xref: /freebsd/sys/net/if_gif.c (revision 721351876cd4d3a8a700f62d2061331fa951a488)
1 /*	$FreeBSD$	*/
2 /*	$KAME: if_gif.c,v 1.87 2001/10/19 08:50:27 itojun 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 "opt_inet.h"
34 #include "opt_inet6.h"
35 #include "opt_mac.h"
36 
37 #include <sys/param.h>
38 #include <sys/systm.h>
39 #include <sys/kernel.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/errno.h>
46 #include <sys/time.h>
47 #include <sys/sysctl.h>
48 #include <sys/syslog.h>
49 #include <sys/proc.h>
50 #include <sys/protosw.h>
51 #include <sys/conf.h>
52 #include <machine/cpu.h>
53 
54 #include <net/if.h>
55 #include <net/if_clone.h>
56 #include <net/if_types.h>
57 #include <net/netisr.h>
58 #include <net/route.h>
59 #include <net/bpf.h>
60 
61 #include <netinet/in.h>
62 #include <netinet/in_systm.h>
63 #include <netinet/ip.h>
64 #ifdef	INET
65 #include <netinet/in_var.h>
66 #include <netinet/in_gif.h>
67 #include <netinet/ip_var.h>
68 #endif	/* INET */
69 
70 #ifdef INET6
71 #ifndef INET
72 #include <netinet/in.h>
73 #endif
74 #include <netinet6/in6_var.h>
75 #include <netinet/ip6.h>
76 #include <netinet6/ip6_var.h>
77 #include <netinet6/scope6_var.h>
78 #include <netinet6/in6_gif.h>
79 #include <netinet6/ip6protosw.h>
80 #endif /* INET6 */
81 
82 #include <netinet/ip_encap.h>
83 #include <net/ethernet.h>
84 #include <net/if_bridgevar.h>
85 #include <net/if_gif.h>
86 
87 #include <security/mac/mac_framework.h>
88 
89 #define GIFNAME		"gif"
90 
91 /*
92  * gif_mtx protects the global gif_softc_list.
93  */
94 static struct mtx gif_mtx;
95 static MALLOC_DEFINE(M_GIF, "gif", "Generic Tunnel Interface");
96 static LIST_HEAD(, gif_softc) gif_softc_list;
97 
98 void	(*ng_gif_input_p)(struct ifnet *ifp, struct mbuf **mp, int af);
99 void	(*ng_gif_input_orphan_p)(struct ifnet *ifp, struct mbuf *m, int af);
100 void	(*ng_gif_attach_p)(struct ifnet *ifp);
101 void	(*ng_gif_detach_p)(struct ifnet *ifp);
102 
103 static void	gif_start(struct ifnet *);
104 static int	gif_clone_create(struct if_clone *, int, caddr_t);
105 static void	gif_clone_destroy(struct ifnet *);
106 
107 IFC_SIMPLE_DECLARE(gif, 0);
108 
109 static int gifmodevent(module_t, int, void *);
110 
111 SYSCTL_DECL(_net_link);
112 SYSCTL_NODE(_net_link, IFT_GIF, gif, CTLFLAG_RW, 0,
113     "Generic Tunnel Interface");
114 #ifndef MAX_GIF_NEST
115 /*
116  * This macro controls the default upper limitation on nesting of gif tunnels.
117  * Since, setting a large value to this macro with a careless configuration
118  * may introduce system crash, we don't allow any nestings by default.
119  * If you need to configure nested gif tunnels, you can define this macro
120  * in your kernel configuration file.  However, if you do so, please be
121  * careful to configure the tunnels so that it won't make a loop.
122  */
123 #define MAX_GIF_NEST 1
124 #endif
125 static int max_gif_nesting = MAX_GIF_NEST;
126 SYSCTL_INT(_net_link_gif, OID_AUTO, max_nesting, CTLFLAG_RW,
127     &max_gif_nesting, 0, "Max nested tunnels");
128 
129 /*
130  * By default, we disallow creation of multiple tunnels between the same
131  * pair of addresses.  Some applications require this functionality so
132  * we allow control over this check here.
