xref: /freebsd/sys/netipsec/ipsec.c (revision c697fb7f)
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 | CTLFLAG_NEEDGIANT,
179     0, 0, sysctl_def_policy, "I",
180     "IPsec default policy.");
181 SYSCTL_INT(_net_inet_ipsec, IPSECCTL_DEF_ESP_TRANSLEV, esp_trans_deflev,
182 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip4_esp_trans_deflev), 0,
183 	"Default ESP transport mode level");
184 SYSCTL_INT(_net_inet_ipsec, IPSECCTL_DEF_ESP_NETLEV, esp_net_deflev,
185 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip4_esp_net_deflev), 0,
186 	"Default ESP tunnel mode level.");
187 SYSCTL_INT(_net_inet_ipsec, IPSECCTL_DEF_AH_TRANSLEV, ah_trans_deflev,
188 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip4_ah_trans_deflev), 0,
189 	"AH transfer mode default level.");
190 SYSCTL_INT(_net_inet_ipsec, IPSECCTL_DEF_AH_NETLEV, ah_net_deflev,
191 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip4_ah_net_deflev), 0,
192 	"AH tunnel mode default level.");
193 SYSCTL_INT(_net_inet_ipsec, IPSECCTL_AH_CLEARTOS, ah_cleartos,
194 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ah_cleartos), 0,
195 	"If set, clear type-of-service field when doing AH computation.");
196 SYSCTL_INT(_net_inet_ipsec, IPSECCTL_DFBIT, dfbit,
197 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip4_ipsec_dfbit), 0,
198 	"Do not fragment bit on encap.");
199 SYSCTL_INT(_net_inet_ipsec, IPSECCTL_ECN, ecn,
200 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip4_ipsec_ecn), 0,
201 	"Explicit Congestion Notification handling.");
202 SYSCTL_INT(_net_inet_ipsec, OID_AUTO, crypto_support,
203 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(crypto_support), 0,
204 	"Crypto driver selection.");
205 SYSCTL_INT(_net_inet_ipsec, OID_AUTO, async_crypto,
206 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(async_crypto), 0,
207 	"Use asynchronous mode to parallelize crypto jobs.");
208 SYSCTL_INT(_net_inet_ipsec, OID_AUTO, check_policy_history,
209 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(check_policy_history), 0,
210 	"Use strict check of inbound packets to security policy compliance.");
211 SYSCTL_INT(_net_inet_ipsec, OID_AUTO, natt_cksum_policy,
212 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(natt_cksum_policy), 0,
213 	"Method to fix TCP/UDP checksum for transport mode IPsec after NAT.");
214 SYSCTL_INT(_net_inet_ipsec, OID_AUTO, filtertunnel,
215 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip4_filtertunnel), 0,
216 	"If set, filter packets from an IPsec tunnel.");
217 SYSCTL_VNET_PCPUSTAT(_net_inet_ipsec, OID_AUTO, ipsecstats, struct ipsecstat,
218     ipsec4stat, "IPsec IPv4 statistics.");
219 
220 struct timeval ipsec_warn_interval = { .tv_sec = 1, .tv_usec = 0 };
221 SYSCTL_TIMEVAL_SEC(_net_inet_ipsec, OID_AUTO, crypto_warn_interval, CTLFLAG_RW,
222     &ipsec_warn_interval,
223     "Delay in seconds between warnings of deprecated IPsec crypto algorithms.");
224 
225 #ifdef REGRESSION
226 /*
227  * When set to 1, IPsec will send packets with the same sequence number.
228  * This allows to verify if the other side has proper replay attacks detection.
229  */
230 VNET_DEFINE(int, ipsec_replay) = 0;
231 SYSCTL_INT(_net_inet_ipsec, OID_AUTO, test_replay,
232 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ipsec_replay), 0,
233 	"Emulate replay attack");
234 /*
235  * When set 1, IPsec will send packets with corrupted HMAC.
236  * This allows to verify if the other side properly detects modified packets.
237  */
238 VNET_DEFINE(int, ipsec_integrity) = 0;
239 SYSCTL_INT(_net_inet_ipsec, OID_AUTO, test_integrity,
240 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ipsec_integrity), 0,
241 	"Emulate man-in-the-middle attack");
242 #endif
243 
244 #ifdef INET6
245 VNET_PCPUSTAT_DEFINE(struct ipsecstat, ipsec6stat);
246 VNET_PCPUSTAT_SYSINIT(ipsec6stat);
247 
248 #ifdef VIMAGE
249 VNET_PCPUSTAT_SYSUNINIT(ipsec6stat);
250 #endif /* VIMAGE */
251 
252 VNET_DEFINE(int, ip6_esp_trans_deflev) = IPSEC_LEVEL_USE;
253 VNET_DEFINE(int, ip6_esp_net_deflev) = IPSEC_LEVEL_USE;
254 VNET_DEFINE(int, ip6_ah_trans_deflev) = IPSEC_LEVEL_USE;
255 VNET_DEFINE(int, ip6_ah_net_deflev) = IPSEC_LEVEL_USE;
256 VNET_DEFINE(int, ip6_ipsec_ecn) = 0;	/* ECN ignore(-1)/forbidden(0)/allowed(1) */
257 
258 VNET_DEFINE_STATIC(int, ip6_filtertunnel) = 0;
259 #define	V_ip6_filtertunnel	VNET(ip6_filtertunnel)
260 
261 SYSCTL_DECL(_net_inet6_ipsec6);
262 
263 /* net.inet6.ipsec6 */
264 SYSCTL_PROC(_net_inet6_ipsec6, IPSECCTL_DEF_POLICY, def_policy,
265     CTLTYPE_INT | CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
266     0, 0, sysctl_def_policy, "I",
267     "IPsec default policy.");
268 SYSCTL_INT(_net_inet6_ipsec6, IPSECCTL_DEF_ESP_TRANSLEV, esp_trans_deflev,
269 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_esp_trans_deflev), 0,
270 	"Default ESP transport mode level.");
271 SYSCTL_INT(_net_inet6_ipsec6, IPSECCTL_DEF_ESP_NETLEV, esp_net_deflev,
272 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_esp_net_deflev), 0,
273 	"Default ESP tunnel mode level.");
274 SYSCTL_INT(_net_inet6_ipsec6, IPSECCTL_DEF_AH_TRANSLEV, ah_trans_deflev,
275 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_ah_trans_deflev), 0,
276 	"AH transfer mode default level.");
277 SYSCTL_INT(_net_inet6_ipsec6, IPSECCTL_DEF_AH_NETLEV, ah_net_deflev,
278 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_ah_net_deflev), 0,
279 	"AH tunnel mode default level.");
280 SYSCTL_INT(_net_inet6_ipsec6, IPSECCTL_ECN, ecn,
281 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_ipsec_ecn), 0,
282 	"Explicit Congestion Notification handling.");
283 SYSCTL_INT(_net_inet6_ipsec6, OID_AUTO, filtertunnel,
284 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_filtertunnel),  0,
285 	"If set, filter packets from an IPsec tunnel.");
286 SYSCTL_VNET_PCPUSTAT(_net_inet6_ipsec6, IPSECCTL_STATS, ipsecstats,
287     struct ipsecstat, ipsec6stat, "IPsec IPv6 statistics.");
288 #endif /* INET6 */
289 
290 static int ipsec_in_reject(struct secpolicy *, struct inpcb *,
291     const struct mbuf *);
292 
293 #ifdef INET
294 static void ipsec4_get_ulp(const struct mbuf *, struct secpolicyindex *, int);
295 static void ipsec4_setspidx_ipaddr(const struct mbuf *,
296     struct secpolicyindex *);
297 #endif
298 #ifdef INET6
299 static void ipsec6_get_ulp(const struct mbuf *m, struct secpolicyindex *, int);
300 static void ipsec6_setspidx_ipaddr(const struct mbuf *,
301     struct secpolicyindex *);
302 #endif
303 
304 /*
305  * Return a held reference to the default SP.
