp_senderr_adj, msg);
868 		}
869 		sc->sc_sendad_success = 0;
870 	} else if (sc->sc_sendad_errors > 0) {
871 		if (++sc->sc_sendad_success >= CARP_SENDAD_MIN_SUCCESS) {
872 			if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) {
873 				static const char fmt[] = "send ok on %s";
874 				char msg[sizeof(fmt) + IFNAMSIZ];
875 
876 				sprintf(msg, fmt, sc->sc_carpdev->if_xname);
877 				carp_demote_adj(-V_carp_senderr_adj, msg);
878 			}
879 			sc->sc_sendad_errors = 0;
880 		}
881 	}
882 }
883 
884 /*
885  * Pick the best ifaddr on the given ifp for sending CARP
886  * advertisements.
887  *
888  * "Best" here is defined by ifa_preferred().  This function is much
889  * much like ifaof_ifpforaddr() except that we just use ifa_preferred().
890  *
891  * (This could be simplified to return the actual address, except that
892  * it has a different format in AF_INET and AF_INET6.)
893  */
894 static struct ifaddr *
895 carp_best_ifa(int af, struct ifnet *ifp)
896 {
897 	struct ifaddr *ifa, *best;
898 
899 	NET_EPOCH_ASSERT();
900 
901 	if (af >= AF_MAX)
902 		return (NULL);
903 	best = NULL;
904 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
905 		if (ifa->ifa_addr->sa_family == af &&
906 		    (best == NULL || ifa_preferred(best, ifa)))
907 			best = ifa;
908 	}
909 	if (best != NULL)
910 		ifa_ref(best);
911 	return (best);
912 }
913 
914 static void
915 carp_send_ad_locked(struct carp_softc *sc)
916 {
917 	struct carp_header ch;
918 	struct timeval tv;
919 	struct ifaddr *ifa;
920 	struct carp_header *ch_ptr;
921 	struct mbuf *m;
922 	int len, advskew;
923 
924 	NET_EPOCH_ASSERT();
925 	CARP_LOCK_ASSERT(sc);
926 
927 	advskew = DEMOTE_ADVSKEW(sc);
928 	tv.tv_sec = sc->sc_advbase;
929 	tv.tv_usec = advskew * 1000000 / 256;
930 
931 	ch.carp_version = CARP_VERSION;
932 	ch.carp_type = CARP_ADVERTISEMENT;
933 	ch.carp_vhid = sc->sc_vhid;
934 	ch.carp_advbase = sc->sc_advbase;
935 	ch.carp_advskew = advskew;
936 	ch.carp_authlen = 7;	/* XXX DEFINE */
937 	ch.carp_pad1 = 0;	/* must be zero */
938 	ch.carp_cksum = 0;
939 
940 	/* XXXGL: OpenBSD picks first ifaddr with needed family. */
941 
942 #ifdef INET
943 	if (sc->sc_naddrs) {
944 		struct ip *ip;
945 
946 		m = m_gethdr(M_NOWAIT, MT_DATA);
947 		if (m == NULL) {
948 			CARPSTATS_INC(carps_onomem);
949 			goto resched;
950 		}
951 		len = sizeof(*ip) + sizeof(ch);
952 		m->m_pkthdr.len = len;
953 		m->m_pkthdr.rcvif = NULL;
954 		m->m_len = len;
955 		M_ALIGN(m, m->m_len);
956 		m->m_flags |= M_MCAST;
957 		ip = mtod(m, struct ip *);
958 		ip->ip_v = IPVERSION;
959 		ip->ip_hl = sizeof(*ip) >> 2;
960 		ip->ip_tos = V_carp_dscp << IPTOS_DSCP_OFFSET;
961 		ip->ip_len = htons(len);
962 		ip->ip_off = htons(IP_DF);
963 		ip->ip_ttl = CARP_DFLTTL;
964 		ip->ip_p = IPPROTO_CARP;
965 		ip->ip_sum = 0;
966 		ip_fillid(ip);
967 
968 		ifa = carp_best_ifa(AF_INET, sc->sc_carpdev);
969 		if (ifa != NULL) {
970 			ip->ip_src.s_addr =
971 			    ifatoia(ifa)->ia_addr.sin_addr.s_addr;
972 			ifa_free(ifa);
973 		} else
974 			ip->ip_src.s_addr = 0;
975 		ip->ip_dst.s_addr = htonl(INADDR_CARP_GROUP);
976 
977 		ch_ptr = (struct carp_header *)(&ip[1]);
978 		bcopy(&ch, ch_ptr, sizeof(ch));
979 		if (carp_prepare_ad(m, sc, ch_ptr))
980 			goto resched;
981 
982 		m->m_data += sizeof(*ip);
983 		ch_ptr->carp_cksum = in_cksum(m, len - sizeof(*ip));
984 		m->m_data -= sizeof(*ip);
985 
986 		CARPSTATS_INC(carps_opackets);
987 
988 		carp_send_ad_error(sc, ip_output(m, NULL, NULL, IP_RAWOUTPUT,
989 		    &sc->sc_carpdev->if_carp->cif_imo, NULL));
990 	}
991 #endif /* INET */
992 #ifdef INET6
993 	if (sc->sc_naddrs6) {
994 		struct ip6_hdr *ip6;
995 
996 		m = m_gethdr(M_NOWAIT, MT_DATA);
997 		if (m == NULL) {
998 			CARPSTATS_INC(carps_onomem);
999 			goto resched;
1000 		}
1001 		len = sizeof(*ip6) + sizeof(ch);
1002 		m->m_pkthdr.len = len;
1003 		m->m_pkthdr.rcvif = NULL;
1004 		m->m_len = len;
1005 		M_ALIGN(m, m->m_len);
1006 		m->m_flags |= M_MCAST;
1007 		ip6 = mtod(m, struct ip6_hdr *);
1008 		bzero(ip6, sizeof(*ip6));
1009 		ip6->ip6_vfc |= IPV6_VERSION;
1010 		/* Traffic class isn't defined in ip6 struct instead
1011 		 * it gets offset into flowid field */
1012 		ip6->ip6_flow |= htonl(V_carp_dscp << (IPV6_FLOWLABEL_LEN +
1013 		    IPTOS_DSCP_OFFSET));
1014 		ip6->ip6_hlim = CARP_DFLTTL;
1015 		ip6->ip6_nxt = IPPROTO_CARP;
