xref: /netbsd-src/sys/net/if_gif.c (revision 63aea4bd5b445e491ff0389fe27ec78b3099dba3)
1 /*	$NetBSD: if_gif.c,v 1.102 2015/12/11 07:59:14 knakahara Exp $	*/
2 /*	$KAME: if_gif.c,v 1.76 2001/08/20 02:01:02 kjc 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 #include <sys/cdefs.h>
34 __KERNEL_RCSID(0, "$NetBSD: if_gif.c,v 1.102 2015/12/11 07:59:14 knakahara Exp $");
35 
36 #ifdef _KERNEL_OPT
37 #include "opt_inet.h"
38 #endif
39 
40 #include <sys/param.h>
41 #include <sys/systm.h>
42 #include <sys/kernel.h>
43 #include <sys/mbuf.h>
44 #include <sys/socket.h>
45 #include <sys/sockio.h>
46 #include <sys/errno.h>
47 #include <sys/ioctl.h>
48 #include <sys/time.h>
49 #include <sys/socketvar.h>
50 #include <sys/syslog.h>
51 #include <sys/proc.h>
52 #include <sys/protosw.h>
53 #include <sys/cpu.h>
54 #include <sys/intr.h>
55 #include <sys/kmem.h>
56 #include <sys/atomic.h>
57 
58 #include <net/if.h>
59 #include <net/if_types.h>
60 #include <net/netisr.h>
61 #include <net/route.h>
62 #include <net/bpf.h>
63 
64 #include <netinet/in.h>
65 #include <netinet/in_systm.h>
66 #include <netinet/ip.h>
67 #ifdef	INET
68 #include <netinet/in_var.h>
69 #endif	/* INET */
70 #include <netinet/in_gif.h>
71 
72 #ifdef INET6
73 #ifndef INET
74 #include <netinet/in.h>
75 #endif
76 #include <netinet6/in6_var.h>
77 #include <netinet/ip6.h>
78 #include <netinet6/ip6_var.h>
79 #include <netinet6/in6_gif.h>
80 #include <netinet6/ip6protosw.h>
81 #endif /* INET6 */
82 
83 #include <netinet/ip_encap.h>
84 #include <net/if_gif.h>
85 
86 #include <net/net_osdep.h>
87 
88 #include "ioconf.h"
89 
90 static void	gifintr(void *);
91 
92 /*
93  * gif global variable definitions
94  */
95 LIST_HEAD(, gif_softc) gif_softc_list;	/* XXX should be static */
96 
97 static int	gif_clone_create(struct if_clone *, int);
98 static int	gif_clone_destroy(struct ifnet *);
99 static int	gif_check_nesting(struct ifnet *, struct mbuf *);
100 
101 static int	gif_encap_attach(struct gif_softc *);
102 static int	gif_encap_detach(struct gif_softc *);
103 
104 static struct if_clone gif_cloner =
105     IF_CLONE_INITIALIZER("gif", gif_clone_create, gif_clone_destroy);
106 
107 #ifndef MAX_GIF_NEST
108 /*
109  * This macro controls the upper limitation on nesting of gif tunnels.
110  * Since, setting a large value to this macro with a careless configuration
111  * may introduce system crash, we don't allow any nestings by default.
112  * If you need to configure nested gif tunnels, you can define this macro
113  * in your kernel configuration file.  However, if you do so, please be
114  * careful to configure the tunnels so that it won't make a loop.
