if_tun.c revision 147611 
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 147611 2005-06-26 18:11:11Z dwmalone $
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/fcntl.h>
33#include <sys/filio.h>
34#include <sys/sockio.h>
35#include <sys/ttycom.h>
36#include <sys/poll.h>
37#include <sys/selinfo.h>
38#include <sys/signalvar.h>
39#include <sys/filedesc.h>
40#include <sys/kernel.h>
41#include <sys/sysctl.h>
42#include <sys/conf.h>
43#include <sys/uio.h>
44#include <sys/malloc.h>
45#include <sys/random.h>
46
47#include <net/if.h>
48#include <net/if_types.h>
49#include <net/netisr.h>
50#include <net/route.h>
51#ifdef INET
52#include <netinet/in.h>
53#endif
54#include <net/bpf.h>
55#include <net/if_tun.h>
56
57#include <sys/queue.h>
58
59/*
60 * tun_list is protected by global tunmtx.  Other mutable fields are
61 * protected by tun->tun_mtx, or by their owning subsystem.  tun_dev is
62 * static for the duration of a tunnel interface.
63 */
64struct tun_softc {
65	TAILQ_ENTRY(tun_softc)	tun_list;
66	struct cdev *tun_dev;
67	u_short	tun_flags;		/* misc flags */
68#define	TUN_OPEN	0x0001
69#define	TUN_INITED	0x0002
70#define	TUN_RCOLL	0x0004
71#define	TUN_IASET	0x0008
72#define	TUN_DSTADDR	0x0010
73#define	TUN_LMODE	0x0020
74#define	TUN_RWAIT	0x0040
75#define	TUN_ASYNC	0x0080
76#define	TUN_IFHEAD	0x0100
77
78#define TUN_READY       (TUN_OPEN | TUN_INITED)
79
80	/*
81	 * XXXRW: tun_pid is used to exclusively lock /dev/tun.  Is this
82	 * actually needed?  Can we just return EBUSY if already open?
83	 * Problem is that this involved inherent races when a tun device
84	 * is handed off from one process to another, as opposed to just
85	 * being slightly stale informationally.
86	 */
87	pid_t	tun_pid;		/* owning pid */
88	struct	ifnet *tun_ifp;		/* the interface */
89	struct  sigio *tun_sigio;	/* information for async I/O */
90	struct	selinfo	tun_rsel;	/* read select */
91	struct mtx	tun_mtx;	/* protect mutable softc fields */
92};
93#define TUN2IFP(sc)	((sc)->tun_ifp)
94
95#define TUNDEBUG	if (tundebug) if_printf
96#define	TUNNAME		"tun"
97
98/*
99 * All mutable global variables in if_tun are locked using tunmtx, with
100 * the exception of tundebug, which is used unlocked, and tunclones,
101 * which is static after setup.
102 */
103static struct mtx tunmtx;
104static MALLOC_DEFINE(M_TUN, TUNNAME, "Tunnel Interface");
105static int tundebug = 0;
106static struct clonedevs *tunclones;
107static TAILQ_HEAD(,tun_softc)	tunhead = TAILQ_HEAD_INITIALIZER(tunhead);
108SYSCTL_INT(_debug, OID_AUTO, if_tun_debug, CTLFLAG_RW, &tundebug, 0, "");
109
110static void	tunclone(void *arg, char *name, int namelen, struct cdev **dev);
111static void	tuncreate(struct cdev *dev);
112static int	tunifioctl(struct ifnet *, u_long, caddr_t);
113static int	tuninit(struct ifnet *);
114static int	tunmodevent(module_t, int, void *);
115static int	tunoutput(struct ifnet *, struct mbuf *, struct sockaddr *,