306  */
307 static struct secpolicy *
308 key_allocsp_default(void)
309 {
310 
311 	key_addref(V_def_policy);
312 	return (V_def_policy);
313 }
314 
315 static void
316 ipsec_invalidate_cache(struct inpcb *inp, u_int dir)
317 {
318 	struct secpolicy *sp;
319 
320 	INP_WLOCK_ASSERT(inp);
321 	if (dir == IPSEC_DIR_OUTBOUND) {
322 		if (inp->inp_sp->flags & INP_INBOUND_POLICY)
323 			return;
324 		sp = inp->inp_sp->sp_in;
325 		inp->inp_sp->sp_in = NULL;
326 	} else {
327 		if (inp->inp_sp->flags & INP_OUTBOUND_POLICY)
328 			return;
329 		sp = inp->inp_sp->sp_out;
330 		inp->inp_sp->sp_out = NULL;
331 	}
332 	if (sp != NULL)
333 		key_freesp(&sp); /* release extra reference */
334 }
335 
336 static void
337 ipsec_cachepolicy(struct inpcb *inp, struct secpolicy *sp, u_int dir)
338 {
339 	uint32_t genid;
340 	int downgrade;
341 
342 	INP_LOCK_ASSERT(inp);
343 
344 	if (dir == IPSEC_DIR_OUTBOUND) {
345 		/* Do we have configured PCB policy? */
346 		if (inp->inp_sp->flags & INP_OUTBOUND_POLICY)
347 			return;
348 		/* Another thread has already set cached policy */
349 		if (inp->inp_sp->sp_out != NULL)
350 			return;
351 		/*
352 		 * Do not cache OUTBOUND policy if PCB isn't connected,
353 		 * i.e. foreign address is INADDR_ANY/UNSPECIFIED.
354 		 */
355 #ifdef INET
356 		if ((inp->inp_vflag & INP_IPV4) != 0 &&
357 		    inp->inp_faddr.s_addr == INADDR_ANY)
358 			return;
359 #endif
360 #ifdef INET6
361 		if ((inp->inp_vflag & INP_IPV6) != 0 &&
362 		    IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr))
363 			return;
364 #endif
365 	} else {
366 		/* Do we have configured PCB policy? */
367 		if (inp->inp_sp->flags & INP_INBOUND_POLICY)
368 			return;
369 		/* Another thread has already set cached policy */
370 		if (inp->inp_sp->sp_in != NULL)
371 			return;
372 		/*
373 		 * Do not cache INBOUND policy for listen socket,
374 		 * that is bound to INADDR_ANY/UNSPECIFIED address.
375 		 */
376 #ifdef INET
377 		if ((inp->inp_vflag & INP_IPV4) != 0 &&
378 		    inp->inp_faddr.s_addr == INADDR_ANY)
379 			return;
380 #endif
381 #ifdef INET6
382 		if ((inp->inp_vflag & INP_IPV6) != 0 &&
383 		    IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr))
384 			return;
385 #endif
386 	}
387 	downgrade = 0;
388 	if (!INP_WLOCKED(inp)) {
389 		if ((downgrade = INP_TRY_UPGRADE(inp)) == 0)
390 			return;
391 	}
392 	if (dir == IPSEC_DIR_OUTBOUND)
393 		inp->inp_sp->sp_out = sp;
394 	else
395 		inp->inp_sp->sp_in = sp;
396 	/*
397 	 * SP is already referenced by the lookup code.
398 	 * We take extra reference here to avoid race in the
399 	 * ipsec_getpcbpolicy() function - SP will not be freed in the
400 	 * time between we take SP pointer from the cache and key_addref()
401 	 * call.
402 	 */
403 	key_addref(sp);
404 	genid = key_getspgen();
405 	if (genid != inp->inp_sp->genid) {
406 		ipsec_invalidate_cache(inp, dir);
407 		inp->inp_sp->genid = genid;
408 	}
409 	KEYDBG(IPSEC_STAMP,
410 	    printf("%s: PCB(%p): cached %s SP(%p)\n",
411 	    __func__, inp, dir == IPSEC_DIR_OUTBOUND ? "OUTBOUND":
412 	    "INBOUND", sp));
413 	if (downgrade != 0)
414 		INP_DOWNGRADE(inp);
415 }
416 
417 static struct secpolicy *
418 ipsec_checkpolicy(struct secpolicy *sp, struct inpcb *inp, int *error)
419 {
420 
421 	/* Save found OUTBOUND policy into PCB SP cache. */
422 	if (inp != NULL && inp->inp_sp != NULL && inp->inp_sp->sp_out == NULL)
423 		ipsec_cachepolicy(inp, sp, IPSEC_DIR_OUTBOUND);
424 
425 	switch (sp->policy) {
426 	default:
427 		printf("%s: invalid policy %u\n", __func__, sp->policy);
428 		/* FALLTHROUGH */
429 	case IPSEC_POLICY_DISCARD:
430 		*error = -EINVAL;	/* Packet is discarded by caller. */
431 		/* FALLTHROUGH */
432 	case IPSEC_POLICY_BYPASS:
433 	case IPSEC_POLICY_NONE:
434 		key_freesp(&sp);
435 		sp = NULL;		/* NB: force NULL result. */
436 		break;
437 	case IPSEC_POLICY_IPSEC:
438 		/* XXXAE: handle LARVAL SP */
439 		break;
440 	}
441 	KEYDBG(IPSEC_DUMP,
442 	    printf("%s: get SP(%p), error %d\n", __func__, sp, *error));
443 	return (sp);
444 }
445 
446 static struct secpolicy *
447 ipsec_getpcbpolicy(struct inpcb *inp, u_int dir)
448 {
449 	struct secpolicy *sp;
450 	int flags, downgrade;
451 
452 	if (inp == NULL || inp->inp_sp == NULL)
453 		return (NULL);
454 
455 	INP_LOCK_ASSERT(inp);
456 
457 	flags = inp->inp_sp->flags;
458 	if (dir == IPSEC_DIR_OUTBOUND) {
459 		sp = inp->inp_sp->sp_out;
460 		flags &= INP_OUTBOUND_POLICY;
461 	} else {
462 		sp = inp->inp_sp->sp_in;
463 		flags &= INP_INBOUND_POLICY;
464 	}
465 	/*
466 	 * Check flags. If we have PCB SP, just return it.
