xref: /freebsd/sys/netipsec/ipsec.c (revision ca987d4641cdcd7f27e153db17c5bf064934faf5)
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 /*
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_AH_OFFSETMASK, ah_offsetmask,
196 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip4_ah_offsetmask), 0,
197 	"If not set, clear offset field mask when doing AH computation.");
198 SYSCTL_INT(_net_inet_ipsec, IPSECCTL_DFBIT, dfbit,
199 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip4_ipsec_dfbit), 0,
200 	"Do not fragment bit on encap.");
201 SYSCTL_INT(_net_inet_ipsec, IPSECCTL_ECN, ecn,
202 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip4_ipsec_ecn), 0,
203 	"Explicit Congestion Notification handling.");
204 SYSCTL_INT(_net_inet_ipsec, OID_AUTO, crypto_support,
205 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(crypto_support), 0,
206 	"Crypto driver selection.");
207 SYSCTL_INT(_net_inet_ipsec, OID_AUTO, async_crypto,
208 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(async_crypto), 0,
209 	"Use asynchronous mode to parallelize crypto jobs.");
210 SYSCTL_INT(_net_inet_ipsec, OID_AUTO, check_policy_history,
211 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(check_policy_history), 0,
212 	"Use strict check of inbound packets to security policy compliance.");
213 SYSCTL_INT(_net_inet_ipsec, OID_AUTO, natt_cksum_policy,
214 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(natt_cksum_policy), 0,
215 	"Method to fix TCP/UDP checksum for transport mode IPsec after NAT.");
216 SYSCTL_INT(_net_inet_ipsec, OID_AUTO, filtertunnel,
217 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip4_filtertunnel), 0,
218 	"If set, filter packets from an IPsec tunnel.");
219 SYSCTL_VNET_PCPUSTAT(_net_inet_ipsec, OID_AUTO, ipsecstats, struct ipsecstat,
220     ipsec4stat, "IPsec IPv4 statistics.");
221 
222 #ifdef REGRESSION
223 /*
224  * When set to 1, IPsec will send packets with the same sequence number.
225  * This allows to verify if the other side has proper replay attacks detection.
226  */
227 VNET_DEFINE(int, ipsec_replay) = 0;
228 SYSCTL_INT(_net_inet_ipsec, OID_AUTO, test_replay,
229 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ipsec_replay), 0,
230 	"Emulate replay attack");
231 /*
232  * When set 1, IPsec will send packets with corrupted HMAC.
233  * This allows to verify if the other side properly detects modified packets.
234  */
235 VNET_DEFINE(int, ipsec_integrity) = 0;
236 SYSCTL_INT(_net_inet_ipsec, OID_AUTO, test_integrity,
237 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ipsec_integrity), 0,
238 	"Emulate man-in-the-middle attack");
239 #endif
240 
241 #ifdef INET6
242 VNET_PCPUSTAT_DEFINE(struct ipsecstat, ipsec6stat);
243 VNET_PCPUSTAT_SYSINIT(ipsec6stat);
244 
245 #ifdef VIMAGE
246 VNET_PCPUSTAT_SYSUNINIT(ipsec6stat);
247 #endif /* VIMAGE */
248 
249 VNET_DEFINE(int, ip6_esp_trans_deflev) = IPSEC_LEVEL_USE;
250 VNET_DEFINE(int, ip6_esp_net_deflev) = IPSEC_LEVEL_USE;
251 VNET_DEFINE(int, ip6_ah_trans_deflev) = IPSEC_LEVEL_USE;
252 VNET_DEFINE(int, ip6_ah_net_deflev) = IPSEC_LEVEL_USE;
253 VNET_DEFINE(int, ip6_ipsec_ecn) = 0;	/* ECN ignore(-1)/forbidden(0)/allowed(1) */
254 
255 static VNET_DEFINE(int, ip6_filtertunnel) = 0;
256 #define	V_ip6_filtertunnel	VNET(ip6_filtertunnel)
257 
258 SYSCTL_DECL(_net_inet6_ipsec6);
259 
260 /* net.inet6.ipsec6 */
261 SYSCTL_PROC(_net_inet6_ipsec6, IPSECCTL_DEF_POLICY, def_policy,
262 	CTLTYPE_INT | CTLFLAG_VNET | CTLFLAG_RW, 0, 0, sysctl_def_policy, "I",
263 	"IPsec default policy.");
264 SYSCTL_INT(_net_inet6_ipsec6, IPSECCTL_DEF_ESP_TRANSLEV, esp_trans_deflev,
265 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_esp_trans_deflev), 0,
266 	"Default ESP transport mode level.");
267 SYSCTL_INT(_net_inet6_ipsec6, IPSECCTL_DEF_ESP_NETLEV, esp_net_deflev,
268 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_esp_net_deflev), 0,
269 	"Default ESP tunnel mode level.");
270 SYSCTL_INT(_net_inet6_ipsec6, IPSECCTL_DEF_AH_TRANSLEV, ah_trans_deflev,
271 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_ah_trans_deflev), 0,
272 	"AH transfer mode default level.");
273 SYSCTL_INT(_net_inet6_ipsec6, IPSECCTL_DEF_AH_NETLEV, ah_net_deflev,
274 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_ah_net_deflev), 0,
275 	"AH tunnel mode default level.");
276 SYSCTL_INT(_net_inet6_ipsec6, IPSECCTL_ECN, ecn,
277 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_ipsec_ecn), 0,
278 	"Explicit Congestion Notification handling.");
279 SYSCTL_INT(_net_inet6_ipsec6, OID_AUTO, filtertunnel,
280 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_filtertunnel),  0,
281 	"If set, filter packets from an IPsec tunnel.");
282 SYSCTL_VNET_PCPUSTAT(_net_inet6_ipsec6, IPSECCTL_STATS, ipsecstats,
283     struct ipsecstat, ipsec6stat, "IPsec IPv6 statistics.");
284 #endif /* INET6 */
285 
286 static int ipsec_in_reject(struct secpolicy *, struct inpcb *,
287     const struct mbuf *);
288 
289 #ifdef INET
290 static void ipsec4_get_ulp(const struct mbuf *, struct secpolicyindex *, int);
291 static void ipsec4_setspidx_ipaddr(const struct mbuf *,
292     struct secpolicyindex *);
293 #endif
294 #ifdef INET6
295 static void ipsec6_get_ulp(const struct mbuf *m, struct secpolicyindex *, int);
296 static void ipsec6_setspidx_ipaddr(const struct mbuf *,
297     struct secpolicyindex *);
298 #endif
299 
300 /*
301  * Return a held reference to the default SP.
