xref: /freebsd/sys/net/if_tuntap.c (revision ec994981447e8a974426660b5071bc405280af73)
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 		tuncreate(dev);
552 		struct tuntap_softc *tp = dev->si_drv1;
553 		*ifpp = tp->tun_ifp;
554 	}
555 
556 	return (i);
557 }
558 
559 static void
560 tunclone(void *arg, struct ucred *cred, char *name, int namelen,
561     struct cdev **dev)
562 {
563 	char devname[SPECNAMELEN + 1];
564 	struct tuntap_driver *drv;
565 	int append_unit, i, u, tunflags;
566 	bool mayclone;
567 
568 	if (*dev != NULL)
569 		return;
570 
571 	tunflags = 0;
572 	CURVNET_SET(CRED_TO_VNET(cred));
573 	if (tuntap_name2info(name, &u, &tunflags) != 0)
574 		goto out;	/* Not recognized */
575 
576 	if (u != -1 && u > IF_MAXUNIT)
577 		goto out;	/* Unit number too high */
578 
579 	mayclone = priv_check_cred(cred, PRIV_NET_IFCREATE) == 0;
580 	if ((tunflags & TUN_L2) != 0) {
581 		/* tap/vmnet allow user open with a sysctl */
582 		mayclone = (mayclone || tap_allow_uopen) && tapdclone;
583 	} else {
584 		mayclone = mayclone && tundclone;
585 	}
586 
587 	/*
588 	 * If tun cloning is enabled, only the superuser can create an
589 	 * interface.
590 	 */
591 	if (!mayclone)
592 		goto out;
593 
594 	if (u == -1)
595 		append_unit = 1;
596 	else
597 		append_unit = 0;
598 
599 	drv = tuntap_driver_from_flags(tunflags);
600 	if (drv == NULL)
601 		goto out;
602 
603 	/* find any existing device, or allocate new unit number */
604 	i = clone_create(&drv->clones, &drv->cdevsw, &u, dev, 0);
605 	if (i) {
606 		if (append_unit) {
607 			namelen = snprintf(devname, sizeof(devname), "%s%d",
608 			    name, u);
609 			name = devname;
610 		}
611 
612 		i = tun_create_device(drv, u, cred, dev, name);
613 	}
614 	if (i == 0)
615 		if_clone_create(name, namelen, NULL);
616 out:
617 	CURVNET_RESTORE();
618 }
619 
620 static void
621 tun_destroy(struct tuntap_softc *tp)
622 {
623 
624 	TUN_LOCK(tp);
625 	tp->tun_flags |= TUN_DYING;
626 	if (tp->tun_busy != 0)
627 		cv_wait_unlock(&tp->tun_cv, &tp->tun_mtx);
628 	else
629 		TUN_UNLOCK(tp);
630 
631 	CURVNET_SET(TUN2IFP(tp)->if_vnet);
632 
633 	/* destroy_dev will take care of any alias. */
634 	destroy_dev(tp->tun_dev);
635 	seldrain(&tp->tun_rsel);
636 	knlist_clear(&tp->tun_rsel.si_note, 0);
637 	knlist_destroy(&tp->tun_rsel.si_note);
638 	if ((tp->tun_flags & TUN_L2) != 0) {
639 		ether_ifdetach(TUN2IFP(tp));
640 	} else {
641 		bpfdetach(TUN2IFP(tp));
642 		if_detach(TUN2IFP(tp));
643 	}
644 	sx_xlock(&tun_ioctl_sx);
645 	TUN2IFP(tp)->if_softc = NULL;
646 	sx_xunlock(&tun_ioctl_sx);
647 	free_unr(tp->tun_drv->unrhdr, TUN2IFP(tp)->if_dunit);
648 	if_free(TUN2IFP(tp));
649 	mtx_destroy(&tp->tun_mtx);
650 	cv_destroy(&tp->tun_cv);
651 	free(tp, M_TUN);
652 	CURVNET_RESTORE();
653 }
654 
655 static int
656 tun_clone_destroy(struct if_clone *ifc __unused, struct ifnet *ifp, uint32_t flags)
657 {
658 	struct tuntap_softc *tp = ifp->if_softc;
659 
660 	mtx_lock(&tunmtx);
661 	TAILQ_REMOVE(&tunhead, tp, tun_list);
662 	mtx_unlock(&tunmtx);
663 	tun_destroy(tp);
664 
665 	return (0);
666 }
667 
668 static void
669 vnet_tun_init(const void *unused __unused)
670 {
671 	struct tuntap_driver *drv;
672 	struct tuntap_driver_cloner *drvc;
673 	int i;
674 
675 	for (i = 0; i < nitems(tuntap_drivers); ++i) {
676 		drv = &tuntap_drivers[i];
677 		drvc = malloc(sizeof(*drvc), M_TUN, M_WAITOK | M_ZERO);
678 
679 		drvc->drv = drv;
680 		struct if_clone_addreq req = {
681 			.match_f = drv->clone_match_fn,
682 			.create_f = drv->clone_create_fn,
683 			.destroy_f = drv->clone_destroy_fn,
684 		};
685 		drvc->cloner = ifc_attach_cloner(drv->cdevsw.d_name, &req);
686 		SLIST_INSERT_HEAD(&V_tuntap_driver_cloners, drvc, link);
687 	};
688 }
689 VNET_SYSINIT(vnet_tun_init, SI_SUB_PROTO_IF, SI_ORDER_ANY,
690 		vnet_tun_init, NULL);
691 
692 static void
693 vnet_tun_uninit(const void *unused __unused)
694 {
695 	struct tuntap_driver_cloner *drvc;
696 
697 	while (!SLIST_EMPTY(&V_tuntap_driver_cloners)) {
698 		drvc = SLIST_FIRST(&V_tuntap_driver_cloners);
699 		SLIST_REMOVE_HEAD(&V_tuntap_driver_cloners, link);
700 
701 		if_clone_detach(drvc->cloner);
702 		free(drvc, M_TUN);
703 	}
704 }
705 VNET_SYSUNINIT(vnet_tun_uninit, SI_SUB_PROTO_IF, SI_ORDER_ANY,
706     vnet_tun_uninit, NULL);
707 
708 static void
709 tun_uninit(const void *unused __unused)
710 {
711 	struct tuntap_driver *drv;
712 	struct tuntap_softc *tp;
713 	int i;
714 
715 	EVENTHANDLER_DEREGISTER(ifnet_arrival_event, arrival_tag);
716 	EVENTHANDLER_DEREGISTER(dev_clone, clone_tag);
717 
718 	mtx_lock(&tunmtx);
719 	while ((tp = TAILQ_FIRST(&tunhead)) != NULL) {