133  */
134 #ifdef XBONEHACK
135 static int parallel_tunnels = 1;
136 #else
137 static int parallel_tunnels = 0;
138 #endif
139 SYSCTL_INT(_net_link_gif, OID_AUTO, parallel_tunnels, CTLFLAG_RW,
140     &parallel_tunnels, 0, "Allow parallel tunnels?");
141 
142 /* copy from src/sys/net/if_ethersubr.c */
143 static const u_char etherbroadcastaddr[ETHER_ADDR_LEN] =
144 			{ 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
145 #ifndef ETHER_IS_BROADCAST
146 #define ETHER_IS_BROADCAST(addr) \
147 	(bcmp(etherbroadcastaddr, (addr), ETHER_ADDR_LEN) == 0)
148 #endif
149 
150 static int
151 gif_clone_create(ifc, unit, params)
152 	struct if_clone *ifc;
153 	int unit;
154 	caddr_t params;
155 {
156 	struct gif_softc *sc;
157 
158 	sc = malloc(sizeof(struct gif_softc), M_GIF, M_WAITOK | M_ZERO);
159 	sc->gif_fibnum = curthread->td_proc->p_fibnum;
160 	GIF2IFP(sc) = if_alloc(IFT_GIF);
161 	if (GIF2IFP(sc) == NULL) {
162 		free(sc, M_GIF);
163 		return (ENOSPC);
164 	}
165 
166 	GIF_LOCK_INIT(sc);
167 
168 	GIF2IFP(sc)->if_softc = sc;
169 	if_initname(GIF2IFP(sc), ifc->ifc_name, unit);
170 
171 	sc->encap_cookie4 = sc->encap_cookie6 = NULL;
172 
173 	GIF2IFP(sc)->if_addrlen = 0;
174 	GIF2IFP(sc)->if_mtu    = GIF_MTU;
175 	GIF2IFP(sc)->if_flags  = IFF_POINTOPOINT | IFF_MULTICAST;
176 #if 0
177 	/* turn off ingress filter */
178 	GIF2IFP(sc)->if_flags  |= IFF_LINK2;
179 #endif
180 	GIF2IFP(sc)->if_ioctl  = gif_ioctl;
181 	GIF2IFP(sc)->if_start  = gif_start;
182 	GIF2IFP(sc)->if_output = gif_output;
183 	GIF2IFP(sc)->if_snd.ifq_maxlen = IFQ_MAXLEN;
184 	if_attach(GIF2IFP(sc));
185 	bpfattach(GIF2IFP(sc), DLT_NULL, sizeof(u_int32_t));
186 	if (ng_gif_attach_p != NULL)
187 		(*ng_gif_attach_p)(GIF2IFP(sc));
188 
189 	mtx_lock(&gif_mtx);
190 	LIST_INSERT_HEAD(&gif_softc_list, sc, gif_list);
191 	mtx_unlock(&gif_mtx);
192 
193 	return (0);
194 }
195 
196 static void
197 gif_clone_destroy(ifp)
198 	struct ifnet *ifp;
199 {
200 	int err;
201 	struct gif_softc *sc = ifp->if_softc;
202 
203 	mtx_lock(&gif_mtx);
204 	LIST_REMOVE(sc, gif_list);
205 	mtx_unlock(&gif_mtx);
206 
207 	gif_delete_tunnel(ifp);
208 #ifdef INET6
209 	if (sc->encap_cookie6 != NULL) {
210 		err = encap_detach(sc->encap_cookie6);
211 		KASSERT(err == 0, ("Unexpected error detaching encap_cookie6"));
212 	}
213 #endif
214 #ifdef INET
215 	if (sc->encap_cookie4 != NULL) {
216 		err = encap_detach(sc->encap_cookie4);
217 		KASSERT(err == 0, ("Unexpected error detaching encap_cookie4"));
218 	}
219 #endif
220 
221 	if (ng_gif_detach_p != NULL)
222 		(*ng_gif_detach_p)(ifp);
223 	bpfdetach(ifp);
224 	if_detach(ifp);
225 	if_free(ifp);
226 
227 	GIF_LOCK_DESTROY(sc);
228 
229 	free(sc, M_GIF);
230 }
231 
232 static int
233 gifmodevent(mod, type, data)
234 	module_t mod;
235 	int type;
236 	void *data;
237 {
238 
239 	switch (type) {
240 	case MOD_LOAD:
241 		mtx_init(&gif_mtx, "gif_mtx", NULL, MTX_DEF);
242 		LIST_INIT(&gif_softc_list);