1016 
1017 		/* set the source address */
1018 		ifa = carp_best_ifa(AF_INET6, sc->sc_carpdev);
1019 		if (ifa != NULL) {
1020 			bcopy(IFA_IN6(ifa), &ip6->ip6_src,
1021 			    sizeof(struct in6_addr));
1022 			ifa_free(ifa);
1023 		} else
1024 			/* This should never happen with IPv6. */
1025 			bzero(&ip6->ip6_src, sizeof(struct in6_addr));
1026 
1027 		/* Set the multicast destination. */
1028 		ip6->ip6_dst.s6_addr16[0] = htons(0xff02);
1029 		ip6->ip6_dst.s6_addr8[15] = 0x12;
1030 		if (in6_setscope(&ip6->ip6_dst, sc->sc_carpdev, NULL) != 0) {
1031 			m_freem(m);
1032 			CARP_DEBUG("%s: in6_setscope failed\n", __func__);
1033 			goto resched;
1034 		}
1035 
1036 		ch_ptr = (struct carp_header *)(&ip6[1]);
1037 		bcopy(&ch, ch_ptr, sizeof(ch));
1038 		if (carp_prepare_ad(m, sc, ch_ptr))
1039 			goto resched;
1040 
1041 		m->m_data += sizeof(*ip6);
1042 		ch_ptr->carp_cksum = in_cksum(m, len - sizeof(*ip6));
1043 		m->m_data -= sizeof(*ip6);
1044 
1045 		CARPSTATS_INC(carps_opackets6);
1046 
1047 		carp_send_ad_error(sc, ip6_output(m, NULL, NULL, 0,
1048 		    &sc->sc_carpdev->if_carp->cif_im6o, NULL, NULL));
1049 	}
1050 #endif /* INET6 */
1051 
1052 resched:
1053 	callout_reset(&sc->sc_ad_tmo, tvtohz(&tv), carp_send_ad, sc);
1054 }
1055 
1056 static void
1057 carp_addroute(struct carp_softc *sc)
1058 {
1059 	struct ifaddr *ifa;
1060 
1061 	CARP_FOREACH_IFA(sc, ifa)
1062 		carp_ifa_addroute(ifa);
1063 }
1064 
1065 static void
1066 carp_ifa_addroute(struct ifaddr *ifa)
1067 {
1068 
1069 	switch (ifa->ifa_addr->sa_family) {
1070 #ifdef INET
1071 	case AF_INET:
1072 		in_addprefix(ifatoia(ifa));
1073 		ifa_add_loopback_route(ifa,
1074 		    (struct sockaddr *)&ifatoia(ifa)->ia_addr);
1075 		break;
1076 #endif
1077 #ifdef INET6
1078 	case AF_INET6:
1079 		ifa_add_loopback_route(ifa,
1080 		    (struct sockaddr *)&ifatoia6(ifa)->ia_addr);
1081 		nd6_add_ifa_lle(ifatoia6(ifa));
1082 		break;
1083 #endif
1084 	}
1085 }
1086 
1087 static void
1088 carp_delroute(struct carp_softc *sc)
1089 {
1090 	struct ifaddr *ifa;
1091 
1092 	CARP_FOREACH_IFA(sc, ifa)
1093 		carp_ifa_delroute(ifa);
1094 }
1095 
1096 static void
1097 carp_ifa_delroute(struct ifaddr *ifa)
1098 {
1099 
1100 	switch (ifa->ifa_addr->sa_family) {
1101 #ifdef INET
1102 	case AF_INET:
1103 		ifa_del_loopback_route(ifa,
1104 		    (struct sockaddr *)&ifatoia(ifa)->ia_addr);
1105 		in_scrubprefix(ifatoia(ifa), LLE_STATIC);
1106 		break;
1107 #endif
1108 #ifdef INET6
1109 	case AF_INET6:
1110 		ifa_del_loopback_route(ifa,
1111 		    (struct sockaddr *)&ifatoia6(ifa)->ia_addr);
1112 		nd6_rem_ifa_lle(ifatoia6(ifa), 1);
1113 		break;
1114 #endif
1115 	}
1116 }
1117 
1118 int
1119 carp_master(struct ifaddr *ifa)
1120 {
1121 	struct carp_softc *sc = ifa->ifa_carp;
1122 
1123 	return (sc->sc_state == MASTER);
1124 }
1125 
1126 #ifdef INET
1127 /*
1128  * Broadcast a gratuitous ARP request containing
1129  * the virtual router MAC address for each IP address
1130  * associated with the virtual router.
1131  */
1132 static void
1133 carp_send_arp(struct carp_softc *sc)
1134 {
1135 	struct ifaddr *ifa;
1136 	struct in_addr addr;
1137 
1138 	NET_EPOCH_ASSERT();
1139 
1140 	CARP_FOREACH_IFA(sc, ifa) {
1141 		if (ifa->ifa_addr->sa_family != AF_INET)
1142 			continue;
1143 		addr = ((struct sockaddr_in *)ifa->ifa_addr)->sin_addr;
1144 		arp_announce_ifaddr(sc->sc_carpdev, addr, LLADDR(&sc->sc_addr));
1145 	}
1146 }
1147 
1148 int
1149 carp_iamatch(struct ifaddr *ifa, uint8_t **enaddr)
1150 {
1151 	struct carp_softc *sc = ifa->ifa_carp;
1152 
1153 	if (sc->sc_state == MASTER) {
1154 		*enaddr = LLADDR(&sc->sc_addr);
1155 		return (1);
1156 	}
1157 
1158 	return (0);
1159 }
1160 #endif
1161 
1162 #ifdef INET6
1163 static void
1164 carp_send_na(struct carp_softc *sc)
1165 {
1166 	static struct in6_addr mcast = IN6ADDR_LINKLOCAL_ALLNODES_INIT;
1167 	struct ifaddr *ifa;
1168 	struct in6_addr *in6;
1169 
1170 	CARP_FOREACH_IFA(sc, ifa) {
1171 		if (ifa->ifa_addr->sa_family != AF_INET6)
1172 			continue;
1173 
1174 		in6 = IFA_IN6(ifa);
1175 		nd6_na_output(sc->sc_carpdev, &mcast, in6,
1176 		    ND_NA_FLAG_OVERRIDE, 1, NULL);
1177 		DELAY(1000);	/* XXX */
1178 	}
1179 }
1180 
1181 /*
1182  * Returns ifa in case it's a carp address and it is MASTER, or if the address
1183  * matches and is not a carp address.  Returns NULL otherwise.
