xref: /freebsd/sys/netipsec/ipsec.c (revision 076ad2f836d5f49dc1375f1677335a48fe0d4b82)
1 /*	$FreeBSD$	*/
2 /*	$KAME: ipsec.c,v 1.103 2001/05/24 07:14:18 sakane Exp $	*/
3 
4 /*-
5  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
6  * All rights reserved.
7  *
8  * Redistribution and use in source and binary forms, with or without
9  * modification, are permitted provided that the following conditions
10  * are met:
11  * 1. Redistributions of source code must retain the above copyright
12  *    notice, this list of conditions and the following disclaimer.
13  * 2. Redistributions in binary form must reproduce the above copyright
14  *    notice, this list of conditions and the following disclaimer in the
15  *    documentation and/or other materials provided with the distribution.
16  * 3. Neither the name of the project nor the names of its contributors
17  *    may be used to endorse or promote products derived from this software
18  *    without specific prior written permission.
19  *
20  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
21  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
22  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
23  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
24  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
25  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
26  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
27  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
28  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
29  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
30  * SUCH DAMAGE.
31  */
32 
33 /*
34  * IPsec controller part.
35  */
36 
37 #include "opt_inet.h"
38 #include "opt_inet6.h"
39 #include "opt_ipsec.h"
40 
41 #include <sys/param.h>
42 #include <sys/systm.h>
43 #include <sys/malloc.h>
44 #include <sys/mbuf.h>
45 #include <sys/domain.h>
46 #include <sys/priv.h>
47 #include <sys/protosw.h>
48 #include <sys/socket.h>
49 #include <sys/socketvar.h>
50 #include <sys/errno.h>
51 #include <sys/hhook.h>
52 #include <sys/time.h>
53 #include <sys/kernel.h>
54 #include <sys/syslog.h>
55 #include <sys/sysctl.h>
56 #include <sys/proc.h>
57 
58 #include <net/if.h>
59 #include <net/if_enc.h>
60 #include <net/if_var.h>
61 #include <net/vnet.h>
62 
63 #include <netinet/in.h>
64 #include <netinet/in_systm.h>
65 #include <netinet/ip.h>
66 #include <netinet/ip_var.h>
67 #include <netinet/in_var.h>
68 #include <netinet/udp.h>
69 #include <netinet/udp_var.h>
70 #include <netinet/tcp.h>
71 #include <netinet/udp.h>
72 
73 #include <netinet/ip6.h>
74 #ifdef INET6
75 #include <netinet6/ip6_var.h>
76 #endif
77 #include <netinet/in_pcb.h>
78 #ifdef INET6
79 #include <netinet/icmp6.h>
80 #endif
81 
82 #include <sys/types.h>
83 #include <netipsec/ipsec.h>
84 #ifdef INET6
85 #include <netipsec/ipsec6.h>
86 #endif
87 #include <netipsec/ah_var.h>
88 #include <netipsec/esp_var.h>
89 #include <netipsec/ipcomp.h>		/*XXX*/
90 #include <netipsec/ipcomp_var.h>
91 #include <netipsec/ipsec_support.h>
92 
93 #include <netipsec/key.h>
94 #include <netipsec/keydb.h>
95 #include <netipsec/key_debug.h>
96 
97 #include <netipsec/xform.h>
98 
99 #include <machine/in_cksum.h>
100 
101 #include <opencrypto/cryptodev.h>
102 
103 /* NB: name changed so netstat doesn't use it. */
104 VNET_PCPUSTAT_DEFINE(struct ipsecstat, ipsec4stat);
105 VNET_PCPUSTAT_SYSINIT(ipsec4stat);
106 
107 #ifdef VIMAGE
108 VNET_PCPUSTAT_SYSUNINIT(ipsec4stat);
109 #endif /* VIMAGE */
110 
111 VNET_DEFINE(int, ip4_ah_offsetmask) = 0;	/* maybe IP_DF? */
112 /* DF bit on encap. 0: clear 1: set 2: copy */
113 VNET_DEFINE(int, ip4_ipsec_dfbit) = 0;
114 VNET_DEFINE(int, ip4_esp_trans_deflev) = IPSEC_LEVEL_USE;
115 VNET_DEFINE(int, ip4_esp_net_deflev) = IPSEC_LEVEL_USE;
116 VNET_DEFINE(int, ip4_ah_trans_deflev) = IPSEC_LEVEL_USE;
117 VNET_DEFINE(int, ip4_ah_net_deflev) = IPSEC_LEVEL_USE;
118 /* ECN ignore(-1)/forbidden(0)/allowed(1) */
119 VNET_DEFINE(int, ip4_ipsec_ecn) = 0;
120 VNET_DEFINE(int, ip4_esp_randpad) = -1;
121 
122 static VNET_DEFINE(int, ip4_filtertunnel) = 0;
123 #define	V_ip4_filtertunnel VNET(ip4_filtertunnel)
124 static VNET_DEFINE(int, check_policy_history) = 0;
125 #define	V_check_policy_history	VNET(check_policy_history)
126 static VNET_DEFINE(struct secpolicy *, def_policy) = NULL;
127 #define	V_def_policy	VNET(def_policy)
128 static int
129 sysctl_def_policy(SYSCTL_HANDLER_ARGS)
130 {
131 	int error, value;
132 
133 	value = V_def_policy->policy;
134 	error = sysctl_handle_int(oidp, &value, 0, req);
135 	if (error == 0) {
136 		if (value != IPSEC_POLICY_DISCARD &&
137 		    value != IPSEC_POLICY_NONE)
138 			return (EINVAL);
139 		V_def_policy->policy = value;
140 	}
141 	return (error);
142 }
143 
144 /*
145  * Crypto support requirements:
146  *
147  *  1	require hardware support
148  * -1	require software support
149  *  0	take anything
150  */
151 VNET_DEFINE(int, crypto_support) = CRYPTOCAP_F_HARDWARE | CRYPTOCAP_F_SOFTWARE;
152 /*
153  * TCP/UDP checksum handling policy for transport mode NAT-T (RFC3948)
154  *
155  * 0 - auto: incrementally recompute, when checksum delta is known;
156  *     if checksum delta isn't known, reset checksum to zero for UDP,
157  *     and mark csum_flags as valid for TCP.
158  * 1 - fully recompute TCP/UDP checksum.
