xref: /freebsd/sys/net/if_tuntap.c (revision 2008043f386721d58158e37e0d7e50df8095942d)
1 /*	$NetBSD: if_tun.c,v 1.14 1994/06/29 06:36:25 cgd Exp $	*/
2 /*-
3  * SPDX-License-Identifier: BSD-2-Clause
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 
47 #include "opt_inet.h"
48 #include "opt_inet6.h"
49 
50 #include <sys/param.h>
51 #include <sys/lock.h>
52 #include <sys/priv.h>
53 #include <sys/proc.h>
54 #include <sys/systm.h>
55 #include <sys/jail.h>
56 #include <sys/mbuf.h>
57 #include <sys/module.h>
58 #include <sys/socket.h>
59 #include <sys/eventhandler.h>
60 #include <sys/fcntl.h>
61 #include <sys/filio.h>
62 #include <sys/sockio.h>
63 #include <sys/sx.h>
64 #include <sys/syslog.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_private.h>
85 #include <net/if_types.h>
86 #include <net/if_vlan_var.h>
87 #include <net/netisr.h>
88 #include <net/route.h>
89 #include <net/vnet.h>
90 #include <netinet/in.h>
91 #ifdef INET
92 #include <netinet/ip.h>
93 #endif
94 #ifdef INET6
95 #include <netinet/ip6.h>
96 #include <netinet6/ip6_var.h>
97 #endif
98 #include <netinet/udp.h>
99 #include <netinet/tcp.h>
100 #include <net/bpf.h>
101 #include <net/if_tap.h>
102 #include <net/if_tun.h>
103 
104 #include <dev/virtio/network/virtio_net.h>
105 
106 #include <sys/queue.h>
107 #include <sys/condvar.h>
108 #include <security/mac/mac_framework.h>
109 
110 struct tuntap_driver;
111 
112 /*
113  * tun_list is protected by global tunmtx.  Other mutable fields are
114  * protected by tun->tun_mtx, or by their owning subsystem.  tun_dev is
115  * static for the duration of a tunnel interface.
116  */
117 struct tuntap_softc {
118 	TAILQ_ENTRY(tuntap_softc)	 tun_list;
119 	struct cdev			*tun_alias;
120 	struct cdev			*tun_dev;
121 	u_short				 tun_flags;	/* misc flags */
122 #define	TUN_OPEN	0x0001
123 #define	TUN_INITED	0x0002
124 #define	TUN_UNUSED1	0x0008
125 #define	TUN_UNUSED2	0x0010
126 #define	TUN_LMODE	0x0020
127 #define	TUN_RWAIT	0x0040
128 #define	TUN_ASYNC	0x0080
129 #define	TUN_IFHEAD	0x0100
130 #define	TUN_DYING	0x0200
131 #define	TUN_L2		0x0400
132 #define	TUN_VMNET	0x0800
133 
134 #define	TUN_DRIVER_IDENT_MASK	(TUN_L2 | TUN_VMNET)
135 #define	TUN_READY		(TUN_OPEN | TUN_INITED)
136 
137 	pid_t			 tun_pid;	/* owning pid */
138 	struct ifnet		*tun_ifp;	/* the interface */
139 	struct sigio		*tun_sigio;	/* async I/O info */
140 	struct tuntap_driver	*tun_drv;	/* appropriate driver */
141 	struct selinfo		 tun_rsel;	/* read select */
142 	struct mtx		 tun_mtx;	/* softc field mutex */
143 	struct cv		 tun_cv;	/* for ref'd dev destroy */
144 	struct ether_addr	 tun_ether;	/* remote address */
145 	int			 tun_busy;	/* busy count */
146 	int			 tun_vhdrlen;	/* virtio-net header length */
147 };
148 #define	TUN2IFP(sc)	((sc)->tun_ifp)
149 
150 #define	TUNDEBUG	if (tundebug) if_printf
151 
152 #define	TUN_LOCK(tp)		mtx_lock(&(tp)->tun_mtx)
153 #define	TUN_UNLOCK(tp)		mtx_unlock(&(tp)->tun_mtx)
154 #define	TUN_LOCK_ASSERT(tp)	mtx_assert(&(tp)->tun_mtx, MA_OWNED);
155 
156 #define	TUN_VMIO_FLAG_MASK	0x0fff
157 
158 /*
159  * Interface capabilities of a tap device that supports the virtio-net
160  * header.
161  */
162 #define TAP_VNET_HDR_CAPS	(IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6	\
163 				| IFCAP_VLAN_HWCSUM			\
164 				| IFCAP_TSO | IFCAP_LRO			\
165 				| IFCAP_VLAN_HWTSO)
166 
167 #define TAP_ALL_OFFLOAD		(CSUM_TSO | CSUM_TCP | CSUM_UDP |\
168 				    CSUM_TCP_IPV6 | CSUM_UDP_IPV6)
169 
170 /*
171  * All mutable global variables in if_tun are locked using tunmtx, with
172  * the exception of tundebug, which is used unlocked, and the drivers' *clones,
173  * which are static after setup.
174  */
175 static struct mtx tunmtx;
176 static eventhandler_tag arrival_tag;
177 static eventhandler_tag clone_tag;
178 static const char tunname[] = "tun";
179 static const char tapname[] = "tap";
180 static const char vmnetname[] = "vmnet";
181 static MALLOC_DEFINE(M_TUN, tunname, "Tunnel Interface");
182 static int tundebug = 0;
183 static int tundclone = 1;
184 static int tap_allow_uopen = 0;	/* allow user devfs cloning */
185 static int tapuponopen = 0;	/* IFF_UP on open() */
186 static int tapdclone = 1;	/* enable devfs cloning */
187 
188 static TAILQ_HEAD(,tuntap_softc)	tunhead = TAILQ_HEAD_INITIALIZER(tunhead);
189 SYSCTL_INT(_debug, OID_AUTO, if_tun_debug, CTLFLAG_RW, &tundebug, 0, "");
190 
191 static struct sx tun_ioctl_sx;
192 SX_SYSINIT(tun_ioctl_sx, &tun_ioctl_sx, "tun_ioctl");
193 
194 SYSCTL_DECL(_net_link);
195 /* tun */
196 static SYSCTL_NODE(_net_link, OID_AUTO, tun, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
197     "IP tunnel software network interface");
198 SYSCTL_INT(_net_link_tun, OID_AUTO, devfs_cloning, CTLFLAG_RWTUN, &tundclone, 0,
199     "Enable legacy devfs interface creation");
200 
201 /* tap */
202 static SYSCTL_NODE(_net_link, OID_AUTO, tap, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
203     "Ethernet tunnel software network interface");
204 SYSCTL_INT(_net_link_tap, OID_AUTO, user_open, CTLFLAG_RW, &tap_allow_uopen, 0,
205     "Enable legacy devfs interface creation for all users");
206 SYSCTL_INT(_net_link_tap, OID_AUTO, up_on_open, CTLFLAG_RW, &tapuponopen, 0,
207     "Bring interface up when /dev/tap is opened");
208 SYSCTL_INT(_net_link_tap, OID_AUTO, devfs_cloning, CTLFLAG_RWTUN, &tapdclone, 0,
209     "Enable legacy devfs interface creation");
210 SYSCTL_INT(_net_link_tap, OID_AUTO, debug, CTLFLAG_RW, &tundebug, 0, "");
211 
212 static int	tun_create_device(struct tuntap_driver *drv, int unit,
213     struct ucred *cr, struct cdev **dev, const char *name);
214 static int	tun_busy_locked(struct tuntap_softc *tp);
215 static void	tun_unbusy_locked(struct tuntap_softc *tp);
216 static int	tun_busy(struct tuntap_softc *tp);
217 static void	tun_unbusy(struct tuntap_softc *tp);
218 
219 static int	tuntap_name2info(const char *name, int *unit, int *flags);
220 static void	tunclone(void *arg, struct ucred *cred, char *name,
221 		    int namelen, struct cdev **dev);
222 static void	tuncreate(struct cdev *dev);
223 static void	tundtor(void *data);
224 static void	tunrename(void *arg, struct ifnet *ifp);
225 static int	tunifioctl(struct ifnet *, u_long, caddr_t);
226 static void	tuninit(struct ifnet *);
227 static void	tunifinit(void *xtp);
228 static int	tuntapmodevent(module_t, int, void *);
229 static int	tunoutput(struct ifnet *, struct mbuf *,
230 		    const struct sockaddr *, struct route *ro);
231 static void	tunstart(struct ifnet *);
232 static void	tunstart_l2(struct ifnet *);
233 
234 static int	tun_clone_match(struct if_clone *ifc, const char *name);
235 static int	tap_clone_match(struct if_clone *ifc, const char *name);
236 static int	vmnet_clone_match(struct if_clone *ifc, const char *name);
237 static int	tun_clone_create(struct if_clone *, char *, size_t,
238 		    struct ifc_data *, struct ifnet **);
239 static int	tun_clone_destroy(struct if_clone *, struct ifnet *, uint32_t);
240 static void	tun_vnethdr_set(struct ifnet *ifp, int vhdrlen);
241 
242 static d_open_t		tunopen;
243 static d_read_t		tunread;
244 static d_write_t	tunwrite;
245 static d_ioctl_t	tunioctl;
246 static d_poll_t		tunpoll;
247 static d_kqfilter_t	tunkqfilter;
248 
249 static int		tunkqread(struct knote *, long);
250 static int		tunkqwrite(struct knote *, long);
251 static void		tunkqdetach(struct knote *);
252 
253 static struct filterops tun_read_filterops = {
254 	.f_isfd =	1,
255 	.f_attach =	NULL,
256 	.f_detach =	tunkqdetach,
257 	.f_event =	tunkqread,
258 };
259 
260 static struct filterops tun_write_filterops = {
261 	.f_isfd =	1,
262 	.f_attach =	NULL,
263 	.f_detach =	tunkqdetach,
264 	.f_event =	tunkqwrite,
265 };
266 
267 static struct tuntap_driver {
268 	struct cdevsw		 cdevsw;
269 	int			 ident_flags;
270 	struct unrhdr		*unrhdr;
271 	struct clonedevs	*clones;
272 	ifc_match_f		*clone_match_fn;
273 	ifc_create_f		*clone_create_fn;
274 	ifc_destroy_f		*clone_destroy_fn;
275 } tuntap_drivers[] = {
276 	{
277 		.ident_flags =	0,
278 		.cdevsw =	{
279 		    .d_version =	D_VERSION,
280 		    .d_flags =		D_NEEDMINOR,
281 		    .d_open =		tunopen,
282 		    .d_read =		tunread,
283 		    .d_write =		tunwrite,
284 		    .d_ioctl =		tunioctl,
285 		    .d_poll =		tunpoll,
