xref: /netbsd-src/sys/net/if_tun.c (revision fdd524d4ccd2bb0c6f67401e938dabf773eb0372)
1 /*	$NetBSD: if_tun.c,v 1.127 2016/07/07 09:32:02 ozaki-r 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 its
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 
17 #include <sys/cdefs.h>
18 __KERNEL_RCSID(0, "$NetBSD: if_tun.c,v 1.127 2016/07/07 09:32:02 ozaki-r Exp $");
19 
20 #ifdef _KERNEL_OPT
21 #include "opt_inet.h"
22 #endif
23 
24 #include <sys/param.h>
25 #include <sys/proc.h>
26 #include <sys/systm.h>
27 #include <sys/mbuf.h>
28 #include <sys/buf.h>
29 #include <sys/protosw.h>
30 #include <sys/socket.h>
31 #include <sys/ioctl.h>
32 #include <sys/errno.h>
33 #include <sys/syslog.h>
34 #include <sys/select.h>
35 #include <sys/poll.h>
36 #include <sys/file.h>
37 #include <sys/signalvar.h>
38 #include <sys/conf.h>
39 #include <sys/kauth.h>
40 #include <sys/mutex.h>
41 #include <sys/cpu.h>
42 
43 #include <net/if.h>
44 #include <net/if_types.h>
45 #include <net/netisr.h>
46 #include <net/route.h>
47 
48 
49 #ifdef INET
50 #include <netinet/in.h>
51 #include <netinet/in_systm.h>
52 #include <netinet/in_var.h>
53 #include <netinet/ip.h>
54 #include <netinet/if_inarp.h>
55 #endif
56 
57 
58 #include <sys/time.h>
59 #include <net/bpf.h>
60 
61 #include <net/if_tun.h>
62 
63 #include "ioconf.h"
64 
65 #define TUNDEBUG	if (tundebug) printf
66 int	tundebug = 0;
67 
68 extern int ifqmaxlen;
69 
70 static LIST_HEAD(, tun_softc) tun_softc_list;
71 static LIST_HEAD(, tun_softc) tunz_softc_list;
72 static kmutex_t tun_softc_lock;
73 
74 static int	tun_ioctl(struct ifnet *, u_long, void *);
75 static int	tun_output(struct ifnet *, struct mbuf *,
76 			const struct sockaddr *, const struct rtentry *rt);
77 static int	tun_clone_create(struct if_clone *, int);
78 static int	tun_clone_destroy(struct ifnet *);
79 
80 static struct if_clone tun_cloner =
81     IF_CLONE_INITIALIZER("tun", tun_clone_create, tun_clone_destroy);
82 
83 static void tunattach0(struct tun_softc *);
84 static void tuninit(struct tun_softc *);
85 static void tun_i_softintr(void *);
86 static void tun_o_softintr(void *);
87 #ifdef ALTQ
88 static void tunstart(struct ifnet *);
89 #endif
90 static struct tun_softc *tun_find_unit(dev_t);
91 static struct tun_softc *tun_find_zunit(int);
92 
93 static dev_type_open(tunopen);
94 static dev_type_close(tunclose);
95 static dev_type_read(tunread);
96 static dev_type_write(tunwrite);
97 static dev_type_ioctl(tunioctl);
98 static dev_type_poll(tunpoll);
99 static dev_type_kqfilter(tunkqfilter);
100 
101 const struct cdevsw tun_cdevsw = {
102 	.d_open = tunopen,
103 	.d_close = tunclose,
104 	.d_read = tunread,
105 	.d_write = tunwrite,
106 	.d_ioctl = tunioctl,
107 	.d_stop = nostop,
108 	.d_tty = notty,
109 	.d_poll = tunpoll,
110 	.d_mmap = nommap,
111 	.d_kqfilter = tunkqfilter,
112 	.d_discard = nodiscard,
113 	.d_flag = D_OTHER
114 };
115 
116 void
117 tunattach(int unused)
118 {
119 
120 	mutex_init(&tun_softc_lock, MUTEX_DEFAULT, IPL_NET);
121 	LIST_INIT(&tun_softc_list);
122 	LIST_INIT(&tunz_softc_list);
123 	if_clone_attach(&tun_cloner);
124 }
125 
126 /*
127  * Find driver instance from dev_t.
128  * Returns with tp locked (if found).
129  */
130 static struct tun_softc *
131 tun_find_unit(dev_t dev)
132 {
133 	struct tun_softc *tp;
134 	int unit = minor(dev);
135 
136 	mutex_enter(&tun_softc_lock);
137 	LIST_FOREACH(tp, &tun_softc_list, tun_list)
138 		if (unit == tp->tun_unit)
139 			break;
140 	if (tp)
141 		mutex_enter(&tp->tun_lock);
142 	mutex_exit(&tun_softc_lock);
143 
144 	return (tp);
145 }
146 
147 /*
148  * Find zombie driver instance by unit number.