243 		if_clone_attach(&gif_cloner);
244 
245 #ifdef INET6
246 		ip6_gif_hlim = GIF_HLIM;
247 #endif
248 
249 		break;
250 	case MOD_UNLOAD:
251 		if_clone_detach(&gif_cloner);
252 		mtx_destroy(&gif_mtx);
253 #ifdef INET6
254 		ip6_gif_hlim = 0;
255 #endif
256 		break;
257 	default:
258 		return EOPNOTSUPP;
259 	}
260 	return 0;
261 }
262 
263 static moduledata_t gif_mod = {
264 	"if_gif",
265 	gifmodevent,
266 	0
267 };
268 
269 DECLARE_MODULE(if_gif, gif_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
270 MODULE_VERSION(if_gif, 1);
271 
272 int
273 gif_encapcheck(m, off, proto, arg)
274 	const struct mbuf *m;
275 	int off;
276 	int proto;
277 	void *arg;
278 {
279 	struct ip ip;
280 	struct gif_softc *sc;
281 
282 	sc = (struct gif_softc *)arg;
283 	if (sc == NULL)
284 		return 0;
285 
286 	if ((GIF2IFP(sc)->if_flags & IFF_UP) == 0)
287 		return 0;
288 
289 	/* no physical address */
290 	if (!sc->gif_psrc || !sc->gif_pdst)
291 		return 0;
292 
293 	switch (proto) {
294 #ifdef INET
295 	case IPPROTO_IPV4:
296 		break;
297 #endif
298 #ifdef INET6
299 	case IPPROTO_IPV6:
300 		break;
301 #endif
302 	case IPPROTO_ETHERIP:
303 		break;
304 
305 	default:
306 		return 0;
307 	}
308 
309 	/* Bail on short packets */
310 	if (m->m_pkthdr.len < sizeof(ip))
311 		return 0;
312 
313 	m_copydata(m, 0, sizeof(ip), (caddr_t)&ip);
314 
315 	switch (ip.ip_v) {
316 #ifdef INET
317 	case 4:
318 		if (sc->gif_psrc->sa_family != AF_INET ||
319 		    sc->gif_pdst->sa_family != AF_INET)
320 			return 0;
321 		return gif_encapcheck4(m, off, proto, arg);
322 #endif
323 #ifdef INET6
324 	case 6:
325 		if (m->m_pkthdr.len < sizeof(struct ip6_hdr))
326 			return 0;
327 		if (sc->gif_psrc->sa_family != AF_INET6 ||
328 		    sc->gif_pdst->sa_family != AF_INET6)
329 			return 0;
330 		return gif_encapcheck6(m, off, proto, arg);
331 #endif
332 	default:
333 		return 0;
334 	}
335 }
336 
337 static void
338 gif_start(struct ifnet *ifp)
339 {
340 	struct gif_softc *sc;
341 	struct mbuf *m;
342 
343 	sc = ifp->if_softc;
344 
345 	ifp->if_drv_flags |= IFF_DRV_OACTIVE;
346 	for (;;) {
347 		IFQ_DEQUEUE(&ifp->if_snd, m);
348 		if (m == 0)
349 			break;
350 
351 		gif_output(ifp, m, sc->gif_pdst, NULL);
352 
353 	}
354 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
355 
356 	return;
357 }
358 
359 int
360 gif_output(ifp, m, dst, rt)
361 	struct ifnet *ifp;
362 	struct mbuf *m;
363 	struct sockaddr *dst;
364 	struct rtentry *rt;	/* added in net2 */
365 {
366 	struct gif_softc *sc = ifp->if_softc;
367 	struct m_tag *mtag;
368 	int error = 0;
369 	int gif_called;
370 	u_int32_t af;
371 
372 #ifdef MAC
373 	error = mac_ifnet_check_transmit(ifp, m);
374 	if (error) {
375 		m_freem(m);
376 		goto end;
377 	}
378 #endif
379 
380 	/*
381 	 * gif may cause infinite recursion calls when misconfigured.
382 	 * We'll prevent this by detecting loops.
383 	 *
384 	 * High nesting level may cause stack exhaustion.
