xref: /freebsd/sys/net/if_tuntap.c (revision b197d4b893974c9eb4d7b38704c6d5c486235d6f)
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/eventhandler.h>
62 #include <sys/fcntl.h>
63 #include <sys/filio.h>
64 #include <sys/sockio.h>
65 #include <sys/sx.h>
66 #include <sys/syslog.h>
67 #include <sys/ttycom.h>
68 #include <sys/poll.h>
69 #include <sys/selinfo.h>
70 #include <sys/signalvar.h>
71 #include <sys/filedesc.h>
72 #include <sys/kernel.h>
73 #include <sys/sysctl.h>
74 #include <sys/conf.h>
75 #include <sys/uio.h>
76 #include <sys/malloc.h>
77 #include <sys/random.h>
78 #include <sys/ctype.h>
79 
80 #include <net/ethernet.h>
81 #include <net/if.h>
82 #include <net/if_var.h>
83 #include <net/if_clone.h>
84 #include <net/if_dl.h>
85 #include <net/if_media.h>
86 #include <net/if_types.h>
87 #include <net/if_vlan_var.h>
88 #include <net/netisr.h>
89 #include <net/route.h>
90 #include <net/vnet.h>
91 #include <netinet/in.h>
92 #ifdef INET
93 #include <netinet/ip.h>
94 #endif
95 #ifdef INET6
96 #include <netinet/ip6.h>
97 #include <netinet6/ip6_var.h>
98 #endif
99 #include <netinet/udp.h>
100 #include <netinet/tcp.h>
101 #include <net/bpf.h>
102 #include <net/if_tap.h>
103 #include <net/if_tun.h>
104 
105 #include <dev/virtio/network/virtio_net.h>
106 
107 #include <sys/queue.h>
108 #include <sys/condvar.h>
109 #include <security/mac/mac_framework.h>
110 
111 struct tuntap_driver;
112 
113 /*
114  * tun_list is protected by global tunmtx.  Other mutable fields are
115  * protected by tun->tun_mtx, or by their owning subsystem.  tun_dev is
116  * static for the duration of a tunnel interface.
117  */
118 struct tuntap_softc {
119 	TAILQ_ENTRY(tuntap_softc)	 tun_list;
120 	struct cdev			*tun_alias;
121 	struct cdev			*tun_dev;
122 	u_short				 tun_flags;	/* misc flags */
123 #define	TUN_OPEN	0x0001
124 #define	TUN_INITED	0x0002
125 #define	TUN_UNUSED1	0x0008
126 #define	TUN_UNUSED2	0x0010
127 #define	TUN_LMODE	0x0020
128 #define	TUN_RWAIT	0x0040
129 #define	TUN_ASYNC	0x0080
130 #define	TUN_IFHEAD	0x0100
131 #define	TUN_DYING	0x0200
132 #define	TUN_L2		0x0400
133 #define	TUN_VMNET	0x0800
134 
135 #define	TUN_DRIVER_IDENT_MASK	(TUN_L2 | TUN_VMNET)
136 #define	TUN_READY		(TUN_OPEN | TUN_INITED)
137 
138 	pid_t			 tun_pid;	/* owning pid */
139 	struct ifnet		*tun_ifp;	/* the interface */
140 	struct sigio		*tun_sigio;	/* async I/O info */
141 	struct tuntap_driver	*tun_drv;	/* appropriate driver */
142 	struct selinfo		 tun_rsel;	/* read select */
143 	struct mtx		 tun_mtx;	/* softc field mutex */
144 	struct cv		 tun_cv;	/* for ref'd dev destroy */
145 	struct ether_addr	 tun_ether;	/* remote address */
146 	int			 tun_busy;	/* busy count */
147 	int			 tun_vhdrlen;	/* virtio-net header length */
148 };
149 #define	TUN2IFP(sc)	((sc)->tun_ifp)
150 
151 #define	TUNDEBUG	if (tundebug) if_printf
152 
153 #define	TUN_LOCK(tp)		mtx_lock(&(tp)->tun_mtx)
154 #define	TUN_UNLOCK(tp)		mtx_unlock(&(tp)->tun_mtx)
155 #define	TUN_LOCK_ASSERT(tp)	mtx_assert(&(tp)->tun_mtx, MA_OWNED);
156 
157 #define	TUN_VMIO_FLAG_MASK	0x0fff
158 
159 /*
160  * Interface capabilities of a tap device that supports the virtio-net
161  * header.
162  */
163 #define TAP_VNET_HDR_CAPS	(IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6	\
164 				| IFCAP_VLAN_HWCSUM			\
165 				| IFCAP_TSO | IFCAP_LRO			\
166 				| IFCAP_VLAN_HWTSO)
167 
168 #define TAP_ALL_OFFLOAD		(CSUM_TSO | CSUM_TCP | CSUM_UDP |\
169 				    CSUM_TCP_IPV6 | CSUM_UDP_IPV6)
170 
171 /*
172  * All mutable global variables in if_tun are locked using tunmtx, with
173  * the exception of tundebug, which is used unlocked, and the drivers' *clones,
174  * which are static after setup.
175  */
176 static struct mtx tunmtx;
177 static eventhandler_tag arrival_tag;
178 static eventhandler_tag clone_tag;
179 static const char tunname[] = "tun";
180 static const char tapname[] = "tap";
181 static const char vmnetname[] = "vmnet";
182 static MALLOC_DEFINE(M_TUN, tunname, "Tunnel Interface");
183 static int tundebug = 0;
184 static int tundclone = 1;
185 static int tap_allow_uopen = 0;	/* allow user devfs cloning */
186 static int tapuponopen = 0;	/* IFF_UP on open() */
187 static int tapdclone = 1;	/* enable devfs cloning */
188 
189 static TAILQ_HEAD(,tuntap_softc)	tunhead = TAILQ_HEAD_INITIALIZER(tunhead);
190 SYSCTL_INT(_debug, OID_AUTO, if_tun_debug, CTLFLAG_RW, &tundebug, 0, "");
191 
192 static struct sx tun_ioctl_sx;
193 SX_SYSINIT(tun_ioctl_sx, &tun_ioctl_sx, "tun_ioctl");
194 
195 SYSCTL_DECL(_net_link);
196 /* tun */
197 static SYSCTL_NODE(_net_link, OID_AUTO, tun, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
198     "IP tunnel software network interface");
199 SYSCTL_INT(_net_link_tun, OID_AUTO, devfs_cloning, CTLFLAG_RWTUN, &tundclone, 0,
200     "Enable legacy devfs interface creation");
201 
202 /* tap */
203 static SYSCTL_NODE(_net_link, OID_AUTO, tap, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
204     "Ethernet tunnel software network interface");
205 SYSCTL_INT(_net_link_tap, OID_AUTO, user_open, CTLFLAG_RW, &tap_allow_uopen, 0,
206     "Enable legacy devfs interface creation for all users");
207 SYSCTL_INT(_net_link_tap, OID_AUTO, up_on_open, CTLFLAG_RW, &tapuponopen, 0,
208     "Bring interface up when /dev/tap is opened");
209 SYSCTL_INT(_net_link_tap, OID_AUTO, devfs_cloning, CTLFLAG_RWTUN, &tapdclone, 0,
210     "Enable legacy devfs interface creation");
211 SYSCTL_INT(_net_link_tap, OID_AUTO, debug, CTLFLAG_RW, &tundebug, 0, "");
212 
213 static int	tun_create_device(struct tuntap_driver *drv, int unit,
214     struct ucred *cr, struct cdev **dev, const char *name);
215 static int	tun_busy_locked(struct tuntap_softc *tp);
216 static void	tun_unbusy_locked(struct tuntap_softc *tp);
217 static int	tun_busy(struct tuntap_softc *tp);
218 static void	tun_unbusy(struct tuntap_softc *tp);
219 
220 static int	tuntap_name2info(const char *name, int *unit, int *flags);
221 static void	tunclone(void *arg, struct ucred *cred, char *name,
222 		    int namelen, struct cdev **dev);
223 static void	tuncreate(struct cdev *dev);
224 static void	tundtor(void *data);
225 static void	tunrename(void *arg, struct ifnet *ifp);
226 static int	tunifioctl(struct ifnet *, u_long, caddr_t);
227 static void	tuninit(struct ifnet *);
228 static void	tunifinit(void *xtp);
229 static int	tuntapmodevent(module_t, int, void *);
230 static int	tunoutput(struct ifnet *, struct mbuf *,
231 		    const struct sockaddr *, struct route *ro);
232 static void	tunstart(struct ifnet *);
233 static void	tunstart_l2(struct ifnet *);
234 
235 static int	tun_clone_match(struct if_clone *ifc, const char *name);
236 static int	tap_clone_match(struct if_clone *ifc, const char *name);
237 static int	vmnet_clone_match(struct if_clone *ifc, const char *name);