286 		    .d_kqfilter =	tunkqfilter,
287 		    .d_name =		tunname,
288 		},
289 		.clone_match_fn =	tun_clone_match,
290 		.clone_create_fn =	tun_clone_create,
291 		.clone_destroy_fn =	tun_clone_destroy,
292 	},
293 	{
294 		.ident_flags =	TUN_L2,
295 		.cdevsw =	{
296 		    .d_version =	D_VERSION,
297 		    .d_flags =		D_NEEDMINOR,
298 		    .d_open =		tunopen,
299 		    .d_read =		tunread,
300 		    .d_write =		tunwrite,
301 		    .d_ioctl =		tunioctl,
302 		    .d_poll =		tunpoll,
303 		    .d_kqfilter =	tunkqfilter,
304 		    .d_name =		tapname,
305 		},
306 		.clone_match_fn =	tap_clone_match,
307 		.clone_create_fn =	tun_clone_create,
308 		.clone_destroy_fn =	tun_clone_destroy,
309 	},
310 	{
311 		.ident_flags =	TUN_L2 | TUN_VMNET,
312 		.cdevsw =	{
313 		    .d_version =	D_VERSION,
314 		    .d_flags =		D_NEEDMINOR,
315 		    .d_open =		tunopen,
316 		    .d_read =		tunread,
317 		    .d_write =		tunwrite,
318 		    .d_ioctl =		tunioctl,
319 		    .d_poll =		tunpoll,
320 		    .d_kqfilter =	tunkqfilter,
321 		    .d_name =		vmnetname,
322 		},
323 		.clone_match_fn =	vmnet_clone_match,
324 		.clone_create_fn =	tun_clone_create,
325 		.clone_destroy_fn =	tun_clone_destroy,
326 	},
327 };
328 
329 struct tuntap_driver_cloner {
330 	SLIST_ENTRY(tuntap_driver_cloner)	 link;
331 	struct tuntap_driver			*drv;
332 	struct if_clone				*cloner;
333 };
334 
335 VNET_DEFINE_STATIC(SLIST_HEAD(, tuntap_driver_cloner), tuntap_driver_cloners) =
336     SLIST_HEAD_INITIALIZER(tuntap_driver_cloners);
337 
338 #define	V_tuntap_driver_cloners	VNET(tuntap_driver_cloners)
339 
340 /*
341  * Mechanism for marking a tunnel device as busy so that we can safely do some
342  * orthogonal operations (such as operations on devices) without racing against
343  * tun_destroy.  tun_destroy will wait on the condvar if we're at all busy or
344  * open, to be woken up when the condition is alleviated.
345  */
346 static int
347 tun_busy_locked(struct tuntap_softc *tp)
348 {
349 
350 	TUN_LOCK_ASSERT(tp);
351 	if ((tp->tun_flags & TUN_DYING) != 0) {
352 		/*
353 		 * Perhaps unintuitive, but the device is busy going away.
354 		 * Other interpretations of EBUSY from tun_busy make little
355 		 * sense, since making a busy device even more busy doesn't
356 		 * sound like a problem.
357 		 */
358 		return (EBUSY);
359 	}
360 
361 	++tp->tun_busy;
362 	return (0);
363 }
364 
365 static void
366 tun_unbusy_locked(struct tuntap_softc *tp)
367 {
368 
369 	TUN_LOCK_ASSERT(tp);
370 	KASSERT(tp->tun_busy != 0, ("tun_unbusy: called for non-busy tunnel"));
371 
372 	--tp->tun_busy;
373 	/* Wake up anything that may be waiting on our busy tunnel. */
374 	if (tp->tun_busy == 0)
375 		cv_broadcast(&tp->tun_cv);
376 }
377 
378 static int
379 tun_busy(struct tuntap_softc *tp)
380 {
381 	int ret;
382 
383 	TUN_LOCK(tp);
384 	ret = tun_busy_locked(tp);
385 	TUN_UNLOCK(tp);
386 	return (ret);
387 }
388 
389 static void
390 tun_unbusy(struct tuntap_softc *tp)
391 {
392 
393 	TUN_LOCK(tp);
394 	tun_unbusy_locked(tp);
395 	TUN_UNLOCK(tp);
396 }
397 
398 /*
399  * Sets unit and/or flags given the device name.  Must be called with correct
400  * vnet context.
401  */
402 static int
403 tuntap_name2info(const char *name, int *outunit, int *outflags)
404 {
405 	struct tuntap_driver *drv;
406 	struct tuntap_driver_cloner *drvc;
407 	char *dname;
408 	int flags, unit;
409 	bool found;
410 
411 	if (name == NULL)
412 		return (EINVAL);
413 
414 	/*
415 	 * Needed for dev_stdclone, but dev_stdclone will not modify, it just
416 	 * wants to be able to pass back a char * through the second param. We
417 	 * will always set that as NULL here, so we'll fake it.
418 	 */
419 	dname = __DECONST(char *, name);
420 	found = false;
421 
422 	KASSERT(!SLIST_EMPTY(&V_tuntap_driver_cloners),
423 	    ("tuntap_driver_cloners failed to initialize"));
424 	SLIST_FOREACH(drvc, &V_tuntap_driver_cloners, link) {
425 		KASSERT(drvc->drv != NULL,
426 		    ("tuntap_driver_cloners entry not properly initialized"));
427 		drv = drvc->drv;
428 
429 		if (strcmp(name, drv->cdevsw.d_name) == 0) {
430 			found = true;
431 			unit = -1;
432 			flags = drv->ident_flags;
433 			break;
434 		}
435 
436 		if (dev_stdclone(dname, NULL, drv->cdevsw.d_name, &unit) == 1) {
437 			found = true;
438 			flags = drv->ident_flags;
439 			break;
440 		}
441 	}
442 
443 	if (!found)
444 		return (ENXIO);
445 
446 	if (outunit != NULL)
447 		*outunit = unit;
448 	if (outflags != NULL)
449 		*outflags = flags;
450 	return (0);
451 }
452 
453 /*
454  * Get driver information from a set of flags specified.  Masks the identifying
455  * part of the flags and compares it against all of the available
456  * tuntap_drivers. Must be called with correct vnet context.
457  */
458 static struct tuntap_driver *
459 tuntap_driver_from_flags(int tun_flags)
460 {
461 	struct tuntap_driver *drv;
462 	struct tuntap_driver_cloner *drvc;
463 
464 	KASSERT(!SLIST_EMPTY(&V_tuntap_driver_cloners),
465 	    ("tuntap_driver_cloners failed to initialize"));
466 	SLIST_FOREACH(drvc, &V_tuntap_driver_cloners, link) {
467 		KASSERT(drvc->drv != NULL,
468 		    ("tuntap_driver_cloners entry not properly initialized"));
469 		drv = drvc->drv;
470 		if ((tun_flags & TUN_DRIVER_IDENT_MASK) == drv->ident_flags)
471 			return (drv);
472 	}
473 
474 	return (NULL);
475 }
476 
477 static int
478 tun_clone_match(struct if_clone *ifc, const char *name)
479 {
480 	int tunflags;
481 
482 	if (tuntap_name2info(name, NULL, &tunflags) == 0) {
483 		if ((tunflags & TUN_L2) == 0)
484 			return (1);
485 	}
486 
487 	return (0);
488 }
489 
490 static int
491 tap_clone_match(struct if_clone *ifc, const char *name)
492 {
493 	int tunflags;
494 
495 	if (tuntap_name2info(name, NULL, &tunflags) == 0) {
496 		if ((tunflags & (TUN_L2 | TUN_VMNET)) == TUN_L2)
497 			return (1);
498 	}
499 
500 	return (0);
501 }
502 
503 static int
504 vmnet_clone_match(struct if_clone *ifc, const char *name)
505 {
506 	int tunflags;
507 
508 	if (tuntap_name2info(name, NULL, &tunflags) == 0) {
509 		if ((tunflags & TUN_VMNET) != 0)
510 			return (1);
511 	}
512 
513 	return (0);
514 }
515 
516 static int
517 tun_clone_create(struct if_clone *ifc, char *name, size_t len,
518     struct ifc_data *ifd, struct ifnet **ifpp)
519 {
520 	struct tuntap_driver *drv;
521 	struct cdev *dev;
522 	int err, i, tunflags, unit;
523 
524 	tunflags = 0;
525 	/* The name here tells us exactly what we're creating */
526 	err = tuntap_name2info(name, &unit, &tunflags);
527 	if (err != 0)
528 		return (err);
529 
530 	drv = tuntap_driver_from_flags(tunflags);
531 	if (drv == NULL)
532 		return (ENXIO);
533 
534 	if (unit != -1) {
535 		/* If this unit number is still available that's okay. */
536 		if (alloc_unr_specific(drv->unrhdr, unit) == -1)
537 			return (EEXIST);
538 	} else {
539 		unit = alloc_unr(drv->unrhdr);
540 	}
541 
542 	snprintf(name, IFNAMSIZ, "%s%d", drv->cdevsw.d_name, unit);
543 
544 	/* find any existing device, or allocate new unit number */
545 	dev = NULL;
546 	i = clone_create(&drv->clones, &drv->cdevsw, &unit, &dev, 0);
547 	/* No preexisting struct cdev *, create one */
548 	if (i != 0)
549 		i = tun_create_device(drv, unit, NULL, &dev, name);
550 	if (i == 0) {
551 		dev_ref(dev);
552 		tuncreate(dev);
553 		struct tuntap_softc *tp = dev->si_drv1;
554 		*ifpp = tp->tun_ifp;
555 	}
556 
557 	return (i);
558 }
559 
560 static void
561 tunclone(void *arg, struct ucred *cred, char *name, int namelen,
562     struct cdev **dev)
563 {
564 	char devname[SPECNAMELEN + 1];
565 	struct tuntap_driver *drv;
566 	int append_unit, i, u, tunflags;
567 	bool mayclone;
568 
569 	if (*dev != NULL)
570 		return;
571 
572 	tunflags = 0;
573 	CURVNET_SET(CRED_TO_VNET(cred));
574 	if (tuntap_name2info(name, &u, &tunflags) != 0)
575 		goto out;	/* Not recognized */
576 
577 	if (u != -1 && u > IF_MAXUNIT)
578 		goto out;	/* Unit number too high */
579 
580 	mayclone = priv_check_cred(cred, PRIV_NET_IFCREATE) == 0;
581 	if ((tunflags & TUN_L2) != 0) {
582 		/* tap/vmnet allow user open with a sysctl */
583 		mayclone = (mayclone || tap_allow_uopen) && tapdclone;
584 	} else {
585 		mayclone = mayclone && tundclone;
586 	}
587 
588 	/*
589 	 * If tun cloning is enabled, only the superuser can create an
590 	 * interface.