159  */
160 VNET_DEFINE(int, natt_cksum_policy) = 0;
161 
162 FEATURE(ipsec, "Internet Protocol Security (IPsec)");
163 FEATURE(ipsec_natt, "UDP Encapsulation of IPsec ESP Packets ('NAT-T')");
164 
165 SYSCTL_DECL(_net_inet_ipsec);
166 
167 /* net.inet.ipsec */
168 SYSCTL_PROC(_net_inet_ipsec, IPSECCTL_DEF_POLICY, def_policy,
169 	CTLTYPE_INT | CTLFLAG_VNET | CTLFLAG_RW, 0, 0, sysctl_def_policy, "I",
170 	"IPsec default policy.");
171 SYSCTL_INT(_net_inet_ipsec, IPSECCTL_DEF_ESP_TRANSLEV, esp_trans_deflev,
172 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip4_esp_trans_deflev), 0,
173 	"Default ESP transport mode level");
174 SYSCTL_INT(_net_inet_ipsec, IPSECCTL_DEF_ESP_NETLEV, esp_net_deflev,
175 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip4_esp_net_deflev), 0,
176 	"Default ESP tunnel mode level.");
177 SYSCTL_INT(_net_inet_ipsec, IPSECCTL_DEF_AH_TRANSLEV, ah_trans_deflev,
178 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip4_ah_trans_deflev), 0,
179 	"AH transfer mode default level.");
180 SYSCTL_INT(_net_inet_ipsec, IPSECCTL_DEF_AH_NETLEV, ah_net_deflev,
181 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip4_ah_net_deflev), 0,
182 	"AH tunnel mode default level.");
183 SYSCTL_INT(_net_inet_ipsec, IPSECCTL_AH_CLEARTOS, ah_cleartos,
184 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ah_cleartos), 0,
185 	"If set, clear type-of-service field when doing AH computation.");
186 SYSCTL_INT(_net_inet_ipsec, IPSECCTL_AH_OFFSETMASK, ah_offsetmask,
187 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip4_ah_offsetmask), 0,
188 	"If not set, clear offset field mask when doing AH computation.");
189 SYSCTL_INT(_net_inet_ipsec, IPSECCTL_DFBIT, dfbit,
190 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip4_ipsec_dfbit), 0,
191 	"Do not fragment bit on encap.");
192 SYSCTL_INT(_net_inet_ipsec, IPSECCTL_ECN, ecn,
193 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip4_ipsec_ecn), 0,
194 	"Explicit Congestion Notification handling.");
195 SYSCTL_INT(_net_inet_ipsec, OID_AUTO, crypto_support,
196 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(crypto_support), 0,
197 	"Crypto driver selection.");
198 SYSCTL_INT(_net_inet_ipsec, OID_AUTO, check_policy_history,
199 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(check_policy_history), 0,
200 	"Use strict check of inbound packets to security policy compliance.");
201 SYSCTL_INT(_net_inet_ipsec, OID_AUTO, natt_cksum_policy,
202 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(natt_cksum_policy), 0,
203 	"Method to fix TCP/UDP checksum for transport mode IPsec after NAT.");
204 SYSCTL_INT(_net_inet_ipsec, OID_AUTO, filtertunnel,
205 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip4_filtertunnel), 0,
206 	"If set, filter packets from an IPsec tunnel.");
207 SYSCTL_VNET_PCPUSTAT(_net_inet_ipsec, OID_AUTO, ipsecstats, struct ipsecstat,
208     ipsec4stat, "IPsec IPv4 statistics.");
209 
210 #ifdef REGRESSION
211 /*
212  * When set to 1, IPsec will send packets with the same sequence number.
213  * This allows to verify if the other side has proper replay attacks detection.
214  */
215 VNET_DEFINE(int, ipsec_replay) = 0;
216 SYSCTL_INT(_net_inet_ipsec, OID_AUTO, test_replay,
217 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ipsec_replay), 0,
218 	"Emulate replay attack");
219 /*
220  * When set 1, IPsec will send packets with corrupted HMAC.
221  * This allows to verify if the other side properly detects modified packets.
222  */
223 VNET_DEFINE(int, ipsec_integrity) = 0;
224 SYSCTL_INT(_net_inet_ipsec, OID_AUTO, test_integrity,
225 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ipsec_integrity), 0,
226 	"Emulate man-in-the-middle attack");
227 #endif
228 
229 #ifdef INET6
230 VNET_PCPUSTAT_DEFINE(struct ipsecstat, ipsec6stat);
231 VNET_PCPUSTAT_SYSINIT(ipsec6stat);
232 
233 #ifdef VIMAGE
234 VNET_PCPUSTAT_SYSUNINIT(ipsec6stat);
235 #endif /* VIMAGE */
236 
237 VNET_DEFINE(int, ip6_esp_trans_deflev) = IPSEC_LEVEL_USE;
238 VNET_DEFINE(int, ip6_esp_net_deflev) = IPSEC_LEVEL_USE;
239 VNET_DEFINE(int, ip6_ah_trans_deflev) = IPSEC_LEVEL_USE;
240 VNET_DEFINE(int, ip6_ah_net_deflev) = IPSEC_LEVEL_USE;
241 VNET_DEFINE(int, ip6_ipsec_ecn) = 0;	/* ECN ignore(-1)/forbidden(0)/allowed(1) */
242 
243 static VNET_DEFINE(int, ip6_filtertunnel) = 0;
244 #define	V_ip6_filtertunnel	VNET(ip6_filtertunnel)
245 
246 SYSCTL_DECL(_net_inet6_ipsec6);
247 
248 /* net.inet6.ipsec6 */
249 SYSCTL_PROC(_net_inet6_ipsec6, IPSECCTL_DEF_POLICY, def_policy,
250 	CTLTYPE_INT | CTLFLAG_VNET | CTLFLAG_RW, 0, 0, sysctl_def_policy, "I",
251 	"IPsec default policy.");
252 SYSCTL_INT(_net_inet6_ipsec6, IPSECCTL_DEF_ESP_TRANSLEV, esp_trans_deflev,
253 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_esp_trans_deflev), 0,
254 	"Default ESP transport mode level.");
255 SYSCTL_INT(_net_inet6_ipsec6, IPSECCTL_DEF_ESP_NETLEV, esp_net_deflev,
256 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_esp_net_deflev), 0,
257 	"Default ESP tunnel mode level.");
258 SYSCTL_INT(_net_inet6_ipsec6, IPSECCTL_DEF_AH_TRANSLEV, ah_trans_deflev,
259 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_ah_trans_deflev), 0,
260 	"AH transfer mode default level.");
261 SYSCTL_INT(_net_inet6_ipsec6, IPSECCTL_DEF_AH_NETLEV, ah_net_deflev,
262 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_ah_net_deflev), 0,
263 	"AH tunnel mode default level.");
264 SYSCTL_INT(_net_inet6_ipsec6, IPSECCTL_ECN, ecn,
265 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_ipsec_ecn), 0,
266 	"Explicit Congestion Notification handling.");
267 SYSCTL_INT(_net_inet6_ipsec6, OID_AUTO, filtertunnel,
268 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_filtertunnel),  0,
269 	"If set, filter packets from an IPsec tunnel.");
270 SYSCTL_VNET_PCPUSTAT(_net_inet6_ipsec6, IPSECCTL_STATS, ipsecstats,
271     struct ipsecstat, ipsec6stat, "IPsec IPv6 statistics.");
272 #endif /* INET6 */
273 
274 static int ipsec_in_reject(struct secpolicy *, struct inpcb *,
275     const struct mbuf *);
276 
277 #ifdef INET
278 static void ipsec4_get_ulp(const struct mbuf *, struct secpolicyindex *, int);
279 static void ipsec4_setspidx_ipaddr(const struct mbuf *,
280     struct secpolicyindex *);
281 #endif
282 #ifdef INET6
283 static void ipsec6_get_ulp(const struct mbuf *m, struct secpolicyindex *, int);
284 static void ipsec6_setspidx_ipaddr(const struct mbuf *,
285     struct secpolicyindex *);
286 #endif
287 
288 /*
289  * Return a held reference to the default SP.
