xref: /freebsd/sys/net/if_ipsec.c (revision a5ada504b340e3e85bf907cbdbb7109ee16edb70)
1 /*-
2  * Copyright (c) 2016-2018 Yandex LLC
3  * Copyright (c) 2016-2018 Andrey V. Elsukov <ae@FreeBSD.org>
4  * All rights reserved.
5  *
6  * Redistribution and use in source and binary forms, with or without
7  * modification, are permitted provided that the following conditions
8  * are met:
9  *
10  * 1. Redistributions of source code must retain the above copyright
11  *    notice, this list of conditions and the following disclaimer.
12  * 2. Redistributions in binary form must reproduce the above copyright
13  *    notice, this list of conditions and the following disclaimer in the
14  *    documentation and/or other materials provided with the distribution.
15  *
16  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
17  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
18  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
19  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
20  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
21  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
22  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
23  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
24  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
25  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
26  */
27 
28 #include <sys/cdefs.h>
29 __FBSDID("$FreeBSD$");
30 
31 #include "opt_inet.h"
32 #include "opt_inet6.h"
33 
34 #include <sys/param.h>
35 #include <sys/systm.h>
36 #include <sys/kernel.h>
37 #include <sys/fnv_hash.h>
38 #include <sys/jail.h>
39 #include <sys/lock.h>
40 #include <sys/malloc.h>
41 #include <sys/mbuf.h>
42 #include <sys/module.h>
43 #include <sys/socket.h>
44 #include <sys/sockio.h>
45 #include <sys/sx.h>
46 #include <sys/errno.h>
47 #include <sys/sysctl.h>
48 #include <sys/priv.h>
49 #include <sys/proc.h>
50 #include <sys/conf.h>
51 
52 #include <net/if.h>
53 #include <net/if_var.h>
54 #include <net/if_clone.h>
55 #include <net/if_types.h>
56 #include <net/bpf.h>
57 #include <net/route.h>
58 #include <net/vnet.h>
59 
60 #include <netinet/in.h>
61 #include <netinet/in_var.h>
62 #include <netinet/ip.h>
63 #include <netinet/ip_encap.h>
64 
65 #include <netinet/ip6.h>
66 #include <netinet6/in6_var.h>
67 #include <netinet6/scope6_var.h>
68 
69 #include <netipsec/ipsec.h>
70 #ifdef INET6
71 #include <netipsec/ipsec6.h>
72 #endif
73 
74 #include <net/if_ipsec.h>
75 #include <netipsec/key.h>
76 
77 #include <security/mac/mac_framework.h>
78 
79 static MALLOC_DEFINE(M_IPSEC, "ipsec", "IPsec Virtual Tunnel Interface");
80 static const char ipsecname[] = "ipsec";
81 
82 #if defined(INET) && defined(INET6)
83 #define	IPSEC_SPCOUNT		4
84 #else
85 #define	IPSEC_SPCOUNT		2
86 #endif
87 
88 struct ipsec_softc {
89 	struct ifnet		*ifp;
90 	struct secpolicy	*sp[IPSEC_SPCOUNT];
91 	uint32_t		reqid;
92 	u_int			family;
93 	u_int			fibnum;
94 
95 	CK_LIST_ENTRY(ipsec_softc) idhash;
96 	CK_LIST_ENTRY(ipsec_softc) srchash;
97 };
98 
99 #define	IPSEC_RLOCK_TRACKER	struct epoch_tracker ipsec_et
100 #define	IPSEC_RLOCK()	epoch_enter_preempt(net_epoch_preempt, &ipsec_et)
101 #define	IPSEC_RUNLOCK()	epoch_exit_preempt(net_epoch_preempt, &ipsec_et)
102 #define	IPSEC_WAIT()	epoch_wait_preempt(net_epoch_preempt)
103 
104 #ifndef IPSEC_HASH_SIZE
105 #define	IPSEC_HASH_SIZE	(1 << 5)
106 #endif
107 
108 CK_LIST_HEAD(ipsec_iflist, ipsec_softc);
109 VNET_DEFINE_STATIC(struct ipsec_iflist *, ipsec_idhtbl) = NULL;
110 #define	V_ipsec_idhtbl		VNET(ipsec_idhtbl)
111 
112 #ifdef INET
113 VNET_DEFINE_STATIC(struct ipsec_iflist *, ipsec4_srchtbl) = NULL;
114 #define	V_ipsec4_srchtbl	VNET(ipsec4_srchtbl)
115 static const struct srcaddrtab *ipsec4_srctab = NULL;
116 #endif
117 
118 #ifdef INET6
119 VNET_DEFINE_STATIC(struct ipsec_iflist *, ipsec6_srchtbl) = NULL;
120 #define	V_ipsec6_srchtbl	VNET(ipsec6_srchtbl)
121 static const struct srcaddrtab *ipsec6_srctab = NULL;
122 #endif
123 
124 static struct ipsec_iflist *
125 ipsec_idhash(uint32_t id)
126 {
127 
128 	return (&V_ipsec_idhtbl[fnv_32_buf(&id, sizeof(id),
129 	    FNV1_32_INIT) & (IPSEC_HASH_SIZE - 1)]);
130 }
131 
132 static struct ipsec_iflist *
133 ipsec_srchash(const struct sockaddr *sa)
134 {
135 	uint32_t hval;
136 
137 	switch (sa->sa_family) {
138 #ifdef INET
139 	case AF_INET:
140 		hval = fnv_32_buf(
141 		    &((const struct sockaddr_in *)sa)->sin_addr.s_addr,
142 		    sizeof(in_addr_t), FNV1_32_INIT);
143 		return (&V_ipsec4_srchtbl[hval & (IPSEC_HASH_SIZE - 1)]);
144 #endif
145 #ifdef INET6
146 	case AF_INET6:
147 		hval = fnv_32_buf(
148 		    &((const struct sockaddr_in6 *)sa)->sin6_addr,
149 		    sizeof(struct in6_addr), FNV1_32_INIT);
150 		return (&V_ipsec6_srchtbl[hval & (IPSEC_HASH_SIZE - 1)]);
151 #endif
152 	}
153 	return (NULL);
154 }
155 
156 /*
157  * ipsec_ioctl_sx protects from concurrent ioctls.