591 	 */
592 	if (!mayclone)
593 		goto out;
594 
595 	if (u == -1)
596 		append_unit = 1;
597 	else
598 		append_unit = 0;
599 
600 	drv = tuntap_driver_from_flags(tunflags);
601 	if (drv == NULL)
602 		goto out;
603 
604 	/* find any existing device, or allocate new unit number */
605 	i = clone_create(&drv->clones, &drv->cdevsw, &u, dev, 0);
606 	if (i) {
607 		if (append_unit) {
608 			namelen = snprintf(devname, sizeof(devname), "%s%d",
609 			    name, u);
610 			name = devname;
611 		}
612 
613 		i = tun_create_device(drv, u, cred, dev, name);
614 	}
615 	if (i == 0) {
616 		dev_ref(*dev);
617 		if_clone_create(name, namelen, NULL);
618 	}
619 out:
620 	CURVNET_RESTORE();
621 }
622 
623 static void
624 tun_destroy(struct tuntap_softc *tp)
625 {
626 
627 	TUN_LOCK(tp);
628 	tp->tun_flags |= TUN_DYING;
629 	if (tp->tun_busy != 0)
630 		cv_wait_unlock(&tp->tun_cv, &tp->tun_mtx);
631 	else
632 		TUN_UNLOCK(tp);
633 
634 	CURVNET_SET(TUN2IFP(tp)->if_vnet);
635 
636 	/* destroy_dev will take care of any alias. */
637 	destroy_dev(tp->tun_dev);
638 	seldrain(&tp->tun_rsel);
639 	knlist_clear(&tp->tun_rsel.si_note, 0);
640 	knlist_destroy(&tp->tun_rsel.si_note);
641 	if ((tp->tun_flags & TUN_L2) != 0) {
642 		ether_ifdetach(TUN2IFP(tp));
643 	} else {
644 		bpfdetach(TUN2IFP(tp));
645 		if_detach(TUN2IFP(tp));
646 	}
647 	sx_xlock(&tun_ioctl_sx);
648 	TUN2IFP(tp)->if_softc = NULL;
649 	sx_xunlock(&tun_ioctl_sx);
650 	free_unr(tp->tun_drv->unrhdr, TUN2IFP(tp)->if_dunit);
651 	if_free(TUN2IFP(tp));
652 	mtx_destroy(&tp->tun_mtx);
653 	cv_destroy(&tp->tun_cv);
654 	free(tp, M_TUN);
655 	CURVNET_RESTORE();
656 }
657 
658 static int
659 tun_clone_destroy(struct if_clone *ifc __unused, struct ifnet *ifp, uint32_t flags)
660 {
661 	struct tuntap_softc *tp = ifp->if_softc;
662 
663 	mtx_lock(&tunmtx);
664 	TAILQ_REMOVE(&tunhead, tp, tun_list);
665 	mtx_unlock(&tunmtx);
666 	tun_destroy(tp);
667 
668 	return (0);
669 }
670 
671 static void
672 vnet_tun_init(const void *unused __unused)
673 {
674 	struct tuntap_driver *drv;
675 	struct tuntap_driver_cloner *drvc;
676 	int i;
677 
678 	for (i = 0; i < nitems(tuntap_drivers); ++i) {
679 		drv = &tuntap_drivers[i];
680 		drvc = malloc(sizeof(*drvc), M_TUN, M_WAITOK | M_ZERO);
681 
682 		drvc->drv = drv;
683 		struct if_clone_addreq req = {
684 			.match_f = drv->clone_match_fn,
685 			.create_f = drv->clone_create_fn,
686 			.destroy_f = drv->clone_destroy_fn,
687 		};
688 		drvc->cloner = ifc_attach_cloner(drv->cdevsw.d_name, &req);
689 		SLIST_INSERT_HEAD(&V_tuntap_driver_cloners, drvc, link);
690 	};
691 }
692 VNET_SYSINIT(vnet_tun_init, SI_SUB_PROTO_IF, SI_ORDER_ANY,
693 		vnet_tun_init, NULL);
694 
695 static void
696 vnet_tun_uninit(const void *unused __unused)
697 {
698 	struct tuntap_driver_cloner *drvc;
699 
700 	while (!SLIST_EMPTY(&V_tuntap_driver_cloners)) {
701 		drvc = SLIST_FIRST(&V_tuntap_driver_cloners);
702 		SLIST_REMOVE_HEAD(&V_tuntap_driver_cloners, link);
703 
704 		if_clone_detach(drvc->cloner);
705 		free(drvc, M_TUN);
706 	}
707 }
708 VNET_SYSUNINIT(vnet_tun_uninit, SI_SUB_PROTO_IF, SI_ORDER_ANY,
709     vnet_tun_uninit, NULL);
710 
711 static void
712 tun_uninit(const void *unused __unused)
713 {
714 	struct tuntap_driver *drv;
715 	struct tuntap_softc *tp;
716 	int i;
717 
718 	EVENTHANDLER_DEREGISTER(ifnet_arrival_event, arrival_tag);
719 	EVENTHANDLER_DEREGISTER(dev_clone, clone_tag);
720 
721 	mtx_lock(&tunmtx);
722 	while ((tp = TAILQ_FIRST(&tunhead)) != NULL) {
723 		TAILQ_REMOVE(&tunhead, tp, tun_list);
724 		mtx_unlock(&tunmtx);
725 		tun_destroy(tp);
726 		mtx_lock(&tunmtx);
727 	}
728 	mtx_unlock(&tunmtx);
729 	for (i = 0; i < nitems(tuntap_drivers); ++i) {
730 		drv = &tuntap_drivers[i];
731 		delete_unrhdr(drv->unrhdr);
732 		clone_cleanup(&drv->clones);
733 	}
734 	mtx_destroy(&tunmtx);
735 }
736 SYSUNINIT(tun_uninit, SI_SUB_PROTO_IF, SI_ORDER_ANY, tun_uninit, NULL);
737 
738 static struct tuntap_driver *
739 tuntap_driver_from_ifnet(const struct ifnet *ifp)
740 {
741 	struct tuntap_driver *drv;
742 	int i;
743 
744 	if (ifp == NULL)
745 		return (NULL);
746 
747 	for (i = 0; i < nitems(tuntap_drivers); ++i) {
748 		drv = &tuntap_drivers[i];
749 		if (strcmp(ifp->if_dname, drv->cdevsw.d_name) == 0)
750 			return (drv);
751 	}
752 
753 	return (NULL);
754 }
755 
756 static int
757 tuntapmodevent(module_t mod, int type, void *data)
758 {
759 	struct tuntap_driver *drv;
760 	int i;
761 
762 	switch (type) {
763 	case MOD_LOAD:
764 		mtx_init(&tunmtx, "tunmtx", NULL, MTX_DEF);
765 		for (i = 0; i < nitems(tuntap_drivers); ++i) {
766 			drv = &tuntap_drivers[i];
767 			clone_setup(&drv->clones);
768 			drv->unrhdr = new_unrhdr(0, IF_MAXUNIT, &tunmtx);
769 		}
770 		arrival_tag = EVENTHANDLER_REGISTER(ifnet_arrival_event,
771 		   tunrename, 0, 1000);
772 		if (arrival_tag == NULL)
773 			return (ENOMEM);
774 		clone_tag = EVENTHANDLER_REGISTER(dev_clone, tunclone, 0, 1000);
775 		if (clone_tag == NULL)
776 			return (ENOMEM);
777 		break;
778 	case MOD_UNLOAD:
779 		/* See tun_uninit, so it's done after the vnet_sysuninit() */
780 		break;
781 	default:
782 		return EOPNOTSUPP;
783 	}
784 	return 0;
785 }
786 
787 static moduledata_t tuntap_mod = {
788 	"if_tuntap",
789 	tuntapmodevent,
790 	0
791 };
792 
793 /* We'll only ever have these two, so no need for a macro. */
794 static moduledata_t tun_mod = { "if_tun", NULL, 0 };
795 static moduledata_t tap_mod = { "if_tap", NULL, 0 };
796 
797 DECLARE_MODULE(if_tuntap, tuntap_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
798 MODULE_VERSION(if_tuntap, 1);
799 DECLARE_MODULE(if_tun, tun_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
800 MODULE_VERSION(if_tun, 1);
801 DECLARE_MODULE(if_tap, tap_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
802 MODULE_VERSION(if_tap, 1);
803 
804 static int
805 tun_create_device(struct tuntap_driver *drv, int unit, struct ucred *cr,
806     struct cdev **dev, const char *name)
807 {
808 	struct make_dev_args args;
809 	struct tuntap_softc *tp;
810 	int error;
811 
812 	tp = malloc(sizeof(*tp), M_TUN, M_WAITOK | M_ZERO);
813 	mtx_init(&tp->tun_mtx, "tun_mtx", NULL, MTX_DEF);
814 	cv_init(&tp->tun_cv, "tun_condvar");
815 	tp->tun_flags = drv->ident_flags;
816 	tp->tun_drv = drv;
817 
818 	make_dev_args_init(&args);