302  */
303 static struct secpolicy *
304 key_allocsp_default(void)
305 {
306 
307 	key_addref(V_def_policy);
308 	return (V_def_policy);
309 }
310 
311 static void
312 ipsec_invalidate_cache(struct inpcb *inp, u_int dir)
313 {
314 	struct secpolicy *sp;
315 
316 	INP_WLOCK_ASSERT(inp);
317 	if (dir == IPSEC_DIR_OUTBOUND) {
318 		if (inp->inp_sp->flags & INP_INBOUND_POLICY)
319 			return;
320 		sp = inp->inp_sp->sp_in;
321 		inp->inp_sp->sp_in = NULL;
322 	} else {
323 		if (inp->inp_sp->flags & INP_OUTBOUND_POLICY)
324 			return;
325 		sp = inp->inp_sp->sp_out;
326 		inp->inp_sp->sp_out = NULL;
327 	}
328 	if (sp != NULL)
329 		key_freesp(&sp); /* release extra reference */
330 }
331 
332 static void
333 ipsec_cachepolicy(struct inpcb *inp, struct secpolicy *sp, u_int dir)
334 {
335 	uint32_t genid;
336 	int downgrade;
337 
338 	INP_LOCK_ASSERT(inp);
339 
340 	if (dir == IPSEC_DIR_OUTBOUND) {
341 		/* Do we have configured PCB policy? */
342 		if (inp->inp_sp->flags & INP_OUTBOUND_POLICY)
343 			return;
344 		/* Another thread has already set cached policy */
345 		if (inp->inp_sp->sp_out != NULL)
346 			return;
347 		/*
348 		 * Do not cache OUTBOUND policy if PCB isn't connected,
349 		 * i.e. foreign address is INADDR_ANY/UNSPECIFIED.
350 		 */
351 #ifdef INET
352 		if ((inp->inp_vflag & INP_IPV4) != 0 &&
353 		    inp->inp_faddr.s_addr == INADDR_ANY)
354 			return;
355 #endif
356 #ifdef INET6
357 		if ((inp->inp_vflag & INP_IPV6) != 0 &&
358 		    IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr))
359 			return;
360 #endif
361 	} else {
362 		/* Do we have configured PCB policy? */
363 		if (inp->inp_sp->flags & INP_INBOUND_POLICY)
364 			return;
365 		/* Another thread has already set cached policy */
366 		if (inp->inp_sp->sp_in != NULL)
367 			return;
368 		/*
369 		 * Do not cache INBOUND policy for listen socket,
370 		 * that is bound to INADDR_ANY/UNSPECIFIED address.
371 		 */
372 #ifdef INET
373 		if ((inp->inp_vflag & INP_IPV4) != 0 &&
374 		    inp->inp_faddr.s_addr == INADDR_ANY)
375 			return;
376 #endif
377 #ifdef INET6
378 		if ((inp->inp_vflag & INP_IPV6) != 0 &&
379 		    IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr))
380 			return;
381 #endif
382 	}
383 	downgrade = 0;
384 	if (!INP_WLOCKED(inp)) {
385 		if ((downgrade = INP_TRY_UPGRADE(inp)) == 0)
386 			return;
387 	}
388 	if (dir == IPSEC_DIR_OUTBOUND)
389 		inp->inp_sp->sp_out = sp;
390 	else
391 		inp->inp_sp->sp_in = sp;
392 	/*
393 	 * SP is already referenced by the lookup code.
394 	 * We take extra reference here to avoid race in the
395 	 * ipsec_getpcbpolicy() function - SP will not be freed in the
396 	 * time between we take SP pointer from the cache and key_addref()
397 	 * call.
398 	 */
399 	key_addref(sp);
400 	genid = key_getspgen();
401 	if (genid != inp->inp_sp->genid) {
402 		ipsec_invalidate_cache(inp, dir);
403 		inp->inp_sp->genid = genid;
404 	}
405 	KEYDBG(IPSEC_STAMP,
406 	    printf("%s: PCB(%p): cached %s SP(%p)\n",
407 	    __func__, inp, dir == IPSEC_DIR_OUTBOUND ? "OUTBOUND":
408 	    "INBOUND", sp));
409 	if (downgrade != 0)
410 		INP_DOWNGRADE(inp);
411 }
412 
413 static struct secpolicy *
414 ipsec_checkpolicy(struct secpolicy *sp, struct inpcb *inp, int *error)
415 {
416 
417 	/* Save found OUTBOUND policy into PCB SP cache. */
418 	if (inp != NULL && inp->inp_sp != NULL && inp->inp_sp->sp_out == NULL)
419 		ipsec_cachepolicy(inp, sp, IPSEC_DIR_OUTBOUND);
420 
421 	switch (sp->policy) {
422 	default:
423 		printf("%s: invalid policy %u\n", __func__, sp->policy);
424 		/* FALLTHROUGH */
425 	case IPSEC_POLICY_DISCARD:
426 		*error = -EINVAL;	/* Packet is discarded by caller. */
427 		/* FALLTHROUGH */
428 	case IPSEC_POLICY_BYPASS:
429 	case IPSEC_POLICY_NONE:
430 		key_freesp(&sp);
431 		sp = NULL;		/* NB: force NULL result. */
432 		break;
433 	case IPSEC_POLICY_IPSEC:
434 		/* XXXAE: handle LARVAL SP */
435 		break;
436 	}
437 	KEYDBG(IPSEC_DUMP,
438 	    printf("%s: get SP(%p), error %d\n", __func__, sp, *error));
439 	return (sp);
440 }
441 
442 static struct secpolicy *
443 ipsec_getpcbpolicy(struct inpcb *inp, u_int dir)
444 {
445 	struct secpolicy *sp;
446 	int flags, downgrade;
447 
448 	if (inp == NULL || inp->inp_sp == NULL)
449 		return (NULL);
450 
451 	INP_LOCK_ASSERT(inp);
452 
453 	flags = inp->inp_sp->flags;
454 	if (dir == IPSEC_DIR_OUTBOUND) {
455 		sp = inp->inp_sp->sp_out;
456 		flags &= INP_OUTBOUND_POLICY;
457 	} else {
458 		sp = inp->inp_sp->sp_in;
459 		flags &= INP_INBOUND_POLICY;
460 	}
461 	/*
462 	 * Check flags. If we have PCB SP, just return it.