149  * Remove tp from list and return it unlocked (if found).
150  */
151 static struct tun_softc *
152 tun_find_zunit(int unit)
153 {
154 	struct tun_softc *tp;
155 
156 	mutex_enter(&tun_softc_lock);
157 	LIST_FOREACH(tp, &tunz_softc_list, tun_list)
158 		if (unit == tp->tun_unit)
159 			break;
160 	if (tp)
161 		LIST_REMOVE(tp, tun_list);
162 	mutex_exit(&tun_softc_lock);
163 #ifdef DIAGNOSTIC
164 	if (tp != NULL && (tp->tun_flags & (TUN_INITED|TUN_OPEN)) != TUN_OPEN)
165 		printf("tun%d: inconsistent flags: %x\n", unit, tp->tun_flags);
166 #endif
167 
168 	return (tp);
169 }
170 
171 static int
172 tun_clone_create(struct if_clone *ifc, int unit)
173 {
174 	struct tun_softc *tp;
175 
176 	if ((tp = tun_find_zunit(unit)) == NULL) {
177 		/* Allocate a new instance */
178 		tp = malloc(sizeof(*tp), M_DEVBUF, M_WAITOK|M_ZERO);
179 
180 		tp->tun_unit = unit;
181 		mutex_init(&tp->tun_lock, MUTEX_DEFAULT, IPL_NET);
182 		selinit(&tp->tun_rsel);
183 		selinit(&tp->tun_wsel);
184 	} else {
185 		/* Revive tunnel instance; clear ifp part */
186 		(void)memset(&tp->tun_if, 0, sizeof(struct ifnet));
187 	}
188 
189 	if_initname(&tp->tun_if, ifc->ifc_name, unit);
190 	tunattach0(tp);
191 	tp->tun_flags |= TUN_INITED;
192 	tp->tun_osih = softint_establish(SOFTINT_CLOCK, tun_o_softintr, tp);
193 	tp->tun_isih = softint_establish(SOFTINT_CLOCK, tun_i_softintr, tp);
194 
195 	mutex_enter(&tun_softc_lock);
196 	LIST_INSERT_HEAD(&tun_softc_list, tp, tun_list);
197 	mutex_exit(&tun_softc_lock);
198 
199 	return (0);
200 }
201 
202 static void
203 tunattach0(struct tun_softc *tp)
204 {
205 	struct ifnet *ifp;
206 
207 	ifp = &tp->tun_if;
208 	ifp->if_softc = tp;
209 	ifp->if_mtu = TUNMTU;
210 	ifp->if_ioctl = tun_ioctl;
211 	ifp->if_output = tun_output;
212 #ifdef ALTQ
213 	ifp->if_start = tunstart;
214 #endif
215 	ifp->if_flags = IFF_POINTOPOINT;
216 	ifp->if_type = IFT_TUNNEL;
217 	ifp->if_snd.ifq_maxlen = ifqmaxlen;
218 	ifp->if_collisions = 0;
219 	ifp->if_ierrors = 0;
220 	ifp->if_oerrors = 0;
221 	ifp->if_ipackets = 0;
222 	ifp->if_opackets = 0;
223 	ifp->if_ibytes   = 0;
224 	ifp->if_obytes   = 0;
225 	ifp->if_dlt = DLT_NULL;
226 	IFQ_SET_READY(&ifp->if_snd);
227 	if_attach(ifp);
228 	if_alloc_sadl(ifp);
229 	bpf_attach(ifp, DLT_NULL, sizeof(uint32_t));
230 }
231 
232 static int
233 tun_clone_destroy(struct ifnet *ifp)
234 {
235 	struct tun_softc *tp = (void *)ifp;
236 	int zombie = 0;
237 
238 	IF_PURGE(&ifp->if_snd);
239 	ifp->if_flags &= ~IFF_RUNNING;
240 
241 	mutex_enter(&tun_softc_lock);
242 	mutex_enter(&tp->tun_lock);
243 	LIST_REMOVE(tp, tun_list);
244 	if (tp->tun_flags & TUN_OPEN) {
245 		/* Hang on to storage until last close */
246 		zombie = 1;
247 		tp->tun_flags &= ~TUN_INITED;
248 		LIST_INSERT_HEAD(&tunz_softc_list, tp, tun_list);
249 	}
250 	mutex_exit(&tun_softc_lock);
251 
252 	if (tp->tun_flags & TUN_RWAIT) {
253 		tp->tun_flags &= ~TUN_RWAIT;
254 		wakeup((void *)tp);
255 	}
256 	selnotify(&tp->tun_rsel, 0, 0);