819 	if (cr != NULL)
820 		args.mda_flags = MAKEDEV_REF;
821 	args.mda_devsw = &drv->cdevsw;
822 	args.mda_cr = cr;
823 	args.mda_uid = UID_UUCP;
824 	args.mda_gid = GID_DIALER;
825 	args.mda_mode = 0600;
826 	args.mda_unit = unit;
827 	args.mda_si_drv1 = tp;
828 	error = make_dev_s(&args, dev, "%s", name);
829 	if (error != 0) {
830 		free(tp, M_TUN);
831 		return (error);
832 	}
833 
834 	KASSERT((*dev)->si_drv1 != NULL,
835 	    ("Failed to set si_drv1 at %s creation", name));
836 	tp->tun_dev = *dev;
837 	knlist_init_mtx(&tp->tun_rsel.si_note, &tp->tun_mtx);
838 	mtx_lock(&tunmtx);
839 	TAILQ_INSERT_TAIL(&tunhead, tp, tun_list);
840 	mtx_unlock(&tunmtx);
841 	return (0);
842 }
843 
844 static void
845 tunstart(struct ifnet *ifp)
846 {
847 	struct tuntap_softc *tp = ifp->if_softc;
848 	struct mbuf *m;
849 
850 	TUNDEBUG(ifp, "starting\n");
851 	if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
852 		IFQ_LOCK(&ifp->if_snd);
853 		IFQ_POLL_NOLOCK(&ifp->if_snd, m);
854 		if (m == NULL) {
855 			IFQ_UNLOCK(&ifp->if_snd);
856 			return;
857 		}
858 		IFQ_UNLOCK(&ifp->if_snd);
859 	}
860 
861 	TUN_LOCK(tp);
862 	if (tp->tun_flags & TUN_RWAIT) {
863 		tp->tun_flags &= ~TUN_RWAIT;
864 		wakeup(tp);
865 	}
866 	selwakeuppri(&tp->tun_rsel, PZERO + 1);
867 	KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
868 	if (tp->tun_flags & TUN_ASYNC && tp->tun_sigio) {
869 		TUN_UNLOCK(tp);
870 		pgsigio(&tp->tun_sigio, SIGIO, 0);
871 	} else
872 		TUN_UNLOCK(tp);
873 }
874 
875 /*
876  * tunstart_l2
877  *
878  * queue packets from higher level ready to put out
879  */
880 static void
881 tunstart_l2(struct ifnet *ifp)
882 {
883 	struct tuntap_softc	*tp = ifp->if_softc;
884 
885 	TUNDEBUG(ifp, "starting\n");
886 
887 	/*
888 	 * do not junk pending output if we are in VMnet mode.
889 	 * XXX: can this do any harm because of queue overflow?
890 	 */
891 
892 	TUN_LOCK(tp);
893 	if (((tp->tun_flags & TUN_VMNET) == 0) &&
894 	    ((tp->tun_flags & TUN_READY) != TUN_READY)) {
895 		struct mbuf *m;
896 
897 		/* Unlocked read. */
898 		TUNDEBUG(ifp, "not ready, tun_flags = 0x%x\n", tp->tun_flags);
899 
900 		for (;;) {
901 			IF_DEQUEUE(&ifp->if_snd, m);
902 			if (m != NULL) {
903 				m_freem(m);
904 				if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
905 			} else
906 				break;
907 		}
908 		TUN_UNLOCK(tp);
909 
910 		return;
911 	}
912 
913 	ifp->if_drv_flags |= IFF_DRV_OACTIVE;
914 
915 	if (!IFQ_IS_EMPTY(&ifp->if_snd)) {
916 		if (tp->tun_flags & TUN_RWAIT) {
917 			tp->tun_flags &= ~TUN_RWAIT;
918 			wakeup(tp);
919 		}
920 
921 		if ((tp->tun_flags & TUN_ASYNC) && (tp->tun_sigio != NULL)) {
922 			TUN_UNLOCK(tp);
923 			pgsigio(&tp->tun_sigio, SIGIO, 0);
924 			TUN_LOCK(tp);
925 		}
926 
927 		selwakeuppri(&tp->tun_rsel, PZERO+1);
928 		KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
929 		if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); /* obytes are counted in ether_output */
930 	}
931 
932 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
933 	TUN_UNLOCK(tp);
934 } /* tunstart_l2 */
935 
936 /* XXX: should return an error code so it can fail. */
937 static void
938 tuncreate(struct cdev *dev)
939 {
940 	struct tuntap_driver *drv;
941 	struct tuntap_softc *tp;
942 	struct ifnet *ifp;
943 	struct ether_addr eaddr;
944 	int iflags;
945 	u_char type;
946 
947 	tp = dev->si_drv1;
948 	KASSERT(tp != NULL,
949 	    ("si_drv1 should have been initialized at creation"));
950 
951 	drv = tp->tun_drv;
952 	iflags = IFF_MULTICAST;
953 	if ((tp->tun_flags & TUN_L2) != 0) {
954 		type = IFT_ETHER;
955 		iflags |= IFF_BROADCAST | IFF_SIMPLEX;
956 	} else {
957 		type = IFT_PPP;
958 		iflags |= IFF_POINTOPOINT;
959 	}
960 	ifp = tp->tun_ifp = if_alloc(type);
961 	if (ifp == NULL)
962 		panic("%s%d: failed to if_alloc() interface.\n",
963 		    drv->cdevsw.d_name, dev2unit(dev));
964 	ifp->if_softc = tp;
965 	if_initname(ifp, drv->cdevsw.d_name, dev2unit(dev));
966 	ifp->if_ioctl = tunifioctl;
967 	ifp->if_flags = iflags;
968 	IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen);
969 	ifp->if_capabilities |= IFCAP_LINKSTATE;
970 	ifp->if_capenable |= IFCAP_LINKSTATE;
971 
972 	if ((tp->tun_flags & TUN_L2) != 0) {
973 		ifp->if_init = tunifinit;
974 		ifp->if_start = tunstart_l2;
975 
976 		ether_gen_addr(ifp, &eaddr);
977 		ether_ifattach(ifp, eaddr.octet);
978 	} else {
979 		ifp->if_mtu = TUNMTU;
980 		ifp->if_start = tunstart;
981 		ifp->if_output = tunoutput;
982 
983 		ifp->if_snd.ifq_drv_maxlen = 0;
984 		IFQ_SET_READY(&ifp->if_snd);
985 
986 		if_attach(ifp);
987 		bpfattach(ifp, DLT_NULL, sizeof(u_int32_t));
988 	}
989 
990 	TUN_LOCK(tp);
991 	tp->tun_flags |= TUN_INITED;
992 	TUN_UNLOCK(tp);
993 
994 	TUNDEBUG(ifp, "interface %s is created, minor = %#x\n",
995 	    ifp->if_xname, dev2unit(dev));
996 }
997 
998 static void
999 tunrename(void *arg __unused, struct ifnet *ifp)
1000 {
1001 	struct tuntap_softc *tp;
1002 	int error;
1003 
1004 	if ((ifp->if_flags & IFF_RENAMING) == 0)
1005 		return;
1006 
1007 	if (tuntap_driver_from_ifnet(ifp) == NULL)
1008 		return;
1009 
1010 	/*
1011 	 * We need to grab the ioctl sx long enough to make sure the softc is
1012 	 * still there.  If it is, we can safely try to busy the tun device.
1013 	 * The busy may fail if the device is currently dying, in which case
1014 	 * we do nothing.  If it doesn't fail, the busy count stops the device
1015 	 * from dying until we've created the alias (that will then be
1016 	 * subsequently destroyed).
1017 	 */
1018 	sx_xlock(&tun_ioctl_sx);
1019 	tp = ifp->if_softc;
1020 	if (tp == NULL) {
1021 		sx_xunlock(&tun_ioctl_sx);
1022 		return;
1023 	}
1024 	error = tun_busy(tp);
1025 	sx_xunlock(&tun_ioctl_sx);
1026 	if (error != 0)
1027 		return;
1028 	if (tp->tun_alias != NULL) {
1029 		destroy_dev(tp->tun_alias);
1030 		tp->tun_alias = NULL;
1031 	}
1032 
1033 	if (strcmp(ifp->if_xname, tp->tun_dev->si_name) == 0)
1034 		goto out;
1035 
1036 	/*
1037 	 * Failure's ok, aliases are created on a best effort basis.  If a
1038 	 * tun user/consumer decides to rename the interface to conflict with
1039 	 * another device (non-ifnet) on the system, we will assume they know
1040 	 * what they are doing.  make_dev_alias_p won't touch tun_alias on
1041 	 * failure, so we use it but ignore the return value.