463 	 * Otherwise we need to check that cached SP entry isn't stale.
464 	 */
465 	if (flags == 0) {
466 		if (sp == NULL)
467 			return (NULL);
468 		if (inp->inp_sp->genid != key_getspgen()) {
469 			/* Invalidate the cache. */
470 			downgrade = 0;
471 			if (!INP_WLOCKED(inp)) {
472 				if ((downgrade = INP_TRY_UPGRADE(inp)) == 0)
473 					return (NULL);
474 			}
475 			ipsec_invalidate_cache(inp, IPSEC_DIR_OUTBOUND);
476 			ipsec_invalidate_cache(inp, IPSEC_DIR_INBOUND);
477 			if (downgrade != 0)
478 				INP_DOWNGRADE(inp);
479 			return (NULL);
480 		}
481 		KEYDBG(IPSEC_STAMP,
482 		    printf("%s: PCB(%p): cache hit SP(%p)\n",
483 		    __func__, inp, sp));
484 		/* Return referenced cached policy */
485 	}
486 	key_addref(sp);
487 	return (sp);
488 }
489 
490 #ifdef INET
491 static void
492 ipsec4_get_ulp(const struct mbuf *m, struct secpolicyindex *spidx,
493     int needport)
494 {
495 	uint8_t nxt;
496 	int off;
497 
498 	/* Sanity check. */
499 	IPSEC_ASSERT(m->m_pkthdr.len >= sizeof(struct ip),
500 	    ("packet too short"));
501 
502 	if (m->m_len >= sizeof (struct ip)) {
503 		const struct ip *ip = mtod(m, const struct ip *);
504 		if (ip->ip_off & htons(IP_MF | IP_OFFMASK))
505 			goto done;
506 		off = ip->ip_hl << 2;
507 		nxt = ip->ip_p;
508 	} else {
509 		struct ip ih;
510 
511 		m_copydata(m, 0, sizeof (struct ip), (caddr_t) &ih);
512 		if (ih.ip_off & htons(IP_MF | IP_OFFMASK))
513 			goto done;
514 		off = ih.ip_hl << 2;
515 		nxt = ih.ip_p;
516 	}
517 
518 	while (off < m->m_pkthdr.len) {
519 		struct ip6_ext ip6e;
520 		struct tcphdr th;
521 		struct udphdr uh;
522 
523 		switch (nxt) {
524 		case IPPROTO_TCP:
525 			spidx->ul_proto = nxt;
526 			if (!needport)
527 				goto done_proto;
528 			if (off + sizeof(struct tcphdr) > m->m_pkthdr.len)
529 				goto done;
530 			m_copydata(m, off, sizeof (th), (caddr_t) &th);
531 			spidx->src.sin.sin_port = th.th_sport;
532 			spidx->dst.sin.sin_port = th.th_dport;
533 			return;
534 		case IPPROTO_UDP:
535 			spidx->ul_proto = nxt;
536 			if (!needport)
537 				goto done_proto;
538 			if (off + sizeof(struct udphdr) > m->m_pkthdr.len)
539 				goto done;
540 			m_copydata(m, off, sizeof (uh), (caddr_t) &uh);
541 			spidx->src.sin.sin_port = uh.uh_sport;
542 			spidx->dst.sin.sin_port = uh.uh_dport;
543 			return;
544 		case IPPROTO_AH:
545 			if (off + sizeof(ip6e) > m->m_pkthdr.len)
546 				goto done;
547 			/* XXX Sigh, this works but is totally bogus. */
548 			m_copydata(m, off, sizeof(ip6e), (caddr_t) &ip6e);
549 			off += (ip6e.ip6e_len + 2) << 2;
550 			nxt = ip6e.ip6e_nxt;
551 			break;
552 		case IPPROTO_ICMP:
553 		default:
554 			/* XXX Intermediate headers??? */
555 			spidx->ul_proto = nxt;
556 			goto done_proto;
557 		}
558 	}
559 done:
560 	spidx->ul_proto = IPSEC_ULPROTO_ANY;
561 done_proto:
562 	spidx->src.sin.sin_port = IPSEC_PORT_ANY;
563 	spidx->dst.sin.sin_port = IPSEC_PORT_ANY;
564 	KEYDBG(IPSEC_DUMP,
565 	    printf("%s: ", __func__); kdebug_secpolicyindex(spidx, NULL));
566 }
567 
568 static void
569 ipsec4_setspidx_ipaddr(const struct mbuf *m, struct secpolicyindex *spidx)
570 {
571 
572 	ipsec4_setsockaddrs(m, &spidx->src, &spidx->dst);
573 	spidx->prefs = sizeof(struct in_addr) << 3;
574 	spidx->prefd = sizeof(struct in_addr) << 3;
575 }
576 
577 static struct secpolicy *
578 ipsec4_getpolicy(const struct mbuf *m, struct inpcb *inp, u_int dir,
579     int needport)
580 {
581 	struct secpolicyindex spidx;
582 	struct secpolicy *sp;
583 
584 	sp = ipsec_getpcbpolicy(inp, dir);
585 	if (sp == NULL && key_havesp(dir)) {
586 		/* Make an index to look for a policy. */
587 		ipsec4_setspidx_ipaddr(m, &spidx);
588 		ipsec4_get_ulp(m, &spidx, needport);
589 		spidx.dir = dir;
590 		sp = key_allocsp(&spidx, dir);
591 	}
592 	if (sp == NULL)		/* No SP found, use system default. */
593 		sp = key_allocsp_default();
594 	return (sp);
595 }
596 
597 /*
598  * Check security policy for *OUTBOUND* IPv4 packet.
599  */
600 struct secpolicy *
601 ipsec4_checkpolicy(const struct mbuf *m, struct inpcb *inp, int *error,
602     int needport)
603 {
604 	struct secpolicy *sp;
605 
606 	*error = 0;
607 	sp = ipsec4_getpolicy(m, inp, IPSEC_DIR_OUTBOUND, needport);
608 	if (sp != NULL)
609 		sp = ipsec_checkpolicy(sp, inp, error);
610 	if (sp == NULL) {
611 		switch (*error) {
612 		case 0: /* No IPsec required: BYPASS or NONE */
613 			break;
614 		case -EINVAL:
615 			IPSECSTAT_INC(ips_out_polvio);
616 			break;
617 		default:
618 			IPSECSTAT_INC(ips_out_inval);
619 		}
620 	}
621 	KEYDBG(IPSEC_STAMP,
622 	    printf("%s: using SP(%p), error %d\n", __func__, sp, *error));
623 	if (sp != NULL)
624 		KEYDBG(IPSEC_DATA, kdebug_secpolicy(sp));
625 	return (sp);
626 }
627 
628 /*
629  * Check IPv4 packet against *INBOUND* security policy.