290  */
291 static struct secpolicy *
292 key_allocsp_default(void)
293 {
294 
295 	key_addref(V_def_policy);
296 	return (V_def_policy);
297 }
298 
299 static void
300 ipsec_invalidate_cache(struct inpcb *inp, u_int dir)
301 {
302 	struct secpolicy *sp;
303 
304 	INP_WLOCK_ASSERT(inp);
305 	if (dir == IPSEC_DIR_OUTBOUND) {
306 		if (inp->inp_sp->flags & INP_INBOUND_POLICY)
307 			return;
308 		sp = inp->inp_sp->sp_in;
309 		inp->inp_sp->sp_in = NULL;
310 	} else {
311 		if (inp->inp_sp->flags & INP_OUTBOUND_POLICY)
312 			return;
313 		sp = inp->inp_sp->sp_out;
314 		inp->inp_sp->sp_out = NULL;
315 	}
316 	if (sp != NULL)
317 		key_freesp(&sp); /* release extra reference */
318 }
319 
320 static void
321 ipsec_cachepolicy(struct inpcb *inp, struct secpolicy *sp, u_int dir)
322 {
323 	uint32_t genid;
324 	int downgrade;
325 
326 	INP_LOCK_ASSERT(inp);
327 
328 	if (dir == IPSEC_DIR_OUTBOUND) {
329 		/* Do we have configured PCB policy? */
330 		if (inp->inp_sp->flags & INP_OUTBOUND_POLICY)
331 			return;
332 		/* Another thread has already set cached policy */
333 		if (inp->inp_sp->sp_out != NULL)
334 			return;
335 		/*
336 		 * Do not cache OUTBOUND policy if PCB isn't connected,
337 		 * i.e. foreign address is INADDR_ANY/UNSPECIFIED.
338 		 */
339 #ifdef INET
340 		if ((inp->inp_vflag & INP_IPV4) != 0 &&
341 		    inp->inp_faddr.s_addr == INADDR_ANY)
342 			return;
343 #endif
344 #ifdef INET6
345 		if ((inp->inp_vflag & INP_IPV6) != 0 &&
346 		    IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr))
347 			return;
348 #endif
349 	} else {
350 		/* Do we have configured PCB policy? */
351 		if (inp->inp_sp->flags & INP_INBOUND_POLICY)
352 			return;
353 		/* Another thread has already set cached policy */
354 		if (inp->inp_sp->sp_in != NULL)
355 			return;
356 		/*
357 		 * Do not cache INBOUND policy for listen socket,
358 		 * that is bound to INADDR_ANY/UNSPECIFIED address.
359 		 */
360 #ifdef INET
361 		if ((inp->inp_vflag & INP_IPV4) != 0 &&
362 		    inp->inp_faddr.s_addr == INADDR_ANY)
363 			return;
364 #endif
365 #ifdef INET6
366 		if ((inp->inp_vflag & INP_IPV6) != 0 &&
367 		    IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr))
368 			return;
369 #endif
370 	}
371 	downgrade = 0;
372 	if (!INP_WLOCKED(inp)) {
373 		if ((downgrade = INP_TRY_UPGRADE(inp)) == 0)
374 			return;
375 	}
376 	if (dir == IPSEC_DIR_OUTBOUND)
377 		inp->inp_sp->sp_out = sp;
378 	else
379 		inp->inp_sp->sp_in = sp;
380 	/*
381 	 * SP is already referenced by the lookup code.
382 	 * We take extra reference here to avoid race in the
383 	 * ipsec_getpcbpolicy() function - SP will not be freed in the
384 	 * time between we take SP pointer from the cache and key_addref()
385 	 * call.
386 	 */
387 	key_addref(sp);
388 	genid = key_getspgen();
389 	if (genid != inp->inp_sp->genid) {
390 		ipsec_invalidate_cache(inp, dir);
391 		inp->inp_sp->genid = genid;
392 	}
393 	KEYDBG(IPSEC_STAMP,
394 	    printf("%s: PCB(%p): cached %s SP(%p)\n",
395 	    __func__, inp, dir == IPSEC_DIR_OUTBOUND ? "OUTBOUND":
396 	    "INBOUND", sp));
397 	if (downgrade != 0)
398 		INP_DOWNGRADE(inp);
399 }
400 
401 static struct secpolicy *
402 ipsec_checkpolicy(struct secpolicy *sp, struct inpcb *inp, int *error)
403 {
404 
405 	/* Save found OUTBOUND policy into PCB SP cache. */
406 	if (inp != NULL && inp->inp_sp != NULL && inp->inp_sp->sp_out == NULL)
407 		ipsec_cachepolicy(inp, sp, IPSEC_DIR_OUTBOUND);
408 
409 	switch (sp->policy) {
410 	default:
411 		printf("%s: invalid policy %u\n", __func__, sp->policy);
412 		/* FALLTHROUGH */
413 	case IPSEC_POLICY_DISCARD:
414 		*error = -EINVAL;	/* Packet is discarded by caller. */
415 		/* FALLTHROUGH */
416 	case IPSEC_POLICY_BYPASS:
417 	case IPSEC_POLICY_NONE:
418 		key_freesp(&sp);
419 		sp = NULL;		/* NB: force NULL result. */
420 		break;
421 	case IPSEC_POLICY_IPSEC:
422 		/* XXXAE: handle LARVAL SP */
423 		break;
424 	}
425 	KEYDBG(IPSEC_DUMP,
426 	    printf("%s: get SP(%p), error %d\n", __func__, sp, *error));
427 	return (sp);
428 }
429 
430 static struct secpolicy *
431 ipsec_getpcbpolicy(struct inpcb *inp, u_int dir)
432 {
433 	struct secpolicy *sp;
434 	int flags, downgrade;
435 
436 	if (inp == NULL || inp->inp_sp == NULL)
437 		return (NULL);
438 
439 	INP_LOCK_ASSERT(inp);
440 
441 	flags = inp->inp_sp->flags;
442 	if (dir == IPSEC_DIR_OUTBOUND) {
443 		sp = inp->inp_sp->sp_out;
444 		flags &= INP_OUTBOUND_POLICY;
445 	} else {
446 		sp = inp->inp_sp->sp_in;
447 		flags &= INP_INBOUND_POLICY;
448 	}
449 	/*
450 	 * Check flags. If we have PCB SP, just return it.
451 	 * Otherwise we need to check that cached SP entry isn't stale.
