xref: /freebsd/sys/netinet/ip_carp.c (revision 9768746b)
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_private.h>
62 #include <net/if_types.h>
63 #include <net/route.h>
64 #include <net/vnet.h>
65 
66 #if defined(INET) || defined(INET6)
67 #include <netinet/in.h>
68 #include <netinet/in_var.h>
69 #include <netinet/ip_carp.h>
70 #include <netinet/ip.h>
71 #include <machine/in_cksum.h>
72 #endif
73 #ifdef INET
74 #include <netinet/ip_var.h>
75 #include <netinet/if_ether.h>
76 #endif
77 
78 #ifdef INET6
79 #include <netinet/icmp6.h>
80 #include <netinet/ip6.h>
81 #include <netinet6/in6_var.h>
82 #include <netinet6/ip6_var.h>
83 #include <netinet6/scope6_var.h>
84 #include <netinet6/nd6.h>
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 	uint32_t	cif_flags;
148 #define	CIF_PROMISC	0x00000001
149 };
150 
151 /*
152  * Brief design of carp(4).
153  *
154  * Any carp-capable ifnet may have a list of carp softcs hanging off
155  * its ifp->if_carp pointer. Each softc represents one unique virtual
156  * host id, or vhid. The softc has a back pointer to the ifnet. All
157  * softcs are joined in a global list, which has quite limited use.
158  *
159  * Any interface address that takes part in CARP negotiation has a
160  * pointer to the softc of its vhid, ifa->ifa_carp. That could be either
161  * AF_INET or AF_INET6 address.
162  *
163  * Although, one can get the softc's backpointer to ifnet and traverse
164  * through its ifp->if_addrhead queue to find all interface addresses
165  * involved in CARP, we keep a growable array of ifaddr pointers. This
166  * allows us to avoid grabbing the IF_ADDR_LOCK() in many traversals that
167  * do calls into the network stack, thus avoiding LORs.
168  *
169  * Locking:
170  *
171  * Each softc has a lock sc_mtx. It is used to synchronise carp_input_c(),
172  * callout-driven events and ioctl()s.
173  *
174  * To traverse the list of softcs on an ifnet we use CIF_LOCK() or carp_sx.
175  * To traverse the global list we use the mutex carp_mtx.
176  *
177  * Known issues with locking:
178  *
179  * - Sending ad, we put the pointer to the softc in an mtag, and no reference
180  *   counting is done on the softc.
181  * - On module unload we may race (?) with packet processing thread
182  *   dereferencing our function pointers.
183  */
184 
185 /* Accept incoming CARP packets. */
186 VNET_DEFINE_STATIC(int, carp_allow) = 1;
187 #define	V_carp_allow	VNET(carp_allow)
188 
189 /* Set DSCP in outgoing CARP packets. */
190 VNET_DEFINE_STATIC(int, carp_dscp) = 56;
191 #define	V_carp_dscp	VNET(carp_dscp)
192 
193 /* Preempt slower nodes. */
194 VNET_DEFINE_STATIC(int, carp_preempt) = 0;
195 #define	V_carp_preempt	VNET(carp_preempt)
196 
197 /* Log level. */
198 VNET_DEFINE_STATIC(int, carp_log) = 1;
199 #define	V_carp_log	VNET(carp_log)
200 
201 /* Global advskew demotion. */
202 VNET_DEFINE_STATIC(int, carp_demotion) = 0;
203 #define	V_carp_demotion	VNET(carp_demotion)
204 
205 /* Send error demotion factor. */
206 VNET_DEFINE_STATIC(int, carp_senderr_adj) = CARP_MAXSKEW;
207 #define	V_carp_senderr_adj	VNET(carp_senderr_adj)
208 
209 /* Iface down demotion factor. */
210 VNET_DEFINE_STATIC(int, carp_ifdown_adj) = CARP_MAXSKEW;
211 #define	V_carp_ifdown_adj	VNET(carp_ifdown_adj)
212 
213 static int carp_allow_sysctl(SYSCTL_HANDLER_ARGS);
214 static int carp_dscp_sysctl(SYSCTL_HANDLER_ARGS);
215 static int carp_demote_adj_sysctl(SYSCTL_HANDLER_ARGS);
216 
217 SYSCTL_NODE(_net_inet, IPPROTO_CARP, carp, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
218     "CARP");
219 SYSCTL_PROC(_net_inet_carp, OID_AUTO, allow,
220     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RWTUN | CTLFLAG_MPSAFE,
221     &VNET_NAME(carp_allow), 0, carp_allow_sysctl, "I",
222     "Accept incoming CARP packets");
223 SYSCTL_PROC(_net_inet_carp, OID_AUTO, dscp,
224     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
225     0, 0, carp_dscp_sysctl, "I",
226     "DSCP value for carp packets");
227 SYSCTL_INT(_net_inet_carp, OID_AUTO, preempt, CTLFLAG_VNET | CTLFLAG_RW,
228     &VNET_NAME(carp_preempt), 0, "High-priority backup preemption mode");
229 SYSCTL_INT(_net_inet_carp, OID_AUTO, log, CTLFLAG_VNET | CTLFLAG_RW,
230     &VNET_NAME(carp_log), 0, "CARP log level");
231 SYSCTL_PROC(_net_inet_carp, OID_AUTO, demotion,
232     CTLFLAG_VNET | CTLTYPE_INT | CTLFLAG_RW | CTLFLAG_MPSAFE,
233     0, 0, carp_demote_adj_sysctl, "I",
234     "Adjust demotion factor (skew of advskew)");
235 SYSCTL_INT(_net_inet_carp, OID_AUTO, senderr_demotion_factor,
236     CTLFLAG_VNET | CTLFLAG_RW,
237     &VNET_NAME(carp_senderr_adj), 0, "Send error demotion factor adjustment");
238 SYSCTL_INT(_net_inet_carp, OID_AUTO, ifdown_demotion_factor,
239     CTLFLAG_VNET | CTLFLAG_RW,
240     &VNET_NAME(carp_ifdown_adj), 0,
241     "Interface down demotion factor adjustment");
242 
243 VNET_PCPUSTAT_DEFINE(struct carpstats, carpstats);
244 VNET_PCPUSTAT_SYSINIT(carpstats);
245 VNET_PCPUSTAT_SYSUNINIT(carpstats);
246 
247 #define	CARPSTATS_ADD(name, val)	\
248     counter_u64_add(VNET(carpstats)[offsetof(struct carpstats, name) / \
249 	sizeof(uint64_t)], (val))
250 #define	CARPSTATS_INC(name)		CARPSTATS_ADD(name, 1)
251 
252 SYSCTL_VNET_PCPUSTAT(_net_inet_carp, OID_AUTO, stats, struct carpstats,
253     carpstats, "CARP statistics (struct carpstats, netinet/ip_carp.h)");
254 
255 #define	CARP_LOCK_INIT(sc)	mtx_init(&(sc)->sc_mtx, "carp_softc",   \
256 	NULL, MTX_DEF)
257 #define	CARP_LOCK_DESTROY(sc)	mtx_destroy(&(sc)->sc_mtx)
258 #define	CARP_LOCK_ASSERT(sc)	mtx_assert(&(sc)->sc_mtx, MA_OWNED)
259 #define	CARP_LOCK(sc)		mtx_lock(&(sc)->sc_mtx)
260 #define	CARP_UNLOCK(sc)		mtx_unlock(&(sc)->sc_mtx)
261 #define	CIF_LOCK_INIT(cif)	mtx_init(&(cif)->cif_mtx, "carp_if",   \
262 	NULL, MTX_DEF)
263 #define	CIF_LOCK_DESTROY(cif)	mtx_destroy(&(cif)->cif_mtx)
264 #define	CIF_LOCK_ASSERT(cif)	mtx_assert(&(cif)->cif_mtx, MA_OWNED)
265 #define	CIF_LOCK(cif)		mtx_lock(&(cif)->cif_mtx)
266 #define	CIF_UNLOCK(cif)		mtx_unlock(&(cif)->cif_mtx)
267 #define	CIF_FREE(cif)	do {				\
268 		CIF_LOCK(cif);				\
269 		if (TAILQ_EMPTY(&(cif)->cif_vrs))	\
270 			carp_free_if(cif);		\
271 		else					\
272 			CIF_UNLOCK(cif);		\
273 } while (0)
274 
275 #define	CARP_LOG(...)	do {				\
276 	if (V_carp_log > 0)				\
277 		log(LOG_INFO, "carp: " __VA_ARGS__);	\
278 } while (0)
279 
280 #define	CARP_DEBUG(...)	do {				\
281 	if (V_carp_log > 1)				\
282 		log(LOG_DEBUG, __VA_ARGS__);		\
283 } while (0)
284 
285 #define	IFNET_FOREACH_IFA(ifp, ifa)					\
286 	CK_STAILQ_FOREACH((ifa), &(ifp)->if_addrhead, ifa_link)	\
287 		if ((ifa)->ifa_carp != NULL)
288 
289 #define	CARP_FOREACH_IFA(sc, ifa)					\
290 	CARP_LOCK_ASSERT(sc);						\
291 	for (int _i = 0;						\
292 		_i < (sc)->sc_naddrs + (sc)->sc_naddrs6 &&		\
293 		((ifa) = sc->sc_ifas[_i]) != NULL;			\
294 		++_i)
295 
296 #define	IFNET_FOREACH_CARP(ifp, sc)					\
297 	KASSERT(mtx_owned(&ifp->if_carp->cif_mtx) ||			\
298 	    sx_xlocked(&carp_sx), ("cif_vrs not locked"));		\
299 	TAILQ_FOREACH((sc), &(ifp)->if_carp->cif_vrs, sc_list)
300 
301 #define	DEMOTE_ADVSKEW(sc)					\
302     (((sc)->sc_advskew + V_carp_demotion > CARP_MAXSKEW) ?	\
303     CARP_MAXSKEW :						\
304         (((sc)->sc_advskew + V_carp_demotion < 0) ?		\
305         0 : ((sc)->sc_advskew + V_carp_demotion)))
306 
307 static void	carp_input_c(struct mbuf *, struct carp_header *, sa_family_t);
308 static struct carp_softc
309 		*carp_alloc(struct ifnet *);
310 static void	carp_destroy(struct carp_softc *);
311 static struct carp_if
312 		*carp_alloc_if(struct ifnet *);
313 static void	carp_free_if(struct carp_if *);
314 static void	carp_set_state(struct carp_softc *, int, const char* reason);
315 static void	carp_sc_state(struct carp_softc *);
316 static void	carp_setrun(struct carp_softc *, sa_family_t);
317 static void	carp_master_down(void *);
318 static void	carp_master_down_locked(struct carp_softc *,
319     		    const char* reason);
320 static void	carp_send_ad(void *);
321 static void	carp_send_ad_locked(struct carp_softc *);