257 
258 	mutex_exit(&tp->tun_lock);
259 
260 	if (tp->tun_flags & TUN_ASYNC && tp->tun_pgid)
261 		fownsignal(tp->tun_pgid, SIGIO, POLL_HUP, 0, NULL);
262 
263 	bpf_detach(ifp);
264 	if_detach(ifp);
265 
266 	if (!zombie) {
267 		seldestroy(&tp->tun_rsel);
268 		seldestroy(&tp->tun_wsel);
269 		softint_disestablish(tp->tun_osih);
270 		softint_disestablish(tp->tun_isih);
271 		mutex_destroy(&tp->tun_lock);
272 		free(tp, M_DEVBUF);
273 	}
274 
275 	return (0);
276 }
277 
278 /*
279  * tunnel open - must be superuser & the device must be
280  * configured in
281  */
282 static int
283 tunopen(dev_t dev, int flag, int mode, struct lwp *l)
284 {
285 	struct ifnet	*ifp;
286 	struct tun_softc *tp;
287 	int	error;
288 
289 	error = kauth_authorize_network(l->l_cred, KAUTH_NETWORK_INTERFACE_TUN,
290 	    KAUTH_REQ_NETWORK_INTERFACE_TUN_ADD, NULL, NULL, NULL);
291 	if (error)
292 		return (error);
293 
294 	tp = tun_find_unit(dev);
295 
296 	if (tp == NULL) {
297 		(void)tun_clone_create(&tun_cloner, minor(dev));
298 		tp = tun_find_unit(dev);
299 		if (tp == NULL) {
300 			error = ENXIO;
301 			goto out_nolock;
302 		}
303 	}
304 
305 	if (tp->tun_flags & TUN_OPEN) {
306 		error = EBUSY;
307 		goto out;
308 	}
309 
310 	ifp = &tp->tun_if;
311 	tp->tun_flags |= TUN_OPEN;
312 	TUNDEBUG("%s: open\n", ifp->if_xname);
313 out:
314 	mutex_exit(&tp->tun_lock);
315 out_nolock:
316 	return (error);
317 }
318 
319 /*
320  * tunclose - close the device - mark i/f down & delete
321  * routing info
322  */
323 int
324 tunclose(dev_t dev, int flag, int mode,
325     struct lwp *l)
326 {
327 	struct tun_softc *tp;
328 	struct ifnet	*ifp;
329 
330 	if ((tp = tun_find_zunit(minor(dev))) != NULL) {
331 		/* interface was "destroyed" before the close */
332 		seldestroy(&tp->tun_rsel);
333 		seldestroy(&tp->tun_wsel);
334 		softint_disestablish(tp->tun_osih);
335 		softint_disestablish(tp->tun_isih);
336 		mutex_destroy(&tp->tun_lock);
337 		free(tp, M_DEVBUF);
338 		goto out_nolock;
339 	}
340 
341 	if ((tp = tun_find_unit(dev)) == NULL)
342 		goto out_nolock;
343 
344 	ifp = &tp->tun_if;
345 
346 	tp->tun_flags &= ~TUN_OPEN;
347 
348 	tp->tun_pgid = 0;
349 	selnotify(&tp->tun_rsel, 0, 0);
350 
351 	TUNDEBUG ("%s: closed\n", ifp->if_xname);
352 	mutex_exit(&tp->tun_lock);
353 
354 	/*
355 	 * junk all pending output
356 	 */
357 	IFQ_PURGE(&ifp->if_snd);
358 
359 	if (ifp->if_flags & IFF_UP) {
360 		if_down(ifp);
361 		if (ifp->if_flags & IFF_RUNNING) {
362 			/* find internet addresses and delete routes */
363 			struct ifaddr *ifa;
364 			IFADDR_READER_FOREACH(ifa, ifp) {
365 #if defined(INET) || defined(INET6)
366 				if (ifa->ifa_addr->sa_family == AF_INET ||
367 				    ifa->ifa_addr->sa_family == AF_INET6) {
368 					rtinit(ifa, (int)RTM_DELETE,
369 					       tp->tun_flags & TUN_DSTADDR
370 							? RTF_HOST
371 							: 0);
372 				}
373 #endif
374 			}
375 		}
376 	}
377 out_nolock:
378 	return (0);
379 }
380 
381 /*
382  * Call at splnet().