630  * This function is called from tcp_input(), udp_input(),
631  * rip_input() and sctp_input().
632  */
633 int
634 ipsec4_in_reject(const struct mbuf *m, struct inpcb *inp)
635 {
636 	struct secpolicy *sp;
637 	int result;
638 
639 	sp = ipsec4_getpolicy(m, inp, IPSEC_DIR_INBOUND, 0);
640 	result = ipsec_in_reject(sp, inp, m);
641 	key_freesp(&sp);
642 	if (result != 0)
643 		IPSECSTAT_INC(ips_in_polvio);
644 	return (result);
645 }
646 
647 /*
648  * IPSEC_CAP() method implementation for IPv4.
649  */
650 int
651 ipsec4_capability(struct mbuf *m, u_int cap)
652 {
653 
654 	switch (cap) {
655 	case IPSEC_CAP_BYPASS_FILTER:
656 		/*
657 		 * Bypass packet filtering for packets previously handled
658 		 * by IPsec.
659 		 */
660 		if (!V_ip4_filtertunnel &&
661 		    m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL) != NULL)
662 			return (1);
663 		return (0);
664 	case IPSEC_CAP_OPERABLE:
665 		/* Do we have active security policies? */
666 		if (key_havesp(IPSEC_DIR_INBOUND) != 0 ||
667 		    key_havesp(IPSEC_DIR_OUTBOUND) != 0)
668 			return (1);
669 		return (0);
670 	};
671 	return (EOPNOTSUPP);
672 }
673 
674 #endif /* INET */
675 
676 #ifdef INET6
677 static void
678 ipsec6_get_ulp(const struct mbuf *m, struct secpolicyindex *spidx,
679     int needport)
680 {
681 	struct tcphdr th;
682 	struct udphdr uh;
683 	struct icmp6_hdr ih;
684 	int off, nxt;
685 
686 	IPSEC_ASSERT(m->m_pkthdr.len >= sizeof(struct ip6_hdr),
687 	    ("packet too short"));
688 
689 	/* Set default. */
690 	spidx->ul_proto = IPSEC_ULPROTO_ANY;
691 	spidx->src.sin6.sin6_port = IPSEC_PORT_ANY;
692 	spidx->dst.sin6.sin6_port = IPSEC_PORT_ANY;
693 
694 	nxt = -1;
695 	off = ip6_lasthdr(m, 0, IPPROTO_IPV6, &nxt);
696 	if (off < 0 || m->m_pkthdr.len < off)
697 		return;
698 
699 	switch (nxt) {
700 	case IPPROTO_TCP:
701 		spidx->ul_proto = nxt;
702 		if (!needport)
703 			break;
704 		if (off + sizeof(struct tcphdr) > m->m_pkthdr.len)
705 			break;
706 		m_copydata(m, off, sizeof(th), (caddr_t)&th);
707 		spidx->src.sin6.sin6_port = th.th_sport;
708 		spidx->dst.sin6.sin6_port = th.th_dport;
709 		break;
710 	case IPPROTO_UDP:
711 		spidx->ul_proto = nxt;
712 		if (!needport)
713 			break;
714 		if (off + sizeof(struct udphdr) > m->m_pkthdr.len)
715 			break;
716 		m_copydata(m, off, sizeof(uh), (caddr_t)&uh);
717 		spidx->src.sin6.sin6_port = uh.uh_sport;
718 		spidx->dst.sin6.sin6_port = uh.uh_dport;
719 		break;
720 	case IPPROTO_ICMPV6:
721 		spidx->ul_proto = nxt;
722 		if (off + sizeof(struct icmp6_hdr) > m->m_pkthdr.len)
723 			break;
724 		m_copydata(m, off, sizeof(ih), (caddr_t)&ih);
725 		spidx->src.sin6.sin6_port = htons((uint16_t)ih.icmp6_type);
726 		spidx->dst.sin6.sin6_port = htons((uint16_t)ih.icmp6_code);
727 		break;
728 	default:
729 		/* XXX Intermediate headers??? */
730 		spidx->ul_proto = nxt;
731 		break;
732 	}
733 	KEYDBG(IPSEC_DUMP,
734 	    printf("%s: ", __func__); kdebug_secpolicyindex(spidx, NULL));
735 }
736 
737 static void
738 ipsec6_setspidx_ipaddr(const struct mbuf *m, struct secpolicyindex *spidx)
739 {
740 
741 	ipsec6_setsockaddrs(m, &spidx->src, &spidx->dst);
742 	spidx->prefs = sizeof(struct in6_addr) << 3;
743 	spidx->prefd = sizeof(struct in6_addr) << 3;
744 }
745 
746 static struct secpolicy *
747 ipsec6_getpolicy(const struct mbuf *m, struct inpcb *inp, u_int dir,
748     int needport)
749 {
750 	struct secpolicyindex spidx;
751 	struct secpolicy *sp;
752 
753 	sp = ipsec_getpcbpolicy(inp, dir);
754 	if (sp == NULL && key_havesp(dir)) {
755 		/* Make an index to look for a policy. */
756 		ipsec6_setspidx_ipaddr(m, &spidx);
757 		ipsec6_get_ulp(m, &spidx, needport);
758 		spidx.dir = dir;
759 		sp = key_allocsp(&spidx, dir);
760 	}
761 	if (sp == NULL)		/* No SP found, use system default. */
762 		sp = key_allocsp_default();
763 	return (sp);
764 }
765 
766 /*
767  * Check security policy for *OUTBOUND* IPv6 packet.