115  */
116 #define MAX_GIF_NEST 1
117 #endif
118 static int max_gif_nesting = MAX_GIF_NEST;
119 
120 /* ARGSUSED */
121 void
122 gifattach(int count)
123 {
124 
125 	LIST_INIT(&gif_softc_list);
126 	if_clone_attach(&gif_cloner);
127 }
128 
129 static int
130 gif_clone_create(struct if_clone *ifc, int unit)
131 {
132 	struct gif_softc *sc;
133 
134 	sc = kmem_zalloc(sizeof(struct gif_softc), KM_SLEEP);
135 	if (sc == NULL)
136 		return ENOMEM;
137 
138 	if_initname(&sc->gif_if, ifc->ifc_name, unit);
139 
140 	gifattach0(sc);
141 
142 	LIST_INSERT_HEAD(&gif_softc_list, sc, gif_list);
143 	return (0);
144 }
145 
146 void
147 gifattach0(struct gif_softc *sc)
148 {
149 
150 	sc->gif_si_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE);
151 	KASSERT(sc->gif_si_lock != NULL);
152 	cv_init(&sc->gif_si_cv, "if_gif_cv");
153 	sc->gif_si_refs = 0;
154 	sc->encap_cookie4 = sc->encap_cookie6 = NULL;
155 
156 	sc->gif_if.if_addrlen = 0;
157 	sc->gif_if.if_mtu    = GIF_MTU;
158 	sc->gif_if.if_flags  = IFF_POINTOPOINT | IFF_MULTICAST;
159 	sc->gif_if.if_ioctl  = gif_ioctl;
160 	sc->gif_if.if_output = gif_output;
161 	sc->gif_if.if_type   = IFT_GIF;
162 	sc->gif_if.if_dlt    = DLT_NULL;
163 	sc->gif_if.if_softc  = sc;
164 	IFQ_SET_READY(&sc->gif_if.if_snd);
165 	if_attach(&sc->gif_if);
166 	if_alloc_sadl(&sc->gif_if);
167 	bpf_attach(&sc->gif_if, DLT_NULL, sizeof(u_int));
168 }
169 
170 static int
171 gif_clone_destroy(struct ifnet *ifp)
172 {
173 	struct gif_softc *sc = (void *) ifp;
174 
175 	LIST_REMOVE(sc, gif_list);
176 
177 	gif_delete_tunnel(&sc->gif_if);
178 	bpf_detach(ifp);
179 	if_detach(ifp);
180 	rtcache_free(&sc->gif_ro);
181 
182 	cv_destroy(&sc->gif_si_cv);
183 	mutex_obj_free(sc->gif_si_lock);
184 	kmem_free(sc, sizeof(struct gif_softc));
185 
186 	return (0);
187 }
188 
189 #ifdef GIF_ENCAPCHECK
190 int
191 gif_encapcheck(struct mbuf *m, int off, int proto, void *arg)
192 {
193 	struct ip ip;
194 	struct gif_softc *sc;
195 
196 	sc = arg;
197 	if (sc == NULL)
198 		return 0;
199 
200 	if ((sc->gif_if.if_flags & IFF_UP) == 0)
201 		return 0;
202 
203 	/* no physical address */
204 	if (!sc->gif_psrc || !sc->gif_pdst)
205 		return 0;
206 
207 	switch (proto) {
208 #ifdef INET
209 	case IPPROTO_IPV4:
210 		break;
211 #endif
212 #ifdef INET6
213 	case IPPROTO_IPV6:
214 		break;
215 #endif
216 	default:
217 		return 0;
218 	}
219 
220 	/* Bail on short packets */
221 	KASSERT(m->m_flags & M_PKTHDR);
222 	if (m->m_pkthdr.len < sizeof(ip))
223 		return 0;
224 
225 	m_copydata(m, 0, sizeof(ip), &ip);
226 
227 	switch (ip.ip_v) {
228 #ifdef INET
229 	case 4:
230 		if (sc->gif_psrc->sa_family != AF_INET ||
231 		    sc->gif_pdst->sa_family != AF_INET)
232 			return 0;
233 		return gif_encapcheck4(m, off, proto, arg);
234 #endif
235 #ifdef INET6
236 	case 6:
237 		if (m->m_pkthdr.len < sizeof(struct ip6_hdr))
238 			return 0;
239 		if (sc->gif_psrc->sa_family != AF_INET6 ||
240 		    sc->gif_pdst->sa_family != AF_INET6)
241 			return 0;
242 		return gif_encapcheck6(m, off, proto, arg);
243 #endif
244 	default:
245 		return 0;
246 	}
247 }
248 #endif
249 
250 /*
251  * gif may cause infinite recursion calls when misconfigured.
