xref: /freebsd/sys/netipsec/ipsec.c (revision e12ff891366cf94db4bfe4c2c810b26a5531053d)
1 /*	$FreeBSD$	*/
2 /*	$KAME: ipsec.c,v 1.103 2001/05/24 07:14:18 sakane Exp $	*/
3 
4 /*-
5  * SPDX-License-Identifier: BSD-3-Clause
6  *
7  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
8  * All rights reserved.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. Neither the name of the project nor the names of its contributors
19  *    may be used to endorse or promote products derived from this software
20  *    without specific prior written permission.
21  *
22  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
23  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
26  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32  * SUCH DAMAGE.
33  */
34 
35 /*
36  * IPsec controller part.
37  */
38 
39 #include "opt_inet.h"
40 #include "opt_inet6.h"
41 #include "opt_ipsec.h"
42 
43 #include <sys/param.h>
44 #include <sys/systm.h>
45 #include <sys/malloc.h>
46 #include <sys/mbuf.h>
47 #include <sys/domain.h>
48 #include <sys/priv.h>
49 #include <sys/protosw.h>
50 #include <sys/socket.h>
51 #include <sys/socketvar.h>
52 #include <sys/errno.h>
53 #include <sys/hhook.h>
54 #include <sys/time.h>
55 #include <sys/kernel.h>
56 #include <sys/syslog.h>
57 #include <sys/sysctl.h>
58 #include <sys/proc.h>
59 
60 #include <net/if.h>
61 #include <net/if_enc.h>
62 #include <net/if_var.h>
63 #include <net/vnet.h>
64 
65 #include <netinet/in.h>
66 #include <netinet/in_systm.h>
67 #include <netinet/ip.h>
68 #include <netinet/ip_var.h>
69 #include <netinet/in_var.h>
70 #include <netinet/udp.h>
71 #include <netinet/udp_var.h>
72 #include <netinet/tcp.h>
73 #include <netinet/udp.h>
74 
75 #include <netinet/ip6.h>
76 #ifdef INET6
77 #include <netinet6/ip6_var.h>
78 #endif
79 #include <netinet/in_pcb.h>
80 #ifdef INET6
81 #include <netinet/icmp6.h>
82 #endif
83 
84 #include <sys/types.h>
85 #include <netipsec/ipsec.h>
86 #ifdef INET6
87 #include <netipsec/ipsec6.h>
88 #endif
89 #include <netipsec/ah_var.h>
90 #include <netipsec/esp_var.h>
91 #include <netipsec/ipcomp.h>		/*XXX*/
92 #include <netipsec/ipcomp_var.h>
93 #include <netipsec/ipsec_support.h>
94 
95 #include <netipsec/key.h>
96 #include <netipsec/keydb.h>
97 #include <netipsec/key_debug.h>
98 
99 #include <netipsec/xform.h>
100 
101 #include <machine/in_cksum.h>
102 
103 #include <opencrypto/cryptodev.h>
104 
105 /* NB: name changed so netstat doesn't use it. */
106 VNET_PCPUSTAT_DEFINE(struct ipsecstat, ipsec4stat);
107 VNET_PCPUSTAT_SYSINIT(ipsec4stat);
108 
109 #ifdef VIMAGE
110 VNET_PCPUSTAT_SYSUNINIT(ipsec4stat);
111 #endif /* VIMAGE */
112 
113 /* DF bit on encap. 0: clear 1: set 2: copy */
114 VNET_DEFINE(int, ip4_ipsec_dfbit) = 0;
115 VNET_DEFINE(int, ip4_esp_trans_deflev) = IPSEC_LEVEL_USE;
116 VNET_DEFINE(int, ip4_esp_net_deflev) = IPSEC_LEVEL_USE;
117 VNET_DEFINE(int, ip4_ah_trans_deflev) = IPSEC_LEVEL_USE;
118 VNET_DEFINE(int, ip4_ah_net_deflev) = IPSEC_LEVEL_USE;
119 /* ECN ignore(-1)/forbidden(0)/allowed(1) */
120 VNET_DEFINE(int, ip4_ipsec_ecn) = 0;
121 
122 VNET_DEFINE_STATIC(int, ip4_filtertunnel) = 0;
123 #define	V_ip4_filtertunnel VNET(ip4_filtertunnel)
124 VNET_DEFINE_STATIC(int, check_policy_history) = 0;
125 #define	V_check_policy_history	VNET(check_policy_history)
126 VNET_DEFINE_STATIC(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 /*
154  * Use asynchronous mode to parallelize crypto jobs:
155  *
156  *  0 - disabled
157  *  1 - enabled
158  */
159 VNET_DEFINE(int, async_crypto) = 0;
160 
161 /*
162  * TCP/UDP checksum handling policy for transport mode NAT-T (RFC3948)
163  *
164  * 0 - auto: incrementally recompute, when checksum delta is known;
165  *     if checksum delta isn't known, reset checksum to zero for UDP,
166  *     and mark csum_flags as valid for TCP.
167  * 1 - fully recompute TCP/UDP checksum.
168  */
169 VNET_DEFINE(int, natt_cksum_policy) = 0;
170 
171 FEATURE(ipsec, "Internet Protocol Security (IPsec)");
172 FEATURE(ipsec_natt, "UDP Encapsulation of IPsec ESP Packets ('NAT-T')");
173 
174 SYSCTL_DECL(_net_inet_ipsec);
175 
176 /* net.inet.ipsec */
177 SYSCTL_PROC(_net_inet_ipsec, IPSECCTL_DEF_POLICY, def_policy,
178 	CTLTYPE_INT | CTLFLAG_VNET | CTLFLAG_RW, 0, 0, sysctl_def_policy, "I",
179 	"IPsec default policy.");
180 SYSCTL_INT(_net_inet_ipsec, IPSECCTL_DEF_ESP_TRANSLEV, esp_trans_deflev,
181 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip4_esp_trans_deflev), 0,
182 	"Default ESP transport mode level");
183 SYSCTL_INT(_net_inet_ipsec, IPSECCTL_DEF_ESP_NETLEV, esp_net_deflev,
184 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip4_esp_net_deflev), 0,
185 	"Default ESP tunnel mode level.");
186 SYSCTL_INT(_net_inet_ipsec, IPSECCTL_DEF_AH_TRANSLEV, ah_trans_deflev,
187 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip4_ah_trans_deflev), 0,
188 	"AH transfer mode default level.");
189 SYSCTL_INT(_net_inet_ipsec, IPSECCTL_DEF_AH_NETLEV, ah_net_deflev,
190 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip4_ah_net_deflev), 0,
191 	"AH tunnel mode default level.");
192 SYSCTL_INT(_net_inet_ipsec, IPSECCTL_AH_CLEARTOS, ah_cleartos,
193 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ah_cleartos), 0,
194 	"If set, clear type-of-service field when doing AH computation.");
195 SYSCTL_INT(_net_inet_ipsec, IPSECCTL_DFBIT, dfbit,
196 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip4_ipsec_dfbit), 0,
197 	"Do not fragment bit on encap.");
198 SYSCTL_INT(_net_inet_ipsec, IPSECCTL_ECN, ecn,
199 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip4_ipsec_ecn), 0,
200 	"Explicit Congestion Notification handling.");
201 SYSCTL_INT(_net_inet_ipsec, OID_AUTO, crypto_support,
202 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(crypto_support), 0,
203 	"Crypto driver selection.");
204 SYSCTL_INT(_net_inet_ipsec, OID_AUTO, async_crypto,
205 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(async_crypto), 0,
206 	"Use asynchronous mode to parallelize crypto jobs.");
207 SYSCTL_INT(_net_inet_ipsec, OID_AUTO, check_policy_history,
208 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(check_policy_history), 0,
209 	"Use strict check of inbound packets to security policy compliance.");
210 SYSCTL_INT(_net_inet_ipsec, OID_AUTO, natt_cksum_policy,
211 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(natt_cksum_policy), 0,
212 	"Method to fix TCP/UDP checksum for transport mode IPsec after NAT.");
213 SYSCTL_INT(_net_inet_ipsec, OID_AUTO, filtertunnel,
214 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip4_filtertunnel), 0,
215 	"If set, filter packets from an IPsec tunnel.");
216 SYSCTL_VNET_PCPUSTAT(_net_inet_ipsec, OID_AUTO, ipsecstats, struct ipsecstat,
217     ipsec4stat, "IPsec IPv4 statistics.");
218 
219 struct timeval ipsec_warn_interval = { .tv_sec = 1, .tv_usec = 0 };
220 SYSCTL_TIMEVAL_SEC(_net_inet_ipsec, OID_AUTO, crypto_warn_interval, CTLFLAG_RW,
221     &ipsec_warn_interval,
222     "Delay in seconds between warnings of deprecated IPsec crypto algorithms.");
223 
224 #ifdef REGRESSION
225 /*
226  * When set to 1, IPsec will send packets with the same sequence number.