768  */
769 struct secpolicy *
770 ipsec6_checkpolicy(const struct mbuf *m, struct inpcb *inp, int *error,
771     int needport)
772 {
773 	struct secpolicy *sp;
774 
775 	*error = 0;
776 	sp = ipsec6_getpolicy(m, inp, IPSEC_DIR_OUTBOUND, needport);
777 	if (sp != NULL)
778 		sp = ipsec_checkpolicy(sp, inp, error);
779 	if (sp == NULL) {
780 		switch (*error) {
781 		case 0: /* No IPsec required: BYPASS or NONE */
782 			break;
783 		case -EINVAL:
784 			IPSEC6STAT_INC(ips_out_polvio);
785 			break;
786 		default:
787 			IPSEC6STAT_INC(ips_out_inval);
788 		}
789 	}
790 	KEYDBG(IPSEC_STAMP,
791 	    printf("%s: using SP(%p), error %d\n", __func__, sp, *error));
792 	if (sp != NULL)
793 		KEYDBG(IPSEC_DATA, kdebug_secpolicy(sp));
794 	return (sp);
795 }
796 
797 /*
798  * Check IPv6 packet against inbound security policy.
799  * This function is called from tcp6_input(), udp6_input(),
800  * rip6_input() and sctp_input().
801  */
802 int
803 ipsec6_in_reject(const struct mbuf *m, struct inpcb *inp)
804 {
805 	struct secpolicy *sp;
806 	int result;
807 
808 	sp = ipsec6_getpolicy(m, inp, IPSEC_DIR_INBOUND, 0);
809 	result = ipsec_in_reject(sp, inp, m);
810 	key_freesp(&sp);
811 	if (result)
812 		IPSEC6STAT_INC(ips_in_polvio);
813 	return (result);
814 }
815 
816 /*
817  * IPSEC_CAP() method implementation for IPv6.
818  */
819 int
820 ipsec6_capability(struct mbuf *m, u_int cap)
821 {
822 
823 	switch (cap) {
824 	case IPSEC_CAP_BYPASS_FILTER:
825 		/*
826 		 * Bypass packet filtering for packets previously handled
827 		 * by IPsec.
828 		 */
829 		if (!V_ip6_filtertunnel &&
830 		    m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL) != NULL)
831 			return (1);
832 		return (0);
833 	case IPSEC_CAP_OPERABLE:
834 		/* Do we have active security policies? */
835 		if (key_havesp(IPSEC_DIR_INBOUND) != 0 ||
836 		    key_havesp(IPSEC_DIR_OUTBOUND) != 0)
837 			return (1);
838 		return (0);
839 	};
840 	return (EOPNOTSUPP);
841 }
842 #endif /* INET6 */
843 
844 int
845 ipsec_run_hhooks(struct ipsec_ctx_data *ctx, int type)
846 {
847 	int idx;
848 
849 	switch (ctx->af) {
850 #ifdef INET
851 	case AF_INET:
852 		idx = HHOOK_IPSEC_INET;
853 		break;
854 #endif
855 #ifdef INET6
856 	case AF_INET6:
857 		idx = HHOOK_IPSEC_INET6;
858 		break;
859 #endif
860 	default:
861 		return (EPFNOSUPPORT);
862 	}
863 	if (type == HHOOK_TYPE_IPSEC_IN)
864 		HHOOKS_RUN_IF(V_ipsec_hhh_in[idx], ctx, NULL);
865 	else
866 		HHOOKS_RUN_IF(V_ipsec_hhh_out[idx], ctx, NULL);
867 	if (*ctx->mp == NULL)
868 		return (EACCES);
869 	return (0);
870 }
871 
872 /*
873  * Return current level.
874  * Either IPSEC_LEVEL_USE or IPSEC_LEVEL_REQUIRE are always returned.
875  */
876 u_int
877 ipsec_get_reqlevel(struct secpolicy *sp, u_int idx)
878 {
879 	struct ipsecrequest *isr;
880 	u_int esp_trans_deflev, esp_net_deflev;
881 	u_int ah_trans_deflev, ah_net_deflev;
882 	u_int level = 0;
883 
884 	IPSEC_ASSERT(idx < sp->tcount, ("Wrong IPsec request index %d", idx));
885 /* XXX Note that we have ipseclog() expanded here - code sync issue. */
886 #define IPSEC_CHECK_DEFAULT(lev) \
887 	(((lev) != IPSEC_LEVEL_USE && (lev) != IPSEC_LEVEL_REQUIRE &&	\
888 	  (lev) != IPSEC_LEVEL_UNIQUE)					\
889 		? (V_ipsec_debug  ?					\
890 		log(LOG_INFO, "fixed system default level " #lev ":%d->%d\n",\
891 		(lev), IPSEC_LEVEL_REQUIRE) : 0),			\
892 		(lev) = IPSEC_LEVEL_REQUIRE, (lev) : (lev))
893 
894 	/*
895 	 * IPsec VTI uses unique security policy with fake spidx filled
896 	 * with zeroes. Just return IPSEC_LEVEL_REQUIRE instead of doing
897 	 * full level lookup for such policies.
898 	 */
899 	if (sp->state == IPSEC_SPSTATE_IFNET) {
900 		IPSEC_ASSERT(sp->req[idx]->level == IPSEC_LEVEL_UNIQUE,
901 		    ("Wrong IPsec request level %d", sp->req[idx]->level));
902 		return (IPSEC_LEVEL_REQUIRE);
903 	}
904 
905 	/* Set default level. */
906 	switch (sp->spidx.src.sa.sa_family) {
907 #ifdef INET
908 	case AF_INET:
909 		esp_trans_deflev = IPSEC_CHECK_DEFAULT(V_ip4_esp_trans_deflev);
910 		esp_net_deflev = IPSEC_CHECK_DEFAULT(V_ip4_esp_net_deflev);
911 		ah_trans_deflev = IPSEC_CHECK_DEFAULT(V_ip4_ah_trans_deflev);
912 		ah_net_deflev = IPSEC_CHECK_DEFAULT(V_ip4_ah_net_deflev);
913 		break;
914 #endif
915 #ifdef INET6
916 	case AF_INET6:
917 		esp_trans_deflev = IPSEC_CHECK_DEFAULT(V_ip6_esp_trans_deflev);
918 		esp_net_deflev = IPSEC_CHECK_DEFAULT(V_ip6_esp_net_deflev);
919 		ah_trans_deflev = IPSEC_CHECK_DEFAULT(V_ip6_ah_trans_deflev);
920 		ah_net_deflev = IPSEC_CHECK_DEFAULT(V_ip6_ah_net_deflev);
921 		break;
922 #endif /* INET6 */
923 	default:
924 		panic("%s: unknown af %u",
925 			__func__, sp->spidx.src.sa.sa_family);
926 	}
927 
928 #undef IPSEC_CHECK_DEFAULT
929 
930 	isr = sp->req[idx];
931 	/* Set level. */
932 	switch (isr->level) {
933 	case IPSEC_LEVEL_DEFAULT:
934 		switch (isr->saidx.proto) {
935 		case IPPROTO_ESP:
936 			if (isr->saidx.mode == IPSEC_MODE_TUNNEL)
937 				level = esp_net_deflev;
938 			else
939 				level = esp_trans_deflev;
940 			break;
941 		case IPPROTO_AH:
942 			if (isr->saidx.mode == IPSEC_MODE_TUNNEL)
943 				level = ah_net_deflev;
944 			else
945 				level = ah_trans_deflev;
946 			break;
947 		case IPPROTO_IPCOMP:
948 			/*
949 			 * We don't really care, as IPcomp document says that
950 			 * we shouldn't compress small packets.