116		    struct rtentry *rt);
117static void	tunstart(struct ifnet *);
118
119static d_open_t		tunopen;
120static d_close_t	tunclose;
121static d_read_t		tunread;
122static d_write_t	tunwrite;
123static d_ioctl_t	tunioctl;
124static d_poll_t		tunpoll;
125
126static struct cdevsw tun_cdevsw = {
127	.d_version =	D_VERSION,
128	.d_flags =	D_PSEUDO | D_NEEDGIANT,
129	.d_open =	tunopen,
130	.d_close =	tunclose,
131	.d_read =	tunread,
132	.d_write =	tunwrite,
133	.d_ioctl =	tunioctl,
134	.d_poll =	tunpoll,
135	.d_name =	TUNNAME,
136};
137
138static void
139tunclone(void *arg, char *name, int namelen, struct cdev **dev)
140{
141	int u, i;
142
143	if (*dev != NULL)
144		return;
145
146	if (strcmp(name, TUNNAME) == 0) {
147		u = -1;
148	} else if (dev_stdclone(name, NULL, TUNNAME, &u) != 1)
149		return;	/* Don't recognise the name */
150	if (u != -1 && u > IF_MAXUNIT)
151		return;	/* Unit number too high */
152
153	/* find any existing device, or allocate new unit number */
154	i = clone_create(&tunclones, &tun_cdevsw, &u, dev, 0);
155	if (i) {
156		/* No preexisting struct cdev *, create one */
157		*dev = make_dev(&tun_cdevsw, unit2minor(u),
158		    UID_UUCP, GID_DIALER, 0600, "tun%d", u);
159		if (*dev != NULL) {
160			dev_ref(*dev);
161			(*dev)->si_flags |= SI_CHEAPCLONE;
162		}
163	}
164}
165
166static void
167tun_destroy(struct tun_softc *tp)
168{
169	struct cdev *dev;
170
171	/* Unlocked read. */
172	KASSERT((tp->tun_flags & TUN_OPEN) == 0,
173	    ("tununits is out of sync - unit %d", TUN2IFP(tp)->if_dunit));
174
175	dev = tp->tun_dev;
176	bpfdetach(TUN2IFP(tp));
177	if_detach(TUN2IFP(tp));
178	if_free(TUN2IFP(tp));
179	destroy_dev(dev);
180	mtx_destroy(&tp->tun_mtx);
181	free(tp, M_TUN);
182}
183
184static int
185tunmodevent(module_t mod, int type, void *data)
186{
187	static eventhandler_tag tag;
188	struct tun_softc *tp;
189
190	switch (type) {
191	case MOD_LOAD:
192		mtx_init(&tunmtx, "tunmtx", NULL, MTX_DEF);
193		clone_setup(&tunclones);
194		tag = EVENTHANDLER_REGISTER(dev_clone, tunclone, 0, 1000);
195		if (tag == NULL)
196			return (ENOMEM);
197		break;
198	case MOD_UNLOAD:
199		EVENTHANDLER_DEREGISTER(dev_clone, tag);
200
201		mtx_lock(&tunmtx);
202		while ((tp = TAILQ_FIRST(&tunhead)) != NULL) {
203			TAILQ_REMOVE(&tunhead, tp, tun_list);
204			mtx_unlock(&tunmtx);
205			tun_destroy(tp);
206			mtx_lock(&tunmtx);
207		}
208		mtx_unlock(&tunmtx);
209		clone_cleanup(&tunclones);
210		mtx_destroy(&tunmtx);
211		break;
212	default:
213		return EOPNOTSUPP;
214	}
215	return 0;
216}
217
218static moduledata_t tun_mod = {
219	"if_tun",
220	tunmodevent,
221	0
222};
223
224DECLARE_MODULE(if_tun, tun_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
225
226static void
227tunstart(struct ifnet *ifp)
228{
229	struct tun_softc *tp = ifp->if_softc;
230	struct mbuf *m;
231
232	if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
