if_tun.c revision 11921
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#ifdef __FreeBSD__
66static void tunattach __P((void *));
67PSEUDO_SET(tunattach, if_tun);
68#endif
69
70#define TUNDEBUG	if (tundebug) printf
71int	tundebug = 0;
72
73struct tun_softc tunctl[NTUN];
74
75d_open_t tunopen;
76d_close_t tunclose;
77int	tunoutput __P((struct ifnet *, struct mbuf *, struct sockaddr *,
78	    struct rtentry *rt));
79d_rdwr_t tunread;
80d_rdwr_t tunwrite;
81d_ioctl_t tunioctl;
82int	tunifioctl __P((struct ifnet *, int, caddr_t));
83d_select_t tunselect;
84
85static struct cdevsw tuncdevsw =
86{ tunopen,      tunclose,       tunread,        tunwrite,
87  tunioctl,     (d_stop_t *)enodev, (d_reset_t *)nullop, (d_ttycv_t *)enodev,
88  tunselect,    (d_mmap_t *)enodev,         NULL };
89extern dev_t tuncdev;
90
91static int tuninit __P((int));
92
93static void
94tunattach(udata)
95	void *udata;
96{
97	register int i;
98	struct ifnet *ifp;
99
100	/*
101	 * In case we are an LKM, set up device switch.
102	 */
103	cdevsw[major(tuncdev)] = tuncdevsw;
104
105	for (i = 0; i < NTUN; i++) {
106		tunctl[i].tun_flags = TUN_INITED;
107
108		ifp = &tunctl[i].tun_if;
109		ifp->if_unit = i;
110		ifp->if_name = "tun";
111		ifp->if_mtu = TUNMTU;
112		ifp->if_ioctl = tunifioctl;
113		ifp->if_output = tunoutput;
114		ifp->if_flags = IFF_POINTOPOINT | IFF_MULTICAST;
115		ifp->if_snd.ifq_maxlen = ifqmaxlen;
116		ifp->if_collisions = 0;
117		ifp->if_ierrors = 0;
118		ifp->if_oerrors = 0;
119		ifp->if_ipackets = 0;
120		ifp->if_opackets = 0;
121		if_attach(ifp);
122#if NBPFILTER > 0
123		bpfattach(&tunctl[i].tun_bpf, ifp, DLT_NULL, sizeof(u_int));
124#endif
125	}
126}
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	register struct ifreq *ifr = (struct ifreq *)data;
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	case SIOCADDMULTI:
256	case SIOCDELMULTI:
257		if (ifr == 0) {
258			error = EAFNOSUPPORT;		/* XXX */
259			break;
260		}
261		switch (ifr->ifr_addr.sa_family) {
262
263#ifdef INET
264		case AF_INET:
265			break;
266#endif
267
268		default:
269			error = EAFNOSUPPORT;
270			break;
271		}
272		break;
273
274
275	default:
276		error = EINVAL;
277	}
278	splx(s);
279	return (error);
280}
281
282/*
283 * tunoutput - queue packets from higher level ready to put out.
284 */
285int
286tunoutput(ifp, m0, dst, rt)
287	struct ifnet   *ifp;
288	struct mbuf    *m0;
289	struct sockaddr *dst;
290	struct rtentry *rt;
291{
292	struct tun_softc *tp = &tunctl[ifp->if_unit];
293	struct proc	*p;
294	int		s;
295
296	TUNDEBUG ("%s%d: tunoutput\n", ifp->if_name, ifp->if_unit);
297
298	if ((tp->tun_flags & TUN_READY) != TUN_READY) {
299		TUNDEBUG ("%s%d: not ready 0%o\n", ifp->if_name,
300			  ifp->if_unit, tp->tun_flags);
301		m_freem (m0);
302		return EHOSTDOWN;
303	}
304
305#if NBPFILTER > 0
306	/* BPF write needs to be handled specially */
307	if (dst->sa_family == AF_UNSPEC) {
308		dst->sa_family = *(mtod(m0, int *));
309		m0->m_len -= sizeof(int);
310		m0->m_pkthdr.len -= sizeof(int);
311		m0->m_data += sizeof(int);
312	}
313
314	if (tp->tun_bpf) {
315		/*
316		 * We need to prepend the address family as
317		 * a four byte field.  Cons up a dummy header
318		 * to pacify bpf.  This is safe because bpf
319		 * will only read from the mbuf (i.e., it won't
320		 * try to free it or keep a pointer to it).
