xref: /freebsd/sys/net/if_gif.c (revision 984485a02eb3e63b4170dd911b72de38b35b2289)
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 
36 #include <sys/param.h>
37 #include <sys/systm.h>
38 #include <sys/kernel.h>
39 #include <sys/malloc.h>
40 #include <sys/mbuf.h>
41 #include <sys/module.h>
42 #include <sys/socket.h>
43 #include <sys/sockio.h>
44 #include <sys/errno.h>
45 #include <sys/time.h>
46 #include <sys/sysctl.h>
47 #include <sys/syslog.h>
48 #include <sys/priv.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 #include <net/vnet.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 
98 static VNET_DEFINE(LIST_HEAD(, gif_softc), gif_softc_list);
99 
100 #define	V_gif_softc_list	VNET(gif_softc_list)
101 
102 #ifdef INET
103 VNET_DEFINE(int, ip_gif_ttl) = GIF_TTL;
104 #define	V_ip_gif_ttl		VNET(ip_gif_ttl)
105 #endif
106 #ifdef INET6
107 VNET_DEFINE(int, ip6_gif_hlim) = GIF_HLIM;
108 #define	V_ip6_gif_hlim		VNET(ip6_gif_hlim)
109 #endif
110 
111 void	(*ng_gif_input_p)(struct ifnet *ifp, struct mbuf **mp, int af);
112 void	(*ng_gif_input_orphan_p)(struct ifnet *ifp, struct mbuf *m, int af);
113 void	(*ng_gif_attach_p)(struct ifnet *ifp);
114 void	(*ng_gif_detach_p)(struct ifnet *ifp);
115 
116 static void	gif_start(struct ifnet *);
117 static int	gif_clone_create(struct if_clone *, int, caddr_t);
118 static void	gif_clone_destroy(struct ifnet *);
119 
120 IFC_SIMPLE_DECLARE(gif, 0);
121 
122 static int gifmodevent(module_t, int, void *);
123 
124 SYSCTL_DECL(_net_link);
125 SYSCTL_NODE(_net_link, IFT_GIF, gif, CTLFLAG_RW, 0,
126     "Generic Tunnel Interface");
127 #ifndef MAX_GIF_NEST
128 /*
129  * This macro controls the default upper limitation on nesting of gif tunnels.
130  * Since, setting a large value to this macro with a careless configuration
131  * may introduce system crash, we don't allow any nestings by default.
132  * If you need to configure nested gif tunnels, you can define this macro
133  * in your kernel configuration file.  However, if you do so, please be
134  * careful to configure the tunnels so that it won't make a loop.
135  */
136 #define MAX_GIF_NEST 1
137 #endif
138 
139 static VNET_DEFINE(int, max_gif_nesting) = MAX_GIF_NEST;
140 #define	V_max_gif_nesting	VNET(max_gif_nesting)
141 
142 SYSCTL_VNET_INT(_net_link_gif, OID_AUTO, max_nesting, CTLFLAG_RW,
143     &VNET_NAME(max_gif_nesting), 0, "Max nested tunnels");
144 
145 #ifdef INET6
146 SYSCTL_DECL(_net_inet6_ip6);
147 SYSCTL_VNET_INT(_net_inet6_ip6, IPV6CTL_GIF_HLIM, gifhlim, CTLFLAG_RW,
148     &VNET_NAME(ip6_gif_hlim), 0, "");
149 #endif
150 
151 /*
152  * By default, we disallow creation of multiple tunnels between the same
153  * pair of addresses.  Some applications require this functionality so
154  * we allow control over this check here.