252  * We'll prevent this by introducing upper limit.
253  */
254 static int
255 gif_check_nesting(struct ifnet *ifp, struct mbuf *m)
256 {
257 	struct m_tag *mtag;
258 	int *count;
259 
260 	mtag = m_tag_find(m, PACKET_TAG_TUNNEL_INFO, NULL);
261 	if (mtag != NULL) {
262 		count = (int *)(mtag + 1);
263 		if (++(*count) > max_gif_nesting) {
264 			log(LOG_NOTICE,
265 			    "%s: recursively called too many times(%d)\n",
266 			    if_name(ifp),
267 			    *count);
268 			return EIO;
269 		}
270 	} else {
271 		mtag = m_tag_get(PACKET_TAG_TUNNEL_INFO, sizeof(*count),
272 		    M_NOWAIT);
273 		if (mtag != NULL) {
274 			m_tag_prepend(m, mtag);
275 			count = (int *)(mtag + 1);
276 			*count = 0;
277 		} else {
278 			log(LOG_DEBUG,
279 			    "%s: m_tag_get() failed, recursion calls are not prevented.\n",
280 			    if_name(ifp));
281 		}
282 	}
283 
284 	return 0;
285 }
286 
287 int
288 gif_output(struct ifnet *ifp, struct mbuf *m, const struct sockaddr *dst,
289     struct rtentry *rt)
290 {
291 	struct gif_softc *sc = ifp->if_softc;
292 	int error = 0;
293 	ALTQ_DECL(struct altq_pktattr pktattr;)
294 	int s;
295 
296 	IFQ_CLASSIFY(&ifp->if_snd, m, dst->sa_family, &pktattr);
297 
298 	if ((error = gif_check_nesting(ifp, m)) != 0) {
299 		m_free(m);
300 		goto end;
301 	}
302 
303 	m->m_flags &= ~(M_BCAST|M_MCAST);
304 	if (!(ifp->if_flags & IFF_UP) ||
305 	    sc->gif_psrc == NULL || sc->gif_pdst == NULL ||
306 	    sc->gif_si == NULL) {
307 		m_freem(m);
308 		error = ENETDOWN;
309 		goto end;
310 	}
311 
312 	/* XXX should we check if our outer source is legal? */
313 
314 	/* use DLT_NULL encapsulation here to pass inner af type */
315 	M_PREPEND(m, sizeof(int), M_DONTWAIT);
316 	if (!m) {
317 		error = ENOBUFS;
318 		goto end;
319 	}
320 	*mtod(m, int *) = dst->sa_family;
321 
322 	/* Clear checksum-offload flags. */
323 	m->m_pkthdr.csum_flags = 0;
324 	m->m_pkthdr.csum_data = 0;
325 
326 	s = splnet();
327 	IFQ_ENQUEUE(&ifp->if_snd, m, &pktattr, error);
328 	if (error) {
329 		splx(s);
330 		goto end;
331 	}
332 
333 	/* softint_schedule() must be called with kpreempt_disabled() */
334 	softint_schedule(sc->gif_si);
335 	splx(s);
336 
337 	error = 0;
338 
339   end:
340 	if (error)
341 		ifp->if_oerrors++;
342 	return error;
343 }
344 
345 static void
346 gifintr(void *arg)
347 {
348 	struct gif_softc *sc;
349 	struct ifnet *ifp;
350 	struct mbuf *m;
351 	int family;
352 	int len;
353 	int s;
354 	int error;
355 
356 	sc = arg;
357 	ifp = &sc->gif_if;
358 
359 	atomic_inc_uint(&sc->gif_si_refs);
360 
361 	/*
362 	 * pattern (a) (see also gif_set_tunnel())
363 	 * other CPUs does {set,delete}_tunnel after curcpu have done
364 	 * softint_schedule().