322 static void	carp_addroute(struct carp_softc *);
323 static void	carp_ifa_addroute(struct ifaddr *);
324 static void	carp_delroute(struct carp_softc *);
325 static void	carp_ifa_delroute(struct ifaddr *);
326 static void	carp_send_ad_all(void *, int);
327 static void	carp_demote_adj(int, char *);
328 
329 static LIST_HEAD(, carp_softc) carp_list;
330 static struct mtx carp_mtx;
331 static struct sx carp_sx;
332 static struct task carp_sendall_task =
333     TASK_INITIALIZER(0, carp_send_ad_all, NULL);
334 
335 static void
336 carp_hmac_prepare(struct carp_softc *sc)
337 {
338 	uint8_t version = CARP_VERSION, type = CARP_ADVERTISEMENT;
339 	uint8_t vhid = sc->sc_vhid & 0xff;
340 	struct ifaddr *ifa;
341 	int i, found;
342 #ifdef INET
343 	struct in_addr last, cur, in;
344 #endif
345 #ifdef INET6
346 	struct in6_addr last6, cur6, in6;
347 #endif
348 
349 	CARP_LOCK_ASSERT(sc);
350 
351 	/* Compute ipad from key. */
352 	bzero(sc->sc_pad, sizeof(sc->sc_pad));
353 	bcopy(sc->sc_key, sc->sc_pad, sizeof(sc->sc_key));
354 	for (i = 0; i < sizeof(sc->sc_pad); i++)
355 		sc->sc_pad[i] ^= 0x36;
356 
357 	/* Precompute first part of inner hash. */
358 	SHA1Init(&sc->sc_sha1);
359 	SHA1Update(&sc->sc_sha1, sc->sc_pad, sizeof(sc->sc_pad));
360 	SHA1Update(&sc->sc_sha1, (void *)&version, sizeof(version));
361 	SHA1Update(&sc->sc_sha1, (void *)&type, sizeof(type));
362 	SHA1Update(&sc->sc_sha1, (void *)&vhid, sizeof(vhid));
363 #ifdef INET
364 	cur.s_addr = 0;
365 	do {
366 		found = 0;
367 		last = cur;
368 		cur.s_addr = 0xffffffff;
369 		CARP_FOREACH_IFA(sc, ifa) {
370 			in.s_addr = ifatoia(ifa)->ia_addr.sin_addr.s_addr;
371 			if (ifa->ifa_addr->sa_family == AF_INET &&
372 			    ntohl(in.s_addr) > ntohl(last.s_addr) &&
373 			    ntohl(in.s_addr) < ntohl(cur.s_addr)) {
374 				cur.s_addr = in.s_addr;
375 				found++;
376 			}
377 		}
378 		if (found)
379 			SHA1Update(&sc->sc_sha1, (void *)&cur, sizeof(cur));
380 	} while (found);
381 #endif /* INET */
382 #ifdef INET6
383 	memset(&cur6, 0, sizeof(cur6));
384 	do {
385 		found = 0;
386 		last6 = cur6;
387 		memset(&cur6, 0xff, sizeof(cur6));
388 		CARP_FOREACH_IFA(sc, ifa) {
389 			in6 = ifatoia6(ifa)->ia_addr.sin6_addr;
390 			if (IN6_IS_SCOPE_EMBED(&in6))
391 				in6.s6_addr16[1] = 0;
392 			if (ifa->ifa_addr->sa_family == AF_INET6 &&
393 			    memcmp(&in6, &last6, sizeof(in6)) > 0 &&
394 			    memcmp(&in6, &cur6, sizeof(in6)) < 0) {
395 				cur6 = in6;
396 				found++;
397 			}
398 		}
399 		if (found)
400 			SHA1Update(&sc->sc_sha1, (void *)&cur6, sizeof(cur6));
401 	} while (found);
402 #endif /* INET6 */
403 
404 	/* convert ipad to opad */
405 	for (i = 0; i < sizeof(sc->sc_pad); i++)
406 		sc->sc_pad[i] ^= 0x36 ^ 0x5c;
407 }
408 
409 static void
410 carp_hmac_generate(struct carp_softc *sc, uint32_t counter[2],
411     unsigned char md[20])
412 {
413 	SHA1_CTX sha1ctx;
414 
415 	CARP_LOCK_ASSERT(sc);
416 
417 	/* fetch first half of inner hash */
418 	bcopy(&sc->sc_sha1, &sha1ctx, sizeof(sha1ctx));
419 
420 	SHA1Update(&sha1ctx, (void *)counter, sizeof(sc->sc_counter));
421 	SHA1Final(md, &sha1ctx);
422 
423 	/* outer hash */
424 	SHA1Init(&sha1ctx);
425 	SHA1Update(&sha1ctx, sc->sc_pad, sizeof(sc->sc_pad));
426 	SHA1Update(&sha1ctx, md, 20);
427 	SHA1Final(md, &sha1ctx);
428 }
429 
430 static int
431 carp_hmac_verify(struct carp_softc *sc, uint32_t counter[2],
432     unsigned char md[20])
433 {
434 	unsigned char md2[20];
435 
436 	CARP_LOCK_ASSERT(sc);
437 
438 	carp_hmac_generate(sc, counter, md2);
439 
440 	return (bcmp(md, md2, sizeof(md2)));
441 }
442 
443 /*
444  * process input packet.
445  * we have rearranged checks order compared to the rfc,
446  * but it seems more efficient this way or not possible otherwise.
447  */
448 #ifdef INET
449 static int
450 carp_input(struct mbuf **mp, int *offp, int proto)
451 {
452 	struct mbuf *m = *mp;
453 	struct ip *ip = mtod(m, struct ip *);
454 	struct carp_header *ch;
455 	int iplen, len;
456 
457 	iplen = *offp;
458 	*mp = NULL;
459 
460 	CARPSTATS_INC(carps_ipackets);
461 
462 	if (!V_carp_allow) {
463 		m_freem(m);
464 		return (IPPROTO_DONE);
465 	}
466 
467 	/* verify that the IP TTL is 255.  */
468 	if (ip->ip_ttl != CARP_DFLTTL) {
469 		CARPSTATS_INC(carps_badttl);
470 		CARP_DEBUG("%s: received ttl %d != 255 on %s\n", __func__,
471 		    ip->ip_ttl,
472 		    m->m_pkthdr.rcvif->if_xname);
473 		m_freem(m);
474 		return (IPPROTO_DONE);
475 	}
476 
477 	iplen = ip->ip_hl << 2;
478 
479 	if (m->m_pkthdr.len < iplen + sizeof(*ch)) {
480 		CARPSTATS_INC(carps_badlen);
481 		CARP_DEBUG("%s: received len %zd < sizeof(struct carp_header) "
482 		    "on %s\n", __func__, m->m_len - sizeof(struct ip),
483 		    m->m_pkthdr.rcvif->if_xname);
484 		m_freem(m);
485 		return (IPPROTO_DONE);
486 	}
487 
488 	if (iplen + sizeof(*ch) < m->m_len) {
489 		if ((m = m_pullup(m, iplen + sizeof(*ch))) == NULL) {
490 			CARPSTATS_INC(carps_hdrops);
491 			CARP_DEBUG("%s: pullup failed\n", __func__);
492 			return (IPPROTO_DONE);
493 		}
494 		ip = mtod(m, struct ip *);
495 	}
496 	ch = (struct carp_header *)((char *)ip + iplen);
497 
498 	/*
499 	 * verify that the received packet length is
500 	 * equal to the CARP header
501 	 */
502 	len = iplen + sizeof(*ch);
503 	if (len > m->m_pkthdr.len) {
504 		CARPSTATS_INC(carps_badlen);
505 		CARP_DEBUG("%s: packet too short %d on %s\n", __func__,
506 		    m->m_pkthdr.len,
507 		    m->m_pkthdr.rcvif->if_xname);
508 		m_freem(m);
509 		return (IPPROTO_DONE);
510 	}
511 
512 	if ((m = m_pullup(m, len)) == NULL) {
513 		CARPSTATS_INC(carps_hdrops);
514 		return (IPPROTO_DONE);
515 	}
516 	ip = mtod(m, struct ip *);
517 	ch = (struct carp_header *)((char *)ip + iplen);
518 
519 	/* verify the CARP checksum */
520 	m->m_data += iplen;
521 	if (in_cksum(m, len - iplen)) {
522 		CARPSTATS_INC(carps_badsum);
523 		CARP_DEBUG("%s: checksum failed on %s\n", __func__,
524 		    m->m_pkthdr.rcvif->if_xname);
525 		m_freem(m);
526 		return (IPPROTO_DONE);
527 	}
528 	m->m_data -= iplen;
529 
530 	carp_input_c(m, ch, AF_INET);
531 	return (IPPROTO_DONE);
532 }
533 #endif
534 
535 #ifdef INET6
536 static int
537 carp6_input(struct mbuf **mp, int *offp, int proto)
538 {
539 	struct mbuf *m = *mp;
540 	struct ip6_hdr *ip6 = mtod(m, struct ip6_hdr *);
541 	struct carp_header *ch;
542 	u_int len;
543 
544 	CARPSTATS_INC(carps_ipackets6);
545 
546 	if (!V_carp_allow) {
547 		m_freem(m);
548 		return (IPPROTO_DONE);
549 	}
550 
551 	/* check if received on a valid carp interface */
552 	if (m->m_pkthdr.rcvif->if_carp == NULL) {
553 		CARPSTATS_INC(carps_badif);
554 		CARP_DEBUG("%s: packet received on non-carp interface: %s\n",
555 		    __func__, m->m_pkthdr.rcvif->if_xname);
556 		m_freem(m);
557 		return (IPPROTO_DONE);
558 	}
559 
560 	/* verify that the IP TTL is 255 */
561 	if (ip6->ip6_hlim != CARP_DFLTTL) {
562 		CARPSTATS_INC(carps_badttl);
563 		CARP_DEBUG("%s: received ttl %d != 255 on %s\n", __func__,
564 		    ip6->ip6_hlim, m->m_pkthdr.rcvif->if_xname);
565 		m_freem(m);
566 		return (IPPROTO_DONE);
567 	}
568 
569 	/* verify that we have a complete carp packet */
570 	if (m->m_len < *offp + sizeof(*ch)) {
571 		len = m->m_len;
572 		m = m_pullup(m, *offp + sizeof(*ch));
573 		if (m == NULL) {
574 			CARPSTATS_INC(carps_badlen);
575 			CARP_DEBUG("%s: packet size %u too small\n", __func__, len);
576 			return (IPPROTO_DONE);
577 		}
578 	}
579 	ch = (struct carp_header *)(mtod(m, char *) + *offp);
580 
581 	/* verify the CARP checksum */
582 	m->m_data += *offp;
583 	if (in_cksum(m, sizeof(*ch))) {
584 		CARPSTATS_INC(carps_badsum);
585 		CARP_DEBUG("%s: checksum failed, on %s\n", __func__,
586 		    m->m_pkthdr.rcvif->if_xname);
587 		m_freem(m);
588 		return (IPPROTO_DONE);
589 	}
590 	m->m_data -= *offp;
591 
592 	carp_input_c(m, ch, AF_INET6);
593 	return (IPPROTO_DONE);
594 }
595 #endif /* INET6 */
596 
597 /*
598  * This routine should not be necessary at all, but some switches
599  * (VMWare ESX vswitches) can echo our own packets back at us,
600  * and we must ignore them or they will cause us to drop out of
601  * MASTER mode.