720 		TAILQ_REMOVE(&tunhead, tp, tun_list);
721 		mtx_unlock(&tunmtx);
722 		tun_destroy(tp);
723 		mtx_lock(&tunmtx);
724 	}
725 	mtx_unlock(&tunmtx);
726 	for (i = 0; i < nitems(tuntap_drivers); ++i) {
727 		drv = &tuntap_drivers[i];
728 		delete_unrhdr(drv->unrhdr);
729 		clone_cleanup(&drv->clones);
730 	}
731 	mtx_destroy(&tunmtx);
732 }
733 SYSUNINIT(tun_uninit, SI_SUB_PROTO_IF, SI_ORDER_ANY, tun_uninit, NULL);
734 
735 static struct tuntap_driver *
736 tuntap_driver_from_ifnet(const struct ifnet *ifp)
737 {
738 	struct tuntap_driver *drv;
739 	int i;
740 
741 	if (ifp == NULL)
742 		return (NULL);
743 
744 	for (i = 0; i < nitems(tuntap_drivers); ++i) {
745 		drv = &tuntap_drivers[i];
746 		if (strcmp(ifp->if_dname, drv->cdevsw.d_name) == 0)
747 			return (drv);
748 	}
749 
750 	return (NULL);
751 }
752 
753 static int
754 tuntapmodevent(module_t mod, int type, void *data)
755 {
756 	struct tuntap_driver *drv;
757 	int i;
758 
759 	switch (type) {
760 	case MOD_LOAD:
761 		mtx_init(&tunmtx, "tunmtx", NULL, MTX_DEF);
762 		for (i = 0; i < nitems(tuntap_drivers); ++i) {
763 			drv = &tuntap_drivers[i];
764 			clone_setup(&drv->clones);
765 			drv->unrhdr = new_unrhdr(0, IF_MAXUNIT, &tunmtx);
766 		}
767 		arrival_tag = EVENTHANDLER_REGISTER(ifnet_arrival_event,
768 		   tunrename, 0, 1000);
769 		if (arrival_tag == NULL)
770 			return (ENOMEM);
771 		clone_tag = EVENTHANDLER_REGISTER(dev_clone, tunclone, 0, 1000);
772 		if (clone_tag == NULL)
773 			return (ENOMEM);
774 		break;
775 	case MOD_UNLOAD:
776 		/* See tun_uninit, so it's done after the vnet_sysuninit() */
777 		break;
778 	default:
779 		return EOPNOTSUPP;
780 	}
781 	return 0;
782 }
783 
784 static moduledata_t tuntap_mod = {
785 	"if_tuntap",
786 	tuntapmodevent,
787 	0
788 };
789 
790 /* We'll only ever have these two, so no need for a macro. */
791 static moduledata_t tun_mod = { "if_tun", NULL, 0 };
792 static moduledata_t tap_mod = { "if_tap", NULL, 0 };
793 
794 DECLARE_MODULE(if_tuntap, tuntap_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
795 MODULE_VERSION(if_tuntap, 1);
796 DECLARE_MODULE(if_tun, tun_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
797 MODULE_VERSION(if_tun, 1);
798 DECLARE_MODULE(if_tap, tap_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
799 MODULE_VERSION(if_tap, 1);
800 
801 static int
802 tun_create_device(struct tuntap_driver *drv, int unit, struct ucred *cr,
803     struct cdev **dev, const char *name)
804 {
805 	struct make_dev_args args;
806 	struct tuntap_softc *tp;
807 	int error;
808 
809 	tp = malloc(sizeof(*tp), M_TUN, M_WAITOK | M_ZERO);
810 	mtx_init(&tp->tun_mtx, "tun_mtx", NULL, MTX_DEF);
811 	cv_init(&tp->tun_cv, "tun_condvar");
812 	tp->tun_flags = drv->ident_flags;
813 	tp->tun_drv = drv;
814 
815 	make_dev_args_init(&args);
816 	if (cr != NULL)
817 		args.mda_flags = MAKEDEV_REF;
818 	args.mda_devsw = &drv->cdevsw;
819 	args.mda_cr = cr;
820 	args.mda_uid = UID_UUCP;
821 	args.mda_gid = GID_DIALER;
822 	args.mda_mode = 0600;
823 	args.mda_unit = unit;
824 	args.mda_si_drv1 = tp;
825 	error = make_dev_s(&args, dev, "%s", name);
826 	if (error != 0) {
827 		free(tp, M_TUN);
828 		return (error);
829 	}
830 
831 	KASSERT((*dev)->si_drv1 != NULL,
832 	    ("Failed to set si_drv1 at %s creation", name));
833 	tp->tun_dev = *dev;
834 	knlist_init_mtx(&tp->tun_rsel.si_note, &tp->tun_mtx);
835 	mtx_lock(&tunmtx);
836 	TAILQ_INSERT_TAIL(&tunhead, tp, tun_list);
837 	mtx_unlock(&tunmtx);
838 	return (0);
839 }
840 
841 static void
842 tunstart(struct ifnet *ifp)
843 {
844 	struct tuntap_softc *tp = ifp->if_softc;
845 	struct mbuf *m;
846 
847 	TUNDEBUG(ifp, "starting\n");
848 	if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
849 		IFQ_LOCK(&ifp->if_snd);
850 		IFQ_POLL_NOLOCK(&ifp->if_snd, m);
851 		if (m == NULL) {
852 			IFQ_UNLOCK(&ifp->if_snd);
853 			return;
854 		}
855 		IFQ_UNLOCK(&ifp->if_snd);
856 	}
857 
858 	TUN_LOCK(tp);
859 	if (tp->tun_flags & TUN_RWAIT) {
860 		tp->tun_flags &= ~TUN_RWAIT;
861 		wakeup(tp);
862 	}
863 	selwakeuppri(&tp->tun_rsel, PZERO + 1);
864 	KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
865 	if (tp->tun_flags & TUN_ASYNC && tp->tun_sigio) {
866 		TUN_UNLOCK(tp);
867 		pgsigio(&tp->tun_sigio, SIGIO, 0);
868 	} else
869 		TUN_UNLOCK(tp);
870 }
871 
872 /*
873  * tunstart_l2
874  *
875  * queue packets from higher level ready to put out
876  */
877 static void
878 tunstart_l2(struct ifnet *ifp)
879 {
880 	struct tuntap_softc	*tp = ifp->if_softc;
881 
882 	TUNDEBUG(ifp, "starting\n");
883 
884 	/*
885 	 * do not junk pending output if we are in VMnet mode.
886 	 * XXX: can this do any harm because of queue overflow?