321		 */
322		struct mbuf m;
323		u_int af = dst->sa_family;
324
325		m.m_next = m0;
326		m.m_len = 4;
327		m.m_data = (char *)&af;
328
329		bpf_mtap(tp->tun_bpf, &m);
330	}
331#endif
332
333	switch(dst->sa_family) {
334#ifdef INET
335	case AF_INET:
336		s = splimp();
337		if (IF_QFULL(&ifp->if_snd)) {
338			IF_DROP(&ifp->if_snd);
339			m_freem(m0);
340			splx(s);
341			ifp->if_collisions++;
342			return (ENOBUFS);
343		}
344		IF_ENQUEUE(&ifp->if_snd, m0);
345		splx(s);
346		ifp->if_opackets++;
347		break;
348#endif
349	default:
350		m_freem(m0);
351		return EAFNOSUPPORT;
352	}
353
354	if (tp->tun_flags & TUN_RWAIT) {
355		tp->tun_flags &= ~TUN_RWAIT;
356		wakeup((caddr_t)tp);
357	}
358	if (tp->tun_flags & TUN_ASYNC && tp->tun_pgrp) {
359		if (tp->tun_pgrp > 0)
360			gsignal(tp->tun_pgrp, SIGIO);
361		else if (p = pfind(-tp->tun_pgrp))
362			psignal(p, SIGIO);
363	}
364	selwakeup(&tp->tun_rsel);
365	return 0;
366}
367
368/*
369 * the cdevsw interface is now pretty minimal.
370 */
371int
372tunioctl(dev, cmd, data, flag, p)
373	dev_t		dev;
374	int		cmd;
375	caddr_t		data;
376	int		flag;
377	struct proc	*p;
378{
379	int		unit = minor(dev), s;
380	struct tun_softc *tp = &tunctl[unit];
381 	struct tuninfo *tunp;
382
383	switch (cmd) {
384 	case TUNSIFINFO:
385 	        tunp = (struct tuninfo *)data;
386 		tp->tun_if.if_mtu = tunp->mtu;
387 		tp->tun_if.if_type = tunp->type;
388 		tp->tun_if.if_baudrate = tunp->baudrate;
389 		break;
390 	case TUNGIFINFO:
391 		tunp = (struct tuninfo *)data;
392 		tunp->mtu = tp->tun_if.if_mtu;
393 		tunp->type = tp->tun_if.if_type;
394 		tunp->baudrate = tp->tun_if.if_baudrate;
395 		break;
396	case TUNSDEBUG:
397		tundebug = *(int *)data;
398		break;
399	case TUNGDEBUG:
400		*(int *)data = tundebug;
401		break;
402	case FIONBIO:
403		if (*(int *)data)
404			tp->tun_flags |= TUN_NBIO;
405		else
406			tp->tun_flags &= ~TUN_NBIO;
407		break;
408	case FIOASYNC:
409		if (*(int *)data)
410			tp->tun_flags |= TUN_ASYNC;
411		else
412			tp->tun_flags &= ~TUN_ASYNC;
413		break;
414	case FIONREAD:
415		s = splimp();
416		if (tp->tun_if.if_snd.ifq_head)
417			*(int *)data = tp->tun_if.if_snd.ifq_head->m_len;
418		else
419			*(int *)data = 0;
420		splx(s);
421		break;
422	case TIOCSPGRP:
423		tp->tun_pgrp = *(int *)data;
424		break;
425	case TIOCGPGRP:
426		*(int *)data = tp->tun_pgrp;
427		break;
428	default:
429		return (ENOTTY);
430	}
431	return (0);
432}
433
434/*
435 * The cdevsw read interface - reads a packet at a time, or at
436 * least as much of a packet as can be read.
437 */
438int
439tunread(dev_t dev, struct uio *uio, int flag)
440{
441	int		unit = minor(dev);
442	struct tun_softc *tp = &tunctl[unit];
443	struct ifnet	*ifp = &tp->tun_if;
444	struct mbuf	*m, *m0;
445	int		error=0, len, s;
446
447	TUNDEBUG ("%s%d: read\n", ifp->if_name, ifp->if_unit);
448	if ((tp->tun_flags & TUN_READY) != TUN_READY) {
449		TUNDEBUG ("%s%d: not ready 0%o\n", ifp->if_name,
450			  ifp->if_unit, tp->tun_flags);
451		return EHOSTDOWN;
452	}
453
454	tp->tun_flags &= ~TUN_RWAIT;
455
456	s = splimp();
457	do {
458		IF_DEQUEUE(&ifp->if_snd, m0);
459		if (m0 == 0) {
460			if (tp->tun_flags & TUN_NBIO) {
461				splx(s);
462				return EWOULDBLOCK;
463			}
464			tp->tun_flags |= TUN_RWAIT;
465			tsleep((caddr_t)tp, PZERO + 1, "tunread", 0);
466		}
467	} while (m0 == 0);
468	splx(s);
469
470	while (m0 && uio->uio_resid > 0 && error == 0) {
471		len = min(uio->uio_resid, m0->m_len);
472		if (len == 0)
473			break;
474		error = uiomove(mtod(m0, caddr_t), len, uio);
475		MFREE(m0, m);
476		m0 = m;
477	}
478
479	if (m0) {
480		TUNDEBUG("Dropping mbuf\n");
481		m_freem(m0);
482	}
483	return error;
484}
485
486/*
487 * the cdevsw write interface - an atomic write is a packet - or else!