602  *
603  * We cannot catch all cases of network loops.  Instead, what we
604  * do here is catch any packet that arrives with a carp header
605  * with a VHID of 0, that comes from an address that is our own.
606  * These packets are by definition "from us" (even if they are from
607  * a misconfigured host that is pretending to be us).
608  *
609  * The VHID test is outside this mini-function.
610  */
611 static int
612 carp_source_is_self(struct mbuf *m, struct ifaddr *ifa, sa_family_t af)
613 {
614 #ifdef INET
615 	struct ip *ip4;
616 	struct in_addr in4;
617 #endif
618 #ifdef INET6
619 	struct ip6_hdr *ip6;
620 	struct in6_addr in6;
621 #endif
622 
623 	switch (af) {
624 #ifdef INET
625 	case AF_INET:
626 		ip4 = mtod(m, struct ip *);
627 		in4 = ifatoia(ifa)->ia_addr.sin_addr;
628 		return (in4.s_addr == ip4->ip_src.s_addr);
629 #endif
630 #ifdef INET6
631 	case AF_INET6:
632 		ip6 = mtod(m, struct ip6_hdr *);
633 		in6 = ifatoia6(ifa)->ia_addr.sin6_addr;
634 		return (memcmp(&in6, &ip6->ip6_src, sizeof(in6)) == 0);
635 #endif
636 	default:
637 		break;
638 	}
639 	return (0);
640 }
641 
642 static void
643 carp_input_c(struct mbuf *m, struct carp_header *ch, sa_family_t af)
644 {
645 	struct ifnet *ifp = m->m_pkthdr.rcvif;
646 	struct ifaddr *ifa, *match;
647 	struct carp_softc *sc;
648 	uint64_t tmp_counter;
649 	struct timeval sc_tv, ch_tv;
650 	int error;
651 
652 	NET_EPOCH_ASSERT();
653 
654 	/*
655 	 * Verify that the VHID is valid on the receiving interface.
656 	 *
657 	 * There should be just one match.  If there are none
658 	 * the VHID is not valid and we drop the packet.  If
659 	 * there are multiple VHID matches, take just the first
660 	 * one, for compatibility with previous code.  While we're
661 	 * scanning, check for obvious loops in the network topology
662 	 * (these should never happen, and as noted above, we may
663 	 * miss real loops; this is just a double-check).
664 	 */
665 	error = 0;
666 	match = NULL;
667 	IFNET_FOREACH_IFA(ifp, ifa) {
668 		if (match == NULL && ifa->ifa_carp != NULL &&
669 		    ifa->ifa_addr->sa_family == af &&
670 		    ifa->ifa_carp->sc_vhid == ch->carp_vhid)
671 			match = ifa;
672 		if (ch->carp_vhid == 0 && carp_source_is_self(m, ifa, af))
673 			error = ELOOP;
674 	}
675 	ifa = error ? NULL : match;
676 	if (ifa != NULL)
677 		ifa_ref(ifa);
678 
679 	if (ifa == NULL) {
680 		if (error == ELOOP) {
681 			CARP_DEBUG("dropping looped packet on interface %s\n",
682 			    ifp->if_xname);
683 			CARPSTATS_INC(carps_badif);	/* ??? */
684 		} else {
685 			CARPSTATS_INC(carps_badvhid);
686 		}
687 		m_freem(m);
688 		return;
689 	}
690 
691 	/* verify the CARP version. */
692 	if (ch->carp_version != CARP_VERSION) {
693 		CARPSTATS_INC(carps_badver);
694 		CARP_DEBUG("%s: invalid version %d\n", ifp->if_xname,
695 		    ch->carp_version);
696 		ifa_free(ifa);
697 		m_freem(m);
698 		return;
699 	}
700 
701 	sc = ifa->ifa_carp;
702 	CARP_LOCK(sc);
703 	ifa_free(ifa);
704 
705 	if (carp_hmac_verify(sc, ch->carp_counter, ch->carp_md)) {
706 		CARPSTATS_INC(carps_badauth);
707 		CARP_DEBUG("%s: incorrect hash for VHID %u@%s\n", __func__,
708 		    sc->sc_vhid, ifp->if_xname);
709 		goto out;
710 	}
711 
712 	tmp_counter = ntohl(ch->carp_counter[0]);
713 	tmp_counter = tmp_counter<<32;
714 	tmp_counter += ntohl(ch->carp_counter[1]);
715 
716 	/* XXX Replay protection goes here */
717 
718 	sc->sc_init_counter = 0;
719 	sc->sc_counter = tmp_counter;
720 
721 	sc_tv.tv_sec = sc->sc_advbase;
722 	sc_tv.tv_usec = DEMOTE_ADVSKEW(sc) * 1000000 / 256;
723 	ch_tv.tv_sec = ch->carp_advbase;
724 	ch_tv.tv_usec = ch->carp_advskew * 1000000 / 256;
725 
726 	switch (sc->sc_state) {
727 	case INIT:
728 		break;
729 	case MASTER:
730 		/*
731 		 * If we receive an advertisement from a master who's going to
732 		 * be more frequent than us, go into BACKUP state.
733 		 */
734 		if (timevalcmp(&sc_tv, &ch_tv, >) ||
735 		    timevalcmp(&sc_tv, &ch_tv, ==)) {
736 			callout_stop(&sc->sc_ad_tmo);
737 			carp_set_state(sc, BACKUP,
738 			    "more frequent advertisement received");
739 			carp_setrun(sc, 0);
740 			carp_delroute(sc);
741 		}
742 		break;
743 	case BACKUP:
744 		/*
745 		 * If we're pre-empting masters who advertise slower than us,
746 		 * and this one claims to be slower, treat him as down.
747 		 */
748 		if (V_carp_preempt && timevalcmp(&sc_tv, &ch_tv, <)) {
749 			carp_master_down_locked(sc,
750 			    "preempting a slower master");
751 			break;
752 		}
753 
754 		/*
755 		 *  If the master is going to advertise at such a low frequency
756 		 *  that he's guaranteed to time out, we'd might as well just
757 		 *  treat him as timed out now.
758 		 */
759 		sc_tv.tv_sec = sc->sc_advbase * 3;
760 		if (timevalcmp(&sc_tv, &ch_tv, <)) {
761 			carp_master_down_locked(sc, "master will time out");
762 			break;
763 		}
764 
765 		/*
766 		 * Otherwise, we reset the counter and wait for the next
767 		 * advertisement.
768 		 */
769 		carp_setrun(sc, af);
770 		break;
771 	}
772 
773 out:
774 	CARP_UNLOCK(sc);
775 	m_freem(m);
776 }
777 
778 static int
779 carp_prepare_ad(struct mbuf *m, struct carp_softc *sc, struct carp_header *ch)
780 {
781 	struct m_tag *mtag;
782 
783 	if (sc->sc_init_counter) {
784 		/* this could also be seconds since unix epoch */
785 		sc->sc_counter = arc4random();
786 		sc->sc_counter = sc->sc_counter << 32;
787 		sc->sc_counter += arc4random();
788 	} else
789 		sc->sc_counter++;
790 
791 	ch->carp_counter[0] = htonl((sc->sc_counter>>32)&0xffffffff);
792 	ch->carp_counter[1] = htonl(sc->sc_counter&0xffffffff);
793 
794 	carp_hmac_generate(sc, ch->carp_counter, ch->carp_md);
795 
796 	/* Tag packet for carp_output */
797 	if ((mtag = m_tag_get(PACKET_TAG_CARP, sizeof(struct carp_softc *),
798 	    M_NOWAIT)) == NULL) {
799 		m_freem(m);
800 		CARPSTATS_INC(carps_onomem);
801 		return (ENOMEM);
802 	}
803 	bcopy(&sc, mtag + 1, sizeof(sc));
804 	m_tag_prepend(m, mtag);
805 
806 	return (0);
807 }
808 
809 /*
810  * To avoid LORs and possible recursions this function shouldn't
811  * be called directly, but scheduled via taskqueue.
812  */
813 static void
814 carp_send_ad_all(void *ctx __unused, int pending __unused)
815 {
816 	struct carp_softc *sc;
817 	struct epoch_tracker et;
818 
819 	NET_EPOCH_ENTER(et);
820 	mtx_lock(&carp_mtx);
821 	LIST_FOREACH(sc, &carp_list, sc_next)
822 		if (sc->sc_state == MASTER) {
823 			CARP_LOCK(sc);
824 			CURVNET_SET(sc->sc_carpdev->if_vnet);
825 			carp_send_ad_locked(sc);
826 			CURVNET_RESTORE();
827 			CARP_UNLOCK(sc);
828 		}
829 	mtx_unlock(&carp_mtx);
830 	NET_EPOCH_EXIT(et);
831 }
832 
833 /* Send a periodic advertisement, executed in callout context. */
834 static void
835 carp_send_ad(void *v)
836 {
837 	struct carp_softc *sc = v;
838 	struct epoch_tracker et;
839 
840 	NET_EPOCH_ENTER(et);
841 	CARP_LOCK_ASSERT(sc);
842 	CURVNET_SET(sc->sc_carpdev->if_vnet);
843 	carp_send_ad_locked(sc);
844 	CURVNET_RESTORE();
845 	CARP_UNLOCK(sc);
846 	NET_EPOCH_EXIT(et);
847 }
848 
849 static void
850 carp_send_ad_error(struct carp_softc *sc, int error)
851 {
852 
853 	/*
854 	 * We track errors and successfull sends with this logic:
855 	 * - Any error resets success counter to 0.
856 	 * - MAX_ERRORS triggers demotion.
857 	 * - MIN_SUCCESS successes resets error counter to 0.
858 	 * - MIN_SUCCESS reverts demotion, if it was triggered before.