238 static int	tun_clone_create(struct if_clone *, char *, size_t, caddr_t);
239 static int	tun_clone_destroy(struct if_clone *, struct ifnet *);
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_t		*clone_match_fn;
273 	ifc_create_t		*clone_create_fn;
274 	ifc_destroy_t		*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, caddr_t params)
518 {
519 	struct tuntap_driver *drv;
520 	struct cdev *dev;
521 	int err, i, tunflags, unit;
522 
523 	tunflags = 0;
524 	/* The name here tells us exactly what we're creating */
525 	err = tuntap_name2info(name, &unit, &tunflags);
526 	if (err != 0)
527 		return (err);
528 
529 	drv = tuntap_driver_from_flags(tunflags);
530 	if (drv == NULL)
531 		return (ENXIO);
532 
533 	if (unit != -1) {
534 		/* If this unit number is still available that's okay. */
535 		if (alloc_unr_specific(drv->unrhdr, unit) == -1)
536 			return (EEXIST);
537 	} else {
538 		unit = alloc_unr(drv->unrhdr);
539 	}
540 
541 	snprintf(name, IFNAMSIZ, "%s%d", drv->cdevsw.d_name, unit);
542 
543 	/* find any existing device, or allocate new unit number */
544 	dev = NULL;
545 	i = clone_create(&drv->clones, &drv->cdevsw, &unit, &dev, 0);
546 	/* No preexisting struct cdev *, create one */
547 	if (i != 0)
548 		i = tun_create_device(drv, unit, NULL, &dev, name);
549 	if (i == 0)
550 		tuncreate(dev);
551 
552 	return (i);
553 }
554 
555 static void
556 tunclone(void *arg, struct ucred *cred, char *name, int namelen,
557     struct cdev **dev)
558 {
559 	char devname[SPECNAMELEN + 1];
560 	struct tuntap_driver *drv;
561 	int append_unit, i, u, tunflags;
562 	bool mayclone;
563 
564 	if (*dev != NULL)
565 		return;
566 
567 	tunflags = 0;
568 	CURVNET_SET(CRED_TO_VNET(cred));
569 	if (tuntap_name2info(name, &u, &tunflags) != 0)
570 		goto out;	/* Not recognized */
571 
572 	if (u != -1 && u > IF_MAXUNIT)
573 		goto out;	/* Unit number too high */
574 
575 	mayclone = priv_check_cred(cred, PRIV_NET_IFCREATE) == 0;
576 	if ((tunflags & TUN_L2) != 0) {
577 		/* tap/vmnet allow user open with a sysctl */
578 		mayclone = (mayclone || tap_allow_uopen) && tapdclone;
579 	} else {
580 		mayclone = mayclone && tundclone;
581 	}
582 
583 	/*
584 	 * If tun cloning is enabled, only the superuser can create an
585 	 * interface.
586 	 */
587 	if (!mayclone)
588 		goto out;
589 
590 	if (u == -1)
591 		append_unit = 1;
592 	else
593 		append_unit = 0;
594 
595 	drv = tuntap_driver_from_flags(tunflags);
596 	if (drv == NULL)
597 		goto out;
598 
599 	/* find any existing device, or allocate new unit number */
600 	i = clone_create(&drv->clones, &drv->cdevsw, &u, dev, 0);
601 	if (i) {
602 		if (append_unit) {
603 			namelen = snprintf(devname, sizeof(devname), "%s%d",
604 			    name, u);
605 			name = devname;
606 		}
607 
608 		i = tun_create_device(drv, u, cred, dev, name);
609 	}
610 	if (i == 0)
611 		if_clone_create(name, namelen, NULL);
612 out:
613 	CURVNET_RESTORE();
614 }
615 
616 static void
617 tun_destroy(struct tuntap_softc *tp)
618 {
619 
620 	TUN_LOCK(tp);
621 	tp->tun_flags |= TUN_DYING;
622 	if (tp->tun_busy != 0)
623 		cv_wait_unlock(&tp->tun_cv, &tp->tun_mtx);
624 	else
625 		TUN_UNLOCK(tp);
626 
627 	CURVNET_SET(TUN2IFP(tp)->if_vnet);
628 
629 	/* destroy_dev will take care of any alias. */
630 	destroy_dev(tp->tun_dev);
631 	seldrain(&tp->tun_rsel);
632 	knlist_clear(&tp->tun_rsel.si_note, 0);
633 	knlist_destroy(&tp->tun_rsel.si_note);
634 	if ((tp->tun_flags & TUN_L2) != 0) {
635 		ether_ifdetach(TUN2IFP(tp));
636 	} else {
637 		bpfdetach(TUN2IFP(tp));
638 		if_detach(TUN2IFP(tp));
639 	}
640 	sx_xlock(&tun_ioctl_sx);
641 	TUN2IFP(tp)->if_softc = NULL;
642 	sx_xunlock(&tun_ioctl_sx);
643 	free_unr(tp->tun_drv->unrhdr, TUN2IFP(tp)->if_dunit);
644 	if_free(TUN2IFP(tp));
645 	mtx_destroy(&tp->tun_mtx);
646 	cv_destroy(&tp->tun_cv);
647 	free(tp, M_TUN);
648 	CURVNET_RESTORE();
649 }
650 
651 static int
652 tun_clone_destroy(struct if_clone *ifc __unused, struct ifnet *ifp)
653 {
654 	struct tuntap_softc *tp = ifp->if_softc;
655 
656 	mtx_lock(&tunmtx);
657 	TAILQ_REMOVE(&tunhead, tp, tun_list);
658 	mtx_unlock(&tunmtx);
659 	tun_destroy(tp);
660 
661 	return (0);
662 }
663 
664 static void
665 vnet_tun_init(const void *unused __unused)
666 {
667 	struct tuntap_driver *drv;
668 	struct tuntap_driver_cloner *drvc;
669 	int i;
670 
671 	for (i = 0; i < nitems(tuntap_drivers); ++i) {
672 		drv = &tuntap_drivers[i];
673 		drvc = malloc(sizeof(*drvc), M_TUN, M_WAITOK | M_ZERO);
674 
675 		drvc->drv = drv;
676 		drvc->cloner = if_clone_advanced(drv->cdevsw.d_name, 0,
677 		    drv->clone_match_fn, drv->clone_create_fn,
678 		    drv->clone_destroy_fn);
679 		SLIST_INSERT_HEAD(&V_tuntap_driver_cloners, drvc, link);
680 	};
681 }
682 VNET_SYSINIT(vnet_tun_init, SI_SUB_PROTO_IF, SI_ORDER_ANY,
683 		vnet_tun_init, NULL);
684 
685 static void
686 vnet_tun_uninit(const void *unused __unused)
687 {
688 	struct tuntap_driver_cloner *drvc;
689 
690 	while (!SLIST_EMPTY(&V_tuntap_driver_cloners)) {
691 		drvc = SLIST_FIRST(&V_tuntap_driver_cloners);
692 		SLIST_REMOVE_HEAD(&V_tuntap_driver_cloners, link);
693 
694 		if_clone_detach(drvc->cloner);
695 		free(drvc, M_TUN);
696 	}
697 }
698 VNET_SYSUNINIT(vnet_tun_uninit, SI_SUB_PROTO_IF, SI_ORDER_ANY,
699     vnet_tun_uninit, NULL);
700 
701 static void
702 tun_uninit(const void *unused __unused)
703 {
704 	struct tuntap_driver *drv;
705 	struct tuntap_softc *tp;
706 	int i;
707 
708 	EVENTHANDLER_DEREGISTER(ifnet_arrival_event, arrival_tag);
709 	EVENTHANDLER_DEREGISTER(dev_clone, clone_tag);
710 
711 	mtx_lock(&tunmtx);
712 	while ((tp = TAILQ_FIRST(&tunhead)) != NULL) {
713 		TAILQ_REMOVE(&tunhead, tp, tun_list);
714 		mtx_unlock(&tunmtx);
715 		tun_destroy(tp);
716 		mtx_lock(&tunmtx);
717 	}
718 	mtx_unlock(&tunmtx);
719 	for (i = 0; i < nitems(tuntap_drivers); ++i) {
720 		drv = &tuntap_drivers[i];
721 		delete_unrhdr(drv->unrhdr);
722 		clone_cleanup(&drv->clones);
723 	}
724 	mtx_destroy(&tunmtx);
725 }
726 SYSUNINIT(tun_uninit, SI_SUB_PROTO_IF, SI_ORDER_ANY, tun_uninit, NULL);
727 
728 static struct tuntap_driver *
729 tuntap_driver_from_ifnet(const struct ifnet *ifp)
730 {
731 	struct tuntap_driver *drv;
732 	int i;
733 
734 	if (ifp == NULL)
735 		return (NULL);
736 
737 	for (i = 0; i < nitems(tuntap_drivers); ++i) {
738 		drv = &tuntap_drivers[i];
739 		if (strcmp(ifp->if_dname, drv->cdevsw.d_name) == 0)
740 			return (drv);
741 	}
742 
743 	return (NULL);
744 }
745 
746 static int
747 tuntapmodevent(module_t mod, int type, void *data)
748 {
749 	struct tuntap_driver *drv;
750 	int i;
751 
752 	switch (type) {
753 	case MOD_LOAD:
754 		mtx_init(&tunmtx, "tunmtx", NULL, MTX_DEF);
755 		for (i = 0; i < nitems(tuntap_drivers); ++i) {
756 			drv = &tuntap_drivers[i];
757 			clone_setup(&drv->clones);
758 			drv->unrhdr = new_unrhdr(0, IF_MAXUNIT, &tunmtx);
759 		}
