xref: /freebsd/sys/net/if_tuntap.c (revision daceb336172a6b0572de864b97e70b28451ca636)
1 /*	$NetBSD: if_tun.c,v 1.14 1994/06/29 06:36:25 cgd Exp $	*/
2 /*-
3  * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4  *
5  * Copyright (C) 1999-2000 by Maksim Yevmenkin <m_evmenkin@yahoo.com>
6  * All rights reserved.
7  * Copyright (c) 2019 Kyle Evans <kevans@FreeBSD.org>
8  *
9  * Redistribution and use in source and binary forms, with or without
10  * modification, are permitted provided that the following conditions
11  * are met:
12  *
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  *
19  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
20  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
21  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
22  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
23  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
24  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
25  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
26  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
27  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
28  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
29  * SUCH DAMAGE.
30  *
31  * BASED ON:
32  * -------------------------------------------------------------------------
33  *
34  * Copyright (c) 1988, Julian Onions <jpo@cs.nott.ac.uk>
35  * Nottingham University 1987.
36  *
37  * This source may be freely distributed, however I would be interested
38  * in any changes that are made.
39  *
40  * This driver takes packets off the IP i/f and hands them up to a
41  * user process to have its wicked way with. This driver has it's
42  * roots in a similar driver written by Phil Cockcroft (formerly) at
43  * UCL. This driver is based much more on read/write/poll mode of
44  * operation though.
45  *
46  * $FreeBSD$
47  */
48 
49 #include "opt_inet.h"
50 #include "opt_inet6.h"
51 
52 #include <sys/param.h>
53 #include <sys/lock.h>
54 #include <sys/priv.h>
55 #include <sys/proc.h>
56 #include <sys/systm.h>
57 #include <sys/jail.h>
58 #include <sys/mbuf.h>
59 #include <sys/module.h>
60 #include <sys/socket.h>
61 #include <sys/fcntl.h>
62 #include <sys/filio.h>
63 #include <sys/sockio.h>
64 #include <sys/sx.h>
65 #include <sys/ttycom.h>
66 #include <sys/poll.h>
67 #include <sys/selinfo.h>
68 #include <sys/signalvar.h>
69 #include <sys/filedesc.h>
70 #include <sys/kernel.h>
71 #include <sys/sysctl.h>
72 #include <sys/conf.h>
73 #include <sys/uio.h>
74 #include <sys/malloc.h>
75 #include <sys/random.h>
76 #include <sys/ctype.h>
77 
78 #include <net/ethernet.h>
79 #include <net/if.h>
80 #include <net/if_var.h>
81 #include <net/if_clone.h>
82 #include <net/if_dl.h>
83 #include <net/if_media.h>
84 #include <net/if_types.h>
85 #include <net/netisr.h>
86 #include <net/route.h>
87 #include <net/vnet.h>
88 #ifdef INET
89 #include <netinet/in.h>
90 #endif
91 #include <net/bpf.h>
92 #include <net/if_tap.h>
93 #include <net/if_tun.h>
94 
95 #include <sys/queue.h>
96 #include <sys/condvar.h>
97 #include <security/mac/mac_framework.h>
98 
99 struct tuntap_driver;
100 
101 /*
102  * tun_list is protected by global tunmtx.  Other mutable fields are
103  * protected by tun->tun_mtx, or by their owning subsystem.  tun_dev is
104  * static for the duration of a tunnel interface.
105  */
106 struct tuntap_softc {
107 	TAILQ_ENTRY(tuntap_softc)	tun_list;
108 	struct cdev *tun_dev;
109 	u_short	tun_flags;		/* misc flags */
110 #define	TUN_OPEN	0x0001
111 #define	TUN_INITED	0x0002
112 #define	TUN_RCOLL	0x0004
113 #define	TUN_IASET	0x0008
114 #define	TUN_DSTADDR	0x0010
115 #define	TUN_LMODE	0x0020
116 #define	TUN_RWAIT	0x0040
117 #define	TUN_ASYNC	0x0080
118 #define	TUN_IFHEAD	0x0100
119 #define	TUN_DYING	0x0200
120 #define	TUN_L2		0x0400
121 #define	TUN_VMNET	0x0800
122 
123 #define TUN_READY       (TUN_OPEN | TUN_INITED)
124 
125 	pid_t	tun_pid;		/* owning pid */
126 	struct	ifnet *tun_ifp;		/* the interface */
127 	struct  sigio *tun_sigio;	/* information for async I/O */
128 	struct  tuntap_driver *tun_drv;	/* appropriate driver */
129 	struct	selinfo	tun_rsel;	/* read select */
130 	struct mtx	tun_mtx;	/* protect mutable softc fields */
131 	struct cv	tun_cv;		/* protect against ref'd dev destroy */
132 	struct ether_addr	tun_ether;	/* remote address */
133 };
134 #define TUN2IFP(sc)	((sc)->tun_ifp)
135 
136 #define TUNDEBUG	if (tundebug) if_printf
137 
138 #define	TUN_LOCK(tp)	mtx_lock(&(tp)->tun_mtx)
139 #define	TUN_UNLOCK(tp)	mtx_unlock(&(tp)->tun_mtx)
140 
141 #define	TUN_VMIO_FLAG_MASK	0x0fff
142 
143 /*
144  * All mutable global variables in if_tun are locked using tunmtx, with
145  * the exception of tundebug, which is used unlocked, and the drivers' *clones,
146  * which are static after setup.
147  */
148 static struct mtx tunmtx;
149 static eventhandler_tag tag;
150 static const char tunname[] = "tun";
151 static const char tapname[] = "tap";
152 static const char vmnetname[] = "vmnet";
153 static MALLOC_DEFINE(M_TUN, tunname, "Tunnel Interface");
154 static int tundebug = 0;
155 static int tundclone = 1;
156 static int tap_allow_uopen = 0;        /* allow user open() */
157 static int tapuponopen = 0;    /* IFF_UP on open() */
158 static int tapdclone = 1;	/* enable devfs cloning */
159 
160 static TAILQ_HEAD(,tuntap_softc)	tunhead = TAILQ_HEAD_INITIALIZER(tunhead);
161 SYSCTL_INT(_debug, OID_AUTO, if_tun_debug, CTLFLAG_RW, &tundebug, 0, "");
162 
163 static struct sx tun_ioctl_sx;
164 SX_SYSINIT(tun_ioctl_sx, &tun_ioctl_sx, "tun_ioctl");
165 
166 SYSCTL_DECL(_net_link);
167 /* tun */
168 static SYSCTL_NODE(_net_link, OID_AUTO, tun, CTLFLAG_RW, 0,
169     "IP tunnel software network interface.");
170 SYSCTL_INT(_net_link_tun, OID_AUTO, devfs_cloning, CTLFLAG_RWTUN, &tundclone, 0,
171     "Enable legacy devfs interface creation.");
172 
173 /* tap */
174 static SYSCTL_NODE(_net_link, OID_AUTO, tap, CTLFLAG_RW, 0,
175     "Ethernet tunnel software network interface");
176 SYSCTL_INT(_net_link_tap, OID_AUTO, user_open, CTLFLAG_RW, &tap_allow_uopen, 0,