859 	 */
860 	if (error) {
861 		if (sc->sc_sendad_errors < INT_MAX)
862 			sc->sc_sendad_errors++;
863 		if (sc->sc_sendad_errors == CARP_SENDAD_MAX_ERRORS) {
864 			static const char fmt[] = "send error %d on %s";
865 			char msg[sizeof(fmt) + IFNAMSIZ];
866 
867 			sprintf(msg, fmt, error, sc->sc_carpdev->if_xname);
868 			carp_demote_adj(V_carp_senderr_adj, msg);
869 		}
870 		sc->sc_sendad_success = 0;
871 	} else if (sc->sc_sendad_errors > 0) {
872 		if (++sc->sc_sendad_success >= CARP_SENDAD_MIN_SUCCESS) {
873 			if (sc->sc_sendad_errors >= CARP_SENDAD_MAX_ERRORS) {
874 				static const char fmt[] = "send ok on %s";
875 				char msg[sizeof(fmt) + IFNAMSIZ];
876 
877 				sprintf(msg, fmt, sc->sc_carpdev->if_xname);
878 				carp_demote_adj(-V_carp_senderr_adj, msg);
879 			}
880 			sc->sc_sendad_errors = 0;
881 		}
882 	}
883 }
884 
885 /*
886  * Pick the best ifaddr on the given ifp for sending CARP
887  * advertisements.
888  *
889  * "Best" here is defined by ifa_preferred().  This function is much
890  * much like ifaof_ifpforaddr() except that we just use ifa_preferred().
891  *
892  * (This could be simplified to return the actual address, except that
893  * it has a different format in AF_INET and AF_INET6.)
894  */
895 static struct ifaddr *
896 carp_best_ifa(int af, struct ifnet *ifp)
897 {
898 	struct ifaddr *ifa, *best;
899 
900 	NET_EPOCH_ASSERT();
901 
902 	if (af >= AF_MAX)
903 		return (NULL);
904 	best = NULL;
905 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
906 		if (ifa->ifa_addr->sa_family == af &&
907 		    (best == NULL || ifa_preferred(best, ifa)))
908 			best = ifa;
909 	}
910 	if (best != NULL)
911 		ifa_ref(best);
912 	return (best);
913 }
914 
915 static void
916 carp_send_ad_locked(struct carp_softc *sc)
917 {
918 	struct carp_header ch;
919 	struct timeval tv;
920 	struct ifaddr *ifa;
921 	struct carp_header *ch_ptr;
922 	struct mbuf *m;
923 	int len, advskew;
924 
925 	NET_EPOCH_ASSERT();
926 	CARP_LOCK_ASSERT(sc);
927 
928 	advskew = DEMOTE_ADVSKEW(sc);
929 	tv.tv_sec = sc->sc_advbase;
930 	tv.tv_usec = advskew * 1000000 / 256;
931 
932 	ch.carp_version = CARP_VERSION;
933 	ch.carp_type = CARP_ADVERTISEMENT;
934 	ch.carp_vhid = sc->sc_vhid;
935 	ch.carp_advbase = sc->sc_advbase;
936 	ch.carp_advskew = advskew;
937 	ch.carp_authlen = 7;	/* XXX DEFINE */
938 	ch.carp_pad1 = 0;	/* must be zero */
939 	ch.carp_cksum = 0;
940 
941 	/* XXXGL: OpenBSD picks first ifaddr with needed family. */
942 
943 #ifdef INET
944 	if (sc->sc_naddrs) {
945 		struct ip *ip;
946 
947 		m = m_gethdr(M_NOWAIT, MT_DATA);
948 		if (m == NULL) {
949 			CARPSTATS_INC(carps_onomem);
950 			goto resched;
951 		}
952 		len = sizeof(*ip) + sizeof(ch);
953 		m->m_pkthdr.len = len;
954 		m->m_pkthdr.rcvif = NULL;
955 		m->m_len = len;
956 		M_ALIGN(m, m->m_len);
957 		m->m_flags |= M_MCAST;
958 		ip = mtod(m, struct ip *);
959 		ip->ip_v = IPVERSION;
960 		ip->ip_hl = sizeof(*ip) >> 2;
961 		ip->ip_tos = V_carp_dscp << IPTOS_DSCP_OFFSET;
962 		ip->ip_len = htons(len);
963 		ip->ip_off = htons(IP_DF);
964 		ip->ip_ttl = CARP_DFLTTL;
965 		ip->ip_p = IPPROTO_CARP;
966 		ip->ip_sum = 0;
967 		ip_fillid(ip);
968 
969 		ifa = carp_best_ifa(AF_INET, sc->sc_carpdev);
970 		if (ifa != NULL) {
971 			ip->ip_src.s_addr =
972 			    ifatoia(ifa)->ia_addr.sin_addr.s_addr;
973 			ifa_free(ifa);
974 		} else
975 			ip->ip_src.s_addr = 0;
976 		ip->ip_dst.s_addr = htonl(INADDR_CARP_GROUP);
977 
978 		ch_ptr = (struct carp_header *)(&ip[1]);
979 		bcopy(&ch, ch_ptr, sizeof(ch));
980 		if (carp_prepare_ad(m, sc, ch_ptr))
981 			goto resched;
982 
983 		m->m_data += sizeof(*ip);
984 		ch_ptr->carp_cksum = in_cksum(m, len - sizeof(*ip));
985 		m->m_data -= sizeof(*ip);
986 
987 		CARPSTATS_INC(carps_opackets);
988 
989 		carp_send_ad_error(sc, ip_output(m, NULL, NULL, IP_RAWOUTPUT,
990 		    &sc->sc_carpdev->if_carp->cif_imo, NULL));
991 	}
992 #endif /* INET */
993 #ifdef INET6
994 	if (sc->sc_naddrs6) {
995 		struct ip6_hdr *ip6;
996 
997 		m = m_gethdr(M_NOWAIT, MT_DATA);
998 		if (m == NULL) {
999 			CARPSTATS_INC(carps_onomem);
1000 			goto resched;
1001 		}
1002 		len = sizeof(*ip6) + sizeof(ch);
1003 		m->m_pkthdr.len = len;
1004 		m->m_pkthdr.rcvif = NULL;
1005 		m->m_len = len;
1006 		M_ALIGN(m, m->m_len);
1007 		m->m_flags |= M_MCAST;
1008 		ip6 = mtod(m, struct ip6_hdr *);
1009 		bzero(ip6, sizeof(*ip6));
1010 		ip6->ip6_vfc |= IPV6_VERSION;
1011 		/* Traffic class isn't defined in ip6 struct instead
1012 		 * it gets offset into flowid field */
1013 		ip6->ip6_flow |= htonl(V_carp_dscp << (IPV6_FLOWLABEL_LEN +
1014 		    IPTOS_DSCP_OFFSET));
1015 		ip6->ip6_hlim = CARP_DFLTTL;
1016 		ip6->ip6_nxt = IPPROTO_CARP;
1017 
1018 		/* set the source address */
1019 		ifa = carp_best_ifa(AF_INET6, sc->sc_carpdev);
1020 		if (ifa != NULL) {
1021 			bcopy(IFA_IN6(ifa), &ip6->ip6_src,
1022 			    sizeof(struct in6_addr));
1023 			ifa_free(ifa);
1024 		} else
1025 			/* This should never happen with IPv6. */
1026 			bzero(&ip6->ip6_src, sizeof(struct in6_addr));
1027 
1028 		/* Set the multicast destination. */
1029 		ip6->ip6_dst.s6_addr16[0] = htons(0xff02);
1030 		ip6->ip6_dst.s6_addr8[15] = 0x12;
1031 		if (in6_setscope(&ip6->ip6_dst, sc->sc_carpdev, NULL) != 0) {
1032 			m_freem(m);
1033 			CARP_DEBUG("%s: in6_setscope failed\n", __func__);
1034 			goto resched;
1035 		}
1036 
1037 		ch_ptr = (struct carp_header *)(&ip6[1]);
1038 		bcopy(&ch, ch_ptr, sizeof(ch));
1039 		if (carp_prepare_ad(m, sc, ch_ptr))
1040 			goto resched;
1041 
1042 		m->m_data += sizeof(*ip6);
1043 		ch_ptr->carp_cksum = in_cksum(m, len - sizeof(*ip6));
1044 		m->m_data -= sizeof(*ip6);
1045 
1046 		CARPSTATS_INC(carps_opackets6);
1047 
1048 		carp_send_ad_error(sc, ip6_output(m, NULL, NULL, 0,
1049 		    &sc->sc_carpdev->if_carp->cif_im6o, NULL, NULL));
1050 	}
1051 #endif /* INET6 */
1052 
1053 resched:
1054 	callout_reset(&sc->sc_ad_tmo, tvtohz(&tv), carp_send_ad, sc);
1055 }
1056 
1057 static void
1058 carp_addroute(struct carp_softc *sc)
1059 {
1060 	struct ifaddr *ifa;
1061 
1062 	CARP_FOREACH_IFA(sc, ifa)
1063 		carp_ifa_addroute(ifa);
1064 }
1065 
1066 static void
1067 carp_ifa_addroute(struct ifaddr *ifa)
1068 {
1069 
1070 	switch (ifa->ifa_addr->sa_family) {
1071 #ifdef INET
1072 	case AF_INET:
1073 		in_addprefix(ifatoia(ifa));
1074 		ifa_add_loopback_route(ifa,
1075 		    (struct sockaddr *)&ifatoia(ifa)->ia_addr);
1076 		break;
1077 #endif
1078 #ifdef INET6
1079 	case AF_INET6:
1080 		ifa_add_loopback_route(ifa,
1081 		    (struct sockaddr *)&ifatoia6(ifa)->ia_addr);
1082 		nd6_add_ifa_lle(ifatoia6(ifa));
1083 		break;
1084 #endif
1085 	}
1086 }
1087 
1088 static void
1089 carp_delroute(struct carp_softc *sc)
1090 {
1091 	struct ifaddr *ifa;
1092 
1093 	CARP_FOREACH_IFA(sc, ifa)
1094 		carp_ifa_delroute(ifa);
1095 }
1096 
1097 static void
1098 carp_ifa_delroute(struct ifaddr *ifa)
1099 {
1100 
1101 	switch (ifa->ifa_addr->sa_family) {
1102 #ifdef INET
1103 	case AF_INET:
1104 		ifa_del_loopback_route(ifa,
1105 		    (struct sockaddr *)&ifatoia(ifa)->ia_addr);
1106 		in_scrubprefix(ifatoia(ifa), LLE_STATIC);
1107 		break;
1108 #endif
1109 #ifdef INET6
1110 	case AF_INET6:
1111 		ifa_del_loopback_route(ifa,
1112 		    (struct sockaddr *)&ifatoia6(ifa)->ia_addr);
1113 		nd6_rem_ifa_lle(ifatoia6(ifa), 1);
1114 		break;
1115 #endif
1116 	}
1117 }
1118 
1119 int
1120 carp_master(struct ifaddr *ifa)
1121 {
1122 	struct carp_softc *sc = ifa->ifa_carp;
1123 
1124 	return (sc->sc_state == MASTER);
1125 }
1126 
1127 #ifdef INET
1128 /*
1129  * Broadcast a gratuitous ARP request containing
1130  * the virtual router MAC address for each IP address
1131  * associated with the virtual router.
