xref: /freebsd/sys/net/if_tuntap.c (revision 5ca8e32633c4ffbbcd6762e5888b6a4ba0708c6c)
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 <netinet/tcp_lro.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 	struct lro_ctrl		 tun_lro;	/* for TCP LRO */
149 	bool			 tun_lro_ready;	/* TCP LRO initialized */
150 };
151 #define	TUN2IFP(sc)	((sc)->tun_ifp)
152 
153 #define	TUNDEBUG	if (tundebug) if_printf
154 
155 #define	TUN_LOCK(tp)		mtx_lock(&(tp)->tun_mtx)
156 #define	TUN_UNLOCK(tp)		mtx_unlock(&(tp)->tun_mtx)
157 #define	TUN_LOCK_ASSERT(tp)	mtx_assert(&(tp)->tun_mtx, MA_OWNED);
158 
159 #define	TUN_VMIO_FLAG_MASK	0x0fff
160 
161 /*
162  * Interface capabilities of a tap device that supports the virtio-net
163  * header.
164  */
165 #define TAP_VNET_HDR_CAPS	(IFCAP_HWCSUM | IFCAP_HWCSUM_IPV6	\
166 				| IFCAP_VLAN_HWCSUM			\
167 				| IFCAP_TSO | IFCAP_LRO			\
168 				| IFCAP_VLAN_HWTSO)
169 
170 #define TAP_ALL_OFFLOAD		(CSUM_TSO | CSUM_TCP | CSUM_UDP |\
171 				    CSUM_TCP_IPV6 | CSUM_UDP_IPV6)
172 
173 /*
174  * All mutable global variables in if_tun are locked using tunmtx, with
175  * the exception of tundebug, which is used unlocked, and the drivers' *clones,
176  * which are static after setup.
177  */
178 static struct mtx tunmtx;
179 static eventhandler_tag arrival_tag;
180 static eventhandler_tag clone_tag;
181 static const char tunname[] = "tun";
182 static const char tapname[] = "tap";
183 static const char vmnetname[] = "vmnet";
184 static MALLOC_DEFINE(M_TUN, tunname, "Tunnel Interface");
185 static int tundebug = 0;
186 static int tundclone = 1;
187 static int tap_allow_uopen = 0;	/* allow user devfs cloning */
188 static int tapuponopen = 0;	/* IFF_UP on open() */
189 static int tapdclone = 1;	/* enable devfs cloning */
190 
191 static TAILQ_HEAD(,tuntap_softc)	tunhead = TAILQ_HEAD_INITIALIZER(tunhead);
192 SYSCTL_INT(_debug, OID_AUTO, if_tun_debug, CTLFLAG_RW, &tundebug, 0, "");
193 
194 static struct sx tun_ioctl_sx;
195 SX_SYSINIT(tun_ioctl_sx, &tun_ioctl_sx, "tun_ioctl");
196 
197 SYSCTL_DECL(_net_link);
198 /* tun */
199 static SYSCTL_NODE(_net_link, OID_AUTO, tun, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
200     "IP tunnel software network interface");
201 SYSCTL_INT(_net_link_tun, OID_AUTO, devfs_cloning, CTLFLAG_RWTUN, &tundclone, 0,
202     "Enable legacy devfs interface creation");
203 
204 /* tap */
205 static SYSCTL_NODE(_net_link, OID_AUTO, tap, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
206     "Ethernet tunnel software network interface");
207 SYSCTL_INT(_net_link_tap, OID_AUTO, user_open, CTLFLAG_RW, &tap_allow_uopen, 0,
208     "Enable legacy devfs interface creation for all users");
209 SYSCTL_INT(_net_link_tap, OID_AUTO, up_on_open, CTLFLAG_RW, &tapuponopen, 0,
210     "Bring interface up when /dev/tap is opened");
211 SYSCTL_INT(_net_link_tap, OID_AUTO, devfs_cloning, CTLFLAG_RWTUN, &tapdclone, 0,
212     "Enable legacy devfs interface creation");
213 SYSCTL_INT(_net_link_tap, OID_AUTO, debug, CTLFLAG_RW, &tundebug, 0, "");
214 
215 static int	tun_create_device(struct tuntap_driver *drv, int unit,
216     struct ucred *cr, struct cdev **dev, const char *name);
217 static int	tun_busy_locked(struct tuntap_softc *tp);
218 static void	tun_unbusy_locked(struct tuntap_softc *tp);
219 static int	tun_busy(struct tuntap_softc *tp);
220 static void	tun_unbusy(struct tuntap_softc *tp);
221 
222 static int	tuntap_name2info(const char *name, int *unit, int *flags);
223 static void	tunclone(void *arg, struct ucred *cred, char *name,
224 		    int namelen, struct cdev **dev);
225 static void	tuncreate(struct cdev *dev);
226 static void	tundtor(void *data);
227 static void	tunrename(void *arg, struct ifnet *ifp);
228 static int	tunifioctl(struct ifnet *, u_long, caddr_t);
229 static void	tuninit(struct ifnet *);
230 static void	tunifinit(void *xtp);
231 static int	tuntapmodevent(module_t, int, void *);
232 static int	tunoutput(struct ifnet *, struct mbuf *,
233 		    const struct sockaddr *, struct route *ro);
234 static void	tunstart(struct ifnet *);
235 static void	tunstart_l2(struct ifnet *);
236 
237 static int	tun_clone_match(struct if_clone *ifc, const char *name);
238 static int	tap_clone_match(struct if_clone *ifc, const char *name);
239 static int	vmnet_clone_match(struct if_clone *ifc, const char *name);
240 static int	tun_clone_create(struct if_clone *, char *, size_t,
241 		    struct ifc_data *, struct ifnet **);
242 static int	tun_clone_destroy(struct if_clone *, struct ifnet *, uint32_t);
243 static void	tun_vnethdr_set(struct ifnet *ifp, int vhdrlen);
244 
245 static d_open_t		tunopen;
246 static d_read_t		tunread;
247 static d_write_t	tunwrite;
248 static d_ioctl_t	tunioctl;
249 static d_poll_t		tunpoll;
250 static d_kqfilter_t	tunkqfilter;
251 
252 static int		tunkqread(struct knote *, long);
253 static int		tunkqwrite(struct knote *, long);
254 static void		tunkqdetach(struct knote *);
255 
256 static struct filterops tun_read_filterops = {
257 	.f_isfd =	1,
258 	.f_attach =	NULL,
259 	.f_detach =	tunkqdetach,
260 	.f_event =	tunkqread,
261 };
262 
263 static struct filterops tun_write_filterops = {
264 	.f_isfd =	1,
265 	.f_attach =	NULL,
266 	.f_detach =	tunkqdetach,
267 	.f_event =	tunkqwrite,
268 };
269 
270 static struct tuntap_driver {
271 	struct cdevsw		 cdevsw;
272 	int			 ident_flags;
273 	struct unrhdr		*unrhdr;
274 	struct clonedevs	*clones;
275 	ifc_match_f		*clone_match_fn;
276 	ifc_create_f		*clone_create_fn;
277 	ifc_destroy_f		*clone_destroy_fn;
278 } tuntap_drivers[] = {
279 	{
280 		.ident_flags =	0,
281 		.cdevsw =	{
282 		    .d_version =	D_VERSION,
283 		    .d_flags =		D_NEEDMINOR,
284 		    .d_open =		tunopen,
285 		    .d_read =		tunread,
286 		    .d_write =		tunwrite,
287 		    .d_ioctl =		tunioctl,
288 		    .d_poll =		tunpoll,
289 		    .d_kqfilter =	tunkqfilter,
290 		    .d_name =		tunname,
291 		},
292 		.clone_match_fn =	tun_clone_match,
293 		.clone_create_fn =	tun_clone_create,
294 		.clone_destroy_fn =	tun_clone_destroy,
295 	},
296 	{
297 		.ident_flags =	TUN_L2,
298 		.cdevsw =	{
299 		    .d_version =	D_VERSION,
300 		    .d_flags =		D_NEEDMINOR,
301 		    .d_open =		tunopen,
302 		    .d_read =		tunread,
303 		    .d_write =		tunwrite,
304 		    .d_ioctl =		tunioctl,
305 		    .d_poll =		tunpoll,
306 		    .d_kqfilter =	tunkqfilter,
307 		    .d_name =		tapname,
308 		},
309 		.clone_match_fn =	tap_clone_match,
310 		.clone_create_fn =	tun_clone_create,
311 		.clone_destroy_fn =	tun_clone_destroy,
312 	},
313 	{
314 		.ident_flags =	TUN_L2 | TUN_VMNET,
315 		.cdevsw =	{
316 		    .d_version =	D_VERSION,
317 		    .d_flags =		D_NEEDMINOR,
318 		    .d_open =		tunopen,
319 		    .d_read =		tunread,
320 		    .d_write =		tunwrite,
321 		    .d_ioctl =		tunioctl,
322 		    .d_poll =		tunpoll,
323 		    .d_kqfilter =	tunkqfilter,
324 		    .d_name =		vmnetname,
325 		},
326 		.clone_match_fn =	vmnet_clone_match,
327 		.clone_create_fn =	tun_clone_create,
328 		.clone_destroy_fn =	tun_clone_destroy,
329 	},
330 };
331 #define	NDRV	nitems(tuntap_drivers)
332 
333 VNET_DEFINE_STATIC(struct if_clone *, tuntap_driver_cloners[NDRV]);
334 #define	V_tuntap_driver_cloners	VNET(tuntap_driver_cloners)
335 
336 /*
337  * Mechanism for marking a tunnel device as busy so that we can safely do some
338  * orthogonal operations (such as operations on devices) without racing against
339  * tun_destroy.  tun_destroy will wait on the condvar if we're at all busy or
340  * open, to be woken up when the condition is alleviated.