467 	 * Otherwise we need to check that cached SP entry isn't stale.
468 	 */
469 	if (flags == 0) {
470 		if (sp == NULL)
471 			return (NULL);
472 		if (inp->inp_sp->genid != key_getspgen()) {
473 			/* Invalidate the cache. */
474 			downgrade = 0;
475 			if (!INP_WLOCKED(inp)) {
476 				if ((downgrade = INP_TRY_UPGRADE(inp)) == 0)
477 					return (NULL);
478 			}
479 			ipsec_invalidate_cache(inp, IPSEC_DIR_OUTBOUND);
480 			ipsec_invalidate_cache(inp, IPSEC_DIR_INBOUND);
481 			if (downgrade != 0)
482 				INP_DOWNGRADE(inp);
483 			return (NULL);
484 		}
485 		KEYDBG(IPSEC_STAMP,
486 		    printf("%s: PCB(%p): cache hit SP(%p)\n",
487 		    __func__, inp, sp));
488 		/* Return referenced cached policy */
489 	}
490 	key_addref(sp);
491 	return (sp);
492 }
493 
494 #ifdef INET
495 static void
496 ipsec4_get_ulp(const struct mbuf *m, struct secpolicyindex *spidx,
497     int needport)
498 {
499 	uint8_t nxt;
500 	int off;
501 
502 	/* Sanity check. */
503 	IPSEC_ASSERT(m->m_pkthdr.len >= sizeof(struct ip),
504 	    ("packet too short"));
505 
506 	if (m->m_len >= sizeof (struct ip)) {
507 		const struct ip *ip = mtod(m, const struct ip *);
508 		if (ip->ip_off & htons(IP_MF | IP_OFFMASK))
509 			goto done;
510 		off = ip->ip_hl << 2;
511 		nxt = ip->ip_p;
512 	} else {
513 		struct ip ih;
514 
515 		m_copydata(m, 0, sizeof (struct ip), (caddr_t) &ih);
516 		if (ih.ip_off & htons(IP_MF | IP_OFFMASK))
517 			goto done;
518 		off = ih.ip_hl << 2;
519 		nxt = ih.ip_p;
520 	}
521 
522 	while (off < m->m_pkthdr.len) {
523 		struct ip6_ext ip6e;
524 		struct tcphdr th;
525 		struct udphdr uh;
526 
527 		switch (nxt) {
528 		case IPPROTO_TCP:
529 			spidx->ul_proto = nxt;
530 			if (!needport)
531 				goto done_proto;
532 			if (off + sizeof(struct tcphdr) > m->m_pkthdr.len)
533 				goto done;
534 			m_copydata(m, off, sizeof (th), (caddr_t) &th);
535 			spidx->src.sin.sin_port = th.th_sport;
536 			spidx->dst.sin.sin_port = th.th_dport;
537 			return;
538 		case IPPROTO_UDP:
539 			spidx->ul_proto = nxt;
540 			if (!needport)
541 				goto done_proto;
542 			if (off + sizeof(struct udphdr) > m->m_pkthdr.len)
543 				goto done;
544 			m_copydata(m, off, sizeof (uh), (caddr_t) &uh);
545 			spidx->src.sin.sin_port = uh.uh_sport;
546 			spidx->dst.sin.sin_port = uh.uh_dport;
547 			return;
548 		case IPPROTO_AH:
549 			if (off + sizeof(ip6e) > m->m_pkthdr.len)
550 				goto done;
551 			/* XXX Sigh, this works but is totally bogus. */
552 			m_copydata(m, off, sizeof(ip6e), (caddr_t) &ip6e);
553 			off += (ip6e.ip6e_len + 2) << 2;
554 			nxt = ip6e.ip6e_nxt;
555 			break;
556 		case IPPROTO_ICMP:
557 		default:
558 			/* XXX Intermediate headers??? */
559 			spidx->ul_proto = nxt;
560 			goto done_proto;
561 		}
562 	}
563 done:
564 	spidx->ul_proto = IPSEC_ULPROTO_ANY;
565 done_proto:
566 	spidx->src.sin.sin_port = IPSEC_PORT_ANY;
567 	spidx->dst.sin.sin_port = IPSEC_PORT_ANY;
568 	KEYDBG(IPSEC_DUMP,
569 	    printf("%s: ", __func__); kdebug_secpolicyindex(spidx, NULL));
570 }
571 
572 static void
573 ipsec4_setspidx_ipaddr(const struct mbuf *m, struct secpolicyindex *spidx)
574 {
575 
576 	ipsec4_setsockaddrs(m, &spidx->src, &spidx->dst);
577 	spidx->prefs = sizeof(struct in_addr) << 3;
578 	spidx->prefd = sizeof(struct in_addr) << 3;
579 }
580 
581 static struct secpolicy *
582 ipsec4_getpolicy(const struct mbuf *m, struct inpcb *inp, u_int dir,
583     int needport)
584 {
585 	struct secpolicyindex spidx;
586 	struct secpolicy *sp;
587 
588 	sp = ipsec_getpcbpolicy(inp, dir);
589 	if (sp == NULL && key_havesp(dir)) {
590 		/* Make an index to look for a policy. */
591 		ipsec4_setspidx_ipaddr(m, &spidx);
592 		ipsec4_get_ulp(m, &spidx, needport);
593 		spidx.dir = dir;
594 		sp = key_allocsp(&spidx, dir);
595 	}
596 	if (sp == NULL)		/* No SP found, use system default. */
597 		sp = key_allocsp_default();
598 	return (sp);
599 }
600 
601 /*
602  * Check security policy for *OUTBOUND* IPv4 packet.