383  */
384 static void
385 tuninit(struct tun_softc *tp)
386 {
387 	struct ifnet	*ifp = &tp->tun_if;
388 	struct ifaddr	*ifa;
389 
390 	TUNDEBUG("%s: tuninit\n", ifp->if_xname);
391 
392 	mutex_enter(&tp->tun_lock);
393 	ifp->if_flags |= IFF_UP | IFF_RUNNING;
394 
395 	tp->tun_flags &= ~(TUN_IASET|TUN_DSTADDR);
396 	IFADDR_READER_FOREACH(ifa, ifp) {
397 #ifdef INET
398 		if (ifa->ifa_addr->sa_family == AF_INET) {
399 			struct sockaddr_in *sin;
400 
401 			sin = satosin(ifa->ifa_addr);
402 			if (sin && sin->sin_addr.s_addr)
403 				tp->tun_flags |= TUN_IASET;
404 
405 			if (ifp->if_flags & IFF_POINTOPOINT) {
406 				sin = satosin(ifa->ifa_dstaddr);
407 				if (sin && sin->sin_addr.s_addr)
408 					tp->tun_flags |= TUN_DSTADDR;
409 			}
410 		}
411 #endif
412 #ifdef INET6
413 		if (ifa->ifa_addr->sa_family == AF_INET6) {
414 			struct sockaddr_in6 *sin;
415 
416 			sin = (struct sockaddr_in6 *)ifa->ifa_addr;
417 			if (!IN6_IS_ADDR_UNSPECIFIED(&sin->sin6_addr))
418 				tp->tun_flags |= TUN_IASET;
419 
420 			if (ifp->if_flags & IFF_POINTOPOINT) {
421 				sin = (struct sockaddr_in6 *)ifa->ifa_dstaddr;
422 				if (sin &&
423 				    !IN6_IS_ADDR_UNSPECIFIED(&sin->sin6_addr))
424 					tp->tun_flags |= TUN_DSTADDR;
425 			} else
426 				tp->tun_flags &= ~TUN_DSTADDR;
427 		}
428 #endif /* INET6 */
429 	}
430 	mutex_exit(&tp->tun_lock);
431 }
432 
433 /*
434  * Process an ioctl request.
435  */
436 static int
437 tun_ioctl(struct ifnet *ifp, u_long cmd, void *data)
438 {
439 	int		error = 0, s;
440 	struct tun_softc *tp = (struct tun_softc *)(ifp->if_softc);
441 	struct ifreq *ifr = (struct ifreq *)data;
442 	struct ifaddr *ifa = (struct ifaddr *)data;
443 
444 	s = splnet();
445 
446 	switch (cmd) {
447 	case SIOCINITIFADDR:
448 		tuninit(tp);
449 		ifa->ifa_rtrequest = p2p_rtrequest;
450 		TUNDEBUG("%s: address set\n", ifp->if_xname);
451 		break;
452 	case SIOCSIFBRDADDR:
453 		TUNDEBUG("%s: broadcast address set\n", ifp->if_xname);
454 		break;
455 	case SIOCSIFMTU:
456 		if (ifr->ifr_mtu > TUNMTU || ifr->ifr_mtu < 576) {
457 			error = EINVAL;
458 			break;
459 		}
460 		TUNDEBUG("%s: interface mtu set\n", ifp->if_xname);
461 		if ((error = ifioctl_common(ifp, cmd, data)) == ENETRESET)
462 			error = 0;
463 		break;
464 	case SIOCADDMULTI:
465 	case SIOCDELMULTI:
466 		if (ifr == NULL) {
467 	        	error = EAFNOSUPPORT;           /* XXX */
468 			break;
469 		}
470 		switch (ifreq_getaddr(cmd, ifr)->sa_family) {
471 #ifdef INET
472 		case AF_INET:
473 			break;
474 #endif
475 #ifdef INET6
476 		case AF_INET6:
477 			break;
478 #endif
479 		default:
480 			error = EAFNOSUPPORT;
481 			break;
482 		}
483 		break;
484 	default:
485 		error = ifioctl_common(ifp, cmd, data);
486 	}
487 
488 	splx(s);
489 	return (error);
490 }
491 
492 /*
493  * tun_output - queue packets from higher level ready to put out.
494  */
495 static int
496 tun_output(struct ifnet *ifp, struct mbuf *m0, const struct sockaddr *dst,
497     const struct rtentry *rt)
498 {
499 	struct tun_softc *tp = ifp->if_softc;
500 	int		s;
501 	int		error;
502 #if defined(INET) || defined(INET6)
503 	int		mlen;
504 	uint32_t	*af;
505 #endif
506 
507 	s = splnet();
508 	mutex_enter(&tp->tun_lock);
509 	TUNDEBUG ("%s: tun_output\n", ifp->if_xname);
510 
511 	if ((tp->tun_flags & TUN_READY) != TUN_READY) {
512 		TUNDEBUG ("%s: not ready 0%o\n", ifp->if_xname,
513 			  tp->tun_flags);
514 		error = EHOSTDOWN;
515 		goto out;
516 	}
517 
518 	/*
519 	 * if the queueing discipline needs packet classification,
520 	 * do it before prepending link headers.