155  */
156 #ifdef XBONEHACK
157 static VNET_DEFINE(int, parallel_tunnels) = 1;
158 #else
159 static VNET_DEFINE(int, parallel_tunnels) = 0;
160 #endif
161 #define	V_parallel_tunnels	VNET(parallel_tunnels)
162 
163 SYSCTL_VNET_INT(_net_link_gif, OID_AUTO, parallel_tunnels, CTLFLAG_RW,
164     &VNET_NAME(parallel_tunnels), 0, "Allow parallel tunnels?");
165 
166 /* copy from src/sys/net/if_ethersubr.c */
167 static const u_char etherbroadcastaddr[ETHER_ADDR_LEN] =
168 			{ 0xff, 0xff, 0xff, 0xff, 0xff, 0xff };
169 #ifndef ETHER_IS_BROADCAST
170 #define ETHER_IS_BROADCAST(addr) \
171 	(bcmp(etherbroadcastaddr, (addr), ETHER_ADDR_LEN) == 0)
172 #endif
173 
174 static int
175 gif_clone_create(ifc, unit, params)
176 	struct if_clone *ifc;
177 	int unit;
178 	caddr_t params;
179 {
180 	struct gif_softc *sc;
181 
182 	sc = malloc(sizeof(struct gif_softc), M_GIF, M_WAITOK | M_ZERO);
183 	sc->gif_fibnum = curthread->td_proc->p_fibnum;
184 	GIF2IFP(sc) = if_alloc(IFT_GIF);
185 	if (GIF2IFP(sc) == NULL) {
186 		free(sc, M_GIF);
187 		return (ENOSPC);
188 	}
189 
190 	GIF_LOCK_INIT(sc);
191 
192 	GIF2IFP(sc)->if_softc = sc;
193 	if_initname(GIF2IFP(sc), ifc->ifc_name, unit);
194 
195 	sc->encap_cookie4 = sc->encap_cookie6 = NULL;
196 	sc->gif_options = GIF_ACCEPT_REVETHIP;
197 
198 	GIF2IFP(sc)->if_addrlen = 0;
199 	GIF2IFP(sc)->if_mtu    = GIF_MTU;
200 	GIF2IFP(sc)->if_flags  = IFF_POINTOPOINT | IFF_MULTICAST;
201 #if 0
202 	/* turn off ingress filter */
203 	GIF2IFP(sc)->if_flags  |= IFF_LINK2;
204 #endif
205 	GIF2IFP(sc)->if_ioctl  = gif_ioctl;
206 	GIF2IFP(sc)->if_start  = gif_start;
207 	GIF2IFP(sc)->if_output = gif_output;
208 	GIF2IFP(sc)->if_snd.ifq_maxlen = IFQ_MAXLEN;
209 	if_attach(GIF2IFP(sc));
210 	bpfattach(GIF2IFP(sc), DLT_NULL, sizeof(u_int32_t));
211 	if (ng_gif_attach_p != NULL)
212 		(*ng_gif_attach_p)(GIF2IFP(sc));
213 
214 	mtx_lock(&gif_mtx);
215 	LIST_INSERT_HEAD(&V_gif_softc_list, sc, gif_list);
216 	mtx_unlock(&gif_mtx);
217 
218 	return (0);
219 }
220 
221 static void
222 gif_clone_destroy(ifp)
223 	struct ifnet *ifp;
224 {
225 #if defined(INET) || defined(INET6)
226 	int err;
227 #endif
228 	struct gif_softc *sc = ifp->if_softc;
229 
230 	mtx_lock(&gif_mtx);
231 	LIST_REMOVE(sc, gif_list);
232 	mtx_unlock(&gif_mtx);
233 
234 	gif_delete_tunnel(ifp);
235 #ifdef INET6
236 	if (sc->encap_cookie6 != NULL) {
237 		err = encap_detach(sc->encap_cookie6);
238 		KASSERT(err == 0, ("Unexpected error detaching encap_cookie6"));
239 	}
240 #endif
241 #ifdef INET
242 	if (sc->encap_cookie4 != NULL) {
243 		err = encap_detach(sc->encap_cookie4);
244 		KASSERT(err == 0, ("Unexpected error detaching encap_cookie4"));
245 	}
246 #endif
247 
248 	if (ng_gif_detach_p != NULL)
249 		(*ng_gif_detach_p)(ifp);
250 	bpfdetach(ifp);
251 	if_detach(ifp);
252 	if_free(ifp);
253 
254 	GIF_LOCK_DESTROY(sc);
255 
256 	free(sc, M_GIF);
257 }
258 
259 static void
260 vnet_gif_init(const void *unused __unused)
261 {
262 
263 	LIST_INIT(&V_gif_softc_list);
264 }
265 VNET_SYSINIT(vnet_gif_init, SI_SUB_PSEUDO, SI_ORDER_MIDDLE, vnet_gif_init,