1184  */
1185 struct ifaddr *
1186 carp_iamatch6(struct ifnet *ifp, struct in6_addr *taddr)
1187 {
1188 	struct ifaddr *ifa;
1189 
1190 	NET_EPOCH_ASSERT();
1191 
1192 	ifa = NULL;
1193 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1194 		if (ifa->ifa_addr->sa_family != AF_INET6)
1195 			continue;
1196 		if (!IN6_ARE_ADDR_EQUAL(taddr, IFA_IN6(ifa)))
1197 			continue;
1198 		if (ifa->ifa_carp && ifa->ifa_carp->sc_state != MASTER)
1199 			ifa = NULL;
1200 		else
1201 			ifa_ref(ifa);
1202 		break;
1203 	}
1204 
1205 	return (ifa);
1206 }
1207 
1208 char *
1209 carp_macmatch6(struct ifnet *ifp, struct mbuf *m, const struct in6_addr *taddr)
1210 {
1211 	struct ifaddr *ifa;
1212 
1213 	NET_EPOCH_ASSERT();
1214 
1215 	IFNET_FOREACH_IFA(ifp, ifa)
1216 		if (ifa->ifa_addr->sa_family == AF_INET6 &&
1217 		    IN6_ARE_ADDR_EQUAL(taddr, IFA_IN6(ifa))) {
1218 			struct carp_softc *sc = ifa->ifa_carp;
1219 			struct m_tag *mtag;
1220 
1221 			mtag = m_tag_get(PACKET_TAG_CARP,
1222 			    sizeof(struct carp_softc *), M_NOWAIT);
1223 			if (mtag == NULL)
1224 				/* Better a bit than nothing. */
1225 				return (LLADDR(&sc->sc_addr));
1226 
1227 			bcopy(&sc, mtag + 1, sizeof(sc));
1228 			m_tag_prepend(m, mtag);
1229 
1230 			return (LLADDR(&sc->sc_addr));
1231 		}
1232 
1233 	return (NULL);
1234 }
1235 #endif /* INET6 */
1236 
1237 int
1238 carp_forus(struct ifnet *ifp, u_char *dhost)
1239 {
1240 	struct carp_softc *sc;
1241 	uint8_t *ena = dhost;
1242 
1243 	if (ena[0] || ena[1] || ena[2] != 0x5e || ena[3] || ena[4] != 1)
1244 		return (0);
1245 
1246 	CIF_LOCK(ifp->if_carp);
1247 	IFNET_FOREACH_CARP(ifp, sc) {
1248 		/*
1249 		 * CARP_LOCK() is not here, since would protect nothing, but
1250 		 * cause deadlock with if_bridge, calling this under its lock.
1251 		 */
1252 		if (sc->sc_state == MASTER && !bcmp(dhost, LLADDR(&sc->sc_addr),
1253 		    ETHER_ADDR_LEN)) {
1254 			CIF_UNLOCK(ifp->if_carp);
1255 			return (1);
1256 		}
1257 	}
1258 	CIF_UNLOCK(ifp->if_carp);
1259 
1260 	return (0);
1261 }
1262 
1263 /* Master down timeout event, executed in callout context. */
1264 static void
1265 carp_master_down(void *v)
1266 {
1267 	struct carp_softc *sc = v;
1268 	struct epoch_tracker et;
1269 
1270 	NET_EPOCH_ENTER(et);
1271 	CARP_LOCK_ASSERT(sc);
1272 
1273 	CURVNET_SET(sc->sc_carpdev->if_vnet);
1274 	if (sc->sc_state == BACKUP) {
1275 		carp_master_down_locked(sc, "master timed out");
1276 	}
1277 	CURVNET_RESTORE();
1278 
1279 	CARP_UNLOCK(sc);
1280 	NET_EPOCH_EXIT(et);
1281 }
1282 
1283 static void
1284 carp_master_down_locked(struct carp_softc *sc, const char *reason)
1285 {
1286 
1287 	NET_EPOCH_ASSERT();
1288 	CARP_LOCK_ASSERT(sc);
1289 
1290 	switch (sc->sc_state) {
1291 	case BACKUP:
1292 		carp_set_state(sc, MASTER, reason);
1293 		carp_send_ad_locked(sc);
1294 #ifdef INET
1295 		carp_send_arp(sc);
1296 #endif
1297 #ifdef INET6
1298 		carp_send_na(sc);
1299 #endif
1300 		carp_setrun(sc, 0);
1301 		carp_addroute(sc);
1302 		break;
1303 	case INIT:
1304 	case MASTER:
1305 #ifdef INVARIANTS
1306 		panic("carp: VHID %u@%s: master_down event in %s state\n",
1307 		    sc->sc_vhid,
1308 		    sc->sc_carpdev->if_xname,
1309 		    sc->sc_state ? "MASTER" : "INIT");
1310 #endif
1311 		break;
1312 	}
1313 }
1314 
1315 /*
1316  * When in backup state, af indicates whether to reset the master down timer
1317  * for v4 or v6. If it's set to zero, reset the ones which are already pending.
1318  */
1319 static void
1320 carp_setrun(struct carp_softc *sc, sa_family_t af)
1321 {
1322 	struct timeval tv;
1323 
1324 	CARP_LOCK_ASSERT(sc);
1325 
1326 	if ((sc->sc_carpdev->if_flags & IFF_UP) == 0 ||
1327 	    sc->sc_carpdev->if_link_state != LINK_STATE_UP ||
1328 	    (sc->sc_naddrs == 0 && sc->sc_naddrs6 == 0) ||
1329 	    !V_carp_allow)
1330 		return;
1331 
1332 	switch (sc->sc_state) {
1333 	case INIT:
1334 		carp_set_state(sc, BACKUP, "initialization complete");
1335 		carp_setrun(sc, 0);
1336 		break;
1337 	case BACKUP:
1338 		callout_stop(&sc->sc_ad_tmo);
1339 		tv.tv_sec = 3 * sc->sc_advbase;
1340 		tv.tv_usec = sc->sc_advskew * 1000000 / 256;
1341 		switch (af) {
1342 #ifdef INET
1343 		case AF_INET:
1344 			callout_reset(&sc->sc_md_tmo, tvtohz(&tv),
1345 			    carp_master_down, sc);
1346 			break;
1347 #endif
1348 #ifdef INET6
1349 		case AF_INET6:
1350 			callout_reset(&sc->sc_md6_tmo, tvtohz(&tv),
1351 			    carp_master_down, sc);
1352 			break;
1353 #endif
1354 		default:
1355 #ifdef INET
1356 			if (sc->sc_naddrs)
1357 				callout_reset(&sc->sc_md_tmo, tvtohz(&tv),
1358 				    carp_master_down, sc);
1359 #endif
1360 #ifdef INET6
1361 			if (sc->sc_naddrs6)
1362 				callout_reset(&sc->sc_md6_tmo, tvtohz(&tv),
1363 				    carp_master_down, sc);
1364 #endif
1365 			break;
1366 		}
1367 		break;
1368 	case MASTER:
1369 		tv.tv_sec = sc->sc_advbase;
1370 		tv.tv_usec = sc->sc_advskew * 1000000 / 256;
1371 		callout_reset(&sc->sc_ad_tmo, tvtohz(&tv),
1372 		    carp_send_ad, sc);
1373 		break;
1374 	}
1375 }
1376 
1377 /*
1378  * Setup multicast structures.
