xref: /netbsd-src/sys/net/if_tun.c (revision d20841bb642898112fe68f0ad3f7b26dddf56f07)
1 /*	$NetBSD: if_tun.c,v 1.67 2003/09/22 20:49:39 cl Exp $	*/
2 
3 /*
4  * Copyright (c) 1988, Julian Onions <jpo@cs.nott.ac.uk>
5  * Nottingham University 1987.
6  *
7  * This source may be freely distributed, however I would be interested
8  * in any changes that are made.
9  *
10  * This driver takes packets off the IP i/f and hands them up to a
11  * user process to have its wicked way with. This driver has its
12  * roots in a similar driver written by Phil Cockcroft (formerly) at
13  * UCL. This driver is based much more on read/write/poll mode of
14  * operation though.
15  */
16 
17 #include <sys/cdefs.h>
18 __KERNEL_RCSID(0, "$NetBSD: if_tun.c,v 1.67 2003/09/22 20:49:39 cl Exp $");
19 
20 #include "tun.h"
21 
22 #include "opt_inet.h"
23 #include "opt_ns.h"
24 
25 #include <sys/param.h>
26 #include <sys/proc.h>
27 #include <sys/systm.h>
28 #include <sys/mbuf.h>
29 #include <sys/buf.h>
30 #include <sys/protosw.h>
31 #include <sys/socket.h>
32 #include <sys/ioctl.h>
33 #include <sys/errno.h>
34 #include <sys/syslog.h>
35 #include <sys/select.h>
36 #include <sys/poll.h>
37 #include <sys/file.h>
38 #include <sys/signalvar.h>
39 #include <sys/conf.h>
40 
41 #include <machine/cpu.h>
42 
43 #include <net/if.h>
44 #include <net/if_ether.h>
45 #include <net/netisr.h>
46 #include <net/route.h>
47 
48 
49 #ifdef INET
50 #include <netinet/in.h>
51 #include <netinet/in_systm.h>
52 #include <netinet/in_var.h>
53 #include <netinet/ip.h>
54 #include <netinet/if_inarp.h>
55 #endif
56 
57 #ifdef NS
58 #include <netns/ns.h>
59 #include <netns/ns_if.h>
60 #endif
61 
62 #include "bpfilter.h"
63 #if NBPFILTER > 0
64 #include <sys/time.h>
65 #include <net/bpf.h>
66 #endif
67 
68 #include <net/if_tun.h>
69 
70 #define TUNDEBUG	if (tundebug) printf
71 int	tundebug = 0;
72 
73 extern int ifqmaxlen;
74 void	tunattach __P((int));
75 LIST_HEAD(, tun_softc) tun_softc_list;
76 static struct simplelock tun_softc_lock;
77 
78 int	tun_ioctl __P((struct ifnet *, u_long, caddr_t));
79 int	tun_output __P((struct ifnet *, struct mbuf *, struct sockaddr *,
80 		       struct rtentry *rt));
81 int	tun_clone_create __P((struct if_clone *, int));
82 void	tun_clone_destroy __P((struct ifnet *));
83 
84 struct if_clone tun_cloner =
85     IF_CLONE_INITIALIZER("tun", tun_clone_create, tun_clone_destroy);
86 
87 static void tunattach0 __P((struct tun_softc *));
88 static void tuninit __P((struct tun_softc *));
89 #ifdef ALTQ
90 static void tunstart __P((struct ifnet *));
91 #endif
92 static struct tun_softc *tun_find_unit __P((dev_t));
93 
94 dev_type_open(tunopen);
95 dev_type_close(tunclose);
96 dev_type_read(tunread);
97 dev_type_write(tunwrite);
98 dev_type_ioctl(tunioctl);
99 dev_type_poll(tunpoll);
100 dev_type_kqfilter(tunkqfilter);
101 
102 const struct cdevsw tun_cdevsw = {