452 	 */
453 	if (flags == 0) {
454 		if (sp == NULL)
455 			return (NULL);
456 		if (inp->inp_sp->genid != key_getspgen()) {
457 			/* Invalidate the cache. */
458 			downgrade = 0;
459 			if (!INP_WLOCKED(inp)) {
460 				if ((downgrade = INP_TRY_UPGRADE(inp)) == 0)
461 					return (NULL);
462 			}
463 			ipsec_invalidate_cache(inp, IPSEC_DIR_OUTBOUND);
464 			ipsec_invalidate_cache(inp, IPSEC_DIR_INBOUND);
465 			if (downgrade != 0)
466 				INP_DOWNGRADE(inp);
467 			return (NULL);
468 		}
469 		KEYDBG(IPSEC_STAMP,
470 		    printf("%s: PCB(%p): cache hit SP(%p)\n",
471 		    __func__, inp, sp));
472 		/* Return referenced cached policy */
473 	}
474 	key_addref(sp);
475 	return (sp);
476 }
477 
478 #ifdef INET
479 static void
480 ipsec4_get_ulp(const struct mbuf *m, struct secpolicyindex *spidx,
481     int needport)
482 {
483 	uint8_t nxt;
484 	int off;
485 
486 	/* Sanity check. */
487 	IPSEC_ASSERT(m->m_pkthdr.len >= sizeof(struct ip),
488 	    ("packet too short"));
489 
490 	if (m->m_len >= sizeof (struct ip)) {
491 		const struct ip *ip = mtod(m, const struct ip *);
492 		if (ip->ip_off & htons(IP_MF | IP_OFFMASK))
493 			goto done;
494 		off = ip->ip_hl << 2;
495 		nxt = ip->ip_p;
496 	} else {
497 		struct ip ih;
498 
499 		m_copydata(m, 0, sizeof (struct ip), (caddr_t) &ih);
500 		if (ih.ip_off & htons(IP_MF | IP_OFFMASK))
501 			goto done;
502 		off = ih.ip_hl << 2;
503 		nxt = ih.ip_p;
504 	}
505 
506 	while (off < m->m_pkthdr.len) {
507 		struct ip6_ext ip6e;
508 		struct tcphdr th;
509 		struct udphdr uh;
510 
511 		switch (nxt) {
512 		case IPPROTO_TCP:
513 			spidx->ul_proto = nxt;
514 			if (!needport)
515 				goto done_proto;
516 			if (off + sizeof(struct tcphdr) > m->m_pkthdr.len)
517 				goto done;
518 			m_copydata(m, off, sizeof (th), (caddr_t) &th);
519 			spidx->src.sin.sin_port = th.th_sport;
520 			spidx->dst.sin.sin_port = th.th_dport;
521 			return;
522 		case IPPROTO_UDP:
523 			spidx->ul_proto = nxt;
524 			if (!needport)
525 				goto done_proto;
526 			if (off + sizeof(struct udphdr) > m->m_pkthdr.len)
527 				goto done;
528 			m_copydata(m, off, sizeof (uh), (caddr_t) &uh);
529 			spidx->src.sin.sin_port = uh.uh_sport;
530 			spidx->dst.sin.sin_port = uh.uh_dport;
531 			return;
532 		case IPPROTO_AH:
533 			if (off + sizeof(ip6e) > m->m_pkthdr.len)
534 				goto done;
535 			/* XXX Sigh, this works but is totally bogus. */
536 			m_copydata(m, off, sizeof(ip6e), (caddr_t) &ip6e);
537 			off += (ip6e.ip6e_len + 2) << 2;
538 			nxt = ip6e.ip6e_nxt;
539 			break;
540 		case IPPROTO_ICMP:
541 		default:
542 			/* XXX Intermediate headers??? */
543 			spidx->ul_proto = nxt;
544 			goto done_proto;
545 		}
546 	}
547 done:
548 	spidx->ul_proto = IPSEC_ULPROTO_ANY;
549 done_proto:
550 	spidx->src.sin.sin_port = IPSEC_PORT_ANY;
551 	spidx->dst.sin.sin_port = IPSEC_PORT_ANY;
552 	KEYDBG(IPSEC_DUMP,
553 	    printf("%s: ", __func__); kdebug_secpolicyindex(spidx, NULL));
554 }
555 
556 static void
557 ipsec4_setspidx_ipaddr(const struct mbuf *m, struct secpolicyindex *spidx)
558 {
559 
560 	ipsec4_setsockaddrs(m, &spidx->src, &spidx->dst);
561 	spidx->prefs = sizeof(struct in_addr) << 3;
562 	spidx->prefd = sizeof(struct in_addr) << 3;
563 }
564 
565 static struct secpolicy *
566 ipsec4_getpolicy(const struct mbuf *m, struct inpcb *inp, u_int dir)
567 {
568 	struct secpolicyindex spidx;
569 	struct secpolicy *sp;
570 
571 	sp = ipsec_getpcbpolicy(inp, dir);
572 	if (sp == NULL && key_havesp(dir)) {
573 		/* Make an index to look for a policy. */
574 		ipsec4_setspidx_ipaddr(m, &spidx);
575 		/* Fill ports in spidx if we have inpcb. */
576 		ipsec4_get_ulp(m, &spidx, inp != NULL);
577 		spidx.dir = dir;
578 		sp = key_allocsp(&spidx, dir);
579 	}
580 	if (sp == NULL)		/* No SP found, use system default. */
581 		sp = key_allocsp_default();
582 	return (sp);
583 }
584 
585 /*
586  * Check security policy for *OUTBOUND* IPv4 packet.
587  */
588 struct secpolicy *
589 ipsec4_checkpolicy(const struct mbuf *m, struct inpcb *inp, int *error)
590 {
591 	struct secpolicy *sp;
592 
593 	*error = 0;
594 	sp = ipsec4_getpolicy(m, inp, IPSEC_DIR_OUTBOUND);
595 	if (sp != NULL)
596 		sp = ipsec_checkpolicy(sp, inp, error);
597 	if (sp == NULL) {
598 		switch (*error) {
599 		case 0: /* No IPsec required: BYPASS or NONE */
600 			break;
601 		case -EINVAL:
602 			IPSECSTAT_INC(ips_out_polvio);
603 			break;
604 		default:
605 			IPSECSTAT_INC(ips_out_inval);
606 		}
607 	}
608 	KEYDBG(IPSEC_STAMP,
609 	    printf("%s: using SP(%p), error %d\n", __func__, sp, *error));
610 	if (sp != NULL)
611 		KEYDBG(IPSEC_DATA, kdebug_secpolicy(sp));
612 	return (sp);
613 }
614 
615 /*
616  * Check IPv4 packet against *INBOUND* security policy.
617  * This function is called from tcp_input(), udp_input(),
618  * rip_input() and sctp_input().
619  */
620 int
621 ipsec4_in_reject(const struct mbuf *m, struct inpcb *inp)
622 {
623 	struct secpolicy *sp;
624 	int result;
625 
626 	sp = ipsec4_getpolicy(m, inp, IPSEC_DIR_INBOUND);
627 	result = ipsec_in_reject(sp, inp, m);
628 	key_freesp(&sp);
629 	if (result != 0)
630 		IPSECSTAT_INC(ips_in_polvio);
631 	return (result);
632 }
633 
634 /*
635  * IPSEC_CAP() method implementation for IPv4.
636  */
637 int
638 ipsec4_capability(struct mbuf *m, u_int cap)
639 {
640 
641 	switch (cap) {
642 	case IPSEC_CAP_BYPASS_FILTER:
643 		/*
644 		 * Bypass packet filtering for packets previously handled
645 		 * by IPsec.