1379  */
1380 static int
1381 carp_multicast_setup(struct carp_if *cif, sa_family_t sa)
1382 {
1383 	struct ifnet *ifp = cif->cif_ifp;
1384 	int error = 0;
1385 
1386 	switch (sa) {
1387 #ifdef INET
1388 	case AF_INET:
1389 	    {
1390 		struct ip_moptions *imo = &cif->cif_imo;
1391 		struct in_mfilter *imf;
1392 		struct in_addr addr;
1393 
1394 		if (ip_mfilter_first(&imo->imo_head) != NULL)
1395 			return (0);
1396 
1397 		imf = ip_mfilter_alloc(M_WAITOK, 0, 0);
1398 		ip_mfilter_init(&imo->imo_head);
1399 		imo->imo_multicast_vif = -1;
1400 
1401 		addr.s_addr = htonl(INADDR_CARP_GROUP);
1402 		if ((error = in_joingroup(ifp, &addr, NULL,
1403 		    &imf->imf_inm)) != 0) {
1404 			ip_mfilter_free(imf);
1405 			break;
1406 		}
1407 
1408 		ip_mfilter_insert(&imo->imo_head, imf);
1409 		imo->imo_multicast_ifp = ifp;
1410 		imo->imo_multicast_ttl = CARP_DFLTTL;
1411 		imo->imo_multicast_loop = 0;
1412 		break;
1413 	   }
1414 #endif
1415 #ifdef INET6
1416 	case AF_INET6:
1417 	    {
1418 		struct ip6_moptions *im6o = &cif->cif_im6o;
1419 		struct in6_mfilter *im6f[2];
1420 		struct in6_addr in6;
1421 
1422 		if (ip6_mfilter_first(&im6o->im6o_head))
1423 			return (0);
1424 
1425 		im6f[0] = ip6_mfilter_alloc(M_WAITOK, 0, 0);
1426 		im6f[1] = ip6_mfilter_alloc(M_WAITOK, 0, 0);
1427 
1428 		ip6_mfilter_init(&im6o->im6o_head);
1429 		im6o->im6o_multicast_hlim = CARP_DFLTTL;
1430 		im6o->im6o_multicast_ifp = ifp;
1431 
1432 		/* Join IPv6 CARP multicast group. */
1433 		bzero(&in6, sizeof(in6));
1434 		in6.s6_addr16[0] = htons(0xff02);
1435 		in6.s6_addr8[15] = 0x12;
1436 		if ((error = in6_setscope(&in6, ifp, NULL)) != 0) {
1437 			ip6_mfilter_free(im6f[0]);
1438 			ip6_mfilter_free(im6f[1]);
1439 			break;
1440 		}
1441 		if ((error = in6_joingroup(ifp, &in6, NULL, &im6f[0]->im6f_in6m, 0)) != 0) {
1442 			ip6_mfilter_free(im6f[0]);
1443 			ip6_mfilter_free(im6f[1]);
1444 			break;
1445 		}
1446 
1447 		/* Join solicited multicast address. */
1448 		bzero(&in6, sizeof(in6));
1449 		in6.s6_addr16[0] = htons(0xff02);
1450 		in6.s6_addr32[1] = 0;
1451 		in6.s6_addr32[2] = htonl(1);
1452 		in6.s6_addr32[3] = 0;
1453 		in6.s6_addr8[12] = 0xff;
1454 
1455 		if ((error = in6_setscope(&in6, ifp, NULL)) != 0) {
1456 			ip6_mfilter_free(im6f[0]);
1457 			ip6_mfilter_free(im6f[1]);
1458 			break;
1459 		}
1460 
1461 		if ((error = in6_joingroup(ifp, &in6, NULL, &im6f[1]->im6f_in6m, 0)) != 0) {
1462 			in6_leavegroup(im6f[0]->im6f_in6m, NULL);
1463 			ip6_mfilter_free(im6f[0]);
1464 			ip6_mfilter_free(im6f[1]);
1465 			break;
1466 		}
1467 		ip6_mfilter_insert(&im6o->im6o_head, im6f[0]);
1468 		ip6_mfilter_insert(&im6o->im6o_head, im6f[1]);
1469 		break;
1470 	    }
1471 #endif
1472 	}
1473 
1474 	return (error);
1475 }
1476 
1477 /*
1478  * Free multicast structures.
1479  */
1480 static void
1481 carp_multicast_cleanup(struct carp_if *cif, sa_family_t sa)
1482 {
1483 #ifdef INET
1484 	struct ip_moptions *imo = &cif->cif_imo;
1485 	struct in_mfilter *imf;
1486 #endif
1487 #ifdef INET6
1488 	struct ip6_moptions *im6o = &cif->cif_im6o;
1489 	struct in6_mfilter *im6f;
1490 #endif
1491 	sx_assert(&carp_sx, SA_XLOCKED);
1492 
1493 	switch (sa) {
1494 #ifdef INET
1495 	case AF_INET:
1496 		if (cif->cif_naddrs != 0)
1497 			break;
1498 
1499 		while ((imf = ip_mfilter_first(&imo->imo_head)) != NULL) {
1500 			ip_mfilter_remove(&imo->imo_head, imf);
1501 			in_leavegroup(imf->imf_inm, NULL);
1502 			ip_mfilter_free(imf);
1503 		}
1504 		break;
1505 #endif
1506 #ifdef INET6
1507 	case AF_INET6:
1508 		if (cif->cif_naddrs6 != 0)
1509 			break;
1510 
1511 		while ((im6f = ip6_mfilter_first(&im6o->im6o_head)) != NULL) {
1512 			ip6_mfilter_remove(&im6o->im6o_head, im6f);
1513 			in6_leavegroup(im6f->im6f_in6m, NULL);
1514 			ip6_mfilter_free(im6f);
1515 		}
1516 		break;
1517 #endif
1518 	}
1519 }
1520 
1521 int
1522 carp_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *sa)
1523 {
1524 	struct m_tag *mtag;
1525 	struct carp_softc *sc;
1526 
1527 	if (!sa)
1528 		return (0);
1529 
1530 	switch (sa->sa_family) {
1531 #ifdef INET
1532 	case AF_INET:
1533 		break;
1534 #endif
1535 #ifdef INET6
1536 	case AF_INET6:
1537 		break;
1538 #endif
1539 	default:
1540 		return (0);
1541 	}
1542 
1543 	mtag = m_tag_find(m, PACKET_TAG_CARP, NULL);
1544 	if (mtag == NULL)
1545 		return (0);
1546 
1547 	bcopy(mtag + 1, &sc, sizeof(sc));
1548 
1549 	/* Set the source MAC address to the Virtual Router MAC Address. */
