if_tun.c revision 10080
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 it's 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/select mode of
14 * operation though.
15 */
16
17#include "tun.h"
18#if NTUN > 0
19
20#include <sys/param.h>
21#include <sys/proc.h>
22#include <sys/systm.h>
23#include <sys/mbuf.h>
24#include <sys/buf.h>
25#include <sys/protosw.h>
26#include <sys/socket.h>
27#include <sys/ioctl.h>
28#include <sys/errno.h>
29#include <sys/syslog.h>
30#include <sys/select.h>
31#include <sys/file.h>
32#include <sys/signalvar.h>
33#ifdef __FreeBSD__
34#include <sys/kernel.h>
35#endif
36#include <sys/conf.h>
37
38#include <machine/cpu.h>
39
40#include <net/if.h>
41#include <net/netisr.h>
42#include <net/route.h>
43
44#ifdef INET
45#include <netinet/in.h>
46#include <netinet/in_systm.h>
47#include <netinet/in_var.h>
48#include <netinet/ip.h>
49#include <netinet/if_ether.h>
50#endif
51
52#ifdef NS
53#include <netns/ns.h>
54#include <netns/ns_if.h>
55#endif
56
57#include "bpfilter.h"
58#if NBPFILTER > 0
59#include <sys/time.h>
60#include <net/bpf.h>
61#endif
62
63#include <net/if_tun.h>
64
65#define TUNDEBUG	if (tundebug) printf
66int	tundebug = 0;
67
68struct tun_softc tunctl[NTUN];
69
70d_open_t tunopen;
71d_close_t tunclose;
72int	tunoutput __P((struct ifnet *, struct mbuf *, struct sockaddr *,
73	    struct rtentry *rt));
74d_rdwr_t tunread;
75d_rdwr_t tunwrite;
76d_ioctl_t tunioctl;
77int	tunifioctl __P((struct ifnet *, int, caddr_t));
78d_select_t tunselect;
79void	tunattach __P((void));
80
81static struct cdevsw tuncdevsw =
82{ tunopen,      tunclose,       tunread,        tunwrite,
83  tunioctl,     (d_stop_t *)enodev, (d_reset_t *)nullop, (d_ttycv_t *)enodev,
84  tunselect,    (d_mmap_t *)enodev,         NULL };
85extern dev_t tuncdev;
86
87static int tuninit __P((int));
88
89void
90tunattach(void)
91{
92	register int i;
93	struct ifnet *ifp;
94	struct sockaddr_in *sin;
95
96	/*
97	 * In case we are an LKM, set up device switch.
98	 */
99	cdevsw[major(tuncdev)] = tuncdevsw;
100
101	for (i = 0; i < NTUN; i++) {
102		tunctl[i].tun_flags = TUN_INITED;
103
104		ifp = &tunctl[i].tun_if;
105		ifp->if_unit = i;
106		ifp->if_name = "tun";
107		ifp->if_mtu = TUNMTU;
108		ifp->if_ioctl = tunifioctl;
109		ifp->if_output = tunoutput;
110		ifp->if_flags = IFF_POINTOPOINT;
111		ifp->if_snd.ifq_maxlen = ifqmaxlen;
112		ifp->if_collisions = 0;
113		ifp->if_ierrors = 0;
114		ifp->if_oerrors = 0;
115		ifp->if_ipackets = 0;
116		ifp->if_opackets = 0;
117		if_attach(ifp);
118#if NBPFILTER > 0
119		bpfattach(&tunctl[i].tun_bpf, ifp, DLT_NULL, sizeof(u_int));
120#endif
121	}
122}
123
124#ifdef __FreeBSD__
125PSEUDO_SET(tunattach, if_tun);
126#endif
127
128/*
129 * tunnel open - must be superuser & the device must be
130 * configured in
131 */
132int
133tunopen(dev, flag, mode, p)
134	dev_t	dev;
135	int	flag, mode;
136	struct proc *p;
137{
138	struct ifnet	*ifp;
139	struct tun_softc *tp;
140	register int	unit, error;
141
142	if (error = suser(p->p_ucred, &p->p_acflag))
143		return (error);
144
145	if ((unit = minor(dev)) >= NTUN)
146		return (ENXIO);
147	tp = &tunctl[unit];
148	if (tp->tun_flags & TUN_OPEN)
149		return ENXIO;
150	ifp = &tp->tun_if;
151	tp->tun_flags |= TUN_OPEN;
152	TUNDEBUG("%s%d: open\n", ifp->if_name, ifp->if_unit);
153	return (0);
154}
155
156/*
