xref: /freebsd-13-stable/sys/netipsec/ipsec.c (revision 4fbf14e22d7b83de7080a8e491ba14a5785a0ff4)
1 /*	$KAME: ipsec.c,v 1.103 2001/05/24 07:14:18 sakane Exp $	*/
2 
3 /*-
4  * SPDX-License-Identifier: BSD-3-Clause
5  *
6  * Copyright (C) 1995, 1996, 1997, and 1998 WIDE Project.
7  * All rights reserved.
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  * 1. Redistributions of source code must retain the above copyright
13  *    notice, this list of conditions and the following disclaimer.
14  * 2. Redistributions in binary form must reproduce the above copyright
15  *    notice, this list of conditions and the following disclaimer in the
16  *    documentation and/or other materials provided with the distribution.
17  * 3. Neither the name of the project nor the names of its contributors
18  *    may be used to endorse or promote products derived from this software
19  *    without specific prior written permission.
20  *
21  * THIS SOFTWARE IS PROVIDED BY THE PROJECT AND CONTRIBUTORS ``AS IS'' AND
22  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
23  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
24  * ARE DISCLAIMED.  IN NO EVENT SHALL THE PROJECT OR CONTRIBUTORS BE LIABLE
25  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
26  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
27  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
28  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
29  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
30  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31  * SUCH DAMAGE.
32  */
33 
34 /*
35  * IPsec controller part.
36  */
37 
38 #include "opt_inet.h"
39 #include "opt_inet6.h"
40 #include "opt_ipsec.h"
41 
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/malloc.h>
45 #include <sys/mbuf.h>
46 #include <sys/domain.h>
47 #include <sys/priv.h>
48 #include <sys/protosw.h>
49 #include <sys/socket.h>
50 #include <sys/socketvar.h>
51 #include <sys/errno.h>
52 #include <sys/hhook.h>
53 #include <sys/time.h>
54 #include <sys/kernel.h>
55 #include <sys/syslog.h>
56 #include <sys/sysctl.h>
57 #include <sys/proc.h>
58 
59 #include <net/if.h>
60 #include <net/if_enc.h>
61 #include <net/if_var.h>
62 #include <net/vnet.h>
63 
64 #include <netinet/in.h>
65 #include <netinet/in_systm.h>
66 #include <netinet/ip.h>
67 #include <netinet/ip_var.h>
68 #include <netinet/in_var.h>
69 #include <netinet/udp.h>
70 #include <netinet/udp_var.h>
71 #include <netinet/tcp.h>
72 #include <netinet/udp.h>
73 
74 #include <netinet/ip6.h>
75 #ifdef INET6
76 #include <netinet6/ip6_var.h>
77 #endif
78 #include <netinet/in_pcb.h>
79 #ifdef INET6
80 #include <netinet/icmp6.h>
81 #endif
82 
83 #include <sys/types.h>
84 #include <netipsec/ipsec.h>
85 #ifdef INET6
86 #include <netipsec/ipsec6.h>
87 #endif
88 #include <netipsec/ah_var.h>
89 #include <netipsec/esp_var.h>
90 #include <netipsec/ipcomp.h>		/*XXX*/
91 #include <netipsec/ipcomp_var.h>
92 #include <netipsec/ipsec_support.h>
93 
94 #include <netipsec/key.h>
95 #include <netipsec/keydb.h>
96 #include <netipsec/key_debug.h>
97 
98 #include <netipsec/xform.h>
99 
100 #include <machine/in_cksum.h>
101 
102 #include <opencrypto/cryptodev.h>
103 
104 /* NB: name changed so netstat doesn't use it. */
105 VNET_PCPUSTAT_DEFINE(struct ipsecstat, ipsec4stat);
106 VNET_PCPUSTAT_SYSINIT(ipsec4stat);
107 
108 #ifdef VIMAGE
109 VNET_PCPUSTAT_SYSUNINIT(ipsec4stat);
110 #endif /* VIMAGE */
111 
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 
121 VNET_DEFINE_STATIC(int, ip4_filtertunnel) = 0;
122 #define	V_ip4_filtertunnel VNET(ip4_filtertunnel)
123 VNET_DEFINE_STATIC(int, check_policy_history) = 0;
124 #define	V_check_policy_history	VNET(check_policy_history)
125 VNET_DEFINE_STATIC(struct secpolicy *, def_policy) = NULL;
126 #define	V_def_policy	VNET(def_policy)
127 static int
sysctl_def_policy(SYSCTL_HANDLER_ARGS)128 sysctl_def_policy(SYSCTL_HANDLER_ARGS)
129 {
130 	int error, value;
131 
132 	value = V_def_policy->policy;
133 	error = sysctl_handle_int(oidp, &value, 0, req);
134 	if (error == 0) {
135 		if (value != IPSEC_POLICY_DISCARD &&
136 		    value != IPSEC_POLICY_NONE)
137 			return (EINVAL);
138 		V_def_policy->policy = value;
139 	}
140 	return (error);
141 }
142 
143 /*
144  * Crypto support requirements:
145  *
146  *  1	require hardware support
147  * -1	require software support
148  *  0	take anything
149  */
150 VNET_DEFINE(int, crypto_support) = CRYPTOCAP_F_HARDWARE | CRYPTOCAP_F_SOFTWARE;
151 
152 /*
153  * Use asynchronous mode to parallelize crypto jobs:
154  *
155  *  0 - disabled
156  *  1 - enabled
157  */
158 VNET_DEFINE(int, async_crypto) = 0;
159 
160 /*
161  * TCP/UDP checksum handling policy for transport mode NAT-T (RFC3948)
162  *
163  * 0 - auto: incrementally recompute, when checksum delta is known;
164  *     if checksum delta isn't known, reset checksum to zero for UDP,
165  *     and mark csum_flags as valid for TCP.
166  * 1 - fully recompute TCP/UDP checksum.
167  */
168 VNET_DEFINE(int, natt_cksum_policy) = 0;
169 
170 FEATURE(ipsec, "Internet Protocol Security (IPsec)");
171 FEATURE(ipsec_natt, "UDP Encapsulation of IPsec ESP Packets ('NAT-T')");
172 
173 SYSCTL_DECL(_net_inet_ipsec);
174 
175 /* net.inet.ipsec */
176 SYSCTL_PROC(_net_inet_ipsec, IPSECCTL_DEF_POLICY, def_policy,
177     CTLTYPE_INT | CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
178     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 #ifdef REGRESSION
220 /*
221  * When set to 1, IPsec will send packets with the same sequence number.
222  * This allows to verify if the other side has proper replay attacks detection.
223  */
224 VNET_DEFINE(int, ipsec_replay) = 0;
225 SYSCTL_INT(_net_inet_ipsec, OID_AUTO, test_replay,
226 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ipsec_replay), 0,
227 	"Emulate replay attack");
228 /*
229  * When set 1, IPsec will send packets with corrupted HMAC.
230  * This allows to verify if the other side properly detects modified packets.