341  */
342 static int
343 tun_busy_locked(struct tuntap_softc *tp)
344 {
345 
346 	TUN_LOCK_ASSERT(tp);
347 	if ((tp->tun_flags & TUN_DYING) != 0) {
348 		/*
349 		 * Perhaps unintuitive, but the device is busy going away.
350 		 * Other interpretations of EBUSY from tun_busy make little
351 		 * sense, since making a busy device even more busy doesn't
352 		 * sound like a problem.
353 		 */
354 		return (EBUSY);
355 	}
356 
357 	++tp->tun_busy;
358 	return (0);
359 }
360 
361 static void
362 tun_unbusy_locked(struct tuntap_softc *tp)
363 {
364 
365 	TUN_LOCK_ASSERT(tp);
366 	KASSERT(tp->tun_busy != 0, ("tun_unbusy: called for non-busy tunnel"));
367 
368 	--tp->tun_busy;
369 	/* Wake up anything that may be waiting on our busy tunnel. */
370 	if (tp->tun_busy == 0)
371 		cv_broadcast(&tp->tun_cv);
372 }
373 
374 static int
375 tun_busy(struct tuntap_softc *tp)
376 {
377 	int ret;
378 
379 	TUN_LOCK(tp);
380 	ret = tun_busy_locked(tp);
381 	TUN_UNLOCK(tp);
382 	return (ret);
383 }
384 
385 static void
386 tun_unbusy(struct tuntap_softc *tp)
387 {
388 
389 	TUN_LOCK(tp);
390 	tun_unbusy_locked(tp);
391 	TUN_UNLOCK(tp);
392 }
393 
394 /*
395  * Sets unit and/or flags given the device name.  Must be called with correct
396  * vnet context.
397  */
398 static int
399 tuntap_name2info(const char *name, int *outunit, int *outflags)
400 {
401 	struct tuntap_driver *drv;
402 	char *dname;
403 	int flags, unit;
404 	bool found;
405 
406 	if (name == NULL)
407 		return (EINVAL);
408 
409 	/*
410 	 * Needed for dev_stdclone, but dev_stdclone will not modify, it just
411 	 * wants to be able to pass back a char * through the second param. We
412 	 * will always set that as NULL here, so we'll fake it.
413 	 */
414 	dname = __DECONST(char *, name);
415 	found = false;
416 
417 	for (u_int i = 0; i < NDRV; i++) {
418 		drv = &tuntap_drivers[i];
419 
420 		if (strcmp(name, drv->cdevsw.d_name) == 0) {
421 			found = true;
422 			unit = -1;
423 			flags = drv->ident_flags;
424 			break;
425 		}
426 
427 		if (dev_stdclone(dname, NULL, drv->cdevsw.d_name, &unit) == 1) {
428 			found = true;
429 			flags = drv->ident_flags;
430 			break;
431 		}
432 	}
433 
434 	if (!found)
435 		return (ENXIO);
436 
437 	if (outunit != NULL)
438 		*outunit = unit;
439 	if (outflags != NULL)
440 		*outflags = flags;
441 	return (0);
442 }
443 
444 /*
445  * Get driver information from a set of flags specified.  Masks the identifying
446  * part of the flags and compares it against all of the available
447  * tuntap_drivers.
448  */
449 static struct tuntap_driver *
450 tuntap_driver_from_flags(int tun_flags)
451 {
452 
453 	for (u_int i = 0; i < NDRV; i++)
454 		if ((tun_flags & TUN_DRIVER_IDENT_MASK) ==
455 		    tuntap_drivers[i].ident_flags)
456 			return (&tuntap_drivers[i]);
457 
458 	return (NULL);
459 }
460 
461 static int
462 tun_clone_match(struct if_clone *ifc, const char *name)
463 {
464 	int tunflags;
465 
466 	if (tuntap_name2info(name, NULL, &tunflags) == 0) {
467 		if ((tunflags & TUN_L2) == 0)
468 			return (1);
469 	}
470 
471 	return (0);
472 }
473 
474 static int
475 tap_clone_match(struct if_clone *ifc, const char *name)
476 {
477 	int tunflags;
478 
479 	if (tuntap_name2info(name, NULL, &tunflags) == 0) {
480 		if ((tunflags & (TUN_L2 | TUN_VMNET)) == TUN_L2)
481 			return (1);
482 	}
483 
484 	return (0);
485 }
486 
487 static int
488 vmnet_clone_match(struct if_clone *ifc, const char *name)
489 {
490 	int tunflags;
491 
492 	if (tuntap_name2info(name, NULL, &tunflags) == 0) {
493 		if ((tunflags & TUN_VMNET) != 0)
494 			return (1);
495 	}
496 
497 	return (0);
498 }
499 
500 static int
501 tun_clone_create(struct if_clone *ifc, char *name, size_t len,
502     struct ifc_data *ifd, struct ifnet **ifpp)
503 {
504 	struct tuntap_driver *drv;
505 	struct cdev *dev;
506 	int err, i, tunflags, unit;
507 
508 	tunflags = 0;
509 	/* The name here tells us exactly what we're creating */
510 	err = tuntap_name2info(name, &unit, &tunflags);
511 	if (err != 0)
512 		return (err);
513 
514 	drv = tuntap_driver_from_flags(tunflags);
515 	if (drv == NULL)
516 		return (ENXIO);
517 
518 	if (unit != -1) {
519 		/* If this unit number is still available that's okay. */
520 		if (alloc_unr_specific(drv->unrhdr, unit) == -1)
521 			return (EEXIST);
522 	} else {
523 		unit = alloc_unr(drv->unrhdr);
524 	}
525 
526 	snprintf(name, IFNAMSIZ, "%s%d", drv->cdevsw.d_name, unit);
527 
528 	/* find any existing device, or allocate new unit number */
529 	dev = NULL;
530 	i = clone_create(&drv->clones, &drv->cdevsw, &unit, &dev, 0);
531 	/* No preexisting struct cdev *, create one */
532 	if (i != 0)
533 		i = tun_create_device(drv, unit, NULL, &dev, name);
534 	if (i == 0) {
535 		dev_ref(dev);
536 		tuncreate(dev);
537 		struct tuntap_softc *tp = dev->si_drv1;
538 		*ifpp = tp->tun_ifp;
539 	}
540 
541 	return (i);
542 }
543 
544 static void
545 tunclone(void *arg, struct ucred *cred, char *name, int namelen,
546     struct cdev **dev)
547 {
548 	char devname[SPECNAMELEN + 1];
549 	struct tuntap_driver *drv;
550 	int append_unit, i, u, tunflags;
551 	bool mayclone;
552 
553 	if (*dev != NULL)
554 		return;
555 
556 	tunflags = 0;
557 	CURVNET_SET(CRED_TO_VNET(cred));
558 	if (tuntap_name2info(name, &u, &tunflags) != 0)
559 		goto out;	/* Not recognized */
560 
561 	if (u != -1 && u > IF_MAXUNIT)
562 		goto out;	/* Unit number too high */
563 
564 	mayclone = priv_check_cred(cred, PRIV_NET_IFCREATE) == 0;
565 	if ((tunflags & TUN_L2) != 0) {
566 		/* tap/vmnet allow user open with a sysctl */
567 		mayclone = (mayclone || tap_allow_uopen) && tapdclone;
568 	} else {
569 		mayclone = mayclone && tundclone;
570 	}
571 
572 	/*
573 	 * If tun cloning is enabled, only the superuser can create an
574 	 * interface.