227  * This allows to verify if the other side has proper replay attacks detection.
228  */
229 VNET_DEFINE(int, ipsec_replay) = 0;
230 SYSCTL_INT(_net_inet_ipsec, OID_AUTO, test_replay,
231 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ipsec_replay), 0,
232 	"Emulate replay attack");
233 /*
234  * When set 1, IPsec will send packets with corrupted HMAC.
235  * This allows to verify if the other side properly detects modified packets.
236  */
237 VNET_DEFINE(int, ipsec_integrity) = 0;
238 SYSCTL_INT(_net_inet_ipsec, OID_AUTO, test_integrity,
239 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ipsec_integrity), 0,
240 	"Emulate man-in-the-middle attack");
241 #endif
242 
243 #ifdef INET6
244 VNET_PCPUSTAT_DEFINE(struct ipsecstat, ipsec6stat);
245 VNET_PCPUSTAT_SYSINIT(ipsec6stat);
246 
247 #ifdef VIMAGE
248 VNET_PCPUSTAT_SYSUNINIT(ipsec6stat);
249 #endif /* VIMAGE */
250 
251 VNET_DEFINE(int, ip6_esp_trans_deflev) = IPSEC_LEVEL_USE;
252 VNET_DEFINE(int, ip6_esp_net_deflev) = IPSEC_LEVEL_USE;
253 VNET_DEFINE(int, ip6_ah_trans_deflev) = IPSEC_LEVEL_USE;
254 VNET_DEFINE(int, ip6_ah_net_deflev) = IPSEC_LEVEL_USE;
255 VNET_DEFINE(int, ip6_ipsec_ecn) = 0;	/* ECN ignore(-1)/forbidden(0)/allowed(1) */
256 
257 VNET_DEFINE_STATIC(int, ip6_filtertunnel) = 0;
258 #define	V_ip6_filtertunnel	VNET(ip6_filtertunnel)
259 
260 SYSCTL_DECL(_net_inet6_ipsec6);
261 
262 /* net.inet6.ipsec6 */
263 SYSCTL_PROC(_net_inet6_ipsec6, IPSECCTL_DEF_POLICY, def_policy,
264 	CTLTYPE_INT | CTLFLAG_VNET | CTLFLAG_RW, 0, 0, sysctl_def_policy, "I",
265 	"IPsec default policy.");
266 SYSCTL_INT(_net_inet6_ipsec6, IPSECCTL_DEF_ESP_TRANSLEV, esp_trans_deflev,
267 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_esp_trans_deflev), 0,
268 	"Default ESP transport mode level.");
269 SYSCTL_INT(_net_inet6_ipsec6, IPSECCTL_DEF_ESP_NETLEV, esp_net_deflev,
270 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_esp_net_deflev), 0,
271 	"Default ESP tunnel mode level.");
272 SYSCTL_INT(_net_inet6_ipsec6, IPSECCTL_DEF_AH_TRANSLEV, ah_trans_deflev,
273 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_ah_trans_deflev), 0,
274 	"AH transfer mode default level.");
275 SYSCTL_INT(_net_inet6_ipsec6, IPSECCTL_DEF_AH_NETLEV, ah_net_deflev,
276 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_ah_net_deflev), 0,
277 	"AH tunnel mode default level.");
278 SYSCTL_INT(_net_inet6_ipsec6, IPSECCTL_ECN, ecn,
279 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_ipsec_ecn), 0,
280 	"Explicit Congestion Notification handling.");
281 SYSCTL_INT(_net_inet6_ipsec6, OID_AUTO, filtertunnel,
282 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_filtertunnel),  0,
283 	"If set, filter packets from an IPsec tunnel.");
284 SYSCTL_VNET_PCPUSTAT(_net_inet6_ipsec6, IPSECCTL_STATS, ipsecstats,
285     struct ipsecstat, ipsec6stat, "IPsec IPv6 statistics.");
286 #endif /* INET6 */
287 
288 static int ipsec_in_reject(struct secpolicy *, struct inpcb *,
289     const struct mbuf *);
290 
291 #ifdef INET
292 static void ipsec4_get_ulp(const struct mbuf *, struct secpolicyindex *, int);
293 static void ipsec4_setspidx_ipaddr(const struct mbuf *,
294     struct secpolicyindex *);
295 #endif
296 #ifdef INET6
297 static void ipsec6_get_ulp(const struct mbuf *m, struct secpolicyindex *, int);
298 static void ipsec6_setspidx_ipaddr(const struct mbuf *,
299     struct secpolicyindex *);
300 #endif
301 
302 /*
303  * Return a held reference to the default SP.
304  */
305 static struct secpolicy *
306 key_allocsp_default(void)
307 {
308 
309 	key_addref(V_def_policy);
310 	return (V_def_policy);
311 }
312 
313 static void
314 ipsec_invalidate_cache(struct inpcb *inp, u_int dir)
315 {
316 	struct secpolicy *sp;
317 
318 	INP_WLOCK_ASSERT(inp);
319 	if (dir == IPSEC_DIR_OUTBOUND) {
320 		if (inp->inp_sp->flags & INP_INBOUND_POLICY)
321 			return;
322 		sp = inp->inp_sp->sp_in;
323 		inp->inp_sp->sp_in = NULL;
324 	} else {
325 		if (inp->inp_sp->flags & INP_OUTBOUND_POLICY)
326 			return;
327 		sp = inp->inp_sp->sp_out;
328 		inp->inp_sp->sp_out = NULL;
329 	}
330 	if (sp != NULL)
331 		key_freesp(&sp); /* release extra reference */
332 }
333 
334 static void
335 ipsec_cachepolicy(struct inpcb *inp, struct secpolicy *sp, u_int dir)
336 {
337 	uint32_t genid;
338 	int downgrade;
339 
340 	INP_LOCK_ASSERT(inp);
341 
342 	if (dir == IPSEC_DIR_OUTBOUND) {
343 		/* Do we have configured PCB policy? */
344 		if (inp->inp_sp->flags & INP_OUTBOUND_POLICY)
345 			return;
346 		/* Another thread has already set cached policy */
347 		if (inp->inp_sp->sp_out != NULL)
348 			return;
349 		/*
350 		 * Do not cache OUTBOUND policy if PCB isn't connected,
351 		 * i.e. foreign address is INADDR_ANY/UNSPECIFIED.