103 	tunopen, tunclose, tunread, tunwrite, tunioctl,
104 	nostop, notty, tunpoll, nommap, tunkqfilter,
105 };
106 
107 void
108 tunattach(unused)
109 	int unused;
110 {
111 
112 	simple_lock_init(&tun_softc_lock);
113 	LIST_INIT(&tun_softc_list);
114 	if_clone_attach(&tun_cloner);
115 }
116 
117 int
118 tun_clone_create(ifc, unit)
119 	struct if_clone *ifc;
120 	int unit;
121 {
122 	struct tun_softc *sc;
123 
124 	sc = malloc(sizeof(struct tun_softc), M_DEVBUF, M_WAITOK);
125 	(void)memset(sc, 0, sizeof(struct tun_softc));
126 
127 	(void)snprintf(sc->tun_if.if_xname, sizeof(sc->tun_if.if_xname),
128 	    "%s%d", ifc->ifc_name, unit);
129 	sc->tun_unit = unit;
130 	simple_lock_init(&sc->tun_lock);
131 
132 	tunattach0(sc);
133 
134 	simple_lock(&tun_softc_lock);
135 	LIST_INSERT_HEAD(&tun_softc_list, sc, tun_list);
136 	simple_unlock(&tun_softc_lock);
137 
138 	return (0);
139 }
140 
141 void
142 tunattach0(sc)
143 	struct tun_softc *sc;
144 {
145 	struct ifnet *ifp = (void *)sc;
146 
147 	sc->tun_flags = TUN_INITED;
148 
149 	ifp = &sc->tun_if;
150 	ifp->if_softc = sc;
151 	ifp->if_mtu = TUNMTU;
152 	ifp->if_ioctl = tun_ioctl;
153 	ifp->if_output = tun_output;
154 #ifdef ALTQ
155 	ifp->if_start = tunstart;
156 #endif
157 	ifp->if_flags = IFF_POINTOPOINT;
158 	ifp->if_snd.ifq_maxlen = ifqmaxlen;
159 	ifp->if_collisions = 0;
160 	ifp->if_ierrors = 0;
161 	ifp->if_oerrors = 0;
162 	ifp->if_ipackets = 0;
163 	ifp->if_opackets = 0;
164 	ifp->if_ibytes   = 0;
165 	ifp->if_obytes   = 0;
166 	ifp->if_dlt = DLT_NULL;
167 	IFQ_SET_READY(&ifp->if_snd);
168 	if_attach(ifp);
169 	if_alloc_sadl(ifp);
170 #if NBPFILTER > 0
171 	bpfattach(ifp, DLT_NULL, sizeof(u_int32_t));
172 #endif
173 }
174 
175 void
176 tun_clone_destroy(ifp)
177 	struct ifnet *ifp;
178 {
179 	struct tun_softc *tp = (void *)ifp;
180 
181 	simple_lock(&tun_softc_lock);
182 	simple_lock(&tp->tun_lock);
183 	LIST_REMOVE(tp, tun_list);
184 	simple_unlock(&tp->tun_lock);
185 	simple_unlock(&tun_softc_lock);
186 
187 	if (tp->tun_flags & TUN_RWAIT) {
188 		tp->tun_flags &= ~TUN_RWAIT;
189 		wakeup((caddr_t)tp);
190 	}
191 	if (tp->tun_flags & TUN_ASYNC && tp->tun_pgid)
192 		fownsignal(tp->tun_pgid, SIGIO, POLL_HUP, 0, NULL);
193 
194 	selwakeup(&tp->tun_rsel);
195 
196 #if NBPFILTER > 0
197 	bpfdetach(ifp);
198 #endif
199 	if_detach(ifp);
200 
201 	free(tp, M_DEVBUF);
202 }
203 
204 static struct tun_softc *
205 tun_find_unit(dev)
206 	dev_t dev;
207 {
208 	struct tun_softc *tp;
209 	int unit = minor(dev);
210 
211 	simple_lock(&tun_softc_lock);
212 	LIST_FOREACH(tp, &tun_softc_list, tun_list)
213 		if (unit == tp->tun_unit)
214 			break;
215 	if (tp)
216 		simple_lock(&tp->tun_lock);
217 	simple_unlock(&tun_softc_lock);
218 
219 	return (tp);
220 }
221 
222 /*
223  * tunnel open - must be superuser & the device must be
224  * configured in
225  */