951 			 */
952 			level = IPSEC_LEVEL_USE;
953 			break;
954 		default:
955 			panic("%s: Illegal protocol defined %u\n", __func__,
956 				isr->saidx.proto);
957 		}
958 		break;
959 
960 	case IPSEC_LEVEL_USE:
961 	case IPSEC_LEVEL_REQUIRE:
962 		level = isr->level;
963 		break;
964 	case IPSEC_LEVEL_UNIQUE:
965 		level = IPSEC_LEVEL_REQUIRE;
966 		break;
967 
968 	default:
969 		panic("%s: Illegal IPsec level %u\n", __func__, isr->level);
970 	}
971 
972 	return (level);
973 }
974 
975 static int
976 ipsec_check_history(const struct mbuf *m, struct secpolicy *sp, u_int idx)
977 {
978 	struct xform_history *xh;
979 	struct m_tag *mtag;
980 
981 	mtag = NULL;
982 	while ((mtag = m_tag_find(__DECONST(struct mbuf *, m),
983 	    PACKET_TAG_IPSEC_IN_DONE, mtag)) != NULL) {
984 		xh = (struct xform_history *)(mtag + 1);
985 		KEYDBG(IPSEC_DATA,
986 		    char buf[IPSEC_ADDRSTRLEN];
987 		    printf("%s: mode %s proto %u dst %s\n", __func__,
988 			kdebug_secasindex_mode(xh->mode), xh->proto,
989 			ipsec_address(&xh->dst, buf, sizeof(buf))));
990 		if (xh->proto != sp->req[idx]->saidx.proto)
991 			continue;
992 		/* If SA had IPSEC_MODE_ANY, consider this as match. */
993 		if (xh->mode != sp->req[idx]->saidx.mode &&
994 		    xh->mode != IPSEC_MODE_ANY)
995 			continue;
996 		/*
997 		 * For transport mode IPsec request doesn't contain
998 		 * addresses. We need to use address from spidx.
999 		 */
1000 		if (sp->req[idx]->saidx.mode == IPSEC_MODE_TRANSPORT) {
1001 			if (key_sockaddrcmp_withmask(&xh->dst.sa,
1002 			    &sp->spidx.dst.sa, sp->spidx.prefd) != 0)
1003 				continue;
1004 		} else {
1005 			if (key_sockaddrcmp(&xh->dst.sa,
1006 			    &sp->req[idx]->saidx.dst.sa, 0) != 0)
1007 				continue;
1008 		}
1009 		return (0); /* matched */
1010 	}
1011 	return (1);
1012 }
1013 
1014 /*
1015  * Check security policy requirements against the actual
1016  * packet contents.  Return one if the packet should be
1017  * reject as "invalid"; otherwiser return zero to have the
1018  * packet treated as "valid".
1019  *
1020  * OUT:
1021  *	0: valid
1022  *	1: invalid
1023  */
1024 static int
1025 ipsec_in_reject(struct secpolicy *sp, struct inpcb *inp, const struct mbuf *m)
1026 {
1027 	int i;
1028 
1029 	KEYDBG(IPSEC_STAMP,
1030 	    printf("%s: PCB(%p): using SP(%p)\n", __func__, inp, sp));
1031 	KEYDBG(IPSEC_DATA, kdebug_secpolicy(sp));
1032 
1033 	if (inp != NULL && inp->inp_sp != NULL && inp->inp_sp->sp_in == NULL)
1034 		ipsec_cachepolicy(inp, sp, IPSEC_DIR_INBOUND);
1035 
1036 	/* Check policy. */
1037 	switch (sp->policy) {
1038 	case IPSEC_POLICY_DISCARD:
1039 		return (1);
1040 	case IPSEC_POLICY_BYPASS:
1041 	case IPSEC_POLICY_NONE:
1042 		return (0);
1043 	}
1044 
1045 	IPSEC_ASSERT(sp->policy == IPSEC_POLICY_IPSEC,
1046 		("invalid policy %u", sp->policy));
1047 
1048 	/*
1049 	 * ipsec[46]_common_input_cb after each transform adds
1050 	 * PACKET_TAG_IPSEC_IN_DONE mbuf tag. It contains SPI, proto, mode
1051 	 * and destination address from saidx. We can compare info from
1052 	 * these tags with requirements in SP.
1053 	 */
1054 	for (i = 0; i < sp->tcount; i++) {
1055 		/*
1056 		 * Do not check IPcomp, since IPcomp document
1057 		 * says that we shouldn't compress small packets.
1058 		 * IPComp policy should always be treated as being
1059 		 * in "use" level.
1060 		 */
1061 		if (sp->req[i]->saidx.proto == IPPROTO_IPCOMP ||
1062 		    ipsec_get_reqlevel(sp, i) != IPSEC_LEVEL_REQUIRE)
1063 			continue;
1064 		if (V_check_policy_history != 0 &&
1065 		    ipsec_check_history(m, sp, i) != 0)
1066 			return (1);
1067 		else switch (sp->req[i]->saidx.proto) {
1068 		case IPPROTO_ESP:
1069 			if ((m->m_flags & M_DECRYPTED) == 0) {
1070 				KEYDBG(IPSEC_DUMP,
1071 				    printf("%s: ESP m_flags:%x\n", __func__,
1072 					    m->m_flags));
1073 				return (1);
1074 			}
1075 			break;
1076 		case IPPROTO_AH:
1077 			if ((m->m_flags & M_AUTHIPHDR) == 0) {
1078 				KEYDBG(IPSEC_DUMP,
1079 				    printf("%s: AH m_flags:%x\n", __func__,
1080 					    m->m_flags));
1081 				return (1);
1082 			}
1083 			break;
1084 		}
1085 	}
1086 	return (0);		/* Valid. */
1087 }
1088 
1089 /*
1090  * Compute the byte size to be occupied by IPsec header.