385 	 * We'll prevent this by introducing upper limit.
386 	 */
387 	gif_called = 1;
388 	mtag = m_tag_locate(m, MTAG_GIF, MTAG_GIF_CALLED, NULL);
389 	while (mtag != NULL) {
390 		if (*(struct ifnet **)(mtag + 1) == ifp) {
391 			log(LOG_NOTICE,
392 			    "gif_output: loop detected on %s\n",
393 			    (*(struct ifnet **)(mtag + 1))->if_xname);
394 			m_freem(m);
395 			error = EIO;	/* is there better errno? */
396 			goto end;
397 		}
398 		mtag = m_tag_locate(m, MTAG_GIF, MTAG_GIF_CALLED, mtag);
399 		gif_called++;
400 	}
401 	if (gif_called > max_gif_nesting) {
402 		log(LOG_NOTICE,
403 		    "gif_output: recursively called too many times(%d)\n",
404 		    gif_called);
405 		m_freem(m);
406 		error = EIO;	/* is there better errno? */
407 		goto end;
408 	}
409 	mtag = m_tag_alloc(MTAG_GIF, MTAG_GIF_CALLED, sizeof(struct ifnet *),
410 	    M_NOWAIT);
411 	if (mtag == NULL) {
412 		m_freem(m);
413 		error = ENOMEM;
414 		goto end;
415 	}
416 	*(struct ifnet **)(mtag + 1) = ifp;
417 	m_tag_prepend(m, mtag);
418 
419 	m->m_flags &= ~(M_BCAST|M_MCAST);
420 
421 	GIF_LOCK(sc);
422 
423 	if (!(ifp->if_flags & IFF_UP) ||
424 	    sc->gif_psrc == NULL || sc->gif_pdst == NULL) {
425 		GIF_UNLOCK(sc);
426 		m_freem(m);
427 		error = ENETDOWN;
428 		goto end;
429 	}
430 
431 	/* BPF writes need to be handled specially. */
432 	if (dst->sa_family == AF_UNSPEC) {
433 		bcopy(dst->sa_data, &af, sizeof(af));
434 		dst->sa_family = af;
435 	}
436 
437 	af = dst->sa_family;
438 	BPF_MTAP2(ifp, &af, sizeof(af), m);
439 	ifp->if_opackets++;
440 	ifp->if_obytes += m->m_pkthdr.len;
441 
442 	/* override to IPPROTO_ETHERIP for bridged traffic */
443 	if (ifp->if_bridge)
444 		af = AF_LINK;
445 
446 	M_SETFIB(m, sc->gif_fibnum);
447 	/* inner AF-specific encapsulation */
448 
449 	/* XXX should we check if our outer source is legal? */
450 
451 	/* dispatch to output logic based on outer AF */
452 	switch (sc->gif_psrc->sa_family) {
453 #ifdef INET
454 	case AF_INET:
455 		error = in_gif_output(ifp, af, m);
456 		break;
457 #endif
458 #ifdef INET6
459 	case AF_INET6:
460 		error = in6_gif_output(ifp, af, m);
461 		break;
462 #endif
463 	default:
464 		m_freem(m);
465 		error = ENETDOWN;
466 	}
467 
468 	GIF_UNLOCK(sc);
469   end:
470 	if (error)
471 		ifp->if_oerrors++;
472 	return (error);
473 }
474 
475 void
476 gif_input(m, af, ifp)
477 	struct mbuf *m;
478 	int af;
479 	struct ifnet *ifp;
480 {
481 	int isr, n;
482 	struct etherip_header *eip;
483 	struct ether_header *eh;
484 	struct ifnet *oldifp;
485 
486 	if (ifp == NULL) {
487 		/* just in case */
488 		m_freem(m);
489 		return;
490 	}
491 
492 	m->m_pkthdr.rcvif = ifp;
493 
494 #ifdef MAC
495 	mac_ifnet_create_mbuf(ifp, m);
496 #endif
497 
498 	if (bpf_peers_present(ifp->if_bpf)) {
499 		u_int32_t af1 = af;
500 		bpf_mtap2(ifp->if_bpf, &af1, sizeof(af1), m);
501 	}
502 
503 	if (ng_gif_input_p != NULL) {
504 		(*ng_gif_input_p)(ifp, &m, af);
505 		if (m == NULL)
506 			return;
507 	}
508 
509 	/*
510 	 * Put the packet to the network layer input queue according to the
511 	 * specified address family.