266     NULL);
267 
268 static int
269 gifmodevent(mod, type, data)
270 	module_t mod;
271 	int type;
272 	void *data;
273 {
274 
275 	switch (type) {
276 	case MOD_LOAD:
277 		mtx_init(&gif_mtx, "gif_mtx", NULL, MTX_DEF);
278 		if_clone_attach(&gif_cloner);
279 		break;
280 
281 	case MOD_UNLOAD:
282 		if_clone_detach(&gif_cloner);
283 		mtx_destroy(&gif_mtx);
284 		break;
285 	default:
286 		return EOPNOTSUPP;
287 	}
288 	return 0;
289 }
290 
291 static moduledata_t gif_mod = {
292 	"if_gif",
293 	gifmodevent,
294 	0
295 };
296 
297 DECLARE_MODULE(if_gif, gif_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
298 MODULE_VERSION(if_gif, 1);
299 
300 int
301 gif_encapcheck(m, off, proto, arg)
302 	const struct mbuf *m;
303 	int off;
304 	int proto;
305 	void *arg;
306 {
307 	struct ip ip;
308 	struct gif_softc *sc;
309 
310 	sc = (struct gif_softc *)arg;
311 	if (sc == NULL)
312 		return 0;
313 
314 	if ((GIF2IFP(sc)->if_flags & IFF_UP) == 0)
315 		return 0;
316 
317 	/* no physical address */
318 	if (!sc->gif_psrc || !sc->gif_pdst)
319 		return 0;
320 
321 	switch (proto) {
322 #ifdef INET
323 	case IPPROTO_IPV4:
324 		break;
325 #endif
326 #ifdef INET6
327 	case IPPROTO_IPV6:
328 		break;
329 #endif
330 	case IPPROTO_ETHERIP:
331 		break;
332 
333 	default:
334 		return 0;
335 	}
336 
337 	/* Bail on short packets */
338 	if (m->m_pkthdr.len < sizeof(ip))
339 		return 0;
340 
341 	m_copydata(m, 0, sizeof(ip), (caddr_t)&ip);
342 
343 	switch (ip.ip_v) {
344 #ifdef INET
345 	case 4:
346 		if (sc->gif_psrc->sa_family != AF_INET ||
347 		    sc->gif_pdst->sa_family != AF_INET)
348 			return 0;
349 		return gif_encapcheck4(m, off, proto, arg);
350 #endif
351 #ifdef INET6
352 	case 6:
353 		if (m->m_pkthdr.len < sizeof(struct ip6_hdr))
354 			return 0;
355 		if (sc->gif_psrc->sa_family != AF_INET6 ||
356 		    sc->gif_pdst->sa_family != AF_INET6)
357 			return 0;
358 		return gif_encapcheck6(m, off, proto, arg);
359 #endif
360 	default:
361 		return 0;
362 	}
363 }
364 
365 static void
366 gif_start(struct ifnet *ifp)
367 {
368 	struct gif_softc *sc;
369 	struct mbuf *m;
370 
371 	sc = ifp->if_softc;
372 
373 	ifp->if_drv_flags |= IFF_DRV_OACTIVE;
374 	for (;;) {
375 		IFQ_DEQUEUE(&ifp->if_snd, m);
376 		if (m == 0)
377 			break;
378 
379 		gif_output(ifp, m, sc->gif_pdst, NULL);
380 
381 	}
382 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
383 
384 	return;
385 }
386 
387 int
388 gif_output(ifp, m, dst, ro)
389 	struct ifnet *ifp;
390 	struct mbuf *m;
391 	struct sockaddr *dst;
392 	struct route *ro;
393 {
394 	struct gif_softc *sc = ifp->if_softc;
395 	struct m_tag *mtag;
396 	int error = 0;
397 	int gif_called;
398 	u_int32_t af;
399 
400 #ifdef MAC
401 	error = mac_ifnet_check_transmit(ifp, m);
402 	if (error) {
403 		m_freem(m);
404 		goto end;
405 	}
406 #endif
407 
408 	/*
409 	 * gif may cause infinite recursion calls when misconfigured.
410 	 * We'll prevent this by detecting loops.
411 	 *
412 	 * High nesting level may cause stack exhaustion.
413 	 * We'll prevent this by introducing upper limit.