352 		 */
353 #ifdef INET
354 		if ((inp->inp_vflag & INP_IPV4) != 0 &&
355 		    inp->inp_faddr.s_addr == INADDR_ANY)
356 			return;
357 #endif
358 #ifdef INET6
359 		if ((inp->inp_vflag & INP_IPV6) != 0 &&
360 		    IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr))
361 			return;
362 #endif
363 	} else {
364 		/* Do we have configured PCB policy? */
365 		if (inp->inp_sp->flags & INP_INBOUND_POLICY)
366 			return;
367 		/* Another thread has already set cached policy */
368 		if (inp->inp_sp->sp_in != NULL)
369 			return;
370 		/*
371 		 * Do not cache INBOUND policy for listen socket,
372 		 * that is bound to INADDR_ANY/UNSPECIFIED address.
373 		 */
374 #ifdef INET
375 		if ((inp->inp_vflag & INP_IPV4) != 0 &&
376 		    inp->inp_faddr.s_addr == INADDR_ANY)
377 			return;
378 #endif
379 #ifdef INET6
380 		if ((inp->inp_vflag & INP_IPV6) != 0 &&
381 		    IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr))
382 			return;
383 #endif
384 	}
385 	downgrade = 0;
386 	if (!INP_WLOCKED(inp)) {
387 		if ((downgrade = INP_TRY_UPGRADE(inp)) == 0)
388 			return;
389 	}
390 	if (dir == IPSEC_DIR_OUTBOUND)
391 		inp->inp_sp->sp_out = sp;
392 	else
393 		inp->inp_sp->sp_in = sp;
394 	/*
395 	 * SP is already referenced by the lookup code.
396 	 * We take extra reference here to avoid race in the
397 	 * ipsec_getpcbpolicy() function - SP will not be freed in the
398 	 * time between we take SP pointer from the cache and key_addref()
399 	 * call.
400 	 */
401 	key_addref(sp);
402 	genid = key_getspgen();
403 	if (genid != inp->inp_sp->genid) {
404 		ipsec_invalidate_cache(inp, dir);
405 		inp->inp_sp->genid = genid;
406 	}
407 	KEYDBG(IPSEC_STAMP,
408 	    printf("%s: PCB(%p): cached %s SP(%p)\n",
409 	    __func__, inp, dir == IPSEC_DIR_OUTBOUND ? "OUTBOUND":
410 	    "INBOUND", sp));
411 	if (downgrade != 0)
412 		INP_DOWNGRADE(inp);
413 }
414 
415 static struct secpolicy *
416 ipsec_checkpolicy(struct secpolicy *sp, struct inpcb *inp, int *error)
417 {
418 
419 	/* Save found OUTBOUND policy into PCB SP cache. */
420 	if (inp != NULL && inp->inp_sp != NULL && inp->inp_sp->sp_out == NULL)
421 		ipsec_cachepolicy(inp, sp, IPSEC_DIR_OUTBOUND);
422 
423 	switch (sp->policy) {
424 	default:
425 		printf("%s: invalid policy %u\n", __func__, sp->policy);
426 		/* FALLTHROUGH */
427 	case IPSEC_POLICY_DISCARD:
428 		*error = -EINVAL;	/* Packet is discarded by caller. */
429 		/* FALLTHROUGH */
430 	case IPSEC_POLICY_BYPASS:
431 	case IPSEC_POLICY_NONE:
432 		key_freesp(&sp);
433 		sp = NULL;		/* NB: force NULL result. */
434 		break;
435 	case IPSEC_POLICY_IPSEC:
436 		/* XXXAE: handle LARVAL SP */
437 		break;
438 	}
439 	KEYDBG(IPSEC_DUMP,
440 	    printf("%s: get SP(%p), error %d\n", __func__, sp, *error));
441 	return (sp);
442 }
443 
444 static struct secpolicy *
445 ipsec_getpcbpolicy(struct inpcb *inp, u_int dir)
446 {
447 	struct secpolicy *sp;
448 	int flags, downgrade;
449 
450 	if (inp == NULL || inp->inp_sp == NULL)
451 		return (NULL);
452 
453 	INP_LOCK_ASSERT(inp);
454 
455 	flags = inp->inp_sp->flags;
456 	if (dir == IPSEC_DIR_OUTBOUND) {
457 		sp = inp->inp_sp->sp_out;
458 		flags &= INP_OUTBOUND_POLICY;
459 	} else {
460 		sp = inp->inp_sp->sp_in;
461 		flags &= INP_INBOUND_POLICY;
462 	}
463 	/*
464 	 * Check flags. If we have PCB SP, just return it.
465 	 * Otherwise we need to check that cached SP entry isn't stale.