887 	 */
888 
889 	TUN_LOCK(tp);
890 	if (((tp->tun_flags & TUN_VMNET) == 0) &&
891 	    ((tp->tun_flags & TUN_READY) != TUN_READY)) {
892 		struct mbuf *m;
893 
894 		/* Unlocked read. */
895 		TUNDEBUG(ifp, "not ready, tun_flags = 0x%x\n", tp->tun_flags);
896 
897 		for (;;) {
898 			IF_DEQUEUE(&ifp->if_snd, m);
899 			if (m != NULL) {
900 				m_freem(m);
901 				if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
902 			} else
903 				break;
904 		}
905 		TUN_UNLOCK(tp);
906 
907 		return;
908 	}
909 
910 	ifp->if_drv_flags |= IFF_DRV_OACTIVE;
911 
912 	if (!IFQ_IS_EMPTY(&ifp->if_snd)) {
913 		if (tp->tun_flags & TUN_RWAIT) {
914 			tp->tun_flags &= ~TUN_RWAIT;
915 			wakeup(tp);
916 		}
917 
918 		if ((tp->tun_flags & TUN_ASYNC) && (tp->tun_sigio != NULL)) {
919 			TUN_UNLOCK(tp);
920 			pgsigio(&tp->tun_sigio, SIGIO, 0);
921 			TUN_LOCK(tp);
922 		}
923 
924 		selwakeuppri(&tp->tun_rsel, PZERO+1);
925 		KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
926 		if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); /* obytes are counted in ether_output */
927 	}
928 
929 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
930 	TUN_UNLOCK(tp);
931 } /* tunstart_l2 */
932 
933 /* XXX: should return an error code so it can fail. */
934 static void
935 tuncreate(struct cdev *dev)
936 {
937 	struct tuntap_driver *drv;
938 	struct tuntap_softc *tp;
939 	struct ifnet *ifp;
940 	struct ether_addr eaddr;
941 	int iflags;
942 	u_char type;
943 
944 	tp = dev->si_drv1;
945 	KASSERT(tp != NULL,
946 	    ("si_drv1 should have been initialized at creation"));
947 
948 	drv = tp->tun_drv;
949 	iflags = IFF_MULTICAST;
950 	if ((tp->tun_flags & TUN_L2) != 0) {
951 		type = IFT_ETHER;
952 		iflags |= IFF_BROADCAST | IFF_SIMPLEX;
953 	} else {
954 		type = IFT_PPP;
955 		iflags |= IFF_POINTOPOINT;
956 	}
957 	ifp = tp->tun_ifp = if_alloc(type);
958 	if (ifp == NULL)
959 		panic("%s%d: failed to if_alloc() interface.\n",
960 		    drv->cdevsw.d_name, dev2unit(dev));
961 	ifp->if_softc = tp;
962 	if_initname(ifp, drv->cdevsw.d_name, dev2unit(dev));
963 	ifp->if_ioctl = tunifioctl;
964 	ifp->if_flags = iflags;
965 	IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen);
966 	ifp->if_capabilities |= IFCAP_LINKSTATE;
967 	ifp->if_capenable |= IFCAP_LINKSTATE;
968 
969 	if ((tp->tun_flags & TUN_L2) != 0) {
970 		ifp->if_init = tunifinit;
971 		ifp->if_start = tunstart_l2;
972 
973 		ether_gen_addr(ifp, &eaddr);
974 		ether_ifattach(ifp, eaddr.octet);
975 	} else {
976 		ifp->if_mtu = TUNMTU;
977 		ifp->if_start = tunstart;
978 		ifp->if_output = tunoutput;
979 
980 		ifp->if_snd.ifq_drv_maxlen = 0;
981 		IFQ_SET_READY(&ifp->if_snd);
982 
983 		if_attach(ifp);
984 		bpfattach(ifp, DLT_NULL, sizeof(u_int32_t));
985 	}
986 
987 	TUN_LOCK(tp);
988 	tp->tun_flags |= TUN_INITED;
989 	TUN_UNLOCK(tp);
990 
991 	TUNDEBUG(ifp, "interface %s is created, minor = %#x\n",
992 	    ifp->if_xname, dev2unit(dev));
993 }
994 
995 static void
996 tunrename(void *arg __unused, struct ifnet *ifp)
997 {
998 	struct tuntap_softc *tp;
999 	int error;
1000 
1001 	if ((ifp->if_flags & IFF_RENAMING) == 0)
1002 		return;
1003 
1004 	if (tuntap_driver_from_ifnet(ifp) == NULL)
1005 		return;
1006 
1007 	/*
1008 	 * We need to grab the ioctl sx long enough to make sure the softc is
1009 	 * still there.  If it is, we can safely try to busy the tun device.
1010 	 * The busy may fail if the device is currently dying, in which case
1011 	 * we do nothing.  If it doesn't fail, the busy count stops the device
1012 	 * from dying until we've created the alias (that will then be
1013 	 * subsequently destroyed).
1014 	 */
1015 	sx_xlock(&tun_ioctl_sx);
1016 	tp = ifp->if_softc;
1017 	if (tp == NULL) {
1018 		sx_xunlock(&tun_ioctl_sx);
1019 		return;
1020 	}
1021 	error = tun_busy(tp);
1022 	sx_xunlock(&tun_ioctl_sx);
1023 	if (error != 0)
1024 		return;
1025 	if (tp->tun_alias != NULL) {
1026 		destroy_dev(tp->tun_alias);
1027 		tp->tun_alias = NULL;
1028 	}
1029 
1030 	if (strcmp(ifp->if_xname, tp->tun_dev->si_name) == 0)
1031 		goto out;
1032 
1033 	/*
1034 	 * Failure's ok, aliases are created on a best effort basis.  If a
1035 	 * tun user/consumer decides to rename the interface to conflict with
1036 	 * another device (non-ifnet) on the system, we will assume they know
1037 	 * what they are doing.  make_dev_alias_p won't touch tun_alias on
1038 	 * failure, so we use it but ignore the return value.
1039 	 */
1040 	make_dev_alias_p(MAKEDEV_CHECKNAME, &tp->tun_alias, tp->tun_dev, "%s",
1041 	    ifp->if_xname);
1042 out:
1043 	tun_unbusy(tp);
1044 }
1045 
1046 static int
1047 tunopen(struct cdev *dev, int flag, int mode, struct thread *td)
1048 {
1049 	struct ifnet	*ifp;
1050 	struct tuntap_softc *tp;
1051 	int error __diagused, tunflags;
1052 
1053 	tunflags = 0;
1054 	CURVNET_SET(TD_TO_VNET(td));
1055 	error = tuntap_name2info(dev->si_name, NULL, &tunflags);
1056 	if (error != 0) {
1057 		CURVNET_RESTORE();
1058 		return (error);	/* Shouldn't happen */
1059 	}
1060 
1061 	tp = dev->si_drv1;
1062 	KASSERT(tp != NULL,
1063 	    ("si_drv1 should have been initialized at creation"));
1064 
1065 	TUN_LOCK(tp);
1066 	if ((tp->tun_flags & TUN_INITED) == 0) {
1067 		TUN_UNLOCK(tp);
1068 		CURVNET_RESTORE();
1069 		return (ENXIO);
1070 	}
1071 	if ((tp->tun_flags & (TUN_OPEN | TUN_DYING)) != 0) {
1072 		TUN_UNLOCK(tp);
1073 		CURVNET_RESTORE();
1074 		return (EBUSY);
1075 	}
1076 
1077 	error = tun_busy_locked(tp);
1078 	KASSERT(error == 0, ("Must be able to busy an unopen tunnel"));
1079 	ifp = TUN2IFP(tp);
1080 
1081 	if ((tp->tun_flags & TUN_L2) != 0) {
1082 		bcopy(IF_LLADDR(ifp), tp->tun_ether.octet,
1083 		    sizeof(tp->tun_ether.octet));
1084 
1085 		ifp->if_drv_flags |= IFF_DRV_RUNNING;
1086 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1087 
1088 		if (tapuponopen)
1089 			ifp->if_flags |= IFF_UP;
1090 	}
1091 
1092 	tp->tun_pid = td->td_proc->p_pid;
1093 	tp->tun_flags |= TUN_OPEN;
1094 
1095 	if_link_state_change(ifp, LINK_STATE_UP);
1096 	TUNDEBUG(ifp, "open\n");
1097 	TUN_UNLOCK(tp);
1098 
1099 	/*
1100 	 * This can fail with either ENOENT or EBUSY.  This is in the middle of
1101 	 * d_open, so ENOENT should not be possible.  EBUSY is possible, but
1102 	 * the only cdevpriv dtor being set will be tundtor and the softc being
1103 	 * passed is constant for a given cdev.  We ignore the possible error
1104 	 * because of this as either "unlikely" or "not actually a problem."
