if_tun.c revision 121816
1/*	$NetBSD: if_tun.c,v 1.14 1994/06/29 06:36:25 cgd 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 it's
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 * $FreeBSD: head/sys/net/if_tun.c 121816 2003-10-31 18:32:15Z brooks $
17 */
18
19#include "opt_atalk.h"
20#include "opt_inet.h"
21#include "opt_inet6.h"
22#include "opt_ipx.h"
23#include "opt_mac.h"
24
25#include <sys/param.h>
26#include <sys/proc.h>
27#include <sys/systm.h>
28#include <sys/mac.h>
29#include <sys/mbuf.h>
30#include <sys/module.h>
31#include <sys/socket.h>
32#include <sys/filio.h>
33#include <sys/sockio.h>
34#include <sys/ttycom.h>
35#include <sys/poll.h>
36#include <sys/signalvar.h>
37#include <sys/filedesc.h>
38#include <sys/kernel.h>
39#include <sys/sysctl.h>
40#include <sys/conf.h>
41#include <sys/uio.h>
42#include <sys/vnode.h>
43#include <sys/malloc.h>
44#include <machine/bus.h>	/* XXX Shouldn't really be required ! */
45#include <sys/random.h>
46#include <sys/rman.h>
47
48#include <net/if.h>
49#include <net/if_types.h>
50#include <net/netisr.h>
51#include <net/route.h>
52#ifdef INET
53#include <netinet/in.h>
54#endif
55#include <net/bpf.h>
56#include <net/if_tunvar.h>
57#include <net/if_tun.h>
58
59#define TUNDEBUG	if (tundebug) if_printf
60#define	TUNNAME		"tun"
61
62static MALLOC_DEFINE(M_TUN, TUNNAME, "Tunnel Interface");
63static int tundebug = 0;
64static struct tun_softc *tunhead = NULL;
65static struct rman tununits;
66static udev_t tunbasedev = NOUDEV;
67SYSCTL_INT(_debug, OID_AUTO, if_tun_debug, CTLFLAG_RW, &tundebug, 0, "");
68
69static void	tunclone(void *arg, char *name, int namelen, dev_t *dev);
70static void	tuncreate(dev_t dev);
71static int	tunifioctl(struct ifnet *, u_long, caddr_t);
72static int	tuninit(struct ifnet *);
73static int	tunmodevent(module_t, int, void *);
74static int	tunoutput(struct ifnet *, struct mbuf *, struct sockaddr *,
75		    struct rtentry *rt);
76static void	tunstart(struct ifnet *);
77
78static d_open_t		tunopen;
79static d_close_t	tunclose;
80static d_read_t		tunread;
81static d_write_t	tunwrite;
82static d_ioctl_t	tunioctl;
83static d_poll_t		tunpoll;
84
85#define CDEV_MAJOR 52
86static struct cdevsw tun_cdevsw = {
87	.d_open =	tunopen,
88	.d_close =	tunclose,
89	.d_read =	tunread,
90	.d_write =	tunwrite,
91	.d_ioctl =	tunioctl,
92	.d_poll =	tunpoll,
93	.d_name =	TUNNAME,
94	.d_maj =	CDEV_MAJOR,
95};
96
97static void
98tunclone(void *arg, char *name, int namelen, dev_t *dev)
99{
100	struct resource *r;
101	int err;
102	int u;
103
104	if (*dev != NODEV)
105		return;
106
107	if (strcmp(name, TUNNAME) == 0) {
108		r = rman_reserve_resource(&tununits, 0, IF_MAXUNIT, 1,
109		    RF_ALLOCATED | RF_ACTIVE, NULL);
110		u = rman_get_start(r);
111		err = rman_release_resource(r);
112		KASSERT(err == 0, ("Unexpected failure releasing resource"));
113		*dev = makedev(CDEV_MAJOR, unit2minor(u));
114		if ((*dev)->si_flags & SI_NAMED)
115			return;	/* Already make_dev()d */
116	} else if (dev_stdclone(name, NULL, TUNNAME, &u) != 1)
117		return;	/* Don't recognise the name */
118
119	*dev = make_dev(&tun_cdevsw, unit2minor(u),
120	    UID_ROOT, GID_WHEEL, 0600, "tun%d", u);
121
122	/*
123	 * All devices depend on tunbasedev so that we can simply
124	 * destroy_dev() this device at module unload time to get
125	 * rid of all our make_dev()d resources.