226 int
227 tunopen(dev, flag, mode, p)
228 	dev_t	dev;
229 	int	flag, mode;
230 	struct proc *p;
231 {
232 	struct ifnet	*ifp;
233 	struct tun_softc *tp;
234 	int	error;
235 
236 	if ((error = suser(p->p_ucred, &p->p_acflag)) != 0)
237 		return (error);
238 
239 	if (NTUN < 1)
240 		return (ENXIO);
241 
242 	tp = tun_find_unit(dev);
243 
244 	if (!tp) {
245 		(void)tun_clone_create(&tun_cloner, minor(dev));
246 		tp = tun_find_unit(dev);
247 	}
248 
249 	if (!tp)
250 		return (ENXIO);
251 
252 	if (tp->tun_flags & TUN_OPEN) {
253 		simple_unlock(&tp->tun_lock);
254 		return (EBUSY);
255 	}
256 
257 	ifp = &tp->tun_if;
258 	tp->tun_flags |= TUN_OPEN;
259 	TUNDEBUG("%s: open\n", ifp->if_xname);
260 	simple_unlock(&tp->tun_lock);
261 	return (0);
262 }
263 
264 /*
265  * tunclose - close the device - mark i/f down & delete
266  * routing info
267  */
268 int
269 tunclose(dev, flag, mode, p)
270 	dev_t	dev;
271 	int	flag;
272 	int	mode;
273 	struct proc *p;
274 {
275 	int	s;
276 	struct tun_softc *tp;
277 	struct ifnet	*ifp;
278 
279 	tp = tun_find_unit(dev);
280 
281 	/* interface was "destroyed" before the close */
282 	if (tp == NULL)
283 		return (0);
284 
285 	ifp = &tp->tun_if;
286 
287 	tp->tun_flags &= ~TUN_OPEN;
288 
289 	/*
290 	 * junk all pending output
291 	 */
292 	s = splnet();
293 	IFQ_PURGE(&ifp->if_snd);
294 	splx(s);
295 
296 	if (ifp->if_flags & IFF_UP) {
297 		s = splnet();
298 		if_down(ifp);
299 		if (ifp->if_flags & IFF_RUNNING) {
300 			/* find internet addresses and delete routes */
301 			struct ifaddr *ifa;
302 			TAILQ_FOREACH(ifa, &ifp->if_addrlist, ifa_list) {
303 #ifdef INET
304 				if (ifa->ifa_addr->sa_family == AF_INET) {
305 					rtinit(ifa, (int)RTM_DELETE,
306 					       tp->tun_flags & TUN_DSTADDR
307 							? RTF_HOST
308 							: 0);
309 				}
310 #endif
311 			}
312 		}
313 		splx(s);
314 	}
315 	tp->tun_pgid = 0;
316 	selnotify(&tp->tun_rsel, 0);
317 
318 	TUNDEBUG ("%s: closed\n", ifp->if_xname);
319 	simple_unlock(&tp->tun_lock);
320 	return (0);
321 }
322 
323 static void
324 tuninit(tp)
325 	struct tun_softc *tp;
326 {
327 	struct ifnet	*ifp = &tp->tun_if;
328 	struct ifaddr	*ifa;
329 
330 	TUNDEBUG("%s: tuninit\n", ifp->if_xname);
331 
332 	ifp->if_flags |= IFF_UP | IFF_RUNNING;
333 
334 	tp->tun_flags &= ~(TUN_IASET|TUN_DSTADDR);
335 	TAILQ_FOREACH(ifa, &ifp->if_addrlist, ifa_list) {
336 #ifdef INET
337 		if (ifa->ifa_addr->sa_family == AF_INET) {
338 			struct sockaddr_in *sin;
339 
340 			sin = satosin(ifa->ifa_addr);
341 			if (sin && sin->sin_addr.s_addr)
342 				tp->tun_flags |= TUN_IASET;
343 
344 			if (ifp->if_flags & IFF_POINTOPOINT) {
345 				sin = satosin(ifa->ifa_dstaddr);
346 				if (sin && sin->sin_addr.s_addr)
347 					tp->tun_flags |= TUN_DSTADDR;
348 			}
349 		}
350 #endif
351 	}
352 
353 	return;
354 }
355 
356 /*
357  * Process an ioctl request.