1042 	 */
1043 	make_dev_alias_p(MAKEDEV_CHECKNAME, &tp->tun_alias, tp->tun_dev, "%s",
1044 	    ifp->if_xname);
1045 out:
1046 	tun_unbusy(tp);
1047 }
1048 
1049 static int
1050 tunopen(struct cdev *dev, int flag, int mode, struct thread *td)
1051 {
1052 	struct ifnet	*ifp;
1053 	struct tuntap_softc *tp;
1054 	int error __diagused, tunflags;
1055 
1056 	tunflags = 0;
1057 	CURVNET_SET(TD_TO_VNET(td));
1058 	error = tuntap_name2info(dev->si_name, NULL, &tunflags);
1059 	if (error != 0) {
1060 		CURVNET_RESTORE();
1061 		return (error);	/* Shouldn't happen */
1062 	}
1063 
1064 	tp = dev->si_drv1;
1065 	KASSERT(tp != NULL,
1066 	    ("si_drv1 should have been initialized at creation"));
1067 
1068 	TUN_LOCK(tp);
1069 	if ((tp->tun_flags & TUN_INITED) == 0) {
1070 		TUN_UNLOCK(tp);
1071 		CURVNET_RESTORE();
1072 		return (ENXIO);
1073 	}
1074 	if ((tp->tun_flags & (TUN_OPEN | TUN_DYING)) != 0) {
1075 		TUN_UNLOCK(tp);
1076 		CURVNET_RESTORE();
1077 		return (EBUSY);
1078 	}
1079 
1080 	error = tun_busy_locked(tp);
1081 	KASSERT(error == 0, ("Must be able to busy an unopen tunnel"));
1082 	ifp = TUN2IFP(tp);
1083 
1084 	if ((tp->tun_flags & TUN_L2) != 0) {
1085 		bcopy(IF_LLADDR(ifp), tp->tun_ether.octet,
1086 		    sizeof(tp->tun_ether.octet));
1087 
1088 		ifp->if_drv_flags |= IFF_DRV_RUNNING;
1089 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1090 
1091 		if (tapuponopen)
1092 			ifp->if_flags |= IFF_UP;
1093 	}
1094 
1095 	tp->tun_pid = td->td_proc->p_pid;
1096 	tp->tun_flags |= TUN_OPEN;
1097 
1098 	if_link_state_change(ifp, LINK_STATE_UP);
1099 	TUNDEBUG(ifp, "open\n");
1100 	TUN_UNLOCK(tp);
1101 
1102 	/*
1103 	 * This can fail with either ENOENT or EBUSY.  This is in the middle of
1104 	 * d_open, so ENOENT should not be possible.  EBUSY is possible, but
1105 	 * the only cdevpriv dtor being set will be tundtor and the softc being
1106 	 * passed is constant for a given cdev.  We ignore the possible error
1107 	 * because of this as either "unlikely" or "not actually a problem."
1108 	 */
1109 	(void)devfs_set_cdevpriv(tp, tundtor);
1110 	CURVNET_RESTORE();
1111 	return (0);
1112 }
1113 
1114 /*
1115  * tundtor - tear down the device - mark i/f down & delete
1116  * routing info
1117  */
1118 static void
1119 tundtor(void *data)
1120 {
1121 	struct proc *p;
1122 	struct tuntap_softc *tp;
1123 	struct ifnet *ifp;
1124 	bool l2tun;
1125 
1126 	tp = data;
1127 	p = curproc;
1128 	ifp = TUN2IFP(tp);
1129 
1130 	TUN_LOCK(tp);
1131 
1132 	/*
1133 	 * Realistically, we can't be obstinate here.  This only means that the
1134 	 * tuntap device was closed out of order, and the last closer wasn't the
1135 	 * controller.  These are still good to know about, though, as software
1136 	 * should avoid multiple processes with a tuntap device open and
1137 	 * ill-defined transfer of control (e.g., handoff, TUNSIFPID, close in
1138 	 * parent).
1139 	 */
1140 	if (p->p_pid != tp->tun_pid) {
1141 		log(LOG_INFO,
1142 		    "pid %d (%s), %s: tun/tap protocol violation, non-controlling process closed last.\n",
1143 		    p->p_pid, p->p_comm, tp->tun_dev->si_name);
1144 	}
1145 
1146 	/*
1147 	 * junk all pending output
1148 	 */
1149 	CURVNET_SET(ifp->if_vnet);
1150 
1151 	l2tun = false;
1152 	if ((tp->tun_flags & TUN_L2) != 0) {
1153 		l2tun = true;
1154 		IF_DRAIN(&ifp->if_snd);
1155 	} else {
1156 		IFQ_PURGE(&ifp->if_snd);
1157 	}
1158 
1159 	/* For vmnet, we won't do most of the address/route bits */
1160 	if ((tp->tun_flags & TUN_VMNET) != 0 ||
1161 	    (l2tun && (ifp->if_flags & IFF_LINK0) != 0))
1162 		goto out;
1163 
1164 	if (ifp->if_flags & IFF_UP) {
1165 		TUN_UNLOCK(tp);
1166 		if_down(ifp);
1167 		TUN_LOCK(tp);
1168 	}
1169 
1170 	/* Delete all addresses and routes which reference this interface. */
1171 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1172 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1173 		TUN_UNLOCK(tp);
1174 		if_purgeaddrs(ifp);
1175 		TUN_LOCK(tp);
1176 	}
1177 
1178 out:
1179 	if_link_state_change(ifp, LINK_STATE_DOWN);
1180 	CURVNET_RESTORE();
1181 
1182 	funsetown(&tp->tun_sigio);
1183 	selwakeuppri(&tp->tun_rsel, PZERO + 1);
1184 	KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
1185 	TUNDEBUG (ifp, "closed\n");
1186 	tp->tun_flags &= ~TUN_OPEN;
1187 	tp->tun_pid = 0;
1188 	tun_vnethdr_set(ifp, 0);
1189 
1190 	tun_unbusy_locked(tp);
1191 	TUN_UNLOCK(tp);
1192 }
1193 
1194 static void
1195 tuninit(struct ifnet *ifp)
1196 {
1197 	struct tuntap_softc *tp = ifp->if_softc;
1198 
1199 	TUNDEBUG(ifp, "tuninit\n");
1200 
1201 	TUN_LOCK(tp);
1202 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1203 	if ((tp->tun_flags & TUN_L2) == 0) {
1204 		ifp->if_flags |= IFF_UP;
1205 		getmicrotime(&ifp->if_lastchange);
1206 		TUN_UNLOCK(tp);
1207 	} else {
1208 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1209 		TUN_UNLOCK(tp);
1210 		/* attempt to start output */
1211 		tunstart_l2(ifp);
1212 	}
1213 
1214 }
1215 
1216 /*
1217  * Used only for l2 tunnel.
1218  */
1219 static void
1220 tunifinit(void *xtp)
1221 {
1222 	struct tuntap_softc *tp;
1223 
1224 	tp = (struct tuntap_softc *)xtp;
1225 	tuninit(tp->tun_ifp);
1226 }
1227 
1228 /*
1229  * To be called under TUN_LOCK. Update ifp->if_hwassist according to the
1230  * current value of ifp->if_capenable.
1231  */
1232 static void
1233 tun_caps_changed(struct ifnet *ifp)
1234 {
1235 	uint64_t hwassist = 0;
1236 
1237 	TUN_LOCK_ASSERT((struct tuntap_softc *)ifp->if_softc);
1238 	if (ifp->if_capenable & IFCAP_TXCSUM)
1239 		hwassist |= CSUM_TCP | CSUM_UDP;
1240 	if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
1241 		hwassist |= CSUM_TCP_IPV6
1242 		    | CSUM_UDP_IPV6;
1243 	if (ifp->if_capenable & IFCAP_TSO4)
1244 		hwassist |= CSUM_IP_TSO;
1245 	if (ifp->if_capenable & IFCAP_TSO6)
1246 		hwassist |= CSUM_IP6_TSO;
1247 	ifp->if_hwassist = hwassist;
1248 }
1249 
1250 /*
1251  * To be called under TUN_LOCK. Update tp->tun_vhdrlen and adjust
1252  * if_capabilities and if_capenable as needed.
1253  */
1254 static void
1255 tun_vnethdr_set(struct ifnet *ifp, int vhdrlen)
1256 {
1257 	struct tuntap_softc *tp = ifp->if_softc;
1258 
1259 	TUN_LOCK_ASSERT(tp);
1260 
1261 	if (tp->tun_vhdrlen == vhdrlen)
1262 		return;
1263 
1264 	/*
1265 	 * Update if_capabilities to reflect the
1266 	 * functionalities offered by the virtio-net
1267 	 * header.
1268 	 */
1269 	if (vhdrlen != 0)
1270 		ifp->if_capabilities |=
1271 			TAP_VNET_HDR_CAPS;
1272 	else
1273 		ifp->if_capabilities &=
1274 			~TAP_VNET_HDR_CAPS;
1275 	/*
1276 	 * Disable any capabilities that we don't
1277 	 * support anymore.
1278 	 */
1279 	ifp->if_capenable &= ifp->if_capabilities;
1280 	tun_caps_changed(ifp);
1281 	tp->tun_vhdrlen = vhdrlen;
1282 
1283 	TUNDEBUG(ifp, "vnet_hdr_len=%d, if_capabilities=%x\n",
1284 	    vhdrlen, ifp->if_capabilities);
1285 }
1286 
1287 /*
1288  * Process an ioctl request.