1091  * In case it is tunnelled, it includes the size of outer IP header.
1092  */
1093 static size_t
1094 ipsec_hdrsiz_internal(struct secpolicy *sp)
1095 {
1096 	size_t size;
1097 	int i;
1098 
1099 	KEYDBG(IPSEC_STAMP, printf("%s: using SP(%p)\n", __func__, sp));
1100 	KEYDBG(IPSEC_DATA, kdebug_secpolicy(sp));
1101 
1102 	switch (sp->policy) {
1103 	case IPSEC_POLICY_DISCARD:
1104 	case IPSEC_POLICY_BYPASS:
1105 	case IPSEC_POLICY_NONE:
1106 		return (0);
1107 	}
1108 
1109 	IPSEC_ASSERT(sp->policy == IPSEC_POLICY_IPSEC,
1110 		("invalid policy %u", sp->policy));
1111 
1112 	/*
1113 	 * XXX: for each transform we need to lookup suitable SA
1114 	 * and use info from SA to calculate headers size.
1115 	 * XXX: for NAT-T we need to cosider UDP header size.
1116 	 */
1117 	size = 0;
1118 	for (i = 0; i < sp->tcount; i++) {
1119 		switch (sp->req[i]->saidx.proto) {
1120 		case IPPROTO_ESP:
1121 			size += esp_hdrsiz(NULL);
1122 			break;
1123 		case IPPROTO_AH:
1124 			size += ah_hdrsiz(NULL);
1125 			break;
1126 		case IPPROTO_IPCOMP:
1127 			size += sizeof(struct ipcomp);
1128 			break;
1129 		}
1130 
1131 		if (sp->req[i]->saidx.mode == IPSEC_MODE_TUNNEL) {
1132 			switch (sp->req[i]->saidx.dst.sa.sa_family) {
1133 #ifdef INET
1134 			case AF_INET:
1135 				size += sizeof(struct ip);
1136 				break;
1137 #endif
1138 #ifdef INET6
1139 			case AF_INET6:
1140 				size += sizeof(struct ip6_hdr);
1141 				break;
1142 #endif
1143 			default:
1144 				ipseclog((LOG_ERR, "%s: unknown AF %d in "
1145 				    "IPsec tunnel SA\n", __func__,
1146 				    sp->req[i]->saidx.dst.sa.sa_family));
1147 				break;
1148 			}
1149 		}
1150 	}
1151 	return (size);
1152 }
1153 
1154 /*
1155  * Compute ESP/AH header size for protocols with PCB, including
1156  * outer IP header. Currently only tcp_output() uses it.
1157  */
1158 size_t
1159 ipsec_hdrsiz_inpcb(struct inpcb *inp)
1160 {
1161 	struct secpolicyindex spidx;
1162 	struct secpolicy *sp;
1163 	size_t sz;
1164 
1165 	sp = ipsec_getpcbpolicy(inp, IPSEC_DIR_OUTBOUND);
1166 	if (sp == NULL && key_havesp(IPSEC_DIR_OUTBOUND)) {
1167 		ipsec_setspidx_inpcb(inp, &spidx, IPSEC_DIR_OUTBOUND);
1168 		sp = key_allocsp(&spidx, IPSEC_DIR_OUTBOUND);
1169 	}
1170 	if (sp == NULL)
1171 		sp = key_allocsp_default();
1172 	sz = ipsec_hdrsiz_internal(sp);
1173 	key_freesp(&sp);
1174 	return (sz);
1175 }
1176 
1177 /*
1178  * Check the variable replay window.
1179  * ipsec_chkreplay() performs replay check before ICV verification.
1180  * ipsec_updatereplay() updates replay bitmap.  This must be called after
1181  * ICV verification (it also performs replay check, which is usually done
1182  * beforehand).
1183  * 0 (zero) is returned if packet disallowed, 1 if packet permitted.
1184  *
1185  * Based on RFC 6479. Blocks are 32 bits unsigned integers
1186  */
1187 
1188 #define IPSEC_BITMAP_INDEX_MASK(w)	(w - 1)
1189 #define IPSEC_REDUNDANT_BIT_SHIFTS	5
1190 #define IPSEC_REDUNDANT_BITS		(1 << IPSEC_REDUNDANT_BIT_SHIFTS)
1191 #define IPSEC_BITMAP_LOC_MASK		(IPSEC_REDUNDANT_BITS - 1)
1192 
1193 int
1194 ipsec_chkreplay(uint32_t seq, struct secasvar *sav)
1195 {
1196 	const struct secreplay *replay;
1197 	uint32_t wsizeb;		/* Constant: window size. */
1198 	int index, bit_location;
1199 
1200 	IPSEC_ASSERT(sav != NULL, ("Null SA"));
1201 	IPSEC_ASSERT(sav->replay != NULL, ("Null replay state"));
1202 
1203 	replay = sav->replay;
1204 
1205 	/* No need to check replay if disabled. */
1206 	if (replay->wsize == 0)
1207 		return (1);
1208 
1209 	/* Constant. */
1210 	wsizeb = replay->wsize << 3;
1211 
1212 	/* Sequence number of 0 is invalid. */
1213 	if (seq == 0)
1214 		return (0);
1215 
1216 	/* First time is always okay. */
1217 	if (replay->count == 0)
1218 		return (1);
1219 
1220 	/* Larger sequences are okay. */
1221 	if (seq > replay->lastseq)
1222 		return (1);
1223 
1224 	/* Over range to check, i.e. too old or wrapped. */
1225 	if (replay->lastseq - seq >= wsizeb)
1226 		return (0);
1227 
1228 	/* The sequence is inside the sliding window
1229 	 * now check the bit in the bitmap
1230 	 * bit location only depends on the sequence number
1231 	 */
1232 	bit_location = seq & IPSEC_BITMAP_LOC_MASK;
1233 	index = (seq >> IPSEC_REDUNDANT_BIT_SHIFTS)
1234 		& IPSEC_BITMAP_INDEX_MASK(replay->bitmap_size);
1235 
1236 	/* This packet already seen? */
1237 	if ((replay->bitmap)[index] & (1 << bit_location))
1238 		return (0);
1239 	return (1);
1240 }
1241 
1242 /*
1243  * Check replay counter whether to update or not.