158  */
159 static struct sx ipsec_ioctl_sx;
160 SX_SYSINIT(ipsec_ioctl_sx, &ipsec_ioctl_sx, "ipsec_ioctl");
161 
162 static int	ipsec_init_reqid(struct ipsec_softc *);
163 static int	ipsec_set_tunnel(struct ipsec_softc *, struct sockaddr *,
164     struct sockaddr *, uint32_t);
165 static void	ipsec_delete_tunnel(struct ipsec_softc *);
166 
167 static int	ipsec_set_addresses(struct ifnet *, struct sockaddr *,
168     struct sockaddr *);
169 static int	ipsec_set_reqid(struct ipsec_softc *, uint32_t);
170 static void	ipsec_set_running(struct ipsec_softc *);
171 
172 static void	ipsec_srcaddr(void *, const struct sockaddr *, int);
173 static int	ipsec_ioctl(struct ifnet *, u_long, caddr_t);
174 static int	ipsec_transmit(struct ifnet *, struct mbuf *);
175 static int	ipsec_output(struct ifnet *, struct mbuf *,
176     const struct sockaddr *, struct route *);
177 static void	ipsec_qflush(struct ifnet *);
178 static int	ipsec_clone_create(struct if_clone *, int, caddr_t);
179 static void	ipsec_clone_destroy(struct ifnet *);
180 
181 VNET_DEFINE_STATIC(struct if_clone *, ipsec_cloner);
182 #define	V_ipsec_cloner		VNET(ipsec_cloner)
183 
184 static int
185 ipsec_clone_create(struct if_clone *ifc, int unit, caddr_t params)
186 {
187 	struct ipsec_softc *sc;
188 	struct ifnet *ifp;
189 
190 	sc = malloc(sizeof(*sc), M_IPSEC, M_WAITOK | M_ZERO);
191 	sc->fibnum = curthread->td_proc->p_fibnum;
192 	sc->ifp = ifp = if_alloc(IFT_TUNNEL);
193 	ifp->if_softc = sc;
194 	if_initname(ifp, ipsecname, unit);
195 
196 	ifp->if_addrlen = 0;
197 	ifp->if_mtu = IPSEC_MTU;
198 	ifp->if_flags  = IFF_POINTOPOINT | IFF_MULTICAST;
199 	ifp->if_ioctl  = ipsec_ioctl;
200 	ifp->if_transmit  = ipsec_transmit;
201 	ifp->if_qflush  = ipsec_qflush;
202 	ifp->if_output = ipsec_output;
203 	if_attach(ifp);
204 	bpfattach(ifp, DLT_NULL, sizeof(uint32_t));
205 
206 	return (0);
207 }
208 
209 static void
210 ipsec_clone_destroy(struct ifnet *ifp)
211 {
212 	struct ipsec_softc *sc;
213 
214 	sx_xlock(&ipsec_ioctl_sx);
215 	sc = ifp->if_softc;
216 	ipsec_delete_tunnel(sc);
217 	/*
218 	 * Delete softc from idhash on interface destroy, since
219 	 * ipsec_delete_tunnel() keeps reqid unchanged.