126	 */
127	if (tunbasedev == NOUDEV)
128		tunbasedev = (*dev)->si_udev;
129	else {
130		(*dev)->si_flags |= SI_CHEAPCLONE;
131		dev_depends(udev2dev(tunbasedev, 0), *dev);
132	}
133}
134
135static int
136tunmodevent(module_t mod, int type, void *data)
137{
138	static eventhandler_tag tag;
139	struct tun_softc *tp;
140	dev_t dev;
141	int err;
142
143	switch (type) {
144	case MOD_LOAD:
145		tag = EVENTHANDLER_REGISTER(dev_clone, tunclone, 0, 1000);
146		if (tag == NULL)
147			return (ENOMEM);
148		tununits.rm_type = RMAN_ARRAY;
149		tununits.rm_descr = "open if_tun units";
150		err = rman_init(&tununits);
151		if (err != 0) {
152			EVENTHANDLER_DEREGISTER(dev_clone, tag);
153			return (err);
154		}
155		err = rman_manage_region(&tununits, 0, IF_MAXUNIT);
156		if (err != 0) {
157			printf("%s: tununits: rman_manage_region: Failed %d\n",
158			    TUNNAME, err);
159			rman_fini(&tununits);
160			EVENTHANDLER_DEREGISTER(dev_clone, tag);
161			return (err);
162		}
163		break;
164	case MOD_UNLOAD:
165		err = rman_fini(&tununits);
166		if (err != 0)
167			return (err);
168		EVENTHANDLER_DEREGISTER(dev_clone, tag);
169
170		while (tunhead != NULL) {
171			KASSERT((tunhead->tun_flags & TUN_OPEN) == 0,
172			    ("tununits is out of sync - unit %d",
173			    tunhead->tun_if.if_dunit));
174			tp = tunhead;
175			dev = makedev(tun_cdevsw.d_maj,
176			    unit2minor(tp->tun_if.if_dunit));
177			KASSERT(dev->si_drv1 == tp, ("Bad makedev result"));
178			tunhead = tp->next;
179			bpfdetach(&tp->tun_if);
180			if_detach(&tp->tun_if);
181			KASSERT(dev->si_flags & SI_NAMED, ("Missing make_dev"));
182			free(tp, M_TUN);
183		}
184
185		/*
186		 * Destroying tunbasedev results in all of our make_dev()s
187		 * conveniently going away.
188		 */
189		if (tunbasedev != NOUDEV)
190			destroy_dev(udev2dev(tunbasedev, 0));
191
192		break;
193	}
194	return 0;
195}
196
197static moduledata_t tun_mod = {
198	"if_tun",
199	tunmodevent,
200	0
201};
202
203DECLARE_MODULE(if_tun, tun_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
204
205static void
206tunstart(struct ifnet *ifp)
207{
208	struct tun_softc *tp = ifp->if_softc;
209
210	if (tp->tun_flags & TUN_RWAIT) {
211		tp->tun_flags &= ~TUN_RWAIT;
212		wakeup(tp);
213	}
214	if (tp->tun_flags & TUN_ASYNC && tp->tun_sigio)
215		pgsigio(&tp->tun_sigio, SIGIO, 0);
216	selwakeup(&tp->tun_rsel);
217}
218
219static void
220tuncreate(dev_t dev)
221{
222	struct tun_softc *sc;
223	struct ifnet *ifp;
224
225	if (!(dev->si_flags & SI_NAMED))
226		dev = make_dev(&tun_cdevsw, minor(dev),
227		    UID_UUCP, GID_DIALER, 0600, "tun%d", dev2unit(dev));
228
229	MALLOC(sc, struct tun_softc *, sizeof(*sc), M_TUN, M_WAITOK | M_ZERO);
230	sc->tun_flags = TUN_INITED;
231	sc->next = tunhead;
232	tunhead = sc;
233
234	ifp = &sc->tun_if;
235	if_initname(ifp, TUNNAME, dev2unit(dev));
236	ifp->if_mtu = TUNMTU;
237	ifp->if_ioctl = tunifioctl;
238	ifp->if_output = tunoutput;
239	ifp->if_start = tunstart;
240	ifp->if_flags = IFF_POINTOPOINT | IFF_MULTICAST;
241	ifp->if_type = IFT_PPP;
242	ifp->if_snd.ifq_maxlen = ifqmaxlen;