603  */
604 struct secpolicy *
605 ipsec4_checkpolicy(const struct mbuf *m, struct inpcb *inp, int *error,
606     int needport)
607 {
608 	struct secpolicy *sp;
609 
610 	*error = 0;
611 	sp = ipsec4_getpolicy(m, inp, IPSEC_DIR_OUTBOUND, needport);
612 	if (sp != NULL)
613 		sp = ipsec_checkpolicy(sp, inp, error);
614 	if (sp == NULL) {
615 		switch (*error) {
616 		case 0: /* No IPsec required: BYPASS or NONE */
617 			break;
618 		case -EINVAL:
619 			IPSECSTAT_INC(ips_out_polvio);
620 			break;
621 		default:
622 			IPSECSTAT_INC(ips_out_inval);
623 		}
624 	}
625 	KEYDBG(IPSEC_STAMP,
626 	    printf("%s: using SP(%p), error %d\n", __func__, sp, *error));
627 	if (sp != NULL)
628 		KEYDBG(IPSEC_DATA, kdebug_secpolicy(sp));
629 	return (sp);
630 }
631 
632 /*
633  * Check IPv4 packet against *INBOUND* security policy.
634  * This function is called from tcp_input(), udp_input(),
635  * rip_input() and sctp_input().
636  */
637 int
638 ipsec4_in_reject(const struct mbuf *m, struct inpcb *inp)
639 {
640 	struct secpolicy *sp;
641 	int result;
642 
643 	sp = ipsec4_getpolicy(m, inp, IPSEC_DIR_INBOUND, 0);
644 	result = ipsec_in_reject(sp, inp, m);
645 	key_freesp(&sp);
646 	if (result != 0)
647 		IPSECSTAT_INC(ips_in_polvio);
648 	return (result);
649 }
650 
651 /*
652  * IPSEC_CAP() method implementation for IPv4.
653  */
654 int
655 ipsec4_capability(struct mbuf *m, u_int cap)
656 {
657 
658 	switch (cap) {
659 	case IPSEC_CAP_BYPASS_FILTER:
660 		/*
661 		 * Bypass packet filtering for packets previously handled
662 		 * by IPsec.
663 		 */
664 		if (!V_ip4_filtertunnel &&
665 		    m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL) != NULL)
666 			return (1);
667 		return (0);
668 	case IPSEC_CAP_OPERABLE:
669 		/* Do we have active security policies? */
670 		if (key_havesp(IPSEC_DIR_INBOUND) != 0 ||
671 		    key_havesp(IPSEC_DIR_OUTBOUND) != 0)
672 			return (1);
673 		return (0);
674 	};
675 	return (EOPNOTSUPP);
676 }
677 
678 #endif /* INET */
679 
680 #ifdef INET6
681 static void
682 ipsec6_get_ulp(const struct mbuf *m, struct secpolicyindex *spidx,
683     int needport)
684 {
685 	struct tcphdr th;
686 	struct udphdr uh;
687 	struct icmp6_hdr ih;
688 	int off, nxt;
689 
690 	IPSEC_ASSERT(m->m_pkthdr.len >= sizeof(struct ip6_hdr),
691 	    ("packet too short"));
692 
693 	/* Set default. */
694 	spidx->ul_proto = IPSEC_ULPROTO_ANY;
695 	spidx->src.sin6.sin6_port = IPSEC_PORT_ANY;
696 	spidx->dst.sin6.sin6_port = IPSEC_PORT_ANY;
697 
698 	nxt = -1;
699 	off = ip6_lasthdr(m, 0, IPPROTO_IPV6, &nxt);
700 	if (off < 0 || m->m_pkthdr.len < off)
701 		return;
702 
703 	switch (nxt) {
704 	case IPPROTO_TCP:
705 		spidx->ul_proto = nxt;
706 		if (!needport)
707 			break;
708 		if (off + sizeof(struct tcphdr) > m->m_pkthdr.len)
709 			break;
710 		m_copydata(m, off, sizeof(th), (caddr_t)&th);
711 		spidx->src.sin6.sin6_port = th.th_sport;
712 		spidx->dst.sin6.sin6_port = th.th_dport;
713 		break;
714 	case IPPROTO_UDP:
715 		spidx->ul_proto = nxt;
716 		if (!needport)
717 			break;
718 		if (off + sizeof(struct udphdr) > m->m_pkthdr.len)
719 			break;
720 		m_copydata(m, off, sizeof(uh), (caddr_t)&uh);
721 		spidx->src.sin6.sin6_port = uh.uh_sport;
722 		spidx->dst.sin6.sin6_port = uh.uh_dport;
723 		break;
724 	case IPPROTO_ICMPV6:
725 		spidx->ul_proto = nxt;
726 		if (off + sizeof(struct icmp6_hdr) > m->m_pkthdr.len)
727 			break;
728 		m_copydata(m, off, sizeof(ih), (caddr_t)&ih);
729 		spidx->src.sin6.sin6_port = htons((uint16_t)ih.icmp6_type);
730 		spidx->dst.sin6.sin6_port = htons((uint16_t)ih.icmp6_code);
731 		break;
732 	default:
733 		/* XXX Intermediate headers??? */
734 		spidx->ul_proto = nxt;
735 		break;
736 	}
737 	KEYDBG(IPSEC_DUMP,
738 	    printf("%s: ", __func__); kdebug_secpolicyindex(spidx, NULL));
739 }
740 
741 static void
742 ipsec6_setspidx_ipaddr(const struct mbuf *m, struct secpolicyindex *spidx)
743 {
744 
745 	ipsec6_setsockaddrs(m, &spidx->src, &spidx->dst);
746 	spidx->prefs = sizeof(struct in6_addr) << 3;
747 	spidx->prefd = sizeof(struct in6_addr) << 3;
748 }
749 
750 static struct secpolicy *
751 ipsec6_getpolicy(const struct mbuf *m, struct inpcb *inp, u_int dir,
752     int needport)
753 {
754 	struct secpolicyindex spidx;
755 	struct secpolicy *sp;
756 
757 	sp = ipsec_getpcbpolicy(inp, dir);
758 	if (sp == NULL && key_havesp(dir)) {
759 		/* Make an index to look for a policy. */
760 		ipsec6_setspidx_ipaddr(m, &spidx);
761 		ipsec6_get_ulp(m, &spidx, needport);
762 		spidx.dir = dir;
763 		sp = key_allocsp(&spidx, dir);
764 	}
765 	if (sp == NULL)		/* No SP found, use system default. */
766 		sp = key_allocsp_default();
767 	return (sp);
768 }
769 
770 /*
771  * Check security policy for *OUTBOUND* IPv6 packet.