220 	 */
221 	if (sc->reqid != 0)
222 		CK_LIST_REMOVE(sc, idhash);
223 	bpfdetach(ifp);
224 	if_detach(ifp);
225 	ifp->if_softc = NULL;
226 	sx_xunlock(&ipsec_ioctl_sx);
227 
228 	IPSEC_WAIT();
229 	if_free(ifp);
230 	free(sc, M_IPSEC);
231 }
232 
233 static struct ipsec_iflist *
234 ipsec_hashinit(void)
235 {
236 	struct ipsec_iflist *hash;
237 	int i;
238 
239 	hash = malloc(sizeof(struct ipsec_iflist) * IPSEC_HASH_SIZE,
240 	    M_IPSEC, M_WAITOK);
241 	for (i = 0; i < IPSEC_HASH_SIZE; i++)
242 		CK_LIST_INIT(&hash[i]);
243 
244 	return (hash);
245 }
246 
247 static void
248 vnet_ipsec_init(const void *unused __unused)
249 {
250 
251 	V_ipsec_idhtbl = ipsec_hashinit();
252 #ifdef INET
253 	V_ipsec4_srchtbl = ipsec_hashinit();
254 	if (IS_DEFAULT_VNET(curvnet))
255 		ipsec4_srctab = ip_encap_register_srcaddr(ipsec_srcaddr,
256 		    NULL, M_WAITOK);
257 #endif
258 #ifdef INET6
259 	V_ipsec6_srchtbl = ipsec_hashinit();
260 	if (IS_DEFAULT_VNET(curvnet))
261 		ipsec6_srctab = ip6_encap_register_srcaddr(ipsec_srcaddr,
262 		    NULL, M_WAITOK);
263 #endif
264 	V_ipsec_cloner = if_clone_simple(ipsecname, ipsec_clone_create,
265 	    ipsec_clone_destroy, 0);
266 }
267 VNET_SYSINIT(vnet_ipsec_init, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
268     vnet_ipsec_init, NULL);
269 
270 static void
271 vnet_ipsec_uninit(const void *unused __unused)
272 {
273 
274 	if_clone_detach(V_ipsec_cloner);
275 	free(V_ipsec_idhtbl, M_IPSEC);
276 	/*
277 	 * Use V_ipsec_idhtbl pointer as indicator that VNET is going to be
278 	 * destroyed, it is used by ipsec_srcaddr() callback.
279 	 */
280 	V_ipsec_idhtbl = NULL;
281 	IPSEC_WAIT();
282 
283 #ifdef INET
284 	if (IS_DEFAULT_VNET(curvnet))
285 		ip_encap_unregister_srcaddr(ipsec4_srctab);
286 	free(V_ipsec4_srchtbl, M_IPSEC);
287 #endif
288 #ifdef INET6
289 	if (IS_DEFAULT_VNET(curvnet))
290 		ip6_encap_unregister_srcaddr(ipsec6_srctab);
291 	free(V_ipsec6_srchtbl, M_IPSEC);
292 #endif
293 }
294 VNET_SYSUNINIT(vnet_ipsec_uninit, SI_SUB_PROTO_IFATTACHDOMAIN, SI_ORDER_ANY,
295     vnet_ipsec_uninit, NULL);
296 
297 static struct secpolicy *
298 ipsec_getpolicy(struct ipsec_softc *sc, int dir, sa_family_t af)
299 {
300 
301 	switch (af) {
302 #ifdef INET
303 	case AF_INET:
304 		return (sc->sp[(dir == IPSEC_DIR_INBOUND ? 0: 1)]);
305 #endif
306 #ifdef INET6
307 	case AF_INET6:
308 		return (sc->sp[(dir == IPSEC_DIR_INBOUND ? 0: 1)
309 #ifdef INET
310 			+ 2
311 #endif
312 		]);
313 #endif
314 	}
315 	return (NULL);
316 }
317 
318 static struct secasindex *
319 ipsec_getsaidx(struct ipsec_softc *sc, int dir, sa_family_t af)
320 {
321 	struct secpolicy *sp;
322 
323 	sp = ipsec_getpolicy(sc, dir, af);
324 	if (sp == NULL)
325 		return (NULL);
326 	return (&sp->req[0]->saidx);
327 }
328 
329 static int
330 ipsec_transmit(struct ifnet *ifp, struct mbuf *m)
331 {
332 	IPSEC_RLOCK_TRACKER;
333 	struct ipsec_softc *sc;
334 	struct secpolicy *sp;
335 	struct ip *ip;
336 	uint32_t af;
337 	int error;
338 
339 	IPSEC_RLOCK();
340 #ifdef MAC
341 	error = mac_ifnet_check_transmit(ifp, m);
342 	if (error) {
343 		m_freem(m);
344 		goto err;
345 	}
346 #endif
347 	error = ENETDOWN;
348 	sc = ifp->if_softc;
349 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
350 	    (ifp->if_flags & IFF_MONITOR) != 0 ||
351 	    (ifp->if_flags & IFF_UP) == 0 || sc->family == 0) {
352 		m_freem(m);
353 		goto err;
354 	}
355 
356 	/* Determine address family to correctly handle packet in BPF */
357 	ip = mtod(m, struct ip *);
358 	switch (ip->ip_v) {
359 #ifdef INET
360 	case IPVERSION:
361 		af = AF_INET;
362 		break;
363 #endif
364 #ifdef INET6
365 	case (IPV6_VERSION >> 4):
366 		af = AF_INET6;
367 		break;
368 #endif
369 	default:
370 		error = EAFNOSUPPORT;
371 		m_freem(m);
372 		goto err;
373 	}
374 
375 	/*
376 	 * Loop prevention.
377 	 * XXX: for now just check presence of IPSEC_OUT_DONE mbuf tag.
378 	 *      We can read full chain and compare destination address,
379 	 *      proto and mode from xform_history with values from softc.