243	ifp->if_softc = sc;
244	if_attach(ifp);
245	bpfattach(ifp, DLT_NULL, sizeof(u_int));
246	dev->si_drv1 = sc;
247}
248
249static int
250tunopen(dev_t dev, int flag, int mode, struct thread *td)
251{
252	struct resource *r;
253	struct ifnet	*ifp;
254	struct tun_softc *tp;
255	int unit;
256
257	unit = dev2unit(dev);
258	if (unit > IF_MAXUNIT)
259		return (ENXIO);
260
261	r = rman_reserve_resource(&tununits, unit, unit, 1,
262	    RF_ALLOCATED | RF_ACTIVE, NULL);
263	if (r == NULL)
264		return (EBUSY);
265
266	dev->si_flags &= ~SI_CHEAPCLONE;
267
268	tp = dev->si_drv1;
269	if (!tp) {
270		tuncreate(dev);
271		tp = dev->si_drv1;
272	}
273	KASSERT(!(tp->tun_flags & TUN_OPEN), ("Resource & flags out-of-sync"));
274	tp->tun_unit = r;
275	tp->tun_pid = td->td_proc->p_pid;
276	ifp = &tp->tun_if;
277	tp->tun_flags |= TUN_OPEN;
278	TUNDEBUG(ifp, "open\n");
279
280	return (0);
281}
282
283/*
284 * tunclose - close the device - mark i/f down & delete
285 * routing info
286 */
287static	int
288tunclose(dev_t dev, int foo, int bar, struct thread *td)
289{
290	struct tun_softc *tp;
291	struct ifnet *ifp;
292	int s;
293	int err;
294
295	tp = dev->si_drv1;
296	ifp = &tp->tun_if;
297
298	KASSERT(tp->tun_unit, ("Unit %d not marked open", tp->tun_if.if_dunit));
299	tp->tun_flags &= ~TUN_OPEN;
300	tp->tun_pid = 0;
301
302	/*
303	 * junk all pending output
304	 */
305	IF_DRAIN(&ifp->if_snd);
306
307	if (ifp->if_flags & IFF_UP) {
308		s = splimp();
309		if_down(ifp);
310		splx(s);
311	}
312
313	if (ifp->if_flags & IFF_RUNNING) {
314		struct ifaddr *ifa;
315
316		s = splimp();
317		/* find internet addresses and delete routes */
318		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
319			if (ifa->ifa_addr->sa_family == AF_INET)
320				rtinit(ifa, (int)RTM_DELETE,
321				    tp->tun_flags & TUN_DSTADDR ? RTF_HOST : 0);
322		ifp->if_flags &= ~IFF_RUNNING;
323		splx(s);
324	}
325
326	funsetown(&tp->tun_sigio);
327	selwakeup(&tp->tun_rsel);
328
329	TUNDEBUG (ifp, "closed\n");
330	err = rman_release_resource(tp->tun_unit);
331	KASSERT(err == 0, ("Unit %d failed to release", tp->tun_if.if_dunit));
332
333	return (0);
334}
335
336static int
337tuninit(struct ifnet *ifp)
338{
339	struct tun_softc *tp = ifp->if_softc;
340	struct ifaddr *ifa;
341	int error = 0;
342
343	TUNDEBUG(ifp, "tuninit\n");
344
345	ifp->if_flags |= IFF_UP | IFF_RUNNING;
346	getmicrotime(&ifp->if_lastchange);
347
348	for (ifa = TAILQ_FIRST(&ifp->if_addrhead); ifa;
349	     ifa = TAILQ_NEXT(ifa, ifa_link)) {
350		if (ifa->ifa_addr == NULL)
351			error = EFAULT;
352			/* XXX: Should maybe return straight off? */
353		else {
354#ifdef INET
355			if (ifa->ifa_addr->sa_family == AF_INET) {
356			    struct sockaddr_in *si;
357
358			    si = (struct sockaddr_in *)ifa->ifa_addr;
359			    if (si->sin_addr.s_addr)
360				    tp->tun_flags |= TUN_IASET;
361
362			    si = (struct sockaddr_in *)ifa->ifa_dstaddr;
363			    if (si && si->sin_addr.s_addr)
364				    tp->tun_flags |= TUN_DSTADDR;
365			}
366#endif
367		}
368	}
369	return (error);
370}
371
372/*
373 * Process an ioctl request.