358  */
359 int
360 tun_ioctl(ifp, cmd, data)
361 	struct ifnet *ifp;
362 	u_long cmd;
363 	caddr_t	data;
364 {
365 	int		error = 0, s;
366 	struct tun_softc *tp = (struct tun_softc *)(ifp->if_softc);
367 
368 	simple_lock(&tp->tun_lock);
369 
370 	s = splnet();
371 	switch (cmd) {
372 	case SIOCSIFADDR:
373 		tuninit((struct tun_softc *)(ifp->if_softc));
374 		TUNDEBUG("%s: address set\n", ifp->if_xname);
375 		break;
376 	case SIOCSIFDSTADDR:
377 		tuninit((struct tun_softc *)(ifp->if_softc));
378 		TUNDEBUG("%s: destination address set\n", ifp->if_xname);
379 		break;
380 	case SIOCSIFBRDADDR:
381 		TUNDEBUG("%s: broadcast address set\n", ifp->if_xname);
382 		break;
383 	case SIOCSIFMTU: {
384 		struct ifreq *ifr = (struct ifreq *) data;
385 		if (ifr->ifr_mtu > TUNMTU || ifr->ifr_mtu < 576) {
386 		    error = EINVAL;
387 		    break;
388 		}
389 		TUNDEBUG("%s: interface mtu set\n", ifp->if_xname);
390 		ifp->if_mtu = ifr->ifr_mtu;
391 		break;
392 	}
393 	case SIOCADDMULTI:
394 	case SIOCDELMULTI: {
395 		struct ifreq *ifr = (struct ifreq *) data;
396 		if (ifr == 0) {
397 	        	error = EAFNOSUPPORT;           /* XXX */
398 			break;
399 		}
400 		switch (ifr->ifr_addr.sa_family) {
401 
402 #ifdef INET
403 		case AF_INET:
404 			break;
405 #endif
406 
407 		default:
408 			error = EAFNOSUPPORT;
409 			break;
410 		}
411 		break;
412 	}
413 	case SIOCSIFFLAGS:
414 		break;
415 	default:
416 		error = EINVAL;
417 	}
418 	splx(s);
419 	simple_unlock(&tp->tun_lock);
420 	return (error);
421 }
422 
423 /*
424  * tun_output - queue packets from higher level ready to put out.
425  */
426 int
427 tun_output(ifp, m0, dst, rt)
428 	struct ifnet   *ifp;
429 	struct mbuf    *m0;
430 	struct sockaddr *dst;
431 	struct rtentry *rt;
432 {
433 	struct tun_softc *tp = ifp->if_softc;
434 #ifdef INET
435 	int		s;
436 	int		error;
437 #endif
438 	int		mlen;
439 	ALTQ_DECL(struct altq_pktattr pktattr;)
440 
441 	simple_lock(&tp->tun_lock);
442 	TUNDEBUG ("%s: tun_output\n", ifp->if_xname);
443 
444 	if ((tp->tun_flags & TUN_READY) != TUN_READY) {
445 		TUNDEBUG ("%s: not ready 0%o\n", ifp->if_xname,
446 			  tp->tun_flags);
447 		m_freem (m0);
448 		simple_unlock(&tp->tun_lock);
449 		return (EHOSTDOWN);
450 	}
451 
452 	/*
453 	 * if the queueing discipline needs packet classification,
454 	 * do it before prepending link headers.
455 	 */
456 	IFQ_CLASSIFY(&ifp->if_snd, m0, dst->sa_family, &pktattr);
457 
458 #if NBPFILTER > 0
459 	if (ifp->if_bpf) {
460 		/*
461 		 * We need to prepend the address family as
462 		 * a four byte field.  Cons up a dummy header
463 		 * to pacify bpf.  This is safe because bpf
464 		 * will only read from the mbuf (i.e., it won't
465 		 * try to free it or keep a pointer to it).