380 	 */
381 	if (m_tag_find(m, PACKET_TAG_IPSEC_OUT_DONE, NULL) != NULL) {
382 		m_freem(m);
383 		goto err;
384 	}
385 
386 	sp = ipsec_getpolicy(sc, IPSEC_DIR_OUTBOUND, af);
387 	key_addref(sp);
388 	M_SETFIB(m, sc->fibnum);
389 
390 	BPF_MTAP2(ifp, &af, sizeof(af), m);
391 	if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
392 	if_inc_counter(ifp, IFCOUNTER_OBYTES, m->m_pkthdr.len);
393 
394 	switch (af) {
395 #ifdef INET
396 	case AF_INET:
397 		error = ipsec4_process_packet(m, sp, NULL);
398 		break;
399 #endif
400 #ifdef INET6
401 	case AF_INET6:
402 		error = ipsec6_process_packet(m, sp, NULL);
403 		break;
404 #endif
405 	default:
406 		panic("%s: unknown address family\n", __func__);
407 	}
408 err:
409 	if (error != 0)
410 		if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
411 	IPSEC_RUNLOCK();
412 	return (error);
413 }
414 
415 static void
416 ipsec_qflush(struct ifnet *ifp __unused)
417 {
418 
419 }
420 
421 static int
422 ipsec_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
423 	struct route *ro)
424 {
425 
426 	return (ifp->if_transmit(ifp, m));
427 }
428 
429 int
430 ipsec_if_input(struct mbuf *m, struct secasvar *sav, uint32_t af)
431 {
432 	IPSEC_RLOCK_TRACKER;
433 	struct secasindex *saidx;
434 	struct ipsec_softc *sc;
435 	struct ifnet *ifp;
436 
437 	if (sav->state != SADB_SASTATE_MATURE &&
438 	    sav->state != SADB_SASTATE_DYING) {
439 		m_freem(m);
440 		return (ENETDOWN);
441 	}
442 
443 	if (sav->sah->saidx.mode != IPSEC_MODE_TUNNEL ||
444 	    sav->sah->saidx.proto != IPPROTO_ESP)
445 		return (0);
446 
447 	IPSEC_RLOCK();
448 	CK_LIST_FOREACH(sc, ipsec_idhash(sav->sah->saidx.reqid), idhash) {
449 		if (sc->family == 0)
450 			continue;
451 		saidx = ipsec_getsaidx(sc, IPSEC_DIR_INBOUND,
452 		    sav->sah->saidx.src.sa.sa_family);
453 		/* SA's reqid should match reqid in SP */
454 		if (saidx == NULL ||
455 		    sav->sah->saidx.reqid != saidx->reqid)
456 			continue;
457 		/* SAH's addresses should match tunnel endpoints. */
458 		if (key_sockaddrcmp(&sav->sah->saidx.dst.sa,
459 		    &saidx->dst.sa, 0) != 0)
460 			continue;
461 		if (key_sockaddrcmp(&sav->sah->saidx.src.sa,
462 		    &saidx->src.sa, 0) == 0)
463 			break;
464 	}
465 	if (sc == NULL) {
466 		IPSEC_RUNLOCK();
467 		/* Tunnel was not found. Nothing to do. */
468 		return (0);
469 	}
470 	ifp = sc->ifp;
471 	if ((ifp->if_drv_flags & IFF_DRV_RUNNING) == 0 ||
472 	    (ifp->if_flags & IFF_UP) == 0) {
473 		IPSEC_RUNLOCK();
474 		m_freem(m);
475 		return (ENETDOWN);
476 	}
477 	/*
478 	 * We found matching and working tunnel.
479 	 * Set its ifnet as receiving interface.