488 */
489int
490tunwrite(dev_t dev, struct uio *uio, int flag)
491{
492	int		unit = minor (dev);
493	struct ifnet	*ifp = &tunctl[unit].tun_if;
494	struct mbuf	*top, **mp, *m;
495	int		error=0, s, tlen, mlen;
496
497	TUNDEBUG("%s%d: tunwrite\n", ifp->if_name, ifp->if_unit);
498
499	if (uio->uio_resid < 0 || uio->uio_resid > TUNMTU) {
500		TUNDEBUG("%s%d: len=%d!\n", ifp->if_name, ifp->if_unit,
501		    uio->uio_resid);
502		return EIO;
503	}
504	tlen = uio->uio_resid;
505
506	/* get a header mbuf */
507	MGETHDR(m, M_DONTWAIT, MT_DATA);
508	if (m == NULL)
509		return ENOBUFS;
510	mlen = MHLEN;
511
512	top = 0;
513	mp = &top;
514	while (error == 0 && uio->uio_resid > 0) {
515		m->m_len = min(mlen, uio->uio_resid);
516		error = uiomove(mtod (m, caddr_t), m->m_len, uio);
517		*mp = m;
518		mp = &m->m_next;
519		if (uio->uio_resid > 0) {
520			MGET (m, M_DONTWAIT, MT_DATA);
521			if (m == 0) {
522				error = ENOBUFS;
523				break;
524			}
525			mlen = MLEN;
526		}
527	}
528	if (error) {
529		if (top)
530			m_freem (top);
531		return error;
532	}
533
534	top->m_pkthdr.len = tlen;
535	top->m_pkthdr.rcvif = ifp;
536
537#if NBPFILTER > 0
538	if (tunctl[unit].tun_bpf) {
539		/*
540		 * We need to prepend the address family as
541		 * a four byte field.  Cons up a dummy header
542		 * to pacify bpf.  This is safe because bpf
543		 * will only read from the mbuf (i.e., it won't
544		 * try to free it or keep a pointer to it).
545		 */
546		struct mbuf m;
547		u_int af = AF_INET;
548
549		m.m_next = top;
550		m.m_len = 4;
551		m.m_data = (char *)&af;
552
553		bpf_mtap(tunctl[unit].tun_bpf, &m);
554	}
555#endif
556
557	s = splimp();
558	if (IF_QFULL (&ipintrq)) {
559		IF_DROP(&ipintrq);
560		splx(s);
561		ifp->if_collisions++;
562		m_freem(top);
563		return ENOBUFS;
564	}
565	IF_ENQUEUE(&ipintrq, top);
566	splx(s);
567	ifp->if_ipackets++;
568	schednetisr(NETISR_IP);
569	return error;
570}
571
572/*
573 * tunselect - the select interface, this is only useful on reads
574 * really. The write detect always returns true, write never blocks
575 * anyway, it either accepts the packet or drops it.
576 */
577int
578tunselect(dev_t dev, int rw, struct proc *p)
579{
580	int		unit = minor(dev), s;
581	struct tun_softc *tp = &tunctl[unit];
582	struct ifnet	*ifp = &tp->tun_if;
583
584	s = splimp();
585	TUNDEBUG("%s%d: tunselect\n", ifp->if_name, ifp->if_unit);
586
587	switch (rw) {
588	case FREAD:
589		if (ifp->if_snd.ifq_len > 0) {
590			splx(s);
591			TUNDEBUG("%s%d: tunselect q=%d\n", ifp->if_name,
592			    ifp->if_unit, ifp->if_snd.ifq_len);
593			return 1;
594		}
595		selrecord(p, &tp->tun_rsel);
596		break;
597	case FWRITE:
598		splx(s);
599		return 1;
600	}
601	splx(s);
602	TUNDEBUG("%s%d: tunselect waiting\n", ifp->if_name, ifp->if_unit);
603	return 0;
604}
605
606#endif  /* NTUN */
607