xref: /freebsd/sys/net/if_gif.c (revision 4a5216a6dc0c3ce4cf5f2d3ee8af0c3ff3402c4f)
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 <sys/vimage.h>
53 #include <machine/cpu.h>
54 
55 #include <net/if.h>
56 #include <net/if_clone.h>
57 #include <net/if_types.h>
58 #include <net/netisr.h>
59 #include <net/route.h>
60 #include <net/bpf.h>
61 
62 #include <netinet/in.h>
63 #include <netinet/in_systm.h>
64 #include <netinet/ip.h>
65 #ifdef	INET
66 #include <netinet/in_var.h>
67 #include <netinet/in_gif.h>
68 #include <netinet/ip_var.h>
69 #endif	/* INET */
70 
71 #ifdef INET6
72 #ifndef INET
73 #include <netinet/in.h>
74 #endif
75 #include <netinet6/in6_var.h>
76 #include <netinet/ip6.h>
77 #include <netinet6/ip6_var.h>
78 #include <netinet6/scope6_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/ethernet.h>
85 #include <net/if_bridgevar.h>
86 #include <net/if_gif.h>
87 
88 #include <security/mac/mac_framework.h>
89 
90 #define GIFNAME		"gif"
91 
92 /*
93  * gif_mtx protects the global gif_softc_list.
94  */
95 static struct mtx gif_mtx;
96 static MALLOC_DEFINE(M_GIF, "gif", "Generic Tunnel Interface");
97 static LIST_HEAD(, gif_softc) gif_softc_list;
98 
99 void	(*ng_gif_input_p)(struct ifnet *ifp, struct mbuf **mp, int af);
100 void	(*ng_gif_input_orphan_p)(struct ifnet *ifp, struct mbuf *m, int af);
101 void	(*ng_gif_attach_p)(struct ifnet *ifp);
102 void	(*ng_gif_detach_p)(struct ifnet *ifp);
103 
104 static void	gif_start(struct ifnet *);
105 static int	gif_clone_create(struct if_clone *, int, caddr_t);
106 static void	gif_clone_destroy(struct ifnet *);
107 
108 IFC_SIMPLE_DECLARE(gif, 0);
109 
110 static int gifmodevent(module_t, int, void *);
111 
112 SYSCTL_DECL(_net_link);
113 SYSCTL_NODE(_net_link, IFT_GIF, gif, CTLFLAG_RW, 0,
114     "Generic Tunnel Interface");
115 #ifndef MAX_GIF_NEST
116 /*
117  * This macro controls the default upper limitation on nesting of gif tunnels.
118  * Since, setting a large value to this macro with a careless configuration
119  * may introduce system crash, we don't allow any nestings by default.
120  * If you need to configure nested gif tunnels, you can define this macro
121  * in your kernel configuration file.  However, if you do so, please be
122  * careful to configure the tunnels so that it won't make a loop.
123  */
124 #define MAX_GIF_NEST 1
125 #endif
126 #ifndef VIMAGE
127 static int max_gif_nesting = MAX_GIF_NEST;
128 #endif
129 SYSCTL_V_INT(V_NET, vnet_gif, _net_link_gif, OID_AUTO, max_nesting,
130     CTLFLAG_RW, max_gif_nesting, 0, "Max nested tunnels");
131 
132 #ifdef INET6
133 SYSCTL_DECL(_net_inet6_ip6);
134 SYSCTL_V_INT(V_NET, vnet_gif, _net_inet6_ip6, IPV6CTL_GIF_HLIM,
135     gifhlim, CTLFLAG_RW, ip6_gif_hlim, 0, "");
136 #endif
137 
138 /*
139  * By default, we disallow creation of multiple tunnels between the same
140  * pair of addresses.  Some applications require this functionality so
141  * we allow control over this check here.