1105 	 */
1106 	(void)devfs_set_cdevpriv(tp, tundtor);
1107 	CURVNET_RESTORE();
1108 	return (0);
1109 }
1110 
1111 /*
1112  * tundtor - tear down the device - mark i/f down & delete
1113  * routing info
1114  */
1115 static void
1116 tundtor(void *data)
1117 {
1118 	struct proc *p;
1119 	struct tuntap_softc *tp;
1120 	struct ifnet *ifp;
1121 	bool l2tun;
1122 
1123 	tp = data;
1124 	p = curproc;
1125 	ifp = TUN2IFP(tp);
1126 
1127 	TUN_LOCK(tp);
1128 
1129 	/*
1130 	 * Realistically, we can't be obstinate here.  This only means that the
1131 	 * tuntap device was closed out of order, and the last closer wasn't the
1132 	 * controller.  These are still good to know about, though, as software
1133 	 * should avoid multiple processes with a tuntap device open and
1134 	 * ill-defined transfer of control (e.g., handoff, TUNSIFPID, close in
1135 	 * parent).
1136 	 */
1137 	if (p->p_pid != tp->tun_pid) {
1138 		log(LOG_INFO,
1139 		    "pid %d (%s), %s: tun/tap protocol violation, non-controlling process closed last.\n",
1140 		    p->p_pid, p->p_comm, tp->tun_dev->si_name);
1141 	}
1142 
1143 	/*
1144 	 * junk all pending output
1145 	 */
1146 	CURVNET_SET(ifp->if_vnet);
1147 
1148 	l2tun = false;
1149 	if ((tp->tun_flags & TUN_L2) != 0) {
1150 		l2tun = true;
1151 		IF_DRAIN(&ifp->if_snd);
1152 	} else {
1153 		IFQ_PURGE(&ifp->if_snd);
1154 	}
1155 
1156 	/* For vmnet, we won't do most of the address/route bits */
1157 	if ((tp->tun_flags & TUN_VMNET) != 0 ||
1158 	    (l2tun && (ifp->if_flags & IFF_LINK0) != 0))
1159 		goto out;
1160 
1161 	if (ifp->if_flags & IFF_UP) {
1162 		TUN_UNLOCK(tp);
1163 		if_down(ifp);
1164 		TUN_LOCK(tp);
1165 	}
1166 
1167 	/* Delete all addresses and routes which reference this interface. */
1168 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1169 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1170 		TUN_UNLOCK(tp);
1171 		if_purgeaddrs(ifp);
1172 		TUN_LOCK(tp);
1173 	}
1174 
1175 out:
1176 	if_link_state_change(ifp, LINK_STATE_DOWN);
1177 	CURVNET_RESTORE();
1178 
1179 	funsetown(&tp->tun_sigio);
1180 	selwakeuppri(&tp->tun_rsel, PZERO + 1);
1181 	KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
1182 	TUNDEBUG (ifp, "closed\n");
1183 	tp->tun_flags &= ~TUN_OPEN;
1184 	tp->tun_pid = 0;
1185 	tun_vnethdr_set(ifp, 0);
1186 
1187 	tun_unbusy_locked(tp);
1188 	TUN_UNLOCK(tp);
1189 }
1190 
1191 static void
1192 tuninit(struct ifnet *ifp)
1193 {
1194 	struct tuntap_softc *tp = ifp->if_softc;
1195 
1196 	TUNDEBUG(ifp, "tuninit\n");
1197 
1198 	TUN_LOCK(tp);
1199 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1200 	if ((tp->tun_flags & TUN_L2) == 0) {
1201 		ifp->if_flags |= IFF_UP;
1202 		getmicrotime(&ifp->if_lastchange);
1203 		TUN_UNLOCK(tp);
1204 	} else {
1205 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1206 		TUN_UNLOCK(tp);
1207 		/* attempt to start output */
1208 		tunstart_l2(ifp);
1209 	}
1210 
1211 }
1212 
1213 /*
1214  * Used only for l2 tunnel.
1215  */
1216 static void
1217 tunifinit(void *xtp)
1218 {
1219 	struct tuntap_softc *tp;
1220 
1221 	tp = (struct tuntap_softc *)xtp;
1222 	tuninit(tp->tun_ifp);
1223 }
1224 
1225 /*
1226  * To be called under TUN_LOCK. Update ifp->if_hwassist according to the
1227  * current value of ifp->if_capenable.
1228  */
1229 static void
1230 tun_caps_changed(struct ifnet *ifp)
1231 {
1232 	uint64_t hwassist = 0;
1233 
1234 	TUN_LOCK_ASSERT((struct tuntap_softc *)ifp->if_softc);
1235 	if (ifp->if_capenable & IFCAP_TXCSUM)
1236 		hwassist |= CSUM_TCP | CSUM_UDP;
1237 	if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
1238 		hwassist |= CSUM_TCP_IPV6
1239 		    | CSUM_UDP_IPV6;
1240 	if (ifp->if_capenable & IFCAP_TSO4)
1241 		hwassist |= CSUM_IP_TSO;
1242 	if (ifp->if_capenable & IFCAP_TSO6)
1243 		hwassist |= CSUM_IP6_TSO;
1244 	ifp->if_hwassist = hwassist;
1245 }
1246 
1247 /*
1248  * To be called under TUN_LOCK. Update tp->tun_vhdrlen and adjust
1249  * if_capabilities and if_capenable as needed.
1250  */
1251 static void
1252 tun_vnethdr_set(struct ifnet *ifp, int vhdrlen)
1253 {
1254 	struct tuntap_softc *tp = ifp->if_softc;
1255 
1256 	TUN_LOCK_ASSERT(tp);
1257 
1258 	if (tp->tun_vhdrlen == vhdrlen)
1259 		return;
1260 
1261 	/*
1262 	 * Update if_capabilities to reflect the
1263 	 * functionalities offered by the virtio-net
1264 	 * header.
1265 	 */
1266 	if (vhdrlen != 0)
1267 		ifp->if_capabilities |=
1268 			TAP_VNET_HDR_CAPS;
1269 	else
1270 		ifp->if_capabilities &=
1271 			~TAP_VNET_HDR_CAPS;
1272 	/*
1273 	 * Disable any capabilities that we don't
1274 	 * support anymore.
1275 	 */
1276 	ifp->if_capenable &= ifp->if_capabilities;
1277 	tun_caps_changed(ifp);
1278 	tp->tun_vhdrlen = vhdrlen;
1279 
1280 	TUNDEBUG(ifp, "vnet_hdr_len=%d, if_capabilities=%x\n",
1281 	    vhdrlen, ifp->if_capabilities);
1282 }
1283 
1284 /*
1285  * Process an ioctl request.