1550 	switch (ifp->if_type) {
1551 	case IFT_ETHER:
1552 	case IFT_BRIDGE:
1553 	case IFT_L2VLAN: {
1554 			struct ether_header *eh;
1555 
1556 			eh = mtod(m, struct ether_header *);
1557 			eh->ether_shost[0] = 0;
1558 			eh->ether_shost[1] = 0;
1559 			eh->ether_shost[2] = 0x5e;
1560 			eh->ether_shost[3] = 0;
1561 			eh->ether_shost[4] = 1;
1562 			eh->ether_shost[5] = sc->sc_vhid;
1563 		}
1564 		break;
1565 	default:
1566 		printf("%s: carp is not supported for the %d interface type\n",
1567 		    ifp->if_xname, ifp->if_type);
1568 		return (EOPNOTSUPP);
1569 	}
1570 
1571 	return (0);
1572 }
1573 
1574 static struct carp_softc*
1575 carp_alloc(struct ifnet *ifp)
1576 {
1577 	struct carp_softc *sc;
1578 	struct carp_if *cif;
1579 
1580 	sx_assert(&carp_sx, SA_XLOCKED);
1581 
1582 	if ((cif = ifp->if_carp) == NULL)
1583 		cif = carp_alloc_if(ifp);
1584 
1585 	sc = malloc(sizeof(*sc), M_CARP, M_WAITOK|M_ZERO);
1586 
1587 	sc->sc_advbase = CARP_DFLTINTV;
1588 	sc->sc_vhid = -1;	/* required setting */
1589 	sc->sc_init_counter = 1;
1590 	sc->sc_state = INIT;
1591 
1592 	sc->sc_ifasiz = sizeof(struct ifaddr *);
1593 	sc->sc_ifas = malloc(sc->sc_ifasiz, M_CARP, M_WAITOK|M_ZERO);
1594 	sc->sc_carpdev = ifp;
1595 
1596 	CARP_LOCK_INIT(sc);
1597 #ifdef INET
1598 	callout_init_mtx(&sc->sc_md_tmo, &sc->sc_mtx, CALLOUT_RETURNUNLOCKED);
1599 #endif
1600 #ifdef INET6
1601 	callout_init_mtx(&sc->sc_md6_tmo, &sc->sc_mtx, CALLOUT_RETURNUNLOCKED);
1602 #endif
1603 	callout_init_mtx(&sc->sc_ad_tmo, &sc->sc_mtx, CALLOUT_RETURNUNLOCKED);
1604 
1605 	CIF_LOCK(cif);
1606 	TAILQ_INSERT_TAIL(&cif->cif_vrs, sc, sc_list);
1607 	CIF_UNLOCK(cif);
1608 
1609 	mtx_lock(&carp_mtx);
1610 	LIST_INSERT_HEAD(&carp_list, sc, sc_next);
1611 	mtx_unlock(&carp_mtx);
1612 
1613 	return (sc);
1614 }
1615 
1616 static void
1617 carp_grow_ifas(struct carp_softc *sc)
1618 {
1619 	struct ifaddr **new;
1620 
1621 	new = malloc(sc->sc_ifasiz * 2, M_CARP, M_WAITOK | M_ZERO);
1622 	CARP_LOCK(sc);
1623 	bcopy(sc->sc_ifas, new, sc->sc_ifasiz);
1624 	free(sc->sc_ifas, M_CARP);
1625 	sc->sc_ifas = new;
1626 	sc->sc_ifasiz *= 2;
1627 	CARP_UNLOCK(sc);
1628 }
1629 
1630 static void
1631 carp_destroy(struct carp_softc *sc)
1632 {
1633 	struct ifnet *ifp = sc->sc_carpdev;
1634 	struct carp_if *cif = ifp->if_carp;
1635 
1636 	sx_assert(&carp_sx, SA_XLOCKED);
1637 
1638 	if (sc->sc_suppress)
1639 		carp_demote_adj(-V_carp_ifdown_adj, "vhid removed");
1640 	CARP_UNLOCK(sc);
1641 
1642 	CIF_LOCK(cif);
1643 	TAILQ_REMOVE(&cif->cif_vrs, sc, sc_list);
1644 	CIF_UNLOCK(cif);
1645 
1646 	mtx_lock(&carp_mtx);
1647 	LIST_REMOVE(sc, sc_next);
1648 	mtx_unlock(&carp_mtx);
1649 
1650 	callout_drain(&sc->sc_ad_tmo);
1651 #ifdef INET
1652 	callout_drain(&sc->sc_md_tmo);
1653 #endif
1654 #ifdef INET6
1655 	callout_drain(&sc->sc_md6_tmo);
1656 #endif
1657 	CARP_LOCK_DESTROY(sc);
1658 
1659 	free(sc->sc_ifas, M_CARP);
1660 	free(sc, M_CARP);
1661 }
1662 
1663 static struct carp_if*
1664 carp_alloc_if(struct ifnet *ifp)
1665 {
1666 	struct carp_if *cif;
1667 	int error;
1668 
1669 	cif = malloc(sizeof(*cif), M_CARP, M_WAITOK|M_ZERO);
1670 
1671 	if ((error = ifpromisc(ifp, 1)) != 0)
1672 		printf("%s: ifpromisc(%s) failed: %d\n",
1673 		    __func__, ifp->if_xname, error);
1674 	else
1675 		cif->cif_flags |= CIF_PROMISC;
1676 
1677 	CIF_LOCK_INIT(cif);
1678 	cif->cif_ifp = ifp;
1679 	TAILQ_INIT(&cif->cif_vrs);
1680 
1681 	IF_ADDR_WLOCK(ifp);
1682 	ifp->if_carp = cif;
1683 	if_ref(ifp);
1684 	IF_ADDR_WUNLOCK(ifp);
1685 
1686 	return (cif);
1687 }
1688 
1689 static void
1690 carp_free_if(struct carp_if *cif)
1691 {
1692 	struct ifnet *ifp = cif->cif_ifp;
1693 
1694 	CIF_LOCK_ASSERT(cif);
1695 	KASSERT(TAILQ_EMPTY(&cif->cif_vrs), ("%s: softc list not empty",
1696 	    __func__));
1697 
1698 	IF_ADDR_WLOCK(ifp);
1699 	ifp->if_carp = NULL;
1700 	IF_ADDR_WUNLOCK(ifp);
1701 
1702 	CIF_LOCK_DESTROY(cif);
1703 
1704 	if (cif->cif_flags & CIF_PROMISC)
1705 		ifpromisc(ifp, 0);
1706 	if_rele(ifp);
1707 
1708 	free(cif, M_CARP);
1709 }
1710 
1711 static void
1712 carp_carprcp(struct carpreq *carpr, struct carp_softc *sc, int priv)
1713 {
1714 
1715 	CARP_LOCK(sc);
1716 	carpr->carpr_state = sc->sc_state;
1717 	carpr->carpr_vhid = sc->sc_vhid;
1718 	carpr->carpr_advbase = sc->sc_advbase;
1719 	carpr->carpr_advskew = sc->sc_advskew;
1720 	if (priv)
1721 		bcopy(sc->sc_key, carpr->carpr_key, sizeof(carpr->carpr_key));
1722 	else