231  */
232 VNET_DEFINE(int, ipsec_integrity) = 0;
233 SYSCTL_INT(_net_inet_ipsec, OID_AUTO, test_integrity,
234 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ipsec_integrity), 0,
235 	"Emulate man-in-the-middle attack");
236 #endif
237 
238 #ifdef INET6
239 VNET_PCPUSTAT_DEFINE(struct ipsecstat, ipsec6stat);
240 VNET_PCPUSTAT_SYSINIT(ipsec6stat);
241 
242 #ifdef VIMAGE
243 VNET_PCPUSTAT_SYSUNINIT(ipsec6stat);
244 #endif /* VIMAGE */
245 
246 VNET_DEFINE(int, ip6_esp_trans_deflev) = IPSEC_LEVEL_USE;
247 VNET_DEFINE(int, ip6_esp_net_deflev) = IPSEC_LEVEL_USE;
248 VNET_DEFINE(int, ip6_ah_trans_deflev) = IPSEC_LEVEL_USE;
249 VNET_DEFINE(int, ip6_ah_net_deflev) = IPSEC_LEVEL_USE;
250 VNET_DEFINE(int, ip6_ipsec_ecn) = 0;	/* ECN ignore(-1)/forbidden(0)/allowed(1) */
251 
252 VNET_DEFINE_STATIC(int, ip6_filtertunnel) = 0;
253 #define	V_ip6_filtertunnel	VNET(ip6_filtertunnel)
254 
255 SYSCTL_DECL(_net_inet6_ipsec6);
256 
257 /* net.inet6.ipsec6 */
258 SYSCTL_PROC(_net_inet6_ipsec6, IPSECCTL_DEF_POLICY, def_policy,
259     CTLTYPE_INT | CTLFLAG_VNET | CTLFLAG_RW | CTLFLAG_NEEDGIANT,
260     0, 0, sysctl_def_policy, "I",
261     "IPsec default policy.");
262 SYSCTL_INT(_net_inet6_ipsec6, IPSECCTL_DEF_ESP_TRANSLEV, esp_trans_deflev,
263 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_esp_trans_deflev), 0,
264 	"Default ESP transport mode level.");
265 SYSCTL_INT(_net_inet6_ipsec6, IPSECCTL_DEF_ESP_NETLEV, esp_net_deflev,
266 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_esp_net_deflev), 0,
267 	"Default ESP tunnel mode level.");
268 SYSCTL_INT(_net_inet6_ipsec6, IPSECCTL_DEF_AH_TRANSLEV, ah_trans_deflev,
269 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_ah_trans_deflev), 0,
270 	"AH transfer mode default level.");
271 SYSCTL_INT(_net_inet6_ipsec6, IPSECCTL_DEF_AH_NETLEV, ah_net_deflev,
272 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_ah_net_deflev), 0,
273 	"AH tunnel mode default level.");
274 SYSCTL_INT(_net_inet6_ipsec6, IPSECCTL_ECN, ecn,
275 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_ipsec_ecn), 0,
276 	"Explicit Congestion Notification handling.");
277 SYSCTL_INT(_net_inet6_ipsec6, OID_AUTO, filtertunnel,
278 	CTLFLAG_VNET | CTLFLAG_RW, &VNET_NAME(ip6_filtertunnel),  0,
279 	"If set, filter packets from an IPsec tunnel.");
280 SYSCTL_VNET_PCPUSTAT(_net_inet6_ipsec6, IPSECCTL_STATS, ipsecstats,
281     struct ipsecstat, ipsec6stat, "IPsec IPv6 statistics.");
282 #endif /* INET6 */
283 
284 static int ipsec_in_reject(struct secpolicy *, struct inpcb *,
285     const struct mbuf *);
286 
287 #ifdef INET
288 static void ipsec4_get_ulp(const struct mbuf *, struct secpolicyindex *, int);
289 static void ipsec4_setspidx_ipaddr(const struct mbuf *,
290     struct secpolicyindex *);
291 #endif
292 #ifdef INET6
293 static void ipsec6_get_ulp(const struct mbuf *m, struct secpolicyindex *, int);
294 static void ipsec6_setspidx_ipaddr(const struct mbuf *,
295     struct secpolicyindex *);
296 #endif
297 
298 /*
299  * Return a held reference to the default SP.
300  */
301 static struct secpolicy *
key_allocsp_default(void)302 key_allocsp_default(void)
303 {
304 
305 	key_addref(V_def_policy);
306 	return (V_def_policy);
307 }
308 
309 static void
ipsec_invalidate_cache(struct inpcb * inp,u_int dir)310 ipsec_invalidate_cache(struct inpcb *inp, u_int dir)
311 {
312 	struct secpolicy *sp;
313 
314 	INP_WLOCK_ASSERT(inp);
315 	if (dir == IPSEC_DIR_OUTBOUND) {
316 		if (inp->inp_sp->flags & INP_INBOUND_POLICY)
317 			return;
318 		sp = inp->inp_sp->sp_in;
319 		inp->inp_sp->sp_in = NULL;
320 	} else {
321 		if (inp->inp_sp->flags & INP_OUTBOUND_POLICY)
322 			return;
323 		sp = inp->inp_sp->sp_out;
324 		inp->inp_sp->sp_out = NULL;
325 	}
326 	if (sp != NULL)
327 		key_freesp(&sp); /* release extra reference */
328 }
329 
330 static void
ipsec_cachepolicy(struct inpcb * inp,struct secpolicy * sp,u_int dir)331 ipsec_cachepolicy(struct inpcb *inp, struct secpolicy *sp, u_int dir)
332 {
333 	uint32_t genid;
334 	int downgrade;
335 
336 	INP_LOCK_ASSERT(inp);
337 
338 	if (dir == IPSEC_DIR_OUTBOUND) {
339 		/* Do we have configured PCB policy? */
340 		if (inp->inp_sp->flags & INP_OUTBOUND_POLICY)
341 			return;
342 		/* Another thread has already set cached policy */
343 		if (inp->inp_sp->sp_out != NULL)
344 			return;
345 		/*
346 		 * Do not cache OUTBOUND policy if PCB isn't connected,
347 		 * i.e. foreign address is INADDR_ANY/UNSPECIFIED.
348 		 */
349 #ifdef INET
350 		if ((inp->inp_vflag & INP_IPV4) != 0 &&
351 		    inp->inp_faddr.s_addr == INADDR_ANY)
352 			return;
353 #endif
354 #ifdef INET6
355 		if ((inp->inp_vflag & INP_IPV6) != 0 &&
356 		    IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr))
357 			return;
358 #endif
359 	} else {
360 		/* Do we have configured PCB policy? */
361 		if (inp->inp_sp->flags & INP_INBOUND_POLICY)
362 			return;
363 		/* Another thread has already set cached policy */
364 		if (inp->inp_sp->sp_in != NULL)
365 			return;
366 		/*
367 		 * Do not cache INBOUND policy for listen socket,
368 		 * that is bound to INADDR_ANY/UNSPECIFIED address.
369 		 */
370 #ifdef INET
371 		if ((inp->inp_vflag & INP_IPV4) != 0 &&
372 		    inp->inp_faddr.s_addr == INADDR_ANY)
373 			return;
374 #endif
375 #ifdef INET6
376 		if ((inp->inp_vflag & INP_IPV6) != 0 &&
377 		    IN6_IS_ADDR_UNSPECIFIED(&inp->in6p_faddr))
378 			return;
379 #endif
380 	}
381 	downgrade = 0;
382 	if (!INP_WLOCKED(inp)) {
383 		if ((downgrade = INP_TRY_UPGRADE(inp)) == 0)
384 			return;
385 	}
386 	if (dir == IPSEC_DIR_OUTBOUND)
387 		inp->inp_sp->sp_out = sp;
388 	else
389 		inp->inp_sp->sp_in = sp;
390 	/*
391 	 * SP is already referenced by the lookup code.
392 	 * We take extra reference here to avoid race in the
393 	 * ipsec_getpcbpolicy() function - SP will not be freed in the
394 	 * time between we take SP pointer from the cache and key_addref()
395 	 * call.