414 	 */
415 	gif_called = 1;
416 	mtag = m_tag_locate(m, MTAG_GIF, MTAG_GIF_CALLED, NULL);
417 	while (mtag != NULL) {
418 		if (*(struct ifnet **)(mtag + 1) == ifp) {
419 			log(LOG_NOTICE,
420 			    "gif_output: loop detected on %s\n",
421 			    (*(struct ifnet **)(mtag + 1))->if_xname);
422 			m_freem(m);
423 			error = EIO;	/* is there better errno? */
424 			goto end;
425 		}
426 		mtag = m_tag_locate(m, MTAG_GIF, MTAG_GIF_CALLED, mtag);
427 		gif_called++;
428 	}
429 	if (gif_called > V_max_gif_nesting) {
430 		log(LOG_NOTICE,
431 		    "gif_output: recursively called too many times(%d)\n",
432 		    gif_called);
433 		m_freem(m);
434 		error = EIO;	/* is there better errno? */
435 		goto end;
436 	}
437 	mtag = m_tag_alloc(MTAG_GIF, MTAG_GIF_CALLED, sizeof(struct ifnet *),
438 	    M_NOWAIT);
439 	if (mtag == NULL) {
440 		m_freem(m);
441 		error = ENOMEM;
442 		goto end;
443 	}
444 	*(struct ifnet **)(mtag + 1) = ifp;
445 	m_tag_prepend(m, mtag);
446 
447 	m->m_flags &= ~(M_BCAST|M_MCAST);
448 
449 	GIF_LOCK(sc);
450 
451 	if (!(ifp->if_flags & IFF_UP) ||
452 	    sc->gif_psrc == NULL || sc->gif_pdst == NULL) {
453 		GIF_UNLOCK(sc);
454 		m_freem(m);
455 		error = ENETDOWN;
456 		goto end;
457 	}
458 
459 	/* BPF writes need to be handled specially. */
460 	if (dst->sa_family == AF_UNSPEC) {
461 		bcopy(dst->sa_data, &af, sizeof(af));
462 		dst->sa_family = af;
463 	}
464 
465 	af = dst->sa_family;
466 	BPF_MTAP2(ifp, &af, sizeof(af), m);
467 	ifp->if_opackets++;
468 	ifp->if_obytes += m->m_pkthdr.len;
469 
470 	/* override to IPPROTO_ETHERIP for bridged traffic */
471 	if (ifp->if_bridge)
472 		af = AF_LINK;
473 
474 	M_SETFIB(m, sc->gif_fibnum);
475 	/* inner AF-specific encapsulation */
476 
477 	/* XXX should we check if our outer source is legal? */
478 
479 	/* dispatch to output logic based on outer AF */
480 	switch (sc->gif_psrc->sa_family) {
481 #ifdef INET
482 	case AF_INET:
483 		error = in_gif_output(ifp, af, m);
484 		break;
485 #endif
486 #ifdef INET6
487 	case AF_INET6:
488 		error = in6_gif_output(ifp, af, m);
489 		break;
490 #endif
491 	default:
492 		m_freem(m);
493 		error = ENETDOWN;
494 	}
495 
496 	GIF_UNLOCK(sc);
497   end:
498 	if (error)
499 		ifp->if_oerrors++;
500 	return (error);
501 }
502 
503 void
504 gif_input(m, af, ifp)
505 	struct mbuf *m;
506 	int af;
507 	struct ifnet *ifp;
508 {
509 	int isr, n;
510 	struct gif_softc *sc;
511 	struct etherip_header *eip;
512 	struct ether_header *eh;
513 	struct ifnet *oldifp;
514 
515 	if (ifp == NULL) {
516 		/* just in case */
517 		m_freem(m);
518 		return;
519 	}
520 	sc = ifp->if_softc;
521 	m->m_pkthdr.rcvif = ifp;
522 
523 #ifdef MAC
524 	mac_ifnet_create_mbuf(ifp, m);
525 #endif
526 
527 	if (bpf_peers_present(ifp->if_bpf)) {
528 		u_int32_t af1 = af;
529 		bpf_mtap2(ifp->if_bpf, &af1, sizeof(af1), m);
530 	}
531 
532 	if (ng_gif_input_p != NULL) {
533 		(*ng_gif_input_p)(ifp, &m, af);
534 		if (m == NULL)
535 			return;
536 	}
537 
538 	/*
539 	 * Put the packet to the network layer input queue according to the
540 	 * specified address family.
541 	 * Note: older versions of gif_input directly called network layer
542 	 * input functions, e.g. ip6_input, here.  We changed the policy to
543 	 * prevent too many recursive calls of such input functions, which
544 	 * might cause kernel panic.  But the change may introduce another
545 	 * problem; if the input queue is full, packets are discarded.
546 	 * The kernel stack overflow really happened, and we believed
547 	 * queue-full rarely occurs, so we changed the policy.