1244  * OUT:	0:	OK
1245  *	1:	NG
1246  */
1247 int
1248 ipsec_updatereplay(uint32_t seq, struct secasvar *sav)
1249 {
1250 	char buf[128];
1251 	struct secreplay *replay;
1252 	uint32_t wsizeb;		/* Constant: window size. */
1253 	int diff, index, bit_location;
1254 
1255 	IPSEC_ASSERT(sav != NULL, ("Null SA"));
1256 	IPSEC_ASSERT(sav->replay != NULL, ("Null replay state"));
1257 
1258 	replay = sav->replay;
1259 
1260 	if (replay->wsize == 0)
1261 		goto ok;	/* No need to check replay. */
1262 
1263 	/* Constant. */
1264 	wsizeb = replay->wsize << 3;
1265 
1266 	/* Sequence number of 0 is invalid. */
1267 	if (seq == 0)
1268 		return (1);
1269 
1270 	/* The packet is too old, no need to update */
1271 	if (wsizeb + seq < replay->lastseq)
1272 		goto ok;
1273 
1274 	/* Now update the bit */
1275 	index = (seq >> IPSEC_REDUNDANT_BIT_SHIFTS);
1276 
1277 	/* First check if the sequence number is in the range */
1278 	if (seq > replay->lastseq) {
1279 		int id;
1280 		int index_cur = replay->lastseq >> IPSEC_REDUNDANT_BIT_SHIFTS;
1281 
1282 		diff = index - index_cur;
1283 		if (diff > replay->bitmap_size) {
1284 			/* something unusual in this case */
1285 			diff = replay->bitmap_size;
1286 		}
1287 
1288 		for (id = 0; id < diff; ++id) {
1289 			replay->bitmap[(id + index_cur + 1)
1290 			& IPSEC_BITMAP_INDEX_MASK(replay->bitmap_size)] = 0;
1291 		}
1292 
1293 		replay->lastseq = seq;
1294 	}
1295 
1296 	index &= IPSEC_BITMAP_INDEX_MASK(replay->bitmap_size);
1297 	bit_location = seq & IPSEC_BITMAP_LOC_MASK;
1298 
1299 	/* this packet has already been received */
1300 	if (replay->bitmap[index] & (1 << bit_location))
1301 		return (1);
1302 
1303 	replay->bitmap[index] |= (1 << bit_location);
1304 
1305 ok:
1306 	if (replay->count == ~0) {
1307 
1308 		/* Set overflow flag. */
1309 		replay->overflow++;
1310 
1311 		/* Don't increment, no more packets accepted. */
1312 		if ((sav->flags & SADB_X_EXT_CYCSEQ) == 0) {
1313 			if (sav->sah->saidx.proto == IPPROTO_AH)
1314 				AHSTAT_INC(ahs_wrap);
1315 			else if (sav->sah->saidx.proto == IPPROTO_ESP)
1316 				ESPSTAT_INC(esps_wrap);
1317 			return (1);
1318 		}
1319 
1320 		ipseclog((LOG_WARNING, "%s: replay counter made %d cycle. %s\n",
1321 		    __func__, replay->overflow,
1322 		    ipsec_sa2str(sav, buf, sizeof(buf))));
1323 	}
1324 	return (0);
1325 }
1326 
1327 int
1328 ipsec_updateid(struct secasvar *sav, uint64_t *new, uint64_t *old)
1329 {
1330 	uint64_t tmp;
1331 
1332 	/*
1333 	 * tdb_cryptoid is initialized by xform_init().
1334 	 * Then it can be changed only when some crypto error occurred or
1335 	 * when SA is deleted. We stored used cryptoid in the xform_data
1336 	 * structure. In case when crypto error occurred and crypto
1337 	 * subsystem has reinited the session, it returns new cryptoid
1338 	 * and EAGAIN error code.
1339 	 *
1340 	 * This function will be called when we got EAGAIN from crypto
1341 	 * subsystem.
1342 	 * *new is cryptoid that was returned by crypto subsystem in
1343 	 * the crp_sid.
1344 	 * *old is the original cryptoid that we stored in xform_data.
1345 	 *
1346 	 * For first failed request *old == sav->tdb_cryptoid, then
1347 	 * we update sav->tdb_cryptoid and redo crypto_dispatch().
1348 	 * For next failed request *old != sav->tdb_cryptoid, then
1349 	 * we store cryptoid from first request into the *new variable
1350 	 * and crp_sid from this second session will be returned via
1351 	 * *old pointer, so caller can release second session.
1352 	 *
1353 	 * XXXAE: check this more carefully.
1354 	 */
1355 	KEYDBG(IPSEC_STAMP,
1356 	    printf("%s: SA(%p) moves cryptoid %jd -> %jd\n",
1357 		__func__, sav, (uintmax_t)(*old), (uintmax_t)(*new)));
1358 	KEYDBG(IPSEC_DATA, kdebug_secasv(sav));
1359 	SECASVAR_LOCK(sav);
1360 	if (sav->tdb_cryptoid != *old) {
1361 		/* cryptoid was already updated */
1362 		tmp = *new;
1363 		*new = sav->tdb_cryptoid;
1364 		*old = tmp;
1365 		SECASVAR_UNLOCK(sav);
1366 		return (1);
1367 	}
1368 	sav->tdb_cryptoid = *new;
1369 	SECASVAR_UNLOCK(sav);
1370 	return (0);
1371 }
1372 
1373 int
1374 ipsec_initialized(void)
1375 {
1376 
1377 	return (V_def_policy != NULL);
1378 }
1379 
1380 static void
1381 def_policy_init(const void *unused __unused)
1382 {
1383 
1384 	V_def_policy = key_newsp();
1385 	if (V_def_policy != NULL) {
1386 		V_def_policy->policy = IPSEC_POLICY_NONE;
1387 		/* Force INPCB SP cache invalidation */
1388 		key_bumpspgen();
1389 	} else
1390 		printf("%s: failed to initialize default policy\n", __func__);
1391 }
1392 
1393 
1394 static void
1395 def_policy_uninit(const void *unused __unused)
1396 {
1397 
1398 	if (V_def_policy != NULL) {
1399 		key_freesp(&V_def_policy);
1400 		key_bumpspgen();
1401 	}
1402 }
1403 
1404 VNET_SYSINIT(def_policy_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST,
1405     def_policy_init, NULL);
1406 VNET_SYSUNINIT(def_policy_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST,
1407     def_policy_uninit, NULL);
1408