xref: /netbsd/sys/net/if_tun.c (revision c4a72b64)
1 /*	$NetBSD: if_tun.c,v 1.57 2002/11/26 18:51:19 christos 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 1.57 2002/11/26 18:51:19 christos 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_dlt = DLT_NULL;
165 	IFQ_SET_READY(&ifp->if_snd);
166 	if_attach(ifp);
167 	if_alloc_sadl(ifp);
168 #if NBPFILTER > 0
169 	bpfattach(ifp, DLT_NULL, sizeof(u_int32_t));
170 #endif
171 }
172 
173 void
174 tun_clone_destroy(ifp)
175 	struct ifnet *ifp;
176 {
177 	struct tun_softc *tp = (void *)ifp;
178 	struct proc *p;
179 
180 	simple_lock(&tun_softc_lock);
181 	simple_lock(&tp->tun_lock);
182 	LIST_REMOVE(tp, tun_list);
183 	simple_unlock(&tp->tun_lock);
184 	simple_unlock(&tun_softc_lock);
185 
186 	if (tp->tun_flags & TUN_RWAIT) {
187 		tp->tun_flags &= ~TUN_RWAIT;
188 		wakeup((caddr_t)tp);
189 	}
190 	if (tp->tun_flags & TUN_ASYNC && tp->tun_pgrp) {
191 		if (tp->tun_pgrp > 0)
192 			gsignal(tp->tun_pgrp, SIGIO);
193 		else if ((p = pfind(-tp->tun_pgrp)) != NULL)
194 			psignal(p, SIGIO);
195 	}
196 	selwakeup(&tp->tun_rsel);
197 
198 #if NBPFILTER > 0
199 	bpfdetach(ifp);
200 #endif
201 	if_detach(ifp);
202 
203 	free(tp, M_DEVBUF);
204 }
205 
206 static struct tun_softc *
207 tun_find_unit(dev)
208 	dev_t dev;
209 {
210 	struct tun_softc *tp;
211 	int unit = minor(dev);
212 
213 	simple_lock(&tun_softc_lock);
214 	LIST_FOREACH(tp, &tun_softc_list, tun_list)
215 		if (unit == tp->tun_unit)
216 			break;
217 	if (tp)
218 		simple_lock(&tp->tun_lock);
219 	simple_unlock(&tun_softc_lock);
220 
221 	return (tp);
222 }
223 
224 /*
225  * tunnel open - must be superuser & the device must be
226  * configured in
227  */
228 int
229 tunopen(dev, flag, mode, p)
230 	dev_t	dev;
231 	int	flag, mode;
232 	struct proc *p;
233 {
234 	struct ifnet	*ifp;
235 	struct tun_softc *tp;
236 	int	error;
237 
238 	if ((error = suser(p->p_ucred, &p->p_acflag)) != 0)
239 		return (error);
240 
241 	if (NTUN < 1)
242 		return (ENXIO);
243 
244 	tp = tun_find_unit(dev);
245 
246 	if (!tp) {
247 		(void)tun_clone_create(&tun_cloner, minor(dev));
248 		tp = tun_find_unit(dev);
249 	}
250 
251 	if (!tp)
252 		return (ENXIO);
253 
254 	if (tp->tun_flags & TUN_OPEN) {
255 		simple_unlock(&tp->tun_lock);
256 		return (EBUSY);
257 	}
258 
259 	ifp = &tp->tun_if;
260 	tp->tun_flags |= TUN_OPEN;
261 	TUNDEBUG("%s: open\n", ifp->if_xname);
262 	simple_unlock(&tp->tun_lock);
263 	return (0);
264 }
265 
266 /*
267  * tunclose - close the device - mark i/f down & delete
268  * routing info
269  */
270 int
271 tunclose(dev, flag, mode, p)
272 	dev_t	dev;
273 	int	flag;
274 	int	mode;
275 	struct proc *p;
276 {
277 	int	s;
278 	struct tun_softc *tp;
279 	struct ifnet	*ifp;
280 
281 	tp = tun_find_unit(dev);
282 
283 	/* interface was "destroyed" before the close */
284 	if (tp == NULL)
285 		return (0);
286 
287 	ifp = &tp->tun_if;
288 
289 	tp->tun_flags &= ~TUN_OPEN;
290 
291 	/*
292 	 * junk all pending output
293 	 */
294 	s = splnet();
295 	IFQ_PURGE(&ifp->if_snd);
296 	splx(s);
297 
298 	if (ifp->if_flags & IFF_UP) {