1286  */
1287 static int
1288 tunifioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1289 {
1290 	struct ifreq *ifr = (struct ifreq *)data;
1291 	struct tuntap_softc *tp;
1292 	struct ifstat *ifs;
1293 	struct ifmediareq	*ifmr;
1294 	int		dummy, error = 0;
1295 	bool		l2tun;
1296 
1297 	ifmr = NULL;
1298 	sx_xlock(&tun_ioctl_sx);
1299 	tp = ifp->if_softc;
1300 	if (tp == NULL) {
1301 		error = ENXIO;
1302 		goto bad;
1303 	}
1304 	l2tun = (tp->tun_flags & TUN_L2) != 0;
1305 	switch(cmd) {
1306 	case SIOCGIFSTATUS:
1307 		ifs = (struct ifstat *)data;
1308 		TUN_LOCK(tp);
1309 		if (tp->tun_pid)
1310 			snprintf(ifs->ascii, sizeof(ifs->ascii),
1311 			    "\tOpened by PID %d\n", tp->tun_pid);
1312 		else
1313 			ifs->ascii[0] = '\0';
1314 		TUN_UNLOCK(tp);
1315 		break;
1316 	case SIOCSIFADDR:
1317 		if (l2tun)
1318 			error = ether_ioctl(ifp, cmd, data);
1319 		else
1320 			tuninit(ifp);
1321 		if (error == 0)
1322 		    TUNDEBUG(ifp, "address set\n");
1323 		break;
1324 	case SIOCSIFMTU:
1325 		ifp->if_mtu = ifr->ifr_mtu;
1326 		TUNDEBUG(ifp, "mtu set\n");
1327 		break;
1328 	case SIOCSIFFLAGS:
1329 	case SIOCADDMULTI:
1330 	case SIOCDELMULTI:
1331 		break;
1332 	case SIOCGIFMEDIA:
1333 		if (!l2tun) {
1334 			error = EINVAL;
1335 			break;
1336 		}
1337 
1338 		ifmr = (struct ifmediareq *)data;
1339 		dummy = ifmr->ifm_count;
1340 		ifmr->ifm_count = 1;
1341 		ifmr->ifm_status = IFM_AVALID;
1342 		ifmr->ifm_active = IFM_ETHER | IFM_FDX | IFM_1000_T;
1343 		if (tp->tun_flags & TUN_OPEN)
1344 			ifmr->ifm_status |= IFM_ACTIVE;
1345 		ifmr->ifm_current = ifmr->ifm_active;
1346 		if (dummy >= 1) {
1347 			int media = IFM_ETHER;
1348 			error = copyout(&media, ifmr->ifm_ulist, sizeof(int));
1349 		}
1350 		break;
1351 	case SIOCSIFCAP:
1352 		TUN_LOCK(tp);
1353 		ifp->if_capenable = ifr->ifr_reqcap;
1354 		tun_caps_changed(ifp);
1355 		TUN_UNLOCK(tp);
1356 		VLAN_CAPABILITIES(ifp);
1357 		break;
1358 	default:
1359 		if (l2tun) {
1360 			error = ether_ioctl(ifp, cmd, data);
1361 		} else {
1362 			error = EINVAL;
1363 		}
1364 	}
1365 bad:
1366 	sx_xunlock(&tun_ioctl_sx);
1367 	return (error);
1368 }
1369 
1370 /*
1371  * tunoutput - queue packets from higher level ready to put out.
1372  */
1373 static int
1374 tunoutput(struct ifnet *ifp, struct mbuf *m0, const struct sockaddr *dst,
1375     struct route *ro)
1376 {
1377 	struct tuntap_softc *tp = ifp->if_softc;
1378 	u_short cached_tun_flags;
1379 	int error;
1380 	u_int32_t af;
1381 
1382 	TUNDEBUG (ifp, "tunoutput\n");
1383 
1384 #ifdef MAC
1385 	error = mac_ifnet_check_transmit(ifp, m0);
1386 	if (error) {
1387 		m_freem(m0);
1388 		return (error);
1389 	}
1390 #endif
1391 
1392 	/* Could be unlocked read? */
1393 	TUN_LOCK(tp);
1394 	cached_tun_flags = tp->tun_flags;
1395 	TUN_UNLOCK(tp);
1396 	if ((cached_tun_flags & TUN_READY) != TUN_READY) {
1397 		TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags);
1398 		m_freem (m0);
1399 		return (EHOSTDOWN);
1400 	}
1401 
1402 	if ((ifp->if_flags & IFF_UP) != IFF_UP) {
1403 		m_freem (m0);
1404 		return (EHOSTDOWN);
1405 	}
1406 
1407 	/* BPF writes need to be handled specially. */
1408 	if (dst->sa_family == AF_UNSPEC)
1409 		bcopy(dst->sa_data, &af, sizeof(af));
1410 	else
1411 		af = RO_GET_FAMILY(ro, dst);
1412 
1413 	if (bpf_peers_present(ifp->if_bpf))
1414 		bpf_mtap2(ifp->if_bpf, &af, sizeof(af), m0);
1415 
1416 	/* prepend sockaddr? this may abort if the mbuf allocation fails */
1417 	if (cached_tun_flags & TUN_LMODE) {
1418 		/* allocate space for sockaddr */
1419 		M_PREPEND(m0, dst->sa_len, M_NOWAIT);
1420 
1421 		/* if allocation failed drop packet */
1422 		if (m0 == NULL) {
1423 			if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1424 			if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1425 			return (ENOBUFS);
1426 		} else {
1427 			bcopy(dst, m0->m_data, dst->sa_len);
1428 		}
1429 	}
1430 
1431 	if (cached_tun_flags & TUN_IFHEAD) {
1432 		/* Prepend the address family */
1433 		M_PREPEND(m0, 4, M_NOWAIT);
1434 
1435 		/* if allocation failed drop packet */
1436 		if (m0 == NULL) {
1437 			if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1438 			if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1439 			return (ENOBUFS);
1440 		} else
1441 			*(u_int32_t *)m0->m_data = htonl(af);
1442 	} else {
1443 #ifdef INET
1444 		if (af != AF_INET)
1445 #endif
1446 		{
1447 			m_freem(m0);
1448 			return (EAFNOSUPPORT);
1449 		}
1450 	}
1451 
1452 	error = (ifp->if_transmit)(ifp, m0);
1453 	if (error)
1454 		return (ENOBUFS);
1455 	if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1456 	return (0);
1457 }
1458 
1459 /*
1460  * the cdevsw interface is now pretty minimal.