760 		arrival_tag = EVENTHANDLER_REGISTER(ifnet_arrival_event,
761 		   tunrename, 0, 1000);
762 		if (arrival_tag == NULL)
763 			return (ENOMEM);
764 		clone_tag = EVENTHANDLER_REGISTER(dev_clone, tunclone, 0, 1000);
765 		if (clone_tag == NULL)
766 			return (ENOMEM);
767 		break;
768 	case MOD_UNLOAD:
769 		/* See tun_uninit, so it's done after the vnet_sysuninit() */
770 		break;
771 	default:
772 		return EOPNOTSUPP;
773 	}
774 	return 0;
775 }
776 
777 static moduledata_t tuntap_mod = {
778 	"if_tuntap",
779 	tuntapmodevent,
780 	0
781 };
782 
783 /* We'll only ever have these two, so no need for a macro. */
784 static moduledata_t tun_mod = { "if_tun", NULL, 0 };
785 static moduledata_t tap_mod = { "if_tap", NULL, 0 };
786 
787 DECLARE_MODULE(if_tuntap, tuntap_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
788 MODULE_VERSION(if_tuntap, 1);
789 DECLARE_MODULE(if_tun, tun_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
790 MODULE_VERSION(if_tun, 1);
791 DECLARE_MODULE(if_tap, tap_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
792 MODULE_VERSION(if_tap, 1);
793 
794 static int
795 tun_create_device(struct tuntap_driver *drv, int unit, struct ucred *cr,
796     struct cdev **dev, const char *name)
797 {
798 	struct make_dev_args args;
799 	struct tuntap_softc *tp;
800 	int error;
801 
802 	tp = malloc(sizeof(*tp), M_TUN, M_WAITOK | M_ZERO);
803 	mtx_init(&tp->tun_mtx, "tun_mtx", NULL, MTX_DEF);
804 	cv_init(&tp->tun_cv, "tun_condvar");
805 	tp->tun_flags = drv->ident_flags;
806 	tp->tun_drv = drv;
807 
808 	make_dev_args_init(&args);
809 	if (cr != NULL)
810 		args.mda_flags = MAKEDEV_REF;
811 	args.mda_devsw = &drv->cdevsw;
812 	args.mda_cr = cr;
813 	args.mda_uid = UID_UUCP;
814 	args.mda_gid = GID_DIALER;
815 	args.mda_mode = 0600;
816 	args.mda_unit = unit;
817 	args.mda_si_drv1 = tp;
818 	error = make_dev_s(&args, dev, "%s", name);
819 	if (error != 0) {
820 		free(tp, M_TUN);
821 		return (error);
822 	}
823 
824 	KASSERT((*dev)->si_drv1 != NULL,
825 	    ("Failed to set si_drv1 at %s creation", name));
826 	tp->tun_dev = *dev;
827 	knlist_init_mtx(&tp->tun_rsel.si_note, &tp->tun_mtx);
828 	mtx_lock(&tunmtx);
829 	TAILQ_INSERT_TAIL(&tunhead, tp, tun_list);
830 	mtx_unlock(&tunmtx);
831 	return (0);
832 }
833 
834 static void
835 tunstart(struct ifnet *ifp)
836 {
837 	struct tuntap_softc *tp = ifp->if_softc;
838 	struct mbuf *m;
839 
840 	TUNDEBUG(ifp, "starting\n");
841 	if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
842 		IFQ_LOCK(&ifp->if_snd);
843 		IFQ_POLL_NOLOCK(&ifp->if_snd, m);
844 		if (m == NULL) {
845 			IFQ_UNLOCK(&ifp->if_snd);
846 			return;
847 		}
848 		IFQ_UNLOCK(&ifp->if_snd);
849 	}
850 
851 	TUN_LOCK(tp);
852 	if (tp->tun_flags & TUN_RWAIT) {
853 		tp->tun_flags &= ~TUN_RWAIT;
854 		wakeup(tp);
855 	}
856 	selwakeuppri(&tp->tun_rsel, PZERO + 1);
857 	KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
858 	if (tp->tun_flags & TUN_ASYNC && tp->tun_sigio) {
859 		TUN_UNLOCK(tp);
860 		pgsigio(&tp->tun_sigio, SIGIO, 0);
861 	} else
862 		TUN_UNLOCK(tp);
863 }
864 
865 /*
866  * tunstart_l2
867  *
868  * queue packets from higher level ready to put out
869  */
870 static void
871 tunstart_l2(struct ifnet *ifp)
872 {
873 	struct tuntap_softc	*tp = ifp->if_softc;
874 
875 	TUNDEBUG(ifp, "starting\n");
876 
877 	/*
878 	 * do not junk pending output if we are in VMnet mode.
879 	 * XXX: can this do any harm because of queue overflow?
880 	 */
881 
882 	TUN_LOCK(tp);
883 	if (((tp->tun_flags & TUN_VMNET) == 0) &&
884 	    ((tp->tun_flags & TUN_READY) != TUN_READY)) {
885 		struct mbuf *m;
886 
887 		/* Unlocked read. */
888 		TUNDEBUG(ifp, "not ready, tun_flags = 0x%x\n", tp->tun_flags);
889 
890 		for (;;) {
891 			IF_DEQUEUE(&ifp->if_snd, m);
892 			if (m != NULL) {
893 				m_freem(m);
894 				if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
895 			} else
896 				break;
897 		}
898 		TUN_UNLOCK(tp);
899 
900 		return;
901 	}
902 
903 	ifp->if_drv_flags |= IFF_DRV_OACTIVE;
904 
905 	if (!IFQ_IS_EMPTY(&ifp->if_snd)) {
906 		if (tp->tun_flags & TUN_RWAIT) {
907 			tp->tun_flags &= ~TUN_RWAIT;
908 			wakeup(tp);
909 		}
910 
911 		if ((tp->tun_flags & TUN_ASYNC) && (tp->tun_sigio != NULL)) {
912 			TUN_UNLOCK(tp);
913 			pgsigio(&tp->tun_sigio, SIGIO, 0);
914 			TUN_LOCK(tp);
915 		}
916 
917 		selwakeuppri(&tp->tun_rsel, PZERO+1);
918 		KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
919 		if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); /* obytes are counted in ether_output */
920 	}
921 
922 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
923 	TUN_UNLOCK(tp);
924 } /* tunstart_l2 */
925 
926 /* XXX: should return an error code so it can fail. */
927 static void
928 tuncreate(struct cdev *dev)
929 {
930 	struct tuntap_driver *drv;
931 	struct tuntap_softc *tp;
932 	struct ifnet *ifp;
933 	struct ether_addr eaddr;
934 	int iflags;
935 	u_char type;
936 
937 	tp = dev->si_drv1;
938 	KASSERT(tp != NULL,
939 	    ("si_drv1 should have been initialized at creation"));
940 
941 	drv = tp->tun_drv;
942 	iflags = IFF_MULTICAST;
943 	if ((tp->tun_flags & TUN_L2) != 0) {
944 		type = IFT_ETHER;
945 		iflags |= IFF_BROADCAST | IFF_SIMPLEX;
946 	} else {
947 		type = IFT_PPP;
948 		iflags |= IFF_POINTOPOINT;
949 	}
950 	ifp = tp->tun_ifp = if_alloc(type);
951 	if (ifp == NULL)
952 		panic("%s%d: failed to if_alloc() interface.\n",
953 		    drv->cdevsw.d_name, dev2unit(dev));
954 	ifp->if_softc = tp;
955 	if_initname(ifp, drv->cdevsw.d_name, dev2unit(dev));
956 	ifp->if_ioctl = tunifioctl;
957 	ifp->if_flags = iflags;
958 	IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen);
959 	ifp->if_capabilities |= IFCAP_LINKSTATE;
960 	ifp->if_capenable |= IFCAP_LINKSTATE;
961 
962 	if ((tp->tun_flags & TUN_L2) != 0) {
963 		ifp->if_init = tunifinit;
964 		ifp->if_start = tunstart_l2;
965 
966 		ether_gen_addr(ifp, &eaddr);
967 		ether_ifattach(ifp, eaddr.octet);
968 	} else {
969 		ifp->if_mtu = TUNMTU;
970 		ifp->if_start = tunstart;
971 		ifp->if_output = tunoutput;
972 
973 		ifp->if_snd.ifq_drv_maxlen = 0;
974 		IFQ_SET_READY(&ifp->if_snd);
975 
976 		if_attach(ifp);
977 		bpfattach(ifp, DLT_NULL, sizeof(u_int32_t));
978 	}
979 
980 	TUN_LOCK(tp);
981 	tp->tun_flags |= TUN_INITED;
982 	TUN_UNLOCK(tp);
983 
984 	TUNDEBUG(ifp, "interface %s is created, minor = %#x\n",
985 	    ifp->if_xname, dev2unit(dev));
986 }
987 
988 static void
989 tunrename(void *arg __unused, struct ifnet *ifp)
990 {
991 	struct tuntap_softc *tp;
992 	int error;
993 
994 	if ((ifp->if_flags & IFF_RENAMING) == 0)
995 		return;
996 
997 	if (tuntap_driver_from_ifnet(ifp) == NULL)
998 		return;
999 
1000 	/*
1001 	 * We need to grab the ioctl sx long enough to make sure the softc is
1002 	 * still there.  If it is, we can safely try to busy the tun device.