1132  */
1133 static void
1134 carp_send_arp(struct carp_softc *sc)
1135 {
1136 	struct ifaddr *ifa;
1137 	struct in_addr addr;
1138 
1139 	NET_EPOCH_ASSERT();
1140 
1141 	CARP_FOREACH_IFA(sc, ifa) {
1142 		if (ifa->ifa_addr->sa_family != AF_INET)
1143 			continue;
1144 		addr = ((struct sockaddr_in *)ifa->ifa_addr)->sin_addr;
1145 		arp_announce_ifaddr(sc->sc_carpdev, addr, LLADDR(&sc->sc_addr));
1146 	}
1147 }
1148 
1149 int
1150 carp_iamatch(struct ifaddr *ifa, uint8_t **enaddr)
1151 {
1152 	struct carp_softc *sc = ifa->ifa_carp;
1153 
1154 	if (sc->sc_state == MASTER) {
1155 		*enaddr = LLADDR(&sc->sc_addr);
1156 		return (1);
1157 	}
1158 
1159 	return (0);
1160 }
1161 #endif
1162 
1163 #ifdef INET6
1164 static void
1165 carp_send_na(struct carp_softc *sc)
1166 {
1167 	static struct in6_addr mcast = IN6ADDR_LINKLOCAL_ALLNODES_INIT;
1168 	struct ifaddr *ifa;
1169 	struct in6_addr *in6;
1170 
1171 	CARP_FOREACH_IFA(sc, ifa) {
1172 		if (ifa->ifa_addr->sa_family != AF_INET6)
1173 			continue;
1174 
1175 		in6 = IFA_IN6(ifa);
1176 		nd6_na_output(sc->sc_carpdev, &mcast, in6,
1177 		    ND_NA_FLAG_OVERRIDE, 1, NULL);
1178 		DELAY(1000);	/* XXX */
1179 	}
1180 }
1181 
1182 /*
1183  * Returns ifa in case it's a carp address and it is MASTER, or if the address
1184  * matches and is not a carp address.  Returns NULL otherwise.
1185  */
1186 struct ifaddr *
1187 carp_iamatch6(struct ifnet *ifp, struct in6_addr *taddr)
1188 {
1189 	struct ifaddr *ifa;
1190 
1191 	NET_EPOCH_ASSERT();
1192 
1193 	ifa = NULL;
1194 	CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1195 		if (ifa->ifa_addr->sa_family != AF_INET6)
1196 			continue;
1197 		if (!IN6_ARE_ADDR_EQUAL(taddr, IFA_IN6(ifa)))
1198 			continue;
1199 		if (ifa->ifa_carp && ifa->ifa_carp->sc_state != MASTER)
1200 			ifa = NULL;
1201 		else
1202 			ifa_ref(ifa);
1203 		break;
1204 	}
1205 
1206 	return (ifa);
1207 }
1208 
1209 char *
1210 carp_macmatch6(struct ifnet *ifp, struct mbuf *m, const struct in6_addr *taddr)
1211 {
1212 	struct ifaddr *ifa;
1213 
1214 	NET_EPOCH_ASSERT();
1215 
1216 	IFNET_FOREACH_IFA(ifp, ifa)
1217 		if (ifa->ifa_addr->sa_family == AF_INET6 &&
1218 		    IN6_ARE_ADDR_EQUAL(taddr, IFA_IN6(ifa))) {
1219 			struct carp_softc *sc = ifa->ifa_carp;
1220 			struct m_tag *mtag;
1221 
1222 			mtag = m_tag_get(PACKET_TAG_CARP,
1223 			    sizeof(struct carp_softc *), M_NOWAIT);
1224 			if (mtag == NULL)
1225 				/* Better a bit than nothing. */
1226 				return (LLADDR(&sc->sc_addr));
1227 
1228 			bcopy(&sc, mtag + 1, sizeof(sc));
1229 			m_tag_prepend(m, mtag);
1230 
1231 			return (LLADDR(&sc->sc_addr));
1232 		}
1233 
1234 	return (NULL);
1235 }
1236 #endif /* INET6 */
1237 
1238 int
1239 carp_forus(struct ifnet *ifp, u_char *dhost)
1240 {
1241 	struct carp_softc *sc;
1242 	uint8_t *ena = dhost;
1243 
1244 	if (ena[0] || ena[1] || ena[2] != 0x5e || ena[3] || ena[4] != 1)
1245 		return (0);
1246 
1247 	CIF_LOCK(ifp->if_carp);
1248 	IFNET_FOREACH_CARP(ifp, sc) {
1249 		/*
1250 		 * CARP_LOCK() is not here, since would protect nothing, but
1251 		 * cause deadlock with if_bridge, calling this under its lock.
1252 		 */
1253 		if (sc->sc_state == MASTER && !bcmp(dhost, LLADDR(&sc->sc_addr),
1254 		    ETHER_ADDR_LEN)) {
1255 			CIF_UNLOCK(ifp->if_carp);
1256 			return (1);
1257 		}
1258 	}
1259 	CIF_UNLOCK(ifp->if_carp);
1260 
1261 	return (0);
1262 }
1263 
1264 /* Master down timeout event, executed in callout context. */
1265 static void
1266 carp_master_down(void *v)
1267 {
1268 	struct carp_softc *sc = v;
1269 	struct epoch_tracker et;
1270 
1271 	NET_EPOCH_ENTER(et);
1272 	CARP_LOCK_ASSERT(sc);
1273 
1274 	CURVNET_SET(sc->sc_carpdev->if_vnet);
1275 	if (sc->sc_state == BACKUP) {
1276 		carp_master_down_locked(sc, "master timed out");
1277 	}
1278 	CURVNET_RESTORE();
1279 
1280 	CARP_UNLOCK(sc);
1281 	NET_EPOCH_EXIT(et);
1282 }
1283 
1284 static void
1285 carp_master_down_locked(struct carp_softc *sc, const char *reason)
1286 {
1287 
1288 	NET_EPOCH_ASSERT();
1289 	CARP_LOCK_ASSERT(sc);
1290 
1291 	switch (sc->sc_state) {
1292 	case BACKUP:
1293 		carp_set_state(sc, MASTER, reason);
1294 		carp_send_ad_locked(sc);
1295 #ifdef INET
1296 		carp_send_arp(sc);
1297 #endif
1298 #ifdef INET6
1299 		carp_send_na(sc);
1300 #endif
1301 		carp_setrun(sc, 0);
1302 		carp_addroute(sc);
1303 		break;
1304 	case INIT:
1305 	case MASTER:
1306 #ifdef INVARIANTS
1307 		panic("carp: VHID %u@%s: master_down event in %s state\n",
1308 		    sc->sc_vhid,
1309 		    sc->sc_carpdev->if_xname,
1310 		    sc->sc_state ? "MASTER" : "INIT");
1311 #endif
1312 		break;
1313 	}
1314 }
1315 
1316 /*
1317  * When in backup state, af indicates whether to reset the master down timer
1318  * for v4 or v6. If it's set to zero, reset the ones which are already pending.
1319  */
1320 static void
1321 carp_setrun(struct carp_softc *sc, sa_family_t af)
1322 {
1323 	struct timeval tv;
1324 
1325 	CARP_LOCK_ASSERT(sc);
1326 
1327 	if ((sc->sc_carpdev->if_flags & IFF_UP) == 0 ||
1328 	    sc->sc_carpdev->if_link_state != LINK_STATE_UP ||
1329 	    (sc->sc_naddrs == 0 && sc->sc_naddrs6 == 0) ||
1330 	    !V_carp_allow)
1331 		return;
1332 
1333 	switch (sc->sc_state) {
1334 	case INIT:
1335 		carp_set_state(sc, BACKUP, "initialization complete");
1336 		carp_setrun(sc, 0);
1337 		break;
1338 	case BACKUP:
1339 		callout_stop(&sc->sc_ad_tmo);
1340 		tv.tv_sec = 3 * sc->sc_advbase;
1341 		tv.tv_usec = sc->sc_advskew * 1000000 / 256;
1342 		switch (af) {
1343 #ifdef INET
1344 		case AF_INET:
1345 			callout_reset(&sc->sc_md_tmo, tvtohz(&tv),
1346 			    carp_master_down, sc);
1347 			break;
1348 #endif
1349 #ifdef INET6
1350 		case AF_INET6:
1351 			callout_reset(&sc->sc_md6_tmo, tvtohz(&tv),
1352 			    carp_master_down, sc);
1353 			break;
1354 #endif
1355 		default:
1356 #ifdef INET
1357 			if (sc->sc_naddrs)
1358 				callout_reset(&sc->sc_md_tmo, tvtohz(&tv),
1359 				    carp_master_down, sc);
1360 #endif
1361 #ifdef INET6
1362 			if (sc->sc_naddrs6)
1363 				callout_reset(&sc->sc_md6_tmo, tvtohz(&tv),
1364 				    carp_master_down, sc);
1365 #endif
1366 			break;
1367 		}
1368 		break;
1369 	case MASTER:
1370 		tv.tv_sec = sc->sc_advbase;
1371 		tv.tv_usec = sc->sc_advskew * 1000000 / 256;
1372 		callout_reset(&sc->sc_ad_tmo, tvtohz(&tv),
1373 		    carp_send_ad, sc);
1374 		break;
1375 	}
1376 }
1377 
1378 /*
1379  * Setup multicast structures.
