if_tun.c revision 12071 
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
75int	tunoutput __P((struct ifnet *, struct mbuf *, struct sockaddr *,
76	    struct rtentry *rt));
77int	tunifioctl __P((struct ifnet *, int, caddr_t));
78
79static struct cdevsw tuncdevsw =
80{ tunopen,      tunclose,       tunread,        tunwrite,
81  tunioctl,     (d_stop_t *)enodev, (d_reset_t *)nullop, (d_ttycv_t *)enodev,
82  tunselect,    (d_mmap_t *)enodev,         NULL };
83extern dev_t tuncdev;
84
85static int tuninit __P((int));
86
87static void
88tunattach(udata)
89	void *udata;
90{
91	register int i;
92	struct ifnet *ifp;
93
94	/*
95	 * In case we are an LKM, set up device switch.
96	 */
97	cdevsw[major(tuncdev)] = tuncdevsw;
98
99	for (i = 0; i < NTUN; i++) {
100		tunctl[i].tun_flags = TUN_INITED;
101
102		ifp = &tunctl[i].tun_if;
103		ifp->if_unit = i;
104		ifp->if_name = "tun";
105		ifp->if_mtu = TUNMTU;
106		ifp->if_ioctl = tunifioctl;
107		ifp->if_output = tunoutput;
108		ifp->if_flags = IFF_POINTOPOINT | IFF_MULTICAST;
109		ifp->if_snd.ifq_maxlen = ifqmaxlen;
110		ifp->if_collisions = 0;
111		ifp->if_ierrors = 0;
112		ifp->if_oerrors = 0;
113		ifp->if_ipackets = 0;
114		ifp->if_opackets = 0;
115		if_attach(ifp);
116#if NBPFILTER > 0
117		bpfattach(&tunctl[i].tun_bpf, ifp, DLT_NULL, sizeof(u_int));
118#endif
119	}
120}
121
122/*
123 * tunnel open - must be superuser & the device must be
124 * configured in
125 */
126int
127tunopen(dev, flag, mode, p)
128	dev_t	dev;
129	int	flag, mode;
130	struct proc *p;
131{
132	struct ifnet	*ifp;
133	struct tun_softc *tp;
134	register int	unit, error;
135
136	if (error = suser(p->p_ucred, &p->p_acflag))
137		return (error);
138
139	if ((unit = minor(dev)) >= NTUN)
140		return (ENXIO);
141	tp = &tunctl[unit];
142	if (tp->tun_flags & TUN_OPEN)
143		return ENXIO;
144	ifp = &tp->tun_if;
145	tp->tun_flags |= TUN_OPEN;
146	TUNDEBUG("%s%d: open\n", ifp->if_name, ifp->if_unit);
147	return (0);
148}
149
150/*
151 * tunclose - close the device - mark i/f down & delete
152 * routing info
153 */
154int
155tunclose(dev_t dev, int foo, int bar, struct proc *p)
156{
157	register int	unit = minor(dev), s;
158	struct tun_softc *tp = &tunctl[unit];
159	struct ifnet	*ifp = &tp->tun_if;
160	struct mbuf	*m;
161
162	tp->tun_flags &= ~TUN_OPEN;
163
164	/*
165	 * junk all pending output
166	 */
167	do {
168		s = splimp();
169		IF_DEQUEUE(&ifp->if_snd, m);
170		splx(s);
171		if (m)
172			m_freem(m);
173	} while (m);
174
175	if (ifp->if_flags & IFF_UP) {
176		s = splimp();
177		if_down(ifp);
178		if (ifp->if_flags & IFF_RUNNING) {
179		    /* find internet addresses and delete routes */
180		    register struct ifaddr *ifa;
181		    for (ifa = ifp->if_addrlist; ifa; ifa = ifa->ifa_next) {
182			if (ifa->ifa_addr->sa_family == AF_INET) {
183			    rtinit(ifa, (int)RTM_DELETE,
184				   tp->tun_flags & TUN_DSTADDR ? RTF_HOST : 0);
185			}
186		    }
187		}
188		splx(s);
189	}
190	tp->tun_pgrp = 0;
191	selwakeup(&tp->tun_rsel);
192
193	TUNDEBUG ("%s%d: closed\n", ifp->if_name, ifp->if_unit);
194	return (0);
195}
196
197static int
198tuninit(unit)
199	int	unit;
200{
201	struct tun_softc *tp = &tunctl[unit];
202	struct ifnet	*ifp = &tp->tun_if;
203	register struct ifaddr *ifa;
204
205	TUNDEBUG("%s%d: tuninit\n", ifp->if_name, ifp->if_unit);
206
207	ifp->if_flags |= IFF_UP | IFF_RUNNING;
208
209	for (ifa = ifp->if_addrlist; ifa; ifa = ifa->ifa_next)
210		if (ifa->ifa_addr->sa_family == AF_INET) {
211		    struct sockaddr_in *si;
212
213		    si = (struct sockaddr_in *)ifa->ifa_addr;
214		    if (si && si->sin_addr.s_addr)
215			    tp->tun_flags |= TUN_IASET;
216
217		    si = (struct sockaddr_in *)ifa->ifa_dstaddr;
218		    if (si && si->sin_addr.s_addr)
219			    tp->tun_flags |= TUN_DSTADDR;
220		}
221
222	return 0;
223}
224
225/*
226 * Process an ioctl request.