480 	 */
481 	m->m_pkthdr.rcvif = ifp;
482 
483 	m_clrprotoflags(m);
484 	M_SETFIB(m, ifp->if_fib);
485 	BPF_MTAP2(ifp, &af, sizeof(af), m);
486 	if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
487 	if_inc_counter(ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
488 	if ((ifp->if_flags & IFF_MONITOR) != 0) {
489 		IPSEC_RUNLOCK();
490 		m_freem(m);
491 		return (ENETDOWN);
492 	}
493 	IPSEC_RUNLOCK();
494 	return (0);
495 }
496 
497 static int
498 ipsec_ioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
499 {
500 	struct ifreq *ifr = (struct ifreq*)data;
501 	struct sockaddr *dst, *src;
502 	struct ipsec_softc *sc;
503 	struct secasindex *saidx;
504 #ifdef INET
505 	struct sockaddr_in *sin = NULL;
506 #endif
507 #ifdef INET6
508 	struct sockaddr_in6 *sin6 = NULL;
509 #endif
510 	uint32_t reqid;
511 	int error;
512 
513 	switch (cmd) {
514 	case SIOCSIFADDR:
515 		ifp->if_flags |= IFF_UP;
516 	case SIOCADDMULTI:
517 	case SIOCDELMULTI:
518 	case SIOCGIFMTU:
519 	case SIOCSIFFLAGS:
520 		return (0);
521 	case SIOCSIFMTU:
522 		if (ifr->ifr_mtu < IPSEC_MTU_MIN ||
523 		    ifr->ifr_mtu > IPSEC_MTU_MAX)
524 			return (EINVAL);
525 		else
526 			ifp->if_mtu = ifr->ifr_mtu;
527 		return (0);
528 	}
529 	sx_xlock(&ipsec_ioctl_sx);
530 	sc = ifp->if_softc;
531 	/* Check that softc is still here */
532 	if (sc == NULL) {
533 		error = ENXIO;
534 		goto bad;
535 	}
536 	error = 0;
537 	switch (cmd) {
538 	case SIOCSIFPHYADDR:
539 #ifdef INET6
540 	case SIOCSIFPHYADDR_IN6:
541 #endif
542 		error = EINVAL;
543 		switch (cmd) {
544 #ifdef INET
545 		case SIOCSIFPHYADDR:
546 			src = (struct sockaddr *)
547 				&(((struct in_aliasreq *)data)->ifra_addr);
548 			dst = (struct sockaddr *)
549 				&(((struct in_aliasreq *)data)->ifra_dstaddr);
550 			break;
551 #endif
552 #ifdef INET6
553 		case SIOCSIFPHYADDR_IN6:
554 			src = (struct sockaddr *)
555 				&(((struct in6_aliasreq *)data)->ifra_addr);
556 			dst = (struct sockaddr *)
557 				&(((struct in6_aliasreq *)data)->ifra_dstaddr);
558 			break;
559 #endif
560 		default:
561 			goto bad;
562 		}
563 		/* sa_family must be equal */
564 		if (src->sa_family != dst->sa_family ||
565 		    src->sa_len != dst->sa_len)
566 			goto bad;
567 
568 		/* validate sa_len */
569 		switch (src->sa_family) {
570 #ifdef INET
571 		case AF_INET:
572 			if (src->sa_len != sizeof(struct sockaddr_in))
573 				goto bad;
574 			break;
575 #endif
576 #ifdef INET6
577 		case AF_INET6:
578 			if (src->sa_len != sizeof(struct sockaddr_in6))
579 				goto bad;
580 			break;
581 #endif
582 		default:
583 			error = EAFNOSUPPORT;
584 			goto bad;
585 		}
586 		/* check sa_family looks sane for the cmd */
587 		error = EAFNOSUPPORT;
588 		switch (cmd) {
589 #ifdef INET
590 		case SIOCSIFPHYADDR:
591 			if (src->sa_family == AF_INET)
592 				break;
593 			goto bad;
594 #endif
595 #ifdef INET6
596 		case SIOCSIFPHYADDR_IN6:
597 			if (src->sa_family == AF_INET6)
598 				break;
599 			goto bad;
600 #endif
601 		}
602 		error = EADDRNOTAVAIL;
603 		switch (src->sa_family) {
604 #ifdef INET
605 		case AF_INET:
606 			if (satosin(src)->sin_addr.s_addr == INADDR_ANY ||
607 			    satosin(dst)->sin_addr.s_addr == INADDR_ANY)
608 				goto bad;
609 			break;
610 #endif
611 #ifdef INET6
612 		case AF_INET6:
613 			if (IN6_IS_ADDR_UNSPECIFIED(
614 			    &satosin6(src)->sin6_addr) ||
615 			    IN6_IS_ADDR_UNSPECIFIED(
616 			    &satosin6(dst)->sin6_addr))
617 				goto bad;
618 			/*
619 			 * Check validity of the scope zone ID of the
620 			 * addresses, and convert it into the kernel
621 			 * internal form if necessary.
622 			 */
623 			error = sa6_embedscope(satosin6(src), 0);
624 			if (error != 0)
625 				goto bad;
626 			error = sa6_embedscope(satosin6(dst), 0);
627 			if (error != 0)
628 				goto bad;
629 #endif
630 		};
631 		error = ipsec_set_addresses(ifp, src, dst);
632 		break;
633 	case SIOCDIFPHYADDR:
634 		ipsec_delete_tunnel(sc);
635 		break;
636 	case SIOCGIFPSRCADDR:
637 	case SIOCGIFPDSTADDR:
638 #ifdef INET6
639 	case SIOCGIFPSRCADDR_IN6:
640 	case SIOCGIFPDSTADDR_IN6:
641 #endif
642 		if (sc->family == 0) {
643 			error = EADDRNOTAVAIL;
644 			break;
645 		}
646 		saidx = ipsec_getsaidx(sc, IPSEC_DIR_OUTBOUND, sc->family);
647 		switch (cmd) {
648 #ifdef INET