512 	 * Note: older versions of gif_input directly called network layer
513 	 * input functions, e.g. ip6_input, here.  We changed the policy to
514 	 * prevent too many recursive calls of such input functions, which
515 	 * might cause kernel panic.  But the change may introduce another
516 	 * problem; if the input queue is full, packets are discarded.
517 	 * The kernel stack overflow really happened, and we believed
518 	 * queue-full rarely occurs, so we changed the policy.
519 	 */
520 	switch (af) {
521 #ifdef INET
522 	case AF_INET:
523 		isr = NETISR_IP;
524 		break;
525 #endif
526 #ifdef INET6
527 	case AF_INET6:
528 		isr = NETISR_IPV6;
529 		break;
530 #endif
531 	case AF_LINK:
532 		n = sizeof(struct etherip_header) + sizeof(struct ether_header);
533 		if (n > m->m_len) {
534 			m = m_pullup(m, n);
535 			if (m == NULL) {
536 				ifp->if_ierrors++;
537 				return;
538 			}
539 		}
540 
541 		eip = mtod(m, struct etherip_header *);
542  		if (eip->eip_ver !=
543 		    (ETHERIP_VERSION & ETHERIP_VER_VERS_MASK)) {
544 			/* discard unknown versions */
545 			m_freem(m);
546 			return;
547 		}
548 		m_adj(m, sizeof(struct etherip_header));
549 
550 		m->m_flags &= ~(M_BCAST|M_MCAST);
551 		m->m_pkthdr.rcvif = ifp;
552 
553 		if (ifp->if_bridge) {
554 			oldifp = ifp;
555 			eh = mtod(m, struct ether_header *);
556 			if (ETHER_IS_MULTICAST(eh->ether_dhost)) {
557 				if (ETHER_IS_BROADCAST(eh->ether_dhost))
558 					m->m_flags |= M_BCAST;
559 				else
560 					m->m_flags |= M_MCAST;
561 				ifp->if_imcasts++;
562 			}
563 			BRIDGE_INPUT(ifp, m);
564 
565 			if (m != NULL && ifp != oldifp) {
566 				/*
567 				 * The bridge gave us back itself or one of the
568 				 * members for which the frame is addressed.
569 				 */
570 				ether_demux(ifp, m);
571 				return;
572 			}
573 		}
574 		if (m != NULL)
575 			m_freem(m);
576 		return;
577 
578 	default:
579 		if (ng_gif_input_orphan_p != NULL)
580 			(*ng_gif_input_orphan_p)(ifp, m, af);
581 		else
582 			m_freem(m);
583 		return;
584 	}
585 
586 	ifp->if_ipackets++;
587 	ifp->if_ibytes += m->m_pkthdr.len;
588 	netisr_dispatch(isr, m);
589 }
590 
591 /* XXX how should we handle IPv6 scope on SIOC[GS]IFPHYADDR? */
592 int
593 gif_ioctl(ifp, cmd, data)
594 	struct ifnet *ifp;
595 	u_long cmd;
596 	caddr_t data;
597 {
598 	struct gif_softc *sc  = ifp->if_softc;
599 	struct ifreq     *ifr = (struct ifreq*)data;
600 	int error = 0, size;
601 	struct sockaddr *dst, *src;
602 #ifdef	SIOCSIFMTU /* xxx */
603 	u_long mtu;
604 #endif
605 
606 	switch (cmd) {
607 	case SIOCSIFADDR:
608 		ifp->if_flags |= IFF_UP;
609 		break;
610 
611 	case SIOCSIFDSTADDR:
612 		break;
613 
614 	case SIOCADDMULTI:
615 	case SIOCDELMULTI:
616 		break;
617 
618 #ifdef	SIOCSIFMTU /* xxx */
619 	case SIOCGIFMTU:
620 		break;
621 
622 	case SIOCSIFMTU:
623 		mtu = ifr->ifr_mtu;
624 		if (mtu < GIF_MTU_MIN || mtu > GIF_MTU_MAX)
625 			return (EINVAL);
626 		ifp->if_mtu = mtu;
627 		break;
628 #endif /* SIOCSIFMTU */
629 
630 #ifdef INET
631 	case SIOCSIFPHYADDR:
632 #endif
633 #ifdef INET6
634 	case SIOCSIFPHYADDR_IN6:
635 #endif /* INET6 */