233		IFQ_LOCK(&ifp->if_snd);
234		IFQ_POLL_NOLOCK(&ifp->if_snd, m);
235		if (m == NULL) {
236			IFQ_UNLOCK(&ifp->if_snd);
237			return;
238		}
239		IFQ_UNLOCK(&ifp->if_snd);
240	}
241
242	mtx_lock(&tp->tun_mtx);
243	if (tp->tun_flags & TUN_RWAIT) {
244		tp->tun_flags &= ~TUN_RWAIT;
245		wakeup(tp);
246	}
247	if (tp->tun_flags & TUN_ASYNC && tp->tun_sigio) {
248		mtx_unlock(&tp->tun_mtx);
249		pgsigio(&tp->tun_sigio, SIGIO, 0);
250	} else
251		mtx_unlock(&tp->tun_mtx);
252	selwakeuppri(&tp->tun_rsel, PZERO + 1);
253}
254
255/* XXX: should return an error code so it can fail. */
256static void
257tuncreate(struct cdev *dev)
258{
259	struct tun_softc *sc;
260	struct ifnet *ifp;
261
262	dev->si_flags &= ~SI_CHEAPCLONE;
263
264	MALLOC(sc, struct tun_softc *, sizeof(*sc), M_TUN, M_WAITOK | M_ZERO);
265	mtx_init(&sc->tun_mtx, "tun_mtx", NULL, MTX_DEF);
266	sc->tun_flags = TUN_INITED;
267	sc->tun_dev = dev;
268	mtx_lock(&tunmtx);
269	TAILQ_INSERT_TAIL(&tunhead, sc, tun_list);
270	mtx_unlock(&tunmtx);
271
272	ifp = sc->tun_ifp = if_alloc(IFT_PPP);
273	if (ifp == NULL)
274		panic("%s%d: failed to if_alloc() interface.\n",
275		    TUNNAME, dev2unit(dev));
276	if_initname(ifp, TUNNAME, dev2unit(dev));
277	ifp->if_mtu = TUNMTU;
278	ifp->if_ioctl = tunifioctl;
279	ifp->if_output = tunoutput;
280	ifp->if_start = tunstart;
281	ifp->if_flags = IFF_POINTOPOINT | IFF_MULTICAST;
282	ifp->if_softc = sc;
283	IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen);
284	ifp->if_snd.ifq_drv_maxlen = 0;
285	IFQ_SET_READY(&ifp->if_snd);
286
287	if_attach(ifp);
288	bpfattach(ifp, DLT_NULL, sizeof(u_int32_t));
289	dev->si_drv1 = sc;
290}
291
292static int
293tunopen(struct cdev *dev, int flag, int mode, struct thread *td)
294{
295	struct ifnet	*ifp;
296	struct tun_softc *tp;
297
298	/*
299	 * XXXRW: Non-atomic test and set of dev->si_drv1 requires
300	 * synchronization.
301	 */
302	tp = dev->si_drv1;
303	if (!tp) {
304		tuncreate(dev);
305		tp = dev->si_drv1;
306	}
307
308	/*
309	 * XXXRW: This use of tun_pid is subject to error due to the
310	 * fact that a reference to the tunnel can live beyond the
311	 * death of the process that created it.  Can we replace this
312	 * with a simple busy flag?
313	 */
314	mtx_lock(&tp->tun_mtx);
315	if (tp->tun_pid != 0 && tp->tun_pid != td->td_proc->p_pid) {
316		mtx_unlock(&tp->tun_mtx);
317		return (EBUSY);
318	}
319	tp->tun_pid = td->td_proc->p_pid;
320
321	tp->tun_flags |= TUN_OPEN;
322	mtx_unlock(&tp->tun_mtx);
323	ifp = TUN2IFP(tp);
324	TUNDEBUG(ifp, "open\n");
325
326	return (0);
327}
328
329/*
330 * tunclose - close the device - mark i/f down & delete
331 * routing info
332 */
333static	int
334tunclose(struct cdev *dev, int foo, int bar, struct thread *td)
335{
336	struct tun_softc *tp;
337	struct ifnet *ifp;
338	int s;
339
340	tp = dev->si_drv1;
341	ifp = TUN2IFP(tp);
342
343	mtx_lock(&tp->tun_mtx);
344	tp->tun_flags &= ~TUN_OPEN;
345	tp->tun_pid = 0;