1461  */
1462 static	int
1463 tunioctl(struct cdev *dev, u_long cmd, caddr_t data, int flag,
1464     struct thread *td)
1465 {
1466 	struct ifreq ifr, *ifrp;
1467 	struct tuntap_softc *tp = dev->si_drv1;
1468 	struct ifnet *ifp = TUN2IFP(tp);
1469 	struct tuninfo *tunp;
1470 	int error, iflags, ival;
1471 	bool	l2tun;
1472 
1473 	l2tun = (tp->tun_flags & TUN_L2) != 0;
1474 	if (l2tun) {
1475 		/* tap specific ioctls */
1476 		switch(cmd) {
1477 		/* VMware/VMnet port ioctl's */
1478 #if defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD5) || \
1479     defined(COMPAT_FREEBSD4)
1480 		case _IO('V', 0):
1481 			ival = IOCPARM_IVAL(data);
1482 			data = (caddr_t)&ival;
1483 			/* FALLTHROUGH */
1484 #endif
1485 		case VMIO_SIOCSIFFLAGS: /* VMware/VMnet SIOCSIFFLAGS */
1486 			iflags = *(int *)data;
1487 			iflags &= TUN_VMIO_FLAG_MASK;
1488 			iflags &= ~IFF_CANTCHANGE;
1489 			iflags |= IFF_UP;
1490 
1491 			TUN_LOCK(tp);
1492 			ifp->if_flags = iflags |
1493 			    (ifp->if_flags & IFF_CANTCHANGE);
1494 			TUN_UNLOCK(tp);
1495 
1496 			return (0);
1497 		case SIOCGIFADDR:	/* get MAC address of the remote side */
1498 			TUN_LOCK(tp);
1499 			bcopy(&tp->tun_ether.octet, data,
1500 			    sizeof(tp->tun_ether.octet));
1501 			TUN_UNLOCK(tp);
1502 
1503 			return (0);
1504 		case SIOCSIFADDR:	/* set MAC address of the remote side */
1505 			TUN_LOCK(tp);
1506 			bcopy(data, &tp->tun_ether.octet,
1507 			    sizeof(tp->tun_ether.octet));
1508 			TUN_UNLOCK(tp);
1509 
1510 			return (0);
1511 		case TAPSVNETHDR:
1512 			ival = *(int *)data;
1513 			if (ival != 0 &&
1514 			    ival != sizeof(struct virtio_net_hdr) &&
1515 			    ival != sizeof(struct virtio_net_hdr_mrg_rxbuf)) {
1516 				return (EINVAL);
1517 			}
1518 			TUN_LOCK(tp);
1519 			tun_vnethdr_set(ifp, ival);
1520 			TUN_UNLOCK(tp);
1521 
1522 			return (0);
1523 		case TAPGVNETHDR:
1524 			TUN_LOCK(tp);
1525 			*(int *)data = tp->tun_vhdrlen;
1526 			TUN_UNLOCK(tp);
1527 
1528 			return (0);
1529 		}
1530 
1531 		/* Fall through to the common ioctls if unhandled */
1532 	} else {
1533 		switch (cmd) {
1534 		case TUNSLMODE:
1535 			TUN_LOCK(tp);
1536 			if (*(int *)data) {
1537 				tp->tun_flags |= TUN_LMODE;
1538 				tp->tun_flags &= ~TUN_IFHEAD;
1539 			} else
1540 				tp->tun_flags &= ~TUN_LMODE;
1541 			TUN_UNLOCK(tp);
1542 
1543 			return (0);
1544 		case TUNSIFHEAD:
1545 			TUN_LOCK(tp);
1546 			if (*(int *)data) {
1547 				tp->tun_flags |= TUN_IFHEAD;
1548 				tp->tun_flags &= ~TUN_LMODE;
1549 			} else
1550 				tp->tun_flags &= ~TUN_IFHEAD;
1551 			TUN_UNLOCK(tp);
1552 
1553 			return (0);
1554 		case TUNGIFHEAD:
1555 			TUN_LOCK(tp);
1556 			*(int *)data = (tp->tun_flags & TUN_IFHEAD) ? 1 : 0;
1557 			TUN_UNLOCK(tp);
1558 
1559 			return (0);
1560 		case TUNSIFMODE:
1561 			/* deny this if UP */
1562 			if (TUN2IFP(tp)->if_flags & IFF_UP)
1563 				return (EBUSY);
1564 
1565 			switch (*(int *)data & ~IFF_MULTICAST) {
1566 			case IFF_POINTOPOINT:
1567 			case IFF_BROADCAST:
1568 				TUN_LOCK(tp);
1569 				TUN2IFP(tp)->if_flags &=
1570 				    ~(IFF_BROADCAST|IFF_POINTOPOINT|IFF_MULTICAST);
1571 				TUN2IFP(tp)->if_flags |= *(int *)data;
1572 				TUN_UNLOCK(tp);
1573 
1574 				break;
1575 			default:
1576 				return (EINVAL);
1577 			}
1578 
1579 			return (0);
1580 		case TUNSIFPID:
1581 			TUN_LOCK(tp);
1582 			tp->tun_pid = curthread->td_proc->p_pid;
1583 			TUN_UNLOCK(tp);
1584 
1585 			return (0);
1586 		}
1587 		/* Fall through to the common ioctls if unhandled */
1588 	}
1589 
1590 	switch (cmd) {
1591 	case TUNGIFNAME:
1592 		ifrp = (struct ifreq *)data;
1593 		strlcpy(ifrp->ifr_name, TUN2IFP(tp)->if_xname, IFNAMSIZ);
1594 
1595 		return (0);
1596 	case TUNSIFINFO:
1597 		tunp = (struct tuninfo *)data;
1598 		if (TUN2IFP(tp)->if_type != tunp->type)
1599 			return (EPROTOTYPE);
1600 		TUN_LOCK(tp);
1601 		if (TUN2IFP(tp)->if_mtu != tunp->mtu) {
1602 			strlcpy(ifr.ifr_name, if_name(TUN2IFP(tp)), IFNAMSIZ);
1603 			ifr.ifr_mtu = tunp->mtu;
1604 			CURVNET_SET(TUN2IFP(tp)->if_vnet);
1605 			error = ifhwioctl(SIOCSIFMTU, TUN2IFP(tp),
1606 			    (caddr_t)&ifr, td);
1607 			CURVNET_RESTORE();
1608 			if (error) {
1609 				TUN_UNLOCK(tp);
1610 				return (error);
1611 			}
1612 		}
1613 		TUN2IFP(tp)->if_baudrate = tunp->baudrate;
1614 		TUN_UNLOCK(tp);
1615 		break;
1616 	case TUNGIFINFO:
1617 		tunp = (struct tuninfo *)data;
1618 		TUN_LOCK(tp);
1619 		tunp->mtu = TUN2IFP(tp)->if_mtu;
1620 		tunp->type = TUN2IFP(tp)->if_type;
1621 		tunp->baudrate = TUN2IFP(tp)->if_baudrate;
1622 		TUN_UNLOCK(tp);
1623 		break;
1624 	case TUNSDEBUG:
1625 		tundebug = *(int *)data;
1626 		break;
1627 	case TUNGDEBUG:
1628 		*(int *)data = tundebug;
1629 		break;
1630 	case FIONBIO:
1631 		break;
1632 	case FIOASYNC:
1633 		TUN_LOCK(tp);
1634 		if (*(int *)data)
1635 			tp->tun_flags |= TUN_ASYNC;
1636 		else
1637 			tp->tun_flags &= ~TUN_ASYNC;
1638 		TUN_UNLOCK(tp);
1639 		break;
1640 	case FIONREAD:
1641 		if (!IFQ_IS_EMPTY(&TUN2IFP(tp)->if_snd)) {
1642 			struct mbuf *mb;
1643 			IFQ_LOCK(&TUN2IFP(tp)->if_snd);
1644 			IFQ_POLL_NOLOCK(&TUN2IFP(tp)->if_snd, mb);
1645 			for (*(int *)data = 0; mb != NULL; mb = mb->m_next)
1646 				*(int *)data += mb->m_len;
1647 			IFQ_UNLOCK(&TUN2IFP(tp)->if_snd);
1648 		} else
1649 			*(int *)data = 0;
1650 		break;
1651 	case FIOSETOWN:
1652 		return (fsetown(*(int *)data, &tp->tun_sigio));
1653 
1654 	case FIOGETOWN:
1655 		*(int *)data = fgetown(&tp->tun_sigio);
1656 		return (0);
1657 
1658 	/* This is deprecated, FIOSETOWN should be used instead. */
1659 	case TIOCSPGRP:
1660 		return (fsetown(-(*(int *)data), &tp->tun_sigio));
1661 
1662 	/* This is deprecated, FIOGETOWN should be used instead. */
1663 	case TIOCGPGRP:
1664 		*(int *)data = -fgetown(&tp->tun_sigio);
1665 		return (0);
1666 
1667 	default:
1668 		return (ENOTTY);
1669 	}
1670 	return (0);
1671 }
1672 
1673 /*
1674  * The cdevsw read interface - reads a packet at a time, or at
1675  * least as much of a packet as can be read.