636 	case SIOCSLIFPHYADDR:
637 		switch (cmd) {
638 #ifdef INET
639 		case SIOCSIFPHYADDR:
640 			src = (struct sockaddr *)
641 				&(((struct in_aliasreq *)data)->ifra_addr);
642 			dst = (struct sockaddr *)
643 				&(((struct in_aliasreq *)data)->ifra_dstaddr);
644 			break;
645 #endif
646 #ifdef INET6
647 		case SIOCSIFPHYADDR_IN6:
648 			src = (struct sockaddr *)
649 				&(((struct in6_aliasreq *)data)->ifra_addr);
650 			dst = (struct sockaddr *)
651 				&(((struct in6_aliasreq *)data)->ifra_dstaddr);
652 			break;
653 #endif
654 		case SIOCSLIFPHYADDR:
655 			src = (struct sockaddr *)
656 				&(((struct if_laddrreq *)data)->addr);
657 			dst = (struct sockaddr *)
658 				&(((struct if_laddrreq *)data)->dstaddr);
659 			break;
660 		default:
661 			return EINVAL;
662 		}
663 
664 		/* sa_family must be equal */
665 		if (src->sa_family != dst->sa_family)
666 			return EINVAL;
667 
668 		/* validate sa_len */
669 		switch (src->sa_family) {
670 #ifdef INET
671 		case AF_INET:
672 			if (src->sa_len != sizeof(struct sockaddr_in))
673 				return EINVAL;
674 			break;
675 #endif
676 #ifdef INET6
677 		case AF_INET6:
678 			if (src->sa_len != sizeof(struct sockaddr_in6))
679 				return EINVAL;
680 			break;
681 #endif
682 		default:
683 			return EAFNOSUPPORT;
684 		}
685 		switch (dst->sa_family) {
686 #ifdef INET
687 		case AF_INET:
688 			if (dst->sa_len != sizeof(struct sockaddr_in))
689 				return EINVAL;
690 			break;
691 #endif
692 #ifdef INET6
693 		case AF_INET6:
694 			if (dst->sa_len != sizeof(struct sockaddr_in6))
695 				return EINVAL;
696 			break;
697 #endif
698 		default:
699 			return EAFNOSUPPORT;
700 		}
701 
702 		/* check sa_family looks sane for the cmd */
703 		switch (cmd) {
704 		case SIOCSIFPHYADDR:
705 			if (src->sa_family == AF_INET)
706 				break;
707 			return EAFNOSUPPORT;
708 #ifdef INET6
709 		case SIOCSIFPHYADDR_IN6:
710 			if (src->sa_family == AF_INET6)
711 				break;
712 			return EAFNOSUPPORT;
713 #endif /* INET6 */
714 		case SIOCSLIFPHYADDR:
715 			/* checks done in the above */
716 			break;
717 		}
718 
719 		error = gif_set_tunnel(GIF2IFP(sc), src, dst);
720 		break;
721 
722 #ifdef SIOCDIFPHYADDR
723 	case SIOCDIFPHYADDR:
724 		gif_delete_tunnel(GIF2IFP(sc));
725 		break;
726 #endif
727 
728 	case SIOCGIFPSRCADDR:
729 #ifdef INET6
730 	case SIOCGIFPSRCADDR_IN6:
731 #endif /* INET6 */
732 		if (sc->gif_psrc == NULL) {
733 			error = EADDRNOTAVAIL;
734 			goto bad;
735 		}
736 		src = sc->gif_psrc;
737 		switch (cmd) {
738 #ifdef INET
739 		case SIOCGIFPSRCADDR:
740 			dst = &ifr->ifr_addr;
741 			size = sizeof(ifr->ifr_addr);
742 			break;
743 #endif /* INET */
744 #ifdef INET6
745 		case SIOCGIFPSRCADDR_IN6:
746 			dst = (struct sockaddr *)
747 				&(((struct in6_ifreq *)data)->ifr_addr);
748 			size = sizeof(((struct in6_ifreq *)data)->ifr_addr);
749 			break;
750 #endif /* INET6 */
751 		default:
752 			error = EADDRNOTAVAIL;
753 			goto bad;
754 		}
755 		if (src->sa_len > size)
756 			return EINVAL;
757 		bcopy((caddr_t)src, (caddr_t)dst, src->sa_len);
758 #ifdef INET6
759 		if (dst->sa_family == AF_INET6) {