177 	"Allow user to open /dev/tap (based on node permissions)");
178 SYSCTL_INT(_net_link_tap, OID_AUTO, up_on_open, CTLFLAG_RW, &tapuponopen, 0,
179 	"Bring interface up when /dev/tap is opened");
180 SYSCTL_INT(_net_link_tap, OID_AUTO, devfs_cloning, CTLFLAG_RWTUN, &tapdclone, 0,
181 	"Enable legacy devfs interface creation");
182 SYSCTL_INT(_net_link_tap, OID_AUTO, debug, CTLFLAG_RW, &tundebug, 0, "");
183 
184 static int	tuntap_name2info(const char *name, int *unit, int *flags);
185 static void	tunclone(void *arg, struct ucred *cred, char *name,
186 		    int namelen, struct cdev **dev);
187 static void	tuncreate(struct cdev *dev, struct tuntap_driver *);
188 static int	tunifioctl(struct ifnet *, u_long, caddr_t);
189 static void	tuninit(struct ifnet *);
190 static void	tunifinit(void *xtp);
191 static int	tuntapmodevent(module_t, int, void *);
192 static int	tunoutput(struct ifnet *, struct mbuf *,
193 		    const struct sockaddr *, struct route *ro);
194 static void	tunstart(struct ifnet *);
195 static void	tunstart_l2(struct ifnet *);
196 
197 static int	tun_clone_match(struct if_clone *ifc, const char *name);
198 static int	tap_clone_match(struct if_clone *ifc, const char *name);
199 static int	vmnet_clone_match(struct if_clone *ifc, const char *name);
200 static int	tun_clone_create(struct if_clone *, char *, size_t, caddr_t);
201 static int	tun_clone_destroy(struct if_clone *, struct ifnet *);
202 
203 static d_open_t		tunopen;
204 static d_close_t	tunclose;
205 static d_read_t		tunread;
206 static d_write_t	tunwrite;
207 static d_ioctl_t	tunioctl;
208 static d_poll_t		tunpoll;
209 static d_kqfilter_t	tunkqfilter;
210 
211 static int		tunkqread(struct knote *, long);
212 static int		tunkqwrite(struct knote *, long);
213 static void		tunkqdetach(struct knote *);
214 
215 static struct filterops tun_read_filterops = {
216 	.f_isfd =	1,
217 	.f_attach =	NULL,
218 	.f_detach =	tunkqdetach,
219 	.f_event =	tunkqread,
220 };
221 
222 static struct filterops tun_write_filterops = {
223 	.f_isfd =	1,
224 	.f_attach =	NULL,
225 	.f_detach =	tunkqdetach,
226 	.f_event =	tunkqwrite,
227 };
228 
229 #define	TUN_DRIVER_IDENT_MASK	(TUN_L2 | TUN_VMNET)
230 
231 static struct tuntap_driver {
232 	int			 tun_flags;
233 	struct unrhdr		*unrhdr;
234 	struct cdevsw		 cdevsw;
235 	struct clonedevs	*clones;
236 	ifc_match_t		*clone_match_fn;
237 	ifc_create_t		*clone_create_fn;
238 	ifc_destroy_t		*clone_destroy_fn;
239 } tuntap_drivers[] = {
240 	{
241 		.tun_flags =	0,
242 		.cdevsw =	{
243 		    .d_version =	D_VERSION,
244 		    .d_flags =		D_NEEDMINOR,
245 		    .d_open =		tunopen,
246 		    .d_close =		tunclose,
247 		    .d_read =		tunread,
248 		    .d_write =		tunwrite,
249 		    .d_ioctl =		tunioctl,
250 		    .d_poll =		tunpoll,
251 		    .d_kqfilter =	tunkqfilter,
252 		    .d_name =		tunname,
253 		},
254 		.clone_match_fn =	tun_clone_match,
255 		.clone_create_fn =	tun_clone_create,
256 		.clone_destroy_fn =	tun_clone_destroy,
257 	},
258 	{
259 		.tun_flags =	TUN_L2,
260 		.cdevsw =	{
261 		    .d_version =	D_VERSION,
262 		    .d_flags =		D_NEEDMINOR,
263 		    .d_open =		tunopen,
264 		    .d_close =		tunclose,
265 		    .d_read =		tunread,
266 		    .d_write =		tunwrite,
267 		    .d_ioctl =		tunioctl,
268 		    .d_poll =		tunpoll,
269 		    .d_kqfilter =	tunkqfilter,
270 		    .d_name =		tapname,
271 		},
272 		.clone_match_fn =	tap_clone_match,
273 		.clone_create_fn =	tun_clone_create,
274 		.clone_destroy_fn =	tun_clone_destroy,
275 	},
276 	{
277 		.tun_flags =	TUN_L2 | TUN_VMNET,
278 		.cdevsw =	{
279 		    .d_version =	D_VERSION,
280 		    .d_flags =		D_NEEDMINOR,
281 		    .d_open =		tunopen,
282 		    .d_close =		tunclose,
283 		    .d_read =		tunread,
284 		    .d_write =		tunwrite,
285 		    .d_ioctl =		tunioctl,
286 		    .d_poll =		tunpoll,
287 		    .d_kqfilter =	tunkqfilter,
288 		    .d_name =		vmnetname,
289 		},
290 		.clone_match_fn =	vmnet_clone_match,
291 		.clone_create_fn =	tun_clone_create,
292 		.clone_destroy_fn =	tun_clone_destroy,
293 	},
294 };
295 
296 struct tuntap_driver_cloner {
297 	SLIST_ENTRY(tuntap_driver_cloner)		 link;
298 	struct tuntap_driver			*drv;
299 	struct if_clone				*cloner;
300 };
301 
302 VNET_DEFINE_STATIC(SLIST_HEAD(, tuntap_driver_cloner), tuntap_driver_cloners) =
303     SLIST_HEAD_INITIALIZER(tuntap_driver_cloners);
304 
305 #define	V_tuntap_driver_cloners	VNET(tuntap_driver_cloners)
306 
307 /*
308  * Sets unit and/or flags given the device name.  Must be called with correct
309  * vnet context.
310  */
311 static int
312 tuntap_name2info(const char *name, int *outunit, int *outflags)
313 {
314 	struct tuntap_driver *drv;
315 	struct tuntap_driver_cloner *drvc;
316 	char *dname;
317 	int flags, unit;
318 	bool found;
319 
320 	if (name == NULL)
321 		return (EINVAL);
322 
323 	/*
324 	 * Needed for dev_stdclone, but dev_stdclone will not modify, it just
325 	 * wants to be able to pass back a char * through the second param. We
326 	 * will always set that as NULL here, so we'll fake it.
327 	 */
328 	dname = __DECONST(char *, name);
329 	found = false;
330 
331 	KASSERT(!SLIST_EMPTY(&V_tuntap_driver_cloners),
332 	    ("tuntap_driver_cloners failed to initialize"));
333 	SLIST_FOREACH(drvc, &V_tuntap_driver_cloners, link) {
334 		KASSERT(drvc->drv != NULL,
335 		    ("tuntap_driver_cloners entry not properly initialized"));
336 		drv = drvc->drv;
337 
338 		if (strcmp(name, drv->cdevsw.d_name) == 0) {
339 			found = true;
340 			unit = -1;
341 			flags = drv->tun_flags;
342 			break;
343 		}
344 
345 		if (dev_stdclone(dname, NULL, drv->cdevsw.d_name, &unit) == 1) {
346 			found = true;
347 			flags = drv->tun_flags;
348 			break;
349 		}
350 	}
351 
352 	if (!found)
353 		return (ENXIO);
354 
355 	if (outunit != NULL)
356 		*outunit = unit;
357 	if (outflags != NULL)
358 		*outflags = flags;
359 	return (0);
360 }
361 
362 /*
363  * Get driver information from a set of flags specified.  Masks the identifying
364  * part of the flags and compares it against all of the available
365  * tuntap_drivers. Must be called with correct vnet context.