1676  */
1677 static	int
1678 tunread(struct cdev *dev, struct uio *uio, int flag)
1679 {
1680 	struct tuntap_softc *tp = dev->si_drv1;
1681 	struct ifnet	*ifp = TUN2IFP(tp);
1682 	struct mbuf	*m;
1683 	size_t		len;
1684 	int		error = 0;
1685 
1686 	TUNDEBUG (ifp, "read\n");
1687 	TUN_LOCK(tp);
1688 	if ((tp->tun_flags & TUN_READY) != TUN_READY) {
1689 		TUN_UNLOCK(tp);
1690 		TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags);
1691 		return (EHOSTDOWN);
1692 	}
1693 
1694 	tp->tun_flags &= ~TUN_RWAIT;
1695 
1696 	for (;;) {
1697 		IFQ_DEQUEUE(&ifp->if_snd, m);
1698 		if (m != NULL)
1699 			break;
1700 		if (flag & O_NONBLOCK) {
1701 			TUN_UNLOCK(tp);
1702 			return (EWOULDBLOCK);
1703 		}
1704 		tp->tun_flags |= TUN_RWAIT;
1705 		error = mtx_sleep(tp, &tp->tun_mtx, PCATCH | (PZERO + 1),
1706 		    "tunread", 0);
1707 		if (error != 0) {
1708 			TUN_UNLOCK(tp);
1709 			return (error);
1710 		}
1711 	}
1712 	TUN_UNLOCK(tp);
1713 
1714 	if ((tp->tun_flags & TUN_L2) != 0)
1715 		BPF_MTAP(ifp, m);
1716 
1717 	len = min(tp->tun_vhdrlen, uio->uio_resid);
1718 	if (len > 0) {
1719 		struct virtio_net_hdr_mrg_rxbuf vhdr;
1720 
1721 		bzero(&vhdr, sizeof(vhdr));
1722 		if (m->m_pkthdr.csum_flags & TAP_ALL_OFFLOAD) {
1723 			m = virtio_net_tx_offload(ifp, m, false, &vhdr.hdr);
1724 		}
1725 
1726 		TUNDEBUG(ifp, "txvhdr: f %u, gt %u, hl %u, "
1727 		    "gs %u, cs %u, co %u\n", vhdr.hdr.flags,
1728 		    vhdr.hdr.gso_type, vhdr.hdr.hdr_len,
1729 		    vhdr.hdr.gso_size, vhdr.hdr.csum_start,
1730 		    vhdr.hdr.csum_offset);
1731 		error = uiomove(&vhdr, len, uio);
1732 	}
1733 
1734 	while (m && uio->uio_resid > 0 && error == 0) {
1735 		len = min(uio->uio_resid, m->m_len);
1736 		if (len != 0)
1737 			error = uiomove(mtod(m, void *), len, uio);
1738 		m = m_free(m);
1739 	}
1740 
1741 	if (m) {
1742 		TUNDEBUG(ifp, "Dropping mbuf\n");
1743 		m_freem(m);
1744 	}
1745 	return (error);
1746 }
1747 
1748 static int
1749 tunwrite_l2(struct tuntap_softc *tp, struct mbuf *m,
1750 	    struct virtio_net_hdr_mrg_rxbuf *vhdr)
1751 {
1752 	struct epoch_tracker et;
1753 	struct ether_header *eh;
1754 	struct ifnet *ifp;
1755 
1756 	ifp = TUN2IFP(tp);
1757 
1758 	/*
1759 	 * Only pass a unicast frame to ether_input(), if it would
1760 	 * actually have been received by non-virtual hardware.
1761 	 */
1762 	if (m->m_len < sizeof(struct ether_header)) {
1763 		m_freem(m);
1764 		return (0);
1765 	}
1766 
1767 	eh = mtod(m, struct ether_header *);
1768 
1769 	if (eh && (ifp->if_flags & IFF_PROMISC) == 0 &&
1770 	    !ETHER_IS_MULTICAST(eh->ether_dhost) &&
1771 	    bcmp(eh->ether_dhost, IF_LLADDR(ifp), ETHER_ADDR_LEN) != 0) {
1772 		m_freem(m);
1773 		return (0);
1774 	}
1775 
1776 	if (vhdr != NULL && virtio_net_rx_csum(m, &vhdr->hdr)) {
1777 		m_freem(m);
1778 		return (0);
1779 	}
1780 
1781 	/* Pass packet up to parent. */
1782 	CURVNET_SET(ifp->if_vnet);
1783 	NET_EPOCH_ENTER(et);
1784 	(*ifp->if_input)(ifp, m);
1785 	NET_EPOCH_EXIT(et);
1786 	CURVNET_RESTORE();
1787 	/* ibytes are counted in parent */
1788 	if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1789 	return (0);
1790 }
1791 
1792 static int
1793 tunwrite_l3(struct tuntap_softc *tp, struct mbuf *m)
1794 {
1795 	struct epoch_tracker et;
1796 	struct ifnet *ifp;
1797 	int family, isr;
1798 
1799 	ifp = TUN2IFP(tp);
1800 	/* Could be unlocked read? */
1801 	TUN_LOCK(tp);
1802 	if (tp->tun_flags & TUN_IFHEAD) {
1803 		TUN_UNLOCK(tp);
1804 		if (m->m_len < sizeof(family) &&
1805 		(m = m_pullup(m, sizeof(family))) == NULL)
1806 			return (ENOBUFS);
1807 		family = ntohl(*mtod(m, u_int32_t *));
1808 		m_adj(m, sizeof(family));
1809 	} else {
1810 		TUN_UNLOCK(tp);
1811 		family = AF_INET;
1812 	}
1813 
1814 	BPF_MTAP2(ifp, &family, sizeof(family), m);
1815 
1816 	switch (family) {
1817 #ifdef INET
1818 	case AF_INET:
1819 		isr = NETISR_IP;
1820 		break;
1821 #endif
1822 #ifdef INET6
1823 	case AF_INET6:
1824 		isr = NETISR_IPV6;
1825 		break;
1826 #endif
1827 	default:
1828 		m_freem(m);
1829 		return (EAFNOSUPPORT);
1830 	}
1831 	random_harvest_queue(m, sizeof(*m), RANDOM_NET_TUN);
1832 	if_inc_counter(ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
1833 	if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1834 	CURVNET_SET(ifp->if_vnet);
1835 	M_SETFIB(m, ifp->if_fib);
1836 	NET_EPOCH_ENTER(et);
1837 	netisr_dispatch(isr, m);
1838 	NET_EPOCH_EXIT(et);
1839 	CURVNET_RESTORE();
1840 	return (0);
1841 }
1842 
1843 /*
1844  * the cdevsw write interface - an atomic write is a packet - or else!