1289  */
1290 static int
1291 tunifioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1292 {
1293 	struct ifreq *ifr = (struct ifreq *)data;
1294 	struct tuntap_softc *tp;
1295 	struct ifstat *ifs;
1296 	struct ifmediareq	*ifmr;
1297 	int		dummy, error = 0;
1298 	bool		l2tun;
1299 
1300 	ifmr = NULL;
1301 	sx_xlock(&tun_ioctl_sx);
1302 	tp = ifp->if_softc;
1303 	if (tp == NULL) {
1304 		error = ENXIO;
1305 		goto bad;
1306 	}
1307 	l2tun = (tp->tun_flags & TUN_L2) != 0;
1308 	switch(cmd) {
1309 	case SIOCGIFSTATUS:
1310 		ifs = (struct ifstat *)data;
1311 		TUN_LOCK(tp);
1312 		if (tp->tun_pid)
1313 			snprintf(ifs->ascii, sizeof(ifs->ascii),
1314 			    "\tOpened by PID %d\n", tp->tun_pid);
1315 		else
1316 			ifs->ascii[0] = '\0';
1317 		TUN_UNLOCK(tp);
1318 		break;
1319 	case SIOCSIFADDR:
1320 		if (l2tun)
1321 			error = ether_ioctl(ifp, cmd, data);
1322 		else
1323 			tuninit(ifp);
1324 		if (error == 0)
1325 		    TUNDEBUG(ifp, "address set\n");
1326 		break;
1327 	case SIOCSIFMTU:
1328 		ifp->if_mtu = ifr->ifr_mtu;
1329 		TUNDEBUG(ifp, "mtu set\n");
1330 		break;
1331 	case SIOCSIFFLAGS:
1332 	case SIOCADDMULTI:
1333 	case SIOCDELMULTI:
1334 		break;
1335 	case SIOCGIFMEDIA:
1336 		if (!l2tun) {
1337 			error = EINVAL;
1338 			break;
1339 		}
1340 
1341 		ifmr = (struct ifmediareq *)data;
1342 		dummy = ifmr->ifm_count;
1343 		ifmr->ifm_count = 1;
1344 		ifmr->ifm_status = IFM_AVALID;
1345 		ifmr->ifm_active = IFM_ETHER | IFM_FDX | IFM_1000_T;
1346 		if (tp->tun_flags & TUN_OPEN)
1347 			ifmr->ifm_status |= IFM_ACTIVE;
1348 		ifmr->ifm_current = ifmr->ifm_active;
1349 		if (dummy >= 1) {
1350 			int media = IFM_ETHER;
1351 			error = copyout(&media, ifmr->ifm_ulist, sizeof(int));
1352 		}
1353 		break;
1354 	case SIOCSIFCAP:
1355 		TUN_LOCK(tp);
1356 		ifp->if_capenable = ifr->ifr_reqcap;
1357 		tun_caps_changed(ifp);
1358 		TUN_UNLOCK(tp);
1359 		VLAN_CAPABILITIES(ifp);
1360 		break;
1361 	default:
1362 		if (l2tun) {
1363 			error = ether_ioctl(ifp, cmd, data);
1364 		} else {
1365 			error = EINVAL;
1366 		}
1367 	}
1368 bad:
1369 	sx_xunlock(&tun_ioctl_sx);
1370 	return (error);
1371 }
1372 
1373 /*
1374  * tunoutput - queue packets from higher level ready to put out.
1375  */
1376 static int
1377 tunoutput(struct ifnet *ifp, struct mbuf *m0, const struct sockaddr *dst,
1378     struct route *ro)
1379 {
1380 	struct tuntap_softc *tp = ifp->if_softc;
1381 	u_short cached_tun_flags;
1382 	int error;
1383 	u_int32_t af;
1384 
1385 	TUNDEBUG (ifp, "tunoutput\n");
1386 
1387 #ifdef MAC
1388 	error = mac_ifnet_check_transmit(ifp, m0);
1389 	if (error) {
1390 		m_freem(m0);
1391 		return (error);
1392 	}
1393 #endif
1394 
1395 	/* Could be unlocked read? */
1396 	TUN_LOCK(tp);
1397 	cached_tun_flags = tp->tun_flags;
1398 	TUN_UNLOCK(tp);
1399 	if ((cached_tun_flags & TUN_READY) != TUN_READY) {
1400 		TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags);
1401 		m_freem (m0);
1402 		return (EHOSTDOWN);
1403 	}
1404 
1405 	if ((ifp->if_flags & IFF_UP) != IFF_UP) {
1406 		m_freem (m0);
1407 		return (EHOSTDOWN);
1408 	}
1409 
1410 	/* BPF writes need to be handled specially. */
1411 	if (dst->sa_family == AF_UNSPEC)
1412 		bcopy(dst->sa_data, &af, sizeof(af));
1413 	else
1414 		af = RO_GET_FAMILY(ro, dst);
1415 
1416 	if (bpf_peers_present(ifp->if_bpf))
1417 		bpf_mtap2(ifp->if_bpf, &af, sizeof(af), m0);
1418 
1419 	/* prepend sockaddr? this may abort if the mbuf allocation fails */
1420 	if (cached_tun_flags & TUN_LMODE) {
1421 		/* allocate space for sockaddr */
1422 		M_PREPEND(m0, dst->sa_len, M_NOWAIT);
1423 
1424 		/* if allocation failed drop packet */
1425 		if (m0 == NULL) {
1426 			if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1427 			if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1428 			return (ENOBUFS);
1429 		} else {
1430 			bcopy(dst, m0->m_data, dst->sa_len);
1431 		}
1432 	}
1433 
1434 	if (cached_tun_flags & TUN_IFHEAD) {
1435 		/* Prepend the address family */
1436 		M_PREPEND(m0, 4, M_NOWAIT);
1437 
1438 		/* if allocation failed drop packet */
1439 		if (m0 == NULL) {
1440 			if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1441 			if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1442 			return (ENOBUFS);
1443 		} else
1444 			*(u_int32_t *)m0->m_data = htonl(af);
1445 	} else {
1446 #ifdef INET
1447 		if (af != AF_INET)
1448 #endif
1449 		{
1450 			m_freem(m0);
1451 			return (EAFNOSUPPORT);
1452 		}
1453 	}
1454 
1455 	error = (ifp->if_transmit)(ifp, m0);
1456 	if (error)
1457 		return (ENOBUFS);
1458 	if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1459 	return (0);
1460 }
1461 
1462 /*
1463  * the cdevsw interface is now pretty minimal.
1464  */
1465 static	int
1466 tunioctl(struct cdev *dev, u_long cmd, caddr_t data, int flag,
1467     struct thread *td)
1468 {
1469 	struct ifreq ifr, *ifrp;
1470 	struct tuntap_softc *tp = dev->si_drv1;
1471 	struct ifnet *ifp = TUN2IFP(tp);
1472 	struct tuninfo *tunp;
1473 	int error, iflags, ival;
1474 	bool	l2tun;
1475 
1476 	l2tun = (tp->tun_flags & TUN_L2) != 0;
1477 	if (l2tun) {
1478 		/* tap specific ioctls */
1479 		switch(cmd) {
1480 		/* VMware/VMnet port ioctl's */
1481 #if defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD5) || \
1482     defined(COMPAT_FREEBSD4)
1483 		case _IO('V', 0):
1484 			ival = IOCPARM_IVAL(data);
1485 			data = (caddr_t)&ival;
1486 			/* FALLTHROUGH */
1487 #endif
1488 		case VMIO_SIOCSIFFLAGS: /* VMware/VMnet SIOCSIFFLAGS */
1489 			iflags = *(int *)data;
1490 			iflags &= TUN_VMIO_FLAG_MASK;
1491 			iflags &= ~IFF_CANTCHANGE;
1492 			iflags |= IFF_UP;
1493 
1494 			TUN_LOCK(tp);
1495 			ifp->if_flags = iflags |
1496 			    (ifp->if_flags & IFF_CANTCHANGE);
1497 			TUN_UNLOCK(tp);
1498 
1499 			return (0);
1500 		case SIOCGIFADDR:	/* get MAC address of the remote side */
1501 			TUN_LOCK(tp);
1502 			bcopy(&tp->tun_ether.octet, data,
1503 			    sizeof(tp->tun_ether.octet));
1504 			TUN_UNLOCK(tp);
1505 
1506 			return (0);
1507 		case SIOCSIFADDR:	/* set MAC address of the remote side */
1508 			TUN_LOCK(tp);
1509 			bcopy(data, &tp->tun_ether.octet,
1510 			    sizeof(tp->tun_ether.octet));
1511 			TUN_UNLOCK(tp);
1512 
1513 			return (0);
1514 		case TAPSVNETHDR:
1515 			ival = *(int *)data;
1516 			if (ival != 0 &&
1517 			    ival != sizeof(struct virtio_net_hdr) &&
1518 			    ival != sizeof(struct virtio_net_hdr_mrg_rxbuf)) {
1519 				return (EINVAL);
1520 			}
1521 			TUN_LOCK(tp);
1522 			tun_vnethdr_set(ifp, ival);