772  */
773 struct secpolicy *
774 ipsec6_checkpolicy(const struct mbuf *m, struct inpcb *inp, int *error,
775     int needport)
776 {
777 	struct secpolicy *sp;
778 
779 	*error = 0;
780 	sp = ipsec6_getpolicy(m, inp, IPSEC_DIR_OUTBOUND, needport);
781 	if (sp != NULL)
782 		sp = ipsec_checkpolicy(sp, inp, error);
783 	if (sp == NULL) {
784 		switch (*error) {
785 		case 0: /* No IPsec required: BYPASS or NONE */
786 			break;
787 		case -EINVAL:
788 			IPSEC6STAT_INC(ips_out_polvio);
789 			break;
790 		default:
791 			IPSEC6STAT_INC(ips_out_inval);
792 		}
793 	}
794 	KEYDBG(IPSEC_STAMP,
795 	    printf("%s: using SP(%p), error %d\n", __func__, sp, *error));
796 	if (sp != NULL)
797 		KEYDBG(IPSEC_DATA, kdebug_secpolicy(sp));
798 	return (sp);
799 }
800 
801 /*
802  * Check IPv6 packet against inbound security policy.
803  * This function is called from tcp6_input(), udp6_input(),
804  * rip6_input() and sctp_input().
805  */
806 int
807 ipsec6_in_reject(const struct mbuf *m, struct inpcb *inp)
808 {
809 	struct secpolicy *sp;
810 	int result;
811 
812 	sp = ipsec6_getpolicy(m, inp, IPSEC_DIR_INBOUND, 0);
813 	result = ipsec_in_reject(sp, inp, m);
814 	key_freesp(&sp);
815 	if (result)
816 		IPSEC6STAT_INC(ips_in_polvio);
817 	return (result);
818 }
819 
820 /*
821  * IPSEC_CAP() method implementation for IPv6.
822  */
823 int
824 ipsec6_capability(struct mbuf *m, u_int cap)
825 {
826 
827 	switch (cap) {
828 	case IPSEC_CAP_BYPASS_FILTER:
829 		/*
830 		 * Bypass packet filtering for packets previously handled
831 		 * by IPsec.
832 		 */
833 		if (!V_ip6_filtertunnel &&
834 		    m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL) != NULL)
835 			return (1);
836 		return (0);
837 	case IPSEC_CAP_OPERABLE:
838 		/* Do we have active security policies? */
839 		if (key_havesp(IPSEC_DIR_INBOUND) != 0 ||
840 		    key_havesp(IPSEC_DIR_OUTBOUND) != 0)
841 			return (1);
842 		return (0);
843 	};
844 	return (EOPNOTSUPP);
845 }
846 #endif /* INET6 */
847 
848 int
849 ipsec_run_hhooks(struct ipsec_ctx_data *ctx, int type)
850 {
851 	int idx;
852 
853 	switch (ctx->af) {
854 #ifdef INET
855 	case AF_INET:
856 		idx = HHOOK_IPSEC_INET;
857 		break;
858 #endif
859 #ifdef INET6
860 	case AF_INET6:
861 		idx = HHOOK_IPSEC_INET6;
862 		break;
863 #endif
864 	default:
865 		return (EPFNOSUPPORT);
866 	}
867 	if (type == HHOOK_TYPE_IPSEC_IN)
868 		HHOOKS_RUN_IF(V_ipsec_hhh_in[idx], ctx, NULL);
869 	else
870 		HHOOKS_RUN_IF(V_ipsec_hhh_out[idx], ctx, NULL);
871 	if (*ctx->mp == NULL)
872 		return (EACCES);
873 	return (0);
874 }
875 
876 /*
877  * Return current level.
878  * Either IPSEC_LEVEL_USE or IPSEC_LEVEL_REQUIRE are always returned.
879  */
880 u_int
881 ipsec_get_reqlevel(struct secpolicy *sp, u_int idx)
882 {
883 	struct ipsecrequest *isr;
884 	u_int esp_trans_deflev, esp_net_deflev;
885 	u_int ah_trans_deflev, ah_net_deflev;
886 	u_int level = 0;
887 
888 	IPSEC_ASSERT(idx < sp->tcount, ("Wrong IPsec request index %d", idx));
889 /* XXX Note that we have ipseclog() expanded here - code sync issue. */
890 #define IPSEC_CHECK_DEFAULT(lev) \
891 	(((lev) != IPSEC_LEVEL_USE && (lev) != IPSEC_LEVEL_REQUIRE &&	\
892 	  (lev) != IPSEC_LEVEL_UNIQUE)					\
893 		? (V_ipsec_debug  ?					\
894 		log(LOG_INFO, "fixed system default level " #lev ":%d->%d\n",\
895 		(lev), IPSEC_LEVEL_REQUIRE) : 0),			\
896 		(lev) = IPSEC_LEVEL_REQUIRE, (lev) : (lev))
897 
898 	/*
899 	 * IPsec VTI uses unique security policy with fake spidx filled
900 	 * with zeroes. Just return IPSEC_LEVEL_REQUIRE instead of doing
901 	 * full level lookup for such policies.
902 	 */
903 	if (sp->state == IPSEC_SPSTATE_IFNET) {
904 		IPSEC_ASSERT(sp->req[idx]->level == IPSEC_LEVEL_UNIQUE,
905 		    ("Wrong IPsec request level %d", sp->req[idx]->level));
906 		return (IPSEC_LEVEL_REQUIRE);
907 	}
908 
909 	/* Set default level. */
910 	switch (sp->spidx.src.sa.sa_family) {
911 #ifdef INET
912 	case AF_INET:
913 		esp_trans_deflev = IPSEC_CHECK_DEFAULT(V_ip4_esp_trans_deflev);
914 		esp_net_deflev = IPSEC_CHECK_DEFAULT(V_ip4_esp_net_deflev);
915 		ah_trans_deflev = IPSEC_CHECK_DEFAULT(V_ip4_ah_trans_deflev);
916 		ah_net_deflev = IPSEC_CHECK_DEFAULT(V_ip4_ah_net_deflev);
917 		break;
918 #endif
919 #ifdef INET6
920 	case AF_INET6:
921 		esp_trans_deflev = IPSEC_CHECK_DEFAULT(V_ip6_esp_trans_deflev);
922 		esp_net_deflev = IPSEC_CHECK_DEFAULT(V_ip6_esp_net_deflev);
923 		ah_trans_deflev = IPSEC_CHECK_DEFAULT(V_ip6_ah_trans_deflev);
924 		ah_net_deflev = IPSEC_CHECK_DEFAULT(V_ip6_ah_net_deflev);
925 		break;
926 #endif /* INET6 */
927 	default:
928 		panic("%s: unknown af %u",
929 			__func__, sp->spidx.src.sa.sa_family);
930 	}
931 
932 #undef IPSEC_CHECK_DEFAULT
933 
934 	isr = sp->req[idx];
935 	/* Set level. */
936 	switch (isr->level) {
937 	case IPSEC_LEVEL_DEFAULT:
938 		switch (isr->saidx.proto) {
939 		case IPPROTO_ESP:
940 			if (isr->saidx.mode == IPSEC_MODE_TUNNEL)
941 				level = esp_net_deflev;
942 			else
943 				level = esp_trans_deflev;
944 			break;
945 		case IPPROTO_AH:
946 			if (isr->saidx.mode == IPSEC_MODE_TUNNEL)
947 				level = ah_net_deflev;
948 			else
949 				level = ah_trans_deflev;
950 			break;
951 		case IPPROTO_IPCOMP:
952 			/*
953 			 * We don't really care, as IPcomp document says that
954 			 * we shouldn't compress small packets.