346
347	/*
348	 * junk all pending output
349	 */
350	s = splimp();
351	IFQ_PURGE(&ifp->if_snd);
352	splx(s);
353	mtx_unlock(&tp->tun_mtx);
354
355	if (ifp->if_flags & IFF_UP) {
356		s = splimp();
357		if_down(ifp);
358		splx(s);
359	}
360
361	if (ifp->if_flags & IFF_RUNNING) {
362		struct ifaddr *ifa;
363
364		s = splimp();
365		/* find internet addresses and delete routes */
366		TAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link)
367			if (ifa->ifa_addr->sa_family == AF_INET)
368				/* Unlocked read. */
369				rtinit(ifa, (int)RTM_DELETE,
370				    tp->tun_flags & TUN_DSTADDR ? RTF_HOST : 0);
371		ifp->if_flags &= ~IFF_RUNNING;
372		splx(s);
373	}
374
375	funsetown(&tp->tun_sigio);
376	selwakeuppri(&tp->tun_rsel, PZERO + 1);
377	TUNDEBUG (ifp, "closed\n");
378	return (0);
379}
380
381static int
382tuninit(struct ifnet *ifp)
383{
384	struct tun_softc *tp = ifp->if_softc;
385	struct ifaddr *ifa;
386	int error = 0;
387
388	TUNDEBUG(ifp, "tuninit\n");
389
390	ifp->if_flags |= IFF_UP | IFF_RUNNING;
391	getmicrotime(&ifp->if_lastchange);
392
393	for (ifa = TAILQ_FIRST(&ifp->if_addrhead); ifa;
394	     ifa = TAILQ_NEXT(ifa, ifa_link)) {
395		if (ifa->ifa_addr == NULL)
396			error = EFAULT;
397			/* XXX: Should maybe return straight off? */
398		else {
399#ifdef INET
400			if (ifa->ifa_addr->sa_family == AF_INET) {
401			    struct sockaddr_in *si;
402
403			    si = (struct sockaddr_in *)ifa->ifa_addr;
404			    mtx_lock(&tp->tun_mtx);
405			    if (si->sin_addr.s_addr)
406				    tp->tun_flags |= TUN_IASET;
407
408			    si = (struct sockaddr_in *)ifa->ifa_dstaddr;
409			    if (si && si->sin_addr.s_addr)
410				    tp->tun_flags |= TUN_DSTADDR;
411			    mtx_unlock(&tp->tun_mtx);
412			}
413#endif
414		}
415	}
416	return (error);
417}
418
419/*
420 * Process an ioctl request.
421 */
422static int
423tunifioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
424{
425	struct ifreq *ifr = (struct ifreq *)data;
426	struct tun_softc *tp = ifp->if_softc;
427	struct ifstat *ifs;
428	int		error = 0, s;
429
430	s = splimp();
431	switch(cmd) {
432	case SIOCGIFSTATUS:
433		ifs = (struct ifstat *)data;
434		mtx_lock(&tp->tun_mtx);
435		if (tp->tun_pid)
436			sprintf(ifs->ascii + strlen(ifs->ascii),
437			    "\tOpened by PID %d\n", tp->tun_pid);
438		mtx_unlock(&tp->tun_mtx);
439		break;
440	case SIOCSIFADDR:
441		error = tuninit(ifp);
442		TUNDEBUG(ifp, "address set, error=%d\n", error);
443		break;
444	case SIOCSIFDSTADDR:
445		error = tuninit(ifp);
446		TUNDEBUG(ifp, "destination address set, error=%d\n", error);
447		break;
448	case SIOCSIFMTU:
449		ifp->if_mtu = ifr->ifr_mtu;
450		TUNDEBUG(ifp, "mtu set\n");
451		break;
452	case SIOCSIFFLAGS:
453	case SIOCADDMULTI:
454	case SIOCDELMULTI:
455		break;
456	default:
457		error = EINVAL;
458	}
459	splx(s);
460	return (error);
461}
462
463/*
464 * tunoutput - queue packets from higher level ready to put out.