1380  */
1381 static int
1382 carp_multicast_setup(struct carp_if *cif, sa_family_t sa)
1383 {
1384 	struct ifnet *ifp = cif->cif_ifp;
1385 	int error = 0;
1386 
1387 	switch (sa) {
1388 #ifdef INET
1389 	case AF_INET:
1390 	    {
1391 		struct ip_moptions *imo = &cif->cif_imo;
1392 		struct in_mfilter *imf;
1393 		struct in_addr addr;
1394 
1395 		if (ip_mfilter_first(&imo->imo_head) != NULL)
1396 			return (0);
1397 
1398 		imf = ip_mfilter_alloc(M_WAITOK, 0, 0);
1399 		ip_mfilter_init(&imo->imo_head);
1400 		imo->imo_multicast_vif = -1;
1401 
1402 		addr.s_addr = htonl(INADDR_CARP_GROUP);
1403 		if ((error = in_joingroup(ifp, &addr, NULL,
1404 		    &imf->imf_inm)) != 0) {
1405 			ip_mfilter_free(imf);
1406 			break;
1407 		}
1408 
1409 		ip_mfilter_insert(&imo->imo_head, imf);
1410 		imo->imo_multicast_ifp = ifp;
1411 		imo->imo_multicast_ttl = CARP_DFLTTL;
1412 		imo->imo_multicast_loop = 0;
1413 		break;
1414 	   }
1415 #endif
1416 #ifdef INET6
1417 	case AF_INET6:
1418 	    {
1419 		struct ip6_moptions *im6o = &cif->cif_im6o;
1420 		struct in6_mfilter *im6f[2];
1421 		struct in6_addr in6;
1422 
1423 		if (ip6_mfilter_first(&im6o->im6o_head))
1424 			return (0);
1425 
1426 		im6f[0] = ip6_mfilter_alloc(M_WAITOK, 0, 0);
1427 		im6f[1] = ip6_mfilter_alloc(M_WAITOK, 0, 0);
1428 
1429 		ip6_mfilter_init(&im6o->im6o_head);
1430 		im6o->im6o_multicast_hlim = CARP_DFLTTL;
1431 		im6o->im6o_multicast_ifp = ifp;
1432 
1433 		/* Join IPv6 CARP multicast group. */
1434 		bzero(&in6, sizeof(in6));
1435 		in6.s6_addr16[0] = htons(0xff02);
1436 		in6.s6_addr8[15] = 0x12;
1437 		if ((error = in6_setscope(&in6, ifp, NULL)) != 0) {
1438 			ip6_mfilter_free(im6f[0]);
1439 			ip6_mfilter_free(im6f[1]);
1440 			break;
1441 		}
1442 		if ((error = in6_joingroup(ifp, &in6, NULL, &im6f[0]->im6f_in6m, 0)) != 0) {
1443 			ip6_mfilter_free(im6f[0]);
1444 			ip6_mfilter_free(im6f[1]);
1445 			break;
1446 		}
1447 
1448 		/* Join solicited multicast address. */
1449 		bzero(&in6, sizeof(in6));
1450 		in6.s6_addr16[0] = htons(0xff02);
1451 		in6.s6_addr32[1] = 0;
1452 		in6.s6_addr32[2] = htonl(1);
1453 		in6.s6_addr32[3] = 0;
1454 		in6.s6_addr8[12] = 0xff;
1455 
1456 		if ((error = in6_setscope(&in6, ifp, NULL)) != 0) {
1457 			ip6_mfilter_free(im6f[0]);
1458 			ip6_mfilter_free(im6f[1]);
1459 			break;
1460 		}
1461 
1462 		if ((error = in6_joingroup(ifp, &in6, NULL, &im6f[1]->im6f_in6m, 0)) != 0) {
1463 			in6_leavegroup(im6f[0]->im6f_in6m, NULL);
1464 			ip6_mfilter_free(im6f[0]);
1465 			ip6_mfilter_free(im6f[1]);
1466 			break;
1467 		}
1468 		ip6_mfilter_insert(&im6o->im6o_head, im6f[0]);
1469 		ip6_mfilter_insert(&im6o->im6o_head, im6f[1]);
1470 		break;
1471 	    }
1472 #endif
1473 	}
1474 
1475 	return (error);
1476 }
1477 
1478 /*
1479  * Free multicast structures.
1480  */
1481 static void
1482 carp_multicast_cleanup(struct carp_if *cif, sa_family_t sa)
1483 {
1484 #ifdef INET
1485 	struct ip_moptions *imo = &cif->cif_imo;
1486 	struct in_mfilter *imf;
1487 #endif
1488 #ifdef INET6
1489 	struct ip6_moptions *im6o = &cif->cif_im6o;
1490 	struct in6_mfilter *im6f;
1491 #endif
1492 	sx_assert(&carp_sx, SA_XLOCKED);
1493 
1494 	switch (sa) {
1495 #ifdef INET
1496 	case AF_INET:
1497 		if (cif->cif_naddrs != 0)
1498 			break;
1499 
1500 		while ((imf = ip_mfilter_first(&imo->imo_head)) != NULL) {
1501 			ip_mfilter_remove(&imo->imo_head, imf);
1502 			in_leavegroup(imf->imf_inm, NULL);
1503 			ip_mfilter_free(imf);
1504 		}
1505 		break;
1506 #endif
1507 #ifdef INET6
1508 	case AF_INET6:
1509 		if (cif->cif_naddrs6 != 0)
1510 			break;
1511 
1512 		while ((im6f = ip6_mfilter_first(&im6o->im6o_head)) != NULL) {
1513 			ip6_mfilter_remove(&im6o->im6o_head, im6f);
1514 			in6_leavegroup(im6f->im6f_in6m, NULL);
1515 			ip6_mfilter_free(im6f);
1516 		}
1517 		break;
1518 #endif
1519 	}
1520 }
1521 
1522 int
1523 carp_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *sa)
1524 {
1525 	struct m_tag *mtag;
1526 	struct carp_softc *sc;
1527 
1528 	if (!sa)
1529 		return (0);
1530 
1531 	switch (sa->sa_family) {
1532 #ifdef INET
1533 	case AF_INET:
1534 		break;
1535 #endif
1536 #ifdef INET6
1537 	case AF_INET6:
1538 		break;
1539 #endif
1540 	default:
1541 		return (0);
1542 	}
1543 
1544 	mtag = m_tag_find(m, PACKET_TAG_CARP, NULL);
1545 	if (mtag == NULL)
1546 		return (0);
1547 
1548 	bcopy(mtag + 1, &sc, sizeof(sc));
1549 
1550 	/* Set the source MAC address to the Virtual Router MAC Address. */
1551 	switch (ifp->if_type) {
1552 	case IFT_ETHER:
1553 	case IFT_BRIDGE:
1554 	case IFT_L2VLAN: {
1555 			struct ether_header *eh;
1556 
1557 			eh = mtod(m, struct ether_header *);
1558 			eh->ether_shost[0] = 0;
1559 			eh->ether_shost[1] = 0;
1560 			eh->ether_shost[2] = 0x5e;
1561 			eh->ether_shost[3] = 0;
1562 			eh->ether_shost[4] = 1;
1563 			eh->ether_shost[5] = sc->sc_vhid;
1564 		}
1565 		break;
1566 	default:
1567 		printf("%s: carp is not supported for the %d interface type\n",
1568 		    ifp->if_xname, ifp->if_type);
1569 		return (EOPNOTSUPP);
1570 	}
1571 
1572 	return (0);
1573 }
1574 
1575 static struct carp_softc*
1576 carp_alloc(struct ifnet *ifp)
1577 {
1578 	struct carp_softc *sc;
1579 	struct carp_if *cif;
1580 
1581 	sx_assert(&carp_sx, SA_XLOCKED);
1582 
1583 	if ((cif = ifp->if_carp) == NULL)
1584 		cif = carp_alloc_if(ifp);
1585 
1586 	sc = malloc(sizeof(*sc), M_CARP, M_WAITOK|M_ZERO);
1587 
1588 	sc->sc_advbase = CARP_DFLTINTV;
1589 	sc->sc_vhid = -1;	/* required setting */
1590 	sc->sc_init_counter = 1;
1591 	sc->sc_state = INIT;
1592 
1593 	sc->sc_ifasiz = sizeof(struct ifaddr *);
1594 	sc->sc_ifas = malloc(sc->sc_ifasiz, M_CARP, M_WAITOK|M_ZERO);
1595 	sc->sc_carpdev = ifp;
1596 
1597 	CARP_LOCK_INIT(sc);
1598 #ifdef INET
1599 	callout_init_mtx(&sc->sc_md_tmo, &sc->sc_mtx, CALLOUT_RETURNUNLOCKED);
1600 #endif
1601 #ifdef INET6
1602 	callout_init_mtx(&sc->sc_md6_tmo, &sc->sc_mtx, CALLOUT_RETURNUNLOCKED);
1603 #endif
1604 	callout_init_mtx(&sc->sc_ad_tmo, &sc->sc_mtx, CALLOUT_RETURNUNLOCKED);
1605 
1606 	CIF_LOCK(cif);
1607 	TAILQ_INSERT_TAIL(&cif->cif_vrs, sc, sc_list);
1608 	CIF_UNLOCK(cif);
1609 
1610 	mtx_lock(&carp_mtx);
1611 	LIST_INSERT_HEAD(&carp_list, sc, sc_next);
1612 	mtx_unlock(&carp_mtx);
1613 
1614 	return (sc);
1615 }
1616 
1617 static void
1618 carp_grow_ifas(struct carp_softc *sc)
1619 {
1620 	struct ifaddr **new;
1621 
1622 	new = malloc(sc->sc_ifasiz * 2, M_CARP, M_WAITOK | M_ZERO);
1623 	CARP_LOCK(sc);
1624 	bcopy(sc->sc_ifas, new, sc->sc_ifasiz);
1625 	free(sc->sc_ifas, M_CARP);
1626 	sc->sc_ifas = new;
1627 	sc->sc_ifasiz *= 2;
1628 	CARP_UNLOCK(sc);
1629 }
1630 
1631 static void
1632 carp_destroy(struct carp_softc *sc)
1633 {
1634 	struct ifnet *ifp = sc->sc_carpdev;
1635 	struct carp_if *cif = ifp->if_carp;
1636 
1637 	sx_assert(&carp_sx, SA_XLOCKED);
1638 
1639 	if (sc->sc_suppress)
1640 		carp_demote_adj(-V_carp_ifdown_adj, "vhid removed");
1641 	CARP_UNLOCK(sc);
1642 
1643 	CIF_LOCK(cif);
1644 	TAILQ_REMOVE(&cif->cif_vrs, sc, sc_list);
1645 	CIF_UNLOCK(cif);
1646 
1647 	mtx_lock(&carp_mtx);
1648 	LIST_REMOVE(sc, sc_next);
1649 	mtx_unlock(&carp_mtx);
1650 
1651 	callout_drain(&sc->sc_ad_tmo);
1652 #ifdef INET
1653 	callout_drain(&sc->sc_md_tmo);
1654 #endif
1655 #ifdef INET6
1656 	callout_drain(&sc->sc_md6_tmo);
1657 #endif
1658 	CARP_LOCK_DESTROY(sc);
1659 
1660 	free(sc->sc_ifas, M_CARP);
1661 	free(sc, M_CARP);
1662 }
1663 
1664 static struct carp_if*
1665 carp_alloc_if(struct ifnet *ifp)
1666 {
1667 	struct carp_if *cif;
1668 	int error;
1669 