1723 		bzero(carpr->carpr_key, sizeof(carpr->carpr_key));
1724 	CARP_UNLOCK(sc);
1725 }
1726 
1727 int
1728 carp_ioctl(struct ifreq *ifr, u_long cmd, struct thread *td)
1729 {
1730 	struct carpreq carpr;
1731 	struct ifnet *ifp;
1732 	struct carp_softc *sc = NULL;
1733 	int error = 0, locked = 0;
1734 
1735 	if ((error = copyin(ifr_data_get_ptr(ifr), &carpr, sizeof carpr)))
1736 		return (error);
1737 
1738 	ifp = ifunit_ref(ifr->ifr_name);
1739 	if (ifp == NULL)
1740 		return (ENXIO);
1741 
1742 	switch (ifp->if_type) {
1743 	case IFT_ETHER:
1744 	case IFT_L2VLAN:
1745 	case IFT_BRIDGE:
1746 		break;
1747 	default:
1748 		error = EOPNOTSUPP;
1749 		goto out;
1750 	}
1751 
1752 	if ((ifp->if_flags & IFF_MULTICAST) == 0) {
1753 		error = EADDRNOTAVAIL;
1754 		goto out;
1755 	}
1756 
1757 	sx_xlock(&carp_sx);
1758 	switch (cmd) {
1759 	case SIOCSVH:
1760 		if ((error = priv_check(td, PRIV_NETINET_CARP)))
1761 			break;
1762 		if (carpr.carpr_vhid <= 0 || carpr.carpr_vhid > CARP_MAXVHID ||
1763 		    carpr.carpr_advbase < 0 || carpr.carpr_advskew < 0) {
1764 			error = EINVAL;
1765 			break;
1766 		}
1767 
1768 		if (ifp->if_carp) {
1769 			IFNET_FOREACH_CARP(ifp, sc)
1770 				if (sc->sc_vhid == carpr.carpr_vhid)
1771 					break;
1772 		}
1773 		if (sc == NULL) {
1774 			sc = carp_alloc(ifp);
1775 			CARP_LOCK(sc);
1776 			sc->sc_vhid = carpr.carpr_vhid;
1777 			LLADDR(&sc->sc_addr)[0] = 0;
1778 			LLADDR(&sc->sc_addr)[1] = 0;
1779 			LLADDR(&sc->sc_addr)[2] = 0x5e;
1780 			LLADDR(&sc->sc_addr)[3] = 0;
1781 			LLADDR(&sc->sc_addr)[4] = 1;
1782 			LLADDR(&sc->sc_addr)[5] = sc->sc_vhid;
1783 		} else
1784 			CARP_LOCK(sc);
1785 		locked = 1;
1786 		if (carpr.carpr_advbase > 0) {
1787 			if (carpr.carpr_advbase > 255 ||
1788 			    carpr.carpr_advbase < CARP_DFLTINTV) {
1789 				error = EINVAL;
1790 				break;
1791 			}
1792 			sc->sc_advbase = carpr.carpr_advbase;
1793 		}
1794 		if (carpr.carpr_advskew >= 255) {
1795 			error = EINVAL;
1796 			break;
1797 		}
1798 		sc->sc_advskew = carpr.carpr_advskew;
1799 		if (carpr.carpr_key[0] != '\0') {
1800 			bcopy(carpr.carpr_key, sc->sc_key, sizeof(sc->sc_key));
1801 			carp_hmac_prepare(sc);
1802 		}
1803 		if (sc->sc_state != INIT &&
1804 		    carpr.carpr_state != sc->sc_state) {
1805 			switch (carpr.carpr_state) {
1806 			case BACKUP:
1807 				callout_stop(&sc->sc_ad_tmo);
1808 				carp_set_state(sc, BACKUP,
1809 				    "user requested via ifconfig");
1810 				carp_setrun(sc, 0);
1811 				carp_delroute(sc);
1812 				break;
1813 			case MASTER:
1814 				carp_master_down_locked(sc,
1815 				    "user requested via ifconfig");
1816 				break;
1817 			default:
1818 				break;
1819 			}
1820 		}
1821 		break;
1822 
1823 	case SIOCGVH:
1824 	    {
1825 		int priveleged;
1826 
1827 		if (carpr.carpr_vhid < 0 || carpr.carpr_vhid > CARP_MAXVHID) {
1828 			error = EINVAL;
1829 			break;
1830 		}
1831 		if (carpr.carpr_count < 1) {
1832 			error = EMSGSIZE;
1833 			break;
1834 		}
1835 		if (ifp->if_carp == NULL) {
1836 			error = ENOENT;
1837 			break;
1838 		}
1839 
1840 		priveleged = (priv_check(td, PRIV_NETINET_CARP) == 0);
1841 		if (carpr.carpr_vhid != 0) {
1842 			IFNET_FOREACH_CARP(ifp, sc)
1843 				if (sc->sc_vhid == carpr.carpr_vhid)
1844 					break;
1845 			if (sc == NULL) {
1846 				error = ENOENT;
1847 				break;
1848 			}
1849 			carp_carprcp(&carpr, sc, priveleged);
1850 			error = copyout(&carpr, ifr_data_get_ptr(ifr),
1851 			    sizeof(carpr));
1852 		} else  {
1853 			int i, count;
1854 
1855 			count = 0;
1856 			IFNET_FOREACH_CARP(ifp, sc)
1857 				count++;
1858 
1859 			if (count > carpr.carpr_count) {
1860 				CIF_UNLOCK(ifp->if_carp);
1861 				error = EMSGSIZE;
1862 				break;
1863 			}
1864 
1865 			i = 0;
1866 			IFNET_FOREACH_CARP(ifp, sc) {
1867 				carp_carprcp(&carpr, sc, priveleged);
1868 				carpr.carpr_count = count;
1869 				error = copyout(&carpr,
1870 				    (char *)ifr_data_get_ptr(ifr) +
1871 				    (i * sizeof(carpr)), sizeof(carpr));
1872 				if (error) {
1873 					CIF_UNLOCK(ifp->if_carp);
1874 					break;
1875 				}
1876 				i++;
1877 			}
1878 		}
1879 		break;
1880 	    }
1881 	default:
1882 		error = EINVAL;
1883 	}
1884 	sx_xunlock(&carp_sx);
1885 
1886 out:
1887 	if (locked)
1888 		CARP_UNLOCK(sc);
1889 	if_rele(ifp);
1890 
1891 	return (error);
1892 }
1893 
1894 static int
1895 carp_get_vhid(struct ifaddr *ifa)
1896 {
1897 
1898 	if (ifa == NULL || ifa->ifa_carp == NULL)
1899 		return (0);
1900 