760 			error = sa6_recoverscope((struct sockaddr_in6 *)dst);
761 			if (error != 0)
762 				return (error);
763 		}
764 #endif
765 		break;
766 
767 	case SIOCGIFPDSTADDR:
768 #ifdef INET6
769 	case SIOCGIFPDSTADDR_IN6:
770 #endif /* INET6 */
771 		if (sc->gif_pdst == NULL) {
772 			error = EADDRNOTAVAIL;
773 			goto bad;
774 		}
775 		src = sc->gif_pdst;
776 		switch (cmd) {
777 #ifdef INET
778 		case SIOCGIFPDSTADDR:
779 			dst = &ifr->ifr_addr;
780 			size = sizeof(ifr->ifr_addr);
781 			break;
782 #endif /* INET */
783 #ifdef INET6
784 		case SIOCGIFPDSTADDR_IN6:
785 			dst = (struct sockaddr *)
786 				&(((struct in6_ifreq *)data)->ifr_addr);
787 			size = sizeof(((struct in6_ifreq *)data)->ifr_addr);
788 			break;
789 #endif /* INET6 */
790 		default:
791 			error = EADDRNOTAVAIL;
792 			goto bad;
793 		}
794 		if (src->sa_len > size)
795 			return EINVAL;
796 		bcopy((caddr_t)src, (caddr_t)dst, src->sa_len);
797 #ifdef INET6
798 		if (dst->sa_family == AF_INET6) {
799 			error = sa6_recoverscope((struct sockaddr_in6 *)dst);
800 			if (error != 0)
801 				return (error);
802 		}
803 #endif
804 		break;
805 
806 	case SIOCGLIFPHYADDR:
807 		if (sc->gif_psrc == NULL || sc->gif_pdst == NULL) {
808 			error = EADDRNOTAVAIL;
809 			goto bad;
810 		}
811 
812 		/* copy src */
813 		src = sc->gif_psrc;
814 		dst = (struct sockaddr *)
815 			&(((struct if_laddrreq *)data)->addr);
816 		size = sizeof(((struct if_laddrreq *)data)->addr);
817 		if (src->sa_len > size)
818 			return EINVAL;
819 		bcopy((caddr_t)src, (caddr_t)dst, src->sa_len);
820 
821 		/* copy dst */
822 		src = sc->gif_pdst;
823 		dst = (struct sockaddr *)
824 			&(((struct if_laddrreq *)data)->dstaddr);
825 		size = sizeof(((struct if_laddrreq *)data)->dstaddr);
826 		if (src->sa_len > size)
827 			return EINVAL;
828 		bcopy((caddr_t)src, (caddr_t)dst, src->sa_len);
829 		break;
830 
831 	case SIOCSIFFLAGS:
832 		/* if_ioctl() takes care of it */
833 		break;
834 
835 	default:
836 		error = EINVAL;
837 		break;
838 	}
839  bad:
840 	return error;
841 }
842 
843 /*
844  * XXXRW: There's a general event-ordering issue here: the code to check
845  * if a given tunnel is already present happens before we perform a
846  * potentially blocking setup of the tunnel.  This code needs to be
847  * re-ordered so that the check and replacement can be atomic using
848  * a mutex.
849  */
850 int
851 gif_set_tunnel(ifp, src, dst)
852 	struct ifnet *ifp;
853 	struct sockaddr *src;
854 	struct sockaddr *dst;
855 {
856 	struct gif_softc *sc = ifp->if_softc;
857 	struct gif_softc *sc2;
858 	struct sockaddr *osrc, *odst, *sa;
859 	int error = 0;
860 
861 	mtx_lock(&gif_mtx);
862 	LIST_FOREACH(sc2, &gif_softc_list, gif_list) {
863 		if (sc2 == sc)
864 			continue;
865 		if (!sc2->gif_pdst || !sc2->gif_psrc)
866 			continue;
867 		if (sc2->gif_pdst->sa_family != dst->sa_family ||
868 		    sc2->gif_pdst->sa_len != dst->sa_len ||
869 		    sc2->gif_psrc->sa_family != src->sa_family ||
870 		    sc2->gif_psrc->sa_len != src->sa_len)
871 			continue;
872 
873 		/*
874 		 * Disallow parallel tunnels unless instructed
875 		 * otherwise.