142  */
143 #ifdef XBONEHACK
144 static int parallel_tunnels = 1;
145 #else
146 static int parallel_tunnels = 0;
147 #endif
148 SYSCTL_V_INT(V_NET, vnet_gif, _net_link_gif, OID_AUTO, parallel_tunnels,
149     CTLFLAG_RW, parallel_tunnels, 0, "Allow parallel tunnels?");
150 
151 /* copy from src/sys/net/if_ethersubr.c */
152 static const u_char etherbroadcastaddr[ETHER_ADDR_LEN] =
153 			{ 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
154 #ifndef ETHER_IS_BROADCAST
155 #define ETHER_IS_BROADCAST(addr) \
156 	(bcmp(etherbroadcastaddr, (addr), ETHER_ADDR_LEN) == 0)
157 #endif
158 
159 static int
160 gif_clone_create(ifc, unit, params)
161 	struct if_clone *ifc;
162 	int unit;
163 	caddr_t params;
164 {
165 	INIT_VNET_GIF(curvnet);
166 	struct gif_softc *sc;
167 
168 	sc = malloc(sizeof(struct gif_softc), M_GIF, M_WAITOK | M_ZERO);
169 	sc->gif_fibnum = curthread->td_proc->p_fibnum;
170 	GIF2IFP(sc) = if_alloc(IFT_GIF);
171 	if (GIF2IFP(sc) == NULL) {
172 		free(sc, M_GIF);
173 		return (ENOSPC);
174 	}
175 
176 	GIF_LOCK_INIT(sc);
177 
178 	GIF2IFP(sc)->if_softc = sc;
179 	if_initname(GIF2IFP(sc), ifc->ifc_name, unit);
180 
181 	sc->encap_cookie4 = sc->encap_cookie6 = NULL;
182 
183 	GIF2IFP(sc)->if_addrlen = 0;
184 	GIF2IFP(sc)->if_mtu    = GIF_MTU;
185 	GIF2IFP(sc)->if_flags  = IFF_POINTOPOINT | IFF_MULTICAST;
186 #if 0
187 	/* turn off ingress filter */
188 	GIF2IFP(sc)->if_flags  |= IFF_LINK2;
189 #endif
190 	GIF2IFP(sc)->if_ioctl  = gif_ioctl;
191 	GIF2IFP(sc)->if_start  = gif_start;
192 	GIF2IFP(sc)->if_output = gif_output;
193 	GIF2IFP(sc)->if_snd.ifq_maxlen = IFQ_MAXLEN;
194 	if_attach(GIF2IFP(sc));
195 	bpfattach(GIF2IFP(sc), DLT_NULL, sizeof(u_int32_t));
196 	if (ng_gif_attach_p != NULL)
197 		(*ng_gif_attach_p)(GIF2IFP(sc));
198 
199 	mtx_lock(&gif_mtx);
200 	LIST_INSERT_HEAD(&V_gif_softc_list, sc, gif_list);
201 	mtx_unlock(&gif_mtx);
202 
203 	return (0);
204 }
205 
206 static void
207 gif_clone_destroy(ifp)
208 	struct ifnet *ifp;
209 {
210 	int err;
211 	struct gif_softc *sc = ifp->if_softc;
212 
213 	mtx_lock(&gif_mtx);
214 	LIST_REMOVE(sc, gif_list);
215 	mtx_unlock(&gif_mtx);
216 
217 	gif_delete_tunnel(ifp);
218 #ifdef INET6
219 	if (sc->encap_cookie6 != NULL) {
220 		err = encap_detach(sc->encap_cookie6);
221 		KASSERT(err == 0, ("Unexpected error detaching encap_cookie6"));
222 	}
223 #endif
224 #ifdef INET
225 	if (sc->encap_cookie4 != NULL) {
226 		err = encap_detach(sc->encap_cookie4);
227 		KASSERT(err == 0, ("Unexpected error detaching encap_cookie4"));
228 	}
229 #endif
230 
231 	if (ng_gif_detach_p != NULL)
232 		(*ng_gif_detach_p)(ifp);
233 	bpfdetach(ifp);
234 	if_detach(ifp);
235 	if_free(ifp);
236 
237 	GIF_LOCK_DESTROY(sc);
238 
239 	free(sc, M_GIF);
240 }
241 
242 static int
243 gifmodevent(mod, type, data)
244 	module_t mod;
245 	int type;
246 	void *data;
247 {
248 
249 	switch (type) {
250 	case MOD_LOAD:
251 		mtx_init(&gif_mtx, "gif_mtx", NULL, MTX_DEF);
252 		LIST_INIT(&V_gif_softc_list);
253 		if_clone_attach(&gif_cloner);