466 	 */
467 	if (flags == 0) {
468 		if (sp == NULL)
469 			return (NULL);
470 		if (inp->inp_sp->genid != key_getspgen()) {
471 			/* Invalidate the cache. */
472 			downgrade = 0;
473 			if (!INP_WLOCKED(inp)) {
474 				if ((downgrade = INP_TRY_UPGRADE(inp)) == 0)
475 					return (NULL);
476 			}
477 			ipsec_invalidate_cache(inp, IPSEC_DIR_OUTBOUND);
478 			ipsec_invalidate_cache(inp, IPSEC_DIR_INBOUND);
479 			if (downgrade != 0)
480 				INP_DOWNGRADE(inp);
481 			return (NULL);
482 		}
483 		KEYDBG(IPSEC_STAMP,
484 		    printf("%s: PCB(%p): cache hit SP(%p)\n",
485 		    __func__, inp, sp));
486 		/* Return referenced cached policy */
487 	}
488 	key_addref(sp);
489 	return (sp);
490 }
491 
492 #ifdef INET
493 static void
494 ipsec4_get_ulp(const struct mbuf *m, struct secpolicyindex *spidx,
495     int needport)
496 {
497 	uint8_t nxt;
498 	int off;
499 
500 	/* Sanity check. */
501 	IPSEC_ASSERT(m->m_pkthdr.len >= sizeof(struct ip),
502 	    ("packet too short"));
503 
504 	if (m->m_len >= sizeof (struct ip)) {
505 		const struct ip *ip = mtod(m, const struct ip *);
506 		if (ip->ip_off & htons(IP_MF | IP_OFFMASK))
507 			goto done;
508 		off = ip->ip_hl << 2;
509 		nxt = ip->ip_p;
510 	} else {
511 		struct ip ih;
512 
513 		m_copydata(m, 0, sizeof (struct ip), (caddr_t) &ih);
514 		if (ih.ip_off & htons(IP_MF | IP_OFFMASK))
515 			goto done;
516 		off = ih.ip_hl << 2;
517 		nxt = ih.ip_p;
518 	}
519 
520 	while (off < m->m_pkthdr.len) {
521 		struct ip6_ext ip6e;
522 		struct tcphdr th;
523 		struct udphdr uh;
524 
525 		switch (nxt) {
526 		case IPPROTO_TCP:
527 			spidx->ul_proto = nxt;
528 			if (!needport)
529 				goto done_proto;
530 			if (off + sizeof(struct tcphdr) > m->m_pkthdr.len)
531 				goto done;
532 			m_copydata(m, off, sizeof (th), (caddr_t) &th);
533 			spidx->src.sin.sin_port = th.th_sport;
534 			spidx->dst.sin.sin_port = th.th_dport;
535 			return;
536 		case IPPROTO_UDP:
537 			spidx->ul_proto = nxt;
538 			if (!needport)
539 				goto done_proto;
540 			if (off + sizeof(struct udphdr) > m->m_pkthdr.len)
541 				goto done;
542 			m_copydata(m, off, sizeof (uh), (caddr_t) &uh);
543 			spidx->src.sin.sin_port = uh.uh_sport;
544 			spidx->dst.sin.sin_port = uh.uh_dport;
545 			return;
546 		case IPPROTO_AH:
547 			if (off + sizeof(ip6e) > m->m_pkthdr.len)
548 				goto done;
549 			/* XXX Sigh, this works but is totally bogus. */
550 			m_copydata(m, off, sizeof(ip6e), (caddr_t) &ip6e);
551 			off += (ip6e.ip6e_len + 2) << 2;
552 			nxt = ip6e.ip6e_nxt;
553 			break;
554 		case IPPROTO_ICMP:
555 		default:
556 			/* XXX Intermediate headers??? */
557 			spidx->ul_proto = nxt;
558 			goto done_proto;
559 		}
560 	}
561 done:
562 	spidx->ul_proto = IPSEC_ULPROTO_ANY;
563 done_proto:
564 	spidx->src.sin.sin_port = IPSEC_PORT_ANY;
565 	spidx->dst.sin.sin_port = IPSEC_PORT_ANY;
566 	KEYDBG(IPSEC_DUMP,
567 	    printf("%s: ", __func__); kdebug_secpolicyindex(spidx, NULL));
568 }
569 
570 static void
571 ipsec4_setspidx_ipaddr(const struct mbuf *m, struct secpolicyindex *spidx)
572 {
573 
574 	ipsec4_setsockaddrs(m, &spidx->src, &spidx->dst);
575 	spidx->prefs = sizeof(struct in_addr) << 3;
576 	spidx->prefd = sizeof(struct in_addr) << 3;
577 }
578 
579 static struct secpolicy *
580 ipsec4_getpolicy(const struct mbuf *m, struct inpcb *inp, u_int dir,
581     int needport)
582 {
583 	struct secpolicyindex spidx;
584 	struct secpolicy *sp;
585 
586 	sp = ipsec_getpcbpolicy(inp, dir);
587 	if (sp == NULL && key_havesp(dir)) {
588 		/* Make an index to look for a policy. */
589 		ipsec4_setspidx_ipaddr(m, &spidx);
590 		ipsec4_get_ulp(m, &spidx, needport);
591 		spidx.dir = dir;
592 		sp = key_allocsp(&spidx, dir);
593 	}
594 	if (sp == NULL)		/* No SP found, use system default. */
595 		sp = key_allocsp_default();
596 	return (sp);
597 }
598 
599 /*
600  * Check security policy for *OUTBOUND* IPv4 packet.
601  */
602 struct secpolicy *
603 ipsec4_checkpolicy(const struct mbuf *m, struct inpcb *inp, int *error,
604     int needport)
605 {
606 	struct secpolicy *sp;
607 
608 	*error = 0;
609 	sp = ipsec4_getpolicy(m, inp, IPSEC_DIR_OUTBOUND, needport);
610 	if (sp != NULL)
611 		sp = ipsec_checkpolicy(sp, inp, error);
612 	if (sp == NULL) {
613 		switch (*error) {
614 		case 0: /* No IPsec required: BYPASS or NONE */
615 			break;
616 		case -EINVAL:
617 			IPSECSTAT_INC(ips_out_polvio);
618 			break;
619 		default:
620 			IPSECSTAT_INC(ips_out_inval);
621 		}
622 	}
623 	KEYDBG(IPSEC_STAMP,
624 	    printf("%s: using SP(%p), error %d\n", __func__, sp, *error));
625 	if (sp != NULL)
626 		KEYDBG(IPSEC_DATA, kdebug_secpolicy(sp));
627 	return (sp);
628 }
629 
630 /*
631  * Check IPv4 packet against *INBOUND* security policy.
632  * This function is called from tcp_input(), udp_input(),
633  * rip_input() and sctp_input().
634  */
635 int
636 ipsec4_in_reject(const struct mbuf *m, struct inpcb *inp)
637 {
638 	struct secpolicy *sp;
639 	int result;
640 
641 	sp = ipsec4_getpolicy(m, inp, IPSEC_DIR_INBOUND, 0);
642 	result = ipsec_in_reject(sp, inp, m);
643 	key_freesp(&sp);
644 	if (result != 0)
645 		IPSECSTAT_INC(ips_in_polvio);
646 	return (result);
647 }
648 
649 /*
650  * IPSEC_CAP() method implementation for IPv4.
651  */
652 int
653 ipsec4_capability(struct mbuf *m, u_int cap)
654 {
655 
656 	switch (cap) {
657 	case IPSEC_CAP_BYPASS_FILTER:
658 		/*
659 		 * Bypass packet filtering for packets previously handled
660 		 * by IPsec.