1901 	return (ifa->ifa_carp->sc_vhid);
1902 }
1903 
1904 int
1905 carp_attach(struct ifaddr *ifa, int vhid)
1906 {
1907 	struct ifnet *ifp = ifa->ifa_ifp;
1908 	struct carp_if *cif = ifp->if_carp;
1909 	struct carp_softc *sc;
1910 	int index, error;
1911 
1912 	KASSERT(ifa->ifa_carp == NULL, ("%s: ifa %p attached", __func__, ifa));
1913 
1914 	switch (ifa->ifa_addr->sa_family) {
1915 #ifdef INET
1916 	case AF_INET:
1917 #endif
1918 #ifdef INET6
1919 	case AF_INET6:
1920 #endif
1921 		break;
1922 	default:
1923 		return (EPROTOTYPE);
1924 	}
1925 
1926 	sx_xlock(&carp_sx);
1927 	if (ifp->if_carp == NULL) {
1928 		sx_xunlock(&carp_sx);
1929 		return (ENOPROTOOPT);
1930 	}
1931 
1932 	IFNET_FOREACH_CARP(ifp, sc)
1933 		if (sc->sc_vhid == vhid)
1934 			break;
1935 	if (sc == NULL) {
1936 		sx_xunlock(&carp_sx);
1937 		return (ENOENT);
1938 	}
1939 
1940 	error = carp_multicast_setup(cif, ifa->ifa_addr->sa_family);
1941 	if (error) {
1942 		CIF_FREE(cif);
1943 		sx_xunlock(&carp_sx);
1944 		return (error);
1945 	}
1946 
1947 	index = sc->sc_naddrs + sc->sc_naddrs6 + 1;
1948 	if (index > sc->sc_ifasiz / sizeof(struct ifaddr *))
1949 		carp_grow_ifas(sc);
1950 
1951 	switch (ifa->ifa_addr->sa_family) {
1952 #ifdef INET
1953 	case AF_INET:
1954 		cif->cif_naddrs++;
1955 		sc->sc_naddrs++;
1956 		break;
1957 #endif
1958 #ifdef INET6
1959 	case AF_INET6:
1960 		cif->cif_naddrs6++;
1961 		sc->sc_naddrs6++;
1962 		break;
1963 #endif
1964 	}
1965 
1966 	ifa_ref(ifa);
1967 
1968 	CARP_LOCK(sc);
1969 	sc->sc_ifas[index - 1] = ifa;
1970 	ifa->ifa_carp = sc;
1971 	carp_hmac_prepare(sc);
1972 	carp_sc_state(sc);
1973 	CARP_UNLOCK(sc);
1974 
1975 	sx_xunlock(&carp_sx);
1976 
1977 	return (0);
1978 }
1979 
1980 void
1981 carp_detach(struct ifaddr *ifa, bool keep_cif)
1982 {
1983 	struct ifnet *ifp = ifa->ifa_ifp;
1984 	struct carp_if *cif = ifp->if_carp;
1985 	struct carp_softc *sc = ifa->ifa_carp;
1986 	int i, index;
1987 
1988 	KASSERT(sc != NULL, ("%s: %p not attached", __func__, ifa));
1989 
1990 	sx_xlock(&carp_sx);
1991 
1992 	CARP_LOCK(sc);
1993 	/* Shift array. */
1994 	index = sc->sc_naddrs + sc->sc_naddrs6;
1995 	for (i = 0; i < index; i++)
1996 		if (sc->sc_ifas[i] == ifa)
1997 			break;
1998 	KASSERT(i < index, ("%s: %p no backref", __func__, ifa));
1999 	for (; i < index - 1; i++)
2000 		sc->sc_ifas[i] = sc->sc_ifas[i+1];
2001 	sc->sc_ifas[index - 1] = NULL;
2002 
2003 	switch (ifa->ifa_addr->sa_family) {
2004 #ifdef INET
2005 	case AF_INET:
2006 		cif->cif_naddrs--;
2007 		sc->sc_naddrs--;
2008 		break;
2009 #endif
2010 #ifdef INET6
2011 	case AF_INET6:
2012 		cif->cif_naddrs6--;
2013 		sc->sc_naddrs6--;
2014 		break;
2015 #endif
2016 	}
2017 
2018 	carp_ifa_delroute(ifa);
2019 	carp_multicast_cleanup(cif, ifa->ifa_addr->sa_family);
2020 
2021 	ifa->ifa_carp = NULL;
2022 	ifa_free(ifa);
2023 
2024 	carp_hmac_prepare(sc);
2025 	carp_sc_state(sc);
2026 
2027 	if (!keep_cif && sc->sc_naddrs == 0 && sc->sc_naddrs6 == 0)
2028 		carp_destroy(sc);
2029 	else
2030 		CARP_UNLOCK(sc);
2031 
2032 	if (!keep_cif)
2033 		CIF_FREE(cif);
2034 
2035 	sx_xunlock(&carp_sx);
2036 }
2037 
2038 static void
2039 carp_set_state(struct carp_softc *sc, int state, const char *reason)
2040 {
2041 
2042 	CARP_LOCK_ASSERT(sc);
2043 
2044 	if (sc->sc_state != state) {
2045 		const char *carp_states[] = { CARP_STATES };
2046 		char subsys[IFNAMSIZ+5];
2047 
2048 		snprintf(subsys, IFNAMSIZ+5, "%u@%s", sc->sc_vhid,
2049 		    sc->sc_carpdev->if_xname);
2050 
2051 		CARP_LOG("%s: %s -> %s (%s)\n", subsys,
2052 		    carp_states[sc->sc_state], carp_states[state], reason);
2053 
2054 		sc->sc_state = state;
2055 
2056 		devctl_notify("CARP", subsys, carp_states[state], NULL);
2057 	}
2058 }
2059 
2060 static void
2061 carp_linkstate(struct ifnet *ifp)
2062 {
2063 	struct carp_softc *sc;
2064 
2065 	CIF_LOCK(ifp->if_carp);
2066 	IFNET_FOREACH_CARP(ifp, sc) {
2067 		CARP_LOCK(sc);
2068 		carp_sc_state(sc);
2069 		CARP_UNLOCK(sc);
2070 	}
2071 	CIF_UNLOCK(ifp->if_carp);
2072 }
2073 
2074 static void
2075 carp_sc_state(struct carp_softc *sc)
2076 {
2077 
2078 	CARP_LOCK_ASSERT(sc);
2079 
2080 	if (sc->sc_carpdev->if_link_state != LINK_STATE_UP ||
2081 	    !(sc->sc_carpdev->if_flags & IFF_UP) ||
2082 	    !V_carp_allow) {
2083 		callout_stop(&sc->sc_ad_tmo);
2084 #ifdef INET
2085 		callout_stop(&sc->sc_md_tmo);
2086 #endif
2087 #ifdef INET6
2088 		callout_stop(&sc->sc_md6_tmo);
2089 #endif
2090 		carp_set_state(sc, INIT, "hardware interface down");