366  */
367 static struct tuntap_driver *
368 tuntap_driver_from_flags(int tun_flags)
369 {
370 	struct tuntap_driver *drv;
371 	struct tuntap_driver_cloner *drvc;
372 
373 	KASSERT(!SLIST_EMPTY(&V_tuntap_driver_cloners),
374 	    ("tuntap_driver_cloners failed to initialize"));
375 	SLIST_FOREACH(drvc, &V_tuntap_driver_cloners, link) {
376 		KASSERT(drvc->drv != NULL,
377 		    ("tuntap_driver_cloners entry not properly initialized"));
378 		drv = drvc->drv;
379 		if ((tun_flags & TUN_DRIVER_IDENT_MASK) == drv->tun_flags)
380 			return (drv);
381 	}
382 
383 	return (NULL);
384 }
385 
386 
387 
388 static int
389 tun_clone_match(struct if_clone *ifc, const char *name)
390 {
391 	int tunflags;
392 
393 	if (tuntap_name2info(name, NULL, &tunflags) == 0) {
394 		if ((tunflags & TUN_L2) == 0)
395 			return (1);
396 	}
397 
398 	return (0);
399 }
400 
401 static int
402 tap_clone_match(struct if_clone *ifc, const char *name)
403 {
404 	int tunflags;
405 
406 	if (tuntap_name2info(name, NULL, &tunflags) == 0) {
407 		if ((tunflags & (TUN_L2 | TUN_VMNET)) == TUN_L2)
408 			return (1);
409 	}
410 
411 	return (0);
412 }
413 
414 static int
415 vmnet_clone_match(struct if_clone *ifc, const char *name)
416 {
417 	int tunflags;
418 
419 	if (tuntap_name2info(name, NULL, &tunflags) == 0) {
420 		if ((tunflags & TUN_VMNET) != 0)
421 			return (1);
422 	}
423 
424 	return (0);
425 }
426 
427 static int
428 tun_clone_create(struct if_clone *ifc, char *name, size_t len, caddr_t params)
429 {
430 	struct tuntap_driver *drv;
431 	struct cdev *dev;
432 	int err, i, tunflags, unit;
433 
434 	tunflags = 0;
435 	/* The name here tells us exactly what we're creating */
436 	err = tuntap_name2info(name, &unit, &tunflags);
437 	if (err != 0)
438 		return (err);
439 
440 	drv = tuntap_driver_from_flags(tunflags);
441 	if (drv == NULL)
442 		return (ENXIO);
443 
444 	if (unit != -1) {
445 		/* If this unit number is still available that/s okay. */
446 		if (alloc_unr_specific(drv->unrhdr, unit) == -1)
447 			return (EEXIST);
448 	} else {
449 		unit = alloc_unr(drv->unrhdr);
450 	}
451 
452 	snprintf(name, IFNAMSIZ, "%s%d", drv->cdevsw.d_name, unit);
453 
454 	/* find any existing device, or allocate new unit number */
455 	i = clone_create(&drv->clones, &drv->cdevsw, &unit, &dev, 0);
456 	if (i) {
457 		/* No preexisting struct cdev *, create one */
458 		dev = make_dev(&drv->cdevsw, unit, UID_UUCP, GID_DIALER, 0600,
459 		    "%s%d", drv->cdevsw.d_name, unit);
460 	}
461 
462 	tuncreate(dev, drv);
463 
464 	return (0);
465 }
466 
467 static void
468 tunclone(void *arg, struct ucred *cred, char *name, int namelen,
469     struct cdev **dev)
470 {
471 	char devname[SPECNAMELEN + 1];
472 	struct tuntap_driver *drv;
473 	int append_unit, i, u, tunflags;
474 	bool mayclone;
475 
476 	if (*dev != NULL)
477 		return;
478 
479 	tunflags = 0;
480 	CURVNET_SET(CRED_TO_VNET(cred));
481 	if (tuntap_name2info(name, &u, &tunflags) != 0)
482 		goto out;	/* Not recognized */
483 
484 	if (u != -1 && u > IF_MAXUNIT)
485 		goto out;	/* Unit number too high */
486 
487 	mayclone = priv_check_cred(cred, PRIV_NET_IFCREATE) == 0;
488 	if ((tunflags & TUN_L2) != 0) {
489 		/* tap/vmnet allow user open with a sysctl */
490 		mayclone = (mayclone || tap_allow_uopen) && tapdclone;
491 	} else {
492 		mayclone = mayclone && tundclone;
493 	}
494 
495 	/*
496 	 * If tun cloning is enabled, only the superuser can create an
497 	 * interface.
498 	 */
499 	if (!mayclone)
500 		goto out;
501 
502 	if (u == -1)
503 		append_unit = 1;
504 	else
505 		append_unit = 0;
506 
507 	drv = tuntap_driver_from_flags(tunflags);
508 	if (drv == NULL)
509 		goto out;
510 
511 	/* find any existing device, or allocate new unit number */
512 	i = clone_create(&drv->clones, &drv->cdevsw, &u, dev, 0);
513 	if (i) {
514 		if (append_unit) {
515 			namelen = snprintf(devname, sizeof(devname), "%s%d",
516 			    name, u);
517 			name = devname;
518 		}
519 		/* No preexisting struct cdev *, create one */
520 		*dev = make_dev_credf(MAKEDEV_REF, &drv->cdevsw, u, cred,
521 		    UID_UUCP, GID_DIALER, 0600, "%s", name);
522 	}
523 
524 	if_clone_create(name, namelen, NULL);
525 out:
526 	CURVNET_RESTORE();
527 }
528 
529 static void
530 tun_destroy(struct tuntap_softc *tp)
531 {
532 	struct cdev *dev;
533 
534 	TUN_LOCK(tp);
535 	tp->tun_flags |= TUN_DYING;
536 	if ((tp->tun_flags & TUN_OPEN) != 0)
537 		cv_wait_unlock(&tp->tun_cv, &tp->tun_mtx);
538 	else
539 		TUN_UNLOCK(tp);
540 
541 	CURVNET_SET(TUN2IFP(tp)->if_vnet);
542 	sx_xlock(&tun_ioctl_sx);
543 	TUN2IFP(tp)->if_softc = NULL;
544 	sx_xunlock(&tun_ioctl_sx);
545 
546 	dev = tp->tun_dev;
547 	bpfdetach(TUN2IFP(tp));
548 	if_detach(TUN2IFP(tp));
549 	free_unr(tp->tun_drv->unrhdr, TUN2IFP(tp)->if_dunit);
550 	if_free(TUN2IFP(tp));
551 	destroy_dev(dev);
552 	seldrain(&tp->tun_rsel);
553 	knlist_clear(&tp->tun_rsel.si_note, 0);
554 	knlist_destroy(&tp->tun_rsel.si_note);
555 	mtx_destroy(&tp->tun_mtx);
556 	cv_destroy(&tp->tun_cv);
557 	free(tp, M_TUN);
558 	CURVNET_RESTORE();
559 }
560 
561 static int
562 tun_clone_destroy(struct if_clone *ifc __unused, struct ifnet *ifp)
563 {
564 	struct tuntap_softc *tp = ifp->if_softc;
565 
566 	mtx_lock(&tunmtx);
567 	TAILQ_REMOVE(&tunhead, tp, tun_list);
568 	mtx_unlock(&tunmtx);
569 	tun_destroy(tp);
570 
571 	return (0);
572 }
573 
574 static void
575 vnet_tun_init(const void *unused __unused)
576 {
577 	struct tuntap_driver *drv;
578 	struct tuntap_driver_cloner *drvc;
579 	int i;
580 
581 	for (i = 0; i < nitems(tuntap_drivers); ++i) {
582 		drv = &tuntap_drivers[i];
583 		drvc = malloc(sizeof(*drvc), M_TUN, M_WAITOK | M_ZERO);
584 
585 		drvc->drv = drv;
586 		drvc->cloner = if_clone_advanced(drv->cdevsw.d_name, 0,
587 		    drv->clone_match_fn, drv->clone_create_fn,
588 		    drv->clone_destroy_fn);
589 		SLIST_INSERT_HEAD(&V_tuntap_driver_cloners, drvc, link);
590 	};
591 }
592 VNET_SYSINIT(vnet_tun_init, SI_SUB_PROTO_IF, SI_ORDER_ANY,
593 		vnet_tun_init, NULL);
594 
595 static void
596 vnet_tun_uninit(const void *unused __unused)
597 {
598 	struct tuntap_driver_cloner *drvc;
599 
600 	while (!SLIST_EMPTY(&V_tuntap_driver_cloners)) {
601 		drvc = SLIST_FIRST(&V_tuntap_driver_cloners);
602 		SLIST_REMOVE_HEAD(&V_tuntap_driver_cloners, link);
603 
604 		if_clone_detach(drvc->cloner);
605 		free(drvc, M_TUN);
606 	}
607 }
608 VNET_SYSUNINIT(vnet_tun_uninit, SI_SUB_PROTO_IF, SI_ORDER_ANY,
609     vnet_tun_uninit, NULL);
610 
611 static void
612 tun_uninit(const void *unused __unused)
613 {
614 	struct tuntap_driver *drv;
615 	struct tuntap_softc *tp;
616 	int i;
617 
618 	EVENTHANDLER_DEREGISTER(dev_clone, tag);