227 */
228int
229tunifioctl(ifp, cmd, data)
230	struct ifnet *ifp;
231	int	cmd;
232	caddr_t	data;
233{
234	register struct ifreq *ifr = (struct ifreq *)data;
235	int		error = 0, s;
236
237	s = splimp();
238	switch(cmd) {
239	case SIOCSIFADDR:
240		tuninit(ifp->if_unit);
241		TUNDEBUG("%s%d: address set\n",
242			 ifp->if_name, ifp->if_unit);
243		break;
244	case SIOCSIFDSTADDR:
245		tuninit(ifp->if_unit);
246		TUNDEBUG("%s%d: destination address set\n",
247			 ifp->if_name, ifp->if_unit);
248		break;
249	case SIOCADDMULTI:
250	case SIOCDELMULTI:
251		if (ifr == 0) {
252			error = EAFNOSUPPORT;		/* XXX */
253			break;
254		}
255		switch (ifr->ifr_addr.sa_family) {
256
257#ifdef INET
258		case AF_INET:
259			break;
260#endif
261
262		default:
263			error = EAFNOSUPPORT;
264			break;
265		}
266		break;
267
268
269	default:
270		error = EINVAL;
271	}
272	splx(s);
273	return (error);
274}
275
276/*
277 * tunoutput - queue packets from higher level ready to put out.
278 */
279int
280tunoutput(ifp, m0, dst, rt)
281	struct ifnet   *ifp;
282	struct mbuf    *m0;
283	struct sockaddr *dst;
284	struct rtentry *rt;
285{
286	struct tun_softc *tp = &tunctl[ifp->if_unit];
287	struct proc	*p;
288	int		s;
289
290	TUNDEBUG ("%s%d: tunoutput\n", ifp->if_name, ifp->if_unit);
291
292	if ((tp->tun_flags & TUN_READY) != TUN_READY) {
293		TUNDEBUG ("%s%d: not ready 0%o\n", ifp->if_name,
294			  ifp->if_unit, tp->tun_flags);
295		m_freem (m0);
296		return EHOSTDOWN;
297	}
298
299#if NBPFILTER > 0
300	/* BPF write needs to be handled specially */
301	if (dst->sa_family == AF_UNSPEC) {
302		dst->sa_family = *(mtod(m0, int *));
303		m0->m_len -= sizeof(int);
304		m0->m_pkthdr.len -= sizeof(int);
305		m0->m_data += sizeof(int);
306	}
307
308	if (tp->tun_bpf) {
309		/*
310		 * We need to prepend the address family as
311		 * a four byte field.  Cons up a dummy header
312		 * to pacify bpf.  This is safe because bpf
313		 * will only read from the mbuf (i.e., it won't
314		 * try to free it or keep a pointer to it).