575 	 */
576 	if (!mayclone)
577 		goto out;
578 
579 	if (u == -1)
580 		append_unit = 1;
581 	else
582 		append_unit = 0;
583 
584 	drv = tuntap_driver_from_flags(tunflags);
585 	if (drv == NULL)
586 		goto out;
587 
588 	/* find any existing device, or allocate new unit number */
589 	i = clone_create(&drv->clones, &drv->cdevsw, &u, dev, 0);
590 	if (i) {
591 		if (append_unit) {
592 			namelen = snprintf(devname, sizeof(devname), "%s%d",
593 			    name, u);
594 			name = devname;
595 		}
596 
597 		i = tun_create_device(drv, u, cred, dev, name);
598 	}
599 	if (i == 0) {
600 		dev_ref(*dev);
601 		if_clone_create(name, namelen, NULL);
602 	}
603 out:
604 	CURVNET_RESTORE();
605 }
606 
607 static void
608 tun_destroy(struct tuntap_softc *tp)
609 {
610 
611 	TUN_LOCK(tp);
612 	tp->tun_flags |= TUN_DYING;
613 	if (tp->tun_busy != 0)
614 		cv_wait_unlock(&tp->tun_cv, &tp->tun_mtx);
615 	else
616 		TUN_UNLOCK(tp);
617 
618 	CURVNET_SET(TUN2IFP(tp)->if_vnet);
619 
620 	/* destroy_dev will take care of any alias. */
621 	destroy_dev(tp->tun_dev);
622 	seldrain(&tp->tun_rsel);
623 	knlist_clear(&tp->tun_rsel.si_note, 0);
624 	knlist_destroy(&tp->tun_rsel.si_note);
625 	if ((tp->tun_flags & TUN_L2) != 0) {
626 		ether_ifdetach(TUN2IFP(tp));
627 	} else {
628 		bpfdetach(TUN2IFP(tp));
629 		if_detach(TUN2IFP(tp));
630 	}
631 	sx_xlock(&tun_ioctl_sx);
632 	TUN2IFP(tp)->if_softc = NULL;
633 	sx_xunlock(&tun_ioctl_sx);
634 	free_unr(tp->tun_drv->unrhdr, TUN2IFP(tp)->if_dunit);
635 	if_free(TUN2IFP(tp));
636 	mtx_destroy(&tp->tun_mtx);
637 	cv_destroy(&tp->tun_cv);
638 	free(tp, M_TUN);
639 	CURVNET_RESTORE();
640 }
641 
642 static int
643 tun_clone_destroy(struct if_clone *ifc __unused, struct ifnet *ifp, uint32_t flags)
644 {
645 	struct tuntap_softc *tp = ifp->if_softc;
646 
647 	mtx_lock(&tunmtx);
648 	TAILQ_REMOVE(&tunhead, tp, tun_list);
649 	mtx_unlock(&tunmtx);
650 	tun_destroy(tp);
651 
652 	return (0);
653 }
654 
655 static void
656 vnet_tun_init(const void *unused __unused)
657 {
658 
659 	for (u_int i = 0; i < NDRV; ++i) {
660 		struct if_clone_addreq req = {
661 			.match_f = tuntap_drivers[i].clone_match_fn,
662 			.create_f = tuntap_drivers[i].clone_create_fn,
663 			.destroy_f = tuntap_drivers[i].clone_destroy_fn,
664 		};
665 		V_tuntap_driver_cloners[i] =
666 		    ifc_attach_cloner(tuntap_drivers[i].cdevsw.d_name, &req);
667 	};
668 }
669 VNET_SYSINIT(vnet_tun_init, SI_SUB_PROTO_IF, SI_ORDER_ANY,
670 		vnet_tun_init, NULL);
671 
672 static void
673 vnet_tun_uninit(const void *unused __unused)
674 {
675 
676 	for (u_int i = 0; i < NDRV; ++i)
677 		if_clone_detach(V_tuntap_driver_cloners[i]);
678 }
679 VNET_SYSUNINIT(vnet_tun_uninit, SI_SUB_PROTO_IF, SI_ORDER_ANY,
680     vnet_tun_uninit, NULL);
681 
682 static void
683 tun_uninit(const void *unused __unused)
684 {
685 	struct tuntap_driver *drv;
686 	struct tuntap_softc *tp;
687 	int i;
688 
689 	EVENTHANDLER_DEREGISTER(ifnet_arrival_event, arrival_tag);
690 	EVENTHANDLER_DEREGISTER(dev_clone, clone_tag);
691 
692 	mtx_lock(&tunmtx);
693 	while ((tp = TAILQ_FIRST(&tunhead)) != NULL) {
694 		TAILQ_REMOVE(&tunhead, tp, tun_list);
695 		mtx_unlock(&tunmtx);
696 		tun_destroy(tp);
697 		mtx_lock(&tunmtx);
698 	}
699 	mtx_unlock(&tunmtx);
700 	for (i = 0; i < nitems(tuntap_drivers); ++i) {
701 		drv = &tuntap_drivers[i];
702 		delete_unrhdr(drv->unrhdr);
703 		clone_cleanup(&drv->clones);
704 	}
705 	mtx_destroy(&tunmtx);
706 }
707 SYSUNINIT(tun_uninit, SI_SUB_PROTO_IF, SI_ORDER_ANY, tun_uninit, NULL);
708 
709 static struct tuntap_driver *
710 tuntap_driver_from_ifnet(const struct ifnet *ifp)
711 {
712 	struct tuntap_driver *drv;
713 	int i;
714 
715 	if (ifp == NULL)
716 		return (NULL);
717 
718 	for (i = 0; i < nitems(tuntap_drivers); ++i) {
719 		drv = &tuntap_drivers[i];
720 		if (strcmp(ifp->if_dname, drv->cdevsw.d_name) == 0)
721 			return (drv);
722 	}
723 
724 	return (NULL);
725 }
726 
727 static int
728 tuntapmodevent(module_t mod, int type, void *data)
729 {
730 	struct tuntap_driver *drv;
731 	int i;
732 
733 	switch (type) {
734 	case MOD_LOAD:
735 		mtx_init(&tunmtx, "tunmtx", NULL, MTX_DEF);
736 		for (i = 0; i < nitems(tuntap_drivers); ++i) {
737 			drv = &tuntap_drivers[i];
738 			clone_setup(&drv->clones);
739 			drv->unrhdr = new_unrhdr(0, IF_MAXUNIT, &tunmtx);
740 		}
741 		arrival_tag = EVENTHANDLER_REGISTER(ifnet_arrival_event,
742 		   tunrename, 0, 1000);
743 		if (arrival_tag == NULL)
744 			return (ENOMEM);
745 		clone_tag = EVENTHANDLER_REGISTER(dev_clone, tunclone, 0, 1000);
746 		if (clone_tag == NULL)
747 			return (ENOMEM);
748 		break;
749 	case MOD_UNLOAD:
750 		/* See tun_uninit, so it's done after the vnet_sysuninit() */
751 		break;
752 	default:
753 		return EOPNOTSUPP;
754 	}
755 	return 0;
756 }
757 
758 static moduledata_t tuntap_mod = {
759 	"if_tuntap",
760 	tuntapmodevent,
761 	0
762 };
763 
764 /* We'll only ever have these two, so no need for a macro. */
765 static moduledata_t tun_mod = { "if_tun", NULL, 0 };
766 static moduledata_t tap_mod = { "if_tap", NULL, 0 };
767 
768 DECLARE_MODULE(if_tuntap, tuntap_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
769 MODULE_VERSION(if_tuntap, 1);
770 DECLARE_MODULE(if_tun, tun_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
771 MODULE_VERSION(if_tun, 1);
772 DECLARE_MODULE(if_tap, tap_mod, SI_SUB_PSEUDO, SI_ORDER_ANY);
773 MODULE_VERSION(if_tap, 1);
774 
775 static int
776 tun_create_device(struct tuntap_driver *drv, int unit, struct ucred *cr,
777     struct cdev **dev, const char *name)
778 {
779 	struct make_dev_args args;
780 	struct tuntap_softc *tp;
781 	int error;
782 
783 	tp = malloc(sizeof(*tp), M_TUN, M_WAITOK | M_ZERO);
784 	mtx_init(&tp->tun_mtx, "tun_mtx", NULL, MTX_DEF);
785 	cv_init(&tp->tun_cv, "tun_condvar");
786 	tp->tun_flags = drv->ident_flags;
787 	tp->tun_drv = drv;
788 
789 	make_dev_args_init(&args);
790 	if (cr != NULL)
791 		args.mda_flags = MAKEDEV_REF | MAKEDEV_CHECKNAME;
792 	args.mda_devsw = &drv->cdevsw;
793 	args.mda_cr = cr;
794 	args.mda_uid = UID_UUCP;
795 	args.mda_gid = GID_DIALER;
796 	args.mda_mode = 0600;
797 	args.mda_unit = unit;
798 	args.mda_si_drv1 = tp;
799 	error = make_dev_s(&args, dev, "%s", name);
800 	if (error != 0) {
801 		free(tp, M_TUN);
802 		return (error);
803 	}
804 
805 	KASSERT((*dev)->si_drv1 != NULL,
806 	    ("Failed to set si_drv1 at %s creation", name));
807 	tp->tun_dev = *dev;
808 	knlist_init_mtx(&tp->tun_rsel.si_note, &tp->tun_mtx);
809 	mtx_lock(&tunmtx);
810 	TAILQ_INSERT_TAIL(&tunhead, tp, tun_list);
811 	mtx_unlock(&tunmtx);
812 	return (0);
813 }
814 
815 static void
816 tunstart(struct ifnet *ifp)
817 {
818 	struct tuntap_softc *tp = ifp->if_softc;
819 	struct mbuf *m;
820 
821 	TUNDEBUG(ifp, "starting\n");
822 	if (ALTQ_IS_ENABLED(&ifp->if_snd)) {
823 		IFQ_LOCK(&ifp->if_snd);
824 		IFQ_POLL_NOLOCK(&ifp->if_snd, m);
825 		if (m == NULL) {
826 			IFQ_UNLOCK(&ifp->if_snd);
827 			return;
828 		}
829 		IFQ_UNLOCK(&ifp->if_snd);
830 	}
831 
832 	TUN_LOCK(tp);
833 	if (tp->tun_flags & TUN_RWAIT) {
834 		tp->tun_flags &= ~TUN_RWAIT;
835 		wakeup(tp);
836 	}
837 	selwakeuppri(&tp->tun_rsel, PZERO + 1);
838 	KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
839 	if (tp->tun_flags & TUN_ASYNC && tp->tun_sigio) {
840 		TUN_UNLOCK(tp);
841 		pgsigio(&tp->tun_sigio, SIGIO, 0);
842 	} else
843 		TUN_UNLOCK(tp);
844 }
845 
846 /*
847  * tunstart_l2
848  *
849  * queue packets from higher level ready to put out
850  */
851 static void
852 tunstart_l2(struct ifnet *ifp)
853 {
854 	struct tuntap_softc	*tp = ifp->if_softc;
855 
856 	TUNDEBUG(ifp, "starting\n");
857 
858 	/*
859 	 * do not junk pending output if we are in VMnet mode.
860 	 * XXX: can this do any harm because of queue overflow?