548 	 */
549 	switch (af) {
550 #ifdef INET
551 	case AF_INET:
552 		isr = NETISR_IP;
553 		break;
554 #endif
555 #ifdef INET6
556 	case AF_INET6:
557 		isr = NETISR_IPV6;
558 		break;
559 #endif
560 	case AF_LINK:
561 		n = sizeof(struct etherip_header) + sizeof(struct ether_header);
562 		if (n > m->m_len) {
563 			m = m_pullup(m, n);
564 			if (m == NULL) {
565 				ifp->if_ierrors++;
566 				return;
567 			}
568 		}
569 
570 		eip = mtod(m, struct etherip_header *);
571 		/*
572 		 * GIF_ACCEPT_REVETHIP (enabled by default) intentionally
573 		 * accepts an EtherIP packet with revered version field in
574 		 * the header.  This is a knob for backward compatibility
575 		 * with FreeBSD 7.2R or prior.
576 		 */
577 		if (sc->gif_options & GIF_ACCEPT_REVETHIP) {
578 			if (eip->eip_resvl != ETHERIP_VERSION
579 			    && eip->eip_ver != ETHERIP_VERSION) {
580 				/* discard unknown versions */
581 				m_freem(m);
582 				return;
583 			}
584 		} else {
585 			if (eip->eip_ver != ETHERIP_VERSION) {
586 				/* discard unknown versions */
587 				m_freem(m);
588 				return;
589 			}
590 		}
591 		m_adj(m, sizeof(struct etherip_header));
592 
593 		m->m_flags &= ~(M_BCAST|M_MCAST);
594 		m->m_pkthdr.rcvif = ifp;
595 
596 		if (ifp->if_bridge) {
597 			oldifp = ifp;
598 			eh = mtod(m, struct ether_header *);
599 			if (ETHER_IS_MULTICAST(eh->ether_dhost)) {
600 				if (ETHER_IS_BROADCAST(eh->ether_dhost))
601 					m->m_flags |= M_BCAST;
602 				else
603 					m->m_flags |= M_MCAST;
604 				ifp->if_imcasts++;
605 			}
606 			BRIDGE_INPUT(ifp, m);
607 
608 			if (m != NULL && ifp != oldifp) {
609 				/*
610 				 * The bridge gave us back itself or one of the
611 				 * members for which the frame is addressed.
612 				 */
613 				ether_demux(ifp, m);
614 				return;
615 			}
616 		}
617 		if (m != NULL)
618 			m_freem(m);
619 		return;
620 
621 	default:
622 		if (ng_gif_input_orphan_p != NULL)
623 			(*ng_gif_input_orphan_p)(ifp, m, af);
624 		else
625 			m_freem(m);
626 		return;
627 	}
628 
629 	ifp->if_ipackets++;
630 	ifp->if_ibytes += m->m_pkthdr.len;
631 	netisr_dispatch(isr, m);
632 }
633 
634 /* XXX how should we handle IPv6 scope on SIOC[GS]IFPHYADDR? */
635 int
636 gif_ioctl(ifp, cmd, data)
637 	struct ifnet *ifp;
638 	u_long cmd;
639 	caddr_t data;
640 {
641 	struct gif_softc *sc  = ifp->if_softc;
642 	struct ifreq     *ifr = (struct ifreq*)data;
643 	int error = 0, size;
644 	u_int	options;
645 	struct sockaddr *dst, *src;
646 #ifdef	SIOCSIFMTU /* xxx */
647 	u_long mtu;
648 #endif
649 
650 	switch (cmd) {
651 	case SIOCSIFADDR:
652 		ifp->if_flags |= IFF_UP;
653 		break;
654 
655 	case SIOCSIFDSTADDR:
656 		break;
657 
658 	case SIOCADDMULTI:
659 	case SIOCDELMULTI:
660 		break;
661 
662 #ifdef	SIOCSIFMTU /* xxx */
663 	case SIOCGIFMTU:
664 		break;
665 
666 	case SIOCSIFMTU:
667 		mtu = ifr->ifr_mtu;
668 		if (mtu < GIF_MTU_MIN || mtu > GIF_MTU_MAX)
669 			return (EINVAL);
670 		ifp->if_mtu = mtu;
671 		break;
672 #endif /* SIOCSIFMTU */
673 
674 #ifdef INET
675 	case SIOCSIFPHYADDR:
676 #endif
677 #ifdef INET6
678 	case SIOCSIFPHYADDR_IN6:
679 #endif /* INET6 */
680 	case SIOCSLIFPHYADDR:
681 		switch (cmd) {
682 #ifdef INET
683 		case SIOCSIFPHYADDR:
684 			src = (struct sockaddr *)