1670 	cif = malloc(sizeof(*cif), M_CARP, M_WAITOK|M_ZERO);
1671 
1672 	if ((error = ifpromisc(ifp, 1)) != 0)
1673 		printf("%s: ifpromisc(%s) failed: %d\n",
1674 		    __func__, ifp->if_xname, error);
1675 	else
1676 		cif->cif_flags |= CIF_PROMISC;
1677 
1678 	CIF_LOCK_INIT(cif);
1679 	cif->cif_ifp = ifp;
1680 	TAILQ_INIT(&cif->cif_vrs);
1681 
1682 	IF_ADDR_WLOCK(ifp);
1683 	ifp->if_carp = cif;
1684 	if_ref(ifp);
1685 	IF_ADDR_WUNLOCK(ifp);
1686 
1687 	return (cif);
1688 }
1689 
1690 static void
1691 carp_free_if(struct carp_if *cif)
1692 {
1693 	struct ifnet *ifp = cif->cif_ifp;
1694 
1695 	CIF_LOCK_ASSERT(cif);
1696 	KASSERT(TAILQ_EMPTY(&cif->cif_vrs), ("%s: softc list not empty",
1697 	    __func__));
1698 
1699 	IF_ADDR_WLOCK(ifp);
1700 	ifp->if_carp = NULL;
1701 	IF_ADDR_WUNLOCK(ifp);
1702 
1703 	CIF_LOCK_DESTROY(cif);
1704 
1705 	if (cif->cif_flags & CIF_PROMISC)
1706 		ifpromisc(ifp, 0);
1707 	if_rele(ifp);
1708 
1709 	free(cif, M_CARP);
1710 }
1711 
1712 static void
1713 carp_carprcp(struct carpreq *carpr, struct carp_softc *sc, int priv)
1714 {
1715 
1716 	CARP_LOCK(sc);
1717 	carpr->carpr_state = sc->sc_state;
1718 	carpr->carpr_vhid = sc->sc_vhid;
1719 	carpr->carpr_advbase = sc->sc_advbase;
1720 	carpr->carpr_advskew = sc->sc_advskew;
1721 	if (priv)
1722 		bcopy(sc->sc_key, carpr->carpr_key, sizeof(carpr->carpr_key));
1723 	else
1724 		bzero(carpr->carpr_key, sizeof(carpr->carpr_key));
1725 	CARP_UNLOCK(sc);
1726 }
1727 
1728 int
1729 carp_ioctl(struct ifreq *ifr, u_long cmd, struct thread *td)
1730 {
1731 	struct carpreq carpr;
1732 	struct ifnet *ifp;
1733 	struct carp_softc *sc = NULL;
1734 	int error = 0, locked = 0;
1735 
1736 	if ((error = copyin(ifr_data_get_ptr(ifr), &carpr, sizeof carpr)))
1737 		return (error);
1738 
1739 	ifp = ifunit_ref(ifr->ifr_name);
1740 	if (ifp == NULL)
1741 		return (ENXIO);
1742 
1743 	switch (ifp->if_type) {
1744 	case IFT_ETHER:
1745 	case IFT_L2VLAN:
1746 	case IFT_BRIDGE:
1747 		break;
1748 	default:
1749 		error = EOPNOTSUPP;
1750 		goto out;
1751 	}
1752 
1753 	if ((ifp->if_flags & IFF_MULTICAST) == 0) {
1754 		error = EADDRNOTAVAIL;
1755 		goto out;
1756 	}
1757 
1758 	sx_xlock(&carp_sx);
1759 	switch (cmd) {
1760 	case SIOCSVH:
1761 		if ((error = priv_check(td, PRIV_NETINET_CARP)))
1762 			break;
1763 		if (carpr.carpr_vhid <= 0 || carpr.carpr_vhid > CARP_MAXVHID ||
1764 		    carpr.carpr_advbase < 0 || carpr.carpr_advskew < 0) {
1765 			error = EINVAL;
1766 			break;
1767 		}
1768 
1769 		if (ifp->if_carp) {
1770 			IFNET_FOREACH_CARP(ifp, sc)
1771 				if (sc->sc_vhid == carpr.carpr_vhid)
1772 					break;
1773 		}
1774 		if (sc == NULL) {
1775 			sc = carp_alloc(ifp);
1776 			CARP_LOCK(sc);
1777 			sc->sc_vhid = carpr.carpr_vhid;
1778 			LLADDR(&sc->sc_addr)[0] = 0;
1779 			LLADDR(&sc->sc_addr)[1] = 0;
1780 			LLADDR(&sc->sc_addr)[2] = 0x5e;
1781 			LLADDR(&sc->sc_addr)[3] = 0;
1782 			LLADDR(&sc->sc_addr)[4] = 1;
1783 			LLADDR(&sc->sc_addr)[5] = sc->sc_vhid;
1784 		} else
1785 			CARP_LOCK(sc);
1786 		locked = 1;
1787 		if (carpr.carpr_advbase > 0) {
1788 			if (carpr.carpr_advbase > 255 ||
1789 			    carpr.carpr_advbase < CARP_DFLTINTV) {
1790 				error = EINVAL;
1791 				break;
1792 			}
1793 			sc->sc_advbase = carpr.carpr_advbase;
1794 		}
1795 		if (carpr.carpr_advskew >= 255) {
1796 			error = EINVAL;
1797 			break;
1798 		}
1799 		sc->sc_advskew = carpr.carpr_advskew;
1800 		if (carpr.carpr_key[0] != '\0') {
1801 			bcopy(carpr.carpr_key, sc->sc_key, sizeof(sc->sc_key));
1802 			carp_hmac_prepare(sc);
1803 		}
1804 		if (sc->sc_state != INIT &&
1805 		    carpr.carpr_state != sc->sc_state) {
1806 			switch (carpr.carpr_state) {
1807 			case BACKUP:
1808 				callout_stop(&sc->sc_ad_tmo);
1809 				carp_set_state(sc, BACKUP,
1810 				    "user requested via ifconfig");
1811 				carp_setrun(sc, 0);
1812 				carp_delroute(sc);
1813 				break;
1814 			case MASTER:
1815 				carp_master_down_locked(sc,
1816 				    "user requested via ifconfig");
1817 				break;
1818 			default:
1819 				break;
1820 			}
1821 		}
1822 		break;
1823 
1824 	case SIOCGVH:
1825 	    {
1826 		int priveleged;
1827 
1828 		if (carpr.carpr_vhid < 0 || carpr.carpr_vhid > CARP_MAXVHID) {
1829 			error = EINVAL;
1830 			break;
1831 		}
1832 		if (carpr.carpr_count < 1) {
1833 			error = EMSGSIZE;
1834 			break;
1835 		}
1836 		if (ifp->if_carp == NULL) {
1837 			error = ENOENT;
1838 			break;
1839 		}
1840 
1841 		priveleged = (priv_check(td, PRIV_NETINET_CARP) == 0);
1842 		if (carpr.carpr_vhid != 0) {
1843 			IFNET_FOREACH_CARP(ifp, sc)
1844 				if (sc->sc_vhid == carpr.carpr_vhid)
1845 					break;
1846 			if (sc == NULL) {
1847 				error = ENOENT;
1848 				break;
1849 			}
1850 			carp_carprcp(&carpr, sc, priveleged);
1851 			error = copyout(&carpr, ifr_data_get_ptr(ifr),
1852 			    sizeof(carpr));
1853 		} else  {
1854 			int i, count;
1855 
1856 			count = 0;
1857 			IFNET_FOREACH_CARP(ifp, sc)
1858 				count++;
1859 
1860 			if (count > carpr.carpr_count) {
1861 				CIF_UNLOCK(ifp->if_carp);
1862 				error = EMSGSIZE;
1863 				break;
1864 			}
1865 
1866 			i = 0;
1867 			IFNET_FOREACH_CARP(ifp, sc) {
1868 				carp_carprcp(&carpr, sc, priveleged);
1869 				carpr.carpr_count = count;
1870 				error = copyout(&carpr,
1871 				    (char *)ifr_data_get_ptr(ifr) +
1872 				    (i * sizeof(carpr)), sizeof(carpr));
1873 				if (error) {
1874 					CIF_UNLOCK(ifp->if_carp);
1875 					break;
1876 				}
1877 				i++;
1878 			}
1879 		}
1880 		break;
1881 	    }
1882 	default:
1883 		error = EINVAL;
1884 	}
1885 	sx_xunlock(&carp_sx);
1886 
1887 out:
1888 	if (locked)
1889 		CARP_UNLOCK(sc);
1890 	if_rele(ifp);
1891 
1892 	return (error);
1893 }
1894 
1895 static int
1896 carp_get_vhid(struct ifaddr *ifa)
1897 {
1898 
1899 	if (ifa == NULL || ifa->ifa_carp == NULL)
1900 		return (0);
1901 
1902 	return (ifa->ifa_carp->sc_vhid);
1903 }
1904 
1905 int
1906 carp_attach(struct ifaddr *ifa, int vhid)
1907 {
1908 	struct ifnet *ifp = ifa->ifa_ifp;
1909 	struct carp_if *cif = ifp->if_carp;
1910 	struct carp_softc *sc;
1911 	int index, error;
1912 
1913 	KASSERT(ifa->ifa_carp == NULL, ("%s: ifa %p attached", __func__, ifa));
1914 
1915 	switch (ifa->ifa_addr->sa_family) {
1916 #ifdef INET
1917 	case AF_INET:
1918 #endif
1919 #ifdef INET6
1920 	case AF_INET6:
1921 #endif
1922 		break;
1923 	default:
1924 		return (EPROTOTYPE);
1925 	}
1926 
1927 	sx_xlock(&carp_sx);
1928 	if (ifp->if_carp == NULL) {
1929 		sx_xunlock(&carp_sx);
1930 		return (ENOPROTOOPT);
1931 	}
1932 
1933 	IFNET_FOREACH_CARP(ifp, sc)
1934 		if (sc->sc_vhid == vhid)
1935 			break;
1936 	if (sc == NULL) {
1937 		sx_xunlock(&carp_sx);
1938 		return (ENOENT);
1939 	}
1940 
1941 	error = carp_multicast_setup(cif, ifa->ifa_addr->sa_family);
1942 	if (error) {
1943 		CIF_FREE(cif);
1944 		sx_xunlock(&carp_sx);
1945 		return (error);
1946 	}
1947 
1948 	index = sc->sc_naddrs + sc->sc_naddrs6 + 1;
1949 	if (index > sc->sc_ifasiz / sizeof(struct ifaddr *))
1950 		carp_grow_ifas(sc);
1951 
1952 	switch (ifa->ifa_addr->sa_family) {
1953 #ifdef INET
1954 	case AF_INET:
1955 		cif->cif_naddrs++;
1956 		sc->sc_naddrs++;
1957 		break;
1958 #endif
1959 #ifdef INET6
1960 	case AF_INET6:
1961 		cif->cif_naddrs6++;
1962 		sc->sc_naddrs6++;
1963 		break;
1964 #endif
1965 	}
1966 
1967 	ifa_ref(ifa);
1968 
1969 	CARP_LOCK(sc);
1970 	sc->sc_ifas[index - 1] = ifa;
1971 	ifa->ifa_carp = sc;
1972 	carp_hmac_prepare(sc);
1973 	carp_sc_state(sc);
1974 	CARP_UNLOCK(sc);
1975 
1976 	sx_xunlock(&carp_sx);
1977 
1978 	return (0);