646 		 */
647 		if (!V_ip4_filtertunnel &&
648 		    m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL) != NULL)
649 			return (1);
650 		return (0);
651 	case IPSEC_CAP_OPERABLE:
652 		/* Do we have active security policies? */
653 		if (key_havesp(IPSEC_DIR_INBOUND) != 0 ||
654 		    key_havesp(IPSEC_DIR_OUTBOUND) != 0)
655 			return (1);
656 		return (0);
657 	};
658 	return (EOPNOTSUPP);
659 }
660 
661 #endif /* INET */
662 
663 #ifdef INET6
664 static void
665 ipsec6_get_ulp(const struct mbuf *m, struct secpolicyindex *spidx,
666     int needport)
667 {
668 	struct tcphdr th;
669 	struct udphdr uh;
670 	struct icmp6_hdr ih;
671 	int off, nxt;
672 
673 	IPSEC_ASSERT(m->m_pkthdr.len >= sizeof(struct ip6_hdr),
674 	    ("packet too short"));
675 
676 	/* Set default. */
677 	spidx->ul_proto = IPSEC_ULPROTO_ANY;
678 	spidx->src.sin6.sin6_port = IPSEC_PORT_ANY;
679 	spidx->dst.sin6.sin6_port = IPSEC_PORT_ANY;
680 
681 	nxt = -1;
682 	off = ip6_lasthdr(m, 0, IPPROTO_IPV6, &nxt);
683 	if (off < 0 || m->m_pkthdr.len < off)
684 		return;
685 
686 	switch (nxt) {
687 	case IPPROTO_TCP:
688 		spidx->ul_proto = nxt;
689 		if (!needport)
690 			break;
691 		if (off + sizeof(struct tcphdr) > m->m_pkthdr.len)
692 			break;
693 		m_copydata(m, off, sizeof(th), (caddr_t)&th);
694 		spidx->src.sin6.sin6_port = th.th_sport;
695 		spidx->dst.sin6.sin6_port = th.th_dport;
696 		break;
697 	case IPPROTO_UDP:
698 		spidx->ul_proto = nxt;
699 		if (!needport)
700 			break;
701 		if (off + sizeof(struct udphdr) > m->m_pkthdr.len)
702 			break;
703 		m_copydata(m, off, sizeof(uh), (caddr_t)&uh);
704 		spidx->src.sin6.sin6_port = uh.uh_sport;
705 		spidx->dst.sin6.sin6_port = uh.uh_dport;
706 		break;
707 	case IPPROTO_ICMPV6:
708 		spidx->ul_proto = nxt;
709 		if (off + sizeof(struct icmp6_hdr) > m->m_pkthdr.len)
710 			break;
711 		m_copydata(m, off, sizeof(ih), (caddr_t)&ih);
712 		spidx->src.sin6.sin6_port = htons((uint16_t)ih.icmp6_type);
713 		spidx->dst.sin6.sin6_port = htons((uint16_t)ih.icmp6_code);
714 		break;
715 	default:
716 		/* XXX Intermediate headers??? */
717 		spidx->ul_proto = nxt;
718 		break;
719 	}
720 	KEYDBG(IPSEC_DUMP,
721 	    printf("%s: ", __func__); kdebug_secpolicyindex(spidx, NULL));
722 }
723 
724 static void
725 ipsec6_setspidx_ipaddr(const struct mbuf *m, struct secpolicyindex *spidx)
726 {
727 
728 	ipsec6_setsockaddrs(m, &spidx->src, &spidx->dst);
729 	spidx->prefs = sizeof(struct in6_addr) << 3;
730 	spidx->prefd = sizeof(struct in6_addr) << 3;
731 }
732 
733 static struct secpolicy *
734 ipsec6_getpolicy(const struct mbuf *m, struct inpcb *inp, u_int dir)
735 {
736 	struct secpolicyindex spidx;
737 	struct secpolicy *sp;
738 
739 	sp = ipsec_getpcbpolicy(inp, dir);
740 	if (sp == NULL && key_havesp(dir)) {
741 		/* Make an index to look for a policy. */
742 		ipsec6_setspidx_ipaddr(m, &spidx);
743 		/* Fill ports in spidx if we have inpcb. */
744 		ipsec6_get_ulp(m, &spidx, inp != NULL);
745 		spidx.dir = dir;
746 		sp = key_allocsp(&spidx, dir);
747 	}
748 	if (sp == NULL)		/* No SP found, use system default. */
749 		sp = key_allocsp_default();
750 	return (sp);
751 }
752 
753 /*
754  * Check security policy for *OUTBOUND* IPv6 packet.
755  */
756 struct secpolicy *
757 ipsec6_checkpolicy(const struct mbuf *m, struct inpcb *inp, int *error)
758 {
759 	struct secpolicy *sp;
760 
761 	*error = 0;
762 	sp = ipsec6_getpolicy(m, inp, IPSEC_DIR_OUTBOUND);
763 	if (sp != NULL)
764 		sp = ipsec_checkpolicy(sp, inp, error);
765 	if (sp == NULL) {
766 		switch (*error) {
767 		case 0: /* No IPsec required: BYPASS or NONE */
768 			break;
769 		case -EINVAL:
770 			IPSEC6STAT_INC(ips_out_polvio);
771 			break;
772 		default:
773 			IPSEC6STAT_INC(ips_out_inval);
774 		}
775 	}
776 	KEYDBG(IPSEC_STAMP,
777 	    printf("%s: using SP(%p), error %d\n", __func__, sp, *error));
778 	if (sp != NULL)
779 		KEYDBG(IPSEC_DATA, kdebug_secpolicy(sp));
780 	return (sp);
781 }
782 
783 /*
784  * Check IPv6 packet against inbound security policy.
785  * This function is called from tcp6_input(), udp6_input(),
786  * rip6_input() and sctp_input().
787  */
788 int
789 ipsec6_in_reject(const struct mbuf *m, struct inpcb *inp)
790 {
791 	struct secpolicy *sp;
792 	int result;
793 
794 	sp = ipsec6_getpolicy(m, inp, IPSEC_DIR_INBOUND);
795 	result = ipsec_in_reject(sp, inp, m);
796 	key_freesp(&sp);
797 	if (result)
798 		IPSEC6STAT_INC(ips_in_polvio);
799 	return (result);
800 }
801 
802 /*
803  * IPSEC_CAP() method implementation for IPv6.
804  */
805 int
806 ipsec6_capability(struct mbuf *m, u_int cap)
807 {
808 
809 	switch (cap) {
810 	case IPSEC_CAP_BYPASS_FILTER:
811 		/*
812 		 * Bypass packet filtering for packets previously handled
813 		 * by IPsec.
814 		 */
815 		if (!V_ip6_filtertunnel &&
816 		    m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL) != NULL)
817 			return (1);
818 		return (0);
819 	case IPSEC_CAP_OPERABLE:
820 		/* Do we have active security policies? */
821 		if (key_havesp(IPSEC_DIR_INBOUND) != 0 ||
822 		    key_havesp(IPSEC_DIR_OUTBOUND) != 0)
823 			return (1);
824 		return (0);
825 	};
826 	return (EOPNOTSUPP);
827 }
828 #endif /* INET6 */
829 
830 int
831 ipsec_run_hhooks(struct ipsec_ctx_data *ctx, int type)
832 {
833 	int idx;
834 
835 	switch (ctx->af) {
836 #ifdef INET
837 	case AF_INET:
838 		idx = HHOOK_IPSEC_INET;
839 		break;
840 #endif
841 #ifdef INET6
842 	case AF_INET6:
843 		idx = HHOOK_IPSEC_INET6;
844 		break;
845 #endif
846 	default:
847 		return (EPFNOSUPPORT);
848 	}
849 	if (type == HHOOK_TYPE_IPSEC_IN)
850 		HHOOKS_RUN_IF(V_ipsec_hhh_in[idx], ctx, NULL);
851 	else
852 		HHOOKS_RUN_IF(V_ipsec_hhh_out[idx], ctx, NULL);
853 	if (*ctx->mp == NULL)
854 		return (EACCES);
855 	return (0);
856 }
857 
858 /*
859  * Return current level.
860  * Either IPSEC_LEVEL_USE or IPSEC_LEVEL_REQUIRE are always returned.