315		 */
316		struct mbuf m;
317		u_int af = dst->sa_family;
318
319		m.m_next = m0;
320		m.m_len = 4;
321		m.m_data = (char *)&af;
322
323		bpf_mtap(tp->tun_bpf, &m);
324	}
325#endif
326
327	switch(dst->sa_family) {
328#ifdef INET
329	case AF_INET:
330		s = splimp();
331		if (IF_QFULL(&ifp->if_snd)) {
332			IF_DROP(&ifp->if_snd);
333			m_freem(m0);
334			splx(s);
335			ifp->if_collisions++;
336			return (ENOBUFS);
337		}
338		IF_ENQUEUE(&ifp->if_snd, m0);
339		splx(s);
340		ifp->if_opackets++;
341		break;
342#endif
343	default:
344		m_freem(m0);
345		return EAFNOSUPPORT;
346	}
347
348	if (tp->tun_flags & TUN_RWAIT) {
349		tp->tun_flags &= ~TUN_RWAIT;
350		wakeup((caddr_t)tp);
351	}
352	if (tp->tun_flags & TUN_ASYNC && tp->tun_pgrp) {
353		if (tp->tun_pgrp > 0)
354			gsignal(tp->tun_pgrp, SIGIO);
355		else if (p = pfind(-tp->tun_pgrp))
356			psignal(p, SIGIO);
357	}
358	selwakeup(&tp->tun_rsel);
359	return 0;
360}
361
362/*
363 * the cdevsw interface is now pretty minimal.
364 */
365int
366tunioctl(dev, cmd, data, flag, p)
367	dev_t		dev;
368	int		cmd;
369	caddr_t		data;
370	int		flag;
371	struct proc	*p;
372{
373	int		unit = minor(dev), s;
374	struct tun_softc *tp = &tunctl[unit];
375 	struct tuninfo *tunp;
376
377	switch (cmd) {
378 	case TUNSIFINFO:
379 	        tunp = (struct tuninfo *)data;
380 		tp->tun_if.if_mtu = tunp->mtu;
381 		tp->tun_if.if_type = tunp->type;
382 		tp->tun_if.if_baudrate = tunp->baudrate;
383 		break;
384 	case TUNGIFINFO:
385 		tunp = (struct tuninfo *)data;
386 		tunp->mtu = tp->tun_if.if_mtu;
387 		tunp->type = tp->tun_if.if_type;
388 		tunp->baudrate = tp->tun_if.if_baudrate;
389 		break;
390	case TUNSDEBUG:
391		tundebug = *(int *)data;
392		break;
393	case TUNGDEBUG:
394		*(int *)data = tundebug;
395		break;
396	case FIONBIO:
397		if (*(int *)data)
398			tp->tun_flags |= TUN_NBIO;
399		else
400			tp->tun_flags &= ~TUN_NBIO;
401		break;
402	case FIOASYNC:
403		if (*(int *)data)
404			tp->tun_flags |= TUN_ASYNC;
405		else
406			tp->tun_flags &= ~TUN_ASYNC;
407		break;
408	case FIONREAD:
409		s = splimp();
410		if (tp->tun_if.if_snd.ifq_head)
411			*(int *)data = tp->tun_if.if_snd.ifq_head->m_len;
412		else
413			*(int *)data = 0;
414		splx(s);
415		break;
416	case TIOCSPGRP:
417		tp->tun_pgrp = *(int *)data;
418		break;
419	case TIOCGPGRP:
420		*(int *)data = tp->tun_pgrp;
421		break;
422	default:
423		return (ENOTTY);
424	}
425	return (0);
426}
427
428/*
429 * The cdevsw read interface - reads a packet at a time, or at
430 * least as much of a packet as can be read.
431 */
432int
433tunread(dev_t dev, struct uio *uio, int flag)
434{
435	int		unit = minor(dev);
436	struct tun_softc *tp = &tunctl[unit];
437	struct ifnet	*ifp = &tp->tun_if;
438	struct mbuf	*m, *m0;
439	int		error=0, len, s;
440
441	TUNDEBUG ("%s%d: read\n", ifp->if_name, ifp->if_unit);
442	if ((tp->tun_flags & TUN_READY) != TUN_READY) {
443		TUNDEBUG ("%s%d: not ready 0%o\n", ifp->if_name,
444			  ifp->if_unit, tp->tun_flags);
445		return EHOSTDOWN;
446	}
447
448	tp->tun_flags &= ~TUN_RWAIT;
449
450	s = splimp();
451	do {
452		IF_DEQUEUE(&ifp->if_snd, m0);
453		if (m0 == 0) {
454			if (tp->tun_flags & TUN_NBIO) {
455				splx(s);
456				return EWOULDBLOCK;
457			}
458			tp->tun_flags |= TUN_RWAIT;
459			tsleep((caddr_t)tp, PZERO + 1, "tunread", 0);
460		}
461	} while (m0 == 0);
462	splx(s);
463
464	while (m0 && uio->uio_resid > 0 && error == 0) {
465		len = min(uio->uio_resid, m0->m_len);
466		if (len == 0)
467			break;
468		error = uiomove(mtod(m0, caddr_t), len, uio);
469		MFREE(m0, m);
470		m0 = m;
471	}
472
473	if (m0) {
474		TUNDEBUG("Dropping mbuf\n");
475		m_freem(m0);
476	}
477	return error;
478}
479
480/*
481 * the cdevsw write interface - an atomic write is a packet - or else!