1003 	 * The busy may fail if the device is currently dying, in which case
1004 	 * we do nothing.  If it doesn't fail, the busy count stops the device
1005 	 * from dying until we've created the alias (that will then be
1006 	 * subsequently destroyed).
1007 	 */
1008 	sx_xlock(&tun_ioctl_sx);
1009 	tp = ifp->if_softc;
1010 	if (tp == NULL) {
1011 		sx_xunlock(&tun_ioctl_sx);
1012 		return;
1013 	}
1014 	error = tun_busy(tp);
1015 	sx_xunlock(&tun_ioctl_sx);
1016 	if (error != 0)
1017 		return;
1018 	if (tp->tun_alias != NULL) {
1019 		destroy_dev(tp->tun_alias);
1020 		tp->tun_alias = NULL;
1021 	}
1022 
1023 	if (strcmp(ifp->if_xname, tp->tun_dev->si_name) == 0)
1024 		goto out;
1025 
1026 	/*
1027 	 * Failure's ok, aliases are created on a best effort basis.  If a
1028 	 * tun user/consumer decides to rename the interface to conflict with
1029 	 * another device (non-ifnet) on the system, we will assume they know
1030 	 * what they are doing.  make_dev_alias_p won't touch tun_alias on
1031 	 * failure, so we use it but ignore the return value.
1032 	 */
1033 	make_dev_alias_p(MAKEDEV_CHECKNAME, &tp->tun_alias, tp->tun_dev, "%s",
1034 	    ifp->if_xname);
1035 out:
1036 	tun_unbusy(tp);
1037 }
1038 
1039 static int
1040 tunopen(struct cdev *dev, int flag, int mode, struct thread *td)
1041 {
1042 	struct ifnet	*ifp;
1043 	struct tuntap_softc *tp;
1044 	int error __diagused, tunflags;
1045 
1046 	tunflags = 0;
1047 	CURVNET_SET(TD_TO_VNET(td));
1048 	error = tuntap_name2info(dev->si_name, NULL, &tunflags);
1049 	if (error != 0) {
1050 		CURVNET_RESTORE();
1051 		return (error);	/* Shouldn't happen */
1052 	}
1053 
1054 	tp = dev->si_drv1;
1055 	KASSERT(tp != NULL,
1056 	    ("si_drv1 should have been initialized at creation"));
1057 
1058 	TUN_LOCK(tp);
1059 	if ((tp->tun_flags & TUN_INITED) == 0) {
1060 		TUN_UNLOCK(tp);
1061 		CURVNET_RESTORE();
1062 		return (ENXIO);
1063 	}
1064 	if ((tp->tun_flags & (TUN_OPEN | TUN_DYING)) != 0) {
1065 		TUN_UNLOCK(tp);
1066 		CURVNET_RESTORE();
1067 		return (EBUSY);
1068 	}
1069 
1070 	error = tun_busy_locked(tp);
1071 	KASSERT(error == 0, ("Must be able to busy an unopen tunnel"));
1072 	ifp = TUN2IFP(tp);
1073 
1074 	if ((tp->tun_flags & TUN_L2) != 0) {
1075 		bcopy(IF_LLADDR(ifp), tp->tun_ether.octet,
1076 		    sizeof(tp->tun_ether.octet));
1077 
1078 		ifp->if_drv_flags |= IFF_DRV_RUNNING;
1079 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1080 
1081 		if (tapuponopen)
1082 			ifp->if_flags |= IFF_UP;
1083 	}
1084 
1085 	tp->tun_pid = td->td_proc->p_pid;
1086 	tp->tun_flags |= TUN_OPEN;
1087 
1088 	if_link_state_change(ifp, LINK_STATE_UP);
1089 	TUNDEBUG(ifp, "open\n");
1090 	TUN_UNLOCK(tp);
1091 
1092 	/*
1093 	 * This can fail with either ENOENT or EBUSY.  This is in the middle of
1094 	 * d_open, so ENOENT should not be possible.  EBUSY is possible, but
1095 	 * the only cdevpriv dtor being set will be tundtor and the softc being
1096 	 * passed is constant for a given cdev.  We ignore the possible error
1097 	 * because of this as either "unlikely" or "not actually a problem."
1098 	 */
1099 	(void)devfs_set_cdevpriv(tp, tundtor);
1100 	CURVNET_RESTORE();
1101 	return (0);
1102 }
1103 
1104 /*
1105  * tundtor - tear down the device - mark i/f down & delete
1106  * routing info
1107  */
1108 static void
1109 tundtor(void *data)
1110 {
1111 	struct proc *p;
1112 	struct tuntap_softc *tp;
1113 	struct ifnet *ifp;
1114 	bool l2tun;
1115 
1116 	tp = data;
1117 	p = curproc;
1118 	ifp = TUN2IFP(tp);
1119 
1120 	TUN_LOCK(tp);
1121 
1122 	/*
1123 	 * Realistically, we can't be obstinate here.  This only means that the
1124 	 * tuntap device was closed out of order, and the last closer wasn't the
1125 	 * controller.  These are still good to know about, though, as software
1126 	 * should avoid multiple processes with a tuntap device open and
1127 	 * ill-defined transfer of control (e.g., handoff, TUNSIFPID, close in
1128 	 * parent).