1523 			TUN_UNLOCK(tp);
1524 
1525 			return (0);
1526 		case TAPGVNETHDR:
1527 			TUN_LOCK(tp);
1528 			*(int *)data = tp->tun_vhdrlen;
1529 			TUN_UNLOCK(tp);
1530 
1531 			return (0);
1532 		}
1533 
1534 		/* Fall through to the common ioctls if unhandled */
1535 	} else {
1536 		switch (cmd) {
1537 		case TUNSLMODE:
1538 			TUN_LOCK(tp);
1539 			if (*(int *)data) {
1540 				tp->tun_flags |= TUN_LMODE;
1541 				tp->tun_flags &= ~TUN_IFHEAD;
1542 			} else
1543 				tp->tun_flags &= ~TUN_LMODE;
1544 			TUN_UNLOCK(tp);
1545 
1546 			return (0);
1547 		case TUNSIFHEAD:
1548 			TUN_LOCK(tp);
1549 			if (*(int *)data) {
1550 				tp->tun_flags |= TUN_IFHEAD;
1551 				tp->tun_flags &= ~TUN_LMODE;
1552 			} else
1553 				tp->tun_flags &= ~TUN_IFHEAD;
1554 			TUN_UNLOCK(tp);
1555 
1556 			return (0);
1557 		case TUNGIFHEAD:
1558 			TUN_LOCK(tp);
1559 			*(int *)data = (tp->tun_flags & TUN_IFHEAD) ? 1 : 0;
1560 			TUN_UNLOCK(tp);
1561 
1562 			return (0);
1563 		case TUNSIFMODE:
1564 			/* deny this if UP */
1565 			if (TUN2IFP(tp)->if_flags & IFF_UP)
1566 				return (EBUSY);
1567 
1568 			switch (*(int *)data & ~IFF_MULTICAST) {
1569 			case IFF_POINTOPOINT:
1570 			case IFF_BROADCAST:
1571 				TUN_LOCK(tp);
1572 				TUN2IFP(tp)->if_flags &=
1573 				    ~(IFF_BROADCAST|IFF_POINTOPOINT|IFF_MULTICAST);
1574 				TUN2IFP(tp)->if_flags |= *(int *)data;
1575 				TUN_UNLOCK(tp);
1576 
1577 				break;
1578 			default:
1579 				return (EINVAL);
1580 			}
1581 
1582 			return (0);
1583 		case TUNSIFPID:
1584 			TUN_LOCK(tp);
1585 			tp->tun_pid = curthread->td_proc->p_pid;
1586 			TUN_UNLOCK(tp);
1587 
1588 			return (0);
1589 		}
1590 		/* Fall through to the common ioctls if unhandled */
1591 	}
1592 
1593 	switch (cmd) {
1594 	case TUNGIFNAME:
1595 		ifrp = (struct ifreq *)data;
1596 		strlcpy(ifrp->ifr_name, TUN2IFP(tp)->if_xname, IFNAMSIZ);
1597 
1598 		return (0);
1599 	case TUNSIFINFO:
1600 		tunp = (struct tuninfo *)data;
1601 		if (TUN2IFP(tp)->if_type != tunp->type)
1602 			return (EPROTOTYPE);
1603 		TUN_LOCK(tp);
1604 		if (TUN2IFP(tp)->if_mtu != tunp->mtu) {
1605 			strlcpy(ifr.ifr_name, if_name(TUN2IFP(tp)), IFNAMSIZ);
1606 			ifr.ifr_mtu = tunp->mtu;
1607 			CURVNET_SET(TUN2IFP(tp)->if_vnet);
1608 			error = ifhwioctl(SIOCSIFMTU, TUN2IFP(tp),
1609 			    (caddr_t)&ifr, td);
1610 			CURVNET_RESTORE();
1611 			if (error) {
1612 				TUN_UNLOCK(tp);
1613 				return (error);
1614 			}
1615 		}
1616 		TUN2IFP(tp)->if_baudrate = tunp->baudrate;
1617 		TUN_UNLOCK(tp);
1618 		break;
1619 	case TUNGIFINFO:
1620 		tunp = (struct tuninfo *)data;
1621 		TUN_LOCK(tp);
1622 		tunp->mtu = TUN2IFP(tp)->if_mtu;
1623 		tunp->type = TUN2IFP(tp)->if_type;
1624 		tunp->baudrate = TUN2IFP(tp)->if_baudrate;
1625 		TUN_UNLOCK(tp);
1626 		break;
1627 	case TUNSDEBUG:
1628 		tundebug = *(int *)data;
1629 		break;
1630 	case TUNGDEBUG:
1631 		*(int *)data = tundebug;
1632 		break;
1633 	case FIONBIO:
1634 		break;
1635 	case FIOASYNC:
1636 		TUN_LOCK(tp);
1637 		if (*(int *)data)
1638 			tp->tun_flags |= TUN_ASYNC;
1639 		else
1640 			tp->tun_flags &= ~TUN_ASYNC;
1641 		TUN_UNLOCK(tp);
1642 		break;
1643 	case FIONREAD:
1644 		if (!IFQ_IS_EMPTY(&TUN2IFP(tp)->if_snd)) {
1645 			struct mbuf *mb;
1646 			IFQ_LOCK(&TUN2IFP(tp)->if_snd);
1647 			IFQ_POLL_NOLOCK(&TUN2IFP(tp)->if_snd, mb);
1648 			for (*(int *)data = 0; mb != NULL; mb = mb->m_next)
1649 				*(int *)data += mb->m_len;
1650 			IFQ_UNLOCK(&TUN2IFP(tp)->if_snd);
1651 		} else
1652 			*(int *)data = 0;
1653 		break;
1654 	case FIOSETOWN:
1655 		return (fsetown(*(int *)data, &tp->tun_sigio));
1656 
1657 	case FIOGETOWN:
1658 		*(int *)data = fgetown(&tp->tun_sigio);
1659 		return (0);
1660 
1661 	/* This is deprecated, FIOSETOWN should be used instead. */
1662 	case TIOCSPGRP:
1663 		return (fsetown(-(*(int *)data), &tp->tun_sigio));
1664 
1665 	/* This is deprecated, FIOGETOWN should be used instead. */
1666 	case TIOCGPGRP:
1667 		*(int *)data = -fgetown(&tp->tun_sigio);
1668 		return (0);
1669 
1670 	default:
1671 		return (ENOTTY);
1672 	}
1673 	return (0);
1674 }
1675 
1676 /*
1677  * The cdevsw read interface - reads a packet at a time, or at
1678  * least as much of a packet as can be read.
1679  */
1680 static	int
1681 tunread(struct cdev *dev, struct uio *uio, int flag)
1682 {
1683 	struct tuntap_softc *tp = dev->si_drv1;
1684 	struct ifnet	*ifp = TUN2IFP(tp);
1685 	struct mbuf	*m;
1686 	size_t		len;
1687 	int		error = 0;
1688 
1689 	TUNDEBUG (ifp, "read\n");
1690 	TUN_LOCK(tp);
1691 	if ((tp->tun_flags & TUN_READY) != TUN_READY) {
1692 		TUN_UNLOCK(tp);
1693 		TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags);
1694 		return (EHOSTDOWN);
1695 	}
1696 
1697 	tp->tun_flags &= ~TUN_RWAIT;
1698 
1699 	for (;;) {
1700 		IFQ_DEQUEUE(&ifp->if_snd, m);
1701 		if (m != NULL)
1702 			break;
1703 		if (flag & O_NONBLOCK) {
1704 			TUN_UNLOCK(tp);
1705 			return (EWOULDBLOCK);
1706 		}
1707 		tp->tun_flags |= TUN_RWAIT;
1708 		error = mtx_sleep(tp, &tp->tun_mtx, PCATCH | (PZERO + 1),
1709 		    "tunread", 0);
1710 		if (error != 0) {
1711 			TUN_UNLOCK(tp);
1712 			return (error);
1713 		}
1714 	}
1715 	TUN_UNLOCK(tp);
1716 
1717 	if ((tp->tun_flags & TUN_L2) != 0)
1718 		BPF_MTAP(ifp, m);
1719 
1720 	len = min(tp->tun_vhdrlen, uio->uio_resid);
1721 	if (len > 0) {
1722 		struct virtio_net_hdr_mrg_rxbuf vhdr;
1723 
1724 		bzero(&vhdr, sizeof(vhdr));
1725 		if (m->m_pkthdr.csum_flags & TAP_ALL_OFFLOAD) {
1726 			m = virtio_net_tx_offload(ifp, m, false, &vhdr.hdr);
1727 		}
1728 
1729 		TUNDEBUG(ifp, "txvhdr: f %u, gt %u, hl %u, "
1730 		    "gs %u, cs %u, co %u\n", vhdr.hdr.flags,
1731 		    vhdr.hdr.gso_type, vhdr.hdr.hdr_len,
1732 		    vhdr.hdr.gso_size, vhdr.hdr.csum_start,
1733 		    vhdr.hdr.csum_offset);
1734 		error = uiomove(&vhdr, len, uio);
1735 	}
1736 
1737 	while (m && uio->uio_resid > 0 && error == 0) {
1738 		len = min(uio->uio_resid, m->m_len);
1739 		if (len != 0)
1740 			error = uiomove(mtod(m, void *), len, uio);
1741 		m = m_free(m);
1742 	}
1743 
1744 	if (m) {
1745 		TUNDEBUG(ifp, "Dropping mbuf\n");
1746 		m_freem(m);
1747 	}
1748 	return (error);
1749 }
1750 
1751 static int
1752 tunwrite_l2(struct tuntap_softc *tp, struct mbuf *m,
1753 	    struct virtio_net_hdr_mrg_rxbuf *vhdr)
1754 {
1755 	struct epoch_tracker et;
1756 	struct ether_header *eh;
1757 	struct ifnet *ifp;
1758 
1759 	ifp = TUN2IFP(tp);
1760 
1761 	/*
1762 	 * Only pass a unicast frame to ether_input(), if it would
1763 	 * actually have been received by non-virtual hardware.