396 	 */
397 	key_addref(sp);
398 	genid = key_getspgen();
399 	if (genid != inp->inp_sp->genid) {
400 		ipsec_invalidate_cache(inp, dir);
401 		inp->inp_sp->genid = genid;
402 	}
403 	KEYDBG(IPSEC_STAMP,
404 	    printf("%s: PCB(%p): cached %s SP(%p)\n",
405 	    __func__, inp, dir == IPSEC_DIR_OUTBOUND ? "OUTBOUND":
406 	    "INBOUND", sp));
407 	if (downgrade != 0)
408 		INP_DOWNGRADE(inp);
409 }
410 
411 static struct secpolicy *
ipsec_checkpolicy(struct secpolicy * sp,struct inpcb * inp,int * error)412 ipsec_checkpolicy(struct secpolicy *sp, struct inpcb *inp, int *error)
413 {
414 
415 	/* Save found OUTBOUND policy into PCB SP cache. */
416 	if (inp != NULL && inp->inp_sp != NULL && inp->inp_sp->sp_out == NULL)
417 		ipsec_cachepolicy(inp, sp, IPSEC_DIR_OUTBOUND);
418 
419 	switch (sp->policy) {
420 	default:
421 		printf("%s: invalid policy %u\n", __func__, sp->policy);
422 		/* FALLTHROUGH */
423 	case IPSEC_POLICY_DISCARD:
424 		*error = -EINVAL;	/* Packet is discarded by caller. */
425 		/* FALLTHROUGH */
426 	case IPSEC_POLICY_BYPASS:
427 	case IPSEC_POLICY_NONE:
428 		key_freesp(&sp);
429 		sp = NULL;		/* NB: force NULL result. */
430 		break;
431 	case IPSEC_POLICY_IPSEC:
432 		/* XXXAE: handle LARVAL SP */
433 		break;
434 	}
435 	KEYDBG(IPSEC_DUMP,
436 	    printf("%s: get SP(%p), error %d\n", __func__, sp, *error));
437 	return (sp);
438 }
439 
440 static struct secpolicy *
ipsec_getpcbpolicy(struct inpcb * inp,u_int dir)441 ipsec_getpcbpolicy(struct inpcb *inp, u_int dir)
442 {
443 	struct secpolicy *sp;
444 	int flags, downgrade;
445 
446 	if (inp == NULL || inp->inp_sp == NULL)
447 		return (NULL);
448 
449 	INP_LOCK_ASSERT(inp);
450 
451 	flags = inp->inp_sp->flags;
452 	if (dir == IPSEC_DIR_OUTBOUND) {
453 		sp = inp->inp_sp->sp_out;
454 		flags &= INP_OUTBOUND_POLICY;
455 	} else {
456 		sp = inp->inp_sp->sp_in;
457 		flags &= INP_INBOUND_POLICY;
458 	}
459 	/*
460 	 * Check flags. If we have PCB SP, just return it.
461 	 * Otherwise we need to check that cached SP entry isn't stale.
462 	 */
463 	if (flags == 0) {
464 		if (sp == NULL)
465 			return (NULL);
466 		if (inp->inp_sp->genid != key_getspgen()) {
467 			/* Invalidate the cache. */
468 			downgrade = 0;
469 			if (!INP_WLOCKED(inp)) {
470 				if ((downgrade = INP_TRY_UPGRADE(inp)) == 0)
471 					return (NULL);
472 			}
473 			ipsec_invalidate_cache(inp, IPSEC_DIR_OUTBOUND);
474 			ipsec_invalidate_cache(inp, IPSEC_DIR_INBOUND);
475 			if (downgrade != 0)
476 				INP_DOWNGRADE(inp);
477 			return (NULL);
478 		}
479 		KEYDBG(IPSEC_STAMP,
480 		    printf("%s: PCB(%p): cache hit SP(%p)\n",
481 		    __func__, inp, sp));
482 		/* Return referenced cached policy */
483 	}
484 	key_addref(sp);
485 	return (sp);
486 }
487 
488 #ifdef INET
489 static void
ipsec4_get_ulp(const struct mbuf * m,struct secpolicyindex * spidx,int needport)490 ipsec4_get_ulp(const struct mbuf *m, struct secpolicyindex *spidx,
491     int needport)
492 {
493 	uint8_t nxt;
494 	int off;
495 
496 	/* Sanity check. */
497 	IPSEC_ASSERT(m->m_pkthdr.len >= sizeof(struct ip),
498 	    ("packet too short"));
499 
500 	if (m->m_len >= sizeof (struct ip)) {
501 		const struct ip *ip = mtod(m, const struct ip *);
502 		if (ip->ip_off & htons(IP_MF | IP_OFFMASK))
503 			goto done;
504 		off = ip->ip_hl << 2;
505 		nxt = ip->ip_p;
506 	} else {
507 		struct ip ih;
508 
509 		m_copydata(m, 0, sizeof (struct ip), (caddr_t) &ih);
510 		if (ih.ip_off & htons(IP_MF | IP_OFFMASK))
511 			goto done;
512 		off = ih.ip_hl << 2;
513 		nxt = ih.ip_p;
514 	}
515 
516 	while (off < m->m_pkthdr.len) {
517 		struct ip6_ext ip6e;
518 		struct tcphdr th;
519 		struct udphdr uh;
520 
521 		switch (nxt) {
522 		case IPPROTO_TCP:
523 			spidx->ul_proto = nxt;
524 			if (!needport)
525 				goto done_proto;
526 			if (off + sizeof(struct tcphdr) > m->m_pkthdr.len)
527 				goto done;
528 			m_copydata(m, off, sizeof (th), (caddr_t) &th);
529 			spidx->src.sin.sin_port = th.th_sport;
530 			spidx->dst.sin.sin_port = th.th_dport;
531 			return;
532 		case IPPROTO_UDP:
533 			spidx->ul_proto = nxt;
534 			if (!needport)
535 				goto done_proto;
536 			if (off + sizeof(struct udphdr) > m->m_pkthdr.len)
537 				goto done;
538 			m_copydata(m, off, sizeof (uh), (caddr_t) &uh);
539 			spidx->src.sin.sin_port = uh.uh_sport;
540 			spidx->dst.sin.sin_port = uh.uh_dport;
541 			return;
542 		case IPPROTO_AH:
543 			if (off + sizeof(ip6e) > m->m_pkthdr.len)
544 				goto done;
545 			/* XXX Sigh, this works but is totally bogus. */
546 			m_copydata(m, off, sizeof(ip6e), (caddr_t) &ip6e);
547 			off += (ip6e.ip6e_len + 2) << 2;
548 			nxt = ip6e.ip6e_nxt;
549 			break;
550 		case IPPROTO_ICMP:
551 		default:
552 			/* XXX Intermediate headers??? */
553 			spidx->ul_proto = nxt;
554 			goto done_proto;
555 		}
556 	}
557 done:
558 	spidx->ul_proto = IPSEC_ULPROTO_ANY;
559 done_proto:
560 	spidx->src.sin.sin_port = IPSEC_PORT_ANY;
561 	spidx->dst.sin.sin_port = IPSEC_PORT_ANY;
562 	KEYDBG(IPSEC_DUMP,
563 	    printf("%s: ", __func__); kdebug_secpolicyindex(spidx, NULL));
564 }
565 
566 static void
ipsec4_setspidx_ipaddr(const struct mbuf * m,struct secpolicyindex * spidx)567 ipsec4_setspidx_ipaddr(const struct mbuf *m, struct secpolicyindex *spidx)
568 {
569 
570 	ipsec4_setsockaddrs(m, &spidx->src, &spidx->dst);
571 	spidx->prefs = sizeof(struct in_addr) << 3;
572 	spidx->prefd = sizeof(struct in_addr) << 3;
573 }
574 
575 static struct secpolicy *
ipsec4_getpolicy(const struct mbuf * m,struct inpcb * inp,u_int dir,int needport)576 ipsec4_getpolicy(const struct mbuf *m, struct inpcb *inp, u_int dir,
577     int needport)
578 {
579 	struct secpolicyindex spidx;
580 	struct secpolicy *sp;
581 
582 	sp = ipsec_getpcbpolicy(inp, dir);
583 	if (sp == NULL && key_havesp(dir)) {
584 		/* Make an index to look for a policy. */
585 		ipsec4_setspidx_ipaddr(m, &spidx);
586 		ipsec4_get_ulp(m, &spidx, needport);
587 		spidx.dir = dir;
588 		sp = key_allocsp(&spidx, dir);
589 	}
590 	if (sp == NULL)		/* No SP found, use system default. */
591 		sp = key_allocsp_default();
592 	return (sp);
593 }
594 
595 /*
596  * Check security policy for *OUTBOUND* IPv4 packet.