861  */
862 u_int
863 ipsec_get_reqlevel(struct secpolicy *sp, u_int idx)
864 {
865 	struct ipsecrequest *isr;
866 	u_int esp_trans_deflev, esp_net_deflev;
867 	u_int ah_trans_deflev, ah_net_deflev;
868 	u_int level = 0;
869 
870 	IPSEC_ASSERT(idx < sp->tcount, ("Wrong IPsec request index %d", idx));
871 /* XXX Note that we have ipseclog() expanded here - code sync issue. */
872 #define IPSEC_CHECK_DEFAULT(lev) \
873 	(((lev) != IPSEC_LEVEL_USE && (lev) != IPSEC_LEVEL_REQUIRE &&	\
874 	  (lev) != IPSEC_LEVEL_UNIQUE)					\
875 		? (V_ipsec_debug  ?					\
876 		log(LOG_INFO, "fixed system default level " #lev ":%d->%d\n",\
877 		(lev), IPSEC_LEVEL_REQUIRE) : 0),			\
878 		(lev) = IPSEC_LEVEL_REQUIRE, (lev) : (lev))
879 
880 	/*
881 	 * IPsec VTI uses unique security policy with fake spidx filled
882 	 * with zeroes. Just return IPSEC_LEVEL_REQUIRE instead of doing
883 	 * full level lookup for such policies.
884 	 */
885 	if (sp->state == IPSEC_SPSTATE_IFNET) {
886 		IPSEC_ASSERT(sp->req[idx]->level == IPSEC_LEVEL_UNIQUE,
887 		    ("Wrong IPsec request level %d", sp->req[idx]->level));
888 		return (IPSEC_LEVEL_REQUIRE);
889 	}
890 
891 	/* Set default level. */
892 	switch (sp->spidx.src.sa.sa_family) {
893 #ifdef INET
894 	case AF_INET:
895 		esp_trans_deflev = IPSEC_CHECK_DEFAULT(V_ip4_esp_trans_deflev);
896 		esp_net_deflev = IPSEC_CHECK_DEFAULT(V_ip4_esp_net_deflev);
897 		ah_trans_deflev = IPSEC_CHECK_DEFAULT(V_ip4_ah_trans_deflev);
898 		ah_net_deflev = IPSEC_CHECK_DEFAULT(V_ip4_ah_net_deflev);
899 		break;
900 #endif
901 #ifdef INET6
902 	case AF_INET6:
903 		esp_trans_deflev = IPSEC_CHECK_DEFAULT(V_ip6_esp_trans_deflev);
904 		esp_net_deflev = IPSEC_CHECK_DEFAULT(V_ip6_esp_net_deflev);
905 		ah_trans_deflev = IPSEC_CHECK_DEFAULT(V_ip6_ah_trans_deflev);
906 		ah_net_deflev = IPSEC_CHECK_DEFAULT(V_ip6_ah_net_deflev);
907 		break;
908 #endif /* INET6 */
909 	default:
910 		panic("%s: unknown af %u",
911 			__func__, sp->spidx.src.sa.sa_family);
912 	}
913 
914 #undef IPSEC_CHECK_DEFAULT
915 
916 	isr = sp->req[idx];
917 	/* Set level. */
918 	switch (isr->level) {
919 	case IPSEC_LEVEL_DEFAULT:
920 		switch (isr->saidx.proto) {
921 		case IPPROTO_ESP:
922 			if (isr->saidx.mode == IPSEC_MODE_TUNNEL)
923 				level = esp_net_deflev;
924 			else
925 				level = esp_trans_deflev;
926 			break;
927 		case IPPROTO_AH:
928 			if (isr->saidx.mode == IPSEC_MODE_TUNNEL)
929 				level = ah_net_deflev;
930 			else
931 				level = ah_trans_deflev;
932 			break;
933 		case IPPROTO_IPCOMP:
934 			/*
935 			 * We don't really care, as IPcomp document says that
936 			 * we shouldn't compress small packets.
937 			 */
938 			level = IPSEC_LEVEL_USE;
939 			break;
940 		default:
941 			panic("%s: Illegal protocol defined %u\n", __func__,
942 				isr->saidx.proto);
943 		}
944 		break;
945 
946 	case IPSEC_LEVEL_USE:
947 	case IPSEC_LEVEL_REQUIRE:
948 		level = isr->level;
949 		break;
950 	case IPSEC_LEVEL_UNIQUE:
951 		level = IPSEC_LEVEL_REQUIRE;
952 		break;
953 
954 	default:
955 		panic("%s: Illegal IPsec level %u\n", __func__, isr->level);
956 	}
957 
958 	return (level);
959 }
960 
961 static int
962 ipsec_check_history(const struct mbuf *m, struct secpolicy *sp, u_int idx)
963 {
964 	struct xform_history *xh;
965 	struct m_tag *mtag;
966 
967 	mtag = NULL;
968 	while ((mtag = m_tag_find(__DECONST(struct mbuf *, m),
969 	    PACKET_TAG_IPSEC_IN_DONE, mtag)) != NULL) {
970 		xh = (struct xform_history *)(mtag + 1);
971 		KEYDBG(IPSEC_DATA,
972 		    char buf[IPSEC_ADDRSTRLEN];
973 		    printf("%s: mode %s proto %u dst %s\n", __func__,
974 			kdebug_secasindex_mode(xh->mode), xh->proto,
975 			ipsec_address(&xh->dst, buf, sizeof(buf))));
976 		if (xh->proto != sp->req[idx]->saidx.proto)
977 			continue;
978 		/* If SA had IPSEC_MODE_ANY, consider this as match. */
979 		if (xh->mode != sp->req[idx]->saidx.mode &&
980 		    xh->mode != IPSEC_MODE_ANY)
981 			continue;
982 		/*
983 		 * For transport mode IPsec request doesn't contain
984 		 * addresses. We need to use address from spidx.
985 		 */
986 		if (sp->req[idx]->saidx.mode == IPSEC_MODE_TRANSPORT) {
987 			if (key_sockaddrcmp_withmask(&xh->dst.sa,
988 			    &sp->spidx.dst.sa, sp->spidx.prefd) != 0)
989 				continue;
990 		} else {
991 			if (key_sockaddrcmp(&xh->dst.sa,
992 			    &sp->req[idx]->saidx.dst.sa, 0) != 0)
993 				continue;
994 		}
995 		return (0); /* matched */
996 	}
997 	return (1);
998 }
999 
1000 /*
1001  * Check security policy requirements against the actual
1002  * packet contents.  Return one if the packet should be
1003  * reject as "invalid"; otherwiser return zero to have the
1004  * packet treated as "valid".
1005  *
1006  * OUT:
1007  *	0: valid
1008  *	1: invalid
1009  */
1010 static int
1011 ipsec_in_reject(struct secpolicy *sp, struct inpcb *inp, const struct mbuf *m)
1012 {
1013 	int i;
1014 
1015 	KEYDBG(IPSEC_STAMP,
1016 	    printf("%s: PCB(%p): using SP(%p)\n", __func__, inp, sp));
1017 	KEYDBG(IPSEC_DATA, kdebug_secpolicy(sp));
1018 
1019 	if (inp != NULL && inp->inp_sp != NULL && inp->inp_sp->sp_in == NULL)
1020 		ipsec_cachepolicy(inp, sp, IPSEC_DIR_INBOUND);
1021 
1022 	/* Check policy. */
1023 	switch (sp->policy) {
1024 	case IPSEC_POLICY_DISCARD:
1025 		return (1);
1026 	case IPSEC_POLICY_BYPASS:
1027 	case IPSEC_POLICY_NONE:
1028 		return (0);
1029 	}
1030 
1031 	IPSEC_ASSERT(sp->policy == IPSEC_POLICY_IPSEC,
1032 		("invalid policy %u", sp->policy));
1033 
1034 	/*
1035 	 * ipsec[46]_common_input_cb after each transform adds
1036 	 * PACKET_TAG_IPSEC_IN_DONE mbuf tag. It contains SPI, proto, mode
1037 	 * and destination address from saidx. We can compare info from
1038 	 * these tags with requirements in SP.