861 	 */
862 
863 	TUN_LOCK(tp);
864 	if (((tp->tun_flags & TUN_VMNET) == 0) &&
865 	    ((tp->tun_flags & TUN_READY) != TUN_READY)) {
866 		struct mbuf *m;
867 
868 		/* Unlocked read. */
869 		TUNDEBUG(ifp, "not ready, tun_flags = 0x%x\n", tp->tun_flags);
870 
871 		for (;;) {
872 			IF_DEQUEUE(&ifp->if_snd, m);
873 			if (m != NULL) {
874 				m_freem(m);
875 				if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
876 			} else
877 				break;
878 		}
879 		TUN_UNLOCK(tp);
880 
881 		return;
882 	}
883 
884 	ifp->if_drv_flags |= IFF_DRV_OACTIVE;
885 
886 	if (!IFQ_IS_EMPTY(&ifp->if_snd)) {
887 		if (tp->tun_flags & TUN_RWAIT) {
888 			tp->tun_flags &= ~TUN_RWAIT;
889 			wakeup(tp);
890 		}
891 
892 		if ((tp->tun_flags & TUN_ASYNC) && (tp->tun_sigio != NULL)) {
893 			TUN_UNLOCK(tp);
894 			pgsigio(&tp->tun_sigio, SIGIO, 0);
895 			TUN_LOCK(tp);
896 		}
897 
898 		selwakeuppri(&tp->tun_rsel, PZERO+1);
899 		KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
900 		if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1); /* obytes are counted in ether_output */
901 	}
902 
903 	ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
904 	TUN_UNLOCK(tp);
905 } /* tunstart_l2 */
906 
907 static int
908 tap_transmit(struct ifnet *ifp, struct mbuf *m)
909 {
910 	int error;
911 
912 	BPF_MTAP(ifp, m);
913 	IFQ_HANDOFF(ifp, m, error);
914 	return (error);
915 }
916 
917 /* XXX: should return an error code so it can fail. */
918 static void
919 tuncreate(struct cdev *dev)
920 {
921 	struct tuntap_driver *drv;
922 	struct tuntap_softc *tp;
923 	struct ifnet *ifp;
924 	struct ether_addr eaddr;
925 	int iflags;
926 	u_char type;
927 
928 	tp = dev->si_drv1;
929 	KASSERT(tp != NULL,
930 	    ("si_drv1 should have been initialized at creation"));
931 
932 	drv = tp->tun_drv;
933 	iflags = IFF_MULTICAST;
934 	if ((tp->tun_flags & TUN_L2) != 0) {
935 		type = IFT_ETHER;
936 		iflags |= IFF_BROADCAST | IFF_SIMPLEX;
937 	} else {
938 		type = IFT_PPP;
939 		iflags |= IFF_POINTOPOINT;
940 	}
941 	ifp = tp->tun_ifp = if_alloc(type);
942 	if (ifp == NULL)
943 		panic("%s%d: failed to if_alloc() interface.\n",
944 		    drv->cdevsw.d_name, dev2unit(dev));
945 	ifp->if_softc = tp;
946 	if_initname(ifp, drv->cdevsw.d_name, dev2unit(dev));
947 	ifp->if_ioctl = tunifioctl;
948 	ifp->if_flags = iflags;
949 	IFQ_SET_MAXLEN(&ifp->if_snd, ifqmaxlen);
950 	ifp->if_capabilities |= IFCAP_LINKSTATE;
951 	if ((tp->tun_flags & TUN_L2) != 0)
952 		ifp->if_capabilities |=
953 		    IFCAP_RXCSUM | IFCAP_RXCSUM_IPV6 | IFCAP_LRO;
954 	ifp->if_capenable |= IFCAP_LINKSTATE;
955 
956 	if ((tp->tun_flags & TUN_L2) != 0) {
957 		ifp->if_init = tunifinit;
958 		ifp->if_start = tunstart_l2;
959 		ifp->if_transmit = tap_transmit;
960 		ifp->if_qflush = if_qflush;
961 
962 		ether_gen_addr(ifp, &eaddr);
963 		ether_ifattach(ifp, eaddr.octet);
964 	} else {
965 		ifp->if_mtu = TUNMTU;
966 		ifp->if_start = tunstart;
967 		ifp->if_output = tunoutput;
968 
969 		ifp->if_snd.ifq_drv_maxlen = 0;
970 		IFQ_SET_READY(&ifp->if_snd);
971 
972 		if_attach(ifp);
973 		bpfattach(ifp, DLT_NULL, sizeof(u_int32_t));
974 	}
975 
976 	TUN_LOCK(tp);
977 	tp->tun_flags |= TUN_INITED;
978 	TUN_UNLOCK(tp);
979 
980 	TUNDEBUG(ifp, "interface %s is created, minor = %#x\n",
981 	    ifp->if_xname, dev2unit(dev));
982 }
983 
984 static void
985 tunrename(void *arg __unused, struct ifnet *ifp)
986 {
987 	struct tuntap_softc *tp;
988 	int error;
989 
990 	if ((ifp->if_flags & IFF_RENAMING) == 0)
991 		return;
992 
993 	if (tuntap_driver_from_ifnet(ifp) == NULL)
994 		return;
995 
996 	/*
997 	 * We need to grab the ioctl sx long enough to make sure the softc is
998 	 * still there.  If it is, we can safely try to busy the tun device.
999 	 * The busy may fail if the device is currently dying, in which case
1000 	 * we do nothing.  If it doesn't fail, the busy count stops the device
1001 	 * from dying until we've created the alias (that will then be
1002 	 * subsequently destroyed).
1003 	 */
1004 	sx_xlock(&tun_ioctl_sx);
1005 	tp = ifp->if_softc;
1006 	if (tp == NULL) {
1007 		sx_xunlock(&tun_ioctl_sx);
1008 		return;
1009 	}
1010 	error = tun_busy(tp);
1011 	sx_xunlock(&tun_ioctl_sx);
1012 	if (error != 0)
1013 		return;
1014 	if (tp->tun_alias != NULL) {
1015 		destroy_dev(tp->tun_alias);
1016 		tp->tun_alias = NULL;
1017 	}
1018 
1019 	if (strcmp(ifp->if_xname, tp->tun_dev->si_name) == 0)
1020 		goto out;
1021 
1022 	/*
1023 	 * Failure's ok, aliases are created on a best effort basis.  If a
1024 	 * tun user/consumer decides to rename the interface to conflict with
1025 	 * another device (non-ifnet) on the system, we will assume they know
1026 	 * what they are doing.  make_dev_alias_p won't touch tun_alias on
1027 	 * failure, so we use it but ignore the return value.
1028 	 */
1029 	make_dev_alias_p(MAKEDEV_CHECKNAME, &tp->tun_alias, tp->tun_dev, "%s",
1030 	    ifp->if_xname);
1031 out:
1032 	tun_unbusy(tp);
1033 }
1034 
1035 static int
1036 tunopen(struct cdev *dev, int flag, int mode, struct thread *td)
1037 {
1038 	struct ifnet	*ifp;
1039 	struct tuntap_softc *tp;
1040 	int error __diagused, tunflags;
1041 
1042 	tunflags = 0;
1043 	CURVNET_SET(TD_TO_VNET(td));
1044 	error = tuntap_name2info(dev->si_name, NULL, &tunflags);
1045 	if (error != 0) {
1046 		CURVNET_RESTORE();
1047 		return (error);	/* Shouldn't happen */
1048 	}
1049 
1050 	tp = dev->si_drv1;
1051 	KASSERT(tp != NULL,
1052 	    ("si_drv1 should have been initialized at creation"));
1053 
1054 	TUN_LOCK(tp);
1055 	if ((tp->tun_flags & TUN_INITED) == 0) {
1056 		TUN_UNLOCK(tp);
1057 		CURVNET_RESTORE();
1058 		return (ENXIO);
1059 	}
1060 	if ((tp->tun_flags & (TUN_OPEN | TUN_DYING)) != 0) {
1061 		TUN_UNLOCK(tp);
1062 		CURVNET_RESTORE();
1063 		return (EBUSY);
1064 	}
1065 
1066 	error = tun_busy_locked(tp);
1067 	KASSERT(error == 0, ("Must be able to busy an unopen tunnel"));
1068 	ifp = TUN2IFP(tp);
1069 
1070 	if ((tp->tun_flags & TUN_L2) != 0) {
1071 		bcopy(IF_LLADDR(ifp), tp->tun_ether.octet,
1072 		    sizeof(tp->tun_ether.octet));
1073 
1074 		ifp->if_drv_flags |= IFF_DRV_RUNNING;
1075 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1076 
1077 		if (tapuponopen)
1078 			ifp->if_flags |= IFF_UP;
1079 	}
1080 
1081 	tp->tun_pid = td->td_proc->p_pid;
1082 	tp->tun_flags |= TUN_OPEN;
1083 
1084 	if_link_state_change(ifp, LINK_STATE_UP);
1085 	TUNDEBUG(ifp, "open\n");
1086 	TUN_UNLOCK(tp);
1087 
1088 	/*
1089 	 * This can fail with either ENOENT or EBUSY.  This is in the middle of
1090 	 * d_open, so ENOENT should not be possible.  EBUSY is possible, but
1091 	 * the only cdevpriv dtor being set will be tundtor and the softc being
1092 	 * passed is constant for a given cdev.  We ignore the possible error
1093 	 * because of this as either "unlikely" or "not actually a problem."
1094 	 */
1095 	(void)devfs_set_cdevpriv(tp, tundtor);
1096 	CURVNET_RESTORE();
1097 	return (0);
1098 }
1099 
1100 /*
1101  * tundtor - tear down the device - mark i/f down & delete
1102  * routing info
1103  */
1104 static void
1105 tundtor(void *data)
1106 {
1107 	struct proc *p;
1108 	struct tuntap_softc *tp;
1109 	struct ifnet *ifp;
1110 	bool l2tun;
1111 
1112 	tp = data;
1113 	p = curproc;
1114 	ifp = TUN2IFP(tp);
1115 
1116 	TUN_LOCK(tp);
1117 
1118 	/*
1119 	 * Realistically, we can't be obstinate here.  This only means that the
1120 	 * tuntap device was closed out of order, and the last closer wasn't the
1121 	 * controller.  These are still good to know about, though, as software
1122 	 * should avoid multiple processes with a tuntap device open and
1123 	 * ill-defined transfer of control (e.g., handoff, TUNSIFPID, close in
1124 	 * parent).