597  */
598 struct secpolicy *
ipsec4_checkpolicy(const struct mbuf * m,struct inpcb * inp,int * error,int needport)599 ipsec4_checkpolicy(const struct mbuf *m, struct inpcb *inp, int *error,
600     int needport)
601 {
602 	struct secpolicy *sp;
603 
604 	*error = 0;
605 	sp = ipsec4_getpolicy(m, inp, IPSEC_DIR_OUTBOUND, needport);
606 	if (sp != NULL)
607 		sp = ipsec_checkpolicy(sp, inp, error);
608 	if (sp == NULL) {
609 		switch (*error) {
610 		case 0: /* No IPsec required: BYPASS or NONE */
611 			break;
612 		case -EINVAL:
613 			IPSECSTAT_INC(ips_out_polvio);
614 			break;
615 		default:
616 			IPSECSTAT_INC(ips_out_inval);
617 		}
618 	}
619 	KEYDBG(IPSEC_STAMP,
620 	    printf("%s: using SP(%p), error %d\n", __func__, sp, *error));
621 	if (sp != NULL)
622 		KEYDBG(IPSEC_DATA, kdebug_secpolicy(sp));
623 	return (sp);
624 }
625 
626 /*
627  * Check IPv4 packet against *INBOUND* security policy.
628  * This function is called from tcp_input(), udp_input(),
629  * rip_input() and sctp_input().
630  */
631 int
ipsec4_in_reject(const struct mbuf * m,struct inpcb * inp)632 ipsec4_in_reject(const struct mbuf *m, struct inpcb *inp)
633 {
634 	struct secpolicy *sp;
635 	int result;
636 
637 	sp = ipsec4_getpolicy(m, inp, IPSEC_DIR_INBOUND, 0);
638 	result = ipsec_in_reject(sp, inp, m);
639 	key_freesp(&sp);
640 	if (result != 0)
641 		IPSECSTAT_INC(ips_in_polvio);
642 	return (result);
643 }
644 
645 /*
646  * IPSEC_CAP() method implementation for IPv4.
647  */
648 int
ipsec4_capability(struct mbuf * m,u_int cap)649 ipsec4_capability(struct mbuf *m, u_int cap)
650 {
651 
652 	switch (cap) {
653 	case IPSEC_CAP_BYPASS_FILTER:
654 		/*
655 		 * Bypass packet filtering for packets previously handled
656 		 * by IPsec.
657 		 */
658 		if (!V_ip4_filtertunnel &&
659 		    m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL) != NULL)
660 			return (1);
661 		return (0);
662 	case IPSEC_CAP_OPERABLE:
663 		/* Do we have active security policies? */
664 		if (key_havesp(IPSEC_DIR_INBOUND) != 0 ||
665 		    key_havesp(IPSEC_DIR_OUTBOUND) != 0)
666 			return (1);
667 		return (0);
668 	};
669 	return (EOPNOTSUPP);
670 }
671 
672 #endif /* INET */
673 
674 #ifdef INET6
675 static void
ipsec6_get_ulp(const struct mbuf * m,struct secpolicyindex * spidx,int needport)676 ipsec6_get_ulp(const struct mbuf *m, struct secpolicyindex *spidx,
677     int needport)
678 {
679 	struct tcphdr th;
680 	struct udphdr uh;
681 	struct icmp6_hdr ih;
682 	int off, nxt;
683 
684 	IPSEC_ASSERT(m->m_pkthdr.len >= sizeof(struct ip6_hdr),
685 	    ("packet too short"));
686 
687 	/* Set default. */
688 	spidx->ul_proto = IPSEC_ULPROTO_ANY;
689 	spidx->src.sin6.sin6_port = IPSEC_PORT_ANY;
690 	spidx->dst.sin6.sin6_port = IPSEC_PORT_ANY;
691 
692 	nxt = -1;
693 	off = ip6_lasthdr(m, 0, IPPROTO_IPV6, &nxt);
694 	if (off < 0 || m->m_pkthdr.len < off)
695 		return;
696 
697 	switch (nxt) {
698 	case IPPROTO_TCP:
699 		spidx->ul_proto = nxt;
700 		if (!needport)
701 			break;
702 		if (off + sizeof(struct tcphdr) > m->m_pkthdr.len)
703 			break;
704 		m_copydata(m, off, sizeof(th), (caddr_t)&th);
705 		spidx->src.sin6.sin6_port = th.th_sport;
706 		spidx->dst.sin6.sin6_port = th.th_dport;
707 		break;
708 	case IPPROTO_UDP:
709 		spidx->ul_proto = nxt;
710 		if (!needport)
711 			break;
712 		if (off + sizeof(struct udphdr) > m->m_pkthdr.len)
713 			break;
714 		m_copydata(m, off, sizeof(uh), (caddr_t)&uh);
715 		spidx->src.sin6.sin6_port = uh.uh_sport;
716 		spidx->dst.sin6.sin6_port = uh.uh_dport;
717 		break;
718 	case IPPROTO_ICMPV6:
719 		spidx->ul_proto = nxt;
720 		if (off + sizeof(struct icmp6_hdr) > m->m_pkthdr.len)
721 			break;
722 		m_copydata(m, off, sizeof(ih), (caddr_t)&ih);
723 		spidx->src.sin6.sin6_port = htons((uint16_t)ih.icmp6_type);
724 		spidx->dst.sin6.sin6_port = htons((uint16_t)ih.icmp6_code);
725 		break;
726 	default:
727 		/* XXX Intermediate headers??? */
728 		spidx->ul_proto = nxt;
729 		break;
730 	}
731 	KEYDBG(IPSEC_DUMP,
732 	    printf("%s: ", __func__); kdebug_secpolicyindex(spidx, NULL));
733 }
734 
735 static void
ipsec6_setspidx_ipaddr(const struct mbuf * m,struct secpolicyindex * spidx)736 ipsec6_setspidx_ipaddr(const struct mbuf *m, struct secpolicyindex *spidx)
737 {
738 
739 	ipsec6_setsockaddrs(m, &spidx->src, &spidx->dst);
740 	spidx->prefs = sizeof(struct in6_addr) << 3;
741 	spidx->prefd = sizeof(struct in6_addr) << 3;
742 }
743 
744 static struct secpolicy *
ipsec6_getpolicy(const struct mbuf * m,struct inpcb * inp,u_int dir,int needport)745 ipsec6_getpolicy(const struct mbuf *m, struct inpcb *inp, u_int dir,
746     int needport)
747 {
748 	struct secpolicyindex spidx;
749 	struct secpolicy *sp;
750 
751 	sp = ipsec_getpcbpolicy(inp, dir);
752 	if (sp == NULL && key_havesp(dir)) {
753 		/* Make an index to look for a policy. */
754 		ipsec6_setspidx_ipaddr(m, &spidx);
755 		ipsec6_get_ulp(m, &spidx, needport);
756 		spidx.dir = dir;
757 		sp = key_allocsp(&spidx, dir);
758 	}
759 	if (sp == NULL)		/* No SP found, use system default. */
760 		sp = key_allocsp_default();
761 	return (sp);
762 }
763 
764 /*
765  * Check security policy for *OUTBOUND* IPv6 packet.
766  */
767 struct secpolicy *
ipsec6_checkpolicy(const struct mbuf * m,struct inpcb * inp,int * error,int needport)768 ipsec6_checkpolicy(const struct mbuf *m, struct inpcb *inp, int *error,
769     int needport)
770 {
771 	struct secpolicy *sp;
772 
773 	*error = 0;
774 	sp = ipsec6_getpolicy(m, inp, IPSEC_DIR_OUTBOUND, needport);
775 	if (sp != NULL)
776 		sp = ipsec_checkpolicy(sp, inp, error);
777 	if (sp == NULL) {
778 		switch (*error) {
779 		case 0: /* No IPsec required: BYPASS or NONE */
780 			break;
781 		case -EINVAL:
782 			IPSEC6STAT_INC(ips_out_polvio);
783 			break;
784 		default:
785 			IPSEC6STAT_INC(ips_out_inval);
786 		}
787 	}
788 	KEYDBG(IPSEC_STAMP,
789 	    printf("%s: using SP(%p), error %d\n", __func__, sp, *error));
790 	if (sp != NULL)
791 		KEYDBG(IPSEC_DATA, kdebug_secpolicy(sp));
792 	return (sp);
793 }
794 
795 /*
796  * Check IPv6 packet against inbound security policy.
797  * This function is called from tcp6_input(), udp6_input(),
798  * rip6_input() and sctp_input().