955 			 */
956 			level = IPSEC_LEVEL_USE;
957 			break;
958 		default:
959 			panic("%s: Illegal protocol defined %u\n", __func__,
960 				isr->saidx.proto);
961 		}
962 		break;
963 
964 	case IPSEC_LEVEL_USE:
965 	case IPSEC_LEVEL_REQUIRE:
966 		level = isr->level;
967 		break;
968 	case IPSEC_LEVEL_UNIQUE:
969 		level = IPSEC_LEVEL_REQUIRE;
970 		break;
971 
972 	default:
973 		panic("%s: Illegal IPsec level %u\n", __func__, isr->level);
974 	}
975 
976 	return (level);
977 }
978 
979 static int
980 ipsec_check_history(const struct mbuf *m, struct secpolicy *sp, u_int idx)
981 {
982 	struct xform_history *xh;
983 	struct m_tag *mtag;
984 
985 	mtag = NULL;
986 	while ((mtag = m_tag_find(__DECONST(struct mbuf *, m),
987 	    PACKET_TAG_IPSEC_IN_DONE, mtag)) != NULL) {
988 		xh = (struct xform_history *)(mtag + 1);
989 		KEYDBG(IPSEC_DATA,
990 		    char buf[IPSEC_ADDRSTRLEN];
991 		    printf("%s: mode %s proto %u dst %s\n", __func__,
992 			kdebug_secasindex_mode(xh->mode), xh->proto,
993 			ipsec_address(&xh->dst, buf, sizeof(buf))));
994 		if (xh->proto != sp->req[idx]->saidx.proto)
995 			continue;
996 		/* If SA had IPSEC_MODE_ANY, consider this as match. */
997 		if (xh->mode != sp->req[idx]->saidx.mode &&
998 		    xh->mode != IPSEC_MODE_ANY)
999 			continue;
1000 		/*
1001 		 * For transport mode IPsec request doesn't contain
1002 		 * addresses. We need to use address from spidx.
1003 		 */
1004 		if (sp->req[idx]->saidx.mode == IPSEC_MODE_TRANSPORT) {
1005 			if (key_sockaddrcmp_withmask(&xh->dst.sa,
1006 			    &sp->spidx.dst.sa, sp->spidx.prefd) != 0)
1007 				continue;
1008 		} else {
1009 			if (key_sockaddrcmp(&xh->dst.sa,
1010 			    &sp->req[idx]->saidx.dst.sa, 0) != 0)
1011 				continue;
1012 		}
1013 		return (0); /* matched */
1014 	}
1015 	return (1);
1016 }
1017 
1018 /*
1019  * Check security policy requirements against the actual
1020  * packet contents.  Return one if the packet should be
1021  * reject as "invalid"; otherwiser return zero to have the
1022  * packet treated as "valid".
1023  *
1024  * OUT:
1025  *	0: valid
1026  *	1: invalid
1027  */
1028 static int
1029 ipsec_in_reject(struct secpolicy *sp, struct inpcb *inp, const struct mbuf *m)
1030 {
1031 	int i;
1032 
1033 	KEYDBG(IPSEC_STAMP,
1034 	    printf("%s: PCB(%p): using SP(%p)\n", __func__, inp, sp));
1035 	KEYDBG(IPSEC_DATA, kdebug_secpolicy(sp));
1036 
1037 	if (inp != NULL && inp->inp_sp != NULL && inp->inp_sp->sp_in == NULL)
1038 		ipsec_cachepolicy(inp, sp, IPSEC_DIR_INBOUND);
1039 
1040 	/* Check policy. */
1041 	switch (sp->policy) {
1042 	case IPSEC_POLICY_DISCARD:
1043 		return (1);
1044 	case IPSEC_POLICY_BYPASS:
1045 	case IPSEC_POLICY_NONE:
1046 		return (0);
1047 	}
1048 
1049 	IPSEC_ASSERT(sp->policy == IPSEC_POLICY_IPSEC,
1050 		("invalid policy %u", sp->policy));
1051 
1052 	/*
1053 	 * ipsec[46]_common_input_cb after each transform adds
1054 	 * PACKET_TAG_IPSEC_IN_DONE mbuf tag. It contains SPI, proto, mode
1055 	 * and destination address from saidx. We can compare info from
1056 	 * these tags with requirements in SP.
1057 	 */
1058 	for (i = 0; i < sp->tcount; i++) {
1059 		/*
1060 		 * Do not check IPcomp, since IPcomp document
1061 		 * says that we shouldn't compress small packets.
1062 		 * IPComp policy should always be treated as being
1063 		 * in "use" level.
1064 		 */
1065 		if (sp->req[i]->saidx.proto == IPPROTO_IPCOMP ||
1066 		    ipsec_get_reqlevel(sp, i) != IPSEC_LEVEL_REQUIRE)
1067 			continue;
1068 		if (V_check_policy_history != 0 &&
1069 		    ipsec_check_history(m, sp, i) != 0)
1070 			return (1);
1071 		else switch (sp->req[i]->saidx.proto) {
1072 		case IPPROTO_ESP:
1073 			if ((m->m_flags & M_DECRYPTED) == 0) {
1074 				KEYDBG(IPSEC_DUMP,
1075 				    printf("%s: ESP m_flags:%x\n", __func__,
1076 					    m->m_flags));
1077 				return (1);
1078 			}
1079 			break;
1080 		case IPPROTO_AH:
1081 			if ((m->m_flags & M_AUTHIPHDR) == 0) {
1082 				KEYDBG(IPSEC_DUMP,
1083 				    printf("%s: AH m_flags:%x\n", __func__,
1084 					    m->m_flags));
1085 				return (1);
1086 			}
1087 			break;
1088 		}
1089 	}
1090 	return (0);		/* Valid. */
1091 }
1092 
1093 /*
1094  * Compute the byte size to be occupied by IPsec header.