365 	 */
366 	if (sc->gif_pdst == NULL || sc->gif_psrc == NULL) {
367 		IFQ_PURGE(&ifp->if_snd);
368 
369 		if (atomic_dec_uint_nv(&sc->gif_si_refs) == 0) {
370 			mutex_enter(sc->gif_si_lock);
371 			cv_broadcast(&sc->gif_si_cv);
372 			mutex_exit(sc->gif_si_lock);
373 		}
374 		return;
375 	}
376 
377 	/* output processing */
378 	while (1) {
379 		s = splnet();
380 		IFQ_DEQUEUE(&sc->gif_if.if_snd, m);
381 		splx(s);
382 		if (m == NULL)
383 			break;
384 
385 		/* grab and chop off inner af type */
386 		if (sizeof(int) > m->m_len) {
387 			m = m_pullup(m, sizeof(int));
388 			if (!m) {
389 				ifp->if_oerrors++;
390 				continue;
391 			}
392 		}
393 		family = *mtod(m, int *);
394 		bpf_mtap(ifp, m);
395 		m_adj(m, sizeof(int));
396 
397 		len = m->m_pkthdr.len;
398 
399 		/* dispatch to output logic based on outer AF */
400 		switch (sc->gif_psrc->sa_family) {
401 #ifdef INET
402 		case AF_INET:
403 			mutex_enter(softnet_lock);
404 			error = in_gif_output(ifp, family, m);
405 			mutex_exit(softnet_lock);
406 			break;
407 #endif
408 #ifdef INET6
409 		case AF_INET6:
410 			mutex_enter(softnet_lock);
411 			error = in6_gif_output(ifp, family, m);
412 			mutex_exit(softnet_lock);
413 			break;
414 #endif
415 		default:
416 			m_freem(m);
417 			error = ENETDOWN;
418 			break;
419 		}
420 
421 		if (error)
422 			ifp->if_oerrors++;
423 		else {
424 			ifp->if_opackets++;
425 			ifp->if_obytes += len;
426 		}
427 	}
428 
429 	/*
430 	 * pattern (b) (see also gif_set_tunnel())
431 	 * other CPUs begin {set,delete}_tunnel while curcpu si doing gifintr.
432 	 */
433 	if (atomic_dec_uint_nv(&sc->gif_si_refs) == 0) {
434 		mutex_enter(sc->gif_si_lock);
435 		cv_broadcast(&sc->gif_si_cv);
436 		mutex_exit(sc->gif_si_lock);
437 	}
438 }
439 
440 void
441 gif_input(struct mbuf *m, int af, struct ifnet *ifp)
442 {
443 	pktqueue_t *pktq;
444 	size_t pktlen;
445 	int s;
446 
447 	if (ifp == NULL) {
448 		/* just in case */
449 		m_freem(m);
450 		return;
451 	}
452 
453 	m->m_pkthdr.rcvif = ifp;
454 	pktlen = m->m_pkthdr.len;
455 
456 	bpf_mtap_af(ifp, af, m);
457 
458 	/*
459 	 * Put the packet to the network layer input queue according to the
460 	 * specified address family.  Note: we avoid direct call to the
461 	 * input function of the network layer in order to avoid recursion.
462 	 * This may be revisited in the future.