876 		 */
877 		if (!parallel_tunnels &&
878 		    bcmp(sc2->gif_pdst, dst, dst->sa_len) == 0 &&
879 		    bcmp(sc2->gif_psrc, src, src->sa_len) == 0) {
880 			error = EADDRNOTAVAIL;
881 			mtx_unlock(&gif_mtx);
882 			goto bad;
883 		}
884 
885 		/* XXX both end must be valid? (I mean, not 0.0.0.0) */
886 	}
887 	mtx_unlock(&gif_mtx);
888 
889 	/* XXX we can detach from both, but be polite just in case */
890 	if (sc->gif_psrc)
891 		switch (sc->gif_psrc->sa_family) {
892 #ifdef INET
893 		case AF_INET:
894 			(void)in_gif_detach(sc);
895 			break;
896 #endif
897 #ifdef INET6
898 		case AF_INET6:
899 			(void)in6_gif_detach(sc);
900 			break;
901 #endif
902 		}
903 
904 	osrc = sc->gif_psrc;
905 	sa = (struct sockaddr *)malloc(src->sa_len, M_IFADDR, M_WAITOK);
906 	bcopy((caddr_t)src, (caddr_t)sa, src->sa_len);
907 	sc->gif_psrc = sa;
908 
909 	odst = sc->gif_pdst;
910 	sa = (struct sockaddr *)malloc(dst->sa_len, M_IFADDR, M_WAITOK);
911 	bcopy((caddr_t)dst, (caddr_t)sa, dst->sa_len);
912 	sc->gif_pdst = sa;
913 
914 	switch (sc->gif_psrc->sa_family) {
915 #ifdef INET
916 	case AF_INET:
917 		error = in_gif_attach(sc);
918 		break;
919 #endif
920 #ifdef INET6
921 	case AF_INET6:
922 		/*
923 		 * Check validity of the scope zone ID of the addresses, and
924 		 * convert it into the kernel internal form if necessary.
925 		 */
926 		error = sa6_embedscope((struct sockaddr_in6 *)sc->gif_psrc, 0);
927 		if (error != 0)
928 			break;
929 		error = sa6_embedscope((struct sockaddr_in6 *)sc->gif_pdst, 0);
930 		if (error != 0)
931 			break;
932 		error = in6_gif_attach(sc);
933 		break;
934 #endif
935 	}
936 	if (error) {
937 		/* rollback */
938 		free((caddr_t)sc->gif_psrc, M_IFADDR);
939 		free((caddr_t)sc->gif_pdst, M_IFADDR);
940 		sc->gif_psrc = osrc;
941 		sc->gif_pdst = odst;
942 		goto bad;
943 	}
944 
945 	if (osrc)
946 		free((caddr_t)osrc, M_IFADDR);
947 	if (odst)
948 		free((caddr_t)odst, M_IFADDR);
949 
950  bad:
951 	if (sc->gif_psrc && sc->gif_pdst)
952 		ifp->if_drv_flags |= IFF_DRV_RUNNING;
953 	else
954 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
955 
956 	return error;
957 }
958 
959 void
960 gif_delete_tunnel(ifp)
961 	struct ifnet *ifp;
962 {
963 	struct gif_softc *sc = ifp->if_softc;
964 
965 	if (sc->gif_psrc) {
966 		free((caddr_t)sc->gif_psrc, M_IFADDR);
967 		sc->gif_psrc = NULL;
968 	}
969 	if (sc->gif_pdst) {
970 		free((caddr_t)sc->gif_pdst, M_IFADDR);
971 		sc->gif_pdst = NULL;
972 	}
973 	/* it is safe to detach from both */
974 #ifdef INET
975 	(void)in_gif_detach(sc);
976 #endif
977 #ifdef INET6
978 	(void)in6_gif_detach(sc);
979 #endif
980 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
981 }
982