157 * tunclose - close the device - mark i/f down & delete
158 * routing info
159 */
160int
161tunclose(dev_t dev, int foo, int bar, struct proc *p)
162{
163	register int	unit = minor(dev), s;
164	struct tun_softc *tp = &tunctl[unit];
165	struct ifnet	*ifp = &tp->tun_if;
166	struct mbuf	*m;
167
168	tp->tun_flags &= ~TUN_OPEN;
169
170	/*
171	 * junk all pending output
172	 */
173	do {
174		s = splimp();
175		IF_DEQUEUE(&ifp->if_snd, m);
176		splx(s);
177		if (m)
178			m_freem(m);
179	} while (m);
180
181	if (ifp->if_flags & IFF_UP) {
182		s = splimp();
183		if_down(ifp);
184		if (ifp->if_flags & IFF_RUNNING) {
185		    /* find internet addresses and delete routes */
186		    register struct ifaddr *ifa;
187		    for (ifa = ifp->if_addrlist; ifa; ifa = ifa->ifa_next) {
188			if (ifa->ifa_addr->sa_family == AF_INET) {
189			    rtinit(ifa, (int)RTM_DELETE,
190				   tp->tun_flags & TUN_DSTADDR ? RTF_HOST : 0);
191			}
192		    }
193		}
194		splx(s);
195	}
196	tp->tun_pgrp = 0;
197	selwakeup(&tp->tun_rsel);
198
199	TUNDEBUG ("%s%d: closed\n", ifp->if_name, ifp->if_unit);
200	return (0);
201}
202
203static int
204tuninit(unit)
205	int	unit;
206{
207	struct tun_softc *tp = &tunctl[unit];
208	struct ifnet	*ifp = &tp->tun_if;
209	register struct ifaddr *ifa;
210
211	TUNDEBUG("%s%d: tuninit\n", ifp->if_name, ifp->if_unit);
212
213	ifp->if_flags |= IFF_UP | IFF_RUNNING;
214
215	for (ifa = ifp->if_addrlist; ifa; ifa = ifa->ifa_next)
216		if (ifa->ifa_addr->sa_family == AF_INET) {
217		    struct sockaddr_in *si;
218
219		    si = (struct sockaddr_in *)ifa->ifa_addr;
220		    if (si && si->sin_addr.s_addr)
221			    tp->tun_flags |= TUN_IASET;
222
223		    si = (struct sockaddr_in *)ifa->ifa_dstaddr;
224		    if (si && si->sin_addr.s_addr)
225			    tp->tun_flags |= TUN_DSTADDR;
226		}
227
228	return 0;
229}
230
231/*
232 * Process an ioctl request.
233 */
234int
235tunifioctl(ifp, cmd, data)
236	struct ifnet *ifp;
237	int	cmd;
238	caddr_t	data;
239{
240	struct tun_softc *tp = &tunctl[ifp->if_unit];
241	int		error = 0, s;
242
243	s = splimp();
244	switch(cmd) {
245	case SIOCSIFADDR:
246		tuninit(ifp->if_unit);
247		TUNDEBUG("%s%d: address set\n",
248			 ifp->if_name, ifp->if_unit);
249		break;
250	case SIOCSIFDSTADDR:
251		tuninit(ifp->if_unit);
252		TUNDEBUG("%s%d: destination address set\n",
253			 ifp->if_name, ifp->if_unit);
254		break;
255	default:
256		error = EINVAL;
257	}
258	splx(s);
259	return (error);
260}
261
262/*
263 * tunoutput - queue packets from higher level ready to put out.
264 */
265int
266tunoutput(ifp, m0, dst, rt)
267	struct ifnet   *ifp;
268	struct mbuf    *m0;
269	struct sockaddr *dst;
270	struct rtentry *rt;
271{
272	struct tun_softc *tp = &tunctl[ifp->if_unit];
273	struct proc	*p;
274	int		s;
275
276	TUNDEBUG ("%s%d: tunoutput\n", ifp->if_name, ifp->if_unit);
277
278	if ((tp->tun_flags & TUN_READY) != TUN_READY) {
279		TUNDEBUG ("%s%d: not ready 0%o\n", ifp->if_name,
280			  ifp->if_unit, tp->tun_flags);
281		m_freem (m0);
282		return EHOSTDOWN;
283	}
284
285#if NBPFILTER > 0
286	if (tp->tun_bpf) {
287		/*
288		 * We need to prepend the address family as
289		 * a four byte field.  Cons up a dummy header
290		 * to pacify bpf.  This is safe because bpf
291		 * will only read from the mbuf (i.e., it won't
292		 * try to free it or keep a pointer to it).