466 		 */
467 		struct mbuf m;
468 		u_int32_t af = dst->sa_family;
469 
470 		m.m_flags = 0;
471 		m.m_next = m0;
472 		m.m_len = sizeof(af);
473 		m.m_data = (char *)&af;
474 
475 		bpf_mtap(ifp->if_bpf, &m);
476 	}
477 #endif
478 
479 	switch(dst->sa_family) {
480 #ifdef INET
481 	case AF_INET:
482 		if (tp->tun_flags & TUN_PREPADDR) {
483 			/* Simple link-layer header */
484 			M_PREPEND(m0, dst->sa_len, M_DONTWAIT);
485 			if (m0 == NULL) {
486 				IF_DROP(&ifp->if_snd);
487 				simple_unlock(&tp->tun_lock);
488 				return (ENOBUFS);
489 			}
490 			bcopy(dst, mtod(m0, char *), dst->sa_len);
491 		}
492 		/* FALLTHROUGH */
493 	case AF_UNSPEC:
494 		s = splnet();
495 		IFQ_ENQUEUE(&ifp->if_snd, m0, &pktattr, error);
496 		if (error) {
497 			splx(s);
498 			ifp->if_collisions++;
499 			return (error);
500 		}
501 		mlen = m0->m_pkthdr.len;
502 		splx(s);
503 		ifp->if_opackets++;
504 		ifp->if_obytes += mlen;
505 		break;
506 #endif
507 	default:
508 		m_freem(m0);
509 		simple_unlock(&tp->tun_lock);
510 		return (EAFNOSUPPORT);
511 	}
512 
513 	if (tp->tun_flags & TUN_RWAIT) {
514 		tp->tun_flags &= ~TUN_RWAIT;
515 		wakeup((caddr_t)tp);
516 	}
517 	if (tp->tun_flags & TUN_ASYNC && tp->tun_pgid)
518 		fownsignal(tp->tun_pgid, SIGIO, POLL_IN, POLLIN|POLLRDNORM,
519 		    NULL);
520 
521 	selnotify(&tp->tun_rsel, 0);
522 	simple_unlock(&tp->tun_lock);
523 	return (0);
524 }
525 
526 /*
527  * the cdevsw interface is now pretty minimal.
528  */
529 int
530 tunioctl(dev, cmd, data, flag, p)
531 	dev_t		dev;
532 	u_long		cmd;
533 	caddr_t		data;
534 	int		flag;
535 	struct proc	*p;
536 {
537 	int		s;
538 	struct tun_softc *tp;
539 	int error=0;
540 
541 	tp = tun_find_unit(dev);
542 
543 	/* interface was "destroyed" already */
544 	if (tp == NULL)
545 		return (ENXIO);
546 
547 	switch (cmd) {
548 	case TUNSDEBUG:
549 		tundebug = *(int *)data;
550 		break;
551 
552 	case TUNGDEBUG:
553 		*(int *)data = tundebug;
554 		break;
555 
556 	case TUNSIFMODE:
557 		switch (*(int *)data & (IFF_POINTOPOINT|IFF_BROADCAST)) {
558 		case IFF_POINTOPOINT:
559 		case IFF_BROADCAST:
560 			s = splnet();
561 			if (tp->tun_if.if_flags & IFF_UP) {
562 				splx(s);
563 				simple_unlock(&tp->tun_lock);
564 				return (EBUSY);
565 			}
566 			tp->tun_if.if_flags &=
567 				~(IFF_BROADCAST|IFF_POINTOPOINT|IFF_MULTICAST);
568 			tp->tun_if.if_flags |= *(int *)data;
569 			splx(s);
570 			break;
571 		default:
572 			simple_unlock(&tp->tun_lock);
573 			return (EINVAL);
574 		}
575 		break;
576 
577 	case TUNSLMODE:
578 		if (*(int *)data)
579 			tp->tun_flags |= TUN_PREPADDR;
580 		else
581 			tp->tun_flags &= ~TUN_PREPADDR;
582 		break;
583 
584 	case FIONBIO:
585 		if (*(int *)data)
586 			tp->tun_flags |= TUN_NBIO;
587 		else
588 			tp->tun_flags &= ~TUN_NBIO;
589 		break;
590 
591 	case FIOASYNC:
592 		if (*(int *)data)
593 			tp->tun_flags |= TUN_ASYNC;
594 		else
595 			tp->tun_flags &= ~TUN_ASYNC;
596 		break;
597 
598 	case FIONREAD:
599 		s = splnet();
600 		if (tp->tun_if.if_snd.ifq_head)
601 			*(int *)data = tp->tun_if.if_snd.ifq_head->m_pkthdr.len;
602 		else
603 			*(int *)data = 0;
604 		splx(s);
605 		break;
606 
607 	case TIOCSPGRP:
608 	case FIOSETOWN:
609 		error = fsetown(p, &tp->tun_pgid, cmd, data);
610 		break;
611 
612 	case TIOCGPGRP:
613 	case FIOGETOWN:
614 		error = fgetown(p, tp->tun_pgid, cmd, data);
615 		break;
616 
617 	default:
618 		simple_unlock(&tp->tun_lock);
619 		return (ENOTTY);
620 	}
621 	simple_unlock(&tp->tun_lock);
622 	return (error);
623 }
624 
625 /*
626  * The cdevsw read interface - reads a packet at a time, or at
627  * least as much of a packet as can be read.