661 		 */
662 		if (!V_ip4_filtertunnel &&
663 		    m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL) != NULL)
664 			return (1);
665 		return (0);
666 	case IPSEC_CAP_OPERABLE:
667 		/* Do we have active security policies? */
668 		if (key_havesp(IPSEC_DIR_INBOUND) != 0 ||
669 		    key_havesp(IPSEC_DIR_OUTBOUND) != 0)
670 			return (1);
671 		return (0);
672 	};
673 	return (EOPNOTSUPP);
674 }
675 
676 #endif /* INET */
677 
678 #ifdef INET6
679 static void
680 ipsec6_get_ulp(const struct mbuf *m, struct secpolicyindex *spidx,
681     int needport)
682 {
683 	struct tcphdr th;
684 	struct udphdr uh;
685 	struct icmp6_hdr ih;
686 	int off, nxt;
687 
688 	IPSEC_ASSERT(m->m_pkthdr.len >= sizeof(struct ip6_hdr),
689 	    ("packet too short"));
690 
691 	/* Set default. */
692 	spidx->ul_proto = IPSEC_ULPROTO_ANY;
693 	spidx->src.sin6.sin6_port = IPSEC_PORT_ANY;
694 	spidx->dst.sin6.sin6_port = IPSEC_PORT_ANY;
695 
696 	nxt = -1;
697 	off = ip6_lasthdr(m, 0, IPPROTO_IPV6, &nxt);
698 	if (off < 0 || m->m_pkthdr.len < off)
699 		return;
700 
701 	switch (nxt) {
702 	case IPPROTO_TCP:
703 		spidx->ul_proto = nxt;
704 		if (!needport)
705 			break;
706 		if (off + sizeof(struct tcphdr) > m->m_pkthdr.len)
707 			break;
708 		m_copydata(m, off, sizeof(th), (caddr_t)&th);
709 		spidx->src.sin6.sin6_port = th.th_sport;
710 		spidx->dst.sin6.sin6_port = th.th_dport;
711 		break;
712 	case IPPROTO_UDP:
713 		spidx->ul_proto = nxt;
714 		if (!needport)
715 			break;
716 		if (off + sizeof(struct udphdr) > m->m_pkthdr.len)
717 			break;
718 		m_copydata(m, off, sizeof(uh), (caddr_t)&uh);
719 		spidx->src.sin6.sin6_port = uh.uh_sport;
720 		spidx->dst.sin6.sin6_port = uh.uh_dport;
721 		break;
722 	case IPPROTO_ICMPV6:
723 		spidx->ul_proto = nxt;
724 		if (off + sizeof(struct icmp6_hdr) > m->m_pkthdr.len)
725 			break;
726 		m_copydata(m, off, sizeof(ih), (caddr_t)&ih);
727 		spidx->src.sin6.sin6_port = htons((uint16_t)ih.icmp6_type);
728 		spidx->dst.sin6.sin6_port = htons((uint16_t)ih.icmp6_code);
729 		break;
730 	default:
731 		/* XXX Intermediate headers??? */
732 		spidx->ul_proto = nxt;
733 		break;
734 	}
735 	KEYDBG(IPSEC_DUMP,
736 	    printf("%s: ", __func__); kdebug_secpolicyindex(spidx, NULL));
737 }
738 
739 static void
740 ipsec6_setspidx_ipaddr(const struct mbuf *m, struct secpolicyindex *spidx)
741 {
742 
743 	ipsec6_setsockaddrs(m, &spidx->src, &spidx->dst);
744 	spidx->prefs = sizeof(struct in6_addr) << 3;
745 	spidx->prefd = sizeof(struct in6_addr) << 3;
746 }
747 
748 static struct secpolicy *
749 ipsec6_getpolicy(const struct mbuf *m, struct inpcb *inp, u_int dir,
750     int needport)
751 {
752 	struct secpolicyindex spidx;
753 	struct secpolicy *sp;
754 
755 	sp = ipsec_getpcbpolicy(inp, dir);
756 	if (sp == NULL && key_havesp(dir)) {
757 		/* Make an index to look for a policy. */
758 		ipsec6_setspidx_ipaddr(m, &spidx);
759 		ipsec6_get_ulp(m, &spidx, needport);
760 		spidx.dir = dir;
761 		sp = key_allocsp(&spidx, dir);
762 	}
763 	if (sp == NULL)		/* No SP found, use system default. */
764 		sp = key_allocsp_default();
765 	return (sp);
766 }
767 
768 /*
769  * Check security policy for *OUTBOUND* IPv6 packet.
770  */
771 struct secpolicy *
772 ipsec6_checkpolicy(const struct mbuf *m, struct inpcb *inp, int *error,
773     int needport)
774 {
775 	struct secpolicy *sp;
776 
777 	*error = 0;
778 	sp = ipsec6_getpolicy(m, inp, IPSEC_DIR_OUTBOUND, needport);
779 	if (sp != NULL)
780 		sp = ipsec_checkpolicy(sp, inp, error);
781 	if (sp == NULL) {
782 		switch (*error) {
783 		case 0: /* No IPsec required: BYPASS or NONE */
784 			break;
785 		case -EINVAL:
786 			IPSEC6STAT_INC(ips_out_polvio);
787 			break;
788 		default:
789 			IPSEC6STAT_INC(ips_out_inval);
790 		}
791 	}
792 	KEYDBG(IPSEC_STAMP,
793 	    printf("%s: using SP(%p), error %d\n", __func__, sp, *error));
794 	if (sp != NULL)
795 		KEYDBG(IPSEC_DATA, kdebug_secpolicy(sp));
796 	return (sp);
797 }
798 
799 /*
800  * Check IPv6 packet against inbound security policy.
801  * This function is called from tcp6_input(), udp6_input(),
802  * rip6_input() and sctp_input().
803  */
804 int
805 ipsec6_in_reject(const struct mbuf *m, struct inpcb *inp)
806 {
807 	struct secpolicy *sp;
808 	int result;
809 
810 	sp = ipsec6_getpolicy(m, inp, IPSEC_DIR_INBOUND, 0);
811 	result = ipsec_in_reject(sp, inp, m);
812 	key_freesp(&sp);
813 	if (result)
814 		IPSEC6STAT_INC(ips_in_polvio);
815 	return (result);
816 }
817 
818 /*
819  * IPSEC_CAP() method implementation for IPv6.
820  */
821 int
822 ipsec6_capability(struct mbuf *m, u_int cap)
823 {
824 
825 	switch (cap) {
826 	case IPSEC_CAP_BYPASS_FILTER:
827 		/*
828 		 * Bypass packet filtering for packets previously handled
829 		 * by IPsec.
830 		 */
831 		if (!V_ip6_filtertunnel &&
832 		    m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL) != NULL)
833 			return (1);
834 		return (0);
835 	case IPSEC_CAP_OPERABLE:
836 		/* Do we have active security policies? */
837 		if (key_havesp(IPSEC_DIR_INBOUND) != 0 ||
838 		    key_havesp(IPSEC_DIR_OUTBOUND) != 0)
839 			return (1);
840 		return (0);
841 	};
842 	return (EOPNOTSUPP);
843 }
844 #endif /* INET6 */
845 
846 int
847 ipsec_run_hhooks(struct ipsec_ctx_data *ctx, int type)
848 {
849 	int idx;
850 
851 	switch (ctx->af) {
852 #ifdef INET
853 	case AF_INET:
854 		idx = HHOOK_IPSEC_INET;
855 		break;
856 #endif
857 #ifdef INET6
858 	case AF_INET6:
859 		idx = HHOOK_IPSEC_INET6;
860 		break;
861 #endif
862 	default:
863 		return (EPFNOSUPPORT);
864 	}
865 	if (type == HHOOK_TYPE_IPSEC_IN)
866 		HHOOKS_RUN_IF(V_ipsec_hhh_in[idx], ctx, NULL);
867 	else
868 		HHOOKS_RUN_IF(V_ipsec_hhh_out[idx], ctx, NULL);
869 	if (*ctx->mp == NULL)
870 		return (EACCES);
871 	return (0);
872 }
873 
874 /*
875  * Return current level.
876  * Either IPSEC_LEVEL_USE or IPSEC_LEVEL_REQUIRE are always returned.