1764 	 */
1765 	if (m->m_len < sizeof(struct ether_header)) {
1766 		m_freem(m);
1767 		return (0);
1768 	}
1769 
1770 	eh = mtod(m, struct ether_header *);
1771 
1772 	if (eh && (ifp->if_flags & IFF_PROMISC) == 0 &&
1773 	    !ETHER_IS_MULTICAST(eh->ether_dhost) &&
1774 	    bcmp(eh->ether_dhost, IF_LLADDR(ifp), ETHER_ADDR_LEN) != 0) {
1775 		m_freem(m);
1776 		return (0);
1777 	}
1778 
1779 	if (vhdr != NULL && virtio_net_rx_csum(m, &vhdr->hdr)) {
1780 		m_freem(m);
1781 		return (0);
1782 	}
1783 
1784 	/* Pass packet up to parent. */
1785 	CURVNET_SET(ifp->if_vnet);
1786 	NET_EPOCH_ENTER(et);
1787 	(*ifp->if_input)(ifp, m);
1788 	NET_EPOCH_EXIT(et);
1789 	CURVNET_RESTORE();
1790 	/* ibytes are counted in parent */
1791 	if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1792 	return (0);
1793 }
1794 
1795 static int
1796 tunwrite_l3(struct tuntap_softc *tp, struct mbuf *m)
1797 {
1798 	struct epoch_tracker et;
1799 	struct ifnet *ifp;
1800 	int family, isr;
1801 
1802 	ifp = TUN2IFP(tp);
1803 	/* Could be unlocked read? */
1804 	TUN_LOCK(tp);
1805 	if (tp->tun_flags & TUN_IFHEAD) {
1806 		TUN_UNLOCK(tp);
1807 		if (m->m_len < sizeof(family) &&
1808 		(m = m_pullup(m, sizeof(family))) == NULL)
1809 			return (ENOBUFS);
1810 		family = ntohl(*mtod(m, u_int32_t *));
1811 		m_adj(m, sizeof(family));
1812 	} else {
1813 		TUN_UNLOCK(tp);
1814 		family = AF_INET;
1815 	}
1816 
1817 	BPF_MTAP2(ifp, &family, sizeof(family), m);
1818 
1819 	switch (family) {
1820 #ifdef INET
1821 	case AF_INET:
1822 		isr = NETISR_IP;
1823 		break;
1824 #endif
1825 #ifdef INET6
1826 	case AF_INET6:
1827 		isr = NETISR_IPV6;
1828 		break;
1829 #endif
1830 	default:
1831 		m_freem(m);
1832 		return (EAFNOSUPPORT);
1833 	}
1834 	random_harvest_queue(m, sizeof(*m), RANDOM_NET_TUN);
1835 	if_inc_counter(ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
1836 	if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1837 	CURVNET_SET(ifp->if_vnet);
1838 	M_SETFIB(m, ifp->if_fib);
1839 	NET_EPOCH_ENTER(et);
1840 	netisr_dispatch(isr, m);
1841 	NET_EPOCH_EXIT(et);
1842 	CURVNET_RESTORE();
1843 	return (0);
1844 }
1845 
1846 /*
1847  * the cdevsw write interface - an atomic write is a packet - or else!
1848  */
1849 static	int
1850 tunwrite(struct cdev *dev, struct uio *uio, int flag)
1851 {
1852 	struct virtio_net_hdr_mrg_rxbuf vhdr;
1853 	struct tuntap_softc *tp;
1854 	struct ifnet	*ifp;
1855 	struct mbuf	*m;
1856 	uint32_t	mru;
1857 	int		align, vhdrlen, error;
1858 	bool		l2tun;
1859 
1860 	tp = dev->si_drv1;
1861 	ifp = TUN2IFP(tp);
1862 	TUNDEBUG(ifp, "tunwrite\n");
1863 	if ((ifp->if_flags & IFF_UP) != IFF_UP)
1864 		/* ignore silently */
1865 		return (0);
1866 
1867 	if (uio->uio_resid == 0)
1868 		return (0);
1869 
1870 	l2tun = (tp->tun_flags & TUN_L2) != 0;
1871 	mru = l2tun ? TAPMRU : TUNMRU;
1872 	vhdrlen = tp->tun_vhdrlen;
1873 	align = 0;
1874 	if (l2tun) {
1875 		align = ETHER_ALIGN;
1876 		mru += vhdrlen;
1877 	} else if ((tp->tun_flags & TUN_IFHEAD) != 0)
1878 		mru += sizeof(uint32_t);	/* family */
1879 	if (uio->uio_resid < 0 || uio->uio_resid > mru) {
1880 		TUNDEBUG(ifp, "len=%zd!\n", uio->uio_resid);
1881 		return (EIO);
1882 	}
1883 
1884 	if (vhdrlen > 0) {
1885 		error = uiomove(&vhdr, vhdrlen, uio);
1886 		if (error != 0)
1887 			return (error);
1888 		TUNDEBUG(ifp, "txvhdr: f %u, gt %u, hl %u, "
1889 		    "gs %u, cs %u, co %u\n", vhdr.hdr.flags,
1890 		    vhdr.hdr.gso_type, vhdr.hdr.hdr_len,
1891 		    vhdr.hdr.gso_size, vhdr.hdr.csum_start,
1892 		    vhdr.hdr.csum_offset);
1893 	}
1894 
1895 	if ((m = m_uiotombuf(uio, M_NOWAIT, 0, align, M_PKTHDR)) == NULL) {
1896 		if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1897 		return (ENOBUFS);
1898 	}
1899 
1900 	m->m_pkthdr.rcvif = ifp;
1901 #ifdef MAC
1902 	mac_ifnet_create_mbuf(ifp, m);
1903 #endif
1904 
1905 	if (l2tun)
1906 		return (tunwrite_l2(tp, m, vhdrlen > 0 ? &vhdr : NULL));
1907 
1908 	return (tunwrite_l3(tp, m));
1909 }
1910 
1911 /*
1912  * tunpoll - the poll interface, this is only useful on reads
1913  * really. The write detect always returns true, write never blocks
1914  * anyway, it either accepts the packet or drops it.
1915  */
1916 static	int
1917 tunpoll(struct cdev *dev, int events, struct thread *td)
1918 {
1919 	struct tuntap_softc *tp = dev->si_drv1;
1920 	struct ifnet	*ifp = TUN2IFP(tp);
1921 	int		revents = 0;
1922 
1923 	TUNDEBUG(ifp, "tunpoll\n");
1924 
1925 	if (events & (POLLIN | POLLRDNORM)) {
1926 		IFQ_LOCK(&ifp->if_snd);
1927 		if (!IFQ_IS_EMPTY(&ifp->if_snd)) {
1928 			TUNDEBUG(ifp, "tunpoll q=%d\n", ifp->if_snd.ifq_len);
1929 			revents |= events & (POLLIN | POLLRDNORM);
1930 		} else {
1931 			TUNDEBUG(ifp, "tunpoll waiting\n");
1932 			selrecord(td, &tp->tun_rsel);
1933 		}
1934 		IFQ_UNLOCK(&ifp->if_snd);
1935 	}
1936 	revents |= events & (POLLOUT | POLLWRNORM);
1937 
1938 	return (revents);
1939 }
1940 
1941 /*
1942  * tunkqfilter - support for the kevent() system call.
1943  */
1944 static int
1945 tunkqfilter(struct cdev *dev, struct knote *kn)
1946 {
1947 	struct tuntap_softc	*tp = dev->si_drv1;
1948 	struct ifnet	*ifp = TUN2IFP(tp);
1949 
1950 	switch(kn->kn_filter) {
1951 	case EVFILT_READ:
1952 		TUNDEBUG(ifp, "%s kqfilter: EVFILT_READ, minor = %#x\n",
1953 		    ifp->if_xname, dev2unit(dev));
1954 		kn->kn_fop = &tun_read_filterops;
1955 		break;
1956 
1957 	case EVFILT_WRITE:
1958 		TUNDEBUG(ifp, "%s kqfilter: EVFILT_WRITE, minor = %#x\n",
1959 		    ifp->if_xname, dev2unit(dev));
1960 		kn->kn_fop = &tun_write_filterops;
1961 		break;
1962 
1963 	default:
1964 		TUNDEBUG(ifp, "%s kqfilter: invalid filter, minor = %#x\n",
1965 		    ifp->if_xname, dev2unit(dev));
1966 		return(EINVAL);
1967 	}
1968 
1969 	kn->kn_hook = tp;
1970 	knlist_add(&tp->tun_rsel.si_note, kn, 0);
1971 
1972 	return (0);
1973 }
1974 
1975 /*
1976  * Return true of there is data in the interface queue.
1977  */
1978 static int
1979 tunkqread(struct knote *kn, long hint)
1980 {
1981 	int			ret;
1982 	struct tuntap_softc	*tp = kn->kn_hook;
1983 	struct cdev		*dev = tp->tun_dev;
1984 	struct ifnet	*ifp = TUN2IFP(tp);
1985 
1986 	if ((kn->kn_data = ifp->if_snd.ifq_len) > 0) {
1987 		TUNDEBUG(ifp,
1988 		    "%s have data in the queue.  Len = %d, minor = %#x\n",
1989 		    ifp->if_xname, ifp->if_snd.ifq_len, dev2unit(dev));
1990 		ret = 1;
1991 	} else {
1992 		TUNDEBUG(ifp,
1993 		    "%s waiting for data, minor = %#x\n", ifp->if_xname,
1994 		    dev2unit(dev));
1995 		ret = 0;
1996 	}
1997 
1998 	return (ret);
1999 }
2000 
2001 /*
2002  * Always can write, always return MTU in kn->data.
2003  */
2004 static int
2005 tunkqwrite(struct knote *kn, long hint)
2006 {
2007 	struct tuntap_softc	*tp = kn->kn_hook;
2008 	struct ifnet	*ifp = TUN2IFP(tp);
2009 
2010 	kn->kn_data = ifp->if_mtu;
2011 
2012 	return (1);
2013 }
2014 
2015 static void
2016 tunkqdetach(struct knote *kn)
2017 {
2018 	struct tuntap_softc	*tp = kn->kn_hook;
2019 
2020 	knlist_remove(&tp->tun_rsel.si_note, kn, 0);
2021 }
2022