463 	 */
464 	switch (af) {
465 #ifdef INET
466 	case AF_INET:
467 		pktq = ip_pktq;
468 		break;
469 #endif
470 #ifdef INET6
471 	case AF_INET6:
472 		pktq = ip6_pktq;
473 		break;
474 #endif
475 	default:
476 		m_freem(m);
477 		return;
478 	}
479 
480 	s = splnet();
481 	if (__predict_true(pktq_enqueue(pktq, m, 0))) {
482 		ifp->if_ibytes += pktlen;
483 		ifp->if_ipackets++;
484 	} else {
485 		m_freem(m);
486 	}
487 	splx(s);
488 }
489 
490 /* XXX how should we handle IPv6 scope on SIOC[GS]IFPHYADDR? */
491 int
492 gif_ioctl(struct ifnet *ifp, u_long cmd, void *data)
493 {
494 	struct gif_softc *sc  = ifp->if_softc;
495 	struct ifreq     *ifr = (struct ifreq*)data;
496 	struct ifaddr    *ifa = (struct ifaddr*)data;
497 	int error = 0, size;
498 	struct sockaddr *dst, *src;
499 
500 	switch (cmd) {
501 	case SIOCINITIFADDR:
502 		ifp->if_flags |= IFF_UP;
503 		ifa->ifa_rtrequest = p2p_rtrequest;
504 		break;
505 
506 	case SIOCADDMULTI:
507 	case SIOCDELMULTI:
508 		switch (ifr->ifr_addr.sa_family) {
509 #ifdef INET
510 		case AF_INET:	/* IP supports Multicast */
511 			break;
512 #endif /* INET */
513 #ifdef INET6
514 		case AF_INET6:	/* IP6 supports Multicast */
515 			break;
516 #endif /* INET6 */
517 		default:  /* Other protocols doesn't support Multicast */
518 			error = EAFNOSUPPORT;
519 			break;
520 		}
521 		break;
522 
523 	case SIOCSIFMTU:
524 		if (ifr->ifr_mtu < GIF_MTU_MIN || ifr->ifr_mtu > GIF_MTU_MAX)
525 			return EINVAL;
526 		else if ((error = ifioctl_common(ifp, cmd, data)) == ENETRESET)
527 			error = 0;
528 		break;
529 
530 #ifdef INET
531 	case SIOCSIFPHYADDR:
532 #endif
533 #ifdef INET6
534 	case SIOCSIFPHYADDR_IN6:
535 #endif /* INET6 */
536 	case SIOCSLIFPHYADDR:
537 		switch (cmd) {
538 #ifdef INET
539 		case SIOCSIFPHYADDR:
540 			src = (struct sockaddr *)
541 				&(((struct in_aliasreq *)data)->ifra_addr);
542 			dst = (struct sockaddr *)
543 				&(((struct in_aliasreq *)data)->ifra_dstaddr);
544 			break;
545 #endif
546 #ifdef INET6
547 		case SIOCSIFPHYADDR_IN6:
548 			src = (struct sockaddr *)
549 				&(((struct in6_aliasreq *)data)->ifra_addr);
550 			dst = (struct sockaddr *)
551 				&(((struct in6_aliasreq *)data)->ifra_dstaddr);
552 			break;
553 #endif
554 		case SIOCSLIFPHYADDR:
555 			src = (struct sockaddr *)
556 				&(((struct if_laddrreq *)data)->addr);
557 			dst = (struct sockaddr *)
558 				&(((struct if_laddrreq *)data)->dstaddr);
559 			break;
560 		default:
561 			return EINVAL;
562 		}
563 
564 		/* sa_family must be equal */
565 		if (src->sa_family != dst->sa_family)
566 			return EINVAL;
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 				return EINVAL;
574 			break;
575 #endif
576 #ifdef INET6
577 		case AF_INET6:
578 			if (src->sa_len != sizeof(struct sockaddr_in6))
579 				return EINVAL;
580 			break;
581 #endif
582 		default:
583 			return EAFNOSUPPORT;
584 		}
585 		switch (dst->sa_family) {
586 #ifdef INET
587 		case AF_INET:
588 			if (dst->sa_len != sizeof(struct sockaddr_in))
589 				return EINVAL;
590 			break;
591 #endif
592 #ifdef INET6
593 		case AF_INET6:
594 			if (dst->sa_len != sizeof(struct sockaddr_in6))
595 				return EINVAL;