877  */
878 u_int
879 ipsec_get_reqlevel(struct secpolicy *sp, u_int idx)
880 {
881 	struct ipsecrequest *isr;
882 	u_int esp_trans_deflev, esp_net_deflev;
883 	u_int ah_trans_deflev, ah_net_deflev;
884 	u_int level = 0;
885 
886 	IPSEC_ASSERT(idx < sp->tcount, ("Wrong IPsec request index %d", idx));
887 /* XXX Note that we have ipseclog() expanded here - code sync issue. */
888 #define IPSEC_CHECK_DEFAULT(lev) \
889 	(((lev) != IPSEC_LEVEL_USE && (lev) != IPSEC_LEVEL_REQUIRE &&	\
890 	  (lev) != IPSEC_LEVEL_UNIQUE)					\
891 		? (V_ipsec_debug  ?					\
892 		log(LOG_INFO, "fixed system default level " #lev ":%d->%d\n",\
893 		(lev), IPSEC_LEVEL_REQUIRE) : 0),			\
894 		(lev) = IPSEC_LEVEL_REQUIRE, (lev) : (lev))
895 
896 	/*
897 	 * IPsec VTI uses unique security policy with fake spidx filled
898 	 * with zeroes. Just return IPSEC_LEVEL_REQUIRE instead of doing
899 	 * full level lookup for such policies.
900 	 */
901 	if (sp->state == IPSEC_SPSTATE_IFNET) {
902 		IPSEC_ASSERT(sp->req[idx]->level == IPSEC_LEVEL_UNIQUE,
903 		    ("Wrong IPsec request level %d", sp->req[idx]->level));
904 		return (IPSEC_LEVEL_REQUIRE);
905 	}
906 
907 	/* Set default level. */
908 	switch (sp->spidx.src.sa.sa_family) {
909 #ifdef INET
910 	case AF_INET:
911 		esp_trans_deflev = IPSEC_CHECK_DEFAULT(V_ip4_esp_trans_deflev);
912 		esp_net_deflev = IPSEC_CHECK_DEFAULT(V_ip4_esp_net_deflev);
913 		ah_trans_deflev = IPSEC_CHECK_DEFAULT(V_ip4_ah_trans_deflev);
914 		ah_net_deflev = IPSEC_CHECK_DEFAULT(V_ip4_ah_net_deflev);
915 		break;
916 #endif
917 #ifdef INET6
918 	case AF_INET6:
919 		esp_trans_deflev = IPSEC_CHECK_DEFAULT(V_ip6_esp_trans_deflev);
920 		esp_net_deflev = IPSEC_CHECK_DEFAULT(V_ip6_esp_net_deflev);
921 		ah_trans_deflev = IPSEC_CHECK_DEFAULT(V_ip6_ah_trans_deflev);
922 		ah_net_deflev = IPSEC_CHECK_DEFAULT(V_ip6_ah_net_deflev);
923 		break;
924 #endif /* INET6 */
925 	default:
926 		panic("%s: unknown af %u",
927 			__func__, sp->spidx.src.sa.sa_family);
928 	}
929 
930 #undef IPSEC_CHECK_DEFAULT
931 
932 	isr = sp->req[idx];
933 	/* Set level. */
934 	switch (isr->level) {
935 	case IPSEC_LEVEL_DEFAULT:
936 		switch (isr->saidx.proto) {
937 		case IPPROTO_ESP:
938 			if (isr->saidx.mode == IPSEC_MODE_TUNNEL)
939 				level = esp_net_deflev;
940 			else
941 				level = esp_trans_deflev;
942 			break;
943 		case IPPROTO_AH:
944 			if (isr->saidx.mode == IPSEC_MODE_TUNNEL)
945 				level = ah_net_deflev;
946 			else
947 				level = ah_trans_deflev;
948 			break;
949 		case IPPROTO_IPCOMP:
950 			/*
951 			 * We don't really care, as IPcomp document says that
952 			 * we shouldn't compress small packets.
953 			 */
954 			level = IPSEC_LEVEL_USE;
955 			break;
956 		default:
957 			panic("%s: Illegal protocol defined %u\n", __func__,
958 				isr->saidx.proto);
959 		}
960 		break;
961 
962 	case IPSEC_LEVEL_USE:
963 	case IPSEC_LEVEL_REQUIRE:
964 		level = isr->level;
965 		break;
966 	case IPSEC_LEVEL_UNIQUE:
967 		level = IPSEC_LEVEL_REQUIRE;
968 		break;
969 
970 	default:
971 		panic("%s: Illegal IPsec level %u\n", __func__, isr->level);
972 	}
973 
974 	return (level);
975 }
976 
977 static int
978 ipsec_check_history(const struct mbuf *m, struct secpolicy *sp, u_int idx)
979 {
980 	struct xform_history *xh;
981 	struct m_tag *mtag;
982 
983 	mtag = NULL;
984 	while ((mtag = m_tag_find(__DECONST(struct mbuf *, m),
985 	    PACKET_TAG_IPSEC_IN_DONE, mtag)) != NULL) {
986 		xh = (struct xform_history *)(mtag + 1);
987 		KEYDBG(IPSEC_DATA,
988 		    char buf[IPSEC_ADDRSTRLEN];
989 		    printf("%s: mode %s proto %u dst %s\n", __func__,
990 			kdebug_secasindex_mode(xh->mode), xh->proto,
991 			ipsec_address(&xh->dst, buf, sizeof(buf))));
992 		if (xh->proto != sp->req[idx]->saidx.proto)
993 			continue;
994 		/* If SA had IPSEC_MODE_ANY, consider this as match. */
995 		if (xh->mode != sp->req[idx]->saidx.mode &&
996 		    xh->mode != IPSEC_MODE_ANY)
997 			continue;
998 		/*
999 		 * For transport mode IPsec request doesn't contain
1000 		 * addresses. We need to use address from spidx.
1001 		 */
1002 		if (sp->req[idx]->saidx.mode == IPSEC_MODE_TRANSPORT) {
1003 			if (key_sockaddrcmp_withmask(&xh->dst.sa,
1004 			    &sp->spidx.dst.sa, sp->spidx.prefd) != 0)
1005 				continue;
1006 		} else {
1007 			if (key_sockaddrcmp(&xh->dst.sa,
1008 			    &sp->req[idx]->saidx.dst.sa, 0) != 0)
1009 				continue;
1010 		}
1011 		return (0); /* matched */
1012 	}
1013 	return (1);
1014 }
1015 
1016 /*
1017  * Check security policy requirements against the actual
1018  * packet contents.  Return one if the packet should be
1019  * reject as "invalid"; otherwiser return zero to have the
1020  * packet treated as "valid".
1021  *
1022  * OUT:
1023  *	0: valid
1024  *	1: invalid
1025  */
1026 static int
1027 ipsec_in_reject(struct secpolicy *sp, struct inpcb *inp, const struct mbuf *m)
1028 {
1029 	int i;
1030 
1031 	KEYDBG(IPSEC_STAMP,
1032 	    printf("%s: PCB(%p): using SP(%p)\n", __func__, inp, sp));
1033 	KEYDBG(IPSEC_DATA, kdebug_secpolicy(sp));
1034 
1035 	if (inp != NULL && inp->inp_sp != NULL && inp->inp_sp->sp_in == NULL)
1036 		ipsec_cachepolicy(inp, sp, IPSEC_DIR_INBOUND);
1037 
1038 	/* Check policy. */
1039 	switch (sp->policy) {
1040 	case IPSEC_POLICY_DISCARD:
1041 		return (1);
1042 	case IPSEC_POLICY_BYPASS:
1043 	case IPSEC_POLICY_NONE:
1044 		return (0);
1045 	}
1046 
1047 	IPSEC_ASSERT(sp->policy == IPSEC_POLICY_IPSEC,
1048 		("invalid policy %u", sp->policy));
1049 
1050 	/*
1051 	 * ipsec[46]_common_input_cb after each transform adds
1052 	 * PACKET_TAG_IPSEC_IN_DONE mbuf tag. It contains SPI, proto, mode
1053 	 * and destination address from saidx. We can compare info from
1054 	 * these tags with requirements in SP.