482 */
483int
484tunwrite(dev_t dev, struct uio *uio, int flag)
485{
486	int		unit = minor (dev);
487	struct ifnet	*ifp = &tunctl[unit].tun_if;
488	struct mbuf	*top, **mp, *m;
489	int		error=0, s, tlen, mlen;
490
491	TUNDEBUG("%s%d: tunwrite\n", ifp->if_name, ifp->if_unit);
492
493	if (uio->uio_resid < 0 || uio->uio_resid > TUNMTU) {
494		TUNDEBUG("%s%d: len=%d!\n", ifp->if_name, ifp->if_unit,
495		    uio->uio_resid);
496		return EIO;
497	}
498	tlen = uio->uio_resid;
499
500	/* get a header mbuf */
501	MGETHDR(m, M_DONTWAIT, MT_DATA);
502	if (m == NULL)
503		return ENOBUFS;
504	mlen = MHLEN;
505
506	top = 0;
507	mp = &top;
508	while (error == 0 && uio->uio_resid > 0) {
509		m->m_len = min(mlen, uio->uio_resid);
510		error = uiomove(mtod (m, caddr_t), m->m_len, uio);
511		*mp = m;
512		mp = &m->m_next;
513		if (uio->uio_resid > 0) {
514			MGET (m, M_DONTWAIT, MT_DATA);
515			if (m == 0) {
516				error = ENOBUFS;
517				break;
518			}
519			mlen = MLEN;
520		}
521	}
522	if (error) {
523		if (top)
524			m_freem (top);
525		return error;
526	}
527
528	top->m_pkthdr.len = tlen;
529	top->m_pkthdr.rcvif = ifp;
530
531#if NBPFILTER > 0
532	if (tunctl[unit].tun_bpf) {
533		/*
534		 * We need to prepend the address family as
535		 * a four byte field.  Cons up a dummy header
536		 * to pacify bpf.  This is safe because bpf
537		 * will only read from the mbuf (i.e., it won't
538		 * try to free it or keep a pointer to it).
539		 */
540		struct mbuf m;
541		u_int af = AF_INET;
542
543		m.m_next = top;
544		m.m_len = 4;
545		m.m_data = (char *)&af;
546
547		bpf_mtap(tunctl[unit].tun_bpf, &m);
548	}
549#endif
550
551	s = splimp();
552	if (IF_QFULL (&ipintrq)) {
553		IF_DROP(&ipintrq);
554		splx(s);
555		ifp->if_collisions++;
556		m_freem(top);
557		return ENOBUFS;
558	}
559	IF_ENQUEUE(&ipintrq, top);
560	splx(s);
561	ifp->if_ipackets++;
562	schednetisr(NETISR_IP);
563	return error;
564}
565
566/*
567 * tunselect - the select interface, this is only useful on reads
568 * really. The write detect always returns true, write never blocks
569 * anyway, it either accepts the packet or drops it.
570 */
571int
572tunselect(dev_t dev, int rw, struct proc *p)
573{
574	int		unit = minor(dev), s;
575	struct tun_softc *tp = &tunctl[unit];
576	struct ifnet	*ifp = &tp->tun_if;
577
578	s = splimp();
579	TUNDEBUG("%s%d: tunselect\n", ifp->if_name, ifp->if_unit);
580
581	switch (rw) {
582	case FREAD:
583		if (ifp->if_snd.ifq_len > 0) {
584			splx(s);
585			TUNDEBUG("%s%d: tunselect q=%d\n", ifp->if_name,
586			    ifp->if_unit, ifp->if_snd.ifq_len);
587			return 1;
588		}
589		selrecord(p, &tp->tun_rsel);
590		break;
591	case FWRITE:
592		splx(s);
593		return 1;
594	}
595	splx(s);
596	TUNDEBUG("%s%d: tunselect waiting\n", ifp->if_name, ifp->if_unit);
597	return 0;
598}
599
600#endif  /* NTUN */
601