799  */
800 int
ipsec6_in_reject(const struct mbuf * m,struct inpcb * inp)801 ipsec6_in_reject(const struct mbuf *m, struct inpcb *inp)
802 {
803 	struct secpolicy *sp;
804 	int result;
805 
806 	sp = ipsec6_getpolicy(m, inp, IPSEC_DIR_INBOUND, 0);
807 	result = ipsec_in_reject(sp, inp, m);
808 	key_freesp(&sp);
809 	if (result)
810 		IPSEC6STAT_INC(ips_in_polvio);
811 	return (result);
812 }
813 
814 /*
815  * IPSEC_CAP() method implementation for IPv6.
816  */
817 int
ipsec6_capability(struct mbuf * m,u_int cap)818 ipsec6_capability(struct mbuf *m, u_int cap)
819 {
820 
821 	switch (cap) {
822 	case IPSEC_CAP_BYPASS_FILTER:
823 		/*
824 		 * Bypass packet filtering for packets previously handled
825 		 * by IPsec.
826 		 */
827 		if (!V_ip6_filtertunnel &&
828 		    m_tag_find(m, PACKET_TAG_IPSEC_IN_DONE, NULL) != NULL)
829 			return (1);
830 		return (0);
831 	case IPSEC_CAP_OPERABLE:
832 		/* Do we have active security policies? */
833 		if (key_havesp(IPSEC_DIR_INBOUND) != 0 ||
834 		    key_havesp(IPSEC_DIR_OUTBOUND) != 0)
835 			return (1);
836 		return (0);
837 	};
838 	return (EOPNOTSUPP);
839 }
840 #endif /* INET6 */
841 
842 int
ipsec_run_hhooks(struct ipsec_ctx_data * ctx,int type)843 ipsec_run_hhooks(struct ipsec_ctx_data *ctx, int type)
844 {
845 	int idx;
846 
847 	switch (ctx->af) {
848 #ifdef INET
849 	case AF_INET:
850 		idx = HHOOK_IPSEC_INET;
851 		break;
852 #endif
853 #ifdef INET6
854 	case AF_INET6:
855 		idx = HHOOK_IPSEC_INET6;
856 		break;
857 #endif
858 	default:
859 		return (EPFNOSUPPORT);
860 	}
861 	if (type == HHOOK_TYPE_IPSEC_IN)
862 		HHOOKS_RUN_IF(V_ipsec_hhh_in[idx], ctx, NULL);
863 	else
864 		HHOOKS_RUN_IF(V_ipsec_hhh_out[idx], ctx, NULL);
865 	if (*ctx->mp == NULL)
866 		return (EACCES);
867 	return (0);
868 }
869 
870 /*
871  * Return current level.
872  * Either IPSEC_LEVEL_USE or IPSEC_LEVEL_REQUIRE are always returned.
873  */
874 u_int
ipsec_get_reqlevel(struct secpolicy * sp,u_int idx)875 ipsec_get_reqlevel(struct secpolicy *sp, u_int idx)
876 {
877 	struct ipsecrequest *isr;
878 	u_int esp_trans_deflev, esp_net_deflev;
879 	u_int ah_trans_deflev, ah_net_deflev;
880 	u_int level = 0;
881 
882 	IPSEC_ASSERT(idx < sp->tcount, ("Wrong IPsec request index %d", idx));
883 /* XXX Note that we have ipseclog() expanded here - code sync issue. */
884 #define IPSEC_CHECK_DEFAULT(lev) \
885 	(((lev) != IPSEC_LEVEL_USE && (lev) != IPSEC_LEVEL_REQUIRE &&	\
886 	  (lev) != IPSEC_LEVEL_UNIQUE)					\
887 		? (V_ipsec_debug  ?					\
888 		log(LOG_INFO, "fixed system default level " #lev ":%d->%d\n",\
889 		(lev), IPSEC_LEVEL_REQUIRE) : 0),			\
890 		(lev) = IPSEC_LEVEL_REQUIRE, (lev) : (lev))
891 
892 	/*
893 	 * IPsec VTI uses unique security policy with fake spidx filled
894 	 * with zeroes. Just return IPSEC_LEVEL_REQUIRE instead of doing
895 	 * full level lookup for such policies.
896 	 */
897 	if (sp->state == IPSEC_SPSTATE_IFNET) {
898 		IPSEC_ASSERT(sp->req[idx]->level == IPSEC_LEVEL_UNIQUE,
899 		    ("Wrong IPsec request level %d", sp->req[idx]->level));
900 		return (IPSEC_LEVEL_REQUIRE);
901 	}
902 
903 	/* Set default level. */
904 	switch (sp->spidx.src.sa.sa_family) {
905 #ifdef INET
906 	case AF_INET:
907 		esp_trans_deflev = IPSEC_CHECK_DEFAULT(V_ip4_esp_trans_deflev);
908 		esp_net_deflev = IPSEC_CHECK_DEFAULT(V_ip4_esp_net_deflev);
909 		ah_trans_deflev = IPSEC_CHECK_DEFAULT(V_ip4_ah_trans_deflev);
910 		ah_net_deflev = IPSEC_CHECK_DEFAULT(V_ip4_ah_net_deflev);
911 		break;
912 #endif
913 #ifdef INET6
914 	case AF_INET6:
915 		esp_trans_deflev = IPSEC_CHECK_DEFAULT(V_ip6_esp_trans_deflev);
916 		esp_net_deflev = IPSEC_CHECK_DEFAULT(V_ip6_esp_net_deflev);
917 		ah_trans_deflev = IPSEC_CHECK_DEFAULT(V_ip6_ah_trans_deflev);
918 		ah_net_deflev = IPSEC_CHECK_DEFAULT(V_ip6_ah_net_deflev);
919 		break;
920 #endif /* INET6 */
921 	default:
922 		panic("%s: unknown af %u",
923 			__func__, sp->spidx.src.sa.sa_family);
924 	}
925 
926 #undef IPSEC_CHECK_DEFAULT
927 
928 	isr = sp->req[idx];
929 	/* Set level. */
930 	switch (isr->level) {
931 	case IPSEC_LEVEL_DEFAULT:
932 		switch (isr->saidx.proto) {
933 		case IPPROTO_ESP:
934 			if (isr->saidx.mode == IPSEC_MODE_TUNNEL)
935 				level = esp_net_deflev;
936 			else
937 				level = esp_trans_deflev;
938 			break;
939 		case IPPROTO_AH:
940 			if (isr->saidx.mode == IPSEC_MODE_TUNNEL)
941 				level = ah_net_deflev;
942 			else
943 				level = ah_trans_deflev;
944 			break;
945 		case IPPROTO_IPCOMP:
946 			/*
947 			 * We don't really care, as IPcomp document says that
948 			 * we shouldn't compress small packets.
949 			 */
950 			level = IPSEC_LEVEL_USE;
951 			break;
952 		default:
953 			panic("%s: Illegal protocol defined %u\n", __func__,
954 				isr->saidx.proto);
955 		}
956 		break;
957 
958 	case IPSEC_LEVEL_USE:
959 	case IPSEC_LEVEL_REQUIRE:
960 		level = isr->level;
961 		break;
962 	case IPSEC_LEVEL_UNIQUE:
963 		level = IPSEC_LEVEL_REQUIRE;
964 		break;
965 
966 	default:
967 		panic("%s: Illegal IPsec level %u\n", __func__, isr->level);
968 	}
969 
970 	return (level);
971 }
972 
973 static int
ipsec_check_history(const struct mbuf * m,struct secpolicy * sp,u_int idx)974 ipsec_check_history(const struct mbuf *m, struct secpolicy *sp, u_int idx)
975 {
976 	struct xform_history *xh;
977 	struct m_tag *mtag;
978 
979 	mtag = NULL;
980 	while ((mtag = m_tag_find(__DECONST(struct mbuf *, m),
981 	    PACKET_TAG_IPSEC_IN_DONE, mtag)) != NULL) {
982 		xh = (struct xform_history *)(mtag + 1);
983 		KEYDBG(IPSEC_DATA,
984 		    char buf[IPSEC_ADDRSTRLEN];
985 		    printf("%s: mode %s proto %u dst %s\n", __func__,
986 			kdebug_secasindex_mode(xh->mode), xh->proto,
987 			ipsec_address(&xh->dst, buf, sizeof(buf))));
988 		if (xh->proto != sp->req[idx]->saidx.proto)
989 			continue;
990 		/* If SA had IPSEC_MODE_ANY, consider this as match. */
991 		if (xh->mode != sp->req[idx]->saidx.mode &&
992 		    xh->mode != IPSEC_MODE_ANY)
993 			continue;
994 		/*
995 		 * For transport mode IPsec request doesn't contain
996 		 * addresses. We need to use address from spidx.