465 */
466static int
467tunoutput(
468	struct ifnet *ifp,
469	struct mbuf *m0,
470	struct sockaddr *dst,
471	struct rtentry *rt)
472{
473	struct tun_softc *tp = ifp->if_softc;
474	u_short cached_tun_flags;
475	int error;
476	u_int32_t af;
477
478	TUNDEBUG (ifp, "tunoutput\n");
479
480#ifdef MAC
481	error = mac_check_ifnet_transmit(ifp, m0);
482	if (error) {
483		m_freem(m0);
484		return (error);
485	}
486#endif
487
488	/* Could be unlocked read? */
489	mtx_lock(&tp->tun_mtx);
490	cached_tun_flags = tp->tun_flags;
491	mtx_unlock(&tp->tun_mtx);
492	if ((cached_tun_flags & TUN_READY) != TUN_READY) {
493		TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags);
494		m_freem (m0);
495		return (EHOSTDOWN);
496	}
497
498	if ((ifp->if_flags & IFF_UP) != IFF_UP) {
499		m_freem (m0);
500		return (EHOSTDOWN);
501	}
502
503	/* BPF writes need to be handled specially. */
504	if (dst->sa_family == AF_UNSPEC) {
505		bcopy(dst->sa_data, &af, sizeof(af));
506		dst->sa_family = af;
507	}
508
509	if (ifp->if_bpf) {
510		af = dst->sa_family;
511		bpf_mtap2(ifp->if_bpf, &af, sizeof(af), m0);
512	}
513
514	/* prepend sockaddr? this may abort if the mbuf allocation fails */
515	if (cached_tun_flags & TUN_LMODE) {
516		/* allocate space for sockaddr */
517		M_PREPEND(m0, dst->sa_len, M_DONTWAIT);
518
519		/* if allocation failed drop packet */
520		if (m0 == NULL) {
521			ifp->if_iqdrops++;
522			ifp->if_oerrors++;
523			return (ENOBUFS);
524		} else {
525			bcopy(dst, m0->m_data, dst->sa_len);
526		}
527	}
528
529	if (cached_tun_flags & TUN_IFHEAD) {
530		/* Prepend the address family */
531		M_PREPEND(m0, 4, M_DONTWAIT);
532
533		/* if allocation failed drop packet */
534		if (m0 == NULL) {
535			ifp->if_iqdrops++;
536			ifp->if_oerrors++;
537			return (ENOBUFS);
538		} else
539			*(u_int32_t *)m0->m_data = htonl(dst->sa_family);
540	} else {
541#ifdef INET
542		if (dst->sa_family != AF_INET)
543#endif
544		{
545			m_freem(m0);
546			return (EAFNOSUPPORT);
547		}
548	}
549
550	IFQ_HANDOFF(ifp, m0, error);
551	if (error) {
552		ifp->if_collisions++;
553		return (ENOBUFS);
554	}
555	ifp->if_opackets++;
556	return (0);
557}
558
559/*
560 * the cdevsw interface is now pretty minimal.
561 */
562static	int
563tunioctl(struct cdev *dev, u_long cmd, caddr_t data, int flag, struct thread *td)
564{
565	int		s;
566	int		error;
567	struct tun_softc *tp = dev->si_drv1;
568	struct tuninfo *tunp;
569
570	switch (cmd) {
571	case TUNSIFINFO:
572		tunp = (struct tuninfo *)data;
573		if (tunp->mtu < IF_MINMTU)
574			return (EINVAL);
575		if (TUN2IFP(tp)->if_mtu != tunp->mtu
576		&& (error = suser(td)) != 0)
577			return (error);
578		TUN2IFP(tp)->if_mtu = tunp->mtu;
579		TUN2IFP(tp)->if_type = tunp->type;
580		TUN2IFP(tp)->if_baudrate = tunp->baudrate;
581		break;
582	case TUNGIFINFO:
583		tunp = (struct tuninfo *)data;
584		tunp->mtu = TUN2IFP(tp)->if_mtu;
585		tunp->type = TUN2IFP(tp)->if_type;
586		tunp->baudrate = TUN2IFP(tp)->if_baudrate;
587		break;
588	case TUNSDEBUG:
589		tundebug = *(int *)data;
590		break;
591	case TUNGDEBUG:
592		*(int *)data = tundebug;
593		break;
594	case TUNSLMODE:
595		mtx_lock(&tp->tun_mtx);
596		if (*(int *)data) {
597			tp->tun_flags |= TUN_LMODE;
598			tp->tun_flags &= ~TUN_IFHEAD;
599		} else
600			tp->tun_flags &= ~TUN_LMODE;
601		mtx_unlock(&tp->tun_mtx);
602		break;
603	case TUNSIFHEAD:
604		mtx_lock(&tp->tun_mtx);
605		if (*(int *)data) {
606			tp->tun_flags |= TUN_IFHEAD;