1129 	 */
1130 	if (p->p_pid != tp->tun_pid) {
1131 		log(LOG_INFO,
1132 		    "pid %d (%s), %s: tun/tap protocol violation, non-controlling process closed last.\n",
1133 		    p->p_pid, p->p_comm, tp->tun_dev->si_name);
1134 	}
1135 
1136 	/*
1137 	 * junk all pending output
1138 	 */
1139 	CURVNET_SET(ifp->if_vnet);
1140 
1141 	l2tun = false;
1142 	if ((tp->tun_flags & TUN_L2) != 0) {
1143 		l2tun = true;
1144 		IF_DRAIN(&ifp->if_snd);
1145 	} else {
1146 		IFQ_PURGE(&ifp->if_snd);
1147 	}
1148 
1149 	/* For vmnet, we won't do most of the address/route bits */
1150 	if ((tp->tun_flags & TUN_VMNET) != 0 ||
1151 	    (l2tun && (ifp->if_flags & IFF_LINK0) != 0))
1152 		goto out;
1153 
1154 	if (ifp->if_flags & IFF_UP) {
1155 		TUN_UNLOCK(tp);
1156 		if_down(ifp);
1157 		TUN_LOCK(tp);
1158 	}
1159 
1160 	/* Delete all addresses and routes which reference this interface. */
1161 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1162 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1163 		TUN_UNLOCK(tp);
1164 		if_purgeaddrs(ifp);
1165 		TUN_LOCK(tp);
1166 	}
1167 
1168 out:
1169 	if_link_state_change(ifp, LINK_STATE_DOWN);
1170 	CURVNET_RESTORE();
1171 
1172 	funsetown(&tp->tun_sigio);
1173 	selwakeuppri(&tp->tun_rsel, PZERO + 1);
1174 	KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
1175 	TUNDEBUG (ifp, "closed\n");
1176 	tp->tun_flags &= ~TUN_OPEN;
1177 	tp->tun_pid = 0;
1178 	tun_vnethdr_set(ifp, 0);
1179 
1180 	tun_unbusy_locked(tp);
1181 	TUN_UNLOCK(tp);
1182 }
1183 
1184 static void
1185 tuninit(struct ifnet *ifp)
1186 {
1187 	struct tuntap_softc *tp = ifp->if_softc;
1188 
1189 	TUNDEBUG(ifp, "tuninit\n");
1190 
1191 	TUN_LOCK(tp);
1192 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1193 	if ((tp->tun_flags & TUN_L2) == 0) {
1194 		ifp->if_flags |= IFF_UP;
1195 		getmicrotime(&ifp->if_lastchange);
1196 		TUN_UNLOCK(tp);
1197 	} else {
1198 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1199 		TUN_UNLOCK(tp);
1200 		/* attempt to start output */
1201 		tunstart_l2(ifp);
1202 	}
1203 
1204 }
1205 
1206 /*
1207  * Used only for l2 tunnel.
1208  */
1209 static void
1210 tunifinit(void *xtp)
1211 {
1212 	struct tuntap_softc *tp;
1213 
1214 	tp = (struct tuntap_softc *)xtp;
1215 	tuninit(tp->tun_ifp);
1216 }
1217 
1218 /*
1219  * To be called under TUN_LOCK. Update ifp->if_hwassist according to the
1220  * current value of ifp->if_capenable.
1221  */
1222 static void
1223 tun_caps_changed(struct ifnet *ifp)
1224 {
1225 	uint64_t hwassist = 0;
1226 
1227 	TUN_LOCK_ASSERT((struct tuntap_softc *)ifp->if_softc);
1228 	if (ifp->if_capenable & IFCAP_TXCSUM)
1229 		hwassist |= CSUM_TCP | CSUM_UDP;
1230 	if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
1231 		hwassist |= CSUM_TCP_IPV6
1232 		    | CSUM_UDP_IPV6;
1233 	if (ifp->if_capenable & IFCAP_TSO4)
1234 		hwassist |= CSUM_IP_TSO;
1235 	if (ifp->if_capenable & IFCAP_TSO6)
1236 		hwassist |= CSUM_IP6_TSO;
1237 	ifp->if_hwassist = hwassist;
1238 }
1239 
1240 /*
1241  * To be called under TUN_LOCK. Update tp->tun_vhdrlen and adjust
1242  * if_capabilities and if_capenable as needed.
1243  */
1244 static void
1245 tun_vnethdr_set(struct ifnet *ifp, int vhdrlen)
1246 {
1247 	struct tuntap_softc *tp = ifp->if_softc;
1248 
1249 	TUN_LOCK_ASSERT(tp);
1250 
1251 	if (tp->tun_vhdrlen == vhdrlen)
1252 		return;
1253 
1254 	/*
1255 	 * Update if_capabilities to reflect the
1256 	 * functionalities offered by the virtio-net
1257 	 * header.
1258 	 */
1259 	if (vhdrlen != 0)
1260 		ifp->if_capabilities |=
1261 			TAP_VNET_HDR_CAPS;
1262 	else
1263 		ifp->if_capabilities &=
1264 			~TAP_VNET_HDR_CAPS;
1265 	/*
1266 	 * Disable any capabilities that we don't
1267 	 * support anymore.
1268 	 */
1269 	ifp->if_capenable &= ifp->if_capabilities;
1270 	tun_caps_changed(ifp);
1271 	tp->tun_vhdrlen = vhdrlen;
1272 
1273 	TUNDEBUG(ifp, "vnet_hdr_len=%d, if_capabilities=%x\n",
1274 	    vhdrlen, ifp->if_capabilities);
1275 }
1276 
1277 /*
1278  * Process an ioctl request.
1279  */
1280 static int
1281 tunifioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1282 {
1283 	struct ifreq *ifr = (struct ifreq *)data;
1284 	struct tuntap_softc *tp;
1285 	struct ifstat *ifs;
1286 	struct ifmediareq	*ifmr;
1287 	int		dummy, error = 0;
1288 	bool		l2tun;
1289 
1290 	ifmr = NULL;
1291 	sx_xlock(&tun_ioctl_sx);
1292 	tp = ifp->if_softc;
1293 	if (tp == NULL) {
1294 		error = ENXIO;
1295 		goto bad;
1296 	}
1297 	l2tun = (tp->tun_flags & TUN_L2) != 0;
1298 	switch(cmd) {
1299 	case SIOCGIFSTATUS:
1300 		ifs = (struct ifstat *)data;
1301 		TUN_LOCK(tp);
1302 		if (tp->tun_pid)
1303 			snprintf(ifs->ascii, sizeof(ifs->ascii),
1304 			    "\tOpened by PID %d\n", tp->tun_pid);
1305 		else
1306 			ifs->ascii[0] = '\0';
1307 		TUN_UNLOCK(tp);
1308 		break;
1309 	case SIOCSIFADDR:
1310 		if (l2tun)
1311 			error = ether_ioctl(ifp, cmd, data);
1312 		else
1313 			tuninit(ifp);
1314 		if (error == 0)
1315 		    TUNDEBUG(ifp, "address set\n");
1316 		break;
1317 	case SIOCSIFMTU:
1318 		ifp->if_mtu = ifr->ifr_mtu;
1319 		TUNDEBUG(ifp, "mtu set\n");
1320 		break;
1321 	case SIOCSIFFLAGS:
1322 	case SIOCADDMULTI:
1323 	case SIOCDELMULTI:
1324 		break;
1325 	case SIOCGIFMEDIA:
1326 		if (!l2tun) {
1327 			error = EINVAL;
1328 			break;
1329 		}
1330 
1331 		ifmr = (struct ifmediareq *)data;
1332 		dummy = ifmr->ifm_count;
1333 		ifmr->ifm_count = 1;
1334 		ifmr->ifm_status = IFM_AVALID;
1335 		ifmr->ifm_active = IFM_ETHER;
1336 		if (tp->tun_flags & TUN_OPEN)
1337 			ifmr->ifm_status |= IFM_ACTIVE;
1338 		ifmr->ifm_current = ifmr->ifm_active;
1339 		if (dummy >= 1) {
1340 			int media = IFM_ETHER;
1341 			error = copyout(&media, ifmr->ifm_ulist, sizeof(int));
1342 		}
1343 		break;
1344 	case SIOCSIFCAP:
1345 		TUN_LOCK(tp);
1346 		ifp->if_capenable = ifr->ifr_reqcap;
1347 		tun_caps_changed(ifp);
1348 		TUN_UNLOCK(tp);
1349 		VLAN_CAPABILITIES(ifp);
1350 		break;
1351 	default:
1352 		if (l2tun) {
1353 			error = ether_ioctl(ifp, cmd, data);
1354 		} else {
1355 			error = EINVAL;
1356 		}
1357 	}
1358 bad:
1359 	sx_xunlock(&tun_ioctl_sx);
1360 	return (error);
1361 }
1362 
1363 /*
1364  * tunoutput - queue packets from higher level ready to put out.