2091 		carp_setrun(sc, 0);
2092 		if (!sc->sc_suppress)
2093 			carp_demote_adj(V_carp_ifdown_adj, "interface down");
2094 		sc->sc_suppress = 1;
2095 	} else {
2096 		carp_set_state(sc, INIT, "hardware interface up");
2097 		carp_setrun(sc, 0);
2098 		if (sc->sc_suppress)
2099 			carp_demote_adj(-V_carp_ifdown_adj, "interface up");
2100 		sc->sc_suppress = 0;
2101 	}
2102 }
2103 
2104 static void
2105 carp_demote_adj(int adj, char *reason)
2106 {
2107 	atomic_add_int(&V_carp_demotion, adj);
2108 	CARP_LOG("demoted by %d to %d (%s)\n", adj, V_carp_demotion, reason);
2109 	taskqueue_enqueue(taskqueue_swi, &carp_sendall_task);
2110 }
2111 
2112 static int
2113 carp_allow_sysctl(SYSCTL_HANDLER_ARGS)
2114 {
2115 	int new, error;
2116 	struct carp_softc *sc;
2117 
2118 	new = V_carp_allow;
2119 	error = sysctl_handle_int(oidp, &new, 0, req);
2120 	if (error || !req->newptr)
2121 		return (error);
2122 
2123 	if (V_carp_allow != new) {
2124 		V_carp_allow = new;
2125 
2126 		mtx_lock(&carp_mtx);
2127 		LIST_FOREACH(sc, &carp_list, sc_next) {
2128 			CARP_LOCK(sc);
2129 			if (curvnet == sc->sc_carpdev->if_vnet)
2130 				carp_sc_state(sc);
2131 			CARP_UNLOCK(sc);
2132 		}
2133 		mtx_unlock(&carp_mtx);
2134 	}
2135 
2136 	return (0);
2137 }
2138 
2139 static int
2140 carp_dscp_sysctl(SYSCTL_HANDLER_ARGS)
2141 {
2142 	int new, error;
2143 
2144 	new = V_carp_dscp;
2145 	error = sysctl_handle_int(oidp, &new, 0, req);
2146 	if (error || !req->newptr)
2147 		return (error);
2148 
2149 	if (new < 0 || new > 63)
2150 		return (EINVAL);
2151 
2152 	V_carp_dscp = new;
2153 
2154 	return (0);
2155 }
2156 
2157 static int
2158 carp_demote_adj_sysctl(SYSCTL_HANDLER_ARGS)
2159 {
2160 	int new, error;
2161 
2162 	new = V_carp_demotion;
2163 	error = sysctl_handle_int(oidp, &new, 0, req);
2164 	if (error || !req->newptr)
2165 		return (error);
2166 
2167 	carp_demote_adj(new, "sysctl");
2168 
2169 	return (0);
2170 }
2171 
2172 static void
2173 carp_mod_cleanup(void)
2174 {
2175 
2176 #ifdef INET
2177 	(void)ipproto_unregister(IPPROTO_CARP);
2178 	carp_iamatch_p = NULL;
2179 #endif
2180 #ifdef INET6
2181 	(void)ip6proto_unregister(IPPROTO_CARP);
2182 	carp_iamatch6_p = NULL;
2183 	carp_macmatch6_p = NULL;
2184 #endif
2185 	carp_ioctl_p = NULL;
2186 	carp_attach_p = NULL;
2187 	carp_detach_p = NULL;
2188 	carp_get_vhid_p = NULL;
2189 	carp_linkstate_p = NULL;
2190 	carp_forus_p = NULL;
2191 	carp_output_p = NULL;
2192 	carp_demote_adj_p = NULL;
2193 	carp_master_p = NULL;
2194 	mtx_unlock(&carp_mtx);
2195 	taskqueue_drain(taskqueue_swi, &carp_sendall_task);
2196 	mtx_destroy(&carp_mtx);
2197 	sx_destroy(&carp_sx);
2198 }
2199 
2200 static int
2201 carp_mod_load(void)
2202 {
2203 	int err;
2204 
2205 	mtx_init(&carp_mtx, "carp_mtx", NULL, MTX_DEF);
2206 	sx_init(&carp_sx, "carp_sx");
2207 	LIST_INIT(&carp_list);
2208 	carp_get_vhid_p = carp_get_vhid;
2209 	carp_forus_p = carp_forus;
2210 	carp_output_p = carp_output;
2211 	carp_linkstate_p = carp_linkstate;
2212 	carp_ioctl_p = carp_ioctl;
2213 	carp_attach_p = carp_attach;
2214 	carp_detach_p = carp_detach;
2215 	carp_demote_adj_p = carp_demote_adj;
2216 	carp_master_p = carp_master;
2217 #ifdef INET6
2218 	carp_iamatch6_p = carp_iamatch6;
2219 	carp_macmatch6_p = carp_macmatch6;
2220 	err = ip6proto_register(IPPROTO_CARP, carp6_input, NULL);
2221 	if (err) {
2222 		printf("carp: error %d registering with INET6\n", err);
2223 		carp_mod_cleanup();
2224 		return (err);
2225 	}
2226 #endif
2227 #ifdef INET
2228 	carp_iamatch_p = carp_iamatch;
2229 	err = ipproto_register(IPPROTO_CARP, carp_input, NULL);
2230 	if (err) {
2231 		printf("carp: error %d registering with INET\n", err);
2232 		carp_mod_cleanup();
2233 		return (err);
2234 	}
2235 #endif
2236 	return (0);
2237 }
2238 
2239 static int
2240 carp_modevent(module_t mod, int type, void *data)
2241 {
2242 	switch (type) {
2243 	case MOD_LOAD:
2244 		return carp_mod_load();
2245 		/* NOTREACHED */
2246 	case MOD_UNLOAD:
2247 		mtx_lock(&carp_mtx);
2248 		if (LIST_EMPTY(&carp_list))
2249 			carp_mod_cleanup();
2250 		else {
2251 			mtx_unlock(&carp_mtx);
2252 			return (EBUSY);
2253 		}
2254 		break;
2255 
2256 	default:
2257 		return (EINVAL);
2258 	}
2259 
2260 	return (0);
2261 }
2262 
2263 static moduledata_t carp_mod = {
2264 	"carp",
2265 	carp_modevent,
2266 	0
2267 };
2268 
2269 DECLARE_MODULE(carp, carp_mod, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY);
2270