293		 */
294		struct mbuf m;
295		u_int af = dst->sa_family;
296
297		m.m_next = m0;
298		m.m_len = 4;
299		m.m_data = (char *)&af;
300
301		bpf_mtap(tp->tun_bpf, &m);
302	}
303#endif
304
305	switch(dst->sa_family) {
306#ifdef INET
307	case AF_INET:
308		s = splimp();
309		if (IF_QFULL(&ifp->if_snd)) {
310			IF_DROP(&ifp->if_snd);
311			m_freem(m0);
312			splx(s);
313			ifp->if_collisions++;
314			return (ENOBUFS);
315		}
316		IF_ENQUEUE(&ifp->if_snd, m0);
317		splx(s);
318		ifp->if_opackets++;
319		break;
320#endif
321	default:
322		m_freem(m0);
323		return EAFNOSUPPORT;
324	}
325
326	if (tp->tun_flags & TUN_RWAIT) {
327		tp->tun_flags &= ~TUN_RWAIT;
328		wakeup((caddr_t)tp);
329	}
330	if (tp->tun_flags & TUN_ASYNC && tp->tun_pgrp) {
331		if (tp->tun_pgrp > 0)
332			gsignal(tp->tun_pgrp, SIGIO);
333		else if (p = pfind(-tp->tun_pgrp))
334			psignal(p, SIGIO);
335	}
336	selwakeup(&tp->tun_rsel);
337	return 0;
338}
339
340/*
341 * the cdevsw interface is now pretty minimal.
342 */
343int
344tunioctl(dev, cmd, data, flag, p)
345	dev_t		dev;
346	int		cmd;
347	caddr_t		data;
348	int		flag;
349	struct proc	*p;
350{
351	int		unit = minor(dev), s;
352	struct tun_softc *tp = &tunctl[unit];
353 	struct tuninfo *tunp;
354
355	switch (cmd) {
356 	case TUNSIFINFO:
357 	        tunp = (struct tuninfo *)data;
358 		tp->tun_if.if_mtu = tunp->mtu;
359 		tp->tun_if.if_type = tunp->type;
360 		tp->tun_if.if_baudrate = tunp->baudrate;
361 		break;
362 	case TUNGIFINFO:
363 		tunp = (struct tuninfo *)data;
364 		tunp->mtu = tp->tun_if.if_mtu;
365 		tunp->type = tp->tun_if.if_type;
366 		tunp->baudrate = tp->tun_if.if_baudrate;
367 		break;
368	case TUNSDEBUG:
369		tundebug = *(int *)data;
370		break;
371	case TUNGDEBUG:
372		*(int *)data = tundebug;
373		break;
374	case FIONBIO:
375		if (*(int *)data)
376			tp->tun_flags |= TUN_NBIO;
377		else
378			tp->tun_flags &= ~TUN_NBIO;
379		break;
380	case FIOASYNC:
381		if (*(int *)data)
382			tp->tun_flags |= TUN_ASYNC;
383		else
384			tp->tun_flags &= ~TUN_ASYNC;
385		break;
386	case FIONREAD:
387		s = splimp();
388		if (tp->tun_if.if_snd.ifq_head)
389			*(int *)data = tp->tun_if.if_snd.ifq_head->m_len;
390		else
391			*(int *)data = 0;
392		splx(s);
393		break;
394	case TIOCSPGRP:
395		tp->tun_pgrp = *(int *)data;
396		break;
397	case TIOCGPGRP:
398		*(int *)data = tp->tun_pgrp;
399		break;
400	default:
401		return (ENOTTY);
402	}
403	return (0);
404}
405
406/*
407 * The cdevsw read interface - reads a packet at a time, or at
408 * least as much of a packet as can be read.
409 */
410int
411tunread(dev_t dev, struct uio *uio, int flag)
412{
413	int		unit = minor(dev);
414	struct tun_softc *tp = &tunctl[unit];
415	struct ifnet	*ifp = &tp->tun_if;
416	struct mbuf	*m, *m0;
417	int		error=0, len, s;
418
419	TUNDEBUG ("%s%d: read\n", ifp->if_name, ifp->if_unit);
420	if ((tp->tun_flags & TUN_READY) != TUN_READY) {
421		TUNDEBUG ("%s%d: not ready 0%o\n", ifp->if_name,
422			  ifp->if_unit, tp->tun_flags);
423		return EHOSTDOWN;
424	}
425
426	tp->tun_flags &= ~TUN_RWAIT;
427
428	s = splimp();
429	do {
430		IF_DEQUEUE(&ifp->if_snd, m0);
431		if (m0 == 0) {
432			if (tp->tun_flags & TUN_NBIO) {
433				splx(s);
434				return EWOULDBLOCK;
435			}
436			tp->tun_flags |= TUN_RWAIT;
437			tsleep((caddr_t)tp, PZERO + 1, "tunread", 0);
438		}
439	} while (m0 == 0);
440	splx(s);
441
442	while (m0 && uio->uio_resid > 0 && error == 0) {
443		len = min(uio->uio_resid, m0->m_len);
444		if (len == 0)
445			break;
446		error = uiomove(mtod(m0, caddr_t), len, uio);
447		MFREE(m0, m);
448		m0 = m;
449	}
450
451	if (m0) {
452		TUNDEBUG("Dropping mbuf\n");
453		m_freem(m0);
454	}
455	return error;
456}
457
458/*
459 * the cdevsw write interface - an atomic write is a packet - or else!