374 */
375static int
376tunifioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
377{
378	struct ifreq *ifr = (struct ifreq *)data;
379	struct tun_softc *tp = ifp->if_softc;
380	struct ifstat *ifs;
381	int		error = 0, s;
382
383	s = splimp();
384	switch(cmd) {
385	case SIOCGIFSTATUS:
386		ifs = (struct ifstat *)data;
387		if (tp->tun_pid)
388			sprintf(ifs->ascii + strlen(ifs->ascii),
389			    "\tOpened by PID %d\n", tp->tun_pid);
390		break;
391	case SIOCSIFADDR:
392		error = tuninit(ifp);
393		TUNDEBUG(ifp, "address set, error=%d\n", error);
394		break;
395	case SIOCSIFDSTADDR:
396		error = tuninit(ifp);
397		TUNDEBUG(ifp, "destination address set, error=%d\n", error);
398		break;
399	case SIOCSIFMTU:
400		ifp->if_mtu = ifr->ifr_mtu;
401		TUNDEBUG(ifp, "mtu set\n");
402		break;
403	case SIOCSIFFLAGS:
404	case SIOCADDMULTI:
405	case SIOCDELMULTI:
406		break;
407	default:
408		error = EINVAL;
409	}
410	splx(s);
411	return (error);
412}
413
414/*
415 * tunoutput - queue packets from higher level ready to put out.
416 */
417static int
418tunoutput(
419	struct ifnet *ifp,
420	struct mbuf *m0,
421	struct sockaddr *dst,
422	struct rtentry *rt)
423{
424	struct tun_softc *tp = ifp->if_softc;
425#ifdef MAC
426	int error;
427#endif
428
429	TUNDEBUG (ifp, "tunoutput\n");
430
431#ifdef MAC
432	error = mac_check_ifnet_transmit(ifp, m0);
433	if (error) {
434		m_freem(m0);
435		return (error);
436	}
437#endif
438
439	if ((tp->tun_flags & TUN_READY) != TUN_READY) {
440		TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags);
441		m_freem (m0);
442		return (EHOSTDOWN);
443	}
444
445	if ((ifp->if_flags & IFF_UP) != IFF_UP) {
446		m_freem (m0);
447		return (EHOSTDOWN);
448	}
449
450	/* BPF write needs to be handled specially */
451	if (dst->sa_family == AF_UNSPEC) {
452		dst->sa_family = *(mtod(m0, int *));
453		m0->m_len -= sizeof(int);
454		m0->m_pkthdr.len -= sizeof(int);
455		m0->m_data += sizeof(int);
456	}
457
458	if (ifp->if_bpf) {
459		/*
460		 * We need to prepend the address family as
461		 * a four byte field.  Cons up a dummy header
462		 * to pacify bpf.  This is safe because bpf
463		 * will only read from the mbuf (i.e., it won't
464		 * try to free it or keep a pointer to it).