619 	drain_dev_clone_events();
620 
621 	mtx_lock(&tunmtx);
622 	while ((tp = TAILQ_FIRST(&tunhead)) != NULL) {
623 		TAILQ_REMOVE(&tunhead, tp, tun_list);
624 		mtx_unlock(&tunmtx);
625 		tun_destroy(tp);
626 		mtx_lock(&tunmtx);
627 	}
628 	mtx_unlock(&tunmtx);
629 	for (i = 0; i < nitems(tuntap_drivers); ++i) {
630 		drv = &tuntap_drivers[i];
631 		delete_unrhdr(drv->unrhdr);
632 		clone_cleanup(&drv->clones);
633 	}
634 	mtx_destroy(&tunmtx);
635 }
636 SYSUNINIT(tun_uninit, SI_SUB_PROTO_IF, SI_ORDER_ANY, tun_uninit, NULL);
637 
638 static int
639 tuntapmodevent(module_t mod, int type, void *data)
640 {
641 	struct tuntap_driver *drv;
642 	int i;
643 
644 	switch (type) {
645 	case MOD_LOAD:
646 		mtx_init(&tunmtx, "tunmtx", NULL, MTX_DEF);
647 		for (i = 0; i < nitems(tuntap_drivers); ++i) {
648 			drv = &tuntap_drivers[i];
649 			clone_setup(&drv->clones);
650 			drv->unrhdr = new_unrhdr(0, IF_MAXUNIT, &tunmtx);
651 		}
652 		tag = EVENTHANDLER_REGISTER(dev_clone, tunclone, 0, 1000);
653 		if (tag == NULL)
654 			return (ENOMEM);
655 		break;
656 	case MOD_UNLOAD:
657 		/* See tun_uninit, so it's done after the vnet_sysuninit() */
658 		break;
659 	default:
660 		return EOPNOTSUPP;
661 	}
662 	return 0;
663 }
664 
665 static moduledata_t tuntap_mod = {
666 	"if_tuntap",
667 	tuntapmodevent,
668 	0
669 };
670 
671 DECLARE_MODULE(if_tuntap, tuntap_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
672 MODULE_VERSION(if_tuntap, 1);
673 
674 static void
675 tunstart(struct ifnet *ifp)
676 {
677 	struct tuntap_softc *tp = ifp->if_softc;
678 	struct mbuf *m;
679 
680 	TUNDEBUG(ifp, "starting\n");
681 	if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
682 		IFQ_LOCK(&ifp->if_snd);
683 		IFQ_POLL_NOLOCK(&ifp->if_snd, m);
684 		if (m == NULL) {
685 			IFQ_UNLOCK(&ifp->if_snd);
686 			return;
687 		}
688 		IFQ_UNLOCK(&ifp->if_snd);
689 	}
690 
691 	TUN_LOCK(tp);
692 	if (tp->tun_flags & TUN_RWAIT) {
693 		tp->tun_flags &= ~TUN_RWAIT;
694 		wakeup(tp);
695 	}
696 	selwakeuppri(&tp->tun_rsel, PZERO + 1);
697 	KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
698 	if (tp->tun_flags & TUN_ASYNC && tp->tun_sigio) {
699 		TUN_UNLOCK(tp);
700 		pgsigio(&tp->tun_sigio, SIGIO, 0);
701 	} else
702 		TUN_UNLOCK(tp);
703 }
704 
705 /*
706  * tunstart_l2
707  *
708  * queue packets from higher level ready to put out
709  */
710 static void
711 tunstart_l2(struct ifnet *ifp)
712 {
713 	struct tuntap_softc	*tp = ifp->if_softc;
714 
715 	TUNDEBUG(ifp, "starting\n");
716 
717 	/*
718 	 * do not junk pending output if we are in VMnet mode.
719 	 * XXX: can this do any harm because of queue overflow?
720 	 */
721 
722 	TUN_LOCK(tp);
723 	if (((tp->tun_flags & TUN_VMNET) == 0) &&
724 	    ((tp->tun_flags & TUN_READY) != TUN_READY)) {
725 		struct mbuf *m;
726 
727 		/* Unlocked read. */
728 		TUNDEBUG(ifp, "not ready, tun_flags = 0x%x\n", tp->tun_flags);
729 
730 		for (;;) {
731 			IF_DEQUEUE(&ifp->if_snd, m);
732 			if (m != NULL) {
733 				m_freem(m);
734 				if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
735 			} else
736 				break;
737 		}
738 		TUN_UNLOCK(tp);
739 
740 		return;
741 	}
742 
743 	ifp->if_drv_flags |= IFF_DRV_OACTIVE;
744 
745 	if (!IFQ_IS_EMPTY(&ifp->if_snd)) {
746 		if (tp->tun_flags & TUN_RWAIT) {
747 			tp->tun_flags &= ~TUN_RWAIT;
748 			wakeup(tp);
749 		}
750 
751 		if ((tp->tun_flags & TUN_ASYNC) && (tp->tun_sigio != NULL)) {
752 			TUN_UNLOCK(tp);
753 			pgsigio(&tp->tun_sigio, SIGIO, 0);
754 			TUN_LOCK(tp);
755 		}
756 
757 		selwakeuppri(&tp->tun_rsel, PZERO+1);
758 		KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
759 		if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); /* obytes are counted in ether_output */
760 	}
761 
762 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
763 	TUN_UNLOCK(tp);
764 } /* tunstart_l2 */
765 
766 
767 /* XXX: should return an error code so it can fail. */
768 static void
769 tuncreate(struct cdev *dev, struct tuntap_driver *drv)
770 {
771 	struct tuntap_softc *sc;
772 	struct ifnet *ifp;
773 	struct ether_addr eaddr;
774 	int iflags;
775 	u_char type;
776 
777 	sc = malloc(sizeof(*sc), M_TUN, M_WAITOK | M_ZERO);
778 	mtx_init(&sc->tun_mtx, "tun_mtx", NULL, MTX_DEF);
779 	cv_init(&sc->tun_cv, "tun_condvar");
780 	sc->tun_flags = drv->tun_flags;
781 	sc->tun_dev = dev;
782 	sc->tun_drv = drv;
783 	mtx_lock(&tunmtx);
784 	TAILQ_INSERT_TAIL(&tunhead, sc, tun_list);
785 	mtx_unlock(&tunmtx);
786 
787 	iflags = IFF_MULTICAST;
788 	if ((sc->tun_flags & TUN_L2) != 0) {
789 		type = IFT_ETHER;
790 		iflags |= IFF_BROADCAST | IFF_SIMPLEX;
791 	} else {
792 		type = IFT_PPP;
793 		iflags |= IFF_POINTOPOINT;
794 	}
795 	ifp = sc->tun_ifp = if_alloc(type);
796 	if (ifp == NULL)
797 		panic("%s%d: failed to if_alloc() interface.\n",
798 		    drv->cdevsw.d_name, dev2unit(dev));
799 	ifp->if_softc = sc;
800 	if_initname(ifp, drv->cdevsw.d_name, dev2unit(dev));
801 	ifp->if_ioctl = tunifioctl;
802 	ifp->if_flags = iflags;
803 	IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen);
804 	knlist_init_mtx(&sc->tun_rsel.si_note, &sc->tun_mtx);
805 	ifp->if_capabilities |= IFCAP_LINKSTATE;
806 	ifp->if_capenable |= IFCAP_LINKSTATE;
807 
808 	if ((sc->tun_flags & TUN_L2) != 0) {
809 		ifp->if_mtu = ETHERMTU;
810 		ifp->if_init = tunifinit;
811 		ifp->if_start = tunstart_l2;
812 
813 		ether_gen_addr(ifp, &eaddr);
814 		ether_ifattach(ifp, eaddr.octet);
815 	} else {
816 		ifp->if_mtu = TUNMTU;
817 		ifp->if_start = tunstart;
818 		ifp->if_output = tunoutput;
819 
820 		ifp->if_snd.ifq_drv_maxlen = 0;
821 		IFQ_SET_READY(&ifp->if_snd);
822 
823 		if_attach(ifp);
824 		bpfattach(ifp, DLT_NULL, sizeof(u_int32_t));
825 	}
826 	dev->si_drv1 = sc;
827 
828 	TUN_LOCK(sc);
829 	sc->tun_flags |= TUN_INITED;
830 	TUN_UNLOCK(sc);
831 
832 	TUNDEBUG(ifp, "interface %s is created, minor = %#x\n",
833 	    ifp->if_xname, dev2unit(dev));
834 }
835 
836 static int
837 tunopen(struct cdev *dev, int flag, int mode, struct thread *td)
838 {
839 	struct ifnet	*ifp;
840 	struct tuntap_driver *drv;
841 	struct tuntap_softc *tp;
842 	int error, tunflags;
843 
844 	tunflags = 0;
845 	CURVNET_SET(TD_TO_VNET(td));
846 	error = tuntap_name2info(dev->si_name, NULL, &tunflags);
847 	if (error != 0) {
848 		CURVNET_RESTORE();
849 		return (error);	/* Shouldn't happen */
850 	}
851 
852 	if ((tunflags & TUN_L2) != 0) {
853 		/* Restrict? */
854 		if (tap_allow_uopen == 0) {
855 			error = priv_check(td, PRIV_NET_TAP);
856 			if (error != 0) {
857 				CURVNET_RESTORE();
858 				return (error);
859 			}
860 		}
861 	}
862 
863 	/*
864 	 * XXXRW: Non-atomic test and set of dev->si_drv1 requires
865 	 * synchronization.