521 	 */
522 	IFQ_CLASSIFY(&ifp->if_snd, m0, dst->sa_family);
523 
524 	bpf_mtap_af(ifp, dst->sa_family, m0);
525 
526 	switch(dst->sa_family) {
527 #ifdef INET6
528 	case AF_INET6:
529 #endif
530 #ifdef INET
531 	case AF_INET:
532 #endif
533 #if defined(INET) || defined(INET6)
534 		if (tp->tun_flags & TUN_PREPADDR) {
535 			/* Simple link-layer header */
536 			M_PREPEND(m0, dst->sa_len, M_DONTWAIT);
537 			if (m0 == NULL) {
538 				IF_DROP(&ifp->if_snd);
539 				error = ENOBUFS;
540 				goto out;
541 			}
542 			bcopy(dst, mtod(m0, char *), dst->sa_len);
543 		}
544 
545 		if (tp->tun_flags & TUN_IFHEAD) {
546 			/* Prepend the address family */
547 			M_PREPEND(m0, sizeof(*af), M_DONTWAIT);
548 			if (m0 == NULL) {
549 				IF_DROP(&ifp->if_snd);
550 				error = ENOBUFS;
551 				goto out;
552 			}
553 			af = mtod(m0,uint32_t *);
554 			*af = htonl(dst->sa_family);
555 		} else {
556 #ifdef INET
557 			if (dst->sa_family != AF_INET)
558 #endif
559 			{
560 				error = EAFNOSUPPORT;
561 				goto out;
562 			}
563 		}
564 		/* FALLTHROUGH */
565 	case AF_UNSPEC:
566 		IFQ_ENQUEUE(&ifp->if_snd, m0, error);
567 		if (error) {
568 			ifp->if_collisions++;
569 			error = EAFNOSUPPORT;
570 			m0 = NULL;
571 			goto out;
572 		}
573 		mlen = m0->m_pkthdr.len;
574 		ifp->if_opackets++;
575 		ifp->if_obytes += mlen;
576 		break;
577 #endif
578 	default:
579 		error = EAFNOSUPPORT;
580 		goto out;
581 	}
582 
583 	if (tp->tun_flags & TUN_RWAIT) {
584 		tp->tun_flags &= ~TUN_RWAIT;
585 		wakeup((void *)tp);
586 	}
587 	if (tp->tun_flags & TUN_ASYNC && tp->tun_pgid)
588 		softint_schedule(tp->tun_isih);
589 
590 	selnotify(&tp->tun_rsel, 0, 0);
591 out:
592 	mutex_exit(&tp->tun_lock);
593 	splx(s);
594 
595 	if (error && m0) {
596 		m_freem(m0);
597 	}
598 	return 0;
599 }
600 
601 static void
602 tun_i_softintr(void *cookie)
603 {
604 	struct tun_softc *tp = cookie;
605 
606 	if (tp->tun_flags & TUN_ASYNC && tp->tun_pgid)
607 		fownsignal(tp->tun_pgid, SIGIO, POLL_IN, POLLIN|POLLRDNORM,
608 		    NULL);
609 }
610 
611 static void
612 tun_o_softintr(void *cookie)
613 {
614 	struct tun_softc *tp = cookie;
615 
616 	if (tp->tun_flags & TUN_ASYNC && tp->tun_pgid)
617 		fownsignal(tp->tun_pgid, SIGIO, POLL_OUT, POLLOUT|POLLWRNORM,
618 		    NULL);
619 }
620 
621 /*
622  * the cdevsw interface is now pretty minimal.
623  */
624 int
625 tunioctl(dev_t dev, u_long cmd, void *data, int flag, struct lwp *l)
626 {
627 	struct tun_softc *tp;
628 	int s, error = 0;
629 
630 	s = splnet();
631 	tp = tun_find_unit(dev);
632 
633 	/* interface was "destroyed" already */
634 	if (tp == NULL) {
635 		error = ENXIO;
636 		goto out_nolock;
637 	}
638 
639 	switch (cmd) {
640 	case TUNSDEBUG:
641 		tundebug = *(int *)data;
642 		break;
643 
644 	case TUNGDEBUG:
645 		*(int *)data = tundebug;
646 		break;
647 
648 	case TUNSIFMODE:
649 		switch (*(int *)data & (IFF_POINTOPOINT|IFF_BROADCAST)) {
650 		case IFF_POINTOPOINT:
651 		case IFF_BROADCAST:
652 			if (tp->tun_if.if_flags & IFF_UP) {
653 				error = EBUSY;
654 				goto out;
655 			}
656 			tp->tun_if.if_flags &=
657 				~(IFF_BROADCAST|IFF_POINTOPOINT|IFF_MULTICAST);
658 			tp->tun_if.if_flags |= *(int *)data;
659 			break;
660 		default:
661 			error = EINVAL;
662 			goto out;
663 		}
664 		break;
665 
666 	case TUNSLMODE:
667 		if (*(int *)data) {
668 			tp->tun_flags |= TUN_PREPADDR;
669 			tp->tun_flags &= ~TUN_IFHEAD;
670 		} else