628  */
629 int
630 tunread(dev, uio, ioflag)
631 	dev_t		dev;
632 	struct uio	*uio;
633 	int		ioflag;
634 {
635 	struct tun_softc *tp;
636 	struct ifnet	*ifp;
637 	struct mbuf	*m, *m0;
638 	int		error=0, len, s, index;
639 
640 	tp = tun_find_unit(dev);
641 
642 	/* interface was "destroyed" already */
643 	if (tp == NULL)
644 		return (ENXIO);
645 
646 	index = tp->tun_if.if_index;
647 	ifp = &tp->tun_if;
648 
649 	TUNDEBUG ("%s: read\n", ifp->if_xname);
650 	if ((tp->tun_flags & TUN_READY) != TUN_READY) {
651 		TUNDEBUG ("%s: not ready 0%o\n", ifp->if_xname, tp->tun_flags);
652 		simple_unlock(&tp->tun_lock);
653 		return EHOSTDOWN;
654 	}
655 
656 	tp->tun_flags &= ~TUN_RWAIT;
657 
658 	s = splnet();
659 	do {
660 		IFQ_DEQUEUE(&ifp->if_snd, m0);
661 		if (m0 == 0) {
662 			if (tp->tun_flags & TUN_NBIO) {
663 				splx(s);
664 				simple_unlock(&tp->tun_lock);
665 				return (EWOULDBLOCK);
666 			}
667 			tp->tun_flags |= TUN_RWAIT;
668 			simple_unlock(&tp->tun_lock);
669 			if (tsleep((caddr_t)tp, PZERO|PCATCH, "tunread", 0)) {
670 				splx(s);
671 				return (EINTR);
672 			} else {
673 				/*
674 				 * Maybe the interface was destroyed while
675 				 * we were sleeping, so let's ensure that
676 				 * we're looking at the same (valid) tun
677 				 * interface before looping.
678 				 */
679 				tp = tun_find_unit(dev);
680 				if (tp == NULL ||
681 				    tp->tun_if.if_index != index) {
682 					splx(s);
683 					if (tp)
684 						simple_unlock(&tp->tun_lock);
685 					return (ENXIO);
686 				}
687 			}
688 		}
689 	} while (m0 == 0);
690 	splx(s);
691 
692 	while (m0 && uio->uio_resid > 0 && error == 0) {
693 		len = min(uio->uio_resid, m0->m_len);
694 		if (len != 0)
695 			error = uiomove(mtod(m0, caddr_t), len, uio);
696 		MFREE(m0, m);
697 		m0 = m;
698 	}
699 
700 	if (m0) {
701 		TUNDEBUG("Dropping mbuf\n");
702 		m_freem(m0);
703 	}
704 	if (error)
705 		ifp->if_ierrors++;
706 	simple_unlock(&tp->tun_lock);
707 	return (error);
708 }
709 
710 /*
711  * the cdevsw write interface - an atomic write is a packet - or else!