465		 */
466		struct mbuf m;
467		uint32_t af = dst->sa_family;
468
469		m.m_next = m0;
470		m.m_len = 4;
471		m.m_data = (char *)&af;
472
473		BPF_MTAP(ifp, &m);
474	}
475
476	/* prepend sockaddr? this may abort if the mbuf allocation fails */
477	if (tp->tun_flags & TUN_LMODE) {
478		/* allocate space for sockaddr */
479		M_PREPEND(m0, dst->sa_len, M_DONTWAIT);
480
481		/* if allocation failed drop packet */
482		if (m0 == NULL) {
483			ifp->if_iqdrops++;
484			ifp->if_oerrors++;
485			return (ENOBUFS);
486		} else {
487			bcopy(dst, m0->m_data, dst->sa_len);
488		}
489	}
490
491	if (tp->tun_flags & TUN_IFHEAD) {
492		/* Prepend the address family */
493		M_PREPEND(m0, 4, M_DONTWAIT);
494
495		/* if allocation failed drop packet */
496		if (m0 == NULL) {
497			ifp->if_iqdrops++;
498			ifp->if_oerrors++;
499			return (ENOBUFS);
500		} else
501			*(u_int32_t *)m0->m_data = htonl(dst->sa_family);
502	} else {
503#ifdef INET
504		if (dst->sa_family != AF_INET)
505#endif
506		{
507			m_freem(m0);
508			return (EAFNOSUPPORT);
509		}
510	}
511
512	if (! IF_HANDOFF(&ifp->if_snd, m0, ifp)) {
513		ifp->if_collisions++;
514		return (ENOBUFS);
515	}
516	ifp->if_opackets++;
517	return (0);
518}
519
520/*
521 * the cdevsw interface is now pretty minimal.
522 */
523static	int
524tunioctl(dev_t dev, u_long cmd, caddr_t data, int flag, struct thread *td)
525{
526	int		s;
527	int		error;
528	struct tun_softc *tp = dev->si_drv1;
529	struct tuninfo *tunp;
530
531	switch (cmd) {
532	case TUNSIFINFO:
533		tunp = (struct tuninfo *)data;
534		if (tunp->mtu < IF_MINMTU)
535			return (EINVAL);
536		if (tp->tun_if.if_mtu != tunp->mtu
537		&& (error = suser(td)) != 0)
538			return (error);
539		tp->tun_if.if_mtu = tunp->mtu;
540		tp->tun_if.if_type = tunp->type;
541		tp->tun_if.if_baudrate = tunp->baudrate;
542		break;
543	case TUNGIFINFO:
544		tunp = (struct tuninfo *)data;
545		tunp->mtu = tp->tun_if.if_mtu;
546		tunp->type = tp->tun_if.if_type;
547		tunp->baudrate = tp->tun_if.if_baudrate;
548		break;
549	case TUNSDEBUG:
550		tundebug = *(int *)data;
551		break;
552	case TUNGDEBUG:
553		*(int *)data = tundebug;
554		break;
555	case TUNSLMODE:
556		if (*(int *)data) {
557			tp->tun_flags |= TUN_LMODE;
558			tp->tun_flags &= ~TUN_IFHEAD;
559		} else
560			tp->tun_flags &= ~TUN_LMODE;
561		break;
562	case TUNSIFHEAD:
563		if (*(int *)data) {
564			tp->tun_flags |= TUN_IFHEAD;
565			tp->tun_flags &= ~TUN_LMODE;
566		} else
567			tp->tun_flags &= ~TUN_IFHEAD;
568		break;
569	case TUNGIFHEAD:
570		*(int *)data = (tp->tun_flags & TUN_IFHEAD) ? 1 : 0;
571		break;
572	case TUNSIFMODE:
573		/* deny this if UP */
574		if (tp->tun_if.if_flags & IFF_UP)
575			return(EBUSY);
576
577		switch (*(int *)data & ~IFF_MULTICAST) {
578		case IFF_POINTOPOINT:
579		case IFF_BROADCAST:
580			tp->tun_if.if_flags &=
581			    ~(IFF_BROADCAST|IFF_POINTOPOINT|IFF_MULTICAST);
582			tp->tun_if.if_flags |= *(int *)data;
583			break;
584		default:
585			return(EINVAL);
586		}
587		break;
588	case TUNSIFPID:
589		tp->tun_pid = curthread->td_proc->p_pid;
590		break;
591	case FIONBIO:
592		break;
593	case FIOASYNC:
594		if (*(int *)data)
595			tp->tun_flags |= TUN_ASYNC;
596		else
597			tp->tun_flags &= ~TUN_ASYNC;
598		break;
599	case FIONREAD:
600		s = splimp();
601		if (tp->tun_if.if_snd.ifq_head) {
602			struct mbuf *mb = tp->tun_if.if_snd.ifq_head;
603			for( *(int *)data = 0; mb != 0; mb = mb->m_next)
604				*(int *)data += mb->m_len;
605		} else
606			*(int *)data = 0;
607		splx(s);
608		break;
609	case FIOSETOWN:
610		return (fsetown(*(int *)data, &tp->tun_sigio));
611
612	case FIOGETOWN:
613		*(int *)data = fgetown(&tp->tun_sigio);
614		return (0);
615
616	/* This is deprecated, FIOSETOWN should be used instead. */
617	case TIOCSPGRP:
618		return (fsetown(-(*(int *)data), &tp->tun_sigio));
619
620	/* This is deprecated, FIOGETOWN should be used instead. */
621	case TIOCGPGRP:
622		*(int *)data = -fgetown(&tp->tun_sigio);
623		return (0);
624
625	default:
626		return (ENOTTY);
627	}
628	return (0);
629}
630
631/*
632 * The cdevsw read interface - reads a packet at a time, or at
633 * least as much of a packet as can be read.