1845  */
1846 static	int
1847 tunwrite(struct cdev *dev, struct uio *uio, int flag)
1848 {
1849 	struct virtio_net_hdr_mrg_rxbuf vhdr;
1850 	struct tuntap_softc *tp;
1851 	struct ifnet	*ifp;
1852 	struct mbuf	*m;
1853 	uint32_t	mru;
1854 	int		align, vhdrlen, error;
1855 	bool		l2tun;
1856 
1857 	tp = dev->si_drv1;
1858 	ifp = TUN2IFP(tp);
1859 	TUNDEBUG(ifp, "tunwrite\n");
1860 	if ((ifp->if_flags & IFF_UP) != IFF_UP)
1861 		/* ignore silently */
1862 		return (0);
1863 
1864 	if (uio->uio_resid == 0)
1865 		return (0);
1866 
1867 	l2tun = (tp->tun_flags & TUN_L2) != 0;
1868 	mru = l2tun ? TAPMRU : TUNMRU;
1869 	vhdrlen = tp->tun_vhdrlen;
1870 	align = 0;
1871 	if (l2tun) {
1872 		align = ETHER_ALIGN;
1873 		mru += vhdrlen;
1874 	} else if ((tp->tun_flags & TUN_IFHEAD) != 0)
1875 		mru += sizeof(uint32_t);	/* family */
1876 	if (uio->uio_resid < 0 || uio->uio_resid > mru) {
1877 		TUNDEBUG(ifp, "len=%zd!\n", uio->uio_resid);
1878 		return (EIO);
1879 	}
1880 
1881 	if (vhdrlen > 0) {
1882 		error = uiomove(&vhdr, vhdrlen, uio);
1883 		if (error != 0)
1884 			return (error);
1885 		TUNDEBUG(ifp, "txvhdr: f %u, gt %u, hl %u, "
1886 		    "gs %u, cs %u, co %u\n", vhdr.hdr.flags,
1887 		    vhdr.hdr.gso_type, vhdr.hdr.hdr_len,
1888 		    vhdr.hdr.gso_size, vhdr.hdr.csum_start,
1889 		    vhdr.hdr.csum_offset);
1890 	}
1891 
1892 	if ((m = m_uiotombuf(uio, M_NOWAIT, 0, align, M_PKTHDR)) == NULL) {
1893 		if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1894 		return (ENOBUFS);
1895 	}
1896 
1897 	m->m_pkthdr.rcvif = ifp;
1898 #ifdef MAC
1899 	mac_ifnet_create_mbuf(ifp, m);
1900 #endif
1901 
1902 	if (l2tun)
1903 		return (tunwrite_l2(tp, m, vhdrlen > 0 ? &vhdr : NULL));
1904 
1905 	return (tunwrite_l3(tp, m));
1906 }
1907 
1908 /*
1909  * tunpoll - the poll interface, this is only useful on reads
1910  * really. The write detect always returns true, write never blocks
1911  * anyway, it either accepts the packet or drops it.
1912  */
1913 static	int
1914 tunpoll(struct cdev *dev, int events, struct thread *td)
1915 {
1916 	struct tuntap_softc *tp = dev->si_drv1;
1917 	struct ifnet	*ifp = TUN2IFP(tp);
1918 	int		revents = 0;
1919 
1920 	TUNDEBUG(ifp, "tunpoll\n");
1921 
1922 	if (events & (POLLIN | POLLRDNORM)) {
1923 		IFQ_LOCK(&ifp->if_snd);
1924 		if (!IFQ_IS_EMPTY(&ifp->if_snd)) {
1925 			TUNDEBUG(ifp, "tunpoll q=%d\n", ifp->if_snd.ifq_len);
1926 			revents |= events & (POLLIN | POLLRDNORM);
1927 		} else {
1928 			TUNDEBUG(ifp, "tunpoll waiting\n");
1929 			selrecord(td, &tp->tun_rsel);
1930 		}
1931 		IFQ_UNLOCK(&ifp->if_snd);
1932 	}
1933 	revents |= events & (POLLOUT | POLLWRNORM);
1934 
1935 	return (revents);
1936 }
1937 
1938 /*
1939  * tunkqfilter - support for the kevent() system call.
1940  */
1941 static int
1942 tunkqfilter(struct cdev *dev, struct knote *kn)
1943 {
1944 	struct tuntap_softc	*tp = dev->si_drv1;
1945 	struct ifnet	*ifp = TUN2IFP(tp);
1946 
1947 	switch(kn->kn_filter) {
1948 	case EVFILT_READ:
1949 		TUNDEBUG(ifp, "%s kqfilter: EVFILT_READ, minor = %#x\n",
1950 		    ifp->if_xname, dev2unit(dev));
1951 		kn->kn_fop = &tun_read_filterops;
1952 		break;
1953 
1954 	case EVFILT_WRITE:
1955 		TUNDEBUG(ifp, "%s kqfilter: EVFILT_WRITE, minor = %#x\n",
1956 		    ifp->if_xname, dev2unit(dev));
1957 		kn->kn_fop = &tun_write_filterops;
1958 		break;
1959 
1960 	default:
1961 		TUNDEBUG(ifp, "%s kqfilter: invalid filter, minor = %#x\n",
1962 		    ifp->if_xname, dev2unit(dev));
1963 		return(EINVAL);
1964 	}
1965 
1966 	kn->kn_hook = tp;
1967 	knlist_add(&tp->tun_rsel.si_note, kn, 0);
1968 
1969 	return (0);
1970 }
1971 
1972 /*
1973  * Return true of there is data in the interface queue.
1974  */
1975 static int
1976 tunkqread(struct knote *kn, long hint)
1977 {
1978 	int			ret;
1979 	struct tuntap_softc	*tp = kn->kn_hook;
1980 	struct cdev		*dev = tp->tun_dev;
1981 	struct ifnet	*ifp = TUN2IFP(tp);
1982 
1983 	if ((kn->kn_data = ifp->if_snd.ifq_len) > 0) {
1984 		TUNDEBUG(ifp,
1985 		    "%s have data in the queue.  Len = %d, minor = %#x\n",
1986 		    ifp->if_xname, ifp->if_snd.ifq_len, dev2unit(dev));
1987 		ret = 1;
1988 	} else {
1989 		TUNDEBUG(ifp,
1990 		    "%s waiting for data, minor = %#x\n", ifp->if_xname,
1991 		    dev2unit(dev));
1992 		ret = 0;
1993 	}
1994 
1995 	return (ret);
1996 }
1997 
1998 /*
1999  * Always can write, always return MTU in kn->data.
2000  */
2001 static int
2002 tunkqwrite(struct knote *kn, long hint)
2003 {
2004 	struct tuntap_softc	*tp = kn->kn_hook;
2005 	struct ifnet	*ifp = TUN2IFP(tp);
2006 
2007 	kn->kn_data = ifp->if_mtu;
2008 
2009 	return (1);
2010 }
2011 
2012 static void
2013 tunkqdetach(struct knote *kn)
2014 {
2015 	struct tuntap_softc	*tp = kn->kn_hook;
2016 
2017 	knlist_remove(&tp->tun_rsel.si_note, kn, 0);
2018 }
2019