1125 	 */
1126 	if (p->p_pid != tp->tun_pid) {
1127 		log(LOG_INFO,
1128 		    "pid %d (%s), %s: tun/tap protocol violation, non-controlling process closed last.\n",
1129 		    p->p_pid, p->p_comm, tp->tun_dev->si_name);
1130 	}
1131 
1132 	/*
1133 	 * junk all pending output
1134 	 */
1135 	CURVNET_SET(ifp->if_vnet);
1136 
1137 	l2tun = false;
1138 	if ((tp->tun_flags & TUN_L2) != 0) {
1139 		l2tun = true;
1140 		IF_DRAIN(&ifp->if_snd);
1141 	} else {
1142 		IFQ_PURGE(&ifp->if_snd);
1143 	}
1144 
1145 	/* For vmnet, we won't do most of the address/route bits */
1146 	if ((tp->tun_flags & TUN_VMNET) != 0 ||
1147 	    (l2tun && (ifp->if_flags & IFF_LINK0) != 0))
1148 		goto out;
1149 #if defined(INET) || defined(INET6)
1150 	if (l2tun && tp->tun_lro_ready) {
1151 		TUNDEBUG (ifp, "LRO disabled\n");
1152 		tcp_lro_free(&tp->tun_lro);
1153 		tp->tun_lro_ready = false;
1154 	}
1155 #endif
1156 	if (ifp->if_flags & IFF_UP) {
1157 		TUN_UNLOCK(tp);
1158 		if_down(ifp);
1159 		TUN_LOCK(tp);
1160 	}
1161 
1162 	/* Delete all addresses and routes which reference this interface. */
1163 	if (ifp->if_drv_flags & IFF_DRV_RUNNING) {
1164 		ifp->if_drv_flags &= ~IFF_DRV_RUNNING;
1165 		TUN_UNLOCK(tp);
1166 		if_purgeaddrs(ifp);
1167 		TUN_LOCK(tp);
1168 	}
1169 
1170 out:
1171 	if_link_state_change(ifp, LINK_STATE_DOWN);
1172 	CURVNET_RESTORE();
1173 
1174 	funsetown(&tp->tun_sigio);
1175 	selwakeuppri(&tp->tun_rsel, PZERO + 1);
1176 	KNOTE_LOCKED(&tp->tun_rsel.si_note, 0);
1177 	TUNDEBUG (ifp, "closed\n");
1178 	tp->tun_flags &= ~TUN_OPEN;
1179 	tp->tun_pid = 0;
1180 	tun_vnethdr_set(ifp, 0);
1181 
1182 	tun_unbusy_locked(tp);
1183 	TUN_UNLOCK(tp);
1184 }
1185 
1186 static void
1187 tuninit(struct ifnet *ifp)
1188 {
1189 	struct tuntap_softc *tp = ifp->if_softc;
1190 
1191 	TUNDEBUG(ifp, "tuninit\n");
1192 
1193 	TUN_LOCK(tp);
1194 	ifp->if_drv_flags |= IFF_DRV_RUNNING;
1195 	if ((tp->tun_flags & TUN_L2) == 0) {
1196 		ifp->if_flags |= IFF_UP;
1197 		getmicrotime(&ifp->if_lastchange);
1198 		TUN_UNLOCK(tp);
1199 	} else {
1200 #if defined(INET) || defined(INET6)
1201 		if (tcp_lro_init(&tp->tun_lro) == 0) {
1202 			TUNDEBUG(ifp, "LRO enabled\n");
1203 			tp->tun_lro.ifp = ifp;
1204 			tp->tun_lro_ready = true;
1205 		} else {
1206 			TUNDEBUG(ifp, "Could not enable LRO\n");
1207 			tp->tun_lro_ready = false;
1208 		}
1209 #endif
1210 		ifp->if_drv_flags &= ~IFF_DRV_OACTIVE;
1211 		TUN_UNLOCK(tp);
1212 		/* attempt to start output */
1213 		tunstart_l2(ifp);
1214 	}
1215 
1216 }
1217 
1218 /*
1219  * Used only for l2 tunnel.
1220  */
1221 static void
1222 tunifinit(void *xtp)
1223 {
1224 	struct tuntap_softc *tp;
1225 
1226 	tp = (struct tuntap_softc *)xtp;
1227 	tuninit(tp->tun_ifp);
1228 }
1229 
1230 /*
1231  * To be called under TUN_LOCK. Update ifp->if_hwassist according to the
1232  * current value of ifp->if_capenable.
1233  */
1234 static void
1235 tun_caps_changed(struct ifnet *ifp)
1236 {
1237 	uint64_t hwassist = 0;
1238 
1239 	TUN_LOCK_ASSERT((struct tuntap_softc *)ifp->if_softc);
1240 	if (ifp->if_capenable & IFCAP_TXCSUM)
1241 		hwassist |= CSUM_TCP | CSUM_UDP;
1242 	if (ifp->if_capenable & IFCAP_TXCSUM_IPV6)
1243 		hwassist |= CSUM_TCP_IPV6
1244 		    | CSUM_UDP_IPV6;
1245 	if (ifp->if_capenable & IFCAP_TSO4)
1246 		hwassist |= CSUM_IP_TSO;
1247 	if (ifp->if_capenable & IFCAP_TSO6)
1248 		hwassist |= CSUM_IP6_TSO;
1249 	ifp->if_hwassist = hwassist;
1250 }
1251 
1252 /*
1253  * To be called under TUN_LOCK. Update tp->tun_vhdrlen and adjust
1254  * if_capabilities and if_capenable as needed.
1255  */
1256 static void
1257 tun_vnethdr_set(struct ifnet *ifp, int vhdrlen)
1258 {
1259 	struct tuntap_softc *tp = ifp->if_softc;
1260 
1261 	TUN_LOCK_ASSERT(tp);
1262 
1263 	if (tp->tun_vhdrlen == vhdrlen)
1264 		return;
1265 
1266 	/*
1267 	 * Update if_capabilities to reflect the
1268 	 * functionalities offered by the virtio-net
1269 	 * header.
1270 	 */
1271 	if (vhdrlen != 0)
1272 		ifp->if_capabilities |=
1273 			TAP_VNET_HDR_CAPS;
1274 	else
1275 		ifp->if_capabilities &=
1276 			~TAP_VNET_HDR_CAPS;
1277 	/*
1278 	 * Disable any capabilities that we don't
1279 	 * support anymore.
1280 	 */
1281 	ifp->if_capenable &= ifp->if_capabilities;
1282 	tun_caps_changed(ifp);
1283 	tp->tun_vhdrlen = vhdrlen;
1284 
1285 	TUNDEBUG(ifp, "vnet_hdr_len=%d, if_capabilities=%x\n",
1286 	    vhdrlen, ifp->if_capabilities);
1287 }
1288 
1289 /*
1290  * Process an ioctl request.
1291  */
1292 static int
1293 tunifioctl(struct ifnet *ifp, u_long cmd, caddr_t data)
1294 {
1295 	struct ifreq *ifr = (struct ifreq *)data;
1296 	struct tuntap_softc *tp;
1297 	struct ifstat *ifs;
1298 	struct ifmediareq	*ifmr;
1299 	int		dummy, error = 0;
1300 	bool		l2tun;
1301 
1302 	ifmr = NULL;
1303 	sx_xlock(&tun_ioctl_sx);
1304 	tp = ifp->if_softc;
1305 	if (tp == NULL) {
1306 		error = ENXIO;
1307 		goto bad;
1308 	}
1309 	l2tun = (tp->tun_flags & TUN_L2) != 0;
1310 	switch(cmd) {
1311 	case SIOCGIFSTATUS:
1312 		ifs = (struct ifstat *)data;
1313 		TUN_LOCK(tp);
1314 		if (tp->tun_pid)
1315 			snprintf(ifs->ascii, sizeof(ifs->ascii),
1316 			    "\tOpened by PID %d\n", tp->tun_pid);
1317 		else
1318 			ifs->ascii[0] = '\0';
1319 		TUN_UNLOCK(tp);
1320 		break;
1321 	case SIOCSIFADDR:
1322 		if (l2tun)
1323 			error = ether_ioctl(ifp, cmd, data);
1324 		else
1325 			tuninit(ifp);
1326 		if (error == 0)
1327 		    TUNDEBUG(ifp, "address set\n");
1328 		break;
1329 	case SIOCSIFMTU:
1330 		ifp->if_mtu = ifr->ifr_mtu;
1331 		TUNDEBUG(ifp, "mtu set\n");
1332 		break;
1333 	case SIOCSIFFLAGS:
1334 	case SIOCADDMULTI:
1335 	case SIOCDELMULTI:
1336 		break;
1337 	case SIOCGIFMEDIA:
1338 		if (!l2tun) {
1339 			error = EINVAL;
1340 			break;
1341 		}
1342 
1343 		ifmr = (struct ifmediareq *)data;
1344 		dummy = ifmr->ifm_count;
1345 		ifmr->ifm_count = 1;
1346 		ifmr->ifm_status = IFM_AVALID;
1347 		ifmr->ifm_active = IFM_ETHER | IFM_FDX | IFM_1000_T;
1348 		if (tp->tun_flags & TUN_OPEN)
1349 			ifmr->ifm_status |= IFM_ACTIVE;
1350 		ifmr->ifm_current = ifmr->ifm_active;
1351 		if (dummy >= 1) {
1352 			int media = IFM_ETHER;
1353 			error = copyout(&media, ifmr->ifm_ulist, sizeof(int));
1354 		}
1355 		break;
1356 	case SIOCSIFCAP:
1357 		TUN_LOCK(tp);
1358 		ifp->if_capenable = ifr->ifr_reqcap;
1359 		tun_caps_changed(ifp);
1360 		TUN_UNLOCK(tp);
1361 		VLAN_CAPABILITIES(ifp);
1362 		break;
1363 	default:
1364 		if (l2tun) {
1365 			error = ether_ioctl(ifp, cmd, data);
1366 		} else {
1367 			error = EINVAL;
1368 		}
1369 	}
1370 bad:
1371 	sx_xunlock(&tun_ioctl_sx);
1372 	return (error);
1373 }
1374 
1375 /*
1376  * tunoutput - queue packets from higher level ready to put out.