1039 	 */
1040 	for (i = 0; i < sp->tcount; i++) {
1041 		/*
1042 		 * Do not check IPcomp, since IPcomp document
1043 		 * says that we shouldn't compress small packets.
1044 		 * IPComp policy should always be treated as being
1045 		 * in "use" level.
1046 		 */
1047 		if (sp->req[i]->saidx.proto == IPPROTO_IPCOMP ||
1048 		    ipsec_get_reqlevel(sp, i) != IPSEC_LEVEL_REQUIRE)
1049 			continue;
1050 		if (V_check_policy_history != 0 &&
1051 		    ipsec_check_history(m, sp, i) != 0)
1052 			return (1);
1053 		else switch (sp->req[i]->saidx.proto) {
1054 		case IPPROTO_ESP:
1055 			if ((m->m_flags & M_DECRYPTED) == 0) {
1056 				KEYDBG(IPSEC_DUMP,
1057 				    printf("%s: ESP m_flags:%x\n", __func__,
1058 					    m->m_flags));
1059 				return (1);
1060 			}
1061 			break;
1062 		case IPPROTO_AH:
1063 			if ((m->m_flags & M_AUTHIPHDR) == 0) {
1064 				KEYDBG(IPSEC_DUMP,
1065 				    printf("%s: AH m_flags:%x\n", __func__,
1066 					    m->m_flags));
1067 				return (1);
1068 			}
1069 			break;
1070 		}
1071 	}
1072 	return (0);		/* Valid. */
1073 }
1074 
1075 /*
1076  * Compute the byte size to be occupied by IPsec header.
1077  * In case it is tunnelled, it includes the size of outer IP header.
1078  */
1079 static size_t
1080 ipsec_hdrsiz_internal(struct secpolicy *sp)
1081 {
1082 	size_t size;
1083 	int i;
1084 
1085 	KEYDBG(IPSEC_STAMP, printf("%s: using SP(%p)\n", __func__, sp));
1086 	KEYDBG(IPSEC_DATA, kdebug_secpolicy(sp));
1087 
1088 	switch (sp->policy) {
1089 	case IPSEC_POLICY_DISCARD:
1090 	case IPSEC_POLICY_BYPASS:
1091 	case IPSEC_POLICY_NONE:
1092 		return (0);
1093 	}
1094 
1095 	IPSEC_ASSERT(sp->policy == IPSEC_POLICY_IPSEC,
1096 		("invalid policy %u", sp->policy));
1097 
1098 	/*
1099 	 * XXX: for each transform we need to lookup suitable SA
1100 	 * and use info from SA to calculate headers size.
1101 	 * XXX: for NAT-T we need to cosider UDP header size.
1102 	 */
1103 	size = 0;
1104 	for (i = 0; i < sp->tcount; i++) {
1105 		switch (sp->req[i]->saidx.proto) {
1106 		case IPPROTO_ESP:
1107 			size += esp_hdrsiz(NULL);
1108 			break;
1109 		case IPPROTO_AH:
1110 			size += ah_hdrsiz(NULL);
1111 			break;
1112 		case IPPROTO_IPCOMP:
1113 			size += sizeof(struct ipcomp);
1114 			break;
1115 		}
1116 
1117 		if (sp->req[i]->saidx.mode == IPSEC_MODE_TUNNEL) {
1118 			switch (sp->req[i]->saidx.dst.sa.sa_family) {
1119 #ifdef INET
1120 			case AF_INET:
1121 				size += sizeof(struct ip);
1122 				break;
1123 #endif
1124 #ifdef INET6
1125 			case AF_INET6:
1126 				size += sizeof(struct ip6_hdr);
1127 				break;
1128 #endif
1129 			default:
1130 				ipseclog((LOG_ERR, "%s: unknown AF %d in "
1131 				    "IPsec tunnel SA\n", __func__,
1132 				    sp->req[i]->saidx.dst.sa.sa_family));
1133 				break;
1134 			}
1135 		}
1136 	}
1137 	return (size);
1138 }
1139 
1140 /*
1141  * Compute ESP/AH header size for protocols with PCB, including
1142  * outer IP header. Currently only tcp_output() uses it.
1143  */
1144 size_t
1145 ipsec_hdrsiz_inpcb(struct inpcb *inp)
1146 {
1147 	struct secpolicyindex spidx;
1148 	struct secpolicy *sp;
1149 	size_t sz;
1150 
1151 	sp = ipsec_getpcbpolicy(inp, IPSEC_DIR_OUTBOUND);
1152 	if (sp == NULL && key_havesp(IPSEC_DIR_OUTBOUND)) {
1153 		ipsec_setspidx_inpcb(inp, &spidx, IPSEC_DIR_OUTBOUND);
1154 		sp = key_allocsp(&spidx, IPSEC_DIR_OUTBOUND);
1155 	}
1156 	if (sp == NULL)
1157 		sp = key_allocsp_default();
1158 	sz = ipsec_hdrsiz_internal(sp);
1159 	key_freesp(&sp);
1160 	return (sz);
1161 }
1162 
1163 /*
1164  * Check the variable replay window.
1165  * ipsec_chkreplay() performs replay check before ICV verification.
1166  * ipsec_updatereplay() updates replay bitmap.  This must be called after
1167  * ICV verification (it also performs replay check, which is usually done
1168  * beforehand).
1169  * 0 (zero) is returned if packet disallowed, 1 if packet permitted.
1170  *
1171  * Based on RFC 6479. Blocks are 32 bits unsigned integers
1172  */
1173 
1174 #define IPSEC_BITMAP_INDEX_MASK(w)	(w - 1)
1175 #define IPSEC_REDUNDANT_BIT_SHIFTS	5
1176 #define IPSEC_REDUNDANT_BITS		(1 << IPSEC_REDUNDANT_BIT_SHIFTS)
1177 #define IPSEC_BITMAP_LOC_MASK		(IPSEC_REDUNDANT_BITS - 1)
1178 
1179 int
1180 ipsec_chkreplay(uint32_t seq, struct secasvar *sav)
1181 {
1182 	const struct secreplay *replay;
1183 	uint32_t wsizeb;		/* Constant: window size. */
1184 	int index, bit_location;
1185 
1186 	IPSEC_ASSERT(sav != NULL, ("Null SA"));
1187 	IPSEC_ASSERT(sav->replay != NULL, ("Null replay state"));
1188 
1189 	replay = sav->replay;
1190 
1191 	/* No need to check replay if disabled. */
1192 	if (replay->wsize == 0)
1193 		return (1);
1194 
1195 	/* Constant. */
1196 	wsizeb = replay->wsize << 3;
1197 
1198 	/* Sequence number of 0 is invalid. */
1199 	if (seq == 0)
1200 		return (0);
1201 
1202 	/* First time is always okay. */
1203 	if (replay->count == 0)
1204 		return (1);
1205 
1206 	/* Larger sequences are okay. */
1207 	if (seq > replay->lastseq)
1208 		return (1);
1209 
1210 	/* Over range to check, i.e. too old or wrapped. */
1211 	if (replay->lastseq - seq >= wsizeb)
1212 		return (0);
1213 
1214 	/* The sequence is inside the sliding window
1215 	 * now check the bit in the bitmap
1216 	 * bit location only depends on the sequence number
1217 	 */
1218 	bit_location = seq & IPSEC_BITMAP_LOC_MASK;
1219 	index = (seq >> IPSEC_REDUNDANT_BIT_SHIFTS)
1220 		& IPSEC_BITMAP_INDEX_MASK(replay->bitmap_size);
1221 
1222 	/* This packet already seen? */
1223 	if ((replay->bitmap)[index] & (1 << bit_location))
1224 		return (0);
1225 	return (1);
1226 }
1227 
1228 /*
1229  * Check replay counter whether to update or not.