634 */
635static	int
636tunread(dev_t dev, struct uio *uio, int flag)
637{
638	struct tun_softc *tp = dev->si_drv1;
639	struct ifnet	*ifp = &tp->tun_if;
640	struct mbuf	*m;
641	int		error=0, len, s;
642
643	TUNDEBUG (ifp, "read\n");
644	if ((tp->tun_flags & TUN_READY) != TUN_READY) {
645		TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags);
646		return (EHOSTDOWN);
647	}
648
649	tp->tun_flags &= ~TUN_RWAIT;
650
651	s = splimp();
652	do {
653		IF_DEQUEUE(&ifp->if_snd, m);
654		if (m == NULL) {
655			if (flag & IO_NDELAY) {
656				splx(s);
657				return (EWOULDBLOCK);
658			}
659			tp->tun_flags |= TUN_RWAIT;
660			if((error = tsleep(tp, PCATCH | (PZERO + 1),
661					"tunread", 0)) != 0) {
662				splx(s);
663				return (error);
664			}
665		}
666	} while (m == NULL);
667	splx(s);
668
669	while (m && uio->uio_resid > 0 && error == 0) {
670		len = min(uio->uio_resid, m->m_len);
671		if (len != 0)
672			error = uiomove(mtod(m, void *), len, uio);
673		m = m_free(m);
674	}
675
676	if (m) {
677		TUNDEBUG(ifp, "Dropping mbuf\n");
678		m_freem(m);
679	}
680	return (error);
681}
682
683/*
684 * the cdevsw write interface - an atomic write is a packet - or else!
685 */
686static	int
687tunwrite(dev_t dev, struct uio *uio, int flag)
688{
689	struct tun_softc *tp = dev->si_drv1;
690	struct ifnet	*ifp = &tp->tun_if;
691	struct mbuf	*top, **mp, *m;
692	int		error=0, tlen, mlen;
693	uint32_t	family;
694	int 		isr;
695
696	TUNDEBUG(ifp, "tunwrite\n");
697
698	if ((ifp->if_flags & IFF_UP) != IFF_UP)
699		/* ignore silently */
700		return (0);
701
702	if (uio->uio_resid == 0)
703		return (0);
704
705	if (uio->uio_resid < 0 || uio->uio_resid > TUNMRU) {
706		TUNDEBUG(ifp, "len=%d!\n", uio->uio_resid);
707		return (EIO);
708	}
709	tlen = uio->uio_resid;
710
711	/* get a header mbuf */
712	MGETHDR(m, M_DONTWAIT, MT_DATA);
713	if (m == NULL)
714		return (ENOBUFS);
715	mlen = MHLEN;
716
717	top = 0;
718	mp = &top;
719	while (error == 0 && uio->uio_resid > 0) {
720		m->m_len = min(mlen, uio->uio_resid);
721		error = uiomove(mtod(m, void *), m->m_len, uio);
722		*mp = m;
723		mp = &m->m_next;
724		if (uio->uio_resid > 0) {
725			MGET (m, M_DONTWAIT, MT_DATA);
726			if (m == 0) {
727				error = ENOBUFS;
728				break;
729			}
730			mlen = MLEN;
731		}
732	}
733	if (error) {
734		if (top)
735			m_freem (top);
736		ifp->if_ierrors++;
737		return (error);
738	}
739
740	top->m_pkthdr.len = tlen;
741	top->m_pkthdr.rcvif = ifp;
742#ifdef MAC
743	mac_create_mbuf_from_ifnet(ifp, top);
744#endif
745
746	if (ifp->if_bpf) {
747		if (tp->tun_flags & TUN_IFHEAD) {
748			/*
749			 * Conveniently, we already have a 4-byte address
750			 * family prepended to our packet !