299 		s = splnet();
300 		if_down(ifp);
301 		if (ifp->if_flags & IFF_RUNNING) {
302 			/* find internet addresses and delete routes */
303 			struct ifaddr *ifa;
304 			TAILQ_FOREACH(ifa, &ifp->if_addrlist, ifa_list) {
305 #ifdef INET
306 				if (ifa->ifa_addr->sa_family == AF_INET) {
307 					rtinit(ifa, (int)RTM_DELETE,
308 					       tp->tun_flags & TUN_DSTADDR
309 							? RTF_HOST
310 							: 0);
311 				}
312 #endif
313 			}
314 		}
315 		splx(s);
316 	}
317 	tp->tun_pgrp = 0;
318 	selnotify(&tp->tun_rsel, 0);
319 
320 	TUNDEBUG ("%s: closed\n", ifp->if_xname);
321 	simple_unlock(&tp->tun_lock);
322 	return (0);
323 }
324 
325 static void
326 tuninit(tp)
327 	struct tun_softc *tp;
328 {
329 	struct ifnet	*ifp = &tp->tun_if;
330 	struct ifaddr	*ifa;
331 
332 	TUNDEBUG("%s: tuninit\n", ifp->if_xname);
333 
334 	ifp->if_flags |= IFF_UP | IFF_RUNNING;
335 
336 	tp->tun_flags &= ~(TUN_IASET|TUN_DSTADDR);
337 	TAILQ_FOREACH(ifa, &ifp->if_addrlist, ifa_list) {
338 #ifdef INET
339 		if (ifa->ifa_addr->sa_family == AF_INET) {
340 			struct sockaddr_in *sin;
341 
342 			sin = satosin(ifa->ifa_addr);
343 			if (sin && sin->sin_addr.s_addr)
344 				tp->tun_flags |= TUN_IASET;
345 
346 			if (ifp->if_flags & IFF_POINTOPOINT) {
347 				sin = satosin(ifa->ifa_dstaddr);
348 				if (sin && sin->sin_addr.s_addr)
349 					tp->tun_flags |= TUN_DSTADDR;
350 			}
351 		}
352 #endif
353 	}
354 
355 	return;
356 }
357 
358 /*
359  * Process an ioctl request.
360  */
361 int
362 tun_ioctl(ifp, cmd, data)
363 	struct ifnet *ifp;
364 	u_long cmd;
365 	caddr_t	data;
366 {
367 	int		error = 0, s;
368 	struct tun_softc *tp = (struct tun_softc *)(ifp->if_softc);
369 
370 	simple_lock(&tp->tun_lock);
371 
372 	s = splnet();
373 	switch(cmd) {
374 	case SIOCSIFADDR:
375 		tuninit((struct tun_softc *)(ifp->if_softc));
376 		TUNDEBUG("%s: address set\n", ifp->if_xname);
377 		break;
378 	case SIOCSIFDSTADDR:
379 		tuninit((struct tun_softc *)(ifp->if_softc));
380 		TUNDEBUG("%s: destination address set\n", ifp->if_xname);
381 		break;
382 	case SIOCSIFBRDADDR:
383 		TUNDEBUG("%s: broadcast address set\n", ifp->if_xname);
384 		break;
385 	case SIOCSIFMTU: {
386 		struct ifreq *ifr = (struct ifreq *) data;
387 		if (ifr->ifr_mtu > TUNMTU || ifr->ifr_mtu < 576) {
388 		    error = EINVAL;
389 		    break;
390 		}
391 		TUNDEBUG("%s: interface mtu set\n", ifp->if_xname);
392 		ifp->if_mtu = ifr->ifr_mtu;
393 		break;
394 	}
395 	case SIOCADDMULTI:
396 	case SIOCDELMULTI: {
397 		struct ifreq *ifr = (struct ifreq *) data;
398 		if (ifr == 0) {
399 	        	error = EAFNOSUPPORT;           /* XXX */
400 			break;
401 		}
402 		switch (ifr->ifr_addr.sa_family) {
403 
404 #ifdef INET
405 		case AF_INET:
406 			break;
407 #endif
408 
409 		default:
410 			error = EAFNOSUPPORT;
411 			break;
412 		}
413 		break;
414 	}
415 	case SIOCSIFFLAGS:
416 		break;
417 	default:
418 		error = EINVAL;
419 	}
420 	splx(s);
421 	simple_unlock(&tp->tun_lock);
422 	return (error);
423 }
424 
425 /*
426  * tun_output - queue packets from higher level ready to put out.