671 			tp->tun_flags &= ~TUN_PREPADDR;
672 		break;
673 
674 	case TUNSIFHEAD:
675 		if (*(int *)data) {
676 			tp->tun_flags |= TUN_IFHEAD;
677 			tp->tun_flags &= ~TUN_PREPADDR;
678 		} else
679 			tp->tun_flags &= ~TUN_IFHEAD;
680 		break;
681 
682 	case TUNGIFHEAD:
683 		*(int *)data = (tp->tun_flags & TUN_IFHEAD);
684 		break;
685 
686 	case FIONBIO:
687 		if (*(int *)data)
688 			tp->tun_flags |= TUN_NBIO;
689 		else
690 			tp->tun_flags &= ~TUN_NBIO;
691 		break;
692 
693 	case FIOASYNC:
694 		if (*(int *)data)
695 			tp->tun_flags |= TUN_ASYNC;
696 		else
697 			tp->tun_flags &= ~TUN_ASYNC;
698 		break;
699 
700 	case FIONREAD:
701 		if (tp->tun_if.if_snd.ifq_head)
702 			*(int *)data = tp->tun_if.if_snd.ifq_head->m_pkthdr.len;
703 		else
704 			*(int *)data = 0;
705 		break;
706 
707 	case TIOCSPGRP:
708 	case FIOSETOWN:
709 		error = fsetown(&tp->tun_pgid, cmd, data);
710 		break;
711 
712 	case TIOCGPGRP:
713 	case FIOGETOWN:
714 		error = fgetown(tp->tun_pgid, cmd, data);
715 		break;
716 
717 	default:
718 		error = ENOTTY;
719 	}
720 
721 out:
722 	mutex_exit(&tp->tun_lock);
723 out_nolock:
724 	splx(s);
725 	return (error);
726 }
727 
728 /*
729  * The cdevsw read interface - reads a packet at a time, or at
730  * least as much of a packet as can be read.
731  */
732 int
733 tunread(dev_t dev, struct uio *uio, int ioflag)
734 {
735 	struct tun_softc *tp;
736 	struct ifnet	*ifp;
737 	struct mbuf	*m, *m0;
738 	int		error = 0, len, s, index;
739 
740 	s = splnet();
741 	tp = tun_find_unit(dev);
742 
743 	/* interface was "destroyed" already */
744 	if (tp == NULL) {
745 		error = ENXIO;
746 		goto out_nolock;
747 	}
748 
749 	index = tp->tun_if.if_index;
750 	ifp = &tp->tun_if;
751 
752 	TUNDEBUG ("%s: read\n", ifp->if_xname);
753 	if ((tp->tun_flags & TUN_READY) != TUN_READY) {
754 		TUNDEBUG ("%s: not ready 0%o\n", ifp->if_xname, tp->tun_flags);
755 		error = EHOSTDOWN;
756 		goto out;
757 	}
758 
759 	tp->tun_flags &= ~TUN_RWAIT;
760 
761 	do {
762 		IFQ_DEQUEUE(&ifp->if_snd, m0);
763 		if (m0 == 0) {
764 			if (tp->tun_flags & TUN_NBIO) {
765 				error = EWOULDBLOCK;
766 				goto out;
767 			}
768 			tp->tun_flags |= TUN_RWAIT;
769 			if (mtsleep((void *)tp, PZERO|PCATCH|PNORELOCK,
770 					"tunread", 0, &tp->tun_lock) != 0) {
771 				error = EINTR;
772 				goto out_nolock;
773 			} else {
774 				/*
775 				 * Maybe the interface was destroyed while
776 				 * we were sleeping, so let's ensure that
777 				 * we're looking at the same (valid) tun
778 				 * interface before looping.
779 				 */
780 				tp = tun_find_unit(dev);
781 				if (tp == NULL) {
782 					error = ENXIO;
783 					goto out_nolock;
784 				}
785 				if (tp->tun_if.if_index != index) {
786 					error = ENXIO;
787 					goto out;
788 				}
789 			}
790 		}
791 	} while (m0 == 0);
792 
793 	mutex_exit(&tp->tun_lock);
794 	splx(s);
795 
796 	/* Copy the mbuf chain */
797 	while (m0 && uio->uio_resid > 0 && error == 0) {
798 		len = min(uio->uio_resid, m0->m_len);
799 		if (len != 0)
800 			error = uiomove(mtod(m0, void *), len, uio);
801 		MFREE(m0, m);
802 		m0 = m;
803 	}
804 
805 	if (m0) {
806 		TUNDEBUG("Dropping mbuf\n");
807 		m_freem(m0);
808 	}
809 	if (error)
810 		ifp->if_ierrors++;
811 
812 	return (error);
813 
814 out:
815 	mutex_exit(&tp->tun_lock);
816 out_nolock:
817 	splx(s);
818 	return (error);
819 }
820 
821 /*
822  * the cdevsw write interface - an atomic write is a packet - or else!