997 		 */
998 		if (sp->req[idx]->saidx.mode == IPSEC_MODE_TRANSPORT) {
999 			if (key_sockaddrcmp_withmask(&xh->dst.sa,
1000 			    &sp->spidx.dst.sa, sp->spidx.prefd) != 0)
1001 				continue;
1002 		} else {
1003 			if (key_sockaddrcmp(&xh->dst.sa,
1004 			    &sp->req[idx]->saidx.dst.sa, 0) != 0)
1005 				continue;
1006 		}
1007 		return (0); /* matched */
1008 	}
1009 	return (1);
1010 }
1011 
1012 /*
1013  * Check security policy requirements against the actual
1014  * packet contents.  Return one if the packet should be
1015  * rejected as "invalid"; otherwise return zero to have the
1016  * packet treated as "valid".
1017  *
1018  * OUT:
1019  *	0: valid
1020  *	1: invalid
1021  */
1022 static int
ipsec_in_reject(struct secpolicy * sp,struct inpcb * inp,const struct mbuf * m)1023 ipsec_in_reject(struct secpolicy *sp, struct inpcb *inp, const struct mbuf *m)
1024 {
1025 	int i;
1026 
1027 	KEYDBG(IPSEC_STAMP,
1028 	    printf("%s: PCB(%p): using SP(%p)\n", __func__, inp, sp));
1029 	KEYDBG(IPSEC_DATA, kdebug_secpolicy(sp));
1030 
1031 	if (inp != NULL && inp->inp_sp != NULL && inp->inp_sp->sp_in == NULL)
1032 		ipsec_cachepolicy(inp, sp, IPSEC_DIR_INBOUND);
1033 
1034 	/* Check policy. */
1035 	switch (sp->policy) {
1036 	case IPSEC_POLICY_DISCARD:
1037 		return (1);
1038 	case IPSEC_POLICY_BYPASS:
1039 	case IPSEC_POLICY_NONE:
1040 		return (0);
1041 	}
1042 
1043 	IPSEC_ASSERT(sp->policy == IPSEC_POLICY_IPSEC,
1044 		("invalid policy %u", sp->policy));
1045 
1046 	/*
1047 	 * ipsec[46]_common_input_cb after each transform adds
1048 	 * PACKET_TAG_IPSEC_IN_DONE mbuf tag. It contains SPI, proto, mode
1049 	 * and destination address from saidx. We can compare info from
1050 	 * these tags with requirements in SP.
1051 	 */
1052 	for (i = 0; i < sp->tcount; i++) {
1053 		/*
1054 		 * Do not check IPcomp, since IPcomp document
1055 		 * says that we shouldn't compress small packets.
1056 		 * IPComp policy should always be treated as being
1057 		 * in "use" level.
1058 		 */
1059 		if (sp->req[i]->saidx.proto == IPPROTO_IPCOMP ||
1060 		    ipsec_get_reqlevel(sp, i) != IPSEC_LEVEL_REQUIRE)
1061 			continue;
1062 		if (V_check_policy_history != 0 &&
1063 		    ipsec_check_history(m, sp, i) != 0)
1064 			return (1);
1065 		else switch (sp->req[i]->saidx.proto) {
1066 		case IPPROTO_ESP:
1067 			if ((m->m_flags & M_DECRYPTED) == 0) {
1068 				KEYDBG(IPSEC_DUMP,
1069 				    printf("%s: ESP m_flags:%x\n", __func__,
1070 					    m->m_flags));
1071 				return (1);
1072 			}
1073 			break;
1074 		case IPPROTO_AH:
1075 			if ((m->m_flags & M_AUTHIPHDR) == 0) {
1076 				KEYDBG(IPSEC_DUMP,
1077 				    printf("%s: AH m_flags:%x\n", __func__,
1078 					    m->m_flags));
1079 				return (1);
1080 			}
1081 			break;
1082 		}
1083 	}
1084 	return (0);		/* Valid. */
1085 }
1086 
1087 /*
1088  * Compute the byte size to be occupied by IPsec header.
1089  * In case it is tunnelled, it includes the size of outer IP header.
1090  */
1091 static size_t
ipsec_hdrsiz_internal(struct secpolicy * sp)1092 ipsec_hdrsiz_internal(struct secpolicy *sp)
1093 {
1094 	size_t size;
1095 	int i;
1096 
1097 	KEYDBG(IPSEC_STAMP, printf("%s: using SP(%p)\n", __func__, sp));
1098 	KEYDBG(IPSEC_DATA, kdebug_secpolicy(sp));
1099 
1100 	switch (sp->policy) {
1101 	case IPSEC_POLICY_DISCARD:
1102 	case IPSEC_POLICY_BYPASS:
1103 	case IPSEC_POLICY_NONE:
1104 		return (0);
1105 	}
1106 
1107 	IPSEC_ASSERT(sp->policy == IPSEC_POLICY_IPSEC,
1108 		("invalid policy %u", sp->policy));
1109 
1110 	/*
1111 	 * XXX: for each transform we need to lookup suitable SA
1112 	 * and use info from SA to calculate headers size.
1113 	 * XXX: for NAT-T we need to cosider UDP header size.
1114 	 */
1115 	size = 0;
1116 	for (i = 0; i < sp->tcount; i++) {
1117 		switch (sp->req[i]->saidx.proto) {
1118 		case IPPROTO_ESP:
1119 			size += esp_hdrsiz(NULL);
1120 			break;
1121 		case IPPROTO_AH:
1122 			size += ah_hdrsiz(NULL);
1123 			break;
1124 		case IPPROTO_IPCOMP:
1125 			size += sizeof(struct ipcomp);
1126 			break;
1127 		}
1128 
1129 		if (sp->req[i]->saidx.mode == IPSEC_MODE_TUNNEL) {
1130 			switch (sp->req[i]->saidx.dst.sa.sa_family) {
1131 #ifdef INET
1132 			case AF_INET:
1133 				size += sizeof(struct ip);
1134 				break;
1135 #endif
1136 #ifdef INET6
1137 			case AF_INET6:
1138 				size += sizeof(struct ip6_hdr);
1139 				break;
1140 #endif
1141 			default:
1142 				ipseclog((LOG_ERR, "%s: unknown AF %d in "
1143 				    "IPsec tunnel SA\n", __func__,
1144 				    sp->req[i]->saidx.dst.sa.sa_family));
1145 				break;
1146 			}
1147 		}
1148 	}
1149 	return (size);
1150 }
1151 
1152 /*
1153  * Compute ESP/AH header size for protocols with PCB, including
1154  * outer IP header. Currently only tcp_output() uses it.
1155  */
1156 size_t
ipsec_hdrsiz_inpcb(struct inpcb * inp)1157 ipsec_hdrsiz_inpcb(struct inpcb *inp)
1158 {
1159 	struct secpolicyindex spidx;
1160 	struct secpolicy *sp;
1161 	size_t sz;
1162 
1163 	sp = ipsec_getpcbpolicy(inp, IPSEC_DIR_OUTBOUND);
1164 	if (sp == NULL && key_havesp(IPSEC_DIR_OUTBOUND)) {
1165 		ipsec_setspidx_inpcb(inp, &spidx, IPSEC_DIR_OUTBOUND);
1166 		sp = key_allocsp(&spidx, IPSEC_DIR_OUTBOUND);
1167 	}
1168 	if (sp == NULL)
1169 		sp = key_allocsp_default();
1170 	sz = ipsec_hdrsiz_internal(sp);
1171 	key_freesp(&sp);
1172 	return (sz);
1173 }
1174 
1175 
1176 #define IPSEC_BITMAP_INDEX_MASK(w)	(w - 1)
1177 #define IPSEC_REDUNDANT_BIT_SHIFTS	5
1178 #define IPSEC_REDUNDANT_BITS		(1 << IPSEC_REDUNDANT_BIT_SHIFTS)
1179 #define IPSEC_BITMAP_LOC_MASK		(IPSEC_REDUNDANT_BITS - 1)
1180 
1181 /*
1182  * Functions below are responsible for checking and updating bitmap.