596 			break;
597 #endif
598 		default:
599 			return EAFNOSUPPORT;
600 		}
601 
602 		/* check sa_family looks sane for the cmd */
603 		switch (cmd) {
604 		case SIOCSIFPHYADDR:
605 			if (src->sa_family == AF_INET)
606 				break;
607 			return EAFNOSUPPORT;
608 #ifdef INET6
609 		case SIOCSIFPHYADDR_IN6:
610 			if (src->sa_family == AF_INET6)
611 				break;
612 			return EAFNOSUPPORT;
613 #endif /* INET6 */
614 		case SIOCSLIFPHYADDR:
615 			/* checks done in the above */
616 			break;
617 		}
618 
619 		error = gif_set_tunnel(&sc->gif_if, src, dst);
620 		break;
621 
622 #ifdef SIOCDIFPHYADDR
623 	case SIOCDIFPHYADDR:
624 		gif_delete_tunnel(&sc->gif_if);
625 		break;
626 #endif
627 
628 	case SIOCGIFPSRCADDR:
629 #ifdef INET6
630 	case SIOCGIFPSRCADDR_IN6:
631 #endif /* INET6 */
632 		if (sc->gif_psrc == NULL) {
633 			error = EADDRNOTAVAIL;
634 			goto bad;
635 		}
636 		src = sc->gif_psrc;
637 		switch (cmd) {
638 #ifdef INET
639 		case SIOCGIFPSRCADDR:
640 			dst = &ifr->ifr_addr;
641 			size = sizeof(ifr->ifr_addr);
642 			break;
643 #endif /* INET */
644 #ifdef INET6
645 		case SIOCGIFPSRCADDR_IN6:
646 			dst = (struct sockaddr *)
647 				&(((struct in6_ifreq *)data)->ifr_addr);
648 			size = sizeof(((struct in6_ifreq *)data)->ifr_addr);
649 			break;
650 #endif /* INET6 */
651 		default:
652 			error = EADDRNOTAVAIL;
653 			goto bad;
654 		}
655 		if (src->sa_len > size)
656 			return EINVAL;
657 		memcpy(dst, src, src->sa_len);
658 		break;
659 
660 	case SIOCGIFPDSTADDR:
661 #ifdef INET6
662 	case SIOCGIFPDSTADDR_IN6:
663 #endif /* INET6 */
664 		if (sc->gif_pdst == NULL) {
665 			error = EADDRNOTAVAIL;
666 			goto bad;
667 		}
668 		src = sc->gif_pdst;
669 		switch (cmd) {
670 #ifdef INET
671 		case SIOCGIFPDSTADDR:
672 			dst = &ifr->ifr_addr;
673 			size = sizeof(ifr->ifr_addr);
674 			break;
675 #endif /* INET */
676 #ifdef INET6
677 		case SIOCGIFPDSTADDR_IN6:
678 			dst = (struct sockaddr *)
679 				&(((struct in6_ifreq *)data)->ifr_addr);
680 			size = sizeof(((struct in6_ifreq *)data)->ifr_addr);
681 			break;
682 #endif /* INET6 */
683 		default:
684 			error = EADDRNOTAVAIL;
685 			goto bad;
686 		}
687 		if (src->sa_len > size)
688 			return EINVAL;
689 		memcpy(dst, src, src->sa_len);
690 		break;
691 
692 	case SIOCGLIFPHYADDR:
693 		if (sc->gif_psrc == NULL || sc->gif_pdst == NULL) {
694 			error = EADDRNOTAVAIL;
695 			goto bad;
696 		}
697 
698 		/* copy src */
699 		src = sc->gif_psrc;
700 		dst = (struct sockaddr *)
701 			&(((struct if_laddrreq *)data)->addr);
702 		size = sizeof(((struct if_laddrreq *)data)->addr);
703 		if (src->sa_len > size)
704 			return EINVAL;
705 		memcpy(dst, src, src->sa_len);
706 
707 		/* copy dst */
708 		src = sc->gif_pdst;
709 		dst = (struct sockaddr *)
710 			&(((struct if_laddrreq *)data)->dstaddr);
711 		size = sizeof(((struct if_laddrreq *)data)->dstaddr);
712 		if (src->sa_len > size)
713 			return EINVAL;
714 		memcpy(dst, src, src->sa_len);
715 		break;
716 
717 	default:
718 		return ifioctl_common(ifp, cmd, data);
719 	}
720  bad:
721 	return error;
722 }
723 
724 static int
725 gif_encap_attach(struct gif_softc *sc)
726 {
727 	int error;