685 				&(((struct in_aliasreq *)data)->ifra_addr);
686 			dst = (struct sockaddr *)
687 				&(((struct in_aliasreq *)data)->ifra_dstaddr);
688 			break;
689 #endif
690 #ifdef INET6
691 		case SIOCSIFPHYADDR_IN6:
692 			src = (struct sockaddr *)
693 				&(((struct in6_aliasreq *)data)->ifra_addr);
694 			dst = (struct sockaddr *)
695 				&(((struct in6_aliasreq *)data)->ifra_dstaddr);
696 			break;
697 #endif
698 		case SIOCSLIFPHYADDR:
699 			src = (struct sockaddr *)
700 				&(((struct if_laddrreq *)data)->addr);
701 			dst = (struct sockaddr *)
702 				&(((struct if_laddrreq *)data)->dstaddr);
703 			break;
704 		default:
705 			return EINVAL;
706 		}
707 
708 		/* sa_family must be equal */
709 		if (src->sa_family != dst->sa_family)
710 			return EINVAL;
711 
712 		/* validate sa_len */
713 		switch (src->sa_family) {
714 #ifdef INET
715 		case AF_INET:
716 			if (src->sa_len != sizeof(struct sockaddr_in))
717 				return EINVAL;
718 			break;
719 #endif
720 #ifdef INET6
721 		case AF_INET6:
722 			if (src->sa_len != sizeof(struct sockaddr_in6))
723 				return EINVAL;
724 			break;
725 #endif
726 		default:
727 			return EAFNOSUPPORT;
728 		}
729 		switch (dst->sa_family) {
730 #ifdef INET
731 		case AF_INET:
732 			if (dst->sa_len != sizeof(struct sockaddr_in))
733 				return EINVAL;
734 			break;
735 #endif
736 #ifdef INET6
737 		case AF_INET6:
738 			if (dst->sa_len != sizeof(struct sockaddr_in6))
739 				return EINVAL;
740 			break;
741 #endif
742 		default:
743 			return EAFNOSUPPORT;
744 		}
745 
746 		/* check sa_family looks sane for the cmd */
747 		switch (cmd) {
748 		case SIOCSIFPHYADDR:
749 			if (src->sa_family == AF_INET)
750 				break;
751 			return EAFNOSUPPORT;
752 #ifdef INET6
753 		case SIOCSIFPHYADDR_IN6:
754 			if (src->sa_family == AF_INET6)
755 				break;
756 			return EAFNOSUPPORT;
757 #endif /* INET6 */
758 		case SIOCSLIFPHYADDR:
759 			/* checks done in the above */
760 			break;
761 		}
762 
763 		error = gif_set_tunnel(GIF2IFP(sc), src, dst);
764 		break;
765 
766 #ifdef SIOCDIFPHYADDR
767 	case SIOCDIFPHYADDR:
768 		gif_delete_tunnel(GIF2IFP(sc));
769 		break;
770 #endif
771 
772 	case SIOCGIFPSRCADDR:
773 #ifdef INET6
774 	case SIOCGIFPSRCADDR_IN6:
775 #endif /* INET6 */
776 		if (sc->gif_psrc == NULL) {
777 			error = EADDRNOTAVAIL;
778 			goto bad;
779 		}
780 		src = sc->gif_psrc;
781 		switch (cmd) {
782 #ifdef INET
783 		case SIOCGIFPSRCADDR:
784 			dst = &ifr->ifr_addr;
785 			size = sizeof(ifr->ifr_addr);
786 			break;
787 #endif /* INET */
788 #ifdef INET6
789 		case SIOCGIFPSRCADDR_IN6:
790 			dst = (struct sockaddr *)
791 				&(((struct in6_ifreq *)data)->ifr_addr);
792 			size = sizeof(((struct in6_ifreq *)data)->ifr_addr);
793 			break;
794 #endif /* INET6 */
795 		default:
796 			error = EADDRNOTAVAIL;
797 			goto bad;
798 		}
799 		if (src->sa_len > size)
800 			return EINVAL;
801 		bcopy((caddr_t)src, (caddr_t)dst, src->sa_len);
802 #ifdef INET6
803 		if (dst->sa_family == AF_INET6) {
804 			error = sa6_recoverscope((struct sockaddr_in6 *)dst);
805 			if (error != 0)
806 				return (error);
807 		}
808 #endif
809 		break;
810 
811 	case SIOCGIFPDSTADDR:
812 #ifdef INET6
813 	case SIOCGIFPDSTADDR_IN6:
814 #endif /* INET6 */
815 		if (sc->gif_pdst == NULL) {
816 			error = EADDRNOTAVAIL;