427  */
428 int
429 tun_output(ifp, m0, dst, rt)
430 	struct ifnet   *ifp;
431 	struct mbuf    *m0;
432 	struct sockaddr *dst;
433 	struct rtentry *rt;
434 {
435 	struct tun_softc *tp = ifp->if_softc;
436 	struct proc	*p;
437 #ifdef INET
438 	int		s;
439 	int		error;
440 #endif
441 	ALTQ_DECL(struct altq_pktattr pktattr;)
442 
443 	simple_lock(&tp->tun_lock);
444 	TUNDEBUG ("%s: tun_output\n", ifp->if_xname);
445 
446 	if ((tp->tun_flags & TUN_READY) != TUN_READY) {
447 		TUNDEBUG ("%s: not ready 0%o\n", ifp->if_xname,
448 			  tp->tun_flags);
449 		m_freem (m0);
450 		simple_unlock(&tp->tun_lock);
451 		return (EHOSTDOWN);
452 	}
453 
454 	/*
455 	 * if the queueing discipline needs packet classification,
456 	 * do it before prepending link headers.
457 	 */
458 	IFQ_CLASSIFY(&ifp->if_snd, m0, dst->sa_family, &pktattr);
459 
460 #if NBPFILTER > 0
461 	if (ifp->if_bpf) {
462 		/*
463 		 * We need to prepend the address family as
464 		 * a four byte field.  Cons up a dummy header
465 		 * to pacify bpf.  This is safe because bpf
466 		 * will only read from the mbuf (i.e., it won't
467 		 * try to free it or keep a pointer to it).
468 		 */
469 		struct mbuf m;
470 		u_int32_t af = dst->sa_family;
471 
472 		m.m_next = m0;
473 		m.m_len = sizeof(af);
474 		m.m_data = (char *)&af;
475 
476 		bpf_mtap(ifp->if_bpf, &m);
477 	}
478 #endif
479 
480 	switch(dst->sa_family) {
481 #ifdef INET
482 	case AF_INET:
483 		if (tp->tun_flags & TUN_PREPADDR) {
484 			/* Simple link-layer header */
485 			M_PREPEND(m0, dst->sa_len, M_DONTWAIT);
486 			if (m0 == NULL) {
487 				IF_DROP(&ifp->if_snd);
488 				simple_unlock(&tp->tun_lock);
489 				return (ENOBUFS);
490 			}
491 			bcopy(dst, mtod(m0, char *), dst->sa_len);
492 		}
493 		/* FALLTHROUGH */
494 	case AF_UNSPEC:
495 		s = splnet();
496 		IFQ_ENQUEUE(&ifp->if_snd, m0, &pktattr, error);
497 		if (error) {
498 			splx(s);
499 			ifp->if_collisions++;
500 			return (error);
501 		}
502 		splx(s);
503 		ifp->if_opackets++;
504 		break;
505 #endif
506 	default:
507 		m_freem(m0);
508 		simple_unlock(&tp->tun_lock);
509 		return (EAFNOSUPPORT);
510 	}
511 
512 	if (tp->tun_flags & TUN_RWAIT) {
513 		tp->tun_flags &= ~TUN_RWAIT;
514 		wakeup((caddr_t)tp);
515 	}
516 	if (tp->tun_flags & TUN_ASYNC && tp->tun_pgrp) {
517 		if (tp->tun_pgrp > 0)
518 			gsignal(tp->tun_pgrp, SIGIO);
519 		else if ((p = pfind(-tp->tun_pgrp)) != NULL)
520 			psignal(p, SIGIO);
521 	}
522 	selnotify(&tp->tun_rsel, 0);
523 	simple_unlock(&tp->tun_lock);
524 	return (0);
525 }
526 
527 /*
528  * the cdevsw interface is now pretty minimal.