460 */
461int
462tunwrite(dev_t dev, struct uio *uio, int flag)
463{
464	int		unit = minor (dev);
465	struct ifnet	*ifp = &tunctl[unit].tun_if;
466	struct mbuf	*top, **mp, *m;
467	int		error=0, s, tlen, mlen;
468
469	TUNDEBUG("%s%d: tunwrite\n", ifp->if_name, ifp->if_unit);
470
471	if (uio->uio_resid < 0 || uio->uio_resid > TUNMTU) {
472		TUNDEBUG("%s%d: len=%d!\n", ifp->if_name, ifp->if_unit,
473		    uio->uio_resid);
474		return EIO;
475	}
476	tlen = uio->uio_resid;
477
478	/* get a header mbuf */
479	MGETHDR(m, M_DONTWAIT, MT_DATA);
480	if (m == NULL)
481		return ENOBUFS;
482	mlen = MHLEN;
483
484	top = 0;
485	mp = &top;
486	while (error == 0 && uio->uio_resid > 0) {
487		m->m_len = min(mlen, uio->uio_resid);
488		error = uiomove(mtod (m, caddr_t), m->m_len, uio);
489		*mp = m;
490		mp = &m->m_next;
491		if (uio->uio_resid > 0) {
492			MGET (m, M_DONTWAIT, MT_DATA);
493			if (m == 0) {
494				error = ENOBUFS;
495				break;
496			}
497			mlen = MLEN;
498		}
499	}
500	if (error) {
501		if (top)
502			m_freem (top);
503		return error;
504	}
505
506	top->m_pkthdr.len = tlen;
507	top->m_pkthdr.rcvif = ifp;
508
509#if NBPFILTER > 0
510	if (tunctl[unit].tun_bpf) {
511		/*
512		 * We need to prepend the address family as
513		 * a four byte field.  Cons up a dummy header
514		 * to pacify bpf.  This is safe because bpf
515		 * will only read from the mbuf (i.e., it won't
516		 * try to free it or keep a pointer to it).
517		 */
518		struct mbuf m;
519		u_int af = AF_INET;
520
521		m.m_next = top;
522		m.m_len = 4;
523		m.m_data = (char *)&af;
524
525		bpf_mtap(tunctl[unit].tun_bpf, &m);
526	}
527#endif
528
529	s = splimp();
530	if (IF_QFULL (&ipintrq)) {
531		IF_DROP(&ipintrq);
532		splx(s);
533		ifp->if_collisions++;
534		m_freem(top);
535		return ENOBUFS;
536	}
537	IF_ENQUEUE(&ipintrq, top);
538	splx(s);
539	ifp->if_ipackets++;
540	schednetisr(NETISR_IP);
541	return error;
542}
543
544/*
545 * tunselect - the select interface, this is only useful on reads
546 * really. The write detect always returns true, write never blocks
547 * anyway, it either accepts the packet or drops it.
548 */
549int
550tunselect(dev_t dev, int rw, struct proc *p)
551{
552	int		unit = minor(dev), s;
553	struct tun_softc *tp = &tunctl[unit];
554	struct ifnet	*ifp = &tp->tun_if;
555
556	s = splimp();
557	TUNDEBUG("%s%d: tunselect\n", ifp->if_name, ifp->if_unit);
558
559	switch (rw) {
560	case FREAD:
561		if (ifp->if_snd.ifq_len > 0) {
562			splx(s);
563			TUNDEBUG("%s%d: tunselect q=%d\n", ifp->if_name,
564			    ifp->if_unit, ifp->if_snd.ifq_len);
565			return 1;
566		}
567		selrecord(p, &tp->tun_rsel);
568		break;
569	case FWRITE:
570		splx(s);
571		return 1;
572	}
573	splx(s);
574	TUNDEBUG("%s%d: tunselect waiting\n", ifp->if_name, ifp->if_unit);
575	return 0;
576}
577
578#endif  /* NTUN */
579