1230  * OUT:	0:	OK
1231  *	1:	NG
1232  */
1233 int
1234 ipsec_updatereplay(uint32_t seq, struct secasvar *sav)
1235 {
1236 	char buf[128];
1237 	struct secreplay *replay;
1238 	uint32_t wsizeb;		/* Constant: window size. */
1239 	int diff, index, bit_location;
1240 
1241 	IPSEC_ASSERT(sav != NULL, ("Null SA"));
1242 	IPSEC_ASSERT(sav->replay != NULL, ("Null replay state"));
1243 
1244 	replay = sav->replay;
1245 
1246 	if (replay->wsize == 0)
1247 		goto ok;	/* No need to check replay. */
1248 
1249 	/* Constant. */
1250 	wsizeb = replay->wsize << 3;
1251 
1252 	/* Sequence number of 0 is invalid. */
1253 	if (seq == 0)
1254 		return (1);
1255 
1256 	/* The packet is too old, no need to update */
1257 	if (wsizeb + seq < replay->lastseq)
1258 		goto ok;
1259 
1260 	/* Now update the bit */
1261 	index = (seq >> IPSEC_REDUNDANT_BIT_SHIFTS);
1262 
1263 	/* First check if the sequence number is in the range */
1264 	if (seq > replay->lastseq) {
1265 		int id;
1266 		int index_cur = replay->lastseq >> IPSEC_REDUNDANT_BIT_SHIFTS;
1267 
1268 		diff = index - index_cur;
1269 		if (diff > replay->bitmap_size) {
1270 			/* something unusual in this case */
1271 			diff = replay->bitmap_size;
1272 		}
1273 
1274 		for (id = 0; id < diff; ++id) {
1275 			replay->bitmap[(id + index_cur + 1)
1276 			& IPSEC_BITMAP_INDEX_MASK(replay->bitmap_size)] = 0;
1277 		}
1278 
1279 		replay->lastseq = seq;
1280 	}
1281 
1282 	index &= IPSEC_BITMAP_INDEX_MASK(replay->bitmap_size);
1283 	bit_location = seq & IPSEC_BITMAP_LOC_MASK;
1284 
1285 	/* this packet has already been received */
1286 	if (replay->bitmap[index] & (1 << bit_location))
1287 		return (1);
1288 
1289 	replay->bitmap[index] |= (1 << bit_location);
1290 
1291 ok:
1292 	if (replay->count == ~0) {
1293 
1294 		/* Set overflow flag. */
1295 		replay->overflow++;
1296 
1297 		/* Don't increment, no more packets accepted. */
1298 		if ((sav->flags & SADB_X_EXT_CYCSEQ) == 0) {
1299 			if (sav->sah->saidx.proto == IPPROTO_AH)
1300 				AHSTAT_INC(ahs_wrap);
1301 			else if (sav->sah->saidx.proto == IPPROTO_ESP)
1302 				ESPSTAT_INC(esps_wrap);
1303 			return (1);
1304 		}
1305 
1306 		ipseclog((LOG_WARNING, "%s: replay counter made %d cycle. %s\n",
1307 		    __func__, replay->overflow,
1308 		    ipsec_sa2str(sav, buf, sizeof(buf))));
1309 	}
1310 	return (0);
1311 }
1312 
1313 int
1314 ipsec_updateid(struct secasvar *sav, uint64_t *new, uint64_t *old)
1315 {
1316 	uint64_t tmp;
1317 
1318 	/*
1319 	 * tdb_cryptoid is initialized by xform_init().
1320 	 * Then it can be changed only when some crypto error occurred or
1321 	 * when SA is deleted. We stored used cryptoid in the xform_data
1322 	 * structure. In case when crypto error occurred and crypto
1323 	 * subsystem has reinited the session, it returns new cryptoid
1324 	 * and EAGAIN error code.
1325 	 *
1326 	 * This function will be called when we got EAGAIN from crypto
1327 	 * subsystem.
1328 	 * *new is cryptoid that was returned by crypto subsystem in
1329 	 * the crp_sid.
1330 	 * *old is the original cryptoid that we stored in xform_data.
1331 	 *
1332 	 * For first failed request *old == sav->tdb_cryptoid, then
1333 	 * we update sav->tdb_cryptoid and redo crypto_dispatch().
1334 	 * For next failed request *old != sav->tdb_cryptoid, then
1335 	 * we store cryptoid from first request into the *new variable
1336 	 * and crp_sid from this second session will be returned via
1337 	 * *old pointer, so caller can release second session.
1338 	 *
1339 	 * XXXAE: check this more carefully.
1340 	 */
1341 	KEYDBG(IPSEC_STAMP,
1342 	    printf("%s: SA(%p) moves cryptoid %jd -> %jd\n",
1343 		__func__, sav, (uintmax_t)(*old), (uintmax_t)(*new)));
1344 	KEYDBG(IPSEC_DATA, kdebug_secasv(sav));
1345 	SECASVAR_LOCK(sav);
1346 	if (sav->tdb_cryptoid != *old) {
1347 		/* cryptoid was already updated */
1348 		tmp = *new;
1349 		*new = sav->tdb_cryptoid;
1350 		*old = tmp;
1351 		SECASVAR_UNLOCK(sav);
1352 		return (1);
1353 	}
1354 	sav->tdb_cryptoid = *new;
1355 	SECASVAR_UNLOCK(sav);
1356 	return (0);
1357 }
1358 
1359 int
1360 ipsec_initialized(void)
1361 {
1362 
1363 	return (V_def_policy != NULL);
1364 }
1365 
1366 static void
1367 def_policy_init(const void *unused __unused)
1368 {
1369 
1370 	V_def_policy = key_newsp();
1371 	if (V_def_policy != NULL) {
1372 		V_def_policy->policy = IPSEC_POLICY_NONE;
1373 		/* Force INPCB SP cache invalidation */
1374 		key_bumpspgen();
1375 	} else
1376 		printf("%s: failed to initialize default policy\n", __func__);
1377 }
1378 
1379 
1380 static void
1381 def_policy_uninit(const void *unused __unused)
1382 {
1383 
1384 	if (V_def_policy != NULL) {
1385 		key_freesp(&V_def_policy);
1386 		key_bumpspgen();
1387 	}
1388 }
1389 
1390 VNET_SYSINIT(def_policy_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST,
1391     def_policy_init, NULL);
1392 VNET_SYSUNINIT(def_policy_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST,
1393     def_policy_uninit, NULL);
1394