823  */
824 int
825 tunwrite(dev_t dev, struct uio *uio, int ioflag)
826 {
827 	struct tun_softc *tp;
828 	struct ifnet	*ifp;
829 	struct mbuf	*top, **mp, *m;
830 	pktqueue_t	*pktq;
831 	struct sockaddr	dst;
832 	int		error = 0, s, tlen, mlen;
833 	uint32_t	family;
834 
835 	s = splnet();
836 	tp = tun_find_unit(dev);
837 
838 	/* interface was "destroyed" already */
839 	if (tp == NULL) {
840 		error = ENXIO;
841 		goto out_nolock;
842 	}
843 
844 	/* Unlock until we've got the data */
845 	mutex_exit(&tp->tun_lock);
846 	splx(s);
847 
848 	ifp = &tp->tun_if;
849 
850 	TUNDEBUG("%s: tunwrite\n", ifp->if_xname);
851 
852 	if (tp->tun_flags & TUN_PREPADDR) {
853 		if (uio->uio_resid < sizeof(dst)) {
854 			error = EIO;
855 			goto out0;
856 		}
857 		error = uiomove((void *)&dst, sizeof(dst), uio);
858 		if (dst.sa_len > sizeof(dst)) {
859 			/* Duh.. */
860 			char discard;
861 			int n = dst.sa_len - sizeof(dst);
862 			while (n--)
863 				if ((error = uiomove(&discard, 1, uio)) != 0) {
864 					goto out0;
865 				}
866 		}
867 	} else if (tp->tun_flags & TUN_IFHEAD) {
868 		if (uio->uio_resid < sizeof(family)){
869 			error = EIO;
870 			goto out0;
871 		}
872 		error = uiomove((void *)&family, sizeof(family), uio);
873 		dst.sa_family = ntohl(family);
874 	} else {
875 #ifdef INET
876 		dst.sa_family = AF_INET;
877 #endif
878 	}
879 
880 	if (uio->uio_resid > TUNMTU) {
881 		TUNDEBUG("%s: len=%lu!\n", ifp->if_xname,
882 		    (unsigned long)uio->uio_resid);
883 		error = EIO;
884 		goto out0;
885 	}
886 
887 	switch (dst.sa_family) {
888 #ifdef INET
889 	case AF_INET:
890 		pktq = ip_pktq;
891 		break;
892 #endif
893 #ifdef INET6
894 	case AF_INET6:
895 		pktq = ip6_pktq;
896 		break;
897 #endif
898 	default:
899 		error = EAFNOSUPPORT;
900 		goto out0;
901 	}
902 
903 	tlen = uio->uio_resid;
904 
905 	/* get a header mbuf */
906 	MGETHDR(m, M_DONTWAIT, MT_DATA);
907 	if (m == NULL) {
908 		error = ENOBUFS;
909 		goto out0;
910 	}
911 	mlen = MHLEN;
912 
913 	top = NULL;
914 	mp = &top;
915 	while (error == 0 && uio->uio_resid > 0) {
916 		m->m_len = min(mlen, uio->uio_resid);
917 		error = uiomove(mtod(m, void *), m->m_len, uio);
918 		*mp = m;
919 		mp = &m->m_next;
920 		if (error == 0 && uio->uio_resid > 0) {
921 			MGET(m, M_DONTWAIT, MT_DATA);
922 			if (m == NULL) {
923 				error = ENOBUFS;
924 				break;
925 			}
926 			mlen = MLEN;
927 		}
928 	}
929 	if (error) {
930 		if (top != NULL)
931 			m_freem (top);
932 		ifp->if_ierrors++;
933 		goto out0;
934 	}
935 
936 	top->m_pkthdr.len = tlen;
937 	m_set_rcvif(top, ifp);
938 
939 	bpf_mtap_af(ifp, dst.sa_family, top);
940 
941 	s = splnet();
942 	mutex_enter(&tp->tun_lock);
943 	if ((tp->tun_flags & TUN_INITED) == 0) {
944 		/* Interface was destroyed */
945 		error = ENXIO;
946 		goto out;
947 	}
948 	if (__predict_false(!pktq_enqueue(pktq, top, 0))) {
949 		ifp->if_collisions++;
950 		mutex_exit(&tp->tun_lock);
951 		error = ENOBUFS;
952 		m_freem(top);
953 		goto out_nolock;
954 	}
955 	ifp->if_ipackets++;
956 	ifp->if_ibytes += tlen;
957 out:
958 	mutex_exit(&tp->tun_lock);
959 out_nolock:
960 	splx(s);
961 out0:
962 	return (error);
963 }
964 
965 #ifdef ALTQ
966 /*
967  * Start packet transmission on the interface.
968  * when the interface queue is rate-limited by ALTQ or TBR,
969  * if_start is needed to drain packets from the queue in order
970  * to notify readers when outgoing packets become ready.
971  *
972  * Should be called at splnet.