1365  */
1366 static int
1367 tunoutput(struct ifnet *ifp, struct mbuf *m0, const struct sockaddr *dst,
1368     struct route *ro)
1369 {
1370 	struct tuntap_softc *tp = ifp->if_softc;
1371 	u_short cached_tun_flags;
1372 	int error;
1373 	u_int32_t af;
1374 
1375 	TUNDEBUG (ifp, "tunoutput\n");
1376 
1377 #ifdef MAC
1378 	error = mac_ifnet_check_transmit(ifp, m0);
1379 	if (error) {
1380 		m_freem(m0);
1381 		return (error);
1382 	}
1383 #endif
1384 
1385 	/* Could be unlocked read? */
1386 	TUN_LOCK(tp);
1387 	cached_tun_flags = tp->tun_flags;
1388 	TUN_UNLOCK(tp);
1389 	if ((cached_tun_flags & TUN_READY) != TUN_READY) {
1390 		TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags);
1391 		m_freem (m0);
1392 		return (EHOSTDOWN);
1393 	}
1394 
1395 	if ((ifp->if_flags & IFF_UP) != IFF_UP) {
1396 		m_freem (m0);
1397 		return (EHOSTDOWN);
1398 	}
1399 
1400 	/* BPF writes need to be handled specially. */
1401 	if (dst->sa_family == AF_UNSPEC)
1402 		bcopy(dst->sa_data, &af, sizeof(af));
1403 	else
1404 		af = RO_GET_FAMILY(ro, dst);
1405 
1406 	if (bpf_peers_present(ifp->if_bpf))
1407 		bpf_mtap2(ifp->if_bpf, &af, sizeof(af), m0);
1408 
1409 	/* prepend sockaddr? this may abort if the mbuf allocation fails */
1410 	if (cached_tun_flags & TUN_LMODE) {
1411 		/* allocate space for sockaddr */
1412 		M_PREPEND(m0, dst->sa_len, M_NOWAIT);
1413 
1414 		/* if allocation failed drop packet */
1415 		if (m0 == NULL) {
1416 			if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1417 			if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1418 			return (ENOBUFS);
1419 		} else {
1420 			bcopy(dst, m0->m_data, dst->sa_len);
1421 		}
1422 	}
1423 
1424 	if (cached_tun_flags & TUN_IFHEAD) {
1425 		/* Prepend the address family */
1426 		M_PREPEND(m0, 4, M_NOWAIT);
1427 
1428 		/* if allocation failed drop packet */
1429 		if (m0 == NULL) {
1430 			if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1431 			if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1432 			return (ENOBUFS);
1433 		} else
1434 			*(u_int32_t *)m0->m_data = htonl(af);
1435 	} else {
1436 #ifdef INET
1437 		if (af != AF_INET)
1438 #endif
1439 		{
1440 			m_freem(m0);
1441 			return (EAFNOSUPPORT);
1442 		}
1443 	}
1444 
1445 	error = (ifp->if_transmit)(ifp, m0);
1446 	if (error)
1447 		return (ENOBUFS);
1448 	if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1449 	return (0);
1450 }
1451 
1452 /*
1453  * the cdevsw interface is now pretty minimal.
1454  */
1455 static	int
1456 tunioctl(struct cdev *dev, u_long cmd, caddr_t data, int flag,
1457     struct thread *td)
1458 {
1459 	struct ifreq ifr, *ifrp;
1460 	struct tuntap_softc *tp = dev->si_drv1;
1461 	struct ifnet *ifp = TUN2IFP(tp);
1462 	struct tuninfo *tunp;
1463 	int error, iflags, ival;
1464 	bool	l2tun;
1465 
1466 	l2tun = (tp->tun_flags & TUN_L2) != 0;
1467 	if (l2tun) {
1468 		/* tap specific ioctls */
1469 		switch(cmd) {
1470 		/* VMware/VMnet port ioctl's */
1471 #if defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD5) || \
1472     defined(COMPAT_FREEBSD4)
1473 		case _IO('V', 0):
1474 			ival = IOCPARM_IVAL(data);
1475 			data = (caddr_t)&ival;
1476 			/* FALLTHROUGH */
1477 #endif
1478 		case VMIO_SIOCSIFFLAGS: /* VMware/VMnet SIOCSIFFLAGS */
1479 			iflags = *(int *)data;
1480 			iflags &= TUN_VMIO_FLAG_MASK;
1481 			iflags &= ~IFF_CANTCHANGE;
1482 			iflags |= IFF_UP;
1483 
1484 			TUN_LOCK(tp);
1485 			ifp->if_flags = iflags |
1486 			    (ifp->if_flags & IFF_CANTCHANGE);
1487 			TUN_UNLOCK(tp);
1488 
1489 			return (0);
1490 		case SIOCGIFADDR:	/* get MAC address of the remote side */
1491 			TUN_LOCK(tp);
1492 			bcopy(&tp->tun_ether.octet, data,
1493 			    sizeof(tp->tun_ether.octet));
1494 			TUN_UNLOCK(tp);
1495 
1496 			return (0);
1497 		case SIOCSIFADDR:	/* set MAC address of the remote side */
1498 			TUN_LOCK(tp);
1499 			bcopy(data, &tp->tun_ether.octet,
1500 			    sizeof(tp->tun_ether.octet));
1501 			TUN_UNLOCK(tp);
1502 
1503 			return (0);
1504 		case TAPSVNETHDR:
1505 			ival = *(int *)data;
1506 			if (ival != 0 &&
1507 			    ival != sizeof(struct virtio_net_hdr) &&
1508 			    ival != sizeof(struct virtio_net_hdr_mrg_rxbuf)) {
1509 				return (EINVAL);
1510 			}
1511 			TUN_LOCK(tp);
1512 			tun_vnethdr_set(ifp, ival);
1513 			TUN_UNLOCK(tp);
1514 
1515 			return (0);
1516 		case TAPGVNETHDR:
1517 			TUN_LOCK(tp);
1518 			*(int *)data = tp->tun_vhdrlen;
1519 			TUN_UNLOCK(tp);
1520 
1521 			return (0);
1522 		}
1523 
1524 		/* Fall through to the common ioctls if unhandled */
1525 	} else {
1526 		switch (cmd) {
1527 		case TUNSLMODE:
1528 			TUN_LOCK(tp);
1529 			if (*(int *)data) {
1530 				tp->tun_flags |= TUN_LMODE;
1531 				tp->tun_flags &= ~TUN_IFHEAD;
1532 			} else
1533 				tp->tun_flags &= ~TUN_LMODE;
1534 			TUN_UNLOCK(tp);
1535 
1536 			return (0);
1537 		case TUNSIFHEAD:
1538 			TUN_LOCK(tp);
1539 			if (*(int *)data) {
1540 				tp->tun_flags |= TUN_IFHEAD;
1541 				tp->tun_flags &= ~TUN_LMODE;
1542 			} else
1543 				tp->tun_flags &= ~TUN_IFHEAD;
1544 			TUN_UNLOCK(tp);
1545 
1546 			return (0);
1547 		case TUNGIFHEAD:
1548 			TUN_LOCK(tp);
1549 			*(int *)data = (tp->tun_flags & TUN_IFHEAD) ? 1 : 0;
1550 			TUN_UNLOCK(tp);
1551 
1552 			return (0);
1553 		case TUNSIFMODE:
1554 			/* deny this if UP */
1555 			if (TUN2IFP(tp)->if_flags & IFF_UP)
1556 				return (EBUSY);
1557 
1558 			switch (*(int *)data & ~IFF_MULTICAST) {
1559 			case IFF_POINTOPOINT:
1560 			case IFF_BROADCAST:
1561 				TUN_LOCK(tp);
1562 				TUN2IFP(tp)->if_flags &=
1563 				    ~(IFF_BROADCAST|IFF_POINTOPOINT|IFF_MULTICAST);
1564 				TUN2IFP(tp)->if_flags |= *(int *)data;
1565 				TUN_UNLOCK(tp);
1566 
1567 				break;
1568 			default:
1569 				return (EINVAL);
1570 			}
1571 
1572 			return (0);
1573 		case TUNSIFPID:
1574 			TUN_LOCK(tp);
1575 			tp->tun_pid = curthread->td_proc->p_pid;
1576 			TUN_UNLOCK(tp);
1577 
1578 			return (0);
1579 		}
1580 		/* Fall through to the common ioctls if unhandled */
1581 	}
1582 
1583 	switch (cmd) {
1584 	case TUNGIFNAME:
1585 		ifrp = (struct ifreq *)data;
1586 		strlcpy(ifrp->ifr_name, TUN2IFP(tp)->if_xname, IFNAMSIZ);
1587 
1588 		return (0);
1589 	case TUNSIFINFO:
1590 		tunp = (struct tuninfo *)data;
1591 		if (TUN2IFP(tp)->if_type != tunp->type)
1592 			return (EPROTOTYPE);
1593 		TUN_LOCK(tp);
1594 		if (TUN2IFP(tp)->if_mtu != tunp->mtu) {
1595 			strlcpy(ifr.ifr_name, if_name(TUN2IFP(tp)), IFNAMSIZ);
1596 			ifr.ifr_mtu = tunp->mtu;
1597 			CURVNET_SET(TUN2IFP(tp)->if_vnet);