1055 	 */
1056 	for (i = 0; i < sp->tcount; i++) {
1057 		/*
1058 		 * Do not check IPcomp, since IPcomp document
1059 		 * says that we shouldn't compress small packets.
1060 		 * IPComp policy should always be treated as being
1061 		 * in "use" level.
1062 		 */
1063 		if (sp->req[i]->saidx.proto == IPPROTO_IPCOMP ||
1064 		    ipsec_get_reqlevel(sp, i) != IPSEC_LEVEL_REQUIRE)
1065 			continue;
1066 		if (V_check_policy_history != 0 &&
1067 		    ipsec_check_history(m, sp, i) != 0)
1068 			return (1);
1069 		else switch (sp->req[i]->saidx.proto) {
1070 		case IPPROTO_ESP:
1071 			if ((m->m_flags & M_DECRYPTED) == 0) {
1072 				KEYDBG(IPSEC_DUMP,
1073 				    printf("%s: ESP m_flags:%x\n", __func__,
1074 					    m->m_flags));
1075 				return (1);
1076 			}
1077 			break;
1078 		case IPPROTO_AH:
1079 			if ((m->m_flags & M_AUTHIPHDR) == 0) {
1080 				KEYDBG(IPSEC_DUMP,
1081 				    printf("%s: AH m_flags:%x\n", __func__,
1082 					    m->m_flags));
1083 				return (1);
1084 			}
1085 			break;
1086 		}
1087 	}
1088 	return (0);		/* Valid. */
1089 }
1090 
1091 /*
1092  * Compute the byte size to be occupied by IPsec header.
1093  * In case it is tunnelled, it includes the size of outer IP header.
1094  */
1095 static size_t
1096 ipsec_hdrsiz_internal(struct secpolicy *sp)
1097 {
1098 	size_t size;
1099 	int i;
1100 
1101 	KEYDBG(IPSEC_STAMP, printf("%s: using SP(%p)\n", __func__, sp));
1102 	KEYDBG(IPSEC_DATA, kdebug_secpolicy(sp));
1103 
1104 	switch (sp->policy) {
1105 	case IPSEC_POLICY_DISCARD:
1106 	case IPSEC_POLICY_BYPASS:
1107 	case IPSEC_POLICY_NONE:
1108 		return (0);
1109 	}
1110 
1111 	IPSEC_ASSERT(sp->policy == IPSEC_POLICY_IPSEC,
1112 		("invalid policy %u", sp->policy));
1113 
1114 	/*
1115 	 * XXX: for each transform we need to lookup suitable SA
1116 	 * and use info from SA to calculate headers size.
1117 	 * XXX: for NAT-T we need to cosider UDP header size.
1118 	 */
1119 	size = 0;
1120 	for (i = 0; i < sp->tcount; i++) {
1121 		switch (sp->req[i]->saidx.proto) {
1122 		case IPPROTO_ESP:
1123 			size += esp_hdrsiz(NULL);
1124 			break;
1125 		case IPPROTO_AH:
1126 			size += ah_hdrsiz(NULL);
1127 			break;
1128 		case IPPROTO_IPCOMP:
1129 			size += sizeof(struct ipcomp);
1130 			break;
1131 		}
1132 
1133 		if (sp->req[i]->saidx.mode == IPSEC_MODE_TUNNEL) {
1134 			switch (sp->req[i]->saidx.dst.sa.sa_family) {
1135 #ifdef INET
1136 			case AF_INET:
1137 				size += sizeof(struct ip);
1138 				break;
1139 #endif
1140 #ifdef INET6
1141 			case AF_INET6:
1142 				size += sizeof(struct ip6_hdr);
1143 				break;
1144 #endif
1145 			default:
1146 				ipseclog((LOG_ERR, "%s: unknown AF %d in "
1147 				    "IPsec tunnel SA\n", __func__,
1148 				    sp->req[i]->saidx.dst.sa.sa_family));
1149 				break;
1150 			}
1151 		}
1152 	}
1153 	return (size);
1154 }
1155 
1156 /*
1157  * Compute ESP/AH header size for protocols with PCB, including
1158  * outer IP header. Currently only tcp_output() uses it.
1159  */
1160 size_t
1161 ipsec_hdrsiz_inpcb(struct inpcb *inp)
1162 {
1163 	struct secpolicyindex spidx;
1164 	struct secpolicy *sp;
1165 	size_t sz;
1166 
1167 	sp = ipsec_getpcbpolicy(inp, IPSEC_DIR_OUTBOUND);
1168 	if (sp == NULL && key_havesp(IPSEC_DIR_OUTBOUND)) {
1169 		ipsec_setspidx_inpcb(inp, &spidx, IPSEC_DIR_OUTBOUND);
1170 		sp = key_allocsp(&spidx, IPSEC_DIR_OUTBOUND);
1171 	}
1172 	if (sp == NULL)
1173 		sp = key_allocsp_default();
1174 	sz = ipsec_hdrsiz_internal(sp);
1175 	key_freesp(&sp);
1176 	return (sz);
1177 }
1178 
1179 /*
1180  * Check the variable replay window.
1181  * ipsec_chkreplay() performs replay check before ICV verification.
1182  * ipsec_updatereplay() updates replay bitmap.  This must be called after
1183  * ICV verification (it also performs replay check, which is usually done
1184  * beforehand).
1185  * 0 (zero) is returned if packet disallowed, 1 if packet permitted.
1186  *
1187  * Based on RFC 6479. Blocks are 32 bits unsigned integers
1188  */
1189 
1190 #define IPSEC_BITMAP_INDEX_MASK(w)	(w - 1)
1191 #define IPSEC_REDUNDANT_BIT_SHIFTS	5
1192 #define IPSEC_REDUNDANT_BITS		(1 << IPSEC_REDUNDANT_BIT_SHIFTS)
1193 #define IPSEC_BITMAP_LOC_MASK		(IPSEC_REDUNDANT_BITS - 1)
1194 
1195 int
1196 ipsec_chkreplay(uint32_t seq, struct secasvar *sav)
1197 {
1198 	const struct secreplay *replay;
1199 	uint32_t wsizeb;		/* Constant: window size. */
1200 	int index, bit_location;
1201 
1202 	IPSEC_ASSERT(sav != NULL, ("Null SA"));
1203 	IPSEC_ASSERT(sav->replay != NULL, ("Null replay state"));
1204 
1205 	replay = sav->replay;
1206 
1207 	/* No need to check replay if disabled. */
1208 	if (replay->wsize == 0)
1209 		return (1);
1210 
1211 	/* Constant. */
1212 	wsizeb = replay->wsize << 3;
1213 
1214 	/* Sequence number of 0 is invalid. */
1215 	if (seq == 0)
1216 		return (0);
1217 
1218 	/* First time is always okay. */
1219 	if (replay->count == 0)
1220 		return (1);
1221 
1222 	/* Larger sequences are okay. */
1223 	if (seq > replay->lastseq)
1224 		return (1);
1225 
1226 	/* Over range to check, i.e. too old or wrapped. */
1227 	if (replay->lastseq - seq >= wsizeb)
1228 		return (0);
1229 
1230 	/* The sequence is inside the sliding window
1231 	 * now check the bit in the bitmap
1232 	 * bit location only depends on the sequence number
1233 	 */
1234 	bit_location = seq & IPSEC_BITMAP_LOC_MASK;
1235 	index = (seq >> IPSEC_REDUNDANT_BIT_SHIFTS)
1236 		& IPSEC_BITMAP_INDEX_MASK(replay->bitmap_size);
1237 
1238 	/* This packet already seen? */
1239 	if ((replay->bitmap)[index] & (1 << bit_location))
1240 		return (0);
1241 	return (1);
1242 }
1243 
1244 /*
1245  * Check replay counter whether to update or not.