1377  */
1378 static int
1379 tunoutput(struct ifnet *ifp, struct mbuf *m0, const struct sockaddr *dst,
1380     struct route *ro)
1381 {
1382 	struct tuntap_softc *tp = ifp->if_softc;
1383 	u_short cached_tun_flags;
1384 	int error;
1385 	u_int32_t af;
1386 
1387 	TUNDEBUG (ifp, "tunoutput\n");
1388 
1389 #ifdef MAC
1390 	error = mac_ifnet_check_transmit(ifp, m0);
1391 	if (error) {
1392 		m_freem(m0);
1393 		return (error);
1394 	}
1395 #endif
1396 
1397 	/* Could be unlocked read? */
1398 	TUN_LOCK(tp);
1399 	cached_tun_flags = tp->tun_flags;
1400 	TUN_UNLOCK(tp);
1401 	if ((cached_tun_flags & TUN_READY) != TUN_READY) {
1402 		TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags);
1403 		m_freem (m0);
1404 		return (EHOSTDOWN);
1405 	}
1406 
1407 	if ((ifp->if_flags & IFF_UP) != IFF_UP) {
1408 		m_freem (m0);
1409 		return (EHOSTDOWN);
1410 	}
1411 
1412 	/* BPF writes need to be handled specially. */
1413 	if (dst->sa_family == AF_UNSPEC)
1414 		bcopy(dst->sa_data, &af, sizeof(af));
1415 	else
1416 		af = RO_GET_FAMILY(ro, dst);
1417 
1418 	BPF_MTAP2(ifp, &af, sizeof(af), m0);
1419 
1420 	/* prepend sockaddr? this may abort if the mbuf allocation fails */
1421 	if (cached_tun_flags & TUN_LMODE) {
1422 		/* allocate space for sockaddr */
1423 		M_PREPEND(m0, dst->sa_len, M_NOWAIT);
1424 
1425 		/* if allocation failed drop packet */
1426 		if (m0 == NULL) {
1427 			if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1428 			if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1429 			return (ENOBUFS);
1430 		} else {
1431 			bcopy(dst, m0->m_data, dst->sa_len);
1432 		}
1433 	}
1434 
1435 	if (cached_tun_flags & TUN_IFHEAD) {
1436 		/* Prepend the address family */
1437 		M_PREPEND(m0, 4, M_NOWAIT);
1438 
1439 		/* if allocation failed drop packet */
1440 		if (m0 == NULL) {
1441 			if_inc_counter(ifp, IFCOUNTER_IQDROPS, 1);
1442 			if_inc_counter(ifp, IFCOUNTER_OERRORS, 1);
1443 			return (ENOBUFS);
1444 		} else
1445 			*(u_int32_t *)m0->m_data = htonl(af);
1446 	} else {
1447 #ifdef INET
1448 		if (af != AF_INET)
1449 #endif
1450 		{
1451 			m_freem(m0);
1452 			return (EAFNOSUPPORT);
1453 		}
1454 	}
1455 
1456 	error = (ifp->if_transmit)(ifp, m0);
1457 	if (error)
1458 		return (ENOBUFS);
1459 	if_inc_counter(ifp, IFCOUNTER_OPACKETS, 1);
1460 	return (0);
1461 }
1462 
1463 /*
1464  * the cdevsw interface is now pretty minimal.
1465  */
1466 static	int
1467 tunioctl(struct cdev *dev, u_long cmd, caddr_t data, int flag,
1468     struct thread *td)
1469 {
1470 	struct ifreq ifr, *ifrp;
1471 	struct tuntap_softc *tp = dev->si_drv1;
1472 	struct ifnet *ifp = TUN2IFP(tp);
1473 	struct tuninfo *tunp;
1474 	int error, iflags, ival;
1475 	bool	l2tun;
1476 
1477 	l2tun = (tp->tun_flags & TUN_L2) != 0;
1478 	if (l2tun) {
1479 		/* tap specific ioctls */
1480 		switch(cmd) {
1481 		/* VMware/VMnet port ioctl's */
1482 #if defined(COMPAT_FREEBSD6) || defined(COMPAT_FREEBSD5) || \
1483     defined(COMPAT_FREEBSD4)
1484 		case _IO('V', 0):
1485 			ival = IOCPARM_IVAL(data);
1486 			data = (caddr_t)&ival;
1487 			/* FALLTHROUGH */
1488 #endif
1489 		case VMIO_SIOCSIFFLAGS: /* VMware/VMnet SIOCSIFFLAGS */
1490 			iflags = *(int *)data;
1491 			iflags &= TUN_VMIO_FLAG_MASK;
1492 			iflags &= ~IFF_CANTCHANGE;
1493 			iflags |= IFF_UP;
1494 
1495 			TUN_LOCK(tp);
1496 			ifp->if_flags = iflags |
1497 			    (ifp->if_flags & IFF_CANTCHANGE);
1498 			TUN_UNLOCK(tp);
1499 
1500 			return (0);
1501 		case SIOCGIFADDR:	/* get MAC address of the remote side */
1502 			TUN_LOCK(tp);
1503 			bcopy(&tp->tun_ether.octet, data,
1504 			    sizeof(tp->tun_ether.octet));
1505 			TUN_UNLOCK(tp);
1506 
1507 			return (0);
1508 		case SIOCSIFADDR:	/* set MAC address of the remote side */
1509 			TUN_LOCK(tp);
1510 			bcopy(data, &tp->tun_ether.octet,
1511 			    sizeof(tp->tun_ether.octet));
1512 			TUN_UNLOCK(tp);
1513 
1514 			return (0);
1515 		case TAPSVNETHDR:
1516 			ival = *(int *)data;
1517 			if (ival != 0 &&
1518 			    ival != sizeof(struct virtio_net_hdr) &&
1519 			    ival != sizeof(struct virtio_net_hdr_mrg_rxbuf)) {
1520 				return (EINVAL);
1521 			}
1522 			TUN_LOCK(tp);
1523 			tun_vnethdr_set(ifp, ival);
1524 			TUN_UNLOCK(tp);
1525 
1526 			return (0);
1527 		case TAPGVNETHDR:
1528 			TUN_LOCK(tp);
1529 			*(int *)data = tp->tun_vhdrlen;
1530 			TUN_UNLOCK(tp);
1531 
1532 			return (0);
1533 		}
1534 
1535 		/* Fall through to the common ioctls if unhandled */
1536 	} else {
1537 		switch (cmd) {
1538 		case TUNSLMODE:
1539 			TUN_LOCK(tp);
1540 			if (*(int *)data) {
1541 				tp->tun_flags |= TUN_LMODE;
1542 				tp->tun_flags &= ~TUN_IFHEAD;
1543 			} else
1544 				tp->tun_flags &= ~TUN_LMODE;
1545 			TUN_UNLOCK(tp);
1546 
1547 			return (0);
1548 		case TUNSIFHEAD:
1549 			TUN_LOCK(tp);
1550 			if (*(int *)data) {
1551 				tp->tun_flags |= TUN_IFHEAD;
1552 				tp->tun_flags &= ~TUN_LMODE;
1553 			} else
1554 				tp->tun_flags &= ~TUN_IFHEAD;
1555 			TUN_UNLOCK(tp);
1556 
1557 			return (0);
1558 		case TUNGIFHEAD:
1559 			TUN_LOCK(tp);
1560 			*(int *)data = (tp->tun_flags & TUN_IFHEAD) ? 1 : 0;
1561 			TUN_UNLOCK(tp);
1562 
1563 			return (0);
1564 		case TUNSIFMODE:
1565 			/* deny this if UP */
1566 			if (TUN2IFP(tp)->if_flags & IFF_UP)
1567 				return (EBUSY);
1568 
1569 			switch (*(int *)data & ~IFF_MULTICAST) {