973  */
974 static void
975 tunstart(struct ifnet *ifp)
976 {
977 	struct tun_softc *tp = ifp->if_softc;
978 
979 	if (!ALTQ_IS_ENABLED(&ifp->if_snd) && !TBR_IS_ENABLED(&ifp->if_snd))
980 		return;
981 
982 	mutex_enter(&tp->tun_lock);
983 	if (!IF_IS_EMPTY(&ifp->if_snd)) {
984 		if (tp->tun_flags & TUN_RWAIT) {
985 			tp->tun_flags &= ~TUN_RWAIT;
986 			wakeup((void *)tp);
987 		}
988 		if (tp->tun_flags & TUN_ASYNC && tp->tun_pgid)
989 			softint_schedule(tp->tun_osih);
990 
991 		selnotify(&tp->tun_rsel, 0, 0);
992 	}
993 	mutex_exit(&tp->tun_lock);
994 }
995 #endif /* ALTQ */
996 /*
997  * tunpoll - the poll interface, this is only useful on reads
998  * really. The write detect always returns true, write never blocks
999  * anyway, it either accepts the packet or drops it.
1000  */
1001 int
1002 tunpoll(dev_t dev, int events, struct lwp *l)
1003 {
1004 	struct tun_softc *tp;
1005 	struct ifnet	*ifp;
1006 	int		s, revents = 0;
1007 
1008 	s = splnet();
1009 	tp = tun_find_unit(dev);
1010 
1011 	/* interface was "destroyed" already */
1012 	if (tp == NULL)
1013 		goto out_nolock;
1014 
1015 	ifp = &tp->tun_if;
1016 
1017 	TUNDEBUG("%s: tunpoll\n", ifp->if_xname);
1018 
1019 	if (events & (POLLIN | POLLRDNORM)) {
1020 		if (!IFQ_IS_EMPTY(&ifp->if_snd)) {
1021 			TUNDEBUG("%s: tunpoll q=%d\n", ifp->if_xname,
1022 			    ifp->if_snd.ifq_len);
1023 			revents |= events & (POLLIN | POLLRDNORM);
1024 		} else {
1025 			TUNDEBUG("%s: tunpoll waiting\n", ifp->if_xname);
1026 			selrecord(l, &tp->tun_rsel);
1027 		}
1028 	}
1029 
1030 	if (events & (POLLOUT | POLLWRNORM))
1031 		revents |= events & (POLLOUT | POLLWRNORM);
1032 
1033 	mutex_exit(&tp->tun_lock);
1034 out_nolock:
1035 	splx(s);
1036 	return (revents);
1037 }
1038 
1039 static void
1040 filt_tunrdetach(struct knote *kn)
1041 {
1042 	struct tun_softc *tp = kn->kn_hook;
1043 	int s;
1044 
1045 	s = splnet();
1046 	SLIST_REMOVE(&tp->tun_rsel.sel_klist, kn, knote, kn_selnext);
1047 	splx(s);
1048 }
1049 
1050 static int
1051 filt_tunread(struct knote *kn, long hint)
1052 {
1053 	struct tun_softc *tp = kn->kn_hook;
1054 	struct ifnet *ifp = &tp->tun_if;
1055 	struct mbuf *m;
1056 	int s;
1057 
1058 	s = splnet();
1059 	IF_POLL(&ifp->if_snd, m);
1060 	if (m == NULL) {
1061 		splx(s);
1062 		return (0);
1063 	}
1064 
1065 	for (kn->kn_data = 0; m != NULL; m = m->m_next)
1066 		kn->kn_data += m->m_len;
1067 
1068 	splx(s);
1069 	return (1);
1070 }
1071 
1072 static const struct filterops tunread_filtops =
1073 	{ 1, NULL, filt_tunrdetach, filt_tunread };
1074 
1075 static const struct filterops tun_seltrue_filtops =
1076 	{ 1, NULL, filt_tunrdetach, filt_seltrue };
1077 
1078 int
1079 tunkqfilter(dev_t dev, struct knote *kn)
1080 {
1081 	struct tun_softc *tp;
1082 	struct klist *klist;
1083 	int rv = 0, s;
1084 
1085 	s = splnet();
1086 	tp = tun_find_unit(dev);
1087 	if (tp == NULL)
1088 		goto out_nolock;
1089 
1090 	switch (kn->kn_filter) {
1091 	case EVFILT_READ:
1092 		klist = &tp->tun_rsel.sel_klist;
1093 		kn->kn_fop = &tunread_filtops;
1094 		break;
1095 
1096 	case EVFILT_WRITE:
1097 		klist = &tp->tun_rsel.sel_klist;
1098 		kn->kn_fop = &tun_seltrue_filtops;
1099 		break;
1100 
1101 	default:
1102 		rv = EINVAL;
1103 		goto out;
1104 	}
1105 
1106 	kn->kn_hook = tp;
1107 
1108 	SLIST_INSERT_HEAD(klist, kn, kn_selnext);
1109 
1110 out:
1111 	mutex_exit(&tp->tun_lock);
1112 out_nolock:
1113 	splx(s);
1114 	return (rv);
1115 }
1116