649 		case SIOCGIFPSRCADDR:
650 		case SIOCGIFPDSTADDR:
651 			if (saidx->src.sa.sa_family != AF_INET) {
652 				error = EADDRNOTAVAIL;
653 				break;
654 			}
655 			sin = (struct sockaddr_in *)&ifr->ifr_addr;
656 			memset(sin, 0, sizeof(*sin));
657 			sin->sin_family = AF_INET;
658 			sin->sin_len = sizeof(*sin);
659 			break;
660 #endif
661 #ifdef INET6
662 		case SIOCGIFPSRCADDR_IN6:
663 		case SIOCGIFPDSTADDR_IN6:
664 			if (saidx->src.sa.sa_family != AF_INET6) {
665 				error = EADDRNOTAVAIL;
666 				break;
667 			}
668 			sin6 = (struct sockaddr_in6 *)
669 				&(((struct in6_ifreq *)data)->ifr_addr);
670 			memset(sin6, 0, sizeof(*sin6));
671 			sin6->sin6_family = AF_INET6;
672 			sin6->sin6_len = sizeof(*sin6);
673 			break;
674 #endif
675 		default:
676 			error = EAFNOSUPPORT;
677 		}
678 		if (error == 0) {
679 			switch (cmd) {
680 #ifdef INET
681 			case SIOCGIFPSRCADDR:
682 				sin->sin_addr = saidx->src.sin.sin_addr;
683 				break;
684 			case SIOCGIFPDSTADDR:
685 				sin->sin_addr = saidx->dst.sin.sin_addr;
686 				break;
687 #endif
688 #ifdef INET6
689 			case SIOCGIFPSRCADDR_IN6:
690 				sin6->sin6_addr = saidx->src.sin6.sin6_addr;
691 				break;
692 			case SIOCGIFPDSTADDR_IN6:
693 				sin6->sin6_addr = saidx->dst.sin6.sin6_addr;
694 				break;
695 #endif
696 			}
697 		}
698 		if (error != 0)
699 			break;
700 		switch (cmd) {
701 #ifdef INET
702 		case SIOCGIFPSRCADDR:
703 		case SIOCGIFPDSTADDR:
704 			error = prison_if(curthread->td_ucred,
705 			    (struct sockaddr *)sin);
706 			if (error != 0)
707 				memset(sin, 0, sizeof(*sin));
708 			break;
709 #endif
710 #ifdef INET6
711 		case SIOCGIFPSRCADDR_IN6:
712 		case SIOCGIFPDSTADDR_IN6:
713 			error = prison_if(curthread->td_ucred,
714 			    (struct sockaddr *)sin6);
715 			if (error == 0)
716 				error = sa6_recoverscope(sin6);
717 			if (error != 0)
718 				memset(sin6, 0, sizeof(*sin6));
719 #endif
720 		}
721 		break;
722 	case SIOCGTUNFIB:
723 		ifr->ifr_fib = sc->fibnum;
724 		break;
725 	case SIOCSTUNFIB:
726 		if ((error = priv_check(curthread, PRIV_NET_SETIFFIB)) != 0)
727 			break;
728 		if (ifr->ifr_fib >= rt_numfibs)
729 			error = EINVAL;
730 		else
731 			sc->fibnum = ifr->ifr_fib;
732 		break;
733 	case IPSECGREQID:
734 		reqid = sc->reqid;
735 		error = copyout(&reqid, ifr_data_get_ptr(ifr), sizeof(reqid));
736 		break;
737 	case IPSECSREQID:
738 		if ((error = priv_check(curthread, PRIV_NET_SETIFCAP)) != 0)
739 			break;
740 		error = copyin(ifr_data_get_ptr(ifr), &reqid, sizeof(reqid));
741 		if (error != 0)
742 			break;
743 		error = ipsec_set_reqid(sc, reqid);
744 		break;
745 	default:
746 		error = EINVAL;
747 		break;
748 	}
749 bad:
750 	sx_xunlock(&ipsec_ioctl_sx);
751 	return (error);
752 }
753 
754 /*
755  * Check that ingress address belongs to local host.
756  */
757 static void
758 ipsec_set_running(struct ipsec_softc *sc)
759 {
760 	struct secasindex *saidx;
761 	int localip;
762 
763 	saidx = ipsec_getsaidx(sc, IPSEC_DIR_OUTBOUND, sc->family);
764 	localip = 0;
765 	switch (sc->family) {
766 #ifdef INET
767 	case AF_INET:
768 		localip = in_localip(saidx->src.sin.sin_addr);
769 		break;
770 #endif
771 #ifdef INET6
772 	case AF_INET6:
773 		localip = in6_localip(&saidx->src.sin6.sin6_addr);
774 		break;
775 #endif
776 	}
777 	if (localip != 0)
778 		sc->ifp->if_drv_flags |= IFF_DRV_RUNNING;
779 	else
780 		sc->ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
781 }
782 
783 /*
784  * ifaddr_event handler.
785  * Clear IFF_DRV_RUNNING flag when ingress address disappears to prevent
786  * source address spoofing.
787  */
788 static void
789 ipsec_srcaddr(void *arg __unused, const struct sockaddr *sa,
790     int event __unused)
791 {
792 	struct ipsec_softc *sc;
793 	struct secasindex *saidx;
794 
795 	/* Check that VNET is ready */
796 	if (V_ipsec_idhtbl == NULL)
797 		return;
798 
799 	MPASS(in_epoch(net_epoch_preempt));
800 	CK_LIST_FOREACH(sc, ipsec_srchash(sa), srchash) {
801 		if (sc->family == 0)
802 			continue;
803 		saidx = ipsec_getsaidx(sc, IPSEC_DIR_OUTBOUND, sa->sa_family);
804 		if (saidx == NULL ||
805 		    key_sockaddrcmp(&saidx->src.sa, sa, 0) != 0)
806 			continue;
807 		ipsec_set_running(sc);
808 	}
809 }
810 
811 /*
812  * Allocate new private security policies for tunneling interface.