1183  * These are used to separate ipsec_chkreplay() and ipsec_updatereplay()
1184  * from window implementation
1185  *
1186  * Based on RFC 6479. Blocks are 32 bits unsigned integers
1187  */
1188 
1189 static inline int
check_window(const struct secreplay * replay,uint64_t seq)1190 check_window(const struct secreplay *replay, uint64_t seq)
1191 {
1192 	int index, bit_location;
1193 
1194 	SECREPLAY_ASSERT(replay);
1195 
1196 	bit_location = seq & IPSEC_BITMAP_LOC_MASK;
1197 	index = (seq >> IPSEC_REDUNDANT_BIT_SHIFTS)
1198 		& IPSEC_BITMAP_INDEX_MASK(replay->bitmap_size);
1199 
1200 	/* This packet already seen? */
1201 	return ((replay->bitmap)[index] & (1 << bit_location));
1202 }
1203 
1204 static inline void
advance_window(const struct secreplay * replay,uint64_t seq)1205 advance_window(const struct secreplay *replay, uint64_t seq)
1206 {
1207 	int i;
1208 	uint64_t index, index_cur, diff;
1209 
1210 	SECREPLAY_ASSERT(replay);
1211 
1212 	index_cur = replay->last >> IPSEC_REDUNDANT_BIT_SHIFTS;
1213 	index = seq >> IPSEC_REDUNDANT_BIT_SHIFTS;
1214 	diff = index - index_cur;
1215 
1216 	if (diff > replay->bitmap_size) {
1217 		/* something unusual in this case */
1218 		diff = replay->bitmap_size;
1219 	}
1220 
1221 	for (i = 0; i < diff; i++) {
1222 		replay->bitmap[(i + index_cur + 1)
1223 		& IPSEC_BITMAP_INDEX_MASK(replay->bitmap_size)] = 0;
1224 	}
1225 }
1226 
1227 static inline void
set_window(const struct secreplay * replay,uint64_t seq)1228 set_window(const struct secreplay *replay, uint64_t seq)
1229 {
1230 	int index, bit_location;
1231 
1232 	SECREPLAY_ASSERT(replay);
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 	replay->bitmap[index] |= (1 << bit_location);
1239 }
1240 
1241 /*
1242  * Check the variable replay window.
1243  * ipsec_chkreplay() performs replay check before ICV verification.
1244  * ipsec_updatereplay() updates replay bitmap.  This must be called after
1245  * ICV verification (it also performs replay check, which is usually done
1246  * beforehand).
1247  * 0 (zero) is returned if packet disallowed, 1 if packet permitted.
1248  *
1249  * Based on RFC 4303
1250  */
1251 
1252 int
ipsec_chkreplay(uint32_t seq,uint32_t * seqhigh,struct secasvar * sav)1253 ipsec_chkreplay(uint32_t seq, uint32_t *seqhigh, struct secasvar *sav)
1254 {
1255 	char buf[128];
1256 	struct secreplay *replay;
1257 	uint32_t window;
1258 	uint32_t tl, th, bl;
1259 	uint32_t seqh;
1260 
1261 	IPSEC_ASSERT(sav != NULL, ("Null SA"));
1262 	IPSEC_ASSERT(sav->replay != NULL, ("Null replay state"));
1263 
1264 	replay = sav->replay;
1265 
1266 	/* No need to check replay if disabled. */
1267 	if (replay->wsize == 0) {
1268 		return (1);
1269 	}
1270 
1271 	SECREPLAY_LOCK(replay);
1272 
1273 	/* Zero sequence number is not allowed. */
1274 	if (seq == 0 && replay->last == 0) {
1275 		SECREPLAY_UNLOCK(replay);
1276 		return (0);
1277 	}
1278 
1279 	window = replay->wsize << 3;		/* Size of window */
1280 	tl = (uint32_t)replay->last;		/* Top of window, lower part */
1281 	th = (uint32_t)(replay->last >> 32);	/* Top of window, high part */
1282 	bl = tl - window + 1;			/* Bottom of window, lower part */
1283 
1284 	/*
1285 	 * We keep the high part intact when:
1286 	 * 1) the seq is within [bl, 0xffffffff] and the whole window is
1287 	 *    within one subspace;
1288 	 * 2) the seq is within [0, bl) and window spans two subspaces.
1289 	 */
1290 	if ((tl >= window - 1 && seq >= bl) ||
1291 	    (tl < window - 1 && seq < bl)) {
1292 		*seqhigh = th;
1293 		if (seq <= tl) {
1294 			/* Sequence number inside window - check against replay */
1295 			if (check_window(replay, seq)) {
1296 				SECREPLAY_UNLOCK(replay);
1297 				return (0);
1298 			}
1299 		}
1300 
1301 		SECREPLAY_UNLOCK(replay);
1302 		/* Sequence number above top of window or not found in bitmap */
1303 		return (1);
1304 	}
1305 
1306 	/*
1307 	 * If ESN is not enabled and packet with highest sequence number
1308 	 * was received we should report overflow
1309 	 */
1310 	if (tl == 0xffffffff && !(sav->flags & SADB_X_SAFLAGS_ESN)) {
1311 		/* Set overflow flag. */
1312 		replay->overflow++;
1313 
1314 		if ((sav->flags & SADB_X_EXT_CYCSEQ) == 0) {
1315 			if (sav->sah->saidx.proto == IPPROTO_ESP)
1316 				ESPSTAT_INC(esps_wrap);
1317 			else if (sav->sah->saidx.proto == IPPROTO_AH)
1318 				AHSTAT_INC(ahs_wrap);
1319 			SECREPLAY_UNLOCK(replay);
1320 			return (0);
1321 		}
1322 
1323 		ipseclog((LOG_WARNING, "%s: replay counter made %d cycle. %s\n",
1324 		    __func__, replay->overflow,
1325 		    ipsec_sa2str(sav, buf, sizeof(buf))));
1326 	}
1327 
1328 	/*
1329 	 * Seq is within [bl, 0xffffffff] and bl is within
1330 	 * [0xffffffff-window, 0xffffffff].  This means we got a seq
1331 	 * which is within our replay window, but in the previous
1332 	 * subspace.
1333 	 */
1334 	if (tl < window - 1 && seq >= bl) {
1335 		if (th == 0)
1336 			return (0);
1337 		*seqhigh = th - 1;
1338 		seqh = th - 1;
1339 		if (check_window(replay, seq)) {
1340 			SECREPLAY_UNLOCK(replay);
1341 			return (0);
1342 		}
1343 		SECREPLAY_UNLOCK(replay);
1344 		return (1);
1345 	}
1346 
1347 	/*
1348 	 * Seq is within [0, bl) but the whole window is within one subspace.
1349 	 * This means that seq has wrapped and is in next subspace
1350 	 */
1351 	*seqhigh = th + 1;
1352 	seqh = th + 1;
1353 
1354 	/* Don't let high part wrap. */
1355 	if (seqh == 0) {
1356 		/* Set overflow flag. */
1357 		replay->overflow++;
1358 
1359 		if ((sav->flags & SADB_X_EXT_CYCSEQ) == 0) {
1360 			if (sav->sah->saidx.proto == IPPROTO_ESP)
1361 				ESPSTAT_INC(esps_wrap);
1362 			else if (sav->sah->saidx.proto == IPPROTO_AH)
1363 				AHSTAT_INC(ahs_wrap);
1364 			SECREPLAY_UNLOCK(replay);
1365 			return (0);
1366 		}
1367 
1368 		ipseclog((LOG_WARNING, "%s: replay counter made %d cycle. %s\n",
1369 		    __func__, replay->overflow,
1370 		    ipsec_sa2str(sav, buf, sizeof(buf))));
1371 	}
1372 
1373 	SECREPLAY_UNLOCK(replay);
1374 	return (1);
1375 }
1376 
1377 /*
1378  * Check replay counter whether to update or not.