1570 			case IFF_POINTOPOINT:
1571 			case IFF_BROADCAST:
1572 				TUN_LOCK(tp);
1573 				TUN2IFP(tp)->if_flags &=
1574 				    ~(IFF_BROADCAST|IFF_POINTOPOINT|IFF_MULTICAST);
1575 				TUN2IFP(tp)->if_flags |= *(int *)data;
1576 				TUN_UNLOCK(tp);
1577 
1578 				break;
1579 			default:
1580 				return (EINVAL);
1581 			}
1582 
1583 			return (0);
1584 		case TUNSIFPID:
1585 			TUN_LOCK(tp);
1586 			tp->tun_pid = curthread->td_proc->p_pid;
1587 			TUN_UNLOCK(tp);
1588 
1589 			return (0);
1590 		}
1591 		/* Fall through to the common ioctls if unhandled */
1592 	}
1593 
1594 	switch (cmd) {
1595 	case TUNGIFNAME:
1596 		ifrp = (struct ifreq *)data;
1597 		strlcpy(ifrp->ifr_name, TUN2IFP(tp)->if_xname, IFNAMSIZ);
1598 
1599 		return (0);
1600 	case TUNSIFINFO:
1601 		tunp = (struct tuninfo *)data;
1602 		if (TUN2IFP(tp)->if_type != tunp->type)
1603 			return (EPROTOTYPE);
1604 		TUN_LOCK(tp);
1605 		if (TUN2IFP(tp)->if_mtu != tunp->mtu) {
1606 			strlcpy(ifr.ifr_name, if_name(TUN2IFP(tp)), IFNAMSIZ);
1607 			ifr.ifr_mtu = tunp->mtu;
1608 			CURVNET_SET(TUN2IFP(tp)->if_vnet);
1609 			error = ifhwioctl(SIOCSIFMTU, TUN2IFP(tp),
1610 			    (caddr_t)&ifr, td);
1611 			CURVNET_RESTORE();
1612 			if (error) {
1613 				TUN_UNLOCK(tp);
1614 				return (error);
1615 			}
1616 		}
1617 		TUN2IFP(tp)->if_baudrate = tunp->baudrate;
1618 		TUN_UNLOCK(tp);
1619 		break;
1620 	case TUNGIFINFO:
1621 		tunp = (struct tuninfo *)data;
1622 		TUN_LOCK(tp);
1623 		tunp->mtu = TUN2IFP(tp)->if_mtu;
1624 		tunp->type = TUN2IFP(tp)->if_type;
1625 		tunp->baudrate = TUN2IFP(tp)->if_baudrate;
1626 		TUN_UNLOCK(tp);
1627 		break;
1628 	case TUNSDEBUG:
1629 		tundebug = *(int *)data;
1630 		break;
1631 	case TUNGDEBUG:
1632 		*(int *)data = tundebug;
1633 		break;
1634 	case FIONBIO:
1635 		break;
1636 	case FIOASYNC:
1637 		TUN_LOCK(tp);
1638 		if (*(int *)data)
1639 			tp->tun_flags |= TUN_ASYNC;
1640 		else
1641 			tp->tun_flags &= ~TUN_ASYNC;
1642 		TUN_UNLOCK(tp);
1643 		break;
1644 	case FIONREAD:
1645 		if (!IFQ_IS_EMPTY(&TUN2IFP(tp)->if_snd)) {
1646 			struct mbuf *mb;
1647 			IFQ_LOCK(&TUN2IFP(tp)->if_snd);
1648 			IFQ_POLL_NOLOCK(&TUN2IFP(tp)->if_snd, mb);
1649 			for (*(int *)data = 0; mb != NULL; mb = mb->m_next)
1650 				*(int *)data += mb->m_len;
1651 			IFQ_UNLOCK(&TUN2IFP(tp)->if_snd);
1652 		} else
1653 			*(int *)data = 0;
1654 		break;
1655 	case FIOSETOWN:
1656 		return (fsetown(*(int *)data, &tp->tun_sigio));
1657 
1658 	case FIOGETOWN:
1659 		*(int *)data = fgetown(&tp->tun_sigio);
1660 		return (0);
1661 
1662 	/* This is deprecated, FIOSETOWN should be used instead. */
1663 	case TIOCSPGRP:
1664 		return (fsetown(-(*(int *)data), &tp->tun_sigio));
1665 
1666 	/* This is deprecated, FIOGETOWN should be used instead. */
1667 	case TIOCGPGRP:
1668 		*(int *)data = -fgetown(&tp->tun_sigio);
1669 		return (0);
1670 
1671 	default:
1672 		return (ENOTTY);
1673 	}
1674 	return (0);
1675 }
1676 
1677 /*
1678  * The cdevsw read interface - reads a packet at a time, or at
1679  * least as much of a packet as can be read.
1680  */
1681 static	int
1682 tunread(struct cdev *dev, struct uio *uio, int flag)
1683 {
1684 	struct tuntap_softc *tp = dev->si_drv1;
1685 	struct ifnet	*ifp = TUN2IFP(tp);
1686 	struct mbuf	*m;
1687 	size_t		len;
1688 	int		error = 0;
1689 
1690 	TUNDEBUG (ifp, "read\n");
1691 	TUN_LOCK(tp);
1692 	if ((tp->tun_flags & TUN_READY) != TUN_READY) {
1693 		TUN_UNLOCK(tp);
1694 		TUNDEBUG (ifp, "not ready 0%o\n", tp->tun_flags);
1695 		return (EHOSTDOWN);
1696 	}
1697 
1698 	tp->tun_flags &= ~TUN_RWAIT;
1699 
1700 	for (;;) {
1701 		IFQ_DEQUEUE(&ifp->if_snd, m);
1702 		if (m != NULL)
1703 			break;
1704 		if (flag & O_NONBLOCK) {
1705 			TUN_UNLOCK(tp);
1706 			return (EWOULDBLOCK);
1707 		}
1708 		tp->tun_flags |= TUN_RWAIT;
1709 		error = mtx_sleep(tp, &tp->tun_mtx, PCATCH | (PZERO + 1),
1710 		    "tunread", 0);
1711 		if (error != 0) {
1712 			TUN_UNLOCK(tp);
1713 			return (error);
1714 		}
1715 	}
1716 	TUN_UNLOCK(tp);
1717 
1718 	len = min(tp->tun_vhdrlen, uio->uio_resid);
1719 	if (len > 0) {
1720 		struct virtio_net_hdr_mrg_rxbuf vhdr;
1721 
1722 		bzero(&vhdr, sizeof(vhdr));
1723 		if (m->m_pkthdr.csum_flags & TAP_ALL_OFFLOAD) {
1724 			m = virtio_net_tx_offload(ifp, m, false, &vhdr.hdr);
1725 		}
1726 
1727 		TUNDEBUG(ifp, "txvhdr: f %u, gt %u, hl %u, "
1728 		    "gs %u, cs %u, co %u\n", vhdr.hdr.flags,
1729 		    vhdr.hdr.gso_type, vhdr.hdr.hdr_len,
1730 		    vhdr.hdr.gso_size, vhdr.hdr.csum_start,
1731 		    vhdr.hdr.csum_offset);
1732 		error = uiomove(&vhdr, len, uio);
1733 	}
1734 
1735 	while (m && uio->uio_resid > 0 && error == 0) {
1736 		len = min(uio->uio_resid, m->m_len);
1737 		if (len != 0)
1738 			error = uiomove(mtod(m, void *), len, uio);
1739 		m = m_free(m);
1740 	}
1741 
1742 	if (m) {
1743 		TUNDEBUG(ifp, "Dropping mbuf\n");
1744 		m_freem(m);
1745 	}
1746 	return (error);
1747 }
1748 
1749 static int
1750 tunwrite_l2(struct tuntap_softc *tp, struct mbuf *m,
1751 	    struct virtio_net_hdr_mrg_rxbuf *vhdr)
1752 {
1753 	struct epoch_tracker et;
1754 	struct ether_header *eh;
1755 	struct ifnet *ifp;
1756 
1757 	ifp = TUN2IFP(tp);
1758 
1759 	/*
1760 	 * Only pass a unicast frame to ether_input(), if it would
1761 	 * actually have been received by non-virtual hardware.