813  * Each tunneling interface has following security policies for
814  * both AF:
815  *   0.0.0.0/0[any] 0.0.0.0/0[any] -P in \
816  *	ipsec esp/tunnel/RemoteIP-LocalIP/unique:reqid
817  *   0.0.0.0/0[any] 0.0.0.0/0[any] -P out \
818  *	ipsec esp/tunnel/LocalIP-RemoteIP/unique:reqid
819  */
820 static int
821 ipsec_newpolicies(struct ipsec_softc *sc, struct secpolicy *sp[IPSEC_SPCOUNT],
822     const struct sockaddr *src, const struct sockaddr *dst, uint32_t reqid)
823 {
824 	struct ipsecrequest *isr;
825 	int i;
826 
827 	memset(sp, 0, sizeof(struct secpolicy *) * IPSEC_SPCOUNT);
828 	for (i = 0; i < IPSEC_SPCOUNT; i++) {
829 		if ((sp[i] = key_newsp()) == NULL)
830 			goto fail;
831 		if ((isr = ipsec_newisr()) == NULL)
832 			goto fail;
833 
834 		sp[i]->policy = IPSEC_POLICY_IPSEC;
835 		sp[i]->state = IPSEC_SPSTATE_DEAD;
836 		sp[i]->req[sp[i]->tcount++] = isr;
837 		sp[i]->created = time_second;
838 		/* Use priority field to store if_index */
839 		sp[i]->priority = sc->ifp->if_index;
840 		isr->level = IPSEC_LEVEL_UNIQUE;
841 		isr->saidx.proto = IPPROTO_ESP;
842 		isr->saidx.mode = IPSEC_MODE_TUNNEL;
843 		isr->saidx.reqid = reqid;
844 		if (i % 2 == 0) {
845 			sp[i]->spidx.dir = IPSEC_DIR_INBOUND;
846 			bcopy(src, &isr->saidx.dst, src->sa_len);
847 			bcopy(dst, &isr->saidx.src, dst->sa_len);
848 		} else {
849 			sp[i]->spidx.dir = IPSEC_DIR_OUTBOUND;
850 			bcopy(src, &isr->saidx.src, src->sa_len);
851 			bcopy(dst, &isr->saidx.dst, dst->sa_len);
852 		}
853 		sp[i]->spidx.ul_proto = IPSEC_ULPROTO_ANY;
854 #ifdef INET
855 		if (i < 2) {
856 			sp[i]->spidx.src.sa.sa_family =
857 			    sp[i]->spidx.dst.sa.sa_family = AF_INET;
858 			sp[i]->spidx.src.sa.sa_len =
859 			    sp[i]->spidx.dst.sa.sa_len =
860 			    sizeof(struct sockaddr_in);
861 			continue;
862 		}
863 #endif
864 #ifdef INET6
865 		sp[i]->spidx.src.sa.sa_family =
866 		    sp[i]->spidx.dst.sa.sa_family = AF_INET6;
867 		sp[i]->spidx.src.sa.sa_len =
868 		    sp[i]->spidx.dst.sa.sa_len = sizeof(struct sockaddr_in6);
869 #endif
870 	}
871 	return (0);
872 fail:
873 	for (i = 0; i < IPSEC_SPCOUNT; i++)
874 		key_freesp(&sp[i]);
875 	return (ENOMEM);
876 }
877 
878 static int
879 ipsec_check_reqid(uint32_t reqid)
880 {
881 	struct ipsec_softc *sc;
882 
883 	sx_assert(&ipsec_ioctl_sx, SA_XLOCKED);
884 	CK_LIST_FOREACH(sc, ipsec_idhash(reqid), idhash) {
885 		if (sc->reqid == reqid)
886 			return (EEXIST);
887 	}
888 	return (0);
889 }
890 
891 /*
892  * We use key_newreqid() to automatically obtain unique reqid.
893  * Then we check that given id is unique, i.e. it is not used by
894  * another if_ipsec(4) interface. This macro limits the number of
895  * tries to get unique id.
896  */
897 #define	IPSEC_REQID_TRYCNT	64
898 static int
899 ipsec_init_reqid(struct ipsec_softc *sc)
900 {
901 	uint32_t reqid;
902 	int trycount;
903 
904 	sx_assert(&ipsec_ioctl_sx, SA_XLOCKED);
905 	if (sc->reqid != 0) /* already initialized */
906 		return (0);
907 
908 	trycount = IPSEC_REQID_TRYCNT;
909 	while (--trycount > 0) {
910 		reqid = key_newreqid();
911 		if (ipsec_check_reqid(reqid) == 0)
912 			break;
913 	}
914 	if (trycount == 0)
915 		return (EEXIST);
916 	sc->reqid = reqid;
917 	CK_LIST_INSERT_HEAD(ipsec_idhash(reqid), sc, idhash);
918 	return (0);
919 }
920 
921 /*
922  * Set or update reqid for given tunneling interface.