751			 * Inconveniently, it's in the wrong byte order !
752			 */
753			if ((top = m_pullup(top, sizeof(family))) == NULL)
754				return (ENOBUFS);
755			*mtod(top, u_int32_t *) =
756			    ntohl(*mtod(top, u_int32_t *));
757			BPF_MTAP(ifp, top);
758			*mtod(top, u_int32_t *) =
759			    htonl(*mtod(top, u_int32_t *));
760		} else {
761			/*
762			 * We need to prepend the address family as
763			 * a four byte field.  Cons up a dummy header
764			 * to pacify bpf.  This is safe because bpf
765			 * will only read from the mbuf (i.e., it won't
766			 * try to free it or keep a pointer to it).
767			 */
768			struct mbuf m;
769			uint32_t af = AF_INET;
770
771			m.m_next = top;
772			m.m_len = 4;
773			m.m_data = (char *)&af;
774
775			BPF_MTAP(ifp, &m);
776		}
777	}
778
779	if (tp->tun_flags & TUN_IFHEAD) {
780		if (top->m_len < sizeof(family) &&
781		    (top = m_pullup(top, sizeof(family))) == NULL)
782			return (ENOBUFS);
783		family = ntohl(*mtod(top, u_int32_t *));
784		m_adj(top, sizeof(family));
785	} else
786		family = AF_INET;
787
788	switch (family) {
789#ifdef INET
790	case AF_INET:
791		isr = NETISR_IP;
792		break;
793#endif
794#ifdef INET6
795	case AF_INET6:
796		isr = NETISR_IPV6;
797		break;
798#endif
799#ifdef IPX
800	case AF_IPX:
801		isr = NETISR_IPX;
802		break;
803#endif
804#ifdef NETATALK
805	case AF_APPLETALK:
806		isr = NETISR_ATALK2;
807		break;
808#endif
809	default:
810		m_freem(m);
811		return (EAFNOSUPPORT);
812	}
813	/* First chunk of an mbuf contains good junk */
814	if (harvest.point_to_point)
815		random_harvest(m, 16, 3, 0, RANDOM_NET);
816	ifp->if_ibytes += top->m_pkthdr.len;
817	ifp->if_ipackets++;
818	netisr_dispatch(isr, top);
819	return (0);
820}
821
822/*
823 * tunpoll - the poll interface, this is only useful on reads
824 * really. The write detect always returns true, write never blocks
825 * anyway, it either accepts the packet or drops it.
826 */
827static	int
828tunpoll(dev_t dev, int events, struct thread *td)
829{
830	int		s;
831	struct tun_softc *tp = dev->si_drv1;
832	struct ifnet	*ifp = &tp->tun_if;
833	int		revents = 0;
834
835	s = splimp();
836	TUNDEBUG(ifp, "tunpoll\n");
837
838	if (events & (POLLIN | POLLRDNORM)) {
839		if (ifp->if_snd.ifq_len > 0) {
840			TUNDEBUG(ifp, "tunpoll q=%d\n", ifp->if_snd.ifq_len);
841			revents |= events & (POLLIN | POLLRDNORM);
842		} else {
843			TUNDEBUG(ifp, "tunpoll waiting\n");
844			selrecord(td, &tp->tun_rsel);
845		}
846	}
847	if (events & (POLLOUT | POLLWRNORM))
848		revents |= events & (POLLOUT | POLLWRNORM);
849
850	splx(s);
851	return (revents);
852}
853