1095  * In case it is tunnelled, it includes the size of outer IP header.
1096  */
1097 static size_t
1098 ipsec_hdrsiz_internal(struct secpolicy *sp)
1099 {
1100 	size_t size;
1101 	int i;
1102 
1103 	KEYDBG(IPSEC_STAMP, printf("%s: using SP(%p)\n", __func__, sp));
1104 	KEYDBG(IPSEC_DATA, kdebug_secpolicy(sp));
1105 
1106 	switch (sp->policy) {
1107 	case IPSEC_POLICY_DISCARD:
1108 	case IPSEC_POLICY_BYPASS:
1109 	case IPSEC_POLICY_NONE:
1110 		return (0);
1111 	}
1112 
1113 	IPSEC_ASSERT(sp->policy == IPSEC_POLICY_IPSEC,
1114 		("invalid policy %u", sp->policy));
1115 
1116 	/*
1117 	 * XXX: for each transform we need to lookup suitable SA
1118 	 * and use info from SA to calculate headers size.
1119 	 * XXX: for NAT-T we need to cosider UDP header size.
1120 	 */
1121 	size = 0;
1122 	for (i = 0; i < sp->tcount; i++) {
1123 		switch (sp->req[i]->saidx.proto) {
1124 		case IPPROTO_ESP:
1125 			size += esp_hdrsiz(NULL);
1126 			break;
1127 		case IPPROTO_AH:
1128 			size += ah_hdrsiz(NULL);
1129 			break;
1130 		case IPPROTO_IPCOMP:
1131 			size += sizeof(struct ipcomp);
1132 			break;
1133 		}
1134 
1135 		if (sp->req[i]->saidx.mode == IPSEC_MODE_TUNNEL) {
1136 			switch (sp->req[i]->saidx.dst.sa.sa_family) {
1137 #ifdef INET
1138 			case AF_INET:
1139 				size += sizeof(struct ip);
1140 				break;
1141 #endif
1142 #ifdef INET6
1143 			case AF_INET6:
1144 				size += sizeof(struct ip6_hdr);
1145 				break;
1146 #endif
1147 			default:
1148 				ipseclog((LOG_ERR, "%s: unknown AF %d in "
1149 				    "IPsec tunnel SA\n", __func__,
1150 				    sp->req[i]->saidx.dst.sa.sa_family));
1151 				break;
1152 			}
1153 		}
1154 	}
1155 	return (size);
1156 }
1157 
1158 /*
1159  * Compute ESP/AH header size for protocols with PCB, including
1160  * outer IP header. Currently only tcp_output() uses it.
1161  */
1162 size_t
1163 ipsec_hdrsiz_inpcb(struct inpcb *inp)
1164 {
1165 	struct secpolicyindex spidx;
1166 	struct secpolicy *sp;
1167 	size_t sz;
1168 
1169 	sp = ipsec_getpcbpolicy(inp, IPSEC_DIR_OUTBOUND);
1170 	if (sp == NULL && key_havesp(IPSEC_DIR_OUTBOUND)) {
1171 		ipsec_setspidx_inpcb(inp, &spidx, IPSEC_DIR_OUTBOUND);
1172 		sp = key_allocsp(&spidx, IPSEC_DIR_OUTBOUND);
1173 	}
1174 	if (sp == NULL)
1175 		sp = key_allocsp_default();
1176 	sz = ipsec_hdrsiz_internal(sp);
1177 	key_freesp(&sp);
1178 	return (sz);
1179 }
1180 
1181 /*
1182  * Check the variable replay window.
1183  * ipsec_chkreplay() performs replay check before ICV verification.
1184  * ipsec_updatereplay() updates replay bitmap.  This must be called after
1185  * ICV verification (it also performs replay check, which is usually done
1186  * beforehand).
1187  * 0 (zero) is returned if packet disallowed, 1 if packet permitted.
1188  *
1189  * Based on RFC 6479. Blocks are 32 bits unsigned integers
1190  */
1191 
1192 #define IPSEC_BITMAP_INDEX_MASK(w)	(w - 1)
1193 #define IPSEC_REDUNDANT_BIT_SHIFTS	5
1194 #define IPSEC_REDUNDANT_BITS		(1 << IPSEC_REDUNDANT_BIT_SHIFTS)
1195 #define IPSEC_BITMAP_LOC_MASK		(IPSEC_REDUNDANT_BITS - 1)
1196 
1197 int
1198 ipsec_chkreplay(uint32_t seq, struct secasvar *sav)
1199 {
1200 	const struct secreplay *replay;
1201 	uint32_t wsizeb;		/* Constant: window size. */
1202 	int index, bit_location;
1203 
1204 	IPSEC_ASSERT(sav != NULL, ("Null SA"));
1205 	IPSEC_ASSERT(sav->replay != NULL, ("Null replay state"));
1206 
1207 	replay = sav->replay;
1208 
1209 	/* No need to check replay if disabled. */
1210 	if (replay->wsize == 0)
1211 		return (1);
1212 
1213 	/* Constant. */
1214 	wsizeb = replay->wsize << 3;
1215 
1216 	/* Sequence number of 0 is invalid. */
1217 	if (seq == 0)
1218 		return (0);
1219 
1220 	/* First time is always okay. */
1221 	if (replay->count == 0)
1222 		return (1);
1223 
1224 	/* Larger sequences are okay. */
1225 	if (seq > replay->lastseq)
1226 		return (1);
1227 
1228 	/* Over range to check, i.e. too old or wrapped. */
1229 	if (replay->lastseq - seq >= wsizeb)
1230 		return (0);
1231 
1232 	/* The sequence is inside the sliding window
1233 	 * now check the bit in the bitmap
1234 	 * bit location only depends on the sequence number
1235 	 */
1236 	bit_location = seq & IPSEC_BITMAP_LOC_MASK;
1237 	index = (seq >> IPSEC_REDUNDANT_BIT_SHIFTS)
1238 		& IPSEC_BITMAP_INDEX_MASK(replay->bitmap_size);
1239 
1240 	/* This packet already seen? */
1241 	if ((replay->bitmap)[index] & (1 << bit_location))
1242 		return (0);
1243 	return (1);
1244 }
1245 
1246 /*
1247  * Check replay counter whether to update or not.