1762 	 */
1763 	if (m->m_len < sizeof(struct ether_header)) {
1764 		m_freem(m);
1765 		return (0);
1766 	}
1767 
1768 	eh = mtod(m, struct ether_header *);
1769 
1770 	if ((ifp->if_flags & IFF_PROMISC) == 0 &&
1771 	    !ETHER_IS_MULTICAST(eh->ether_dhost) &&
1772 	    bcmp(eh->ether_dhost, IF_LLADDR(ifp), ETHER_ADDR_LEN) != 0) {
1773 		m_freem(m);
1774 		return (0);
1775 	}
1776 
1777 	if (vhdr != NULL) {
1778 		if (virtio_net_rx_csum(m, &vhdr->hdr)) {
1779 			m_freem(m);
1780 			return (0);
1781 		}
1782 	} else {
1783 		switch (ntohs(eh->ether_type)) {
1784 #ifdef INET
1785 		case ETHERTYPE_IP:
1786 			if (ifp->if_capenable & IFCAP_RXCSUM) {
1787 				m->m_pkthdr.csum_flags |=
1788 				    CSUM_IP_CHECKED | CSUM_IP_VALID |
1789 				    CSUM_DATA_VALID | CSUM_SCTP_VALID |
1790 				    CSUM_PSEUDO_HDR;
1791 				m->m_pkthdr.csum_data = 0xffff;
1792 			}
1793 			break;
1794 #endif
1795 #ifdef INET6
1796 		case ETHERTYPE_IPV6:
1797 			if (ifp->if_capenable & IFCAP_RXCSUM_IPV6) {
1798 				m->m_pkthdr.csum_flags |=
1799 				    CSUM_DATA_VALID_IPV6 | CSUM_SCTP_VALID |
1800 				    CSUM_PSEUDO_HDR;
1801 				m->m_pkthdr.csum_data = 0xffff;
1802 			}
1803 			break;
1804 #endif
1805 		}
1806 	}
1807 
1808 	/* Pass packet up to parent. */
1809 	CURVNET_SET(ifp->if_vnet);
1810 	NET_EPOCH_ENTER(et);
1811 #if defined(INET) || defined(INET6)
1812 	if (tp->tun_lro_ready && ifp->if_capenable & IFCAP_LRO &&
1813 	    tcp_lro_rx(&tp->tun_lro, m, 0) == 0)
1814 		tcp_lro_flush_all(&tp->tun_lro);
1815 	else
1816 #endif
1817 		(*ifp->if_input)(ifp, m);
1818 	NET_EPOCH_EXIT(et);
1819 	CURVNET_RESTORE();
1820 	/* ibytes are counted in parent */
1821 	if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1822 	return (0);
1823 }
1824 
1825 static int
1826 tunwrite_l3(struct tuntap_softc *tp, struct mbuf *m)
1827 {
1828 	struct epoch_tracker et;
1829 	struct ifnet *ifp;
1830 	int family, isr;
1831 
1832 	ifp = TUN2IFP(tp);
1833 	/* Could be unlocked read? */
1834 	TUN_LOCK(tp);
1835 	if (tp->tun_flags & TUN_IFHEAD) {
1836 		TUN_UNLOCK(tp);
1837 		if (m->m_len < sizeof(family) &&
1838 		(m = m_pullup(m, sizeof(family))) == NULL)
1839 			return (ENOBUFS);
1840 		family = ntohl(*mtod(m, u_int32_t *));
1841 		m_adj(m, sizeof(family));
1842 	} else {
1843 		TUN_UNLOCK(tp);
1844 		family = AF_INET;
1845 	}
1846 
1847 	BPF_MTAP2(ifp, &family, sizeof(family), m);
1848 
1849 	switch (family) {
1850 #ifdef INET
1851 	case AF_INET:
1852 		isr = NETISR_IP;
1853 		break;
1854 #endif
1855 #ifdef INET6
1856 	case AF_INET6:
1857 		isr = NETISR_IPV6;
1858 		break;
1859 #endif
1860 	default:
1861 		m_freem(m);
1862 		return (EAFNOSUPPORT);
1863 	}
1864 	random_harvest_queue(m, sizeof(*m), RANDOM_NET_TUN);
1865 	if_inc_counter(ifp, IFCOUNTER_IBYTES, m->m_pkthdr.len);
1866 	if_inc_counter(ifp, IFCOUNTER_IPACKETS, 1);
1867 	CURVNET_SET(ifp->if_vnet);
1868 	M_SETFIB(m, ifp->if_fib);
1869 	NET_EPOCH_ENTER(et);
1870 	netisr_dispatch(isr, m);
1871 	NET_EPOCH_EXIT(et);
1872 	CURVNET_RESTORE();
1873 	return (0);
1874 }
1875 
1876 /*
1877  * the cdevsw write interface - an atomic write is a packet - or else!
1878  */
1879 static	int
1880 tunwrite(struct cdev *dev, struct uio *uio, int flag)
1881 {
1882 	struct virtio_net_hdr_mrg_rxbuf vhdr;
1883 	struct tuntap_softc *tp;
1884 	struct ifnet	*ifp;
1885 	struct mbuf	*m;
1886 	uint32_t	mru;
1887 	int		align, vhdrlen, error;
1888 	bool		l2tun;
1889 
1890 	tp = dev->si_drv1;
1891 	ifp = TUN2IFP(tp);
1892 	TUNDEBUG(ifp, "tunwrite\n");
1893 	if ((ifp->if_flags & IFF_UP) != IFF_UP)
1894 		/* ignore silently */
1895 		return (0);
1896 
1897 	if (uio->uio_resid == 0)
1898 		return (0);
1899 
1900 	l2tun = (tp->tun_flags & TUN_L2) != 0;
1901 	mru = l2tun ? TAPMRU : TUNMRU;
1902 	vhdrlen = tp->tun_vhdrlen;
1903 	align = 0;
1904 	if (l2tun) {
1905 		align = ETHER_ALIGN;
1906 		mru += vhdrlen;
1907 	} else if ((tp->tun_flags & TUN_IFHEAD) != 0)
1908 		mru += sizeof(uint32_t);	/* family */
1909 	if (uio->uio_resid < 0 || uio->uio_resid > mru) {
1910 		TUNDEBUG(ifp, "len=%zd!\n", uio->uio_resid);
1911 		return (EIO);
1912 	}
1913 
1914 	if (vhdrlen > 0) {
1915 		error = uiomove(&vhdr, vhdrlen, uio);
1916 		if (error != 0)
1917 			return (error);
1918 		TUNDEBUG(ifp, "txvhdr: f %u, gt %u, hl %u, "
1919 		    "gs %u, cs %u, co %u\n", vhdr.hdr.flags,
1920 		    vhdr.hdr.gso_type, vhdr.hdr.hdr_len,
1921 		    vhdr.hdr.gso_size, vhdr.hdr.csum_start,
1922 		    vhdr.hdr.csum_offset);
1923 	}
1924 
1925 	if ((m = m_uiotombuf(uio, M_NOWAIT, 0, align, M_PKTHDR)) == NULL) {
1926 		if_inc_counter(ifp, IFCOUNTER_IERRORS, 1);
1927 		return (ENOBUFS);
1928 	}
1929 
1930 	m->m_pkthdr.rcvif = ifp;
1931 #ifdef MAC
1932 	mac_ifnet_create_mbuf(ifp, m);
1933 #endif
1934 
1935 	if (l2tun)
1936 		return (tunwrite_l2(tp, m, vhdrlen > 0 ? &vhdr : NULL));
1937 
1938 	return (tunwrite_l3(tp, m));
1939 }
1940 
1941 /*
1942  * tunpoll - the poll interface, this is only useful on reads
1943  * really. The write detect always returns true, write never blocks
1944  * anyway, it either accepts the packet or drops it.
1945  */
1946 static	int
1947 tunpoll(struct cdev *dev, int events, struct thread *td)
1948 {
1949 	struct tuntap_softc *tp = dev->si_drv1;
1950 	struct ifnet	*ifp = TUN2IFP(tp);
1951 	int		revents = 0;
1952 
1953 	TUNDEBUG(ifp, "tunpoll\n");
1954 
1955 	if (events & (POLLIN | POLLRDNORM)) {
1956 		IFQ_LOCK(&ifp->if_snd);
1957 		if (!IFQ_IS_EMPTY(&ifp->if_snd)) {
1958 			TUNDEBUG(ifp, "tunpoll q=%d\n", ifp->if_snd.ifq_len);
1959 			revents |= events & (POLLIN | POLLRDNORM);
1960 		} else {
1961 			TUNDEBUG(ifp, "tunpoll waiting\n");
1962 			selrecord(td, &tp->tun_rsel);
1963 		}
1964 		IFQ_UNLOCK(&ifp->if_snd);
1965 	}
1966 	revents |= events & (POLLOUT | POLLWRNORM);
1967 
1968 	return (revents);
1969 }
1970 
1971 /*
1972  * tunkqfilter - support for the kevent() system call.
1973  */
1974 static int
1975 tunkqfilter(struct cdev *dev, struct knote *kn)
1976 {
1977 	struct tuntap_softc	*tp = dev->si_drv1;
1978 	struct ifnet	*ifp = TUN2IFP(tp);
1979 
1980 	switch(kn->kn_filter) {
1981 	case EVFILT_READ:
1982 		TUNDEBUG(ifp, "%s kqfilter: EVFILT_READ, minor = %#x\n",
1983 		    ifp->if_xname, dev2unit(dev));
1984 		kn->kn_fop = &tun_read_filterops;
1985 		break;
1986 
1987 	case EVFILT_WRITE:
1988 		TUNDEBUG(ifp, "%s kqfilter: EVFILT_WRITE, minor = %#x\n",
1989 		    ifp->if_xname, dev2unit(dev));
1990 		kn->kn_fop = &tun_write_filterops;
1991 		break;
1992 
1993 	default:
1994 		TUNDEBUG(ifp, "%s kqfilter: invalid filter, minor = %#x\n",
1995 		    ifp->if_xname, dev2unit(dev));
1996 		return(EINVAL);
1997 	}
1998 
1999 	kn->kn_hook = tp;
2000 	knlist_add(&tp->tun_rsel.si_note, kn, 0);
2001 
2002 	return (0);
2003 }
2004 
2005 /*
2006  * Return true of there is data in the interface queue.
2007  */
2008 static int
2009 tunkqread(struct knote *kn, long hint)
2010 {
2011 	int			ret;
2012 	struct tuntap_softc	*tp = kn->kn_hook;
2013 	struct cdev		*dev = tp->tun_dev;
2014 	struct ifnet	*ifp = TUN2IFP(tp);
2015 
2016 	if ((kn->kn_data = ifp->if_snd.ifq_len) > 0) {
2017 		TUNDEBUG(ifp,
2018 		    "%s have data in the queue.  Len = %d, minor = %#x\n",
2019 		    ifp->if_xname, ifp->if_snd.ifq_len, dev2unit(dev));
2020 		ret = 1;
2021 	} else {
2022 		TUNDEBUG(ifp,
2023 		    "%s waiting for data, minor = %#x\n", ifp->if_xname,
2024 		    dev2unit(dev));
2025 		ret = 0;
2026 	}
2027 
2028 	return (ret);
2029 }
2030 
2031 /*
2032  * Always can write, always return MTU in kn->data.
2033  */
2034 static int
2035 tunkqwrite(struct knote *kn, long hint)
2036 {
2037 	struct tuntap_softc	*tp = kn->kn_hook;
2038 	struct ifnet	*ifp = TUN2IFP(tp);
2039 
2040 	kn->kn_data = ifp->if_mtu;
2041 
2042 	return (1);
2043 }
2044 
2045 static void
2046 tunkqdetach(struct knote *kn)
2047 {
2048 	struct tuntap_softc	*tp = kn->kn_hook;
2049 
2050 	knlist_remove(&tp->tun_rsel.si_note, kn, 0);
2051 }
2052