923  * When specified reqid is zero, generate new one.
924  * We are protected by ioctl_sx lock from concurrent id generation.
925  * Also softc would not disappear while we hold ioctl_sx lock.
926  */
927 static int
928 ipsec_set_reqid(struct ipsec_softc *sc, uint32_t reqid)
929 {
930 	struct secasindex *saidx;
931 
932 	sx_assert(&ipsec_ioctl_sx, SA_XLOCKED);
933 
934 	if (sc->reqid == reqid && reqid != 0)
935 		return (0);
936 
937 	if (reqid != 0) {
938 		/* Check that specified reqid doesn't exist */
939 		if (ipsec_check_reqid(reqid) != 0)
940 			return (EEXIST);
941 		if (sc->reqid != 0) {
942 			CK_LIST_REMOVE(sc, idhash);
943 			IPSEC_WAIT();
944 		}
945 		sc->reqid = reqid;
946 		CK_LIST_INSERT_HEAD(ipsec_idhash(reqid), sc, idhash);
947 	} else {
948 		/* Generate new reqid */
949 		if (ipsec_init_reqid(sc) != 0)
950 			return (EEXIST);
951 	}
952 
953 	/* Tunnel isn't fully configured, just return. */
954 	if (sc->family == 0)
955 		return (0);
956 
957 	saidx = ipsec_getsaidx(sc, IPSEC_DIR_OUTBOUND, sc->family);
958 	KASSERT(saidx != NULL,
959 	    ("saidx is NULL, but family is %d", sc->family));
960 	return (ipsec_set_tunnel(sc, &saidx->src.sa, &saidx->dst.sa,
961 	    sc->reqid));
962 }
963 
964 /*
965  * Set tunnel endpoints addresses.
966  */
967 static int
968 ipsec_set_addresses(struct ifnet *ifp, struct sockaddr *src,
969     struct sockaddr *dst)
970 {
971 	struct ipsec_softc *sc;
972 	struct secasindex *saidx;
973 
974 	sx_assert(&ipsec_ioctl_sx, SA_XLOCKED);
975 
976 	sc = ifp->if_softc;
977 	if (sc->family != 0) {
978 		saidx = ipsec_getsaidx(sc, IPSEC_DIR_OUTBOUND,
979 		    src->sa_family);
980 		if (saidx != NULL && saidx->reqid == sc->reqid &&
981 		    key_sockaddrcmp(&saidx->src.sa, src, 0) == 0 &&
982 		    key_sockaddrcmp(&saidx->dst.sa, dst, 0) == 0)
983 			return (0); /* Nothing has been changed. */
984 
985 	}
986 	/* If reqid is not set, generate new one. */
987 	if (ipsec_init_reqid(sc) != 0)
988 		return (EEXIST);
989 	return (ipsec_set_tunnel(sc, src, dst, sc->reqid));
990 }
991 
992 static int
993 ipsec_set_tunnel(struct ipsec_softc *sc, struct sockaddr *src,
994     struct sockaddr *dst, uint32_t reqid)
995 {
996 	struct secpolicy *sp[IPSEC_SPCOUNT];
997 	int i;
998 
999 	sx_assert(&ipsec_ioctl_sx, SA_XLOCKED);
1000 
1001 	/* Allocate SP with new addresses. */
1002 	if (ipsec_newpolicies(sc, sp, src, dst, reqid) == 0) {
1003 		/* Add new policies to SPDB */
1004 		if (key_register_ifnet(sp, IPSEC_SPCOUNT) != 0) {
1005 			for (i = 0; i < IPSEC_SPCOUNT; i++)
1006 				key_freesp(&sp[i]);
1007 			return (EAGAIN);
1008 		}
1009 		if (sc->family != 0)
1010 			ipsec_delete_tunnel(sc);
1011 		for (i = 0; i < IPSEC_SPCOUNT; i++)
1012 			sc->sp[i] = sp[i];
1013 		sc->family = src->sa_family;
1014 		CK_LIST_INSERT_HEAD(ipsec_srchash(src), sc, srchash);
1015 	} else {
1016 		sc->ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1017 		return (ENOMEM);
1018 	}
1019 	ipsec_set_running(sc);
1020 	return (0);
1021 }
1022 
1023 static void
1024 ipsec_delete_tunnel(struct ipsec_softc *sc)
1025 {
1026 	int i;
1027 
1028 	sx_assert(&ipsec_ioctl_sx, SA_XLOCKED);
1029 
1030 	sc->ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1031 	if (sc->family != 0) {
1032 		CK_LIST_REMOVE(sc, srchash);
1033 		sc->family = 0;
1034 		/*
1035 		 * Make sure that ipsec_if_input() will not do access
1036 		 * to softc's policies.
1037 		 */
1038 		IPSEC_WAIT();
1039 
1040 		key_unregister_ifnet(sc->sp, IPSEC_SPCOUNT);
1041 		for (i = 0; i < IPSEC_SPCOUNT; i++)
1042 			key_freesp(&sc->sp[i]);
1043 	}
1044 }
1045