1598 			error = ifhwioctl(SIOCSIFMTU, TUN2IFP(tp),
1599 			    (caddr_t)&ifr, td);
1600 			CURVNET_RESTORE();
1601 			if (error) {
1602 				TUN_UNLOCK(tp);
1603 				return (error);
1604 			}
1605 		}
1606 		TUN2IFP(tp)->if_baudrate = tunp->baudrate;
1607 		TUN_UNLOCK(tp);
1608 		break;
1609 	case TUNGIFINFO:
1610 		tunp = (struct tuninfo *)data;
1611 		TUN_LOCK(tp);
1612 		tunp->mtu = TUN2IFP(tp)->if_mtu;
1613 		tunp->type = TUN2IFP(tp)->if_type;
1614 		tunp->baudrate = TUN2IFP(tp)->if_baudrate;
1615 		TUN_UNLOCK(tp);
1616 		break;
1617 	case TUNSDEBUG:
1618 		tundebug = *(int *)data;
1619 		break;
1620 	case TUNGDEBUG:
1621 		*(int *)data = tundebug;
1622 		break;
1623 	case FIONBIO:
1624 		break;
1625 	case FIOASYNC:
1626 		TUN_LOCK(tp);
1627 		if (*(int *)data)
1628 			tp->tun_flags |= TUN_ASYNC;
1629 		else
1630 			tp->tun_flags &= ~TUN_ASYNC;
1631 		TUN_UNLOCK(tp);
1632 		break;
1633 	case FIONREAD:
1634 		if (!IFQ_IS_EMPTY(&TUN2IFP(tp)->if_snd)) {
1635 			struct mbuf *mb;
1636 			IFQ_LOCK(&TUN2IFP(tp)->if_snd);
1637 			IFQ_POLL_NOLOCK(&TUN2IFP(tp)->if_snd, mb);
1638 			for (*(int *)data = 0; mb != NULL; mb = mb->m_next)
1639 				*(int *)data += mb->m_len;
1640 			IFQ_UNLOCK(&TUN2IFP(tp)->if_snd);
1641 		} else
1642 			*(int *)data = 0;
1643 		break;
1644 	case FIOSETOWN:
1645 		return (fsetown(*(int *)data, &tp->tun_sigio));
1646 
1647 	case FIOGETOWN:
1648 		*(int *)data = fgetown(&tp->tun_sigio);
1649 		return (0);
1650 
1651 	/* This is deprecated, FIOSETOWN should be used instead. */
1652 	case TIOCSPGRP:
1653 		return (fsetown(-(*(int *)data), &tp->tun_sigio));
1654 
1655 	/* This is deprecated, FIOGETOWN should be used instead. */
1656 	case TIOCGPGRP:
1657 		*(int *)data = -fgetown(&tp->tun_sigio);
1658 		return (0);
1659 
1660 	default:
1661 		return (ENOTTY);
1662 	}
1663 	return (0);
1664 }
1665 
1666 /*
1667  * The cdevsw read interface - reads a packet at a time, or at
1668  * least as much of a packet as can be read.
1669  */
1670 static	int
1671 tunread(struct cdev *dev, struct uio *uio, int flag)
1672 {
1673 	struct tuntap_softc *tp = dev->si_drv1;
1674 	struct ifnet	*ifp = TUN2IFP(tp);
1675 	struct mbuf	*m;
1676 	size_t		len;
1677 	int		error = 0;
1678 
1679 	TUNDEBUG (ifp, "read\n");
1680 	TUN_LOCK(tp);
1681 	if ((tp->tun_flags & TUN_READY) != TUN_READY) {
1682 		TUN_UNLOCK(tp);
1683 		TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags);
1684 		return (EHOSTDOWN);
1685 	}
1686 
1687 	tp->tun_flags &= ~TUN_RWAIT;
1688 
1689 	for (;;) {
1690 		IFQ_DEQUEUE(&ifp->if_snd, m);
1691 		if (m != NULL)
1692 			break;
1693 		if (flag & O_NONBLOCK) {
1694 			TUN_UNLOCK(tp);
1695 			return (EWOULDBLOCK);
1696 		}
1697 		tp->tun_flags |= TUN_RWAIT;
1698 		error = mtx_sleep(tp, &tp->tun_mtx, PCATCH | (PZERO + 1),
1699 		    "tunread", 0);
1700 		if (error != 0) {
1701 			TUN_UNLOCK(tp);
1702 			return (error);
1703 		}
1704 	}
1705 	TUN_UNLOCK(tp);
1706 
1707 	if ((tp->tun_flags & TUN_L2) != 0)
1708 		BPF_MTAP(ifp, m);
1709 
1710 	len = min(tp->tun_vhdrlen, uio->uio_resid);
1711 	if (len > 0) {
1712 		struct virtio_net_hdr_mrg_rxbuf vhdr;
1713 
1714 		bzero(&vhdr, sizeof(vhdr));
1715 		if (m->m_pkthdr.csum_flags & TAP_ALL_OFFLOAD) {
1716 			m = virtio_net_tx_offload(ifp, m, false, &vhdr.hdr);
1717 		}
1718 
1719 		TUNDEBUG(ifp, "txvhdr: f %u, gt %u, hl %u, "
1720 		    "gs %u, cs %u, co %u\n", vhdr.hdr.flags,
1721 		    vhdr.hdr.gso_type, vhdr.hdr.hdr_len,
1722 		    vhdr.hdr.gso_size, vhdr.hdr.csum_start,
1723 		    vhdr.hdr.csum_offset);
1724 		error = uiomove(&vhdr, len, uio);
1725 	}
1726 
1727 	while (m && uio->uio_resid > 0 && error == 0) {
1728 		len = min(uio->uio_resid, m->m_len);
1729 		if (len != 0)
1730 			error = uiomove(mtod(m, void *), len, uio);
1731 		m = m_free(m);
1732 	}
1733 
1734 	if (m) {
1735 		TUNDEBUG(ifp, "Dropping mbuf\n");
1736 		m_freem(m);
1737 	}
1738 	return (error);
1739 }
1740 
1741 static int
1742 tunwrite_l2(struct tuntap_softc *tp, struct mbuf *m,
1743 	    struct virtio_net_hdr_mrg_rxbuf *vhdr)
1744 {
1745 	struct epoch_tracker et;
1746 	struct ether_header *eh;
1747 	struct ifnet *ifp;
1748 
1749 	ifp = TUN2IFP(tp);
1750 
1751 	/*
1752 	 * Only pass a unicast frame to ether_input(), if it would
1753 	 * actually have been received by non-virtual hardware.
1754 	 */
1755 	if (m->m_len < sizeof(struct ether_header)) {
1756 		m_freem(m);
1757 		return (0);
1758 	}
1759 
1760 	eh = mtod(m, struct ether_header *);
1761 
1762 	if (eh && (ifp->if_flags & IFF_PROMISC) == 0 &&
1763 	    !ETHER_IS_MULTICAST(eh->ether_dhost) &&
1764 	    bcmp(eh->ether_dhost, IF_LLADDR(ifp), ETHER_ADDR_LEN) != 0) {
1765 		m_freem(m);
1766 		return (0);
1767 	}
1768 
1769 	if (vhdr != NULL && virtio_net_rx_csum(m, &vhdr->hdr)) {
1770 		m_freem(m);
1771 		return (0);
1772 	}
1773 
1774 	/* Pass packet up to parent. */
1775 	CURVNET_SET(ifp->if_vnet);
1776 	NET_EPOCH_ENTER(et);
1777 	(*ifp->if_input)(ifp, m);
1778 	NET_EPOCH_EXIT(et);
1779 	CURVNET_RESTORE();
1780 	/* ibytes are counted in parent */
1781 	if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1782 	return (0);
1783 }
1784 
1785 static int
1786 tunwrite_l3(struct tuntap_softc *tp, struct mbuf *m)
1787 {
1788 	struct epoch_tracker et;
1789 	struct ifnet *ifp;
1790 	int family, isr;
1791 
1792 	ifp = TUN2IFP(tp);
1793 	/* Could be unlocked read? */
1794 	TUN_LOCK(tp);
1795 	if (tp->tun_flags & TUN_IFHEAD) {
1796 		TUN_UNLOCK(tp);
1797 		if (m->m_len < sizeof(family) &&
1798 		(m = m_pullup(m, sizeof(family))) == NULL)
1799 			return (ENOBUFS);
1800 		family = ntohl(*mtod(m, u_int32_t *));
1801 		m_adj(m, sizeof(family));
1802 	} else {
1803 		TUN_UNLOCK(tp);
1804 		family = AF_INET;
1805 	}
1806 
1807 	BPF_MTAP2(ifp, &family, sizeof(family), m);
1808 
1809 	switch (family) {
1810 #ifdef INET
1811 	case AF_INET:
1812 		isr = NETISR_IP;
1813 		break;
1814 #endif
1815 #ifdef INET6
1816 	case AF_INET6:
1817 		isr = NETISR_IPV6;
1818 		break;
1819 #endif
1820 	default:
1821 		m_freem(m);
1822 		return (EAFNOSUPPORT);
1823 	}
1824 	random_harvest_queue(m, sizeof(*m), RANDOM_NET_TUN);
1825 	if_inc_counter(ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
1826 	if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1827 	CURVNET_SET(ifp->if_vnet);
1828 	M_SETFIB(m, ifp->if_fib);
1829 	NET_EPOCH_ENTER(et);
1830 	netisr_dispatch(isr, m);
1831 	NET_EPOCH_EXIT(et);
1832 	CURVNET_RESTORE();
1833 	return (0);
1834 }
1835 
1836 /*
1837  * the cdevsw write interface - an atomic write is a packet - or else!