1379  * OUT:	0:	OK
1380  *	1:	NG
1381  */
1382 int
ipsec_updatereplay(uint32_t seq,struct secasvar * sav)1383 ipsec_updatereplay(uint32_t seq, struct secasvar *sav)
1384 {
1385 	struct secreplay *replay;
1386 	uint32_t window;
1387 	uint32_t tl, th, bl;
1388 	uint32_t seqh;
1389 
1390 	IPSEC_ASSERT(sav != NULL, ("Null SA"));
1391 	IPSEC_ASSERT(sav->replay != NULL, ("Null replay state"));
1392 
1393 	replay = sav->replay;
1394 
1395 	/* No need to check replay if disabled. */
1396 	if (replay->wsize == 0)
1397 		return (0);
1398 
1399 	SECREPLAY_LOCK(replay);
1400 
1401 	/* Zero sequence number is not allowed. */
1402 	if (seq == 0 && replay->last == 0) {
1403 		SECREPLAY_UNLOCK(replay);
1404 		return (1);
1405 	}
1406 
1407 	window = replay->wsize << 3;		/* Size of window */
1408 	tl = (uint32_t)replay->last;		/* Top of window, lower part */
1409 	th = (uint32_t)(replay->last >> 32);	/* Top of window, high part */
1410 	bl = tl - window + 1;			/* Bottom of window, lower part */
1411 
1412 	/*
1413 	 * We keep the high part intact when:
1414 	 * 1) the seq is within [bl, 0xffffffff] and the whole window is
1415 	 *    within one subspace;
1416 	 * 2) the seq is within [0, bl) and window spans two subspaces.
1417 	 */
1418 	if ((tl >= window - 1 && seq >= bl) ||
1419 	    (tl < window - 1 && seq < bl)) {
1420 		seqh = th;
1421 		if (seq <= tl) {
1422 			/* Sequence number inside window - check against replay */
1423 			if (check_window(replay, seq)) {
1424 				SECREPLAY_UNLOCK(replay);
1425 				return (1);
1426 			}
1427 			set_window(replay, seq);
1428 		} else {
1429 			advance_window(replay, ((uint64_t)seqh << 32) | seq);
1430 			set_window(replay, seq);
1431 			replay->last = ((uint64_t)seqh << 32) | seq;
1432 		}
1433 
1434 		/* Sequence number above top of window or not found in bitmap */
1435 		replay->count++;
1436 		SECREPLAY_UNLOCK(replay);
1437 		return (0);
1438 	}
1439 
1440 	if (!(sav->flags & SADB_X_SAFLAGS_ESN)) {
1441 		SECREPLAY_UNLOCK(replay);
1442 		return (1);
1443 	}
1444 
1445 	/*
1446 	 * Seq is within [bl, 0xffffffff] and bl is within
1447 	 * [0xffffffff-window, 0xffffffff].  This means we got a seq
1448 	 * which is within our replay window, but in the previous
1449 	 * subspace.
1450 	 */
1451 	if (tl < window - 1 && seq >= bl) {
1452 		if (th == 0) {
1453 			SECREPLAY_UNLOCK(replay);
1454 			return (1);
1455 		}
1456 		if (check_window(replay, seq)) {
1457 			SECREPLAY_UNLOCK(replay);
1458 			return (1);
1459 		}
1460 
1461 		set_window(replay, seq);
1462 		replay->count++;
1463 		SECREPLAY_UNLOCK(replay);
1464 		return (0);
1465 	}
1466 
1467 	/*
1468 	 * Seq is within [0, bl) but the whole window is within one subspace.
1469 	 * This means that seq has wrapped and is in next subspace
1470 	 */
1471 	seqh = th + 1;
1472 
1473 	/* Don't let high part wrap. */
1474 	if (seqh == 0) {
1475 		SECREPLAY_UNLOCK(replay);
1476 		return (1);
1477 	}
1478 
1479 	advance_window(replay, ((uint64_t)seqh << 32) | seq);
1480 	set_window(replay, seq);
1481 	replay->last = ((uint64_t)seqh << 32) | seq;
1482 	replay->count++;
1483 
1484 	SECREPLAY_UNLOCK(replay);
1485 	return (0);
1486 }
1487 int
ipsec_updateid(struct secasvar * sav,crypto_session_t * new,crypto_session_t * old)1488 ipsec_updateid(struct secasvar *sav, crypto_session_t *new,
1489     crypto_session_t *old)
1490 {
1491 	crypto_session_t tmp;
1492 
1493 	/*
1494 	 * tdb_cryptoid is initialized by xform_init().
1495 	 * Then it can be changed only when some crypto error occurred or
1496 	 * when SA is deleted. We stored used cryptoid in the xform_data
1497 	 * structure. In case when crypto error occurred and crypto
1498 	 * subsystem has reinited the session, it returns new cryptoid
1499 	 * and EAGAIN error code.
1500 	 *
1501 	 * This function will be called when we got EAGAIN from crypto
1502 	 * subsystem.
1503 	 * *new is cryptoid that was returned by crypto subsystem in
1504 	 * the crp_sid.
1505 	 * *old is the original cryptoid that we stored in xform_data.
1506 	 *
1507 	 * For first failed request *old == sav->tdb_cryptoid, then
1508 	 * we update sav->tdb_cryptoid and redo crypto_dispatch().
1509 	 * For next failed request *old != sav->tdb_cryptoid, then
1510 	 * we store cryptoid from first request into the *new variable
1511 	 * and crp_sid from this second session will be returned via
1512 	 * *old pointer, so caller can release second session.
1513 	 *
1514 	 * XXXAE: check this more carefully.
1515 	 */
1516 	KEYDBG(IPSEC_STAMP,
1517 	    printf("%s: SA(%p) moves cryptoid %p -> %p\n",
1518 		__func__, sav, *old, *new));
1519 	KEYDBG(IPSEC_DATA, kdebug_secasv(sav));
1520 	SECASVAR_WLOCK(sav);
1521 	if (sav->tdb_cryptoid != *old) {
1522 		/* cryptoid was already updated */
1523 		tmp = *new;
1524 		*new = sav->tdb_cryptoid;
1525 		*old = tmp;
1526 		SECASVAR_WUNLOCK(sav);
1527 		return (1);
1528 	}
1529 	sav->tdb_cryptoid = *new;
1530 	SECASVAR_WUNLOCK(sav);
1531 	return (0);
1532 }
1533 
1534 int
ipsec_initialized(void)1535 ipsec_initialized(void)
1536 {
1537 
1538 	return (V_def_policy != NULL);
1539 }
1540 
1541 static void
def_policy_init(const void * unused __unused)1542 def_policy_init(const void *unused __unused)
1543 {
1544 
1545 	V_def_policy = key_newsp();
1546 	if (V_def_policy != NULL) {
1547 		V_def_policy->policy = IPSEC_POLICY_NONE;
1548 		/* Force INPCB SP cache invalidation */
1549 		key_bumpspgen();
1550 	} else
1551 		printf("%s: failed to initialize default policy\n", __func__);
1552 }
1553 
1554 static void
def_policy_uninit(const void * unused __unused)1555 def_policy_uninit(const void *unused __unused)
1556 {
1557 
1558 	if (V_def_policy != NULL) {
1559 		key_freesp(&V_def_policy);
1560 		key_bumpspgen();
1561 	}
1562 }
1563 
1564 VNET_SYSINIT(def_policy_init, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST,
1565     def_policy_init, NULL);
1566 VNET_SYSUNINIT(def_policy_uninit, SI_SUB_PROTO_DOMAIN, SI_ORDER_FIRST,
1567     def_policy_uninit, NULL);
1568