529  */
530 int
531 tunioctl(dev, cmd, data, flag, p)
532 	dev_t		dev;
533 	u_long		cmd;
534 	caddr_t		data;
535 	int		flag;
536 	struct proc	*p;
537 {
538 	int		s;
539 	struct tun_softc *tp;
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 		tp->tun_pgrp = *(int *)data;
609 		break;
610 
611 	case TIOCGPGRP:
612 		*(int *)data = tp->tun_pgrp;
613 		break;
614 
615 	default:
616 		simple_unlock(&tp->tun_lock);
617 		return (ENOTTY);
618 	}
619 	simple_unlock(&tp->tun_lock);
620 	return (0);
621 }
622 
623 /*
624  * The cdevsw read interface - reads a packet at a time, or at
625  * least as much of a packet as can be read.
626  */
627 int
628 tunread(dev, uio, ioflag)
629 	dev_t		dev;
630 	struct uio	*uio;
631 	int		ioflag;
632 {
633 	struct tun_softc *tp;
634 	struct ifnet	*ifp;
635 	struct mbuf	*m, *m0;
636 	int		error=0, len, s, index;
637 
638 	tp = tun_find_unit(dev);
639 
640 	/* interface was "destroyed" already */
641 	if (tp == NULL)
642 		return (ENXIO);
643 
644 	index = tp->tun_if.if_index;
645 	ifp = &tp->tun_if;
646 
647 	TUNDEBUG ("%s: read\n", ifp->if_xname);
648 	if ((tp->tun_flags & TUN_READY) != TUN_READY) {
649 		TUNDEBUG ("%s: not ready 0%o\n", ifp->if_xname, tp->tun_flags);
650 		simple_unlock(&tp->tun_lock);
651 		return EHOSTDOWN;
652 	}
653 
654 	tp->tun_flags &= ~TUN_RWAIT;
655 
656 	s = splnet();
657 	do {
658 		IFQ_DEQUEUE(&ifp->if_snd, m0);
659 		if (m0 == 0) {
660 			if (tp->tun_flags & TUN_NBIO) {
661 				splx(s);
662 				simple_unlock(&tp->tun_lock);
663 				return (EWOULDBLOCK);
664 			}
665 			tp->tun_flags |= TUN_RWAIT;
666 			simple_unlock(&tp->tun_lock);
667 			if (tsleep((caddr_t)tp, PZERO|PCATCH, "tunread", 0)) {
668 				splx(s);
669 				return (EINTR);
670 			} else {
671 				/*
672 				 * Maybe the interface was destroyed while
673 				 * we were sleeping, so let's ensure that
674 				 * we're looking at the same (valid) tun
675 				 * interface before looping.
676 				 */
677 				tp = tun_find_unit(dev);
678 				if (tp == NULL ||
679 				    tp->tun_if.if_index != index) {
680 					splx(s);
681 					if (tp)
682 						simple_unlock(&tp->tun_lock);
683 					return (ENXIO);
684 				}
685 			}
686 		}
687 	} while (m0 == 0);
688 	splx(s);
689 
690 	while (m0 && uio->uio_resid > 0 && error == 0) {
691 		len = min(uio->uio_resid, m0->m_len);
692 		if (len != 0)
693 			error = uiomove(mtod(m0, caddr_t), len, uio);
694 		MFREE(m0, m);
695 		m0 = m;
696 	}
697 
698 	if (m0) {
699 		TUNDEBUG("Dropping mbuf\n");
700 		m_freem(m0);
701 	}
702 	if (error)
703 		ifp->if_ierrors++;
704 	simple_unlock(&tp->tun_lock);
705 	return (error);
706 }
707 
708 /*
709  * the cdevsw write interface - an atomic write is a packet - or else!