712  */
713 int
714 tunwrite(dev, uio, ioflag)
715 	dev_t		dev;
716 	struct uio	*uio;
717 	int		ioflag;
718 {
719 	struct tun_softc *tp;
720 	struct ifnet	*ifp;
721 	struct mbuf	*top, **mp, *m;
722 	struct ifqueue	*ifq;
723 	struct sockaddr	dst;
724 	int		isr, error=0, s, tlen, mlen;
725 
726 	tp = tun_find_unit(dev);
727 
728 	/* interface was "destroyed" already */
729 	if (tp == NULL)
730 		return (ENXIO);
731 
732 	ifp = &tp->tun_if;
733 
734 	TUNDEBUG("%s: tunwrite\n", ifp->if_xname);
735 
736 	if (tp->tun_flags & TUN_PREPADDR) {
737 		if (uio->uio_resid < sizeof(dst)) {
738 			simple_unlock(&tp->tun_lock);
739 			return (EIO);
740 		}
741 		error = uiomove((caddr_t)&dst, sizeof(dst), uio);
742 		if (dst.sa_len > sizeof(dst)) {
743 			/* Duh.. */
744 			char discard;
745 			int n = dst.sa_len - sizeof(dst);
746 			while (n--)
747 				if ((error = uiomove(&discard, 1, uio)) != 0) {
748 					simple_unlock(&tp->tun_lock);
749 					return (error);
750 				}
751 		}
752 	} else {
753 #ifdef INET
754 		dst.sa_family = AF_INET;
755 #endif
756 	}
757 
758 	if (uio->uio_resid > TUNMTU) {
759 		TUNDEBUG("%s: len=%lu!\n", ifp->if_xname,
760 		    (unsigned long)uio->uio_resid);
761 		simple_unlock(&tp->tun_lock);
762 		return (EIO);
763 	}
764 
765 	switch (dst.sa_family) {
766 #ifdef INET
767 	case AF_INET:
768 		ifq = &ipintrq;
769 		isr = NETISR_IP;
770 		break;
771 #endif
772 	default:
773 		simple_unlock(&tp->tun_lock);
774 		return (EAFNOSUPPORT);
775 	}
776 
777 	tlen = uio->uio_resid;
778 
779 	/* get a header mbuf */
780 	MGETHDR(m, M_DONTWAIT, MT_DATA);
781 	if (m == NULL) {
782 		simple_unlock(&tp->tun_lock);
783 		return (ENOBUFS);
784 	}
785 	mlen = MHLEN;
786 
787 	top = 0;
788 	mp = &top;
789 	while (error == 0 && uio->uio_resid > 0) {
790 		m->m_len = min(mlen, uio->uio_resid);
791 		error = uiomove(mtod (m, caddr_t), m->m_len, uio);
792 		*mp = m;
793 		mp = &m->m_next;
794 		if (uio->uio_resid > 0) {
795 			MGET (m, M_DONTWAIT, MT_DATA);
796 			if (m == 0) {
797 				error = ENOBUFS;
798 				break;
799 			}
800 			mlen = MLEN;
801 		}
802 	}
803 	if (error) {
804 		if (top)
805 			m_freem (top);
806 		ifp->if_ierrors++;
807 		simple_unlock(&tp->tun_lock);
808 		return (error);
809 	}
810 
811 	top->m_pkthdr.len = tlen;
812 	top->m_pkthdr.rcvif = ifp;
813 
814 #if NBPFILTER > 0
815 	if (ifp->if_bpf) {
816 		/*
817 		 * We need to prepend the address family as
818 		 * a four byte field.  Cons up a dummy header
819 		 * to pacify bpf.  This is safe because bpf
820 		 * will only read from the mbuf (i.e., it won't
821 		 * try to free it or keep a pointer to it).
822 		 */
823 		struct mbuf m;
824 		u_int32_t af = AF_INET;
825 
826 		m.m_flags = 0;
827 		m.m_next = top;
828 		m.m_len = sizeof(af);
829 		m.m_data = (char *)&af;
830 
831 		bpf_mtap(ifp->if_bpf, &m);
832 	}
833 #endif
834 
835 	s = splnet();
836 	if (IF_QFULL(ifq)) {
837 		IF_DROP(ifq);
838 		splx(s);
839 		ifp->if_collisions++;
840 		m_freem(top);
841 		simple_unlock(&tp->tun_lock);
842 		return (ENOBUFS);
843 	}
844 	IF_ENQUEUE(ifq, top);
845 	splx(s);
846 	ifp->if_ipackets++;
847 	ifp->if_ibytes += tlen;
848 	schednetisr(isr);
849 	simple_unlock(&tp->tun_lock);
850 	return (error);
851 }
852 
853 #ifdef ALTQ
854 /*
855  * Start packet transmission on the interface.
856  * when the interface queue is rate-limited by ALTQ or TBR,
857  * if_start is needed to drain packets from the queue in order
858  * to notify readers when outgoing packets become ready.