866 	 */
867 	tp = dev->si_drv1;
868 	if (!tp) {
869 		drv = tuntap_driver_from_flags(tunflags);
870 		if (drv == NULL) {
871 			CURVNET_RESTORE();
872 			return (ENXIO);
873 		}
874 		tuncreate(dev, drv);
875 		tp = dev->si_drv1;
876 	}
877 
878 	TUN_LOCK(tp);
879 	if ((tp->tun_flags & (TUN_OPEN | TUN_DYING)) != 0) {
880 		TUN_UNLOCK(tp);
881 		CURVNET_RESTORE();
882 		return (EBUSY);
883 	}
884 
885 	ifp = TUN2IFP(tp);
886 
887 	if ((tp->tun_flags & TUN_L2) != 0) {
888 		bcopy(IF_LLADDR(ifp), tp->tun_ether.octet,
889 		    sizeof(tp->tun_ether.octet));
890 
891 		ifp->if_drv_flags |= IFF_DRV_RUNNING;
892 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
893 
894 		if (tapuponopen)
895 			ifp->if_flags |= IFF_UP;
896 	}
897 
898 	tp->tun_pid = td->td_proc->p_pid;
899 	tp->tun_flags |= TUN_OPEN;
900 
901 	if_link_state_change(ifp, LINK_STATE_UP);
902 	TUNDEBUG(ifp, "open\n");
903 	TUN_UNLOCK(tp);
904 	CURVNET_RESTORE();
905 	return (0);
906 }
907 
908 /*
909  * tunclose - close the device - mark i/f down & delete
910  * routing info
911  */
912 static	int
913 tunclose(struct cdev *dev, int foo, int bar, struct thread *td)
914 {
915 	struct tuntap_softc *tp;
916 	struct ifnet *ifp;
917 	bool l2tun;
918 
919 	tp = dev->si_drv1;
920 	ifp = TUN2IFP(tp);
921 
922 	TUN_LOCK(tp);
923 	/*
924 	 * Simply close the device if this isn't the controlling process.  This
925 	 * may happen if, for instance, the tunnel has been handed off to
926 	 * another process.  The original controller should be able to close it
927 	 * without putting us into an inconsistent state.
928 	 */
929 	if (td->td_proc->p_pid != tp->tun_pid) {
930 		TUN_UNLOCK(tp);
931 		return (0);
932 	}
933 
934 	/*
935 	 * junk all pending output
936 	 */
937 	CURVNET_SET(ifp->if_vnet);
938 
939 	l2tun = false;
940 	if ((tp->tun_flags & TUN_L2) != 0) {
941 		l2tun = true;
942 		IF_DRAIN(&ifp->if_snd);
943 	} else {
944 		IFQ_PURGE(&ifp->if_snd);
945 	}
946 
947 	/* For vmnet, we won't do most of the address/route bits */
948 	if ((tp->tun_flags & TUN_VMNET) != 0)
949 		goto out;
950 
951 	if (ifp->if_flags & IFF_UP) {
952 		TUN_UNLOCK(tp);
953 		if_down(ifp);
954 		TUN_LOCK(tp);
955 	}
956 
957 	/* Delete all addresses and routes which reference this interface. */
958 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
959 		struct ifaddr *ifa;
960 
961 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
962 		TUN_UNLOCK(tp);
963 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
964 			/* deal w/IPv4 PtP destination; unlocked read */
965 			if (!l2tun && ifa->ifa_addr->sa_family == AF_INET) {
966 				rtinit(ifa, (int)RTM_DELETE,
967 				    tp->tun_flags & TUN_DSTADDR ? RTF_HOST : 0);
968 			} else {
969 				rtinit(ifa, (int)RTM_DELETE, 0);
970 			}
971 		}
972 		if_purgeaddrs(ifp);
973 		TUN_LOCK(tp);
974 	}
975 
976 out:
977 	if_link_state_change(ifp, LINK_STATE_DOWN);
978 	CURVNET_RESTORE();
979 
980 	funsetown(&tp->tun_sigio);
981 	selwakeuppri(&tp->tun_rsel, PZERO + 1);
982 	KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
983 	TUNDEBUG (ifp, "closed\n");
984 	tp->tun_flags &= ~TUN_OPEN;
985 	tp->tun_pid = 0;
986 
987 	cv_broadcast(&tp->tun_cv);
988 	TUN_UNLOCK(tp);
989 	return (0);
990 }
991 
992 static void
993 tuninit(struct ifnet *ifp)
994 {
995 	struct tuntap_softc *tp = ifp->if_softc;
996 #ifdef INET
997 	struct ifaddr *ifa;
998 #endif
999 
1000 	TUNDEBUG(ifp, "tuninit\n");
1001 
1002 	TUN_LOCK(tp);
1003 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1004 	if ((tp->tun_flags & TUN_L2) == 0) {
1005 		ifp->if_flags |= IFF_UP;
1006 		getmicrotime(&ifp->if_lastchange);
1007 #ifdef INET
1008 		if_addr_rlock(ifp);
1009 		CK_STAILQ_FOREACH(ifa, &ifp->if_addrhead, ifa_link) {
1010 			if (ifa->ifa_addr->sa_family == AF_INET) {
1011 				struct sockaddr_in *si;
1012 
1013 				si = (struct sockaddr_in *)ifa->ifa_addr;
1014 				if (si->sin_addr.s_addr)
1015 					tp->tun_flags |= TUN_IASET;
1016 
1017 				si = (struct sockaddr_in *)ifa->ifa_dstaddr;
1018 				if (si && si->sin_addr.s_addr)
1019 					tp->tun_flags |= TUN_DSTADDR;
1020 			}
1021 		}
1022 		if_addr_runlock(ifp);
1023 #endif
1024 		TUN_UNLOCK(tp);
1025 	} else {
1026 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1027 		TUN_UNLOCK(tp);
1028 		/* attempt to start output */
1029 		tunstart_l2(ifp);
1030 	}
1031 
1032 }
1033 
1034 /*
1035  * Used only for l2 tunnel.
1036  */
1037 static void
1038 tunifinit(void *xtp)
1039 {
1040 	struct tuntap_softc *tp;
1041 
1042 	tp = (struct tuntap_softc *)xtp;
1043 	tuninit(tp->tun_ifp);
1044 }
1045 
1046 /*
1047  * Process an ioctl request.