710  */
711 int
712 tunwrite(dev, uio, ioflag)
713 	dev_t		dev;
714 	struct uio	*uio;
715 	int		ioflag;
716 {
717 	struct tun_softc *tp;
718 	struct ifnet	*ifp;
719 	struct mbuf	*top, **mp, *m;
720 	struct ifqueue	*ifq;
721 	struct sockaddr	dst;
722 	int		isr, error=0, s, tlen, mlen;
723 
724 	tp = tun_find_unit(dev);
725 
726 	/* interface was "destroyed" already */
727 	if (tp == NULL)
728 		return (ENXIO);
729 
730 	ifp = &tp->tun_if;
731 
732 	TUNDEBUG("%s: tunwrite\n", ifp->if_xname);
733 
734 	if (tp->tun_flags & TUN_PREPADDR) {
735 		if (uio->uio_resid < sizeof(dst)) {
736 			simple_unlock(&tp->tun_lock);
737 			return (EIO);
738 		}
739 		error = uiomove((caddr_t)&dst, sizeof(dst), uio);
740 		if (dst.sa_len > sizeof(dst)) {
741 			/* Duh.. */
742 			char discard;
743 			int n = dst.sa_len - sizeof(dst);
744 			while (n--)
745 				if ((error = uiomove(&discard, 1, uio)) != 0) {
746 					simple_unlock(&tp->tun_lock);
747 					return (error);
748 				}
749 		}
750 	} else {
751 #ifdef INET
752 		dst.sa_family = AF_INET;
753 #endif
754 	}
755 
756 	if (uio->uio_resid > TUNMTU) {
757 		TUNDEBUG("%s: len=%lu!\n", ifp->if_xname,
758 		    (unsigned long)uio->uio_resid);
759 		simple_unlock(&tp->tun_lock);
760 		return (EIO);
761 	}
762 
763 	switch (dst.sa_family) {
764 #ifdef INET
765 	case AF_INET:
766 		ifq = &ipintrq;
767 		isr = NETISR_IP;
768 		break;
769 #endif
770 	default:
771 		simple_unlock(&tp->tun_lock);
772 		return (EAFNOSUPPORT);
773 	}
774 
775 	tlen = uio->uio_resid;
776 
777 	/* get a header mbuf */
778 	MGETHDR(m, M_DONTWAIT, MT_DATA);
779 	if (m == NULL) {
780 		simple_unlock(&tp->tun_lock);
781 		return (ENOBUFS);
782 	}
783 	mlen = MHLEN;
784 
785 	top = 0;
786 	mp = &top;
787 	while (error == 0 && uio->uio_resid > 0) {
788 		m->m_len = min(mlen, uio->uio_resid);
789 		error = uiomove(mtod (m, caddr_t), m->m_len, uio);
790 		*mp = m;
791 		mp = &m->m_next;
792 		if (uio->uio_resid > 0) {
793 			MGET (m, M_DONTWAIT, MT_DATA);
794 			if (m == 0) {
795 				error = ENOBUFS;
796 				break;
797 			}
798 			mlen = MLEN;
799 		}
800 	}
801 	if (error) {
802 		if (top)
803 			m_freem (top);
804 		ifp->if_ierrors++;
805 		simple_unlock(&tp->tun_lock);
806 		return (error);
807 	}
808 
809 	top->m_pkthdr.len = tlen;
810 	top->m_pkthdr.rcvif = ifp;
811 
812 #if NBPFILTER > 0
813 	if (ifp->if_bpf) {
814 		/*
815 		 * We need to prepend the address family as
816 		 * a four byte field.  Cons up a dummy header
817 		 * to pacify bpf.  This is safe because bpf
818 		 * will only read from the mbuf (i.e., it won't
819 		 * try to free it or keep a pointer to it).
820 		 */
821 		struct mbuf m;
822 		u_int32_t af = AF_INET;
823 
824 		m.m_next = top;
825 		m.m_len = sizeof(af);
826 		m.m_data = (char *)&af;
827 
828 		bpf_mtap(ifp->if_bpf, &m);
829 	}
830 #endif
831 
832 	s = splnet();
833 	if (IF_QFULL(ifq)) {
834 		IF_DROP(ifq);
835 		splx(s);
836 		ifp->if_collisions++;
837 		m_freem(top);
838 		simple_unlock(&tp->tun_lock);
839 		return (ENOBUFS);
840 	}
841 	IF_ENQUEUE(ifq, top);
842 	splx(s);
843 	ifp->if_ipackets++;
844 	schednetisr(isr);
845 	simple_unlock(&tp->tun_lock);
846 	return (error);
847 }
848 
849 #ifdef ALTQ
850 /*
851  * Start packet transmission on the interface.
852  * when the interface queue is rate-limited by ALTQ or TBR,
853  * if_start is needed to drain packets from the queue in order
854  * to notify readers when outgoing packets become ready.