859  */
860 static void
861 tunstart(ifp)
862 	struct ifnet *ifp;
863 {
864 	struct tun_softc *tp = ifp->if_softc;
865 	struct mbuf *m;
866 
867 	if (!ALTQ_IS_ENABLED(&ifp->if_snd) && !TBR_IS_ENABLED(&ifp->if_snd))
868 		return;
869 
870 	IFQ_POLL(&ifp->if_snd, m);
871 	if (m != NULL) {
872 		if (tp->tun_flags & TUN_RWAIT) {
873 			tp->tun_flags &= ~TUN_RWAIT;
874 			wakeup((caddr_t)tp);
875 		}
876 		if (tp->tun_flags & TUN_ASYNC && tp->tun_pgid)
877 			fownsignal(tp->tun_pgid, SIGIO, POLL_OUT,
878 				POLLOUT|POLLWRNORM, NULL);
879 
880 		selwakeup(&tp->tun_rsel);
881 	}
882 }
883 #endif /* ALTQ */
884 /*
885  * tunpoll - the poll interface, this is only useful on reads
886  * really. The write detect always returns true, write never blocks
887  * anyway, it either accepts the packet or drops it.
888  */
889 int
890 tunpoll(dev, events, p)
891 	dev_t		dev;
892 	int		events;
893 	struct proc	*p;
894 {
895 	struct tun_softc *tp;
896 	struct ifnet	*ifp;
897 	int		s, revents = 0;
898 
899 	tp = tun_find_unit(dev);
900 
901 	/* interface was "destroyed" already */
902 	if (tp == NULL)
903 		return (0);
904 
905 	ifp = &tp->tun_if;
906 
907 	s = splnet();
908 	TUNDEBUG("%s: tunpoll\n", ifp->if_xname);
909 
910 	if (events & (POLLIN | POLLRDNORM)) {
911 		if (IFQ_IS_EMPTY(&ifp->if_snd) == 0) {
912 			TUNDEBUG("%s: tunpoll q=%d\n", ifp->if_xname,
913 			    ifp->if_snd.ifq_len);
914 			revents |= events & (POLLIN | POLLRDNORM);
915 		} else {
916 			TUNDEBUG("%s: tunpoll waiting\n", ifp->if_xname);
917 			selrecord(p, &tp->tun_rsel);
918 		}
919 	}
920 
921 	if (events & (POLLOUT | POLLWRNORM))
922 		revents |= events & (POLLOUT | POLLWRNORM);
923 
924 	splx(s);
925 	simple_unlock(&tp->tun_lock);
926 	return (revents);
927 }
928 
929 static void
930 filt_tunrdetach(struct knote *kn)
931 {
932 	struct tun_softc *tp = kn->kn_hook;
933 	int s;
934 
935 	s = splnet();
936 	SLIST_REMOVE(&tp->tun_rsel.sel_klist, kn, knote, kn_selnext);
937 	splx(s);
938 }
939 
940 static int
941 filt_tunread(struct knote *kn, long hint)
942 {
943 	struct tun_softc *tp = kn->kn_hook;
944 	struct ifnet *ifp = &tp->tun_if;
945 	struct mbuf *m;
946 	int s;
947 
948 	s = splnet();
949 	IF_POLL(&ifp->if_snd, m);
950 	if (m == NULL) {
951 		splx(s);
952 		return (0);
953 	}
954 
955 	for (kn->kn_data = 0; m != NULL; m = m->m_next)
956 		kn->kn_data += m->m_len;
957 
958 	splx(s);
959 	return (1);
960 }
961 
962 static const struct filterops tunread_filtops =
963 	{ 1, NULL, filt_tunrdetach, filt_tunread };
964 
965 static const struct filterops tun_seltrue_filtops =
966 	{ 1, NULL, filt_tunrdetach, filt_seltrue };
967 
968 int
969 tunkqfilter(dev_t dev, struct knote *kn)
970 {
971 	struct tun_softc *tp = tun_find_unit(dev);
972 	struct klist *klist;
973 	int s;
974 
975 	switch (kn->kn_filter) {
976 	case EVFILT_READ:
977 		klist = &tp->tun_rsel.sel_klist;
978 		kn->kn_fop = &tunread_filtops;
979 		break;
980 
981 	case EVFILT_WRITE:
982 		klist = &tp->tun_rsel.sel_klist;
983 		kn->kn_fop = &tun_seltrue_filtops;
984 		break;
985 
986 	default:
987 		return (1);
988 	}
989 
990 	kn->kn_hook = tp;
991 
992 	s = splnet();
993 	SLIST_INSERT_HEAD(klist, kn, kn_selnext);
994 	splx(s);
995 
996 	return (0);
997 }
998