817 			goto bad;
818 		}
819 		src = sc->gif_pdst;
820 		switch (cmd) {
821 #ifdef INET
822 		case SIOCGIFPDSTADDR:
823 			dst = &ifr->ifr_addr;
824 			size = sizeof(ifr->ifr_addr);
825 			break;
826 #endif /* INET */
827 #ifdef INET6
828 		case SIOCGIFPDSTADDR_IN6:
829 			dst = (struct sockaddr *)
830 				&(((struct in6_ifreq *)data)->ifr_addr);
831 			size = sizeof(((struct in6_ifreq *)data)->ifr_addr);
832 			break;
833 #endif /* INET6 */
834 		default:
835 			error = EADDRNOTAVAIL;
836 			goto bad;
837 		}
838 		if (src->sa_len > size)
839 			return EINVAL;
840 		bcopy((caddr_t)src, (caddr_t)dst, src->sa_len);
841 #ifdef INET6
842 		if (dst->sa_family == AF_INET6) {
843 			error = sa6_recoverscope((struct sockaddr_in6 *)dst);
844 			if (error != 0)
845 				return (error);
846 		}
847 #endif
848 		break;
849 
850 	case SIOCGLIFPHYADDR:
851 		if (sc->gif_psrc == NULL || sc->gif_pdst == NULL) {
852 			error = EADDRNOTAVAIL;
853 			goto bad;
854 		}
855 
856 		/* copy src */
857 		src = sc->gif_psrc;
858 		dst = (struct sockaddr *)
859 			&(((struct if_laddrreq *)data)->addr);
860 		size = sizeof(((struct if_laddrreq *)data)->addr);
861 		if (src->sa_len > size)
862 			return EINVAL;
863 		bcopy((caddr_t)src, (caddr_t)dst, src->sa_len);
864 
865 		/* copy dst */
866 		src = sc->gif_pdst;
867 		dst = (struct sockaddr *)
868 			&(((struct if_laddrreq *)data)->dstaddr);
869 		size = sizeof(((struct if_laddrreq *)data)->dstaddr);
870 		if (src->sa_len > size)
871 			return EINVAL;
872 		bcopy((caddr_t)src, (caddr_t)dst, src->sa_len);
873 		break;
874 
875 	case SIOCSIFFLAGS:
876 		/* if_ioctl() takes care of it */
877 		break;
878 
879 	case GIFGOPTS:
880 		options = sc->gif_options;
881 		error = copyout(&options, ifr->ifr_data,
882 				sizeof(options));
883 		break;
884 
885 	case GIFSOPTS:
886 		if ((error = priv_check(curthread, PRIV_NET_GIF)) != 0)
887 			break;
888 		error = copyin(ifr->ifr_data, &options, sizeof(options));
889 		if (error)
890 			break;
891 		if (options & ~GIF_OPTMASK)
892 			error = EINVAL;
893 		else
894 			sc->gif_options = options;
895 		break;
896 
897 	default:
898 		error = EINVAL;
899 		break;
900 	}
901  bad:
902 	return error;
903 }
904 
905 /*
906  * XXXRW: There's a general event-ordering issue here: the code to check
907  * if a given tunnel is already present happens before we perform a
908  * potentially blocking setup of the tunnel.  This code needs to be
909  * re-ordered so that the check and replacement can be atomic using
910  * a mutex.
911  */
912 int
913 gif_set_tunnel(ifp, src, dst)
914 	struct ifnet *ifp;
915 	struct sockaddr *src;
916 	struct sockaddr *dst;
917 {
918 	struct gif_softc *sc = ifp->if_softc;
919 	struct gif_softc *sc2;
920 	struct sockaddr *osrc, *odst, *sa;
921 	int error = 0;
922 
923 	mtx_lock(&gif_mtx);
924 	LIST_FOREACH(sc2, &V_gif_softc_list, gif_list) {
925 		if (sc2 == sc)
926 			continue;
927 		if (!sc2->gif_pdst || !sc2->gif_psrc)
928 			continue;
929 		if (sc2->gif_pdst->sa_family != dst->sa_family ||
930 		    sc2->gif_pdst->sa_len != dst->sa_len ||
931 		    sc2->gif_psrc->sa_family != src->sa_family ||
932 		    sc2->gif_psrc->sa_len != src->sa_len)
933 			continue;
934 
935 		/*
936 		 * Disallow parallel tunnels unless instructed
937 		 * otherwise.