855  */
856 static void
857 tunstart(ifp)
858 	struct ifnet *ifp;
859 {
860 	struct tun_softc *tp = ifp->if_softc;
861 	struct mbuf *m;
862 	struct proc	*p;
863 
864 	if (!ALTQ_IS_ENABLED(&ifp->if_snd) && !TBR_IS_ENABLED(&ifp->if_snd))
865 		return;
866 
867 	IFQ_POLL(&ifp->if_snd, m);
868 	if (m != NULL) {
869 		if (tp->tun_flags & TUN_RWAIT) {
870 			tp->tun_flags &= ~TUN_RWAIT;
871 			wakeup((caddr_t)tp);
872 		}
873 		if (tp->tun_flags & TUN_ASYNC && tp->tun_pgrp) {
874 			if (tp->tun_pgrp > 0)
875 				gsignal(tp->tun_pgrp, SIGIO);
876 			else if ((p = pfind(-tp->tun_pgrp)) != NULL)
877 				psignal(p, SIGIO);
878 		}
879 		selwakeup(&tp->tun_rsel);
880 	}
881 }
882 #endif /* ALTQ */
883 /*
884  * tunpoll - the poll interface, this is only useful on reads
885  * really. The write detect always returns true, write never blocks
886  * anyway, it either accepts the packet or drops it.
887  */
888 int
889 tunpoll(dev, events, p)
890 	dev_t		dev;
891 	int		events;
892 	struct proc	*p;
893 {
894 	struct tun_softc *tp;
895 	struct ifnet	*ifp;
896 	int		s, revents = 0;
897 
898 	tp = tun_find_unit(dev);
899 
900 	/* interface was "destroyed" already */
901 	if (tp == NULL)
902 		return (0);
903 
904 	ifp = &tp->tun_if;
905 
906 	s = splnet();
907 	TUNDEBUG("%s: tunpoll\n", ifp->if_xname);
908 
909 	if (events & (POLLIN | POLLRDNORM)) {
910 		if (IFQ_IS_EMPTY(&ifp->if_snd) == 0) {
911 			TUNDEBUG("%s: tunpoll q=%d\n", ifp->if_xname,
912 			    ifp->if_snd.ifq_len);
913 			revents |= events & (POLLIN | POLLRDNORM);
914 		} else {
915 			TUNDEBUG("%s: tunpoll waiting\n", ifp->if_xname);
916 			selrecord(p, &tp->tun_rsel);
917 		}
918 	}
919 
920 	if (events & (POLLOUT | POLLWRNORM))
921 		revents |= events & (POLLOUT | POLLWRNORM);
922 
923 	splx(s);
924 	simple_unlock(&tp->tun_lock);
925 	return (revents);
926 }
927 
928 static void
929 filt_tunrdetach(struct knote *kn)
930 {
931 	struct tun_softc *tp = kn->kn_hook;
932 	int s;
933 
934 	s = splnet();
935 	SLIST_REMOVE(&tp->tun_rsel.sel_klist, kn, knote, kn_selnext);
936 	splx(s);
937 }
938 
939 static int
940 filt_tunread(struct knote *kn, long hint)
941 {
942 	struct tun_softc *tp = kn->kn_hook;
943 	struct ifnet *ifp = &tp->tun_if;
944 	struct mbuf *m;
945 	int s;
946 
947 	s = splnet();
948 	IF_POLL(&ifp->if_snd, m);
949 	if (m == NULL) {
950 		splx(s);
951 		return (0);
952 	}
953 
954 	for (kn->kn_data = 0; m != NULL; m = m->m_next)
955 		kn->kn_data += m->m_len;
956 
957 	splx(s);
958 	return (1);
959 }
960 
961 static const struct filterops tunread_filtops =
962 	{ 1, NULL, filt_tunrdetach, filt_tunread };
963 
964 static const struct filterops tun_seltrue_filtops =
965 	{ 1, NULL, filt_tunrdetach, filt_seltrue };
966 
967 int
968 tunkqfilter(dev_t dev, struct knote *kn)
969 {
970 	struct tun_softc *tp = tun_find_unit(dev);
971 	struct klist *klist;
972 	int s;
973 
974 	switch (kn->kn_filter) {
975 	case EVFILT_READ:
976 		klist = &tp->tun_rsel.sel_klist;
977 		kn->kn_fop = &tunread_filtops;
978 		break;
979 
980 	case EVFILT_WRITE:
981 		klist = &tp->tun_rsel.sel_klist;
982 		kn->kn_fop = &tun_seltrue_filtops;
983 		break;
984 
985 	default:
986 		return (1);
987 	}
988 
989 	kn->kn_hook = tp;
990 
991 	s = splnet();
992 	SLIST_INSERT_HEAD(klist, kn, kn_selnext);
993 	splx(s);
994 
995 	return (0);
996 }
997