728 
729 	if (sc == NULL || sc->gif_psrc == NULL)
730 		return EINVAL;
731 
732 	switch (sc->gif_psrc->sa_family) {
733 #ifdef INET
734 	case AF_INET:
735 		error = in_gif_attach(sc);
736 		break;
737 #endif
738 #ifdef INET6
739 	case AF_INET6:
740 		error = in6_gif_attach(sc);
741 		break;
742 #endif
743 	default:
744 		error = EINVAL;
745 		break;
746 	}
747 
748 	return error;
749 }
750 
751 static int
752 gif_encap_detach(struct gif_softc *sc)
753 {
754 	int error;
755 
756 	if (sc == NULL || sc->gif_psrc == NULL)
757 		return EINVAL;
758 
759 	switch (sc->gif_psrc->sa_family) {
760 #ifdef INET
761 	case AF_INET:
762 		error = in_gif_detach(sc);
763 		break;
764 #endif
765 #ifdef INET6
766 	case AF_INET6:
767 		error = in6_gif_detach(sc);
768 		break;
769 #endif
770 	default:
771 		error = EINVAL;
772 		break;
773 	}
774 
775 	return error;
776 }
777 
778 int
779 gif_set_tunnel(struct ifnet *ifp, struct sockaddr *src, struct sockaddr *dst)
780 {
781 	struct gif_softc *sc = ifp->if_softc;
782 	struct gif_softc *sc2;
783 	struct sockaddr *osrc, *odst;
784 	struct sockaddr *nsrc, *ndst;
785 	void *osi;
786 	int s;
787 	int error;
788 
789 	s = splsoftnet();
790 
791 	LIST_FOREACH(sc2, &gif_softc_list, gif_list) {
792 		if (sc2 == sc)
793 			continue;
794 		if (!sc2->gif_pdst || !sc2->gif_psrc)
795 			continue;
796 		/* can't configure same pair of address onto two gifs */
797 		if (sockaddr_cmp(sc2->gif_pdst, dst) == 0 &&
798 		    sockaddr_cmp(sc2->gif_psrc, src) == 0) {
799 			/* continue to use the old configureation. */
800 			splx(s);
801 			return EADDRNOTAVAIL;
802 		}
803 
804 		/* XXX both end must be valid? (I mean, not 0.0.0.0) */
805 	}
806 
807 	if ((nsrc = sockaddr_dup(src, M_WAITOK)) == NULL) {
808 		splx(s);
809 		return ENOMEM;
810 	}
811 	if ((ndst = sockaddr_dup(dst, M_WAITOK)) == NULL) {
812 		sockaddr_free(nsrc);
813 		splx(s);
814 		return ENOMEM;
815 	}
816 
817 	/* Firstly, clear old configurations. */
818 	if (sc->gif_si) {
819 		osrc = sc->gif_psrc;
820 		odst = sc->gif_pdst;
821 		osi = sc->gif_si;
822 		sc->gif_psrc = NULL;
823 		sc->gif_pdst = NULL;
824 		sc->gif_si = NULL;
825 
826 		/*
827 		 * At this point, gif_output() does not softint_schedule()
828 		 * any more. However, there are below 2 fears of other CPUs.
829 		 *     (a) gif_output() has done softint_schedule(),and softint
830 		 *         (gifintr()) is waiting for execution
831 		 *     (b) gifintr() is already running
832 		 * see also gifintr()
833 		 */
834 
835 		/*
836 		 * To avoid the above fears, wait for gifintr() completion of
837 		 * all CPUs here.
838 		 */
839 		mutex_enter(sc->gif_si_lock);
840 		while (sc->gif_si_refs > 0) {
841 			aprint_debug("%s: cv_wait on gif_softc\n", __func__);
842 			cv_wait(&sc->gif_si_cv, sc->gif_si_lock);
843 		}
844 		mutex_exit(sc->gif_si_lock);
845 
846 		softint_disestablish(osi);
847 		sc->gif_psrc = osrc;
848 		sc->gif_pdst = odst;
849 		osrc = NULL;
850 		odst = NULL;
851 	}
852 	/* XXX we can detach from both, but be polite just in case */
853 	if (sc->gif_psrc)
854 		(void)gif_encap_detach(sc);
855 
856 	/*
857 	 * Secondly, try to set new configurations.