1979 }
1980 
1981 void
1982 carp_detach(struct ifaddr *ifa, bool keep_cif)
1983 {
1984 	struct ifnet *ifp = ifa->ifa_ifp;
1985 	struct carp_if *cif = ifp->if_carp;
1986 	struct carp_softc *sc = ifa->ifa_carp;
1987 	int i, index;
1988 
1989 	KASSERT(sc != NULL, ("%s: %p not attached", __func__, ifa));
1990 
1991 	sx_xlock(&carp_sx);
1992 
1993 	CARP_LOCK(sc);
1994 	/* Shift array. */
1995 	index = sc->sc_naddrs + sc->sc_naddrs6;
1996 	for (i = 0; i < index; i++)
1997 		if (sc->sc_ifas[i] == ifa)
1998 			break;
1999 	KASSERT(i < index, ("%s: %p no backref", __func__, ifa));
2000 	for (; i < index - 1; i++)
2001 		sc->sc_ifas[i] = sc->sc_ifas[i+1];
2002 	sc->sc_ifas[index - 1] = NULL;
2003 
2004 	switch (ifa->ifa_addr->sa_family) {
2005 #ifdef INET
2006 	case AF_INET:
2007 		cif->cif_naddrs--;
2008 		sc->sc_naddrs--;
2009 		break;
2010 #endif
2011 #ifdef INET6
2012 	case AF_INET6:
2013 		cif->cif_naddrs6--;
2014 		sc->sc_naddrs6--;
2015 		break;
2016 #endif
2017 	}
2018 
2019 	carp_ifa_delroute(ifa);
2020 	carp_multicast_cleanup(cif, ifa->ifa_addr->sa_family);
2021 
2022 	ifa->ifa_carp = NULL;
2023 	ifa_free(ifa);
2024 
2025 	carp_hmac_prepare(sc);
2026 	carp_sc_state(sc);
2027 
2028 	if (!keep_cif && sc->sc_naddrs == 0 && sc->sc_naddrs6 == 0)
2029 		carp_destroy(sc);
2030 	else
2031 		CARP_UNLOCK(sc);
2032 
2033 	if (!keep_cif)
2034 		CIF_FREE(cif);
2035 
2036 	sx_xunlock(&carp_sx);
2037 }
2038 
2039 static void
2040 carp_set_state(struct carp_softc *sc, int state, const char *reason)
2041 {
2042 
2043 	CARP_LOCK_ASSERT(sc);
2044 
2045 	if (sc->sc_state != state) {
2046 		const char *carp_states[] = { CARP_STATES };
2047 		char subsys[IFNAMSIZ+5];
2048 
2049 		snprintf(subsys, IFNAMSIZ+5, "%u@%s", sc->sc_vhid,
2050 		    sc->sc_carpdev->if_xname);
2051 
2052 		CARP_LOG("%s: %s -> %s (%s)\n", subsys,
2053 		    carp_states[sc->sc_state], carp_states[state], reason);
2054 
2055 		sc->sc_state = state;
2056 
2057 		devctl_notify("CARP", subsys, carp_states[state], NULL);
2058 	}
2059 }
2060 
2061 static void
2062 carp_linkstate(struct ifnet *ifp)
2063 {
2064 	struct carp_softc *sc;
2065 
2066 	CIF_LOCK(ifp->if_carp);
2067 	IFNET_FOREACH_CARP(ifp, sc) {
2068 		CARP_LOCK(sc);
2069 		carp_sc_state(sc);
2070 		CARP_UNLOCK(sc);
2071 	}
2072 	CIF_UNLOCK(ifp->if_carp);
2073 }
2074 
2075 static void
2076 carp_sc_state(struct carp_softc *sc)
2077 {
2078 
2079 	CARP_LOCK_ASSERT(sc);
2080 
2081 	if (sc->sc_carpdev->if_link_state != LINK_STATE_UP ||
2082 	    !(sc->sc_carpdev->if_flags & IFF_UP) ||
2083 	    !V_carp_allow) {
2084 		callout_stop(&sc->sc_ad_tmo);
2085 #ifdef INET
2086 		callout_stop(&sc->sc_md_tmo);
2087 #endif
2088 #ifdef INET6
2089 		callout_stop(&sc->sc_md6_tmo);
2090 #endif
2091 		carp_set_state(sc, INIT, "hardware interface down");
2092 		carp_setrun(sc, 0);
2093 		if (!sc->sc_suppress)
2094 			carp_demote_adj(V_carp_ifdown_adj, "interface down");
2095 		sc->sc_suppress = 1;
2096 	} else {
2097 		carp_set_state(sc, INIT, "hardware interface up");
2098 		carp_setrun(sc, 0);
2099 		if (sc->sc_suppress)
2100 			carp_demote_adj(-V_carp_ifdown_adj, "interface up");
2101 		sc->sc_suppress = 0;
2102 	}
2103 }
2104 
2105 static void
2106 carp_demote_adj(int adj, char *reason)
2107 {
2108 	atomic_add_int(&V_carp_demotion, adj);
2109 	CARP_LOG("demoted by %d to %d (%s)\n", adj, V_carp_demotion, reason);
2110 	taskqueue_enqueue(taskqueue_swi, &carp_sendall_task);
2111 }
2112 
2113 static int
2114 carp_allow_sysctl(SYSCTL_HANDLER_ARGS)
2115 {
2116 	int new, error;
2117 	struct carp_softc *sc;
2118 
2119 	new = V_carp_allow;
2120 	error = sysctl_handle_int(oidp, &new, 0, req);
2121 	if (error || !req->newptr)
2122 		return (error);
2123 
2124 	if (V_carp_allow != new) {
2125 		V_carp_allow = new;
2126 
2127 		mtx_lock(&carp_mtx);
2128 		LIST_FOREACH(sc, &carp_list, sc_next) {
2129 			CARP_LOCK(sc);
2130 			if (curvnet == sc->sc_carpdev->if_vnet)
2131 				carp_sc_state(sc);
2132 			CARP_UNLOCK(sc);
2133 		}
2134 		mtx_unlock(&carp_mtx);
2135 	}
2136 
2137 	return (0);
2138 }
2139 
2140 static int
2141 carp_dscp_sysctl(SYSCTL_HANDLER_ARGS)
2142 {
2143 	int new, error;
2144 
2145 	new = V_carp_dscp;
2146 	error = sysctl_handle_int(oidp, &new, 0, req);
2147 	if (error || !req->newptr)
2148 		return (error);
2149 
2150 	if (new < 0 || new > 63)
2151 		return (EINVAL);
2152 
2153 	V_carp_dscp = new;
2154 
2155 	return (0);
2156 }
2157 
2158 static int
2159 carp_demote_adj_sysctl(SYSCTL_HANDLER_ARGS)
2160 {
2161 	int new, error;
2162 
2163 	new = V_carp_demotion;
2164 	error = sysctl_handle_int(oidp, &new, 0, req);
2165 	if (error || !req->newptr)
2166 		return (error);
2167 
2168 	carp_demote_adj(new, "sysctl");
2169 
2170 	return (0);
2171 }
2172 
2173 static void
2174 carp_mod_cleanup(void)
2175 {
2176 
2177 #ifdef INET
2178 	(void)ipproto_unregister(IPPROTO_CARP);
2179 	carp_iamatch_p = NULL;
2180 #endif
2181 #ifdef INET6
2182 	(void)ip6proto_unregister(IPPROTO_CARP);
2183 	carp_iamatch6_p = NULL;
2184 	carp_macmatch6_p = NULL;
2185 #endif
2186 	carp_ioctl_p = NULL;
2187 	carp_attach_p = NULL;
2188 	carp_detach_p = NULL;
2189 	carp_get_vhid_p = NULL;
2190 	carp_linkstate_p = NULL;
2191 	carp_forus_p = NULL;
2192 	carp_output_p = NULL;
2193 	carp_demote_adj_p = NULL;
2194 	carp_master_p = NULL;
2195 	mtx_unlock(&carp_mtx);
2196 	taskqueue_drain(taskqueue_swi, &carp_sendall_task);
2197 	mtx_destroy(&carp_mtx);
2198 	sx_destroy(&carp_sx);
2199 }
2200 
2201 static void
2202 ipcarp_sysinit(void)
2203 {
2204 
2205 	/* Load allow as tunable so to postpone carp start after module load */
2206 	TUNABLE_INT_FETCH("net.inet.carp.allow", &V_carp_allow);
2207 }
2208 VNET_SYSINIT(ip_carp, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY, ipcarp_sysinit, NULL);
2209 
2210 static int
2211 carp_mod_load(void)
2212 {
2213 	int err;
2214 
2215 	mtx_init(&carp_mtx, "carp_mtx", NULL, MTX_DEF);
2216 	sx_init(&carp_sx, "carp_sx");
2217 	LIST_INIT(&carp_list);
2218 	carp_get_vhid_p = carp_get_vhid;
2219 	carp_forus_p = carp_forus;
2220 	carp_output_p = carp_output;
2221 	carp_linkstate_p = carp_linkstate;
2222 	carp_ioctl_p = carp_ioctl;
2223 	carp_attach_p = carp_attach;
2224 	carp_detach_p = carp_detach;
2225 	carp_demote_adj_p = carp_demote_adj;
2226 	carp_master_p = carp_master;
2227 #ifdef INET6
2228 	carp_iamatch6_p = carp_iamatch6;
2229 	carp_macmatch6_p = carp_macmatch6;
2230 	err = ip6proto_register(IPPROTO_CARP, carp6_input, NULL);
2231 	if (err) {
2232 		printf("carp: error %d registering with INET6\n", err);
2233 		carp_mod_cleanup();
2234 		return (err);
2235 	}
2236 #endif
2237 #ifdef INET
2238 	carp_iamatch_p = carp_iamatch;
2239 	err = ipproto_register(IPPROTO_CARP, carp_input, NULL);
2240 	if (err) {
2241 		printf("carp: error %d registering with INET\n", err);
2242 		carp_mod_cleanup();
2243 		return (err);
2244 	}
2245 #endif
2246 	return (0);
2247 }
2248 
2249 static int
2250 carp_modevent(module_t mod, int type, void *data)
2251 {
2252 	switch (type) {
2253 	case MOD_LOAD:
2254 		return carp_mod_load();
2255 		/* NOTREACHED */
2256 	case MOD_UNLOAD:
2257 		mtx_lock(&carp_mtx);
2258 		if (LIST_EMPTY(&carp_list))
2259 			carp_mod_cleanup();
2260 		else {
2261 			mtx_unlock(&carp_mtx);
2262 			return (EBUSY);
2263 		}
2264 		break;
2265 
2266 	default:
2267 		return (EINVAL);
2268 	}
2269 
2270 	return (0);
2271 }
2272 
2273 static moduledata_t carp_mod = {
2274 	"carp",
2275 	carp_modevent,
2276 	0
2277 };
2278 
2279 DECLARE_MODULE(carp, carp_mod, SI_SUB_PROTO_DOMAIN, SI_ORDER_ANY);
2280