607			tp->tun_flags &= ~TUN_LMODE;
608		} else
609			tp->tun_flags &= ~TUN_IFHEAD;
610		mtx_unlock(&tp->tun_mtx);
611		break;
612	case TUNGIFHEAD:
613		/* Could be unlocked read? */
614		mtx_lock(&tp->tun_mtx);
615		*(int *)data = (tp->tun_flags & TUN_IFHEAD) ? 1 : 0;
616		mtx_unlock(&tp->tun_mtx);
617		break;
618	case TUNSIFMODE:
619		/* deny this if UP */
620		if (TUN2IFP(tp)->if_flags & IFF_UP)
621			return(EBUSY);
622
623		switch (*(int *)data & ~IFF_MULTICAST) {
624		case IFF_POINTOPOINT:
625		case IFF_BROADCAST:
626			TUN2IFP(tp)->if_flags &=
627			    ~(IFF_BROADCAST|IFF_POINTOPOINT|IFF_MULTICAST);
628			TUN2IFP(tp)->if_flags |= *(int *)data;
629			break;
630		default:
631			return(EINVAL);
632		}
633		break;
634	case TUNSIFPID:
635		mtx_lock(&tp->tun_mtx);
636		tp->tun_pid = curthread->td_proc->p_pid;
637		mtx_unlock(&tp->tun_mtx);
638		break;
639	case FIONBIO:
640		break;
641	case FIOASYNC:
642		mtx_lock(&tp->tun_mtx);
643		if (*(int *)data)
644			tp->tun_flags |= TUN_ASYNC;
645		else
646			tp->tun_flags &= ~TUN_ASYNC;
647		mtx_unlock(&tp->tun_mtx);
648		break;
649	case FIONREAD:
650		s = splimp();
651		if (!IFQ_IS_EMPTY(&TUN2IFP(tp)->if_snd)) {
652			struct mbuf *mb;
653			IFQ_LOCK(&TUN2IFP(tp)->if_snd);
654			IFQ_POLL_NOLOCK(&TUN2IFP(tp)->if_snd, mb);
655			for( *(int *)data = 0; mb != 0; mb = mb->m_next)
656				*(int *)data += mb->m_len;
657			IFQ_UNLOCK(&TUN2IFP(tp)->if_snd);
658		} else
659			*(int *)data = 0;
660		splx(s);
661		break;
662	case FIOSETOWN:
663		return (fsetown(*(int *)data, &tp->tun_sigio));
664
665	case FIOGETOWN:
666		*(int *)data = fgetown(&tp->tun_sigio);
667		return (0);
668
669	/* This is deprecated, FIOSETOWN should be used instead. */
670	case TIOCSPGRP:
671		return (fsetown(-(*(int *)data), &tp->tun_sigio));
672
673	/* This is deprecated, FIOGETOWN should be used instead. */
674	case TIOCGPGRP:
675		*(int *)data = -fgetown(&tp->tun_sigio);
676		return (0);
677
678	default:
679		return (ENOTTY);
680	}
681	return (0);
682}
683
684/*
685 * The cdevsw read interface - reads a packet at a time, or at
686 * least as much of a packet as can be read.
687 */
688static	int
689tunread(struct cdev *dev, struct uio *uio, int flag)
690{
691	struct tun_softc *tp = dev->si_drv1;
692	struct ifnet	*ifp = TUN2IFP(tp);
693	struct mbuf	*m;
694	int		error=0, len, s;
695
696	TUNDEBUG (ifp, "read\n");
697	mtx_lock(&tp->tun_mtx);
698	if ((tp->tun_flags & TUN_READY) != TUN_READY) {
699		mtx_unlock(&tp->tun_mtx);
700		TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags);
701		return (EHOSTDOWN);
702	}
703
704	tp->tun_flags &= ~TUN_RWAIT;
705	mtx_unlock(&tp->tun_mtx);
706
707	s = splimp();
708	do {
709		IFQ_DEQUEUE(&ifp->if_snd, m);
710		if (m == NULL) {
711			if (flag & O_NONBLOCK) {
712				splx(s);
713				return (EWOULDBLOCK);
714			}
715			mtx_lock(&tp->tun_mtx);
716			tp->tun_flags |= TUN_RWAIT;
717			mtx_unlock(&tp->tun_mtx);
718			if((error = tsleep(tp, PCATCH | (PZERO + 1),
719					"tunread", 0)) != 0) {
720				splx(s);
721				return (error);
722			}
723		}
724	} while (m == NULL);
725	splx(s);
726
727	while (m && uio->uio_resid > 0 && error == 0) {
728		len = min(uio->uio_resid, m->m_len);
729		if (len != 0)
730			error = uiomove(mtod(m, void *), len, uio);
731		m = m_free(m);
732	}
733
734	if (m) {
735		TUNDEBUG(ifp, "Dropping mbuf\n");
736		m_freem(m);
737	}
738	return (error);
739}
740
741/*
742 * the cdevsw write interface - an atomic write is a packet - or else!