858 	 * If the setup failed, rollback to old configurations.
859 	 */
860 	do {
861 		osrc = sc->gif_psrc;
862 		odst = sc->gif_pdst;
863 		sc->gif_psrc = nsrc;
864 		sc->gif_pdst = ndst;
865 
866 		error = gif_encap_attach(sc);
867 		if (error) {
868 			/* rollback to the last configuration. */
869 			nsrc = osrc;
870 			ndst = odst;
871 			osrc = sc->gif_psrc;
872 			odst = sc->gif_pdst;
873 
874 			continue;
875 		}
876 
877 		sc->gif_si = softint_establish(SOFTINT_NET, gifintr, sc);
878 		if (sc->gif_si == NULL) {
879 			(void)gif_encap_detach(sc);
880 
881 			/* rollback to the last configuration. */
882 			nsrc = osrc;
883 			ndst = odst;
884 			osrc = sc->gif_psrc;
885 			odst = sc->gif_pdst;
886 
887 			error = ENOMEM;
888 			continue;
889 		}
890 	} while (error != 0 && (nsrc != NULL && ndst != NULL));
891 	/* Thirdly, even rollback failed, clear configurations. */
892 	if (error) {
893 		osrc = sc->gif_psrc;
894 		odst = sc->gif_pdst;
895 		sc->gif_psrc = NULL;
896 		sc->gif_pdst = NULL;
897 	}
898 
899 	if (osrc)
900 		sockaddr_free(osrc);
901 	if (odst)
902 		sockaddr_free(odst);
903 
904 	if (sc->gif_psrc && sc->gif_pdst)
905 		ifp->if_flags |= IFF_RUNNING;
906 	else
907 		ifp->if_flags &= ~IFF_RUNNING;
908 
909 	splx(s);
910 	return error;
911 }
912 
913 void
914 gif_delete_tunnel(struct ifnet *ifp)
915 {
916 	struct gif_softc *sc = ifp->if_softc;
917 	struct sockaddr *osrc, *odst;
918 	void *osi;
919 	int s;
920 
921 	s = splsoftnet();
922 
923 	if (sc->gif_si) {
924 		osrc = sc->gif_psrc;
925 		odst = sc->gif_pdst;
926 		osi = sc->gif_si;
927 
928 		sc->gif_psrc = NULL;
929 		sc->gif_pdst = NULL;
930 		sc->gif_si = NULL;
931 
932 		mutex_enter(sc->gif_si_lock);
933 		while (sc->gif_si_refs > 0) {
934 			aprint_debug("%s: cv_wait on gif_softc\n", __func__);
935 			cv_wait(&sc->gif_si_cv, sc->gif_si_lock);
936 		}
937 		mutex_exit(sc->gif_si_lock);
938 
939 		softint_disestablish(osi);
940 		sc->gif_psrc = osrc;
941 		sc->gif_pdst = odst;
942 	}
943 	if (sc->gif_psrc) {
944 		sockaddr_free(sc->gif_psrc);
945 		sc->gif_psrc = NULL;
946 	}
947 	if (sc->gif_pdst) {
948 		sockaddr_free(sc->gif_pdst);
949 		sc->gif_pdst = NULL;
950 	}
951 	/* it is safe to detach from both */
952 #ifdef INET
953 	(void)in_gif_detach(sc);
954 #endif
955 #ifdef INET6
956 	(void)in6_gif_detach(sc);
957 #endif
958 
959 	if (sc->gif_psrc && sc->gif_pdst)
960 		ifp->if_flags |= IFF_RUNNING;
961 	else
962 		ifp->if_flags &= ~IFF_RUNNING;
963 	splx(s);
964 }
965