1048  */
1049 static int
1050 tunifioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1051 {
1052 	struct ifreq *ifr = (struct ifreq *)data;
1053 	struct tuntap_softc *tp;
1054 	struct ifstat *ifs;
1055 	struct ifmediareq	*ifmr;
1056 	int		dummy, error = 0;
1057 	bool		l2tun;
1058 
1059 	ifmr = NULL;
1060 	sx_xlock(&tun_ioctl_sx);
1061 	tp = ifp->if_softc;
1062 	if (tp == NULL) {
1063 		error = ENXIO;
1064 		goto bad;
1065 	}
1066 	l2tun = (tp->tun_flags & TUN_L2) != 0;
1067 	switch(cmd) {
1068 	case SIOCGIFSTATUS:
1069 		ifs = (struct ifstat *)data;
1070 		TUN_LOCK(tp);
1071 		if (tp->tun_pid)
1072 			snprintf(ifs->ascii, sizeof(ifs->ascii),
1073 			    "\tOpened by PID %d\n", tp->tun_pid);
1074 		else
1075 			ifs->ascii[0] = '\0';
1076 		TUN_UNLOCK(tp);
1077 		break;
1078 	case SIOCSIFADDR:
1079 		if (l2tun)
1080 			error = ether_ioctl(ifp, cmd, data);
1081 		else
1082 			tuninit(ifp);
1083 		if (error == 0)
1084 		    TUNDEBUG(ifp, "address set\n");
1085 		break;
1086 	case SIOCSIFMTU:
1087 		ifp->if_mtu = ifr->ifr_mtu;
1088 		TUNDEBUG(ifp, "mtu set\n");
1089 		break;
1090 	case SIOCSIFFLAGS:
1091 	case SIOCADDMULTI:
1092 	case SIOCDELMULTI:
1093 		break;
1094 	case SIOCGIFMEDIA:
1095 		if (!l2tun) {
1096 			error = EINVAL;
1097 			break;
1098 		}
1099 
1100 		ifmr = (struct ifmediareq *)data;
1101 		dummy = ifmr->ifm_count;
1102 		ifmr->ifm_count = 1;
1103 		ifmr->ifm_status = IFM_AVALID;
1104 		ifmr->ifm_active = IFM_ETHER;
1105 		if (tp->tun_flags & TUN_OPEN)
1106 			ifmr->ifm_status |= IFM_ACTIVE;
1107 		ifmr->ifm_current = ifmr->ifm_active;
1108 		if (dummy >= 1) {
1109 			int media = IFM_ETHER;
1110 			error = copyout(&media, ifmr->ifm_ulist, sizeof(int));
1111 		}
1112 		break;
1113 	default:
1114 		if (l2tun) {
1115 			error = ether_ioctl(ifp, cmd, data);
1116 		} else {
1117 			error = EINVAL;
1118 		}
1119 	}
1120 bad:
1121 	sx_xunlock(&tun_ioctl_sx);
1122 	return (error);
1123 }
1124 
1125 /*
1126  * tunoutput - queue packets from higher level ready to put out.
1127  */
1128 static int
1129 tunoutput(struct ifnet *ifp, struct mbuf *m0, const struct sockaddr *dst,
1130     struct route *ro)
1131 {
1132 	struct tuntap_softc *tp = ifp->if_softc;
1133 	u_short cached_tun_flags;
1134 	int error;
1135 	u_int32_t af;
1136 
1137 	TUNDEBUG (ifp, "tunoutput\n");
1138 
1139 #ifdef MAC
1140 	error = mac_ifnet_check_transmit(ifp, m0);
1141 	if (error) {
1142 		m_freem(m0);
1143 		return (error);
1144 	}
1145 #endif
1146 
1147 	/* Could be unlocked read? */
1148 	TUN_LOCK(tp);
1149 	cached_tun_flags = tp->tun_flags;
1150 	TUN_UNLOCK(tp);
1151 	if ((cached_tun_flags & TUN_READY) != TUN_READY) {
1152 		TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags);
1153 		m_freem (m0);
1154 		return (EHOSTDOWN);
1155 	}
1156 
1157 	if ((ifp->if_flags & IFF_UP) != IFF_UP) {
1158 		m_freem (m0);
1159 		return (EHOSTDOWN);
1160 	}
1161 
1162 	/* BPF writes need to be handled specially. */
1163 	if (dst->sa_family == AF_UNSPEC)
1164 		bcopy(dst->sa_data, &af, sizeof(af));
1165 	else
1166 		af = dst->sa_family;
1167 
1168 	if (bpf_peers_present(ifp->if_bpf))
1169 		bpf_mtap2(ifp->if_bpf, &af, sizeof(af), m0);
1170 
1171 	/* prepend sockaddr? this may abort if the mbuf allocation fails */
1172 	if (cached_tun_flags & TUN_LMODE) {
1173 		/* allocate space for sockaddr */
1174 		M_PREPEND(m0, dst->sa_len, M_NOWAIT);
1175 
1176 		/* if allocation failed drop packet */
1177 		if (m0 == NULL) {
1178 			if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1179 			if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1180 			return (ENOBUFS);
1181 		} else {
1182 			bcopy(dst, m0->m_data, dst->sa_len);
1183 		}
1184 	}
1185 
1186 	if (cached_tun_flags & TUN_IFHEAD) {
1187 		/* Prepend the address family */
1188 		M_PREPEND(m0, 4, M_NOWAIT);
1189 
1190 		/* if allocation failed drop packet */
1191 		if (m0 == NULL) {
1192 			if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1193 			if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1194 			return (ENOBUFS);
1195 		} else
1196 			*(u_int32_t *)m0->m_data = htonl(af);
1197 	} else {
1198 #ifdef INET
1199 		if (af != AF_INET)
1200 #endif
1201 		{
1202 			m_freem(m0);
1203 			return (EAFNOSUPPORT);
1204 		}
1205 	}
1206 
1207 	error = (ifp->if_transmit)(ifp, m0);
1208 	if (error)
1209 		return (ENOBUFS);
1210 	if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1211 	return (0);
1212 }
1213 
1214 /*
1215  * the cdevsw interface is now pretty minimal.
1216  */
1217 static	int
1218 tunioctl(struct cdev *dev, u_long cmd, caddr_t data, int flag,
1219     struct thread *td)
1220 {
1221 	struct ifreq ifr, *ifrp;
1222 	struct tuntap_softc *tp = dev->si_drv1;
1223 	struct tuninfo *tunp;
1224 	int error, iflags;
1225 #if defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD5) || \
1226     defined(COMPAT_FREEBSD4)
1227 	int	ival;
1228 #endif
1229 	bool	l2tun;
1230 
1231 	l2tun = (tp->tun_flags & TUN_L2) != 0;
1232 	if (l2tun) {
1233 		/* tap specific ioctls */
1234 		switch(cmd) {
1235 		case TAPGIFNAME:
1236 			ifrp = (struct ifreq *)data;
1237 			strlcpy(ifrp->ifr_name, TUN2IFP(tp)->if_xname,
1238 			    IFNAMSIZ);
1239 
1240 			return (0);
1241 		/* VMware/VMnet port ioctl's */
1242 #if defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD5) || \
1243     defined(COMPAT_FREEBSD4)
1244 		case _IO('V', 0):
1245 			ival = IOCPARM_IVAL(data);
1246 			data = (caddr_t)&ival;
1247 			/* FALLTHROUGH */
1248 #endif
1249 		case VMIO_SIOCSIFFLAGS: /* VMware/VMnet SIOCSIFFLAGS */
1250 			iflags = *(int *)data;
1251 			iflags &= TUN_VMIO_FLAG_MASK;
1252 			iflags &= ~IFF_CANTCHANGE;
1253 			iflags |= IFF_UP;
1254 
1255 			TUN_LOCK(tp);
1256 			TUN2IFP(tp)->if_flags = iflags |
1257 			    (TUN2IFP(tp)->if_flags & IFF_CANTCHANGE);
1258 			TUN_UNLOCK(tp);
1259 
1260 			return (0);
1261 		case SIOCGIFADDR:	/* get MAC address of the remote side */