743 */
744static	int
745tunwrite(struct cdev *dev, struct uio *uio, int flag)
746{
747	struct tun_softc *tp = dev->si_drv1;
748	struct ifnet	*ifp = TUN2IFP(tp);
749	struct mbuf	*m;
750	int		error = 0;
751	uint32_t	family;
752	int 		isr;
753
754	TUNDEBUG(ifp, "tunwrite\n");
755
756	if ((ifp->if_flags & IFF_UP) != IFF_UP)
757		/* ignore silently */
758		return (0);
759
760	if (uio->uio_resid == 0)
761		return (0);
762
763	if (uio->uio_resid < 0 || uio->uio_resid > TUNMRU) {
764		TUNDEBUG(ifp, "len=%d!\n", uio->uio_resid);
765		return (EIO);
766	}
767
768	if ((m = m_uiotombuf(uio, M_DONTWAIT, 0, 0)) == NULL) {
769		ifp->if_ierrors++;
770		return (error);
771	}
772
773	m->m_pkthdr.rcvif = ifp;
774#ifdef MAC
775	mac_create_mbuf_from_ifnet(ifp, m);
776#endif
777
778	/* Could be unlocked read? */
779	mtx_lock(&tp->tun_mtx);
780	if (tp->tun_flags & TUN_IFHEAD) {
781		mtx_unlock(&tp->tun_mtx);
782		if (m->m_len < sizeof(family) &&
783		    (m = m_pullup(m, sizeof(family))) == NULL)
784			return (ENOBUFS);
785		family = ntohl(*mtod(m, u_int32_t *));
786		m_adj(m, sizeof(family));
787	} else {
788		mtx_unlock(&tp->tun_mtx);
789		family = AF_INET;
790	}
791
792	BPF_MTAP2(ifp, &family, sizeof(family), m);
793
794	switch (family) {
795#ifdef INET
796	case AF_INET:
797		isr = NETISR_IP;
798		break;
799#endif
800#ifdef INET6
801	case AF_INET6:
802		isr = NETISR_IPV6;
803		break;
804#endif
805#ifdef IPX
806	case AF_IPX:
807		isr = NETISR_IPX;
808		break;
809#endif
810#ifdef NETATALK
811	case AF_APPLETALK:
812		isr = NETISR_ATALK2;
813		break;
814#endif
815	default:
816		m_freem(m);
817		return (EAFNOSUPPORT);
818	}
819	/* First chunk of an mbuf contains good junk */
820	if (harvest.point_to_point)
821		random_harvest(m, 16, 3, 0, RANDOM_NET);
822	ifp->if_ibytes += m->m_pkthdr.len;
823	ifp->if_ipackets++;
824	netisr_dispatch(isr, m);
825	return (0);
826}
827
828/*
829 * tunpoll - the poll interface, this is only useful on reads
830 * really. The write detect always returns true, write never blocks
831 * anyway, it either accepts the packet or drops it.
832 */
833static	int
834tunpoll(struct cdev *dev, int events, struct thread *td)
835{
836	int		s;
837	struct tun_softc *tp = dev->si_drv1;
838	struct ifnet	*ifp = TUN2IFP(tp);
839	int		revents = 0;
840	struct mbuf	*m;
841
842	s = splimp();
843	TUNDEBUG(ifp, "tunpoll\n");
844
845	if (events & (POLLIN | POLLRDNORM)) {
846		IFQ_LOCK(&ifp->if_snd);
847		IFQ_POLL_NOLOCK(&ifp->if_snd, m);
848		if (m != NULL) {
849			TUNDEBUG(ifp, "tunpoll q=%d\n", ifp->if_snd.ifq_len);
850			revents |= events & (POLLIN | POLLRDNORM);
851		} else {
852			TUNDEBUG(ifp, "tunpoll waiting\n");
853			selrecord(td, &tp->tun_rsel);
854		}
855		IFQ_UNLOCK(&ifp->if_snd);
856	}
857	if (events & (POLLOUT | POLLWRNORM))
858		revents |= events & (POLLOUT | POLLWRNORM);
859
860	splx(s);
861	return (revents);
862}
863