1248  * OUT:	0:	OK
1249  *	1:	NG
1250  */
1251 int
1252 ipsec_updatereplay(uint32_t seq, struct secasvar *sav)
1253 {
1254 	char buf[128];
1255 	struct secreplay *replay;
1256 	uint32_t wsizeb;		/* Constant: window size. */
1257 	int diff, index, bit_location;
1258 
1259 	IPSEC_ASSERT(sav != NULL, ("Null SA"));
1260 	IPSEC_ASSERT(sav->replay != NULL, ("Null replay state"));
1261 
1262 	replay = sav->replay;
1263 
1264 	if (replay->wsize == 0)
1265 		goto ok;	/* No need to check replay. */
1266 
1267 	/* Constant. */
1268 	wsizeb = replay->wsize << 3;
1269 
1270 	/* Sequence number of 0 is invalid. */
1271 	if (seq == 0)
1272 		return (1);
1273 
1274 	/* The packet is too old, no need to update */
1275 	if (wsizeb + seq < replay->lastseq)
1276 		goto ok;
1277 
1278 	/* Now update the bit */
1279 	index = (seq >> IPSEC_REDUNDANT_BIT_SHIFTS);
1280 
1281 	/* First check if the sequence number is in the range */
1282 	if (seq > replay->lastseq) {
1283 		int id;
1284 		int index_cur = replay->lastseq >> IPSEC_REDUNDANT_BIT_SHIFTS;
1285 
1286 		diff = index - index_cur;
1287 		if (diff > replay->bitmap_size) {
1288 			/* something unusual in this case */
1289 			diff = replay->bitmap_size;
1290 		}
1291 
1292 		for (id = 0; id < diff; ++id) {
1293 			replay->bitmap[(id + index_cur + 1)
1294 			& IPSEC_BITMAP_INDEX_MASK(replay->bitmap_size)] = 0;
1295 		}
1296 
1297 		replay->lastseq = seq;
1298 	}
1299 
1300 	index &= IPSEC_BITMAP_INDEX_MASK(replay->bitmap_size);
1301 	bit_location = seq & IPSEC_BITMAP_LOC_MASK;
1302 
1303 	/* this packet has already been received */
1304 	if (replay->bitmap[index] & (1 << bit_location))
1305 		return (1);
1306 
1307 	replay->bitmap[index] |= (1 << bit_location);
1308 
1309 ok:
1310 	if (replay->count == ~0) {
1311 
1312 		/* Set overflow flag. */
1313 		replay->overflow++;
1314 
1315 		/* Don't increment, no more packets accepted. */
1316 		if ((sav->flags & SADB_X_EXT_CYCSEQ) == 0) {
1317 			if (sav->sah->saidx.proto == IPPROTO_AH)
1318 				AHSTAT_INC(ahs_wrap);
1319 			else if (sav->sah->saidx.proto == IPPROTO_ESP)
1320 				ESPSTAT_INC(esps_wrap);
1321 			return (1);
1322 		}
1323 
1324 		ipseclog((LOG_WARNING, "%s: replay counter made %d cycle. %s\n",
1325 		    __func__, replay->overflow,
1326 		    ipsec_sa2str(sav, buf, sizeof(buf))));
1327 	}
1328 
1329 	replay->count++;
1330 	return (0);
1331 }
1332 
1333 int
1334 ipsec_updateid(struct secasvar *sav, crypto_session_t *new,
1335     crypto_session_t *old)
1336 {
1337 	crypto_session_t tmp;
1338 
1339 	/*
1340 	 * tdb_cryptoid is initialized by xform_init().
1341 	 * Then it can be changed only when some crypto error occurred or
1342 	 * when SA is deleted. We stored used cryptoid in the xform_data
1343 	 * structure. In case when crypto error occurred and crypto
1344 	 * subsystem has reinited the session, it returns new cryptoid
1345 	 * and EAGAIN error code.
1346 	 *
1347 	 * This function will be called when we got EAGAIN from crypto
1348 	 * subsystem.
1349 	 * *new is cryptoid that was returned by crypto subsystem in
1350 	 * the crp_sid.
1351 	 * *old is the original cryptoid that we stored in xform_data.
1352 	 *
1353 	 * For first failed request *old == sav->tdb_cryptoid, then
1354 	 * we update sav->tdb_cryptoid and redo crypto_dispatch().
1355 	 * For next failed request *old != sav->tdb_cryptoid, then
1356 	 * we store cryptoid from first request into the *new variable
1357 	 * and crp_sid from this second session will be returned via
1358 	 * *old pointer, so caller can release second session.
1359 	 *
1360 	 * XXXAE: check this more carefully.
1361 	 */
1362 	KEYDBG(IPSEC_STAMP,
1363 	    printf("%s: SA(%p) moves cryptoid %p -> %p\n",
1364 		__func__, sav, *old, *new));
1365 	KEYDBG(IPSEC_DATA, kdebug_secasv(sav));
1366 	SECASVAR_LOCK(sav);
1367 	if (sav->tdb_cryptoid != *old) {
1368 		/* cryptoid was already updated */
1369 		tmp = *new;
1370 		*new = sav->tdb_cryptoid;
1371 		*old = tmp;
1372 		SECASVAR_UNLOCK(sav);
1373 		return (1);
1374 	}
1375 	sav->tdb_cryptoid = *new;
1376 	SECASVAR_UNLOCK(sav);
1377 	return (0);
1378 }
1379 
1380 int
1381 ipsec_initialized(void)
1382 {
1383 
1384 	return (V_def_policy != NULL);
1385 }
1386 
1387 static void
1388 def_policy_init(const void *unused __unused)
1389 {
1390 
1391 	V_def_policy = key_newsp();
1392 	if (V_def_policy != NULL) {
1393 		V_def_policy->policy = IPSEC_POLICY_NONE;
1394 		/* Force INPCB SP cache invalidation */
1395 		key_bumpspgen();
1396 	} else
1397 		printf("%s: failed to initialize default policy\n", __func__);
1398 }
1399 
1400 
1401 static void
1402 def_policy_uninit(const void *unused __unused)
1403 {
1404 
1405 	if (V_def_policy != NULL) {
1406 		key_freesp(&V_def_policy);
1407 		key_bumpspgen();
1408 	}
1409 }
1410 
1411 VNET_SYSINIT(def_policy_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST,
1412     def_policy_init, NULL);
1413 VNET_SYSUNINIT(def_policy_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST,
1414     def_policy_uninit, NULL);
1415