1262 			TUN_LOCK(tp);
1263 			bcopy(&tp->tun_ether.octet, data,
1264 			    sizeof(tp->tun_ether.octet));
1265 			TUN_UNLOCK(tp);
1266 
1267 			return (0);
1268 		case SIOCSIFADDR:	/* set MAC address of the remote side */
1269 			TUN_LOCK(tp);
1270 			bcopy(data, &tp->tun_ether.octet,
1271 			    sizeof(tp->tun_ether.octet));
1272 			TUN_UNLOCK(tp);
1273 
1274 			return (0);
1275 		}
1276 
1277 		/* Fall through to the common ioctls if unhandled */
1278 	} else {
1279 		switch (cmd) {
1280 		case TUNSLMODE:
1281 			TUN_LOCK(tp);
1282 			if (*(int *)data) {
1283 				tp->tun_flags |= TUN_LMODE;
1284 				tp->tun_flags &= ~TUN_IFHEAD;
1285 			} else
1286 				tp->tun_flags &= ~TUN_LMODE;
1287 			TUN_UNLOCK(tp);
1288 
1289 			return (0);
1290 		case TUNSIFHEAD:
1291 			TUN_LOCK(tp);
1292 			if (*(int *)data) {
1293 				tp->tun_flags |= TUN_IFHEAD;
1294 				tp->tun_flags &= ~TUN_LMODE;
1295 			} else
1296 				tp->tun_flags &= ~TUN_IFHEAD;
1297 			TUN_UNLOCK(tp);
1298 
1299 			return (0);
1300 		case TUNGIFHEAD:
1301 			TUN_LOCK(tp);
1302 			*(int *)data = (tp->tun_flags & TUN_IFHEAD) ? 1 : 0;
1303 			TUN_UNLOCK(tp);
1304 
1305 			return (0);
1306 		case TUNSIFMODE:
1307 			/* deny this if UP */
1308 			if (TUN2IFP(tp)->if_flags & IFF_UP)
1309 				return (EBUSY);
1310 
1311 			switch (*(int *)data & ~IFF_MULTICAST) {
1312 			case IFF_POINTOPOINT:
1313 			case IFF_BROADCAST:
1314 				TUN_LOCK(tp);
1315 				TUN2IFP(tp)->if_flags &=
1316 				    ~(IFF_BROADCAST|IFF_POINTOPOINT|IFF_MULTICAST);
1317 				TUN2IFP(tp)->if_flags |= *(int *)data;
1318 				TUN_UNLOCK(tp);
1319 
1320 				break;
1321 			default:
1322 				return (EINVAL);
1323 			}
1324 
1325 			return (0);
1326 		case TUNSIFPID:
1327 			TUN_LOCK(tp);
1328 			tp->tun_pid = curthread->td_proc->p_pid;
1329 			TUN_UNLOCK(tp);
1330 
1331 			return (0);
1332 		}
1333 		/* Fall through to the common ioctls if unhandled */
1334 	}
1335 
1336 	switch (cmd) {
1337 	case TUNSIFINFO:
1338 		tunp = (struct tuninfo *)data;
1339 		if (TUN2IFP(tp)->if_type != tunp->type)
1340 			return (EPROTOTYPE);
1341 		TUN_LOCK(tp);
1342 		if (TUN2IFP(tp)->if_mtu != tunp->mtu) {
1343 			strlcpy(ifr.ifr_name, if_name(TUN2IFP(tp)), IFNAMSIZ);
1344 			ifr.ifr_mtu = tunp->mtu;
1345 			CURVNET_SET(TUN2IFP(tp)->if_vnet);
1346 			error = ifhwioctl(SIOCSIFMTU, TUN2IFP(tp),
1347 			    (caddr_t)&ifr, td);
1348 			CURVNET_RESTORE();
1349 			if (error) {
1350 				TUN_UNLOCK(tp);
1351 				return (error);
1352 			}
1353 		}
1354 		TUN2IFP(tp)->if_baudrate = tunp->baudrate;
1355 		TUN_UNLOCK(tp);
1356 		break;
1357 	case TUNGIFINFO:
1358 		tunp = (struct tuninfo *)data;
1359 		TUN_LOCK(tp);
1360 		tunp->mtu = TUN2IFP(tp)->if_mtu;
1361 		tunp->type = TUN2IFP(tp)->if_type;
1362 		tunp->baudrate = TUN2IFP(tp)->if_baudrate;
1363 		TUN_UNLOCK(tp);
1364 		break;
1365 	case TUNSDEBUG:
1366 		tundebug = *(int *)data;
1367 		break;
1368 	case TUNGDEBUG:
1369 		*(int *)data = tundebug;
1370 		break;
1371 	case FIONBIO:
1372 		break;
1373 	case FIOASYNC:
1374 		TUN_LOCK(tp);
1375 		if (*(int *)data)
1376 			tp->tun_flags |= TUN_ASYNC;
1377 		else
1378 			tp->tun_flags &= ~TUN_ASYNC;
1379 		TUN_UNLOCK(tp);
1380 		break;
1381 	case FIONREAD:
1382 		if (!IFQ_IS_EMPTY(&TUN2IFP(tp)->if_snd)) {
1383 			struct mbuf *mb;
1384 			IFQ_LOCK(&TUN2IFP(tp)->if_snd);
1385 			IFQ_POLL_NOLOCK(&TUN2IFP(tp)->if_snd, mb);
1386 			for (*(int *)data = 0; mb != NULL; mb = mb->m_next)
1387 				*(int *)data += mb->m_len;
1388 			IFQ_UNLOCK(&TUN2IFP(tp)->if_snd);
1389 		} else
1390 			*(int *)data = 0;
1391 		break;
1392 	case FIOSETOWN:
1393 		return (fsetown(*(int *)data, &tp->tun_sigio));
1394 
1395 	case FIOGETOWN:
1396 		*(int *)data = fgetown(&tp->tun_sigio);
1397 		return (0);
1398 
1399 	/* This is deprecated, FIOSETOWN should be used instead. */
1400 	case TIOCSPGRP:
1401 		return (fsetown(-(*(int *)data), &tp->tun_sigio));
1402 
1403 	/* This is deprecated, FIOGETOWN should be used instead. */
1404 	case TIOCGPGRP:
1405 		*(int *)data = -fgetown(&tp->tun_sigio);
1406 		return (0);
1407 
1408 	default:
1409 		return (ENOTTY);
1410 	}
1411 	return (0);
1412 }
1413 
1414 /*
1415  * The cdevsw read interface - reads a packet at a time, or at
1416  * least as much of a packet as can be read.
1417  */
1418 static	int
1419 tunread(struct cdev *dev, struct uio *uio, int flag)
1420 {
1421 	struct tuntap_softc *tp = dev->si_drv1;
1422 	struct ifnet	*ifp = TUN2IFP(tp);
1423 	struct mbuf	*m;
1424 	int		error=0, len;
1425 
1426 	TUNDEBUG (ifp, "read\n");
1427 	TUN_LOCK(tp);
1428 	if ((tp->tun_flags & TUN_READY) != TUN_READY) {
1429 		TUN_UNLOCK(tp);
1430 		TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags);
1431 		return (EHOSTDOWN);
1432 	}
1433 
1434 	tp->tun_flags &= ~TUN_RWAIT;
1435 
1436 	do {
1437 		IFQ_DEQUEUE(&ifp->if_snd, m);
1438 		if (m == NULL) {
1439 			if (flag & O_NONBLOCK) {
1440 				TUN_UNLOCK(tp);
1441 				return (EWOULDBLOCK);
1442 			}
1443 			tp->tun_flags |= TUN_RWAIT;
1444 			error = mtx_sleep(tp, &tp->tun_mtx, PCATCH | (PZERO + 1),
1445 			    "tunread", 0);
1446 			if (error != 0) {
1447 				TUN_UNLOCK(tp);
1448 				return (error);
1449 			}
1450 		}
1451 	} while (m == NULL);
1452 	TUN_UNLOCK(tp);
1453 
1454 	if ((tp->tun_flags & TUN_L2) != 0)
1455 		BPF_MTAP(ifp, m);
1456 
1457 	while (m && uio->uio_resid > 0 && error == 0) {
1458 		len = min(uio->uio_resid, m->m_len);
1459 		if (len != 0)
1460 			error = uiomove(mtod(m, void *), len, uio);
1461 		m = m_free(m);
1462 	}
1463 
1464 	if (m) {
1465 		TUNDEBUG(ifp, "Dropping mbuf\n");
1466 		m_freem(m);
1467 	}
1468 	return (error);
1469 }
1470 
1471 static int
1472 tunwrite_l2(struct tuntap_softc *tp, struct mbuf *m)
1473 {
1474 	struct ether_header *eh;
1475 	struct ifnet *ifp;
1476 
1477 	ifp = TUN2IFP(tp);
1478 
1479 	/*
1480 	 * Only pass a unicast frame to ether_input(), if it would
1481 	 * actually have been received by non-virtual hardware.