938 		 */
939 		if (!V_parallel_tunnels &&
940 		    bcmp(sc2->gif_pdst, dst, dst->sa_len) == 0 &&
941 		    bcmp(sc2->gif_psrc, src, src->sa_len) == 0) {
942 			error = EADDRNOTAVAIL;
943 			mtx_unlock(&gif_mtx);
944 			goto bad;
945 		}
946 
947 		/* XXX both end must be valid? (I mean, not 0.0.0.0) */
948 	}
949 	mtx_unlock(&gif_mtx);
950 
951 	/* XXX we can detach from both, but be polite just in case */
952 	if (sc->gif_psrc)
953 		switch (sc->gif_psrc->sa_family) {
954 #ifdef INET
955 		case AF_INET:
956 			(void)in_gif_detach(sc);
957 			break;
958 #endif
959 #ifdef INET6
960 		case AF_INET6:
961 			(void)in6_gif_detach(sc);
962 			break;
963 #endif
964 		}
965 
966 	osrc = sc->gif_psrc;
967 	sa = (struct sockaddr *)malloc(src->sa_len, M_IFADDR, M_WAITOK);
968 	bcopy((caddr_t)src, (caddr_t)sa, src->sa_len);
969 	sc->gif_psrc = sa;
970 
971 	odst = sc->gif_pdst;
972 	sa = (struct sockaddr *)malloc(dst->sa_len, M_IFADDR, M_WAITOK);
973 	bcopy((caddr_t)dst, (caddr_t)sa, dst->sa_len);
974 	sc->gif_pdst = sa;
975 
976 	switch (sc->gif_psrc->sa_family) {
977 #ifdef INET
978 	case AF_INET:
979 		error = in_gif_attach(sc);
980 		break;
981 #endif
982 #ifdef INET6
983 	case AF_INET6:
984 		/*
985 		 * Check validity of the scope zone ID of the addresses, and
986 		 * convert it into the kernel internal form if necessary.
987 		 */
988 		error = sa6_embedscope((struct sockaddr_in6 *)sc->gif_psrc, 0);
989 		if (error != 0)
990 			break;
991 		error = sa6_embedscope((struct sockaddr_in6 *)sc->gif_pdst, 0);
992 		if (error != 0)
993 			break;
994 		error = in6_gif_attach(sc);
995 		break;
996 #endif
997 	}
998 	if (error) {
999 		/* rollback */
1000 		free((caddr_t)sc->gif_psrc, M_IFADDR);
1001 		free((caddr_t)sc->gif_pdst, M_IFADDR);
1002 		sc->gif_psrc = osrc;
1003 		sc->gif_pdst = odst;
1004 		goto bad;
1005 	}
1006 
1007 	if (osrc)
1008 		free((caddr_t)osrc, M_IFADDR);
1009 	if (odst)
1010 		free((caddr_t)odst, M_IFADDR);
1011 
1012  bad:
1013 	if (sc->gif_psrc && sc->gif_pdst)
1014 		ifp->if_drv_flags |= IFF_DRV_RUNNING;
1015 	else
1016 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1017 
1018 	return error;
1019 }
1020 
1021 void
1022 gif_delete_tunnel(ifp)
1023 	struct ifnet *ifp;
1024 {
1025 	struct gif_softc *sc = ifp->if_softc;
1026 
1027 	if (sc->gif_psrc) {
1028 		free((caddr_t)sc->gif_psrc, M_IFADDR);
1029 		sc->gif_psrc = NULL;
1030 	}
1031 	if (sc->gif_pdst) {
1032 		free((caddr_t)sc->gif_pdst, M_IFADDR);
1033 		sc->gif_pdst = NULL;
1034 	}
1035 	/* it is safe to detach from both */
1036 #ifdef INET
1037 	(void)in_gif_detach(sc);
1038 #endif
1039 #ifdef INET6
1040 	(void)in6_gif_detach(sc);
1041 #endif
1042 	ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1043 }
1044