1838  */
1839 static	int
1840 tunwrite(struct cdev *dev, struct uio *uio, int flag)
1841 {
1842 	struct virtio_net_hdr_mrg_rxbuf vhdr;
1843 	struct tuntap_softc *tp;
1844 	struct ifnet	*ifp;
1845 	struct mbuf	*m;
1846 	uint32_t	mru;
1847 	int		align, vhdrlen, error;
1848 	bool		l2tun;
1849 
1850 	tp = dev->si_drv1;
1851 	ifp = TUN2IFP(tp);
1852 	TUNDEBUG(ifp, "tunwrite\n");
1853 	if ((ifp->if_flags & IFF_UP) != IFF_UP)
1854 		/* ignore silently */
1855 		return (0);
1856 
1857 	if (uio->uio_resid == 0)
1858 		return (0);
1859 
1860 	l2tun = (tp->tun_flags & TUN_L2) != 0;
1861 	mru = l2tun ? TAPMRU : TUNMRU;
1862 	vhdrlen = tp->tun_vhdrlen;
1863 	align = 0;
1864 	if (l2tun) {
1865 		align = ETHER_ALIGN;
1866 		mru += vhdrlen;
1867 	} else if ((tp->tun_flags & TUN_IFHEAD) != 0)
1868 		mru += sizeof(uint32_t);	/* family */
1869 	if (uio->uio_resid < 0 || uio->uio_resid > mru) {
1870 		TUNDEBUG(ifp, "len=%zd!\n", uio->uio_resid);
1871 		return (EIO);
1872 	}
1873 
1874 	if (vhdrlen > 0) {
1875 		error = uiomove(&vhdr, vhdrlen, uio);
1876 		if (error != 0)
1877 			return (error);
1878 		TUNDEBUG(ifp, "txvhdr: f %u, gt %u, hl %u, "
1879 		    "gs %u, cs %u, co %u\n", vhdr.hdr.flags,
1880 		    vhdr.hdr.gso_type, vhdr.hdr.hdr_len,
1881 		    vhdr.hdr.gso_size, vhdr.hdr.csum_start,
1882 		    vhdr.hdr.csum_offset);
1883 	}
1884 
1885 	if ((m = m_uiotombuf(uio, M_NOWAIT, 0, align, M_PKTHDR)) == NULL) {
1886 		if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1887 		return (ENOBUFS);
1888 	}
1889 
1890 	m->m_pkthdr.rcvif = ifp;
1891 #ifdef MAC
1892 	mac_ifnet_create_mbuf(ifp, m);
1893 #endif
1894 
1895 	if (l2tun)
1896 		return (tunwrite_l2(tp, m, vhdrlen > 0 ? &vhdr : NULL));
1897 
1898 	return (tunwrite_l3(tp, m));
1899 }
1900 
1901 /*
1902  * tunpoll - the poll interface, this is only useful on reads
1903  * really. The write detect always returns true, write never blocks
1904  * anyway, it either accepts the packet or drops it.
1905  */
1906 static	int
1907 tunpoll(struct cdev *dev, int events, struct thread *td)
1908 {
1909 	struct tuntap_softc *tp = dev->si_drv1;
1910 	struct ifnet	*ifp = TUN2IFP(tp);
1911 	int		revents = 0;
1912 
1913 	TUNDEBUG(ifp, "tunpoll\n");
1914 
1915 	if (events & (POLLIN | POLLRDNORM)) {
1916 		IFQ_LOCK(&ifp->if_snd);
1917 		if (!IFQ_IS_EMPTY(&ifp->if_snd)) {
1918 			TUNDEBUG(ifp, "tunpoll q=%d\n", ifp->if_snd.ifq_len);
1919 			revents |= events & (POLLIN | POLLRDNORM);
1920 		} else {
1921 			TUNDEBUG(ifp, "tunpoll waiting\n");
1922 			selrecord(td, &tp->tun_rsel);
1923 		}
1924 		IFQ_UNLOCK(&ifp->if_snd);
1925 	}
1926 	revents |= events & (POLLOUT | POLLWRNORM);
1927 
1928 	return (revents);
1929 }
1930 
1931 /*
1932  * tunkqfilter - support for the kevent() system call.
1933  */
1934 static int
1935 tunkqfilter(struct cdev *dev, struct knote *kn)
1936 {
1937 	struct tuntap_softc	*tp = dev->si_drv1;
1938 	struct ifnet	*ifp = TUN2IFP(tp);
1939 
1940 	switch(kn->kn_filter) {
1941 	case EVFILT_READ:
1942 		TUNDEBUG(ifp, "%s kqfilter: EVFILT_READ, minor = %#x\n",
1943 		    ifp->if_xname, dev2unit(dev));
1944 		kn->kn_fop = &tun_read_filterops;
1945 		break;
1946 
1947 	case EVFILT_WRITE:
1948 		TUNDEBUG(ifp, "%s kqfilter: EVFILT_WRITE, minor = %#x\n",
1949 		    ifp->if_xname, dev2unit(dev));
1950 		kn->kn_fop = &tun_write_filterops;
1951 		break;
1952 
1953 	default:
1954 		TUNDEBUG(ifp, "%s kqfilter: invalid filter, minor = %#x\n",
1955 		    ifp->if_xname, dev2unit(dev));
1956 		return(EINVAL);
1957 	}
1958 
1959 	kn->kn_hook = tp;
1960 	knlist_add(&tp->tun_rsel.si_note, kn, 0);
1961 
1962 	return (0);
1963 }
1964 
1965 /*
1966  * Return true of there is data in the interface queue.
1967  */
1968 static int
1969 tunkqread(struct knote *kn, long hint)
1970 {
1971 	int			ret;
1972 	struct tuntap_softc	*tp = kn->kn_hook;
1973 	struct cdev		*dev = tp->tun_dev;
1974 	struct ifnet	*ifp = TUN2IFP(tp);
1975 
1976 	if ((kn->kn_data = ifp->if_snd.ifq_len) > 0) {
1977 		TUNDEBUG(ifp,
1978 		    "%s have data in the queue.  Len = %d, minor = %#x\n",
1979 		    ifp->if_xname, ifp->if_snd.ifq_len, dev2unit(dev));
1980 		ret = 1;
1981 	} else {
1982 		TUNDEBUG(ifp,
1983 		    "%s waiting for data, minor = %#x\n", ifp->if_xname,
1984 		    dev2unit(dev));
1985 		ret = 0;
1986 	}
1987 
1988 	return (ret);
1989 }
1990 
1991 /*
1992  * Always can write, always return MTU in kn->data.
1993  */
1994 static int
1995 tunkqwrite(struct knote *kn, long hint)
1996 {
1997 	struct tuntap_softc	*tp = kn->kn_hook;
1998 	struct ifnet	*ifp = TUN2IFP(tp);
1999 
2000 	kn->kn_data = ifp->if_mtu;
2001 
2002 	return (1);
2003 }
2004 
2005 static void
2006 tunkqdetach(struct knote *kn)
2007 {
2008 	struct tuntap_softc	*tp = kn->kn_hook;
2009 
2010 	knlist_remove(&tp->tun_rsel.si_note, kn, 0);
2011 }
2012