1246  * OUT:	0:	OK
1247  *	1:	NG
1248  */
1249 int
1250 ipsec_updatereplay(uint32_t seq, struct secasvar *sav)
1251 {
1252 	char buf[128];
1253 	struct secreplay *replay;
1254 	uint32_t wsizeb;		/* Constant: window size. */
1255 	int diff, index, bit_location;
1256 
1257 	IPSEC_ASSERT(sav != NULL, ("Null SA"));
1258 	IPSEC_ASSERT(sav->replay != NULL, ("Null replay state"));
1259 
1260 	replay = sav->replay;
1261 
1262 	if (replay->wsize == 0)
1263 		goto ok;	/* No need to check replay. */
1264 
1265 	/* Constant. */
1266 	wsizeb = replay->wsize << 3;
1267 
1268 	/* Sequence number of 0 is invalid. */
1269 	if (seq == 0)
1270 		return (1);
1271 
1272 	/* The packet is too old, no need to update */
1273 	if (wsizeb + seq < replay->lastseq)
1274 		goto ok;
1275 
1276 	/* Now update the bit */
1277 	index = (seq >> IPSEC_REDUNDANT_BIT_SHIFTS);
1278 
1279 	/* First check if the sequence number is in the range */
1280 	if (seq > replay->lastseq) {
1281 		int id;
1282 		int index_cur = replay->lastseq >> IPSEC_REDUNDANT_BIT_SHIFTS;
1283 
1284 		diff = index - index_cur;
1285 		if (diff > replay->bitmap_size) {
1286 			/* something unusual in this case */
1287 			diff = replay->bitmap_size;
1288 		}
1289 
1290 		for (id = 0; id < diff; ++id) {
1291 			replay->bitmap[(id + index_cur + 1)
1292 			& IPSEC_BITMAP_INDEX_MASK(replay->bitmap_size)] = 0;
1293 		}
1294 
1295 		replay->lastseq = seq;
1296 	}
1297 
1298 	index &= IPSEC_BITMAP_INDEX_MASK(replay->bitmap_size);
1299 	bit_location = seq & IPSEC_BITMAP_LOC_MASK;
1300 
1301 	/* this packet has already been received */
1302 	if (replay->bitmap[index] & (1 << bit_location))
1303 		return (1);
1304 
1305 	replay->bitmap[index] |= (1 << bit_location);
1306 
1307 ok:
1308 	if (replay->count == ~0) {
1309 
1310 		/* Set overflow flag. */
1311 		replay->overflow++;
1312 
1313 		/* Don't increment, no more packets accepted. */
1314 		if ((sav->flags & SADB_X_EXT_CYCSEQ) == 0) {
1315 			if (sav->sah->saidx.proto == IPPROTO_AH)
1316 				AHSTAT_INC(ahs_wrap);
1317 			else if (sav->sah->saidx.proto == IPPROTO_ESP)
1318 				ESPSTAT_INC(esps_wrap);
1319 			return (1);
1320 		}
1321 
1322 		ipseclog((LOG_WARNING, "%s: replay counter made %d cycle. %s\n",
1323 		    __func__, replay->overflow,
1324 		    ipsec_sa2str(sav, buf, sizeof(buf))));
1325 	}
1326 
1327 	replay->count++;
1328 	return (0);
1329 }
1330 
1331 int
1332 ipsec_updateid(struct secasvar *sav, crypto_session_t *new,
1333     crypto_session_t *old)
1334 {
1335 	crypto_session_t tmp;
1336 
1337 	/*
1338 	 * tdb_cryptoid is initialized by xform_init().
1339 	 * Then it can be changed only when some crypto error occurred or
1340 	 * when SA is deleted. We stored used cryptoid in the xform_data
1341 	 * structure. In case when crypto error occurred and crypto
1342 	 * subsystem has reinited the session, it returns new cryptoid
1343 	 * and EAGAIN error code.
1344 	 *
1345 	 * This function will be called when we got EAGAIN from crypto
1346 	 * subsystem.
1347 	 * *new is cryptoid that was returned by crypto subsystem in
1348 	 * the crp_sid.
1349 	 * *old is the original cryptoid that we stored in xform_data.
1350 	 *
1351 	 * For first failed request *old == sav->tdb_cryptoid, then
1352 	 * we update sav->tdb_cryptoid and redo crypto_dispatch().
1353 	 * For next failed request *old != sav->tdb_cryptoid, then
1354 	 * we store cryptoid from first request into the *new variable
1355 	 * and crp_sid from this second session will be returned via
1356 	 * *old pointer, so caller can release second session.
1357 	 *
1358 	 * XXXAE: check this more carefully.
1359 	 */
1360 	KEYDBG(IPSEC_STAMP,
1361 	    printf("%s: SA(%p) moves cryptoid %p -> %p\n",
1362 		__func__, sav, *old, *new));
1363 	KEYDBG(IPSEC_DATA, kdebug_secasv(sav));
1364 	SECASVAR_LOCK(sav);
1365 	if (sav->tdb_cryptoid != *old) {
1366 		/* cryptoid was already updated */
1367 		tmp = *new;
1368 		*new = sav->tdb_cryptoid;
1369 		*old = tmp;
1370 		SECASVAR_UNLOCK(sav);
1371 		return (1);
1372 	}
1373 	sav->tdb_cryptoid = *new;
1374 	SECASVAR_UNLOCK(sav);
1375 	return (0);
1376 }
1377 
1378 int
1379 ipsec_initialized(void)
1380 {
1381 
1382 	return (V_def_policy != NULL);
1383 }
1384 
1385 static void
1386 def_policy_init(const void *unused __unused)
1387 {
1388 
1389 	V_def_policy = key_newsp();
1390 	if (V_def_policy != NULL) {
1391 		V_def_policy->policy = IPSEC_POLICY_NONE;
1392 		/* Force INPCB SP cache invalidation */
1393 		key_bumpspgen();
1394 	} else
1395 		printf("%s: failed to initialize default policy\n", __func__);
1396 }
1397 
1398 
1399 static void
1400 def_policy_uninit(const void *unused __unused)
1401 {
1402 
1403 	if (V_def_policy != NULL) {
1404 		key_freesp(&V_def_policy);
1405 		key_bumpspgen();
1406 	}
1407 }
1408 
1409 VNET_SYSINIT(def_policy_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST,
1410     def_policy_init, NULL);
1411 VNET_SYSUNINIT(def_policy_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST,
1412     def_policy_uninit, NULL);
1413