254 
255 #ifdef INET6
256 		V_ip6_gif_hlim = GIF_HLIM;
257 #endif
258 
259 		break;
260 	case MOD_UNLOAD:
261 		if_clone_detach(&gif_cloner);
262 		mtx_destroy(&gif_mtx);
263 #ifdef INET6
264 		V_ip6_gif_hlim = 0;
265 #endif
266 		break;
267 	default:
268 		return EOPNOTSUPP;
269 	}
270 	return 0;
271 }
272 
273 static moduledata_t gif_mod = {
274 	"if_gif",
275 	gifmodevent,
276 	0
277 };
278 
279 DECLARE_MODULE(if_gif, gif_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
280 MODULE_VERSION(if_gif, 1);
281 
282 int
283 gif_encapcheck(m, off, proto, arg)
284 	const struct mbuf *m;
285 	int off;
286 	int proto;
287 	void *arg;
288 {
289 	struct ip ip;
290 	struct gif_softc *sc;
291 
292 	sc = (struct gif_softc *)arg;
293 	if (sc == NULL)
294 		return 0;
295 
296 	if ((GIF2IFP(sc)->if_flags & IFF_UP) == 0)
297 		return 0;
298 
299 	/* no physical address */
300 	if (!sc->gif_psrc || !sc->gif_pdst)
301 		return 0;
302 
303 	switch (proto) {
304 #ifdef INET
305 	case IPPROTO_IPV4:
306 		break;
307 #endif
308 #ifdef INET6
309 	case IPPROTO_IPV6:
310 		break;
311 #endif
312 	case IPPROTO_ETHERIP:
313 		break;
314 
315 	default:
316 		return 0;
317 	}
318 
319 	/* Bail on short packets */
320 	if (m->m_pkthdr.len < sizeof(ip))
321 		return 0;
322 
323 	m_copydata(m, 0, sizeof(ip), (caddr_t)&ip);
324 
325 	switch (ip.ip_v) {
326 #ifdef INET
327 	case 4:
328 		if (sc->gif_psrc->sa_family != AF_INET ||
329 		    sc->gif_pdst->sa_family != AF_INET)
330 			return 0;
331 		return gif_encapcheck4(m, off, proto, arg);
332 #endif
333 #ifdef INET6
334 	case 6:
335 		if (m->m_pkthdr.len < sizeof(struct ip6_hdr))
336 			return 0;
337 		if (sc->gif_psrc->sa_family != AF_INET6 ||
338 		    sc->gif_pdst->sa_family != AF_INET6)
339 			return 0;
340 		return gif_encapcheck6(m, off, proto, arg);
341 #endif
342 	default:
343 		return 0;
344 	}
345 }
346 
347 static void
348 gif_start(struct ifnet *ifp)
349 {
350 	struct gif_softc *sc;
351 	struct mbuf *m;
352 
353 	sc = ifp->if_softc;
354 
355 	ifp->if_drv_flags |= IFF_DRV_OACTIVE;
356 	for (;;) {
357 		IFQ_DEQUEUE(&ifp->if_snd, m);
358 		if (m == 0)
359 			break;
360 
361 		gif_output(ifp, m, sc->gif_pdst, NULL);
362 
363 	}
364 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
365 
366 	return;
367 }
368 
369 int
370 gif_output(ifp, m, dst, rt)
371 	struct ifnet *ifp;
372 	struct mbuf *m;
373 	struct sockaddr *dst;
374 	struct rtentry *rt;	/* added in net2 */
375 {
376 	INIT_VNET_GIF(ifp->if_vnet);
377 	struct gif_softc *sc = ifp->if_softc;
378 	struct m_tag *mtag;
379 	int error = 0;
380 	int gif_called;
381 	u_int32_t af;
382 
383 #ifdef MAC
384 	error = mac_ifnet_check_transmit(ifp, m);
385 	if (error) {
386 		m_freem(m);
387 		goto end;
388 	}
389 #endif
390 
391 	/*
392 	 * gif may cause infinite recursion calls when misconfigured.
393 	 * We'll prevent this by detecting loops.
394 	 *
395 	 * High nesting level may cause stack exhaustion.
396 	 * We'll prevent this by introducing upper limit.