1482 	 */
1483 	if (m->m_len < sizeof(struct ether_header)) {
1484 		m_freem(m);
1485 		return (0);
1486 	}
1487 
1488 	eh = mtod(m, struct ether_header *);
1489 
1490 	if (eh && (ifp->if_flags & IFF_PROMISC) == 0 &&
1491 	    !ETHER_IS_MULTICAST(eh->ether_dhost) &&
1492 	    bcmp(eh->ether_dhost, IF_LLADDR(ifp), ETHER_ADDR_LEN) != 0) {
1493 		m_freem(m);
1494 		return (0);
1495 	}
1496 
1497 	/* Pass packet up to parent. */
1498 	CURVNET_SET(ifp->if_vnet);
1499 	(*ifp->if_input)(ifp, m);
1500 	CURVNET_RESTORE();
1501 	/* ibytes are counted in parent */
1502 	if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1503 	return (0);
1504 }
1505 
1506 static int
1507 tunwrite_l3(struct tuntap_softc *tp, struct mbuf *m)
1508 {
1509 	struct ifnet *ifp;
1510 	int family, isr;
1511 
1512 	ifp = TUN2IFP(tp);
1513 	/* Could be unlocked read? */
1514 	TUN_LOCK(tp);
1515 	if (tp->tun_flags & TUN_IFHEAD) {
1516 		TUN_UNLOCK(tp);
1517 		if (m->m_len < sizeof(family) &&
1518 		(m = m_pullup(m, sizeof(family))) == NULL)
1519 			return (ENOBUFS);
1520 		family = ntohl(*mtod(m, u_int32_t *));
1521 		m_adj(m, sizeof(family));
1522 	} else {
1523 		TUN_UNLOCK(tp);
1524 		family = AF_INET;
1525 	}
1526 
1527 	BPF_MTAP2(ifp, &family, sizeof(family), m);
1528 
1529 	switch (family) {
1530 #ifdef INET
1531 	case AF_INET:
1532 		isr = NETISR_IP;
1533 		break;
1534 #endif
1535 #ifdef INET6
1536 	case AF_INET6:
1537 		isr = NETISR_IPV6;
1538 		break;
1539 #endif
1540 	default:
1541 		m_freem(m);
1542 		return (EAFNOSUPPORT);
1543 	}
1544 	random_harvest_queue(m, sizeof(*m), RANDOM_NET_TUN);
1545 	if_inc_counter(ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
1546 	if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1547 	CURVNET_SET(ifp->if_vnet);
1548 	M_SETFIB(m, ifp->if_fib);
1549 	netisr_dispatch(isr, m);
1550 	CURVNET_RESTORE();
1551 	return (0);
1552 }
1553 
1554 /*
1555  * the cdevsw write interface - an atomic write is a packet - or else!
1556  */
1557 static	int
1558 tunwrite(struct cdev *dev, struct uio *uio, int flag)
1559 {
1560 	struct tuntap_softc *tp;
1561 	struct ifnet	*ifp;
1562 	struct mbuf	*m;
1563 	uint32_t	mru;
1564 	int		align;
1565 	bool		l2tun;
1566 
1567 	tp = dev->si_drv1;
1568 	ifp = TUN2IFP(tp);
1569 	TUNDEBUG(ifp, "tunwrite\n");
1570 	if ((ifp->if_flags & IFF_UP) != IFF_UP)
1571 		/* ignore silently */
1572 		return (0);
1573 
1574 	if (uio->uio_resid == 0)
1575 		return (0);
1576 
1577 	l2tun = (tp->tun_flags & TUN_L2) != 0;
1578 	align = 0;
1579 	mru = l2tun ? TAPMRU : TUNMRU;
1580 	if (l2tun)
1581 		align = ETHER_ALIGN;
1582 	else if ((tp->tun_flags & TUN_IFHEAD) != 0)
1583 		mru += sizeof(uint32_t);	/* family */
1584 	if (uio->uio_resid < 0 || uio->uio_resid > mru) {
1585 		TUNDEBUG(ifp, "len=%zd!\n", uio->uio_resid);
1586 		return (EIO);
1587 	}
1588 
1589 	if ((m = m_uiotombuf(uio, M_NOWAIT, 0, align, M_PKTHDR)) == NULL) {
1590 		if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1591 		return (ENOBUFS);
1592 	}
1593 
1594 	m->m_pkthdr.rcvif = ifp;
1595 #ifdef MAC
1596 	mac_ifnet_create_mbuf(ifp, m);
1597 #endif
1598 
1599 	if (l2tun)
1600 		return (tunwrite_l2(tp, m));
1601 
1602 	return (tunwrite_l3(tp, m));
1603 }
1604 
1605 /*
1606  * tunpoll - the poll interface, this is only useful on reads
1607  * really. The write detect always returns true, write never blocks
1608  * anyway, it either accepts the packet or drops it.
1609  */
1610 static	int
1611 tunpoll(struct cdev *dev, int events, struct thread *td)
1612 {
1613 	struct tuntap_softc *tp = dev->si_drv1;
1614 	struct ifnet	*ifp = TUN2IFP(tp);
1615 	int		revents = 0;
1616 
1617 	TUNDEBUG(ifp, "tunpoll\n");
1618 
1619 	if (events & (POLLIN | POLLRDNORM)) {
1620 		IFQ_LOCK(&ifp->if_snd);
1621 		if (!IFQ_IS_EMPTY(&ifp->if_snd)) {
1622 			TUNDEBUG(ifp, "tunpoll q=%d\n", ifp->if_snd.ifq_len);
1623 			revents |= events & (POLLIN | POLLRDNORM);
1624 		} else {
1625 			TUNDEBUG(ifp, "tunpoll waiting\n");
1626 			selrecord(td, &tp->tun_rsel);
1627 		}
1628 		IFQ_UNLOCK(&ifp->if_snd);
1629 	}
1630 	if (events & (POLLOUT | POLLWRNORM))
1631 		revents |= events & (POLLOUT | POLLWRNORM);
1632 
1633 	return (revents);
1634 }
1635 
1636 /*
1637  * tunkqfilter - support for the kevent() system call.
1638  */
1639 static int
1640 tunkqfilter(struct cdev *dev, struct knote *kn)
1641 {
1642 	struct tuntap_softc	*tp = dev->si_drv1;
1643 	struct ifnet	*ifp = TUN2IFP(tp);
1644 
1645 	switch(kn->kn_filter) {
1646 	case EVFILT_READ:
1647 		TUNDEBUG(ifp, "%s kqfilter: EVFILT_READ, minor = %#x\n",
1648 		    ifp->if_xname, dev2unit(dev));
1649 		kn->kn_fop = &tun_read_filterops;
1650 		break;
1651 
1652 	case EVFILT_WRITE:
1653 		TUNDEBUG(ifp, "%s kqfilter: EVFILT_WRITE, minor = %#x\n",
1654 		    ifp->if_xname, dev2unit(dev));
1655 		kn->kn_fop = &tun_write_filterops;
1656 		break;
1657 
1658 	default:
1659 		TUNDEBUG(ifp, "%s kqfilter: invalid filter, minor = %#x\n",
1660 		    ifp->if_xname, dev2unit(dev));
1661 		return(EINVAL);
1662 	}
1663 
1664 	kn->kn_hook = tp;
1665 	knlist_add(&tp->tun_rsel.si_note, kn, 0);
1666 
1667 	return (0);
1668 }
1669 
1670 /*
1671  * Return true of there is data in the interface queue.
1672  */
1673 static int
1674 tunkqread(struct knote *kn, long hint)
1675 {
1676 	int			ret;
1677 	struct tuntap_softc	*tp = kn->kn_hook;
1678 	struct cdev		*dev = tp->tun_dev;
1679 	struct ifnet	*ifp = TUN2IFP(tp);
1680 
1681 	if ((kn->kn_data = ifp->if_snd.ifq_len) > 0) {
1682 		TUNDEBUG(ifp,
1683 		    "%s have data in the queue.  Len = %d, minor = %#x\n",
1684 		    ifp->if_xname, ifp->if_snd.ifq_len, dev2unit(dev));
1685 		ret = 1;
1686 	} else {
1687 		TUNDEBUG(ifp,
1688 		    "%s waiting for data, minor = %#x\n", ifp->if_xname,
1689 		    dev2unit(dev));
1690 		ret = 0;
1691 	}
1692 
1693 	return (ret);
1694 }
1695 
1696 /*
1697  * Always can write, always return MTU in kn->data.
1698  */
1699 static int
1700 tunkqwrite(struct knote *kn, long hint)
1701 {
1702 	struct tuntap_softc	*tp = kn->kn_hook;
1703 	struct ifnet	*ifp = TUN2IFP(tp);
1704 
1705 	kn->kn_data = ifp->if_mtu;
1706 
1707 	return (1);
1708 }
1709 
1710 static void
1711 tunkqdetach(struct knote *kn)
1712 {
1713 	struct tuntap_softc	*tp = kn->kn_hook;
1714 
1715 	knlist_remove(&tp->tun_rsel.si_note, kn, 0);
1716 }
1717