397 	 */
398 	gif_called = 1;
399 	mtag = m_tag_locate(m, MTAG_GIF, MTAG_GIF_CALLED, NULL);
400 	while (mtag != NULL) {
401 		if (*(struct ifnet **)(mtag + 1) == ifp) {
402 			log(LOG_NOTICE,
403 			    "gif_output: loop detected on %s\n",
404 			    (*(struct ifnet **)(mtag + 1))->if_xname);
405 			m_freem(m);
406 			error = EIO;	/* is there better errno? */
407 			goto end;
408 		}
409 		mtag = m_tag_locate(m, MTAG_GIF, MTAG_GIF_CALLED, mtag);
410 		gif_called++;
411 	}
412 	if (gif_called > V_max_gif_nesting) {
413 		log(LOG_NOTICE,
414 		    "gif_output: recursively called too many times(%d)\n",
415 		    gif_called);
416 		m_freem(m);
417 		error = EIO;	/* is there better errno? */
418 		goto end;
419 	}
420 	mtag = m_tag_alloc(MTAG_GIF, MTAG_GIF_CALLED, sizeof(struct ifnet *),
421 	    M_NOWAIT);
422 	if (mtag == NULL) {
423 		m_freem(m);
424 		error = ENOMEM;
425 		goto end;
426 	}
427 	*(struct ifnet **)(mtag + 1) = ifp;
428 	m_tag_prepend(m, mtag);
429 
430 	m->m_flags &= ~(M_BCAST|M_MCAST);
431 
432 	GIF_LOCK(sc);
433 
434 	if (!(ifp->if_flags & IFF_UP) ||
435 	    sc->gif_psrc == NULL || sc->gif_pdst == NULL) {
436 		GIF_UNLOCK(sc);
437 		m_freem(m);
438 		error = ENETDOWN;
439 		goto end;
440 	}
441 
442 	/* BPF writes need to be handled specially. */
443 	if (dst->sa_family == AF_UNSPEC) {
444 		bcopy(dst->sa_data, &af, sizeof(af));
445 		dst->sa_family = af;
446 	}
447 
448 	af = dst->sa_family;
449 	BPF_MTAP2(ifp, &af, sizeof(af), m);
450 	ifp->if_opackets++;
451 	ifp->if_obytes += m->m_pkthdr.len;
452 
453 	/* override to IPPROTO_ETHERIP for bridged traffic */
454 	if (ifp->if_bridge)
455 		af = AF_LINK;
456 
457 	M_SETFIB(m, sc->gif_fibnum);
458 	/* inner AF-specific encapsulation */
459 
460 	/* XXX should we check if our outer source is legal? */
461 
462 	/* dispatch to output logic based on outer AF */
463 	switch (sc->gif_psrc->sa_family) {
464 #ifdef INET
465 	case AF_INET:
466 		error = in_gif_output(ifp, af, m);
467 		break;
468 #endif
469 #ifdef INET6
470 	case AF_INET6:
471 		error = in6_gif_output(ifp, af, m);
472 		break;
473 #endif
474 	default:
475 		m_freem(m);
476 		error = ENETDOWN;
477 	}
478 
479 	GIF_UNLOCK(sc);
480   end:
481 	if (error)
482 		ifp->if_oerrors++;
483 	return (error);
484 }
485 
486 void
487 gif_input(m, af, ifp)
488 	struct mbuf *m;
489 	int af;
490 	struct ifnet *ifp;
491 {
492 	int isr, n;
493 	struct etherip_header *eip;
494 	struct ether_header *eh;
495 	struct ifnet *oldifp;
496 
497 	if (ifp == NULL) {
498 		/* just in case */
499 		m_freem(m);
500 		return;
501 	}
502 
503 	m->m_pkthdr.rcvif = ifp;
504 
505 #ifdef MAC
506 	mac_ifnet_create_mbuf(ifp, m);
507 #endif
508 
509 	if (bpf_peers_present(ifp->if_bpf)) {
510 		u_int32_t af1 = af;
511 		bpf_mtap2(ifp->if_bpf, &af1, sizeof(af1), m);
512 	}
513 
514 	if (ng_gif_input_p != NULL) {
515 		(*ng_gif_input_p)(ifp, &m, af);
516 		if (m == NULL)
517 			return;
518 	}
519 
520 	/*
521 	 * Put the packet to the network layer input queue according to the
522 	 * specified address family.
523 	 * Note: older versions of gif_input directly called network layer
524 	 * input functions, e.g. ip6_input, here.  We changed the policy to
525 	 * prevent too many recursive calls of such input functions, which
526 	 * might cause kernel panic.  But the change may introduce another
527 	 * problem; if the input queue is full, packets are discarded.
528 	 * The kernel stack overflow really happened, and we believed
529 	 * queue-full rarely occurs, so we changed the policy.
530 	 */
531 	switch (af) {
532 #ifdef INET
533 	case AF_INET:
534 		isr = NETISR_IP;
535 		break;
536 #endif
537 #ifdef INET6
538 	case AF_INET6:
539 		isr = NETISR_IPV6;
540 		break;
541 #endif
542 	case AF_LINK:
543 		n = sizeof(struct etherip_header) + sizeof(struct ether_header);
544 		if (n > m->m_len) {
545 			m = m_pullup(m, n);
546 			if (m == NULL) {
547 				ifp->if_ierrors++;
548 				return;
549 			}
550 		}
551 
552 		eip = mtod(m, struct etherip_header *);
553  		if (eip->eip_ver !=
554 		    (ETHERIP_VERSION & ETHERIP_VER_VERS_MASK)) {
555 			/* discard unknown versions */
556 			m_freem(m);
557 			return;
558 		}
559 		m_adj(m, sizeof(struct etherip_header));
560 
561 		m->m_flags &= ~(M_BCAST|M_MCAST);
562 		m->m_pkthdr.rcvif = ifp;
563 
564 		if (ifp->if_bridge) {
565 			oldifp = ifp;
566 			eh = mtod(m, struct ether_header *);
567 			if (ETHER_IS_MULTICAST(eh->ether_dhost)) {
568 				if (ETHER_IS_BROADCAST(eh->ether_dhost))
569 					m->m_flags |= M_BCAST;
570 				else
571 					m->m_flags |= M_MCAST;
572 				ifp->if_imcasts++;
573 			}
574 			BRIDGE_INPUT(ifp, m);
575 
576 			if (m != NULL && ifp != oldifp) {
577 				/*
578 				 * The bridge gave us back itself or one of the
579 				 * members for which the frame is addressed.
580 				 */
581 				ether_demux(ifp, m);
582 				return;
583 			}
584 		}
585 		if (m != NULL)
586 			m_freem(m);
587 		return;
588 
589 	default:
590 		if (ng_gif_input_orphan_p != NULL)
591 			(*ng_gif_input_orphan_p)(ifp, m, af);
592 		else
593 			m_freem(m);
594 		return;
595 	}
596 
597 	ifp->if_ipackets++;
598 	ifp->if_ibytes += m->m_pkthdr.len;
599 	netisr_dispatch(isr, m);
600 }
601 
602 /* XXX how should we handle IPv6 scope on SIOC[GS]IFPHYADDR? */
603 int
604 gif_ioctl(ifp, cmd, data)
605 	struct ifnet *ifp;
606 	u_long cmd;
607 	caddr_t data;
608 {
609 	struct gif_softc *sc  = ifp->if_softc;
610 	struct ifreq     *ifr = (struct ifreq*)data;
611 	int error = 0, size;
612 	struct sockaddr *dst, *src;
613 #ifdef	SIOCSIFMTU /* xxx */
614 	u_long mtu;
615 #endif
616 
617 	switch (cmd) {
618 	case SIOCSIFADDR:
619 		ifp->if_flags |= IFF_UP;
620 		break;
621 
622 	case SIOCSIFDSTADDR:
623 		break;
624 
625 	case SIOCADDMULTI:
626 	case SIOCDELMULTI:
627 		break;
628 
629 #ifdef	SIOCSIFMTU /* xxx */
630 	case SIOCGIFMTU:
631 		break;
632 
633 	case SIOCSIFMTU:
634 		mtu = ifr->ifr_mtu;
635 		if (mtu < GIF_MTU_MIN || mtu > GIF_MTU_MAX)
636 			return (EINVAL);
637 		ifp->if_mtu = mtu;
638 		break;
639 #endif /* SIOCSIFMTU */
640 
641 #ifdef INET
642 	case SIOCSIFPHYADDR:
643 #endif
644 #ifdef INET6
645 	case SIOCSIFPHYADDR_IN6:
646 #endif /* INET6 */
647 	case SIOCSLIFPHYADDR:
648 		switch (cmd) {
649 #ifdef INET
650 		case SIOCSIFPHYADDR:
651 			src = (struct sockaddr *)
652 				&(((struct in_aliasreq *)data)->ifra_addr);
653 			dst = (struct sockaddr *)
654 				&(((struct in_aliasreq *)data)->ifra_dstaddr);
655 			break;
656 #endif
657 #ifdef INET6
658 		case SIOCSIFPHYADDR_IN6:
659 			src = (struct sockaddr *)
660 				&(((struct in6_aliasreq *)data)->ifra_addr);
661 			dst = (struct sockaddr *)
662 				&(((struct in6_aliasreq *)data)->ifra_dstaddr);
663 			break;
664 #endif
665 		case SIOCSLIFPHYADDR:
666 			src = (struct sockaddr *)
667 				&(((struct if_laddrreq *)data)->addr);
668 			dst = (struct sockaddr *)
669 				&(((struct if_laddrreq *)data)->dstaddr);
670 			break;
671 		default:
672 			return EINVAL;
673 		}
674 
675 		/* sa_family must be equal */
676 		if (src->sa_family != dst->sa_family)
677 			return EINVAL;
678 
679 		/* validate sa_len */
680 		switch (src->sa_family) {
681 #ifdef INET
682 		case AF_INET:
683 			if (src->sa_len != sizeof(struct sockaddr_in))
684 				return EINVAL;
685 			break;
686 #endif
687 #ifdef INET6
688 		case AF_INET6:
689 			if (src->sa_len != sizeof(struct sockaddr_in6))
690 				return EINVAL;
691 			break;
692 #endif
693 		default:
694 			return EAFNOSUPPORT;
695 		}
696 		switch (dst->sa_family) {
697 #ifdef INET
698 		case AF_INET:
699 			if (dst->sa_len != sizeof(struct sockaddr_in))
700 				return EINVAL;
701 			break;
702 #endif
703 #ifdef INET6
704 		case AF_INET6:
705 			if (dst->sa_len != sizeof(struct sockaddr_in6))
706 				return EINVAL;
707 			break;
708 #endif
709 		default:
710 			return EAFNOSUPPORT;
711 		}
712 
713 		/* check sa_family looks sane for the cmd */
714 		switch (cmd) {
715 		case SIOCSIFPHYADDR:
716 			if (src->sa_family == AF_INET)
717 				break;
718 			return EAFNOSUPPORT;
719 #ifdef INET6
720 		case SIOCSIFPHYADDR_IN6:
721 			if (src->sa_family == AF_INET6)
722 				break;
723 			return EAFNOSUPPORT;
724 #endif /* INET6 */
725 		case SIOCSLIFPHYADDR:
726 			/* checks done in the above */
727 			break;
728 		}
729 
730 		error = gif_set_tunnel(GIF2IFP(sc), src, dst);
731 		break;
732 
733 #ifdef SIOCDIFPHYADDR
734 	case SIOCDIFPHYADDR:
735 		gif_delete_tunnel(GIF2IFP(sc));
736 		break;
737 #endif
738 
739 	case SIOCGIFPSRCADDR:
740 #ifdef INET6
741 	case SIOCGIFPSRCADDR_IN6:
742 #endif /* INET6 */
743 		if (sc->gif_psrc == NULL) {
744 			error = EADDRNOTAVAIL;
745 			goto bad;
746 		}
747 		src = sc->gif_psrc;
748 		switch (cmd) {
749 #ifdef INET
750 		case SIOCGIFPSRCADDR:
751 			dst = &ifr->ifr_addr;
752 			size = sizeof(ifr->ifr_addr);
753 			break;
754 #endif /* INET */
755 #ifdef INET6
756 		case SIOCGIFPSRCADDR_IN6:
757 			dst = (struct sockaddr *)
758 				&(((struct in6_ifreq *)data)->ifr_addr);
759 			size = sizeof(((struct in6_ifreq *)data)->ifr_addr);
760 			break;
761 #endif /* INET6 */
762 		default:
763 			error = EADDRNOTAVAIL;
764 			goto bad;
765 		}
766 		if (src->sa_len > size)
767 			return EINVAL;
768 		bcopy((caddr_t)src, (caddr_t)dst, src->sa_len);
769 #ifdef INET6
770 		if (dst->sa_family == AF_INET6) {
771 			error = sa6_recoverscope((struct sockaddr_in6 *)dst);
772 			if (error != 0)
773 				return (error);
774 		}
775 #endif
776 		break;
777 
778 	case SIOCGIFPDSTADDR:
779 #ifdef INET6
780 	case SIOCGIFPDSTADDR_IN6:
781 #endif /* INET6 */
782 		if (sc->gif_pdst == NULL) {
783 			error = EADDRNOTAVAIL;
784 			goto bad;
785 		}
786 		src = sc->gif_pdst;
787 		switch (cmd) {
788 #ifdef INET
789 		case SIOCGIFPDSTADDR:
790 			dst = &ifr->ifr_addr;
791 			size = sizeof(ifr->ifr_addr);
792 			break;
793 #endif /* INET */
794 #ifdef INET6
795 		case SIOCGIFPDSTADDR_IN6:
796 			dst = (struct sockaddr *)
797 				&(((struct in6_ifreq *)data)->ifr_addr);
798 			size = sizeof(((struct in6_ifreq *)data)->ifr_addr);
799 			break;
800 #endif /* INET6 */
801 		default:
802 			error = EADDRNOTAVAIL;
803 			goto bad;
804 		}
805 		if (src->sa_len > size)
806 			return EINVAL;
807 		bcopy((caddr_t)src, (caddr_t)dst, src->sa_len);
808 #ifdef INET6
809 		if (dst->sa_family == AF_INET6) {
810 			error = sa6_recoverscope((struct sockaddr_in6 *)dst);
811 			if (error != 0)
812 				return (error);
813 		}
814 #endif
815 		break;
816 
817 	case SIOCGLIFPHYADDR:
818 		if (sc->gif_psrc == NULL || sc->gif_pdst == NULL) {
819 			error = EADDRNOTAVAIL;
820 			goto bad;
821 		}
822 
823 		/* copy src */
824 		src = sc->gif_psrc;
825 		dst = (struct sockaddr *)
826 			&(((struct if_laddrreq *)data)->addr);
827 		size = sizeof(((struct if_laddrreq *)data)->addr);
828 		if (src->sa_len > size)
829 			return EINVAL;
830 		bcopy((caddr_t)src, (caddr_t)dst, src->sa_len);
831 
832 		/* copy dst */
833 		src = sc->gif_pdst;
834 		dst = (struct sockaddr *)
835 			&(((struct if_laddrreq *)data)->dstaddr);
836 		size = sizeof(((struct if_laddrreq *)data)->dstaddr);
837 		if (src->sa_len > size)
838 			return EINVAL;
839 		bcopy((caddr_t)src, (caddr_t)dst, src->sa_len);
840 		break;
841 
842 	case SIOCSIFFLAGS:
843 		/* if_ioctl() takes care of it */
844 		break;
845 
846 	default:
847 		error = EINVAL;
848 		break;
849 	}
850  bad:
851 	return error;
852 }
853 
854 /*
855  * XXXRW: There's a general event-ordering issue here: the code to check
856  * if a given tunnel is already present happens before we perform a
857  * potentially blocking setup of the tunnel.  This code needs to be
858  * re-ordered so that the check and replacement can be atomic using
859  * a mutex.
860  */
861 int
862 gif_set_tunnel(ifp, src, dst)
863 	struct ifnet *ifp;
864 	struct sockaddr *src;
865 	struct sockaddr *dst;
866 {
867 	INIT_VNET_GIF(ifp->if_vnet);
868 	struct gif_softc *sc = ifp->if_softc;
869 	struct gif_softc *sc2;
870 	struct sockaddr *osrc, *odst, *sa;
871 	int error = 0;
872 
873 	mtx_lock(&gif_mtx);
874 	LIST_FOREACH(sc2, &V_gif_softc_list, gif_list) {
875 		if (sc2 == sc)
876 			continue;
877 		if (!sc2->gif_pdst || !sc2->gif_psrc)
878 			continue;
879 		if (sc2->gif_pdst->sa_family != dst->sa_family ||
880 		    sc2->gif_pdst->sa_len != dst->sa_len ||
881 		    sc2->gif_psrc->sa_family != src->sa_family ||
882 		    sc2->gif_psrc->sa_len != src->sa_len)
883 			continue;
884 
885 		/*
886 		 * Disallow parallel tunnels unless instructed
887 		 * otherwise.
888 		 */
889 		if (!V_parallel_tunnels &&
890 		    bcmp(sc2->gif_pdst, dst, dst->sa_len) == 0 &&
891 		    bcmp(sc2->gif_psrc, src, src->sa_len) == 0) {
892 			error = EADDRNOTAVAIL;
893 			mtx_unlock(&gif_mtx);
894 			goto bad;
895 		}
896 
897 		/* XXX both end must be valid? (I mean, not 0.0.0.0) */
898 	}
899 	mtx_unlock(&gif_mtx);
900 
901 	/* XXX we can detach from both, but be polite just in case */
902 	if (sc->gif_psrc)
903 		switch (sc->gif_psrc->sa_family) {
904 #ifdef INET
905 		case AF_INET:
906 			(void)in_gif_detach(sc);
907 			break;
908 #endif
909 #ifdef INET6
910 		case AF_INET6:
911 			(void)in6_gif_detach(sc);
912 			break;
913 #endif
914 		}
915 
916 	osrc = sc->gif_psrc;
917 	sa = (struct sockaddr *)malloc(src->sa_len, M_IFADDR, M_WAITOK);
918 	bcopy((caddr_t)src, (caddr_t)sa, src->sa_len);
919 	sc->gif_psrc = sa;
920 
921 	odst = sc->gif_pdst;
922 	sa = (struct sockaddr *)malloc(dst->sa_len, M_IFADDR, M_WAITOK);
923 	bcopy((caddr_t)dst, (caddr_t)sa, dst->sa_len);
924 	sc->gif_pdst = sa;
925 
926 	switch (sc->gif_psrc->sa_family) {
927 #ifdef INET
928 	case AF_INET:
929 		error = in_gif_attach(sc);
930 		break;
931 #endif
932 #ifdef INET6
933 	case AF_INET6:
934 		/*
935 		 * Check validity of the scope zone ID of the addresses, and
936 		 * convert it into the kernel internal form if necessary.
937 		 */
938 		error = sa6_embedscope((struct sockaddr_in6 *)sc->gif_psrc, 0);
939 		if (error != 0)
940 			break;
941 		error = sa6_embedscope((struct sockaddr_in6 *)sc->gif_pdst, 0);
942 		if (error != 0)
943 			break;
944 		error = in6_gif_attach(sc);
945 		break;
946 #endif
947 	}
948 	if (error) {
949 		/* rollback */
950 		free((caddr_t)sc->gif_psrc, M_IFADDR);
951 		free((caddr_t)sc->gif_pdst, M_IFADDR);
952 		sc->gif_psrc = osrc;
953 		sc->gif_pdst = odst;
954 		goto bad;
955 	}
956 
957 	if (osrc)
958 		free((caddr_t)osrc, M_IFADDR);
959 	if (odst)
960 		free((caddr_t)odst, M_IFADDR);
961 
962  bad:
963 	if (sc->gif_psrc && sc->gif_pdst)
964 		ifp->if_drv_flags |= IFF_DRV_RUNNING;
965 	else
966 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
967 
968 	return error;
969 }
970 
971 void
972 gif_delete_tunnel(ifp)
973 	struct ifnet *ifp;
974 {
975 	struct gif_softc *sc = ifp->if_softc;
976 
977 	if (sc->gif_psrc) {
978 		free((caddr_t)sc->gif_psrc, M_IFADDR);
979 		sc->gif_psrc = NULL;
980 	}
981 	if (sc->gif_pdst) {
982 		free((caddr_t)sc->gif_pdst, M_IFADDR);
983 		sc->gif_pdst = NULL;
984 	}
985 	/* it is safe to detach from both */
986 #ifdef INET
987 	(void)in_gif_detach(sc);
988 #endif
989 #ifdef INET6
990 	(void)in6_gif_